diff options
author | grog <grog@FreeBSD.org> | 1998-09-16 05:56:21 +0000 |
---|---|---|
committer | grog <grog@FreeBSD.org> | 1998-09-16 05:56:21 +0000 |
commit | 633c70539f9a75caebb97293bd237b0611df1427 (patch) | |
tree | 06e58703e575814c95eee8c95d24a99e35d8eb69 | |
download | FreeBSD-src-633c70539f9a75caebb97293bd237b0611df1427.zip FreeBSD-src-633c70539f9a75caebb97293bd237b0611df1427.tar.gz |
Import base vinum lkm sources
69 files changed, 24705 insertions, 0 deletions
diff --git a/lkm/vinum/COPYRIGHT b/lkm/vinum/COPYRIGHT new file mode 100644 index 0000000..4ee95f2 --- /dev/null +++ b/lkm/vinum/COPYRIGHT @@ -0,0 +1,37 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: COPYRIGHT,v 1.1 1998/03/05 06:07:05 grog Exp grog $ + */ diff --git a/lkm/vinum/Makefile b/lkm/vinum/Makefile new file mode 100644 index 0000000..4b88182 --- /dev/null +++ b/lkm/vinum/Makefile @@ -0,0 +1,26 @@ +# $Id: Makefile.lkm.lite,v 1.2 1998/08/13 06:07:29 grog Exp grog $ + +.PATH: ${.CURDIR}/../../sys/dev/ccd +KMOD= vinum_mod +SRCS= vinum.c vinum.h vnode_if.h parser.c config.c io.c util.c vinumhdr.h request.h \ + state.c memory.c request.c lock.c vinumext.h vinumio.h vinumkw.h \ + vinumstate.h vinumvar.h revive.c vinumioctl.c interrupt.c +NOMAN= +PSEUDO_LKM= +CFLAGS = -I. -O -g -I/usr/include/machine -DDEBUG -Wall -Wno-unused -Wno-parentheses + +CLEANFILES+= vinum.h vnode_if.h vnode_if.c + +all: + +# We don't need this, but the Makefile wants it +vinum.h: + touch $@ + +state.h: maketabs vinumstate.h + ./maketabs >state.h + +maketabs: maketabs.c + ${CC} -g -o maketabs maketabs.c + +.include <bsd.kmod.mk> diff --git a/lkm/vinum/config.c b/lkm/vinum/config.c new file mode 100644 index 0000000..fd31c5e --- /dev/null +++ b/lkm/vinum/config.c @@ -0,0 +1,1712 @@ +/* To do: + + * Don't store drive configuration on the config DB: read each drive's header + * to decide where it is. + * + * Accept any old crap in the config_<foo> functions, and complain when + * we try to bring it up. + * + * When trying to bring volumes up, check that the complete address range + * is covered. + */ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: config.c,v 1.17 1998/08/14 04:49:26 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#define REALLYKERNEL +#include "vinumhdr.h" + +extern jmp_buf command_fail; /* return on a failed command */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +void longjmp(jmp_buf, int); +#endif + +#define MAXTOKEN 64 /* maximum number of tokens in a line */ + +/* We can afford the luxury of global variables here, + * since start_config ensures that these functions + * are single-threaded. */ + +/* These are indices in vinum_conf of the last-mentioned of each kind of object */ +static int current_drive = -1; /* note the last drive we mention, for + * some defaults */ +static int current_plex = -1; /* and the same for the last plex */ +static int current_volume = -1; /* and the last volme */ +static struct _ioctl_reply *ioctl_reply; /* struct to return via ioctl */ + + +/* These values are used by most of these routines, so set them as globals */ +static char *token[MAXTOKEN]; /* pointers to individual tokens */ +static int tokens; /* number of tokens */ + +#define TOCONS 0x01 +#define TOTTY 0x02 +#define TOLOG 0x04 + +struct putchar_arg { + int flags; + struct tty *tty; +}; + +#define MSG_MAX 1024 /* maximum length of a formatted message */ +/* Format an error message and return to the user in the reply. + * CARE: This routine is designed to be called only from the + * configuration routines, so it assumes it's the owner of + * the configuration lock, and unlocks it on exit */ +void +throw_rude_remark(int error, char *msg,...) +{ + BROKEN_GDB; + int retval; + va_list ap; + char *text; + static int finishing; /* don't recurse */ + int was_finishing; + + va_start(ap, msg); + if ((ioctl_reply != NULL) /* we're called from the user */ + &&(!(vinum_conf.flags & VF_KERNELOP))) { /* and we're not doing kernel things: return msg */ + /* XXX We can't just format to ioctl_reply, since it + * may contain our input parameters */ + text = Malloc(MSG_MAX); + if (text == NULL) { + printf("vinum: can't allocate error message buffer"); + printf("vinum: "); + vprintf(msg, ap); /* print to the console */ + printf("\n"); + } else { + retval = kvprintf(msg, NULL, (void *) text, 10, ap); + text[retval] = '\0'; /* delimit */ + strcpy(ioctl_reply->msg, text); + ioctl_reply->error = error; /* first byte is the error number */ + Free(text); + } + } else { + printf("vinum: "); + vprintf(msg, ap); /* print to the console */ + printf("\n"); + } + va_end(ap); + + if (vinum_conf.flags & VF_READING_CONFIG) /* go through to the bitter end, */ + return; + /* We have a problem here: we want to unlock the + * configuration, which implies tidying up, but + * if we find an error while tidying up, we could + * recurse for ever. Use this kludge to only try + * once */ + was_finishing = finishing; + finishing = 1; + finish_config(was_finishing); /* unlock anything we may be holding */ + finishing = was_finishing; + longjmp(command_fail, error); +} + +/* Function declarations */ +int atoi(char *); /* no atoi in the kernel */ + +/* Minimal version of atoi */ +int +atoi(char *s) +{ /* no atoi in the kernel */ + BROKEN_GDB; + int r = 0; + int sign = 1; + + while (((*s >= '0') && (*s <= '9')) || (*s == '-')) { + if (*s == '-') + sign = -sign; + else + r = r * 10 + (*s - '0'); + } + return r; +} + +/* Find index of volume in vinum_conf. Return the index + * if found, or -1 if not */ +int +volume_index(struct volume *vol) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.volumes_used; i++) + if (&VOL[i] == vol) + return i; + return -1; +} + +/* Find index of plex in vinum_conf. Return the index + * if found, or -1 if not */ +int +plex_index(struct plex *plex) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.plexes_used; i++) + if (&PLEX[i] == plex) + return i; + return -1; +} + +/* Find index of subdisk in vinum_conf. Return the index + * if found, or -1 if not */ +int +sd_index(struct sd *sd) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.subdisks_used; i++) + if (&SD[i] == sd) + return i; + return -1; +} + +/* Find index of drive in vinum_conf. Return the index + * if found, or -1 if not */ +int +drive_index(struct drive *drive) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.drives_used; i++) + if (&DRIVE[i] == drive) + return i; + return -1; +} + +/* Check a volume to see if the plex is already assigned to it. + * Return index in volume->plex, or -1 if not assigned */ +int +my_plex(int volno, int plexno) +{ + BROKEN_GDB; + int i; + struct volume *vol; + + vol = &VOL[volno]; /* point to volno */ + for (i = 0; i < vol->plexes; i++) + if (vol->plex[i] == plexno) + return i; + return -1; /* not found */ +} + +/* Check a plex to see if the subdisk is already assigned to it. + * Return index in plex->sd, or -1 if not assigned */ +int +my_sd(int plexno, int sdno) +{ + BROKEN_GDB; + int i; + struct plex *plex; + + plex = &PLEX[plexno]; + for (i = 0; i < plex->subdisks; i++) + if (plex->sdnos[i] == sdno) + return i; + return -1; /* not found */ +} + +/* Check that this operation is being done in the kernel. + * longjmp out if not. op the name of the operation. */ +void +checkkernel(char *op) +{ + BROKEN_GDB; + if (vinum_conf.flags & VF_KERNELOP == 0) + throw_rude_remark(EPERM, "Can't perform '%s' from user space", op); +} + +/* Add plex to the volume if possible */ +int +give_plex_to_volume(int volno, int plexno) +{ + BROKEN_GDB; + struct volume *vol; + + /* XXX It's not an error for the plex to already + * belong to the volume, but we need to check a + * number of things to make sure it's done right. + * Some day. */ + if (my_plex(volno, plexno) >= 0) + return plexno; /* that's it */ + + vol = &VOL[volno]; /* point to volume */ + if (vol->plexes == MAXPLEX) /* all plexes allocated */ + throw_rude_remark(ENOSPC, + "Too many plexes for volume %s", + vol->name); + vol->plex[vol->plexes] = plexno; /* this one */ + vol->plexes++; /* add another plex */ + PLEX[plexno].volno = volno; /* note the number of our volume */ + + return vol->plexes - 1; /* and return its index */ +} + +/* Add subdisk to a plex if possible */ +int +give_sd_to_plex(int plexno, int sdno) +{ + BROKEN_GDB; + int i; + struct plex *plex; + struct sd *sd; + + /* XXX It's not an error for the sd to already + * belong to the plex, but we need to check a + * number of things to make sure it's done right. + * Some day. */ + i = my_sd(plexno, sdno); + if (i >= 0) /* does it already belong to us? */ + return i; /* that's it */ + + plex = &PLEX[plexno]; /* point to the plex */ + sd = &SD[sdno]; /* and the subdisk */ + + /* Do we have an offset? Otherwise put it after the last one */ + if (sd->plexoffset < 0) { /* no offset specified */ + if (plex->subdisks > 0) { + struct sd *lastsd = &SD[plex->sdnos[plex->subdisks - 1]]; /* last subdisk */ + sd->plexoffset = lastsd->sectors + lastsd->plexoffset; /* take it */ + } else /* first subdisk */ + sd->plexoffset = 0; /* start at the beginning */ + } + plex->subdisks++; /* another entry */ + if (plex->subdisks >= plex->subdisks_allocated) /* need more space */ + EXPAND(plex->sdnos, int, plex->subdisks_allocated, INITIAL_SUBDISKS_IN_PLEX); + + /* XXX I'm not sure this makes any sense + * for anything except concatenated plexes, + * and it comes up with the wrong answer for + * RAID-5 plexes, but it's currently needed + * for the calculations. We'll adjust for + * RAID-5 in config_plex */ + if ((sd->sectors + sd->plexoffset) > plex->length) { /* gone beyond the end of the plex */ + plex->length = sd->sectors + sd->plexoffset; /* adjust the length */ + + if ((plex->volno >= 0) /* we have a volume */ + &&(plex->length > VOL[plex->volno].size)) /* and we're now the longest plex */ + VOL[plex->volno].size = plex->length; /* increase the size of the volume */ + } + /* We need to check that the subdisks don't overlap, + * but we can't do that until a point where we *must* + * know the size of all the subdisks. That's not + * here. But we need to sort them by offset */ + for (i = 0; i < plex->subdisks - 1; i++) { + if (sd->plexoffset < SD[plex->sdnos[i]].plexoffset) { /* it fits before this one */ + /* First move any remaining subdisks by one */ + int j; + + for (j = plex->subdisks - 1; j > i; j--) /* move up one at a time */ + plex->sdnos[j] = plex->sdnos[j - 1]; + plex->sdnos[i] = sdno; + return i; + } + } + + /* The plex doesn't have any subdisk with a larger + * offset. Insert it */ + plex->sdnos[i] = sdno; + return i; +} + +/* Add a subdisk to drive if possible. The pointer to the drive + * must already be stored in the sd structure, but the drive + * doesn't know about the subdisk yet. */ +static void +give_sd_to_drive(int sdno) +{ + BROKEN_GDB; + struct sd *sd; /* pointer to subdisk */ + struct drive *drive; /* and drive */ + int fe; /* index in free list */ + + sd = &SD[sdno]; /* point to sd */ + drive = &DRIVE[sd->driveno]; /* and drive */ + + if (drive->state != drive_up) /* not up */ + throw_rude_remark(EIO, "Drive %s is not accessible", drive->label.name); + else if (sd->sectors > drive->sectors_available) { /* too big, */ + sd->driveoffset = -1; /* don't be confusing */ + throw_rude_remark(ENOSPC, "No space for %s on %s", sd->name, drive->label.name); + } + drive->subdisks_used++; /* one more subdisk */ + + /* no offset specified, find one */ + if (sd->driveoffset < 0) { + for (fe = 0; fe < drive->freelist_entries; fe++) { + if (drive->freelist[fe].sectors >= sd->sectors) { /* it'll fit here */ + sd->driveoffset = drive->freelist[fe].offset; + if (sd->sectors == drive->freelist[fe].sectors) { /* used up the entire entry */ + if (fe < (drive->freelist_entries - 1)) /* not the last one, */ + bcopy(&drive->freelist[fe + 1], + &drive->freelist[fe], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries--; /* one less entry */ + } else { + drive->freelist[fe].sectors -= sd->sectors; /* this much less space */ + drive->freelist[fe].offset += sd->sectors; /* this much further on */ + } + drive->sectors_available -= sd->sectors; /* and note how much less space we have */ + break; + } + } + if (fe == drive->freelist_entries) + /* Didn't find anything. Although the drive has + * enough space, it's too fragmented */ + { + sd->driveoffset = -1; /* don't be confusing */ + throw_rude_remark(ENOSPC, "No space for %s on %s", sd->name, drive->label.name); + } + } else { /* specific offset */ + /* For a specific offset to work, the space must be + * entirely in a single freelist entry. Look for it. */ + u_int64_t sdend = sd->driveoffset + sd->sectors; /* end of our subdisk */ + for (fe = 0; fe < drive->freelist_entries; fe++) { + u_int64_t dend = drive->freelist[fe].offset + drive->freelist[fe].sectors; /* end of entry */ + if (dend >= sdend) { /* fits before here */ + if (drive->freelist[fe].offset > sd->driveoffset) /* starts after the beginning of sd area */ + throw_rude_remark(ENOSPC, + "No space for subdisk %s on drive %s at offset %qd\n", + sd->name, + drive->label.name); + + /* We've found the space, and we can allocate it. + * We don't need to say that to the subdisk, which + * already knows about it. We need to tell it to + * the free list, though. We have four possibilities: + * + * 1. The subdisk exactly eats up the entry. That's the + * same as above. + * 2. The subdisk starts at the beginning and leaves space + * at the end. + * 3. The subdisk starts after the beginning and leaves + * space at the end as well: we end up with another + * fragment. + * 4. The subdisk leaves space at the beginning and finishes + * at the end. + */ + drive->sectors_available -= sd->sectors; /* note how much less space we have */ + if (sd->driveoffset == drive->freelist[fe].offset) { /* 1 or 2 */ + if (sd->sectors == drive->freelist[fe].sectors) { /* 1: used up the entire entry */ + if (fe < (drive->freelist_entries - 1)) /* not the last one, */ + bcopy(&drive->freelist[fe + 1], + &drive->freelist[fe], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries--; /* one less entry */ + } else { /* 2: space at the end */ + drive->freelist[fe].sectors -= sd->sectors; /* this much less space */ + drive->freelist[fe].offset += sd->sectors; /* this much further on */ + } + } else { /* 3 or 4 */ + drive->freelist[fe].sectors = sd->driveoffset - drive->freelist[fe].offset; + if (dend > sdend) { /* 3: space at the end as well */ + if (fe < (drive->freelist_entries - 1)) /* not the last one */ + bcopy(&drive->freelist[fe], /* move the rest down */ + &drive->freelist[fe + 1], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries++; /* one less entry */ + drive->freelist[fe + 1].offset = sdend; /* second entry starts after sd */ + drive->freelist[fe + 1].sectors = dend - sdend; /* and is this long */ + } + } + break; + } + } + } + drive->opencount++; /* one more subdisk attached */ +} + +/* Get an empty drive entry from the drive table */ +int +get_empty_drive(void) +{ + BROKEN_GDB; + int driveno; + struct drive *drive; + + /* first see if we have one which has been deallocated */ + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + if (DRIVE[driveno].state == drive_unallocated) /* bingo */ + break; + } + + if (driveno >= vinum_conf.drives_used) + /* Couldn't find a deallocated drive. Allocate a new one */ + { + vinum_conf.drives_used++; + if (vinum_conf.drives_used > vinum_conf.drives_allocated) /* we've used all our allocation */ + EXPAND(DRIVE, struct drive, vinum_conf.drives_allocated, INITIAL_DRIVES); + } + /* got a drive entry. Make it pretty */ + drive = &DRIVE[driveno]; + bzero(drive, sizeof(struct drive)); + drive->driveno = driveno; /* put number in structure */ + return driveno; /* return the index */ +} + +/* Find the named drive in vinum_conf.drive, return a pointer + * return the index in vinum_conf.drive. + * Don't mark the drive as allocated (XXX MP) + * If create != 0, create an entry if it doesn't exist + */ +/* XXX check if we have it open from attach */ +int +find_drive(const char *name, int create) +{ + BROKEN_GDB; + int driveno; + struct drive *drive; + + if (name != NULL) { + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + drive = &DRIVE[driveno]; /* point to drive */ + if ((drive->label.name[0] != '\0') /* it has a name */ + &&(strcmp(drive->label.name, name) == 0)) /* and it's this one: found */ + return driveno; + } + } + /* the drive isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + driveno = get_empty_drive(); + drive = &DRIVE[driveno]; + if (name != NULL) + bcopy(name, /* put in its name */ + drive->label.name, + min(sizeof(drive->label.name), + strlen(name))); + drive->state = drive_uninit; /* in use, nothing worthwhile there */ + return driveno; /* return the index */ +} + +/* Find a drive given its device name. + * devname must be valid. + * Otherwise the same as find_drive above */ +int +find_drive_by_dev(const char *devname, int create) +{ + BROKEN_GDB; + int driveno; + struct drive *drive; + + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + drive = &DRIVE[driveno]; /* point to drive */ + if ((drive->label.name[0] != '\0') /* it has a name */ + &&(strcmp(drive->label.name, devname) == 0)) /* and it's this one: found */ + return driveno; + } + + /* the drive isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + driveno = get_empty_drive(); + drive = &DRIVE[driveno]; + bcopy(devname, /* put in its name */ + drive->devicename, + min(sizeof(drive->devicename), + strlen(devname))); + drive->state = drive_uninit; /* in use, nothing worthwhile there */ + return driveno; /* return the index */ +} + +/* Find an empty subdisk in the subdisk table */ +int +get_empty_sd(void) +{ + BROKEN_GDB; + int sdno; + struct sd *sd; + + /* first see if we have one which has been deallocated */ + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) { + if (SD[sdno].state == sd_unallocated) /* bingo */ + break; + } + + if (sdno >= vinum_conf.subdisks_used) { /* No unused sd found. Allocate a new one */ + vinum_conf.subdisks_used++; + if (vinum_conf.subdisks_used > vinum_conf.subdisks_allocated) + EXPAND(SD, struct sd, vinum_conf.subdisks_allocated, INITIAL_SUBDISKS); + } + /* initialize some things */ + sd = &SD[sdno]; /* point to it */ + bzero(sd, sizeof(struct sd)); /* initialize */ + sd->plexno = -1; /* no plex */ + sd->driveno = -1; /* and no drive */ + sd->plexoffset = -1; /* and no offsets */ + sd->driveoffset = -1; + return sdno; /* return the index */ +} + +/* return a drive to the free pool */ +void +free_drive(struct drive *drive) +{ + BROKEN_GDB; + if (drive->vp != NULL) /* device open */ + vn_close(drive->vp, FREAD | FWRITE, FSCRED, drive->p); + bzero(drive, sizeof(struct drive)); /* this also sets drive_unallocated */ +} + +/* Find the named subdisk in vinum_conf.sd. + + * If create != 0, create an entry if it doesn't exist + * + * Return index in vinum_conf.sd + */ +int +find_subdisk(const char *name, int create) +{ + BROKEN_GDB; + int sdno; + struct sd *sd; + + for (sdno = 0; sdno < vinum_conf.subdisks_allocated; sdno++) { + if (strcmp(SD[sdno].name, name) == 0) /* found it */ + return sdno; + } + + /* the subdisk isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + /* Allocate one and insert the name */ + sdno = get_empty_sd(); + sd = &SD[sdno]; + bcopy(name, sd->name, min(sizeof(sd->name), strlen(name))); /* put in its name */ + return sdno; /* return the pointer */ +} + +/* Free an allocated sd entry + * This performs memory management only. remove() + * is responsible for checking relationships. + */ +void +free_sd(int sdno) +{ + BROKEN_GDB; + struct sd *sd; + struct drive *drive; + int fe; /* free list entry */ + u_int64_t sdend; /* end of our subdisk */ + u_int64_t dend; /* end of our freelist entry */ + + sd = &SD[sdno]; + if ((sd->driveno >= 0) /* we have a drive, */ + &&(sd->sectors > 0)) { /* and some space on it */ + drive = &DRIVE[sd->driveno]; + sdend = sd->driveoffset + sd->sectors; /* end of our subdisk */ + + /* Look for where to return the sd address space */ + for (fe = 0; + (fe < drive->freelist_entries) && (drive->freelist[fe].offset < sd->driveoffset); + fe++); + /* Now we are pointing to the last entry, the first + * with a higher offset than the subdisk, or both. */ + if ((fe > 1) /* not the first entry */ + &&((fe == drive->freelist_entries) /* gone past the end */ + ||(drive->freelist[fe].offset > sd->driveoffset))) /* or past the block were looking for */ + fe--; /* point to the block before */ + dend = drive->freelist[fe].offset + drive->freelist[fe].sectors; /* end of the entry */ + + /* At this point, we are pointing to the correct + * place in the free list. A number of possibilities + * exist: + * + * 1. The block to be freed immediately follows + * the block to which we are pointing. Just + * enlarge it. + * 2. The block to be freed starts at the end of + * the current block and ends at the beginning + * of the following block. Merge the three + * areas into a single block. + * 3. The block to be freed starts after the end + * of the block and ends before the start of + * the following block. Create a new free block. + * 4. The block to be freed starts after the end + * of the block, but ends at the start of the + * following block. Enlarge the following block + * downwards. + * + */ + if (sd->driveoffset == dend) { /* it starts after the end of this block */ + if ((fe < drive->freelist_entries - 1) /* we're not the last block in the free list */ + &&(sdend == drive->freelist[fe + 1].offset)) { /* and the subdisk ends at the start of the + * next block */ + drive->freelist[fe].sectors = drive->freelist[fe + 1].sectors; /* 2: merge all three blocks */ + if (fe < drive->freelist_entries - 2) /* still more blocks after next */ + bcopy(&drive->freelist[fe + 2], /* move down one */ + &drive->freelist[fe + 1], + (drive->freelist_entries - 2 - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries--; /* one less entry in the free list */ + } else /* 1: just enlarge this block */ + drive->freelist[fe].sectors += sd->sectors; + } else { + if (sd->driveoffset > dend) /* it starts after this block */ + fe++; /* so look at the next block */ + if ((fe < drive->freelist_entries) /* we're not the last block in the free list */ + &&(sdend == drive->freelist[fe].offset)) { /* and the subdisk ends at the start of + * this block: case 4 */ + drive->freelist[fe].offset = sd->driveoffset; /* it starts where the sd was */ + drive->freelist[fe].sectors += sd->sectors; /* and it's this much bigger */ + } else { /* case 3: non-contiguous */ + if (fe < drive->freelist_entries) /* not after the last block, */ + bcopy(&drive->freelist[fe], /* move the rest up one entry */ + &drive->freelist[fe + 1], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries++; /* one less entry */ + drive->freelist[fe].offset = sd->driveoffset; /* this entry represents the sd */ + drive->freelist[fe].sectors = sd->sectors; + } + } + drive->opencount--; /* one less subdisk attached */ + } + bzero(sd, sizeof(struct sd)); /* and clear it out */ + sd->state = sd_unallocated; +} + +/* Find an empty plex in the plex table */ +int +get_empty_plex(void) +{ + BROKEN_GDB; + int plexno; + struct plex *plex; /* if we allocate one */ + + /* first see if we have one which has been deallocated */ + for (plexno = 0; plexno < vinum_conf.plexes_used; plexno++) { + if (PLEX[plexno].state == plex_unallocated) /* bingo */ + break; /* and get out of here */ + } + + if (plexno >= vinum_conf.plexes_used) { + /* Couldn't find a deallocated plex. Allocate a new one */ + vinum_conf.plexes_used++; + if (vinum_conf.plexes_used > vinum_conf.plexes_allocated) + EXPAND(PLEX, struct plex, vinum_conf.plexes_allocated, INITIAL_PLEXES); + } + /* Found a plex. Give it an sd structure */ + plex = &PLEX[plexno]; /* this one is ours */ + bzero(plex, sizeof(struct plex)); /* polish it up */ + plex->sdnos = (int *) Malloc(sizeof(int) * INITIAL_SUBDISKS_IN_PLEX); /* allocate sd table */ + CHECKALLOC(plex->sdnos, "vinum: Can't allocate plex subdisk table"); + bzero(plex->sdnos, (sizeof(int) * INITIAL_SUBDISKS_IN_PLEX)); /* do we need this? */ + plex->subdisks = 0; /* no subdisks in use */ + plex->subdisks_allocated = INITIAL_SUBDISKS_IN_PLEX; /* and we have space for this many */ + plex->organization = plex_disorg; /* and it's not organized */ + plex->volno = -1; /* no volume yet */ + return plexno; /* return the index */ +} + +/* Find the named plex in vinum_conf.plex + + * If create != 0, create an entry if it doesn't exist + * return index in vinum_conf.plex + */ +int +find_plex(const char *name, int create) +{ + BROKEN_GDB; + int plexno; + struct plex *plex; + + for (plexno = 0; plexno < vinum_conf.plexes_allocated; plexno++) { + if (strcmp(PLEX[plexno].name, name) == 0) /* found it */ + return plexno; + } + + /* the plex isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + /* Allocate one and insert the name */ + plexno = get_empty_plex(); + plex = &PLEX[plexno]; /* point to it */ + bcopy(name, plex->name, min(sizeof(plex->name), strlen(name))); /* put in its name */ + return plexno; /* return the pointer */ +} + +/* Free an allocated plex entry + * and its associated memory areas */ +void +free_plex(int plexno) +{ + BROKEN_GDB; + struct plex *plex; + + plex = &PLEX[plexno]; + if (plex->sdnos) + Free(plex->sdnos); + if (plex->lock) + Free(plex->lock); + if (plex->defective_region) + Free(plex->defective_region); + if (plex->unmapped_region) + Free(plex->unmapped_region); + bzero(plex, sizeof(struct plex)); /* and clear it out */ + plex->state = plex_unallocated; +} + +/* Find an empty volume in the volume table */ +int +get_empty_volume(void) +{ + BROKEN_GDB; + int volno; + struct volume *vol; + + /* first see if we have one which has been deallocated */ + for (volno = 0; volno < vinum_conf.volumes_used; volno++) { + if (VOL[volno].state == volume_unallocated) /* bingo */ + break; + } + + if (volno >= vinum_conf.volumes_used) + /* Couldn't find a deallocated volume. Allocate a new one */ + { + vinum_conf.volumes_used++; + if (vinum_conf.volumes_used > vinum_conf.volumes_allocated) + EXPAND(VOL, struct volume, vinum_conf.volumes_allocated, INITIAL_VOLUMES); + } + /* Now initialize fields */ + vol = &VOL[volno]; + bzero(vol, sizeof(struct volume)); + vol->preferred_plex = -1; /* default to round robin */ + vol->preferred_plex = ROUND_ROBIN_READPOL; /* round robin */ + + return volno; /* return the index */ +} + +/* Find the named volume in vinum_conf.volume. + + * If create != 0, create an entry if it doesn't exist + * return the index in vinum_conf + */ +int +find_volume(const char *name, int create) +{ + BROKEN_GDB; + int volno; + struct volume *vol; + + for (volno = 0; volno < vinum_conf.volumes_used; volno++) { + if (strcmp(VOL[volno].name, name) == 0) /* found it */ + return volno; + } + + /* the volume isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + /* Allocate one and insert the name */ + volno = get_empty_volume(); + vol = &VOL[volno]; + bcopy(name, vol->name, min(sizeof(vol->name), strlen(name))); /* put in its name */ + vol->blocksize = DEV_BSIZE; /* block size of this volume */ + return volno; /* return the pointer */ +} + +/* Free an allocated volume entry + * and its associated memory areas */ +void +free_volume(int volno) +{ + BROKEN_GDB; + struct volume *vol; + + vol = &VOL[volno]; + bzero(vol, sizeof(struct volume)); /* and clear it out */ + vol->state = volume_unallocated; +} + +/* Handle a drive definition. We store the information in the global variable + * drive, so we don't need to allocate. + * + * If we find an error, print a message and return + */ +void +config_drive(void) +{ + BROKEN_GDB; + enum drive_label_info partition_status; /* info about the partition */ + int parameter; + int driveno; /* index of drive in vinum_conf */ + struct drive *drive; /* and pointer to it */ + + if (tokens < 2) /* not enough tokens */ + throw_rude_remark(EINVAL, "Drive has no name"); + driveno = find_drive(token[1], 1); /* allocate a drive to initialize */ + drive = &DRIVE[driveno]; /* and get a pointer */ + + if (drive->state != drive_uninit) { /* we already know this drive */ + /* XXX Check which definition is more up-to-date. Give + * preference for the definition on its own drive */ + return; /* XXX */ + } + for (parameter = 2; parameter < tokens; parameter++) { /* look at the other tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_device: + parameter++; + if (drive->devicename[0] != '\0') { /* we know this drive... */ + if (strcmp(drive->devicename, token[parameter])) /* different name */ + close_drive(drive); /* close it if it's open */ + else /* no change */ + break; + } + bcopy(token[parameter], /* insert device information */ + drive->devicename, + min(sizeof(drive->devicename), + strlen(token[parameter]))); + /* open the device and get the configuration */ + partition_status = read_drive_label(drive); + if (partition_status == DL_CANT_OPEN) { /* not our kind */ + close_drive(drive); + if (drive->lasterror == EFTYPE) /* wrong kind of partition */ + throw_rude_remark(drive->lasterror, + "Drive %s has invalid partition type", + drive->label.name); + else /* I/O error of some kind */ + throw_rude_remark(drive->lasterror, + "Can't initialize drive %s", + drive->label.name); + } else if (partition_status == DL_WRONG_DRIVE) { /* valid drive, not ours */ + close_drive(drive); + throw_rude_remark(drive->lasterror, + "Incorrect drive name %s specified for drive %s", + token[1], + drive->label.name); + } + break; + + case kw_state: + checkkernel(token[++parameter]); /* must be a kernel user */ + drive->state = DriveState(token[parameter]); /* set the state */ + break; + + default: + close_drive(drive); + throw_rude_remark(EINVAL, + "Drive %s, invalid keyword: %s", + token[1], + token[parameter]); + } + } + + if (drive->devicename[0] == '\0') + throw_rude_remark(EINVAL, "No device name for %s", drive->label.name); + +} + +/* Handle a subdisk definition. We store the information in the global variable + * sd, so we don't need to allocate. + * + * If we find an error, print a message and return + */ +void +config_subdisk(void) +{ + BROKEN_GDB; + int parameter; + int sdno; /* index of sd in vinum_conf */ + struct sd *sd; /* and pointer to it */ + u_int64_t size; + int sectors; /* sector offset value */ + int detached = 0; /* set to 1 if this is a detached subdisk */ + int sdindex = -1; /* index in plexes subdisk table */ + + sdno = get_empty_sd(); /* allocate an SD to initialize */ + sd = &SD[sdno]; /* and get a pointer */ + for (parameter = 1; parameter < tokens; parameter++) { /* look at the other tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_detached: + detached = 1; + break; + + case kw_plexoffset: + size = sizespec(token[++parameter]); + if ((size % DEV_BSIZE) != 0) + throw_rude_remark(EINVAL, "sd %s, bad plex offset alignment: %qd", sd->name, size); + else + sd->plexoffset = size / DEV_BSIZE; + break; + + case kw_driveoffset: + size = sizespec(token[++parameter]); + if ((size % DEV_BSIZE) != 0) + throw_rude_remark(EINVAL, "sd %s, bad drive offset alignment: %qd", sd->name, size); + else + sd->driveoffset = size / DEV_BSIZE; + break; + + case kw_name: + ++parameter; + bcopy(token[parameter], + sd->name, + min(sizeof(sd->name), strlen(token[parameter]))); + break; + + case kw_len: + size = sizespec(token[++parameter]); + if ((size % DEV_BSIZE) != 0) + throw_rude_remark(EINVAL, "sd %s, length %d not multiple of sector size", sd->name, size); + else + sd->sectors = size / DEV_BSIZE; + break; + + case kw_drive: + sd->driveno = find_drive(token[++parameter], 1); /* insert drive information */ + break; + + case kw_plex: + sd->plexno = find_plex(token[++parameter], 1); /* insert plex information */ + break; + + case kw_state: + checkkernel(token[++parameter]); /* must be a kernel user */ + sd->state = SdState(token[parameter]); /* set the state */ + break; + + default: + throw_rude_remark(EINVAL, "sd %s, invalid keyword: %s", sd->name, token[parameter]); + } + } + + /* Check we have a drive name */ + if (sd->driveno < 0) { /* didn't specify a drive */ + sd->driveno = current_drive; /* set to the current drive */ + if (sd->driveno < 0) /* no current drive? */ + throw_rude_remark(EINVAL, "Subdisk %s is not associated with a drive", sd->name); + } + /* Check for a plex name */ + if ((sd->plexno < 0) /* didn't specify a plex */ + &&(!detached)) /* and didn't say not to, */ + sd->plexno = current_plex; /* set to the current plex */ + + if (sd->plexno >= 0) + sdindex = give_sd_to_plex(sd->plexno, sdno); /* now tell the plex that it has this sd */ + + sd->sdno = sdno; /* point to our entry in the table */ + + /* Does the subdisk have a name? If not, give it one */ + if (sd->name[0] == '\0') { /* no name */ + char sdsuffix[8]; /* form sd name suffix here */ + + /* Do we have a plex name? */ + if (sdindex >= 0) /* we have a plex */ + strcpy(sd->name, PLEX[sd->plexno].name); /* take it from there */ + else /* no way */ + throw_rude_remark(EINVAL, "Unnamed sd is not associated with a plex"); + sprintf(sdsuffix, ".s%d", sdindex); /* form the suffix */ + strcat(sd->name, sdsuffix); /* and add it to the name */ + } + /* do we have complete info for this subdisk? */ + if (sd->sectors == 0) + throw_rude_remark(EINVAL, "sd %s has no length spec", sd->name); + + if (sd->state == sd_unallocated) /* no state decided, */ + sd->state = sd_init; /* at least we're in the game */ + + /* register the subdisk with the drive. This action + * will have the side effect of setting the offset if + * we haven't specified one, and causing an error + * message if it overlaps with another subdisk. */ + give_sd_to_drive(sdno); +} + +/* Handle a plex definition. + * If we find an error, print a message, deallocate the nascent plex, and return + */ +void +config_plex(void) +{ + BROKEN_GDB; + int parameter; + int plexno; /* index of plex in vinum_conf */ + struct plex *plex; /* and pointer to it */ + int pindex = MAXPLEX; /* index in volume's plex list */ + int detached = 0; /* don't give it to a volume */ + + current_plex = -1; /* forget the previous plex */ + plexno = get_empty_plex(); /* allocate a plex */ + plex = &PLEX[plexno]; /* and point to it */ + plex->plexno = plexno; /* and back to the config */ + for (parameter = 1; parameter < tokens; parameter++) { /* look at the other tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_detached: + detached = 1; + break; + + case kw_name: + { + int namedplexno; + + namedplexno = find_plex(token[++parameter], 0); /* find an existing plex with this name */ + if (namedplexno >= 0) + throw_rude_remark(EINVAL, "Duplicate plex %s", token[parameter]); + } + bcopy(token[parameter], /* put in the name */ + plex->name, + min(MAXPLEXNAME, strlen(token[parameter]))); + break; + + case kw_org: /* plex organization */ + switch (get_keyword(token[++parameter], &keyword_set)) { + case kw_concat: + plex->organization = plex_concat; + break; + + case kw_striped: + { + int stripesize = sizespec(token[++parameter]); + + plex->organization = plex_striped; + if (stripesize % DEV_BSIZE != 0) /* not a multiple of block size, */ + throw_rude_remark(EINVAL, "plex %s: stripe size %d not a multiple of sector size", + plex->name, + stripesize); + else + plex->stripesize = stripesize / DEV_BSIZE; + break; + } + + + default: + throw_rude_remark(EINVAL, "Invalid plex organization"); + } + if (((plex->organization == plex_striped) + ) + && (plex->stripesize == 0)) /* didn't specify a valid stripe size */ + throw_rude_remark(EINVAL, "Need a stripe size parameter"); + break; + + case kw_volume: + plex->volno = find_volume(token[++parameter], 1); /* insert a pointer to the volume */ + break; + + case kw_sd: /* add a subdisk */ + { + int sdno; + + sdno = find_subdisk(token[++parameter], 1); /* find a subdisk */ + SD[sdno].plexoffset = sizespec(token[++parameter]); /* get the offset */ + give_sd_to_plex(plexno, sdno); /* and insert it there */ + break; + } + + case kw_state: + checkkernel(token[++parameter]); /* only for kernel use */ + plex->state = PlexState(token[parameter]); /* set the state */ + break; + + default: + throw_rude_remark(EINVAL, "plex %s, invalid keyword: %s", + plex->name, + token[parameter]); + } + } + + if ((plex->volno < 0) /* we don't have a volume */ + &&(!detached)) /* and we wouldn't object */ + plex->volno = current_volume; + + if (plex->volno >= 0) + pindex = give_plex_to_volume(plex->volno, plexno); /* Now tell the volume that it has this plex */ + + /* Does the plex have a name? If not, give it one */ + if (plex->name[0] == '\0') { /* no name */ + char plexsuffix[8]; /* form plex name suffix here */ + /* Do we have a volume name? */ + if (plex->volno >= 0) /* we have a volume */ + strcpy(plex->name, /* take it from there */ + VOL[plex->volno].name); + else /* no way */ + throw_rude_remark(EINVAL, "Unnamed plex is not associated with a volume"); + sprintf(plexsuffix, ".p%d", pindex); /* form the suffix */ + strcat(plex->name, plexsuffix); /* and add it to the name */ + } + /* Note the last plex we configured */ + current_plex = plexno; + if (plex->state == plex_unallocated) /* we haven't changed the state, */ + plex->state = plex_init; /* we're initialized now */ +} + +/* Handle a volume definition. + * If we find an error, print a message, deallocate the nascent volume, and return + */ +void +config_volume(void) +{ + BROKEN_GDB; + int parameter; + int volno; + struct volume *vol; /* collect volume info here */ + int i; + + if (tokens < 2) /* not enough tokens */ + throw_rude_remark(EINVAL, "Volume has no name"); + current_volume = -1; /* forget the previous volume */ + volno = find_volume(token[1], 1); /* allocate a volume to initialize */ + vol = &VOL[volno]; /* and get a pointer */ + + for (parameter = 2; parameter < tokens; parameter++) { /* look at all tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_plex: + { + int plexno; /* index of this plex */ + + plexno = find_plex(token[++parameter], 1); /* find a plex */ + if (plexno < 0) /* couldn't */ + break; /* we've already had an error message */ + plexno = my_plex(volno, plexno); /* does it already belong to us? */ + if (plexno > 0) /* yes, shouldn't get it again */ + throw_rude_remark(EINVAL, + "Plex %s already belongs to volume %s", + token[parameter], + vol->name); + else if (++vol->plexes > 8) /* another entry */ + throw_rude_remark(EINVAL, + "Too many plexes for volume %s", + vol->name); + vol->plex[vol->plexes - 1] = plexno; + } + break; + + case kw_readpol: + switch (get_keyword(token[++parameter], &keyword_set)) { /* decide what to do */ + case kw_round: + vol->preferred_plex = ROUND_ROBIN_READPOL; /* default */ + break; + + case kw_prefer: + { + int myplexno; /* index of this plex */ + + myplexno = find_plex(token[++parameter], 1); /* find a plex */ + if (myplexno < 0) /* couldn't */ + break; /* we've already had an error message */ + myplexno = my_plex(volno, myplexno); /* does it already belong to us? */ + if (myplexno > 0) /* yes */ + vol->preferred_plex = myplexno; /* just note the index */ + else if (++vol->plexes > 8) /* another entry */ + throw_rude_remark(EINVAL, "Too many plexes"); + else { /* space for the new plex */ + vol->plex[vol->plexes - 1] = myplexno; /* add it to our list */ + vol->preferred_plex = vol->plexes - 1; /* and note the index */ + } + } + break; + + default: + throw_rude_remark(EINVAL, "Invalid read policy"); + } + + case kw_setupstate: + vol->flags |= VF_CONFIG_SETUPSTATE; /* set the volume up later on */ + break; + + case kw_state: + checkkernel(token[++parameter]); /* must be a kernel user */ + vol->state = VolState(token[parameter]); /* set the state */ + break; + + /* XXX experimental ideas. These are not + * documented, and will not be until I + * decide they're worth keeping */ + case kw_writethrough: /* set writethrough mode */ + vol->flags |= VF_WRITETHROUGH; + break; + + case kw_writeback: /* set writeback mode */ + vol->flags &= ~VF_WRITETHROUGH; + break; + + case kw_raw: + vol->flags |= VF_RAW; /* raw volume (no label) */ + break; + + default: + throw_rude_remark(EINVAL, "volume %s, invalid keyword: %s", + vol->name, + token[parameter]); + } + } + + current_volume = volno; /* note last referred volume */ + vol->devno = VINUMBDEV(volno, 0, 0, VINUM_VOLUME_TYPE); /* also note device number */ + + /* Before we can actually use the volume, we need + * a volume label. We could start to fake one here, + * but it will be a lot easier when we have some + * to copy from the drives, so defer it until we + * set up the configuration. XXX */ + if (vol->state == volume_unallocated) + vol->state = volume_down; /* now ready to bring up at the end */ + + /* Find out how big our volume is */ + for (i = 0; i < vol->plexes; i++) + vol->size = max(vol->size, PLEX[vol->plex[i]].length); +} + +/* Parse a config entry. CARE! This destroys the original contents of the + * config entry, which we don't really need after this. More specifically, it + * places \0 characters at the end of each token. + * + * Return 0 if all is well, otherwise EINVAL */ +int +parse_config(char *cptr, struct keywordset *keyset) +{ + BROKEN_GDB; + int status; + + status = 0; /* until proven otherwise */ + tokens = tokenize(cptr, token); /* chop up into tokens */ + + if (tokens <= 0) /* screwed up or empty line */ + return tokens; /* give up */ + + if (token[0][0] == '#') /* comment line */ + return 0; + + switch (get_keyword(token[0], keyset)) { /* decide what to do */ + case kw_read: /* read config from a specified drive */ + vinum_conf.flags |= VF_KERNELOP | VF_READING_CONFIG; /* kernel operation: reading config */ + status = check_drive(token[1]); /* check the drive info */ + vinum_conf.flags &= ~(VF_KERNELOP | VF_READING_CONFIG); + if (status != 0) { + char *msg = "Can't read configuration from %s"; + if (status == ENODEV) + msg = "No vinum configuration on %s"; + throw_rude_remark(status, msg, token[1]); + } + updateconfig(VF_KERNELOP); /* update from kernel space */ + break; + + case kw_drive: + config_drive(); + break; + + case kw_subdisk: + config_subdisk(); + break; + + case kw_plex: + config_plex(); + break; + + case kw_volume: + config_volume(); + break; + + /* Anything else is invalid in this context */ + default: + throw_rude_remark(EINVAL, /* should we die? */ + "Invalid configuration information: %s", + token[0]); + } + return status; +} + +/* parse a line handed in from userland via ioctl. + * This differs only by the error reporting mechanism: + * we return the error indication in the reply to the + * ioctl, so we need to set a global static pointer in + * this file. This technique works because we have + * ensured that configuration is performed in a single- + * threaded manner */ +int +parse_user_config(char *cptr, struct keywordset *keyset) +{ + BROKEN_GDB; + int status; + + ioctl_reply = (struct _ioctl_reply *) cptr; + status = parse_config(cptr, keyset); + ioctl_reply = NULL; /* don't do this again */ + return status; +} + +/* Remove an object */ +void +remove(struct vinum_ioctl_msg *msg) +{ + struct vinum_ioctl_msg message = *msg; /* make a copy to hand on */ + + ioctl_reply = (struct _ioctl_reply *) msg; /* reinstate the address to reply to */ + ioctl_reply->error = 0; /* no error, */ + ioctl_reply->msg[0] = '\0'; /* no message */ + + switch (message.type) { + case drive_object: + remove_drive_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + case sd_object: + remove_sd_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + case plex_object: + remove_plex_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + case volume_object: + remove_volume_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + default: + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "Invalid object type"); + } +} + +/* Remove a drive. */ +void +remove_drive_entry(int driveno, int force, int recurse) +{ + struct drive *drive = &DRIVE[driveno]; + + if ((driveno > vinum_conf.drives_used) /* not a valid drive */ + ||(drive->state == drive_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such drive"); + } else if (drive->opencount > 0) { /* we have subdisks */ + if (force) { /* do it at any cost */ + int sdno; + struct vinum_ioctl_msg sdmsg; + + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) { + if ((SD[sdno].state != sd_unallocated) /* subdisk is allocated */ + &&(SD[sdno].driveno == driveno)) { /* and it belongs to this drive */ + sdmsg.type = sd_object; + sdmsg.recurse = 1; + sdmsg.force = force; + remove(&sdmsg); /* remove the subdisk by force */ + } + } + remove_drive(driveno); /* now remove it */ + } else + ioctl_reply->error = EBUSY; /* can't do that */ + } else + remove_drive(driveno); /* just remove it */ +} + +/* remove a subdisk */ +void +remove_sd_entry(int sdno, int force, int recurse) +{ + struct sd *sd = &SD[sdno]; + + if ((sdno > vinum_conf.subdisks_used) /* not a valid sd */ + ||(sd->state == sd_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such subdisk"); + } else if (sd->plexno >= 0) { /* we have a plex */ + if (force) { /* do it at any cost */ + struct plex *plex = &PLEX[sd->plexno]; /* point to our plex */ + int mysdno; + + for (mysdno = 0; /* look for ourselves */ + mysdno < plex->subdisks && &SD[plex->sdnos[mysdno]] != sd; + mysdno++); + if (mysdno == plex->subdisks) /* didn't find it */ + throw_rude_remark(ENOENT, "plex %s does not contain subdisk %s", plex->name, sd->name); + if (mysdno < (plex->subdisks - 1)) /* not the last subdisk */ + bcopy(&plex->sdnos[mysdno + 1], + &plex->sdnos[mysdno], + (plex->subdisks - 1 - mysdno) * sizeof(int)); + plex->subdisks--; + /* removing a subdisk from a striped or + * RAID-5 plex really tears the hell out + * of the structure, and it needs to be + * reinitialized */ + if (plex->organization != plex_concat) /* not concatenated, */ + set_plex_state(plex->plexno, plex_faulty, setstate_force); /* need to reinitialize */ + rebuild_plex_unmappedlist(plex); /* and see what remains */ + free_sd(sdno); + } else + ioctl_reply->error = EBUSY; /* can't do that */ + } else + free_sd(sdno); +} + +/* remove a plex */ +void +remove_plex_entry(int plexno, int force, int recurse) +{ + struct plex *plex = &PLEX[plexno]; + int sdno; + + if ((plexno > vinum_conf.plexes_used) /* not a valid plex */ + ||(plex->state == plex_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such plex"); + } else if (plex->pid) { /* we're open */ + ioctl_reply->error = EBUSY; /* no getting around that */ + return; + } + if (plex->subdisks) { + if (force) { /* do it anyway */ + if (recurse) { /* remove all below */ + for (sdno = 0; sdno < plex->subdisks; sdno++) + free_sd(plex->sdnos[sdno]); /* free all subdisks */ + } else { /* just tear them out */ + for (sdno = 0; sdno < plex->subdisks; sdno++) + SD[plex->sdnos[sdno]].plexno = -1; /* no plex any more */ + } + } else { /* can't do it without force */ + ioctl_reply->error = EBUSY; /* can't do that */ + return; + } + } + if (plex->volno >= 0) { /* we are part of a volume */ + /* XXX This should be more intelligent. We should + * be able to remove a plex as long as the volume + * does not lose any data, which is normally the + * case when it has more than one plex. To do it + * right we must compare the completeness of the + * mapping of all the plexes in the volume */ + if (force) { /* do it at any cost */ + struct volume *vol = &VOL[plex->volno]; + int myplexno; + + for (myplexno = 0; myplexno < vol->plexes; myplexno++) + if (vol->plex[myplexno] == plexno) /* found it */ + break; + if (myplexno == vol->plexes) /* didn't find it. Huh? */ + throw_rude_remark(ENOENT, "volume %s does not contain plex %s", vol->name, plex->name); + if (myplexno < (vol->plexes - 1)) /* not the last plex in the list */ + bcopy(&vol->plex[myplexno + 1], &vol->plex[myplexno], vol->plexes - 1 - myplexno); + vol->plexes--; + } else { + ioctl_reply->error = EBUSY; /* can't do that */ + return; + } + } + free_plex(plexno); +} + +/* remove a volume */ +void +remove_volume_entry(int volno, int force, int recurse) +{ + struct volume *vol = &VOL[volno]; + int plexno; + + if ((volno > vinum_conf.volumes_used) /* not a valid volume */ + ||(vol->state == volume_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such volume"); + } else if (vol->opencount) /* we're open */ + ioctl_reply->error = EBUSY; /* no getting around that */ + else if (vol->plexes) { + if (recurse && force) { /* remove all below */ + struct vinum_ioctl_msg plexmsg; + + plexmsg.type = plex_object; + plexmsg.recurse = 1; + plexmsg.force = force; + for (plexno = 0; plexno < vol->plexes; plexno++) { + plexmsg.index = vol->plex[plexno]; /* plex number */ + remove(&plexmsg); + } + free_volume(volno); + } else + ioctl_reply->error = EBUSY; /* can't do that */ + } else + free_volume(volno); +} + +void +update_sd_config(int sdno, int kernelstate) +{ + if (!kernelstate) + set_sd_state(sdno, sd_up, setstate_configuring | setstate_norecurse); +} + +void +update_plex_config(int plexno, int kernelstate) +{ + int error = 0; + int size; + int sdno; + struct plex *plex = &PLEX[plexno]; + enum plexstate state = plex_up; /* state we want the plex in */ + + /* XXX Insert checks here for sparse plexes and volumes */ + + /* Check that our subdisks make sense. For + * striped and RAID5 plexes, we need at least + * two subdisks, and they must all be the same + * size */ + if (((plex->organization == plex_striped) + ) + && (plex->subdisks < 2)) { + error = 1; + printf("vinum: plex %s does not have at least 2 subdisks\n", plex->name); + if (!kernelstate) + set_plex_state(plexno, plex_down, setstate_force | setstate_configuring | setstate_norecurse); + } + size = 0; + for (sdno = 0; sdno < plex->subdisks; sdno++) { + if (((plex->organization == plex_striped) + ) + && (sdno > 0) + && (SD[plex->sdnos[sdno]].sectors != SD[plex->sdnos[sdno - 1]].sectors)) { + error = 1; + printf("vinum: plex %s must have equal sized subdisks\n", plex->name); + set_plex_state(plexno, plex_down, setstate_force | setstate_configuring | setstate_norecurse); + } + size += SD[plex->sdnos[sdno]].sectors; + } + + if (plex->subdisks) { /* plex has subdisks, calculate size */ + rebuild_plex_unmappedlist(plex); /* rebuild the unmapped list first */ + + plex->length = size; + } else { /* no subdisks, */ + plex->length = 0; /* no size */ + state = plex_down; /* take it down */ + } + if (!(kernelstate || error)) + set_plex_state(plexno, state, setstate_none | setstate_configuring | setstate_norecurse); +} + +void +update_volume_config(int volno, int kernelstate) +{ + struct volume *vol = &VOL[volno]; + struct plex *plex; + int plexno; + + if (vol->state != volume_unallocated) + /* Recalculate the size of the volume */ + { + vol->size = 0; + for (plexno = 0; plexno < vol->plexes; plexno++) { + plex = &PLEX[vol->plex[plexno]]; + vol->size = max(plex->length, vol->size); /* maximum size */ + plex->volplexno = plexno; /* note it in the plex */ + } + } + if (!kernelstate) /* try to bring it up */ + set_volume_state(volno, volume_up, setstate_configuring | setstate_norecurse); +} + +/* Update the global configuration. + * kernelstate is != 0 if we're reading in a config + * from disk. In this case, we don't try to + * bring the devices up, though we will bring + * them down if there's some error which got + * missed when writing to disk. + */ +void +updateconfig(int kernelstate) +{ + BROKEN_GDB; + int sdno; + int plexno; + int volno; + struct volume *vol; + struct plex *plex; + + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) + update_sd_config(sdno, kernelstate); + + for (plexno = 0; plexno < vinum_conf.plexes_used; plexno++) + update_plex_config(plexno, kernelstate); + + for (volno = 0; volno < vinum_conf.volumes_used; volno++) + update_volume_config(volno, kernelstate); + save_config(); +} + +/* Start manual changes to the configuration and lock out + * others who may wish to do so. + * XXX why do we need this and lock_config too? */ +int +start_config(void) +{ + int error; + + while ((vinum_conf.flags & VF_CONFIGURING) != 0) { + vinum_conf.flags |= VF_WILL_CONFIGURE; + if ((error = tsleep(&vinum_conf, PRIBIO | PCATCH, "vincfg", 0)) != 0) + return error; + } + /* We need two flags here: VF_CONFIGURING + * tells other processes to hold off (this + * function), and VF_CONFIG_INCOMPLETE + * tells the state change routines not to + * propagate incrememntal state changes */ + vinum_conf.flags |= VF_CONFIGURING | VF_CONFIG_INCOMPLETE; + current_drive = -1; /* reset the defaults */ + current_plex = -1; /* and the same for the last plex */ + current_volume = -1; /* and the last volme */ + return 0; +} + +/* Update the config if update is 1, and unlock + * it. We won't update the configuration if we + * are called in a recursive loop via throw_rude_remark. + */ +void +finish_config(int update) +{ + vinum_conf.flags &= ~VF_CONFIG_INCOMPLETE; /* we've finished our config */ + if (update) + updateconfig(0); /* so update things */ + else + updateconfig(1); /* do some updates only */ + vinum_conf.flags &= ~VF_CONFIGURING; /* and now other people can take a turn */ + if ((vinum_conf.flags & VF_WILL_CONFIGURE) != 0) { + vinum_conf.flags &= ~VF_WILL_CONFIGURE; + wakeup(&vinum_conf); + } +} diff --git a/lkm/vinum/interrupt.c b/lkm/vinum/interrupt.c new file mode 100644 index 0000000..e7eb034 --- /dev/null +++ b/lkm/vinum/interrupt.c @@ -0,0 +1,190 @@ +/* interrupt.c: bottom half of the driver */ + +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: interrupt.c,v 1.1 1998/08/13 06:12:27 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" +#include <miscfs/specfs/specdev.h> +#include <sys/resourcevar.h> + +void complete_raid5_write(struct rqelement *); +void freerq(struct request *rq); +void free_rqg(struct rqgroup *rqg); +void complete_rqe(struct buf *bp); +void sdio_done(struct buf *bp); + +/* Take a completed buffer, transfer the data back if + * it's a read, and complete the high-level request + * if this is the last subrequest. + * + * The bp parameter is in fact a struct rqelement, which + * includes a couple of extras at the end. + */ +void +complete_rqe(struct buf *bp) +{ + BROKEN_GDB; + struct rqelement *rqe; + struct request *rq; + struct rqgroup *rqg; + struct buf *ubp; /* user buffer */ + + rqe = (struct rqelement *) bp; /* point to the element element that completed */ + rqg = rqe->rqg; /* and the request group */ + rq = rqg->rq; /* and the complete request */ + + if ((bp->b_flags & B_ERROR) != 0) { /* transfer in error */ + if (bp->b_error != 0) /* did it return a number? */ + rq->error = bp->b_error; /* yes, put it in. */ + else if (rq->error == 0) /* no: do we have one already? */ + rq->error = EIO; /* no: catchall "I/O error" */ + if (rq->error == EIO) /* I/O error, */ + set_sd_state(rqe->sdno, sd_crashed, setstate_force); /* take the subdisk down */ + } + /* Now update the statistics */ + if (bp->b_flags & B_READ) { /* read operation */ + DRIVE[rqe->driveno].reads++; + DRIVE[rqe->driveno].bytes_read += bp->b_bcount; + SD[rqe->sdno].reads++; + SD[rqe->sdno].bytes_read += bp->b_bcount; + PLEX[rqe->rqg->plexno].reads++; + PLEX[rqe->rqg->plexno].bytes_read += bp->b_bcount; + } else { /* write operation */ + DRIVE[rqe->driveno].writes++; + DRIVE[rqe->driveno].bytes_written += bp->b_bcount; + SD[rqe->sdno].writes++; + SD[rqe->sdno].bytes_written += bp->b_bcount; + PLEX[rqe->rqg->plexno].writes++; + PLEX[rqe->rqg->plexno].bytes_written += bp->b_bcount; + } + ubp = rq->bp; /* user buffer */ + rqg->active--; /* one less request active */ + if (rqg->active == 0) /* request group finished, */ + rq->active--; /* one less */ + if (rq->active == 0) { /* request finished, */ +#if DEBUG + if (debug & 4) { + if (ubp->b_resid != 0) /* still something to transfer? */ + Debugger("resid"); + + { + int i; + for (i = 0; i < ubp->b_bcount; i += 512) /* XXX debug */ + if (((char *) ubp->b_data)[i] != '<') { /* and not what we expected */ + printf("At 0x%x (offset 0x%x): '%c' (0x%x)\n", + (int) (&((char *) ubp->b_data)[i]), + i, + ((char *) ubp->b_data)[i], + ((char *) ubp->b_data)[i]); + Debugger("complete_request checksum"); + } + } + } +#endif + + if (rq->error) { /* did we have an error? */ + ubp->b_flags |= B_ERROR; /* yes, propagate to user */ + ubp->b_error = rq->error; + } else + ubp->b_resid = 0; /* completed our transfer */ + if (rq->isplex == 0) /* volume request, */ + VOL[rq->volplex.volno].active--; /* another request finished */ + biodone(ubp); /* top level buffer completed */ + freerq(rq); /* return the request storage */ + } +} + + +/* Free a request block and anything hanging off it */ +void +freerq(struct request *rq) +{ + BROKEN_GDB; + struct rqgroup *rqg; + struct rqgroup *nrqg; /* next in chain */ + int rqno; + + for (rqg = rq->rqg; rqg != NULL; rqg = nrqg) { /* through the whole request chain */ + for (rqno = 0; rqno < rqg->count; rqno++) + if ((rqg->rqe[rqno].flags & XFR_MALLOCED) /* data buffer was malloced, */ + &&rqg->rqe[rqno].b.b_data) /* and the allocation succeeded */ + Free(rqg->rqe[rqno].b.b_data); /* free it */ + nrqg = rqg->next; /* note the next one */ + Free(rqg); /* and free this one */ + } + Free(rq); /* free the request itself */ +} + +void +free_rqg(struct rqgroup *rqg) +{ + if ((rqg->flags & XFR_GROUPOP) /* RAID 5 request */ + &&(rqg->rqe) /* got a buffer structure */ + &&(rqg->rqe->b.b_data)) /* and it has a buffer allocated */ + Free(rqg->rqe->b.b_data); /* free it */ +} + +/* I/O on subdisk completed */ +void +sdio_done(struct buf *bp) +{ + struct sdbuf *sbp; + + sbp = (struct sdbuf *) bp; + if (sbp->b.b_flags & B_ERROR) { /* had an error */ + bp->b_flags |= B_ERROR; + bp->b_error = sbp->b.b_error; + } + bp->b_resid = sbp->b.b_resid; + biodone(sbp->bp); /* complete the caller's I/O */ + /* Now update the statistics */ + if (bp->b_flags & B_READ) { /* read operation */ + DRIVE[sbp->driveno].reads++; + DRIVE[sbp->driveno].bytes_read += bp->b_bcount; + SD[sbp->sdno].reads++; + SD[sbp->sdno].bytes_read += bp->b_bcount; + } else { /* write operation */ + DRIVE[sbp->driveno].writes++; + DRIVE[sbp->driveno].bytes_written += bp->b_bcount; + SD[sbp->sdno].writes++; + SD[sbp->sdno].bytes_written += bp->b_bcount; + } + Free(sbp); +} diff --git a/lkm/vinum/io.c b/lkm/vinum/io.c new file mode 100644 index 0000000..a16a33a --- /dev/null +++ b/lkm/vinum/io.c @@ -0,0 +1,886 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: io.c,v 1.16 1998/08/10 23:47:21 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#if __FreeBSD__ < 3 /* this is in sys/disklabel.h in 3.0 and on */ +#define DTYPE_VINUM 12 /* vinum volume */ +#endif + +#define REALLYKERNEL +#include "vinumhdr.h" +#include <miscfs/specfs/specdev.h> + +extern jmp_buf command_fail; /* return on a failed command */ +struct _ioctl_reply *ioctl_reply; /* data pointer, for returning error messages */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +/* pointer to ioctl p parameter, to save passing it around */ +extern struct proc *myproc; + +/* Open the device associated with the drive, and set drive's vp */ +int +open_drive(struct drive *drive, struct proc *p) +{ + BROKEN_GDB; + struct nameidata nd; + struct vattr va; + int error; + + if (drive->devicename[0] == '\0') /* no device name */ + sprintf(drive->devicename, "/dev/%s", drive->label.name); /* get it from the drive name */ + NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, drive->devicename, p); + error = vn_open(&nd, FREAD | FWRITE, 0); /* open the device */ + if (error != 0) { /* can't open? */ + set_drive_state(drive->driveno, drive_down, 1); + drive->lasterror = error; + printf("vinum open_drive %s: failed with error %d\n", drive->devicename, error); /* XXX */ + return error; + } + drive->vp = nd.ni_vp; + drive->p = p; + + if (drive->vp->v_usecount > 1) { /* already in use? */ +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); + drive->lasterror = EBUSY; + printf("vinum open_drive %s: Drive in use\n", drive->devicename); /* XXX */ + return EBUSY; + } + error = VOP_GETATTR(drive->vp, &va, NOCRED, p); + if (error) { +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); + drive->lasterror = error; + printf("vinum open_drive %s: GETAATTR returns error %d\n", drive->devicename, error); /* XXX */ + return error; + } + drive->dev = va.va_rdev; /* device */ + + if (va.va_type != VBLK) { /* only consider block devices */ +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); /* this also closes the drive */ + drive->lasterror = ENOTBLK; + printf("vinum open_drive %s: Not a block device\n", drive->devicename); /* XXX */ + return ENOTBLK; + } + drive->vp->v_numoutput = 0; +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + return 0; +} + +/* Set some variables in the drive struct + * in more convenient form. Return error indication */ +int +set_drive_parms(struct drive *drive) +{ + drive->blocksize = BLKDEV_IOSIZE; /* XXX do we need this? */ + drive->secsperblock = drive->blocksize /* number of sectors per block */ + / drive->partinfo.disklab->d_secsize; + + /* Now update the label part */ + bcopy(hostname, drive->label.sysname, VINUMHOSTNAMELEN); /* put in host name */ +#if __FreeBSD__ >= 3 + getmicrotime(&drive->label.date_of_birth); /* and current time */ +#else + drive->label.date_of_birth = time; /* and current time */ +#endif + drive->label.drive_size = ((u_int64_t) drive->partinfo.part->p_size) /* size of the drive in bytes */ + *((u_int64_t) drive->partinfo.disklab->d_secsize); + + /* number of sectors available for subdisks */ + drive->sectors_available = drive->label.drive_size / DEV_BSIZE - DATASTART; + + /* XXX Bug in 3.0 as of January 1998: you can open + * non-existent slices. They have a length of 0 */ + if (drive->label.drive_size < MINVINUMSLICE) { /* too small to worry about */ + set_drive_state(drive->driveno, drive_down, 1); + printf("vinum open_drive %s: Drive too small\n", drive->devicename); /* XXX */ + drive->lasterror = ENOSPC; + return ENOSPC; + } + drive->freelist_size = INITIAL_DRIVE_FREELIST; /* initial number of entries */ + drive->freelist = (struct drive_freelist *) + Malloc(INITIAL_DRIVE_FREELIST * sizeof(struct drive_freelist)); + if (drive->freelist == NULL) /* can't malloc, dammit */ + return ENOSPC; + drive->freelist_entries = 1; /* just (almost) the complete drive */ + drive->freelist[0].offset = DATASTART; /* starts here */ + drive->freelist[0].sectors = (drive->label.drive_size >> DEV_BSHIFT) - DATASTART; /* and it's this long */ + set_drive_state(drive->driveno, drive_up, 1); /* our drive is accessible */ + return 0; +} + +/* Initialize a drive: open the device and add device + * information */ +int +init_drive(struct drive *drive) +{ + BROKEN_GDB; + int error; + + if (drive->devicename[0] == '\0') { /* no device name yet, default to drive name */ + drive->lasterror = EINVAL; + printf("vinum: Can't open drive without drive name\n"); /* XXX */ + return EINVAL; + } + error = open_drive(drive, myproc); /* open the drive */ + if (error) + return error; + + error = VOP_IOCTL(drive->vp, /* get the partition information */ + DIOCGPART, + (caddr_t) & drive->partinfo, + FREAD, + NOCRED, + myproc); + if (error) { + printf("vinum open_drive %s: Can't get partition information, error %d\n", + drive->devicename, + error); /* XXX */ + close_drive(drive); + drive->lasterror = error; + set_drive_state(drive->driveno, drive_down, 1); + return error; + } + if (drive->partinfo.part->p_fstype != 0) { /* not plain */ + drive->lasterror = EFTYPE; + printf("vinum open_drive %s: Wrong partition type for vinum\n", drive->devicename); /* XXX */ + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); + return EFTYPE; + } + return set_drive_parms(drive); /* set various odds and ends */ +} + +/* Close a drive if it's open. No errors */ +void +close_drive(struct drive *drive) +{ + if (drive->vp) { + vn_close(drive->vp, FREAD | FWRITE, NOCRED, drive->p); + drive->vp = NULL; + } +} + +/* Remove drive from the configuration. + * Caller must ensure that it isn't active + */ +void +remove_drive(int driveno) +{ + BROKEN_GDB; + struct drive *drive = &vinum_conf.drive[driveno]; + long long int nomagic = VINUM_NOMAGIC; /* no magic number */ + + write_drive(drive, /* obliterate the magic, but leave a hint */ + (char *) &nomagic, + 8, + VINUM_LABEL_OFFSET); + close_drive(drive); /* and close it */ + drive->state = drive_unallocated; /* and forget everything we knew about it */ + save_config(); /* and save the updated configuration */ +} + +/* Transfer drive data. Usually called from one of these defines; + + * #define read_drive(a, b, c, d) driveio (a, b, c, d, B_READ) + * #define write_drive(a, b, c, d) driveio (a, b, c, d, B_WRITE) + * + * Return error number + */ +int +driveio(struct drive *drive, void *buf, size_t length, off_t offset, int flag) +{ + BROKEN_GDB; + int error; + struct buf *bp; + int spl; + + error = 0; + + /* Get a buffer */ + bp = (struct buf *) Malloc(sizeof(struct buf)); /* get a buffer */ + CHECKALLOC(bp, "Can't allocate memory"); + + bzero(&buf, sizeof(buf)); + bp->b_flags = B_BUSY | flag; /* tell us when it's done */ + bp->b_iodone = drive_io_done; /* here */ + bp->b_proc = myproc; /* process */ + bp->b_dev = drive->vp->v_un.vu_specinfo->si_rdev; /* device */ + if (offset & (drive->partinfo.disklab->d_secsize - 1)) /* not on a block boundary */ + bp->b_blkno = offset / drive->partinfo.disklab->d_secsize; /* block number */ + bp->b_data = buf; + bp->b_vp = drive->vp; /* vnode */ + bp->b_bcount = length; + bp->b_bufsize = length; + + (*bdevsw[major(bp->b_dev)]->d_strategy) (bp); /* initiate the transfer */ + + spl = splbio(); + while ((bp->b_flags & B_DONE) == 0) { + bp->b_flags |= B_CALL; /* wake me again */ + tsleep((caddr_t) bp, PRIBIO, "driveio", 0); /* and wait for it to complete */ + } + splx(spl); + if (bp->b_flags & B_ERROR) /* didn't work */ + error = bp->b_error; /* get the error return */ + Free(bp); /* then return the buffer */ + return error; +} + +/* Read data from a drive + + * Return error number + */ +int +read_drive(struct drive *drive, void *buf, size_t length, off_t offset) +{ + BROKEN_GDB; + int error; + struct buf *bp; + daddr_t nextbn; + long bscale; + + struct uio uio; + struct iovec iov; + daddr_t blocknum; /* block number */ + int blockoff; /* offset in block */ + int count; /* amount to transfer */ + + iov.iov_base = buf; + iov.iov_len = length; + + uio.uio_iov = &iov; + uio.uio_iovcnt = length; + uio.uio_offset = offset; + uio.uio_resid = length; + uio.uio_segflg = UIO_SYSSPACE; + uio.uio_rw = UIO_READ; + uio.uio_procp = myproc; + + bscale = btodb(drive->blocksize); /* mask off offset from block number */ + do { + blocknum = btodb(uio.uio_offset) & ~(bscale - 1); /* get the block number */ + blockoff = uio.uio_offset % drive->blocksize; /* offset in block */ + count = min((unsigned) (drive->blocksize - blockoff), /* amount to transfer in this block */ + uio.uio_resid); + + /* XXX Check this. I think the test is wrong */ + if (drive->vp->v_lastr + bscale == blocknum) { /* did our last read finish in this block? */ + nextbn = blocknum + bscale; /* note the end of the transfer */ + error = breadn(drive->vp, /* and read with read-ahead */ + blocknum, + (int) drive->blocksize, + &nextbn, + (int *) &drive->blocksize, + 1, + NOCRED, + &bp); + } else /* random read: just read this block */ + error = bread(drive->vp, blocknum, (int) drive->blocksize, NOCRED, &bp); + drive->vp->v_lastr = blocknum; /* note the last block we read */ + count = min(count, drive->blocksize - bp->b_resid); + if (error) { + brelse(bp); + return error; + } + error = uiomove((char *) bp->b_data + blockoff, count, &uio); /* move the data */ + brelse(bp); + } + while (error == 0 && uio.uio_resid > 0 && count != 0); + return error; +} + +/* Write data to a drive + + * Return error number + */ +int +write_drive(struct drive *drive, void *buf, size_t length, off_t offset) +{ + BROKEN_GDB; + int error; + struct buf *bp; + struct uio uio; + struct iovec iov; + daddr_t blocknum; /* block number */ + int blockoff; /* offset in block */ + int count; /* amount to transfer */ + int blockshift; + + if (drive->state == drive_down) /* currently down */ + return 0; /* ignore */ + if (drive->vp == NULL) { + drive->lasterror = ENODEV; + return ENODEV; /* not configured yet */ + } + iov.iov_base = buf; + iov.iov_len = length; + + uio.uio_iov = &iov; + uio.uio_iovcnt = length; + uio.uio_offset = offset; + uio.uio_resid = length; + uio.uio_segflg = UIO_SYSSPACE; + uio.uio_rw = UIO_WRITE; + uio.uio_procp = myproc; + + error = 0; + blockshift = btodb(drive->blocksize) - 1; /* amount to shift block number + * to get sector number */ + do { + blocknum = btodb(uio.uio_offset) & ~blockshift; /* get the block number */ + blockoff = uio.uio_offset % drive->blocksize; /* offset in block */ + count = min((unsigned) (drive->blocksize - blockoff), /* amount to transfer in this block */ + uio.uio_resid); + if (count == drive->blocksize) /* the whole block */ + bp = getblk(drive->vp, blocknum, drive->blocksize, 0, 0); /* just get it */ + else /* partial block: */ + error = bread(drive->vp, /* read it first */ + blocknum, + drive->blocksize, + NOCRED, + &bp); + count = min(count, drive->blocksize - bp->b_resid); /* how much will we transfer now? */ + if (error == 0) + error = uiomove((char *) bp->b_data + blockoff, /* move the data to the block */ + count, + &uio); + if (error) { + brelse(bp); + drive->lasterror = error; + switch (error) { + case EIO: + set_drive_state(drive->driveno, drive_down, 1); + break; + + /* XXX Add other possibilities here */ + default: + } + return error; + } + if (count + blockoff == drive->blocksize) + /* The transfer goes to the end of the block. There's + * no need to wait for any more data to arrive. */ + bawrite(bp); /* start the write now */ + else + bdwrite(bp); /* do a delayed write */ + } + while (error == 0 && uio.uio_resid > 0 && count != 0); + if (error) + drive->lasterror = error; + return error; /* OK */ +} + +/* Wake up on completion */ +void +drive_io_done(struct buf *bp) +{ + BROKEN_GDB; + wakeup((caddr_t) bp); /* Wachet auf! */ + bp->b_flags &= ~B_CALL; /* don't do this again */ +} + +/* Check a drive for a vinum header. If found, + * update the drive information. We come here + * with a partially populated drive structure + * which includes the device name. + * + * Return information on what we found + */ +enum drive_label_info +read_drive_label(struct drive *drive) +{ + BROKEN_GDB; + int error; + int result; /* result of our search */ + struct vinum_hdr *vhdr; /* and as header */ + + error = init_drive(drive); /* find the drive */ + if (error) /* find the drive */ + return DL_CANT_OPEN; /* not ours */ + + vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* allocate buffers */ + CHECKALLOC(vhdr, "Can't allocate memory"); + + error = read_drive(drive, (void *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); + if (vhdr->magic == VINUM_MAGIC) { /* ours! */ + if (drive->label.name[0] /* we have a name for this drive */ + &&(strcmp(drive->label.name, vhdr->label.name))) { /* but it doesn't match the real name */ + drive->lasterror = EINVAL; + result = DL_WRONG_DRIVE; /* it's the wrong drive */ + } else { + set_drive_parms(drive); /* and set other parameters */ + result = DL_OURS; + } + /* We copy the drive anyway so that we have + * the correct name in the drive info. This + * may not be the name specified */ + drive->label = vhdr->label; /* put in the label information */ + } else if (vhdr->magic == VINUM_NOMAGIC) /* was ours, but we gave it away */ + result = DL_DELETED_LABEL; + else + result = DL_NOT_OURS; /* we could have it, but we don't yet */ + Free(vhdr); /* that's all. */ + return result; +} + +/* Check a drive for a vinum header. If found, + * read configuration information from the drive and + * incorporate the data into the configuration. + * + * Return error number + */ +int +check_drive(char *drivename) +{ + BROKEN_GDB; + int error; + struct nameidata nd; /* mount point credentials */ + char *config_text; /* read the config info from disk into here */ + volatile char *cptr; /* pointer into config information */ + char *eptr; /* end pointer into config information */ + int driveno; + struct drive *drive; + char *config_line; /* copy the config line to */ + + driveno = find_drive_by_dev(drivename, 1); /* doesn't exist, create it */ + drive = &vinum_conf.drive[driveno]; /* and get a pointer */ + strcpy(drive->devicename, drivename); /* put in device name */ + + if (read_drive_label(drive) == DL_OURS) { /* ours! */ + config_text = (char *) Malloc(MAXCONFIG * 2); /* allocate buffers */ + CHECKALLOC(config_text, "Can't allocate memory"); + config_line = (char *) Malloc(MAXCONFIGLINE * 2); /* allocate buffers */ + CHECKALLOC(config_line, "Can't allocate memory"); + + /* Read in both copies of the configuration information */ + error = read_drive(drive, config_text, MAXCONFIG * 2, VINUM_CONFIG_OFFSET); + + if (error != 0) { + printf("vinum: Can't read device %s, error %d\n", drive->devicename, error); + Free(config_text); + Free(config_line); + free_drive(drive); /* give it back */ + return error; + } + /* XXX At this point, check that the two copies are the same, and do something useful if not. + * In particular, consider which is newer, and what this means for the integrity of the + * data on the drive */ + + /* Parse the configuration, and add it to the global configuration */ + for (cptr = config_text; *cptr != '\0';) { /* love this style(9) */ + volatile int parse_status; /* return value from parse_config */ + + for (eptr = config_line; (*cptr != '\n') && (*cptr != '\0');) /* until the end of the line */ + *eptr++ = *cptr++; + *eptr = '\0'; /* and delimit */ + if (setjmp(command_fail) == 0) { /* come back here on error and continue */ + parse_status = parse_config(config_line, &keyword_set); /* parse the config line */ + if (parse_status < 0) { /* error in config */ + /* This config should have been parsed in user + * space. If we run into problems here, something + * serious is afoot. Complain and let the user + * snarf the config to see what's wrong */ + printf("vinum: Config error on drive %s, aborting integration\n", nd.ni_dirp); + Free(config_text); + Free(config_line); + free_drive(drive); /* give it back */ + return EINVAL; + } + } + while (*cptr == '\n') + cptr++; /* skip to next line */ + } + Free(config_text); + if ((vinum_conf.flags & VF_READING_CONFIG) == 0) /* not reading config */ + updateconfig(0); /* update object states */ + printf("vinum: read configuration from %s\n", drivename); + return 0; /* it all worked */ + } else { /* no vinum label found */ + if (drive->lasterror) { + set_drive_state(drive->driveno, drive_down, 1); + return drive->lasterror; + } else + return ENODEV; /* not our device */ + } +} + +/* Kludge: kernel printf doesn't handle longs correctly XXX */ +static char *lltoa(long long l, char *s); +static char *sappend(char *txt, char *s); + +static char * +lltoa(long long l, char *s) +{ + if (l < 0) { + *s++ = '-'; + l = -l; + } + if (l > 9) { + s = lltoa(l / 10, s); + l %= 10; + } + *s++ = l + '0'; + return s; +} + +static char * +sappend(char *txt, char *s) +{ + while (*s++ = *txt++); + return s - 1; +} + +/* Format the configuration in text form into the buffer + * at config. Don't go beyond len bytes + * XXX this stinks. Fix soon. */ +void +format_config(char *config, int len) +{ + BROKEN_GDB; + int i; + int j; + char *s = config; + + bzero(config, len); + + /* First write the drive configuration */ + for (i = 0; i < vinum_conf.drives_used; i++) { + struct drive *drive; + + drive = &vinum_conf.drive[i]; + if (drive->state != drive_unallocated) { + sprintf(s, + "drive %s state %s device %s\n", + drive->label.name, + drive_state(drive->state), + drive->devicename); + while (*s) + s++; /* find the end */ + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } + + /* Then the volume configuration */ + for (i = 0; i < vinum_conf.volumes_used; i++) { + struct volume *vol; + + vol = &vinum_conf.volume[i]; + if (vol->state != volume_unallocated) { + if (vol->preferred_plex >= 0) /* preferences, */ + sprintf(s, + "volume %s state %s readpol prefer %s", + vol->name, + volume_state(vol->state), + vinum_conf.plex[vol->preferred_plex].name); + else /* default round-robin */ + sprintf(s, + "volume %s state %s", + vol->name, + volume_state(vol->state)); + while (*s) + s++; /* find the end */ + s = sappend("\n", s); + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } + + /* Then the plex configuration */ + for (i = 0; i < vinum_conf.plexes_used; i++) { + struct plex *plex; + + plex = &vinum_conf.plex[i]; + if (plex->state != plex_unallocated) { + sprintf(s, "plex name %s state %s org %s ", + plex->name, + plex_state(plex->state), + plex_org(plex->organization)); + while (*s) + s++; /* find the end */ + if ((plex->organization == plex_striped) + ) { + sprintf(s, "%db ", (int) plex->stripesize); + while (*s) + s++; /* find the end */ + } + if (plex->volno >= 0) /* we have a volume */ + sprintf(s, "vol %s ", vinum_conf.volume[plex->volno].name); + while (*s) + s++; /* find the end */ + for (j = 0; j < plex->subdisks; j++) { + sprintf(s, " sd %s", vinum_conf.sd[plex->sdnos[j]].name); + } + s = sappend("\n", s); + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } + + /* And finally the subdisk configuration */ + for (i = 0; i < vinum_conf.subdisks_used; i++) { + struct sd *sd = &vinum_conf.sd[i]; /* XXX */ + if (vinum_conf.sd[i].state != sd_unallocated) { + sprintf(s, + "sd name %s drive %s plex %s state %s len ", + sd->name, + vinum_conf.drive[sd->driveno].label.name, + vinum_conf.plex[sd->plexno].name, + sd_state(sd->state)); + while (*s) + s++; /* find the end */ + s = lltoa(sd->sectors, s); + s = sappend("b driveoffset ", s); + s = lltoa(sd->driveoffset, s); + s = sappend("b plexoffset ", s); + s = lltoa(sd->plexoffset, s); + s = sappend("b\n", s); + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } +} + +/* Write the configuration to all vinum slices */ +int +save_config(void) +{ + BROKEN_GDB; + int error; + int written_config; /* set when we firstnwrite the config to disk */ + int driveno; + struct drive *drive; /* point to current drive info */ + struct vinum_hdr *vhdr; /* and as header */ + char *config; /* point to config data */ + int wlabel_on; /* to set writing label on/off */ + + /* don't save the configuration while we're still working on it */ + if (vinum_conf.flags & VF_CONFIGURING) + return 0; + written_config = 0; /* no config written yet */ + /* Build a volume header */ + vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* get space for the config data */ + CHECKALLOC(vhdr, "Can't allocate config data"); + vhdr->magic = VINUM_MAGIC; /* magic number */ + vhdr->config_length = MAXCONFIG; /* length of following config info */ + + config = Malloc(MAXCONFIG); /* get space for the config data */ + CHECKALLOC(config, "Can't allocate config data"); + + format_config(config, MAXCONFIG); + error = 0; /* no errors yet */ + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + drive = &vinum_conf.drive[driveno]; /* point to drive */ + + if (drive->state != drive_down) { +#if (__FreeBSD__ >= 3) + getmicrotime(&drive->label.last_update); /* time of last update is now */ +#else + drive->label.last_update = time; /* time of last update is now */ +#endif + bcopy((char *) &drive->label, /* and the label info from the drive structure */ + (char *) &vhdr->label, + sizeof(vhdr->label)); + if ((drive->state != drive_unallocated) + && (drive->state != drive_uninit)) { + wlabel_on = 1; /* enable writing the label */ + error = VOP_IOCTL(drive->vp, /* make the label writeable */ + DIOCWLABEL, + (caddr_t) & wlabel_on, + FWRITE, + NOCRED, + myproc); + if (error == 0) + error = write_drive(drive, vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); + if (error == 0) + error = write_drive(drive, config, MAXCONFIG, VINUM_CONFIG_OFFSET); + wlabel_on = 0; /* enable writing the label */ + VOP_IOCTL(drive->vp, /* make the label non-writeable again */ + DIOCWLABEL, + (caddr_t) & wlabel_on, + FWRITE, + NOCRED, + myproc); + if (error) { + printf("vinum: Can't write config to %s, error %d\n", drive->devicename, error); + set_drive_state(drive->driveno, drive_down, 1); + } else + written_config = 1; /* we've written it on at least one drive */ + } + } + } + Free(vhdr); + Free(config); + return written_config == 0; /* return 1 if we failed to write config */ +} + +/* Disk labels are a mess. The correct way to access them + * is with the DIOC[GSW]DINFO ioctls, but some programs, such + * as newfs, access the disk directly, so we have to write + * things there. We do this only on request. If a user + * request tries to read it directly, we fake up one on the fly. + */ + +/* get_volume_label returns a label structure to lp, which + * is allocated by the caller */ +void +get_volume_label(struct volume *vol, struct disklabel *lp) +{ + bzero(lp, sizeof(struct disklabel)); + + strncpy(lp->d_typename, "vinum", sizeof(lp->d_typename)); + lp->d_type = DTYPE_VINUM; + strncpy(lp->d_packname, vol->name, min(sizeof(lp->d_packname), sizeof(vol->name))); + lp->d_rpm = 14400 * vol->plexes; /* to keep them guessing */ + lp->d_interleave = 1; + lp->d_flags = 0; + + /* Fitting unto the vine, a vinum has a single + * track with all its sectors */ + lp->d_secsize = DEV_BSIZE; /* bytes per sector */ + lp->d_nsectors = vol->size; /* data sectors per track */ + lp->d_ntracks = 1; /* tracks per cylinder */ + lp->d_ncylinders = 1; /* data cylinders per unit */ + lp->d_secpercyl = vol->size; /* data sectors per cylinder */ + lp->d_secperunit = vol->size; /* data sectors per unit */ + + lp->d_bbsize = BBSIZE; + lp->d_sbsize = SBSIZE; + + lp->d_magic = DISKMAGIC; + lp->d_magic2 = DISKMAGIC; + + /* Set up partitions a, b and c to be identical + * and the size of the volume. a is UFS, b is + * swap, c is nothing */ + lp->d_partitions[0].p_size = vol->size; + lp->d_partitions[0].p_fsize = 1024; + lp->d_partitions[0].p_fstype = FS_BSDFFS; /* FreeBSD File System :-) */ + lp->d_partitions[0].p_fsize = 1024; /* FS fragment size */ + lp->d_partitions[0].p_frag = 8; /* and fragments per block */ + lp->d_partitions[SWAP_PART].p_size = vol->size; + lp->d_partitions[SWAP_PART].p_fstype = FS_SWAP; /* swap partition */ + lp->d_partitions[LABEL_PART].p_size = vol->size; + lp->d_npartitions = LABEL_PART + 1; + strncpy(lp->d_packname, vol->name, min(sizeof(lp->d_packname), sizeof(vol->name))); + lp->d_checksum = dkcksum(lp); +} + +int +write_volume_label(int volno) +{ + struct disklabel *lp; + struct buf *bp; + struct disklabel *dlp; + struct volume *vol; + int error; + + lp = (struct disklabel *) Malloc((sizeof(struct disklabel) + (DEV_BSIZE - 1)) & (DEV_BSIZE - 1)); + if (lp == 0) + return ENOMEM; + + if ((unsigned) (volno) >= (unsigned) vinum_conf.volumes_used) /* invalid volume */ + return ENOENT; + + vol = &VOL[volno]; /* volume in question */ + if (vol->state == volume_unallocated) /* nothing there */ + return ENOENT; + + get_volume_label(vol, lp); /* get the label */ + + /* Now write to disk. This code is derived from the + * system writedisklabel (), which does silly things + * like reading the label and refusing to write + * unless it's already there. */ + bp = geteblk((int) lp->d_secsize); /* get a buffer */ + bp->b_dev = minor(vol->devno) | (CDEV_MAJOR << MAJORDEV_SHIFT); /* our own raw volume */ + bp->b_blkno = LABELSECTOR * ((int) lp->d_secsize / DEV_BSIZE); + bp->b_bcount = lp->d_secsize; + bzero(bp->b_data, lp->d_secsize); + dlp = (struct disklabel *) bp->b_data; + *dlp = *lp; + bp->b_flags &= ~B_INVAL; + bp->b_flags |= B_BUSY | B_WRITE; + vinumstrategy(bp); /* write it out */ + error = biowait(bp); + bp->b_flags |= B_INVAL | B_AGE; + brelse(bp); + return error; +} + +/* Initialize a subdisk */ +int +initsd(int sdno) +{ + return 0; +} diff --git a/lkm/vinum/lock.c b/lkm/vinum/lock.c new file mode 100644 index 0000000..ed69649 --- /dev/null +++ b/lkm/vinum/lock.c @@ -0,0 +1,137 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: lock.c,v 1.6 1998/07/28 06:32:57 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" + +/* Lock routines. Currently, we lock either an individual volume + * or the global configuration. I don't think tsleep and + * wakeup are SMP safe. FIXME XXX */ + +/* Lock a volume, wait if it's in use */ +int +lockvol(struct volume *vol) +{ + int error; + + while ((vol->flags & VF_LOCKED) != 0) { + vol->flags |= VF_LOCKING; + /* It would seem to make more sense to sleep on + * the address 'vol'. Unfortuntaly we can't + * guarantee that this address won't change due to + * table expansion. The address we choose won't change. */ + if ((error = tsleep(&vinum_conf.volume + vol->devno, + PRIBIO | PCATCH, + "volock", + 0)) != 0) + return error; + } + vol->flags |= VF_LOCKED; + return 0; +} + +/* Unlock a volume and let the next one at it */ +void +unlockvol(struct volume *vol) +{ + vol->flags &= ~VF_LOCKED; + if ((vol->flags & VF_LOCKING) != 0) { + vol->flags &= ~VF_LOCKING; + wakeup(&vinum_conf.volume + vol->devno); + } +} + +/* Lock a plex, wait if it's in use */ +int +lockplex(struct plex *plex) +{ + int error; + + while ((plex->flags & VF_LOCKED) != 0) { + plex->flags |= VF_LOCKING; + /* It would seem to make more sense to sleep on + * the address 'plex'. Unfortuntaly we can't + * guarantee that this address won't change due to + * table expansion. The address we choose won't change. */ + if ((error = tsleep(&vinum_conf.plex + plex->sdnos[0], + PRIBIO | PCATCH, + "plexlk", + 0)) != 0) + return error; + } + plex->flags |= VF_LOCKED; + return 0; +} + +/* Unlock a plex and let the next one at it */ +void +unlockplex(struct plex *plex) +{ + plex->flags &= ~VF_LOCKED; + if ((plex->flags & VF_LOCKING) != 0) { + plex->flags &= ~VF_LOCKING; + wakeup(&vinum_conf.plex + plex->plexno); + } +} + + +/* Get a lock for the global config, wait if it's not available */ +int +lock_config(void) +{ + int error; + + while ((vinum_conf.flags & VF_LOCKED) != 0) { + vinum_conf.flags |= VF_LOCKING; + if ((error = tsleep(&vinum_conf, PRIBIO | PCATCH, "vincfg", 0)) != 0) + return error; + } + vinum_conf.flags |= VF_LOCKED; + return 0; +} + +/* Unlock and wake up any waiters */ +void +unlock_config(void) +{ + vinum_conf.flags &= ~VF_LOCKED; + if ((vinum_conf.flags & VF_LOCKING) != 0) { + vinum_conf.flags &= ~VF_LOCKING; + wakeup(&vinum_conf); + } +} diff --git a/lkm/vinum/makestatetext b/lkm/vinum/makestatetext new file mode 100755 index 0000000..ca5eb2d --- /dev/null +++ b/lkm/vinum/makestatetext @@ -0,0 +1,40 @@ +#!/bin/sh +# Make statetexts.h from vinumstate.h +# $Id: makestatetext,v 1.4 1998/03/13 05:36:16 grog Exp grog $ +infile=vinumstate.h +ofile=statetexts.h +cat <COPYRIGHT > $ofile + +echo >>$ofile "/* Created by $0 on" `date`. "Do not edit */" +echo >>$ofile +echo >>$ofile "/* Drive state texts */" +echo >>$ofile "char *drivestatetext [] = + { " +egrep -e 'drive_[A-z0-9]*,' <$infile | grep -v = | sed 's: *drive_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; + +/* Subdisk state texts */ +char *sdstatetext [] = + { +FOO +egrep -e 'sd_[A-z0-9]*,' $infile | grep -v = | sed 's: *sd_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; + +/* Plex state texts */ +char *plexstatetext [] = + { +FOO +egrep -e 'plex_[A-z0-9]*,' $infile | grep -v = | sed 's: *plex_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; + +/* Volume state texts */ +char *volstatetext [] = + { +FOO +egrep -e 'volume_[A-z0-9]*,' $infile | grep -v = | sed 's: *volume_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; +FOO diff --git a/lkm/vinum/memory.c b/lkm/vinum/memory.c new file mode 100644 index 0000000..5dee671 --- /dev/null +++ b/lkm/vinum/memory.c @@ -0,0 +1,186 @@ + +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: memory.c,v 1.16 1998/08/08 04:43:22 grog Exp grog $ + */ + +#define REALLYKERNEL +#define USES_VM +#include "vinumhdr.h" + +extern jmp_buf command_fail; /* return on a failed command */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +void freedatabuf(struct mc *me); +caddr_t allocdatabuf(struct mc *me); + +void +expand_table(void **table, int oldsize, int newsize) +{ + if (newsize > oldsize) { + int *temp; + + temp = (int *) Malloc(newsize); /* allocate a new table */ + CHECKALLOC(temp, "vinum: Can't expand table\n"); + if (*table != NULL) { /* already something there, */ + bcopy((char *) *table, (char *) temp, oldsize); /* copy it to the old table */ + Free(*table); + } + *table = temp; + } +} + +#ifndef DEBUG +/* increase the size of a request block */ +void +expandrq(struct plexrq *prq) +{ + expand_table((void **) &prq->rqe, + prq->requests * sizeof(struct rqelement), + (prq->requests + RQELTS) * sizeof(struct rqelement)); + bzero(&prq->rqe[prq->requests], RQELTS * sizeof(struct rqelement)); /* clear the new part */ + prq->rqcount += RQELTS; +} + +#endif + +#if DEBUG /* XXX debug */ +#define MALLOCENTRIES 16384 +int malloccount = 0; +int highwater = 0; /* highest index ever allocated */ +static struct mc malloced[MALLOCENTRIES]; + +static total_malloced; + +caddr_t +MMalloc(int size, char *file, int line) +{ + caddr_t result; + int i; + static int seq = 0; + int s; + struct mc me; /* information to pass to allocdatabuf */ + + if (malloccount >= MALLOCENTRIES) { /* too many */ + printf("vinum: can't allocate table space to trace memory allocation"); + return 0; /* can't continue */ + } + result = malloc(size, M_DEVBUF, M_WAITOK); /* use malloc for smaller and irregular stuff */ + if (result == NULL) + printf("vinum: can't allocate %d bytes from %s:%d\n", size, file, line); + else { + me.flags = 0; /* allocation via malloc */ + s = splhigh(); + for (i = 0; i < malloccount; i++) { + if (((result + size) > malloced[i].address) + && (result < malloced[i].address + malloced[i].size)) /* overlap */ + Debugger("Malloc overlap"); + } + if (result) { + i = malloccount++; + total_malloced += size; + malloced[i].address = result; + malloced[i].size = size; + malloced[i].line = line; + malloced[i].seq = seq++; + malloced[i].flags = me.flags; + malloced[i].databuf = me.databuf; /* only used with kva alloc */ + bcopy(file, malloced[i].file, min(strlen(file) + 1, 16)); + } + if (malloccount > highwater) + highwater = malloccount; + splx(s); + } + return result; +} + +void +FFree(void *mem, char *file, int line) +{ + int i; + int s; + + s = splhigh(); + for (i = 0; i < malloccount; i++) { + if ((caddr_t) mem == malloced[i].address) { /* found it */ + bzero(mem, malloced[i].size); /* XXX */ + free(mem, M_DEVBUF); + malloccount--; + total_malloced -= malloced[i].size; + if (i < malloccount) /* more coming after */ + bcopy(&malloced[i + 1], &malloced[i], (malloccount - i) * sizeof(struct mc)); + splx(s); + return; + } + } + splx(s); + printf("Freeing unallocated data at 0x%08x from %s, line %d\n", (int) mem, file, line); + Debugger("Free"); +} + +void +vinum_meminfo(caddr_t data) +{ + struct meminfo *m = (struct meminfo *) data; + + m->mallocs = malloccount; + m->total_malloced = total_malloced; + m->malloced = malloced; + m->highwater = highwater; +} + +int +vinum_mallocinfo(caddr_t data) +{ + struct mc *m = (struct mc *) data; + unsigned int ent = *(int *) data; /* 1st word is index */ + + if (ent >= malloccount) + return ENOENT; + m->address = malloced[ent].address; + m->size = malloced[ent].size; + m->line = malloced[ent].line; + m->seq = malloced[ent].seq; + bcopy(malloced[ent].file, m->file, 16); + return 0; +} + +#endif diff --git a/lkm/vinum/parser.c b/lkm/vinum/parser.c new file mode 100644 index 0000000..9e7f791 --- /dev/null +++ b/lkm/vinum/parser.c @@ -0,0 +1,206 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: parser.c,v 1.11 1998/08/10 08:50:42 grog Exp grog $ + */ + +/* This file contains the parser for the configuration routines. It's used + * both in the kernel and in the user interface program, thus the separate file. */ + +/* Go through a text and split up into text tokens. These are either non-blank + * sequences, or any sequence (except \0) enclosed in ' or ". Embedded ' or + * " characters may be escaped by \, which otherwise has no special meaning. + * + * Delimit by following with a \0, and return pointers to the starts at token []. + * Return the number of tokens found as the return value. + * + * This method has the restriction that a closing " or ' must be followed by + * grey space. + * + * Error conditions are end of line before end of quote, or no space after + * a closing quote. In this case, tokenize() returns -1. */ + +#include <sys/param.h> +#ifdef KERNEL +#undef KERNEL /* XXX */ +#define REALLYKERNEL +#else +#include <ctype.h> +#include <errno.h> +#include <fcntl.h> +#endif +/* All this mess for a single struct definition */ +#include <sys/uio.h> +#include <sys/namei.h> +#include <sys/disklabel.h> +#include <sys/mount.h> +#include <sys/device.h> +#include <sys/disk.h> +#include "sys/buf.h" + +#include <vinumvar.h> +#include "vinumkw.h" +#include "vinumio.h" +#include "vinumext.h" + +#ifdef REALLYKERNEL +#define isspace(c) ((c == ' ') || (c == '\t')) /* check for white space */ +#else /* get it from the headers */ +#include <ctype.h> +#endif + +/* enum keyword is defined in vinumvar.h */ + +#define keypair(x) { #x, kw_##x } /* create pair "foo", kw_foo */ +#define flagkeypair(x) { "-"#x, kw_##x } /* create pair "-foo", kw_foo */ +#define KEYWORDSET(x) {sizeof (x) / sizeof (struct _keywords), x} + +/* Normal keywords. These are all the words that vinum knows. */ +struct _keywords keywords[] = +{keypair(drive), + keypair(sd), + keypair(subdisk), + keypair(plex), + keypair(volume), + keypair(vol), + keypair(setupstate), + keypair(readpol), + keypair(org), + keypair(name), + keypair(writethrough), + keypair(writeback), + keypair(raw), + keypair(device), + keypair(concat), + keypair(raid5), + keypair(striped), + keypair(plexoffset), + keypair(driveoffset), + keypair(length), + keypair(len), + keypair(state), + keypair(round), + keypair(prefer), + keypair(rename), + keypair(detached), +#ifndef KERNEL /* for vinum(8) only */ +#ifdef DEBUG + keypair(debug), +#endif + keypair(attach), + keypair(detach), + keypair(printconfig), + keypair(replace), + keypair(create), + keypair(read), + keypair(modify), + keypair(list), + keypair(l), + keypair(ld), + keypair(ls), + keypair(lp), + keypair(lv), + keypair(info), + keypair(set), + keypair(rm), + keypair(init), + keypair(label), + keypair(resetconfig), + keypair(start), + keypair(stop), + keypair(resetstats) +#endif +}; +struct keywordset keyword_set = KEYWORDSET(keywords); + +#ifndef KERNEL +struct _keywords flag_keywords[] = +{flagkeypair(f), + flagkeypair(d), + flagkeypair(v), + flagkeypair(s), + flagkeypair(r) +}; +struct keywordset flag_set = KEYWORDSET(flag_keywords); + +#endif + +int +tokenize(char *cptr, char *token[]) +{ + char delim; /* delimiter for searching for the partner */ + int tokennr; /* index of this token */ + tokennr = 0; /* none found yet */ + + for (;;) { + while (isspace(*cptr)) + cptr++; /* skip initial white space */ + if ((*cptr == '\0') || (*cptr == '\n') || (*cptr == '#')) /* end of line */ + return tokennr; /* return number of tokens found */ + delim = *cptr; + token[tokennr] = cptr; /* point to it */ + tokennr++; /* one more */ + /* XXX this is broken. It leaves superfluous \\ characters in the text */ + if ((delim == '\'') || (delim == '"')) { /* delimitered */ + for (;;) { + cptr++; + if ((*cptr == delim) && (cptr[-1] != '\\')) { /* found the partner */ + cptr++; /* move on past */ + if (!isspace(*cptr)) /* error, no space after closing quote */ + return -1; + *cptr++ = '\0'; /* delimit */ + } else if ((*cptr == '\0') || (*cptr == '\n')) /* end of line */ + return -1; + } + } else { /* not quoted */ + while ((*cptr != '\0') && (!isspace(*cptr)) && (*cptr != '\n')) + cptr++; + if (*cptr != '\0') /* not end of the line, */ + *cptr++ = '\0'; /* delimit and move to the next */ + } + } +} + +/* Find a keyword and return an index */ +enum keyword +get_keyword(char *name, struct keywordset *keywordset) +{ + int i; + struct _keywords *keywords = keywordset->k; /* point to the keywords */ + for (i = 0; i < keywordset->size; i++) + if (!strcmp(name, keywords[i].name)) + return (enum keyword) keywords[i].keyword; + return kw_invalid_keyword; +} diff --git a/lkm/vinum/request.c b/lkm/vinum/request.c new file mode 100644 index 0000000..589eb3f --- /dev/null +++ b/lkm/vinum/request.c @@ -0,0 +1,882 @@ +/* XXX to do: + + * Decide where we need splbio () + */ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: request.c,v 1.17 1998/08/13 06:04:47 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" +#include <miscfs/specfs/specdev.h> +#include <sys/resourcevar.h> + +/* pointer to ioctl p parameter, to save passing it around */ +extern struct proc *myproc; + +enum requeststatus bre(struct request *rq, + int plexno, + daddr_t * diskstart, + daddr_t diskend); +enum requeststatus bre5(struct request *rq, + int plexno, + daddr_t * diskstart, + daddr_t diskend); +enum requeststatus build_read_request(struct request *rq, int volplexno); +enum requeststatus build_write_request(struct request *rq); +enum requeststatus build_rq_buffer(struct rqelement *rqe, struct plex *plex); +void freerq(struct request *rq); +void free_rqg(struct rqgroup *rqg); +int find_alternate_sd(struct request *rq); +int check_range_covered(struct request *); +void complete_rqe(struct buf *bp); +void complete_raid5_write(struct rqelement *); +int abortrequest(struct request *rq, int error); +void sdio(struct buf *bp); +void sdio_done(struct buf *bp); +int vinum_bounds_check(struct buf *bp, struct volume *vol); +caddr_t allocdatabuf(struct rqelement *rqe); +void freedatabuf(struct rqelement *rqe); + +void +vinumstrategy(struct buf *bp) +{ + BROKEN_GDB; + int volno; + struct volume *vol = NULL; + int s; + struct devcode *device = (struct devcode *) &bp->b_dev; /* decode device number */ + enum requeststatus status; + + switch (device->type) { + case VINUM_SD_TYPE: + sdio(bp); + return; + + /* In fact, vinum doesn't handle drives: they're + * handled directly by the disk drivers */ + case VINUM_DRIVE_TYPE: + default: + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + biodone(bp); + return; + + case VINUM_VOLUME_TYPE: /* volume I/O */ + volno = VOLNO(bp->b_dev); + vol = &VOL[volno]; + if (vol->state != volume_up) { /* can't access this volume */ + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + biodone(bp); + return; + } + if (vinum_bounds_check(bp, vol) <= 0) { /* don't like them bounds */ + biodone(bp); /* have nothing to do with this */ + return; + } + /* FALLTHROUGH */ + /* Plex I/O is pretty much the same as volume I/O + * for a single plex. Indicate this by passing a NULL + * pointer (set above) for the volume */ + case VINUM_PLEX_TYPE: + bp->b_resid = bp->b_bcount; /* transfer everything */ + vinumstart(bp, 0); + return; + } +} + +/* Start a transfer. Return -1 on error, + * 0 if OK, 1 if we need to retry. + * Parameter reviveok is set when doing + * transfers for revives: it allows transfers to + * be started immediately when a revive is in + * progress. During revive, normal transfers + * are queued if they share address space with + * a currently active revive operation. */ +int +vinumstart(struct buf *bp, int reviveok) +{ + BROKEN_GDB; + int plexno; + int maxplex; /* maximum number of plexes to handle */ + struct volume *vol; + struct rqgroup *rqg; /* current plex's requests */ + struct rqelement *rqe; /* individual element */ + struct request *rq; /* build up our request here */ + int rqno; /* index in request list */ + enum requeststatus status; + + /* XXX In these routines, we're assuming that + * we will always be called with bp->b_bcount + * which is a multiple of the sector size. This + * is a reasonable assumption, since we are only + * called from system routines. Should we check + * anyway? */ + + if ((bp->b_bcount % DEV_BSIZE) != 0) { /* bad length */ + bp->b_error = EINVAL; /* invalid size */ + bp->b_flags |= B_ERROR; + biodone(bp); + return -1; + } + rq = (struct request *) Malloc(sizeof(struct request)); /* allocate a request struct */ + if (rq == NULL) { /* can't do it */ + bp->b_error = ENOMEM; /* can't get memory */ + bp->b_flags |= B_ERROR; + biodone(bp); + return -1; + } + bzero(rq, sizeof(struct request)); + + /* Note the volume ID. This can be NULL, which + * the request building functions use as an + * indication for single plex I/O */ + rq->bp = bp; /* and the user buffer struct */ + + if (DEVTYPE(bp->b_dev) == VINUM_VOLUME_TYPE) { /* it's a volume, */ + rq->volplex.volno = VOLNO(bp->b_dev); /* get the volume number */ + vol = &VOL[rq->volplex.volno]; /* and point to it */ + vol->active++; /* one more active request */ + maxplex = vol->plexes; /* consider all its plexes */ + } else { + vol = NULL; /* no volume */ + rq->volplex.plexno = PLEXNO(bp->b_dev); /* point to the plex */ + rq->isplex = 1; /* note that it's a plex */ + maxplex = 1; /* just the one plex */ + } + + if (bp->b_flags & B_READ) { + /* This is a read request. Decide + * which plex to read from. + * + * There's a potential race condition here, + * since we're not locked, and we could end + * up multiply incrementing the round-robin + * counter. This doesn't have any serious + * effects, however. */ + if (vol != NULL) { + vol->reads++; + vol->bytes_read += bp->b_bcount; + plexno = vol->preferred_plex; /* get the plex to use */ + if (plexno < 0) { /* round robin */ + plexno = vol->last_plex_read; + vol->last_plex_read++; + if (vol->last_plex_read == vol->plexes) /* got the the end? */ + vol->last_plex_read = 0; /* wrap around */ + } + status = build_read_request(rq, plexno); /* build a request */ + } else { + daddr_t diskaddr = bp->b_blkno; /* start offset of transfer */ + status = bre(rq, /* build a request list */ + rq->volplex.plexno, + &diskaddr, + diskaddr + (bp->b_bcount / DEV_BSIZE)); + } + + if ((status > REQUEST_RECOVERED) /* can't satisfy it */ + ||(bp->b_flags & B_DONE)) { /* XXX shouldn't get this without bad status */ + if (status == REQUEST_DOWN) { /* not enough subdisks */ + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + } + biodone(bp); + freerq(rq); + return -1; + } + return launch_requests(rq, reviveok); /* now start the requests if we can */ + } else + /* This is a write operation. We write to all + * plexes. If this is a RAID 5 plex, we must also + * update the parity stripe. */ + { + if (vol != NULL) { + vol->writes++; + vol->bytes_written += bp->b_bcount; + status = build_write_request(rq); /* Not all the subdisks are up */ + } else { /* plex I/O */ + daddr_t diskstart; + + diskstart = bp->b_blkno; /* start offset of transfer */ + status = bre(rq, + PLEXNO(bp->b_dev), + &diskstart, + bp->b_blkno + (bp->b_bcount / DEV_BSIZE)); /* build requests for the plex */ + } + if ((status > REQUEST_RECOVERED) /* can't satisfy it */ + ||(bp->b_flags & B_DONE)) { /* XXX shouldn't get this without bad status */ + if (status == REQUEST_DOWN) { /* not enough subdisks */ + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + } + if ((bp->b_flags & B_DONE) == 0) + biodone(bp); + freerq(rq); + return -1; + } + return launch_requests(rq, reviveok); /* start the requests */ + } +} + +/* Call the low-level strategy routines to + * perform the requests in a struct request */ +int +launch_requests(struct request *rq, int reviveok) +{ + struct rqgroup *rqg; + int rqno; /* loop index */ + struct rqelement *rqe; /* current element */ + int s; + + /* First find out whether we're reviving, and the + * request contains a conflict. If so, we hang + * the request off plex->waitlist of the first + * plex we find which is reviving */ + if ((rq->flags & XFR_REVIVECONFLICT) /* possible revive conflict */ + &&(!reviveok)) { /* and we don't want to do it now, */ + struct volume *vol = &VOL[VOLNO(rq->bp->b_dev)]; + struct plex *plex; + int plexno; + + for (plexno = 0; plexno < vol->plexes; plexno++) { /* find the reviving plex */ + plex = &PLEX[vol->plex[plexno]]; + if (plex->state == plex_reviving) /* found it */ + break; + } + if (plexno < vol->plexes) { /* found it? */ + struct request *waitlist = plex->waitlist; /* point to the waiting list */ + + while (waitlist->next != NULL) /* find the end */ + waitlist = waitlist->next; + waitlist->next = rq; /* hook our request there */ + return 0; /* and get out of here */ + } else /* bad vinum, bad */ + printf("vinum: can't find reviving plex for volume %s\n", vol->name); + } + rq->active = 0; /* nothing yet */ + /* XXX This is probably due to a bug */ + if (rq->rqg == NULL) { /* no request */ + abortrequest(rq, EINVAL); + return -1; + } +#if DEBUG + if (debug & DEBUG_ADDRESSES) + printf("Request: %x\nWrite dev 0x%x, offset 0x%x, length %ld\n", + (u_int) rq, + rq->bp->b_dev, + rq->bp->b_blkno, + rq->bp->b_bcount); /* XXX */ + vinum_conf.lastrq = (int) rq; + vinum_conf.lastbuf = rq->bp; +#endif + for (rqg = rq->rqg; rqg != NULL; rqg = rqg->next) { /* through the whole request chain */ + rqg->active = rqg->count; /* they're all active */ + rq->active++; /* one more active request group */ + for (rqno = 0; rqno < rqg->count; rqno++) { + rqe = &rqg->rqe[rqno]; + if (rqe->flags & XFR_BAD_SUBDISK) /* this subdisk is bad, */ + rqg->active--; /* one less active request */ + else { + struct drive *drive = &DRIVE[rqe->driveno]; /* drive to access */ + if ((rqe->b.b_flags & B_READ) == 0) + rqe->b.b_vp->v_numoutput++; /* one more output going */ +#if DEBUG + if (debug & DEBUG_ADDRESSES) + printf(" %s dev 0x%x, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n", + rqe->b.b_flags & B_READ ? "Read" : "Write", + rqe->b.b_dev, + rqe->sdno, + (u_int) (rqe->b.b_blkno - SD[rqe->sdno].driveoffset), + rqe->b.b_blkno, + rqe->b.b_bcount); /* XXX */ + if (debug & DEBUG_NUMOUTPUT) + printf(" vinumstart sd %d numoutput %ld\n", + rqe->sdno, + rqe->b.b_vp->v_numoutput); +#endif + /* fire off the request */ + s = splbio(); + (*bdevsw[major(rqe->b.b_dev)]->d_strategy) (&rqe->b); + splx(s); + } + /* XXX Do we need caching? Think about this more */ + } + } + return 0; +} + +/* define the low-level requests needed to perform a + * high-level I/O operation for a specific plex 'plexno'. + * + * Return 0 if all subdisks involved in the request are up, 1 if some + * subdisks are not up, and -1 if the request is at least partially + * outside the bounds of the subdisks. + * + * Modify the pointer *diskstart to point to the end address. On + * read, return on the first bad subdisk, so that the caller + * (build_read_request) can try alternatives. + * + * On entry to this routine, the rqg structures are not assigned. The + * assignment is performed by expandrq(). Strictly speaking, the + * elements rqe->sdno of all entries should be set to -1, since 0 + * (from bzero) is a valid subdisk number. We avoid this problem by + * initializing the ones we use, and not looking at the others (index + * >= rqg->requests). + */ +enum requeststatus +bre(struct request *rq, + int plexno, + daddr_t * diskaddr, + daddr_t diskend) +{ + BROKEN_GDB; + int sdno; + struct sd *sd; + struct rqgroup *rqg; + struct buf *bp; /* user's bp */ + struct plex *plex; + enum requeststatus status; /* return value */ + daddr_t plexoffset; /* offset of transfer in plex */ + daddr_t stripebase; /* base address of stripe (1st subdisk) */ + daddr_t stripeoffset; /* offset in stripe */ + daddr_t blockoffset; /* offset in stripe on subdisk */ + struct rqelement *rqe; /* point to this request information */ + daddr_t diskstart = *diskaddr; /* remember where this transfer starts */ + + bp = rq->bp; /* buffer pointer */ + status = REQUEST_OK; /* return value: OK until proven otherwise */ + plex = &PLEX[plexno]; /* point to the plex */ + + switch (plex->organization) { + case plex_concat: + for (sdno = 0; sdno < plex->subdisks; sdno++) { + sd = &SD[plex->sdnos[sdno]]; + if ((*diskaddr < (sd->plexoffset + sd->sectors)) /* The request starts before the end of this */ + &&(diskend > sd->plexoffset)) { /* subdisk and ends after the start of this sd */ + if ((sd->state != sd_up) || (plex->state != plex_up)) { + enum requeststatus s; + + s = checksdstate(sd, rq, *diskaddr, diskend); /* do we need to change state? */ + if (s) /* give up? */ + return s; /* yup */ + } + rqg = allocrqg(rq, 1); /* space for the request */ + if (rqg == NULL) { /* malloc failed */ + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; + } + rqg->plexno = plexno; + + rqe = &rqg->rqe[0]; /* point to the element */ + rqe->rqg = rqg; /* group */ + rqe->sdno = sd->sdno; /* put in the subdisk number */ + plexoffset = max(sd->plexoffset, *diskaddr); /* start offset in plex */ + rqe->sdoffset = plexoffset - sd->plexoffset; /* start offset in subdisk */ + rqe->useroffset = plexoffset - diskstart; /* start offset in user buffer */ + rqe->dataoffset = 0; + rqe->datalen = min(diskend - *diskaddr, /* number of sectors to transfer in this sd */ + sd->sectors - rqe->sdoffset); + rqe->groupoffset = 0; /* no groups for concatenated plexes */ + rqe->grouplen = 0; + rqe->buflen = rqe->datalen; /* buffer length is data buffer length */ + rqe->flags = 0; + rqe->driveno = sd->driveno; + *diskaddr += rqe->datalen; /* bump the address */ + if (build_rq_buffer(rqe, plex)) { /* build the buffer */ + deallocrqg(rqg); + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; /* can't do it */ + } + } + if (*diskaddr > diskend) /* we're finished, */ + break; /* get out of here */ + } + break; + + case plex_striped: + { + while (*diskaddr < diskend) { /* until we get it all sorted out */ + /* The offset of the start address from + * the start of the stripe */ + stripeoffset = *diskaddr % (plex->stripesize * plex->subdisks); + + /* The plex-relative address of the + * start of the stripe */ + stripebase = *diskaddr - stripeoffset; + + /* The number of the subdisk in which + * the start is located */ + sdno = stripeoffset / plex->stripesize; + + /* The offset from the beginning of the stripe + * on this subdisk */ + blockoffset = stripeoffset % plex->stripesize; + + sd = &SD[plex->sdnos[sdno]]; /* the subdisk in question */ + if ((sd->state != sd_up) || (plex->state != plex_up)) { + enum requeststatus s; + + s = checksdstate(sd, rq, *diskaddr, diskend); /* do we need to change state? */ + if (s) /* give up? */ + return s; /* yup */ + } + rqg = allocrqg(rq, 1); /* space for the request */ + if (rqg == NULL) { /* malloc failed */ + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; + } + rqg->plexno = plexno; + + rqe = &rqg->rqe[0]; /* point to the element */ + rqe->rqg = rqg; + rqe->sdoffset = stripebase / plex->subdisks + blockoffset; /* start offset in this subdisk */ + rqe->useroffset = *diskaddr - diskstart; /* The offset of the start in the user buffer */ + rqe->dataoffset = 0; + rqe->datalen = min(diskend - *diskaddr, /* the amount remaining to transfer */ + plex->stripesize - blockoffset); /* and the amount left in this stripe */ + rqe->groupoffset = 0; /* no groups for striped plexes */ + rqe->grouplen = 0; + rqe->buflen = rqe->datalen; /* buffer length is data buffer length */ + rqe->flags = 0; + rqe->sdno = sd->sdno; /* put in the subdisk number */ + rqe->driveno = sd->driveno; + + if (rqe->sdoffset >= sd->sectors) { /* starts beyond the end of the subdisk? */ + deallocrqg(rqg); + return REQUEST_EOF; + } else if (rqe->sdoffset + rqe->datalen > sd->sectors) /* ends beyond the end of the subdisk? */ + rqe->datalen = sd->sectors - rqe->sdoffset; /* yes, truncate */ + + if (build_rq_buffer(rqe, plex)) { /* build the buffer */ + deallocrqg(rqg); + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; /* can't do it */ + } + *diskaddr += rqe->datalen; /* look at the remainder */ + if (*diskaddr < diskend) { /* didn't finish the request on this stripe */ + plex->multiblock++; /* count another one */ + if (sdno == plex->subdisks - 1) /* last subdisk, */ + plex->multistripe++; /* another stripe as well */ + } + } + } + break; + + + default: + printf("vinum: invalid plex type in bre"); + } + + return status; +} + +/* Build up a request structure for reading volumes. + * This function is not needed for plex reads, since there's + * no recovery if a plex read can't be satisified. */ +enum requeststatus +build_read_request(struct request *rq, /* request */ + int plexindex) +{ /* index in the volume's plex table */ + BROKEN_GDB; + struct buf *bp; + daddr_t startaddr; /* offset of previous part of transfer */ + daddr_t diskaddr; /* offset of current part of transfer */ + daddr_t diskend; /* and end offset of transfer */ + int plexno; /* plex index in vinum_conf */ + struct rqgroup *rqg; /* point to the request we're working on */ + struct volume *vol; /* volume in question */ + off_t oldstart; /* note where we started */ + int recovered = 0; /* set if we recover a read */ + enum requeststatus status = REQUEST_OK; + + bp = rq->bp; /* buffer pointer */ + diskaddr = bp->b_blkno; /* start offset of transfer */ + diskend = diskaddr + (bp->b_bcount / DEV_BSIZE); /* and end offset of transfer */ + rqg = &rq->rqg[plexindex]; /* plex request */ + vol = &VOL[rq->volplex.volno]; /* point to volume */ + + while (diskaddr < diskend) { /* build up request components */ + startaddr = diskaddr; + status = bre(rq, vol->plex[plexindex], &diskaddr, diskend); /* build up a request */ + switch (status) { + case REQUEST_OK: + continue; + + case REQUEST_RECOVERED: + recovered = 1; + break; + + case REQUEST_EOF: + case REQUEST_ENOMEM: + return status; + + /* if we get here, we have either had a failure or + * a RAID 5 recovery. We don't want to use the + * recovery, because it's expensive, so first we + * check if we have alternatives */ + case REQUEST_DOWN: /* can't access the plex */ + if (vol != NULL) { /* and this is volume I/O */ + /* Try to satisfy the request + * from another plex */ + for (plexno = 0; plexno < vol->plexes; plexno++) { + diskaddr = startaddr; /* start at the beginning again */ + oldstart = startaddr; /* and note where that was */ + if (plexno != plexindex) { /* don't try this plex again */ + bre(rq, vol->plex[plexno], &diskaddr, diskend); /* try a request */ + if (diskaddr > oldstart) { /* we satisfied another part */ + recovered = 1; /* we recovered from the problem */ + status = REQUEST_OK; /* don't complain about it */ + break; + } + } + if (plexno == (vol->plexes - 1)) /* couldn't satisfy the request */ + return REQUEST_DOWN; /* failed */ + } + } else + return REQUEST_DOWN; /* bad luck */ + } + if (recovered) + vol->recovered_reads += recovered; /* adjust our recovery count */ + } + return status; +} + +/* Build up a request structure for writes. + * Return 0 if all subdisks involved in the request are up, 1 if some + * subdisks are not up, and -1 if the request is at least partially + * outside the bounds of the subdisks. */ +enum requeststatus +build_write_request(struct request *rq) +{ /* request */ + BROKEN_GDB; + struct buf *bp; + daddr_t diskstart; /* offset of current part of transfer */ + daddr_t diskend; /* and end offset of transfer */ + int plexno; /* plex index in vinum_conf */ + struct volume *vol; /* volume in question */ + enum requeststatus status; + + bp = rq->bp; /* buffer pointer */ + vol = &VOL[rq->volplex.volno]; /* point to volume */ + diskend = bp->b_blkno + (bp->b_bcount / DEV_BSIZE); /* end offset of transfer */ + status = REQUEST_OK; + for (plexno = 0; plexno < vol->plexes; plexno++) { + diskstart = bp->b_blkno; /* start offset of transfer */ + status = min(status, bre(rq, /* build requests for the plex */ + vol->plex[plexno], + &diskstart, + diskend)); + } + return status; +} + +/* Fill in the struct buf part of a request element. */ +enum requeststatus +build_rq_buffer(struct rqelement *rqe, struct plex *plex) +{ + BROKEN_GDB; + struct sd *sd; /* point to subdisk */ + struct volume *vol; + struct buf *bp; + struct buf *ubp; /* user (high level) buffer header */ + + vol = &VOL[rqe->rqg->rq->volplex.volno]; + sd = &SD[rqe->sdno]; /* point to subdisk */ + bp = &rqe->b; + ubp = rqe->rqg->rq->bp; /* pointer to user buffer header */ + + /* Initialize the buf struct */ + bzero(&rqe->b, sizeof(struct buf)); + bp->b_proc = ubp->b_proc; /* process pointer */ + bp->b_flags = ubp->b_flags & (B_NOCACHE | B_READ | B_ASYNC); /* copy these flags from user bp */ + bp->b_flags |= B_CALL | B_BUSY; /* inform us when it's done */ + if (plex->state == plex_reviving) + bp->b_flags |= B_ORDERED; /* keep request order if we're reviving */ + bp->b_iodone = complete_rqe; /* by calling us here */ + bp->b_dev = DRIVE[rqe->driveno].dev; /* drive device */ + bp->b_blkno = rqe->sdoffset + sd->driveoffset; /* start address */ + bp->b_bcount = rqe->buflen << DEV_BSHIFT; /* number of bytes to transfer */ + bp->b_resid = bp->b_bcount; /* and it's still all waiting */ + bp->b_bufsize = bp->b_bcount; /* and buffer size */ + bp->b_vp = DRIVE[rqe->driveno].vp; /* drive vnode */ + bp->b_rcred = FSCRED; /* we have the file system credentials */ + bp->b_wcred = FSCRED; /* we have the file system credentials */ + + if (rqe->flags & XFR_MALLOCED) { /* this operation requires a malloced buffer */ + bp->b_data = Malloc(bp->b_bcount); /* get a buffer to put it in */ + if (bp->b_data == NULL) { /* failed */ + Debugger("XXX"); + abortrequest(rqe->rqg->rq, ENOMEM); + return REQUEST_ENOMEM; /* no memory */ + } + } else + /* Point directly to user buffer data. This means + * that we don't need to do anything when we have + * finished the transfer */ + bp->b_data = ubp->b_data + rqe->useroffset * DEV_BSIZE; + return 0; +} +/* Abort a request: free resources and complete the + * user request with the specified error */ +int +abortrequest(struct request *rq, int error) +{ + struct buf *bp = rq->bp; /* user buffer */ + + bp->b_flags |= B_ERROR; + bp->b_error = error; + freerq(rq); /* free everything we're doing */ + biodone(bp); + return error; /* and give up */ +} + +/* Check that our transfer will cover the + * complete address space of the user request. + * + * Return 1 if it can, otherwise 0 */ +int +check_range_covered(struct request *rq) +{ + /* XXX */ + return 1; +} + +/* Perform I/O on a subdisk */ +void +sdio(struct buf *bp) +{ + int s; /* spl */ + struct sd *sd; + struct sdbuf *sbp; + daddr_t endoffset; + struct drive *drive; + + sd = &SD[SDNO(bp->b_dev)]; /* point to the subdisk */ + drive = &DRIVE[sd->driveno]; + + if (drive->state != drive_up) { /* XXX until we get the states fixed */ + set_sd_state(SDNO(bp->b_dev), sd_obsolete, setstate_force); + bp->b_flags |= B_ERROR; + bp->b_error = EIO; + biodone(bp); + return; + } + /* XXX decide which states we will really accept here. up + * implies it could be involved with a plex, in which + * case we don't want to dick with it */ + if ((sd->state != sd_up) + && (sd->state != sd_initializing) + && (sd->state != sd_reborn)) { /* we can't access it */ + bp->b_flags |= B_ERROR; + bp->b_flags = EIO; + if (bp->b_flags & B_BUSY) /* XXX why isn't this always the case? */ + biodone(bp); + return; + } + /* Get a buffer */ + sbp = (struct sdbuf *) Malloc(sizeof(struct sdbuf)); + if (sbp == NULL) { + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return; + } + bcopy(bp, &sbp->b, sizeof(struct buf)); /* start with the user's buffer */ + sbp->b.b_flags |= B_CALL; /* tell us when it's done */ + sbp->b.b_iodone = sdio_done; /* here */ + sbp->b.b_dev = DRIVE[sd->driveno].dev; /* device */ + sbp->b.b_vp = DRIVE[sd->driveno].vp; /* vnode */ + sbp->b.b_blkno += sd->driveoffset; + sbp->bp = bp; /* note the address of the original header */ + sbp->sdno = sd->sdno; /* note for statistics */ + sbp->driveno = sd->driveno; + endoffset = bp->b_blkno + sbp->b.b_bcount / DEV_BSIZE; /* final sector offset */ + if (endoffset > sd->sectors) { /* beyond the end */ + sbp->b.b_bcount -= (endoffset - sd->sectors) * DEV_BSIZE; /* trim */ + if (sbp->b.b_bcount <= 0) { /* nothing to transfer */ + bp->b_resid = bp->b_bcount; /* nothing transferred */ + /* XXX Grrr. This doesn't seem to work. Return + * an error after all */ + bp->b_flags |= B_ERROR; + bp->b_error = ENOSPC; + biodone(bp); + Free(sbp); + return; + } + } + if ((sbp->b.b_flags & B_READ) == 0) /* write */ + sbp->b.b_vp->v_numoutput++; /* one more output going */ +#if DEBUG + if (debug & DEBUG_ADDRESSES) + printf(" %s dev 0x%x, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n", + sbp->b.b_flags & B_READ ? "Read" : "Write", + sbp->b.b_dev, + sbp->sdno, + (u_int) (sbp->b.b_blkno - SD[sbp->sdno].driveoffset), + (int) sbp->b.b_blkno, + sbp->b.b_bcount); /* XXX */ + if (debug & DEBUG_NUMOUTPUT) + printf(" vinumstart sd %d numoutput %ld\n", + sbp->sdno, + sbp->b.b_vp->v_numoutput); +#endif + s = splbio(); + (*bdevsw[major(sbp->b.b_dev)]->d_strategy) (&sbp->b); + splx(s); +} + +/* Simplified version of bounds_check_with_label + * Determine the size of the transfer, and make sure it is + * within the boundaries of the partition. Adjust transfer + * if needed, and signal errors or early completion. + * + * Volumes are simpler than disk slices: they only contain + * one component (though we call them a, b and c to make + * system utilities happy), and they always take up the + * complete space of the "partition". + * + * I'm still not happy with this: why should the label be + * protected? If it weren't so damned difficult to write + * one in the first pleace (because it's protected), it wouldn't + * be a problem. + */ +int +vinum_bounds_check(struct buf *bp, struct volume *vol) +{ + int maxsize = vol->size; /* size of the partition (sectors) */ + int size = (bp->b_bcount + DEV_BSIZE - 1) >> DEV_BSHIFT; /* size of this request (sectors) */ + + /* Would this transfer overwrite the disk label? */ + if (bp->b_blkno <= LABELSECTOR /* starts before or at the label */ +#if LABELSECTOR != 0 + && bp->b_blkno + size > LABELSECTOR /* and finishes after */ +#endif + && (!(vol->flags & VF_RAW)) /* and it's not raw */ + &&major(bp->b_dev) == BDEV_MAJOR /* and it's the block device */ + && (bp->b_flags & B_READ) == 0 /* and it's a write */ + && (!vol->flags & (VF_WLABEL | VF_LABELLING))) { /* and we're not allowed to write the label */ + bp->b_error = EROFS; /* read-only */ + bp->b_flags |= B_ERROR; + return -1; + } + if (size == 0) /* no transfer specified, */ + return 0; /* treat as EOF */ + /* beyond partition? */ + if (bp->b_blkno < 0 /* negative start */ + || bp->b_blkno + size > maxsize) { /* or goes beyond the end of the partition */ + /* if exactly at end of disk, return an EOF */ + if (bp->b_blkno == maxsize) { + bp->b_resid = bp->b_bcount; + return 0; + } + /* or truncate if part of it fits */ + size = maxsize - bp->b_blkno; + if (size <= 0) { /* nothing to transfer */ + bp->b_error = EINVAL; + bp->b_flags |= B_ERROR; + return -1; + } + bp->b_bcount = size << DEV_BSHIFT; + } + bp->b_pblkno = bp->b_blkno; + return 1; +} + +/* Allocate a request group and hook + * it in in the list for rq */ +struct rqgroup * +allocrqg(struct request *rq, int elements) +{ + struct rqgroup *rqg; /* the one we're going to allocate */ + int size = sizeof(struct rqgroup) + elements * sizeof(struct rqelement); + + rqg = (struct rqgroup *) Malloc(size); + if (rqg != NULL) { /* malloc OK, */ + if (rq->rqg) /* we already have requests */ + rq->lrqg->next = rqg; /* hang it off the end */ + else /* first request */ + rq->rqg = rqg; /* at the start */ + rq->lrqg = rqg; /* this one is the last in the list */ + + bzero(rqg, size); /* no old junk */ + rqg->rq = rq; /* point back to the parent request */ + rqg->count = elements; /* number of requests in the group */ + } else + Debugger("XXX"); + return rqg; +} + +/* Deallocate a request group out of a chain. We do + * this by linear search: the chain is short, this + * almost never happens, and currently it can only + * happen to the first member of the chain. */ +void +deallocrqg(struct rqgroup *rqg) +{ + struct rqgroup *rqgc = rqg->rq->rqg; /* point to the request chain */ + + if (rqg->rq->rqg == rqg) /* we're first in line */ + rqg->rq->rqg = rqg->next; /* unhook ourselves */ + else { + while (rqgc->next != rqg) /* find the group */ + rqgc = rqgc->next; + rqgc->next = rqg->next; + } + Free(rqgc); +} + +/* Character device interface */ +int +vinumread(dev_t dev, struct uio *uio, int ioflag) +{ + return (physio(vinumstrategy, NULL, dev, 1, minphys, uio)); +} + +int +vinumwrite(dev_t dev, struct uio *uio, int ioflag) +{ + return (physio(vinumstrategy, NULL, dev, 0, minphys, uio)); +} diff --git a/lkm/vinum/request.h b/lkm/vinum/request.h new file mode 100644 index 0000000..b4beccc --- /dev/null +++ b/lkm/vinum/request.h @@ -0,0 +1,159 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: request.h,v 1.10 1998/08/03 07:15:26 grog Exp grog $ + */ + +/* Information needed to set up a transfer */ + +/* struct buf is surprisingly big (about 300 + * bytes), and it's part of the request, so this + * value is really important. Most requests + * don't need more than 2 subrequests per + * plex. The table is automatically extended if + * this value is too small. */ +#define RQELTS 2 /* default of 2 requests per transfer */ + +enum xferinfo { + XFR_NORMAL_READ = 1, + XFR_NORMAL_WRITE = 2, /* write request in normal mode */ + XFR_RECOVERY_READ = 4, + XFR_DEGRADED_WRITE = 8, + XFR_PARITYLESS_WRITE = 0x10, + XFR_NO_PARITY_STRIPE = 0x20, /* parity stripe is not available */ + XFR_DATA_BLOCK = 0x40, /* data block in request */ + XFR_PARITY_BLOCK = 0x80, /* parity block in request */ + XFR_BAD_SUBDISK = 0x100, /* this subdisk is dead */ + XFR_MALLOCED = 0x200, /* this buffer is malloced */ +#if DEBUG + XFR_PHASE2 = 0x800, /* documentation only: 2nd phase write */ +#endif + XFR_REVIVECONFLICT = 0x1000, /* possible conflict with a revive operation */ + /* operations that need a parity block */ + XFR_PARITYOP = (XFR_NORMAL_WRITE | XFR_RECOVERY_READ | XFR_DEGRADED_WRITE), + /* operations that use the group parameters */ + XFR_GROUPOP = (XFR_DEGRADED_WRITE | XFR_RECOVERY_READ), + /* operations that that use the data parameters */ + XFR_DATAOP = (XFR_NORMAL_READ | XFR_NORMAL_WRITE | XFR_PARITYLESS_WRITE), + /* operations requiring read before write */ + XFR_RBW = (XFR_NORMAL_WRITE | XFR_DEGRADED_WRITE), + /* operations that need a malloced buffer */ + XFR_NEEDS_MALLOC = (XFR_NORMAL_WRITE | XFR_RECOVERY_READ | XFR_DEGRADED_WRITE) +}; + +/* Describe one low-level request, part + * of a high-level request. This is an + * extended struct buf buffer, and the first + * element *must* be a struct buf. We pass this structure + * to the I/O routines instead of a struct buf in oder + * to be able to locate the high-level request when it + * completes. + * + * All offsets and lengths are in "blocks", i.e. sectors */ +struct rqelement { + struct buf b; /* buf structure */ + struct rqgroup *rqg; /* pointer to our group */ + /* Information about the transfer */ + daddr_t sdoffset; /* offset in subdisk */ + int useroffset; /* offset in user buffer of normal data */ + /* dataoffset and datalen refer to "individual" + * data transfers (normal read, parityless write) + * and also degraded write. + * + * groupoffset and grouplen refer to the other + * "group" operations (normal write, recovery read) + * Both the offsets are relative to the start of the + * local buffer */ + int dataoffset; /* offset in buffer of the normal data */ + int groupoffset; /* offset in buffer of group data */ + short datalen; /* length of normal data (sectors) */ + short grouplen; /* length of group data (sectors) */ + short buflen; /* total buffer length to allocate */ + short flags; /* really enum xferinfo (see above) */ + /* Ways to find other components */ + short sdno; /* subdisk number */ + short driveno; /* drive number */ +}; + +/* A group of requests built to satisfy a certain + * component of a user request */ +struct rqgroup { + struct rqgroup *next; /* pointer to next group */ + struct request *rq; /* pointer to the request */ + short count; /* number of requests in this group */ + short active; /* and number active */ + short plexno; /* index of plex */ + int badsdno; /* index of bad subdisk or -1 */ + enum xferinfo flags; /* description of transfer */ + struct rqelement rqe[0]; /* and the elements of this request */ +}; + +/* Describe one high-level request and the + * work we have to do to satisfy it */ +struct request { + struct buf *bp; /* pointer to the high-level request */ + int flags; + union { + int volno; /* volume index */ + int plexno; /* or plex index */ + } volplex; + int error; /* current error indication */ + short isplex; /* set if this is a plex request */ + short active; /* number of subrequests still active */ + struct rqgroup *rqg; /* pointer to the first group of requests */ + struct rqgroup *lrqg; /* and to the first group of requests */ + struct request *next; /* link of waiting requests */ +}; + +/* Extended buffer header for subdisk I/O. Includes + * a pointer to the user I/O request. */ +struct sdbuf { + struct buf b; /* our buffer */ + struct buf *bp; /* and pointer to parent */ + short driveno; /* drive index */ + short sdno; /* and subdisk index */ +}; + +/* Values returned by rqe and friends. + * Be careful with these: they are in order of increasing + * seriousness. Some routines check for > REQUEST_RECOVERED + * to indicate a completely failed request. */ +enum requeststatus { + REQUEST_OK, /* request built OK */ + REQUEST_RECOVERED, /* request OK, but involves RAID5 recovery */ + REQUEST_EOF, /* request failed: outside plex */ + REQUEST_DOWN, /* request failed: subdisk down */ + REQUEST_ENOMEM /* ran out of memory */ +}; diff --git a/lkm/vinum/revive.c b/lkm/vinum/revive.c new file mode 100644 index 0000000..2f8117a --- /dev/null +++ b/lkm/vinum/revive.c @@ -0,0 +1,128 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: revive.c,v 1.1 1998/08/14 06:16:59 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" + +/* revive a block of a plex. Return an error + * indication. EAGAIN means successful copy, but + * that more blocks remain to be copied. + * XXX We should specify a block size here. At the moment, + * just take a default value. FIXME */ +int +revive_block(int plexno) +{ + struct plex *plex = &PLEX[plexno]; + struct buf *bp; + int error = EAGAIN; + int size; + int s; /* priority level */ + + if (plex->revive_blocksize == 0) { + if (plex->stripesize != 0) /* we're striped, don't revive more than */ + plex->revive_blocksize = min(DEFAULT_REVIVE_BLOCKSIZE, plex->stripesize); /* one block at a time */ + else + plex->revive_blocksize = DEFAULT_REVIVE_BLOCKSIZE; + } + size = min(plex->revive_blocksize, plex->length - plex->revived) << DEV_BSHIFT; + + s = splbio(); + /* Get a buffer */ + bp = geteblk(size); + if (bp == NULL) { + splx(s); + return ENOMEM; + } + if (bp->b_qindex != 0) /* on a queue, */ + bremfree(bp); /* remove it */ + splx(s); + + /* Amount to transfer: block size, unless it + * would overlap the end */ + bp->b_bufsize = size; + bp->b_bcount = bp->b_bufsize; + bp->b_resid = 0x0; + bp->b_blkno = plex->revived; /* we've got this far */ + + /* XXX what about reviving anonymous plexes? */ + + /* First, read the data from the volume. We don't + * care which plex, that's bre's job */ + bp->b_dev = VINUMBDEV(plex->volno, 0, 0, VINUM_VOLUME_TYPE); /* create the device number */ + bp->b_flags = B_BUSY | B_READ; + vinumstart(bp, 1); + biowait(bp); + if (bp->b_flags & B_ERROR) + error = bp->b_error; + else + /* Now write to the plex */ + { + s = splbio(); + if (bp->b_qindex != 0) /* on a queue, */ + bremfree(bp); /* remove it */ + splx(s); + bp->b_dev = VINUMBDEV(plex->volno, plex->volplexno, 0, VINUM_PLEX_TYPE); /* create the device number */ + + bp->b_flags = B_BUSY; /* make this a write */ + bp->b_resid = 0x0; + vinumstart(bp, 1); + biowait(bp); + if (bp->b_flags & B_ERROR) + error = bp->b_error; + else { + plex->revived += bp->b_bcount >> DEV_BSHIFT; /* moved this much further down */ + if (plex->revived >= plex->length) { /* finished */ + plex->revived = 0; + plex->state = plex_up; /* do we need to do more? */ + if (plex->volno >= 0) /* we have a volume, */ + set_volume_state(plex->volno, volume_up, 0); + printf("vinum: plex %s is %s\n", plex->name, plex_state(plex->state)); + save_config(); /* and save the updated configuration */ + error = 0; /* we're done */ + } + } + while (plex->waitlist) { /* we have waiting requests */ + launch_requests(plex->waitlist, 1); /* do them now */ + plex->waitlist = plex->waitlist->next; /* and move on to the next */ + } + } + if (bp->b_qindex == 0) /* not on a queue, */ + brelse(bp); /* is this kosher? */ + return error; +} diff --git a/lkm/vinum/state.c b/lkm/vinum/state.c new file mode 100644 index 0000000..2ce2ed0 --- /dev/null +++ b/lkm/vinum/state.c @@ -0,0 +1,755 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: state.c,v 2.6 1998/08/19 08:04:47 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" + +/* Update drive state */ +/* Return 1 if the state changes, otherwise 0 */ +int +set_drive_state(int driveno, enum drivestate state, int flags) +{ + struct drive *drive = &DRIVE[driveno]; + int oldstate = drive->state; + int sdno; + + if (drive->state == drive_unallocated) /* no drive to do anything with, */ + return 0; + + if (state != oldstate) { /* don't change it if it's not different */ + if (state == drive_down) { /* the drive's going down */ + if (flags || (drive->opencount == 0)) { /* we can do it */ + close_drive(drive); + drive->state = state; + printf("vinum: drive %s is %s\n", drive->label.name, drive_state(drive->state)); + } else + return 0; /* don't do it */ + } + drive->state = state; /* set the state */ + if (((drive->state == drive_up) + || ((drive->state == drive_coming_up))) + && (drive->vp == NULL)) /* should be open, but we're not */ + init_drive(drive); /* which changes the state again */ + if ((state != oldstate) /* state has changed */ + &&((flags & setstate_norecurse) == 0)) { /* and we want to recurse, */ + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) { /* find this drive's subdisks */ + if (SD[sdno].driveno == driveno) /* belongs to this drive */ + set_sd_state(sdno, sd_down, setstate_force | setstate_recursing); /* take it down */ + } + save_config(); /* and save the updated configuration */ + return 1; + } + } + return 0; +} + +/* Try to set the subdisk state. Return 1 if state changed to + * what we wanted, -1 if it changed to something else, and 0 + * if no change. + * + * This routine is called both from the user (up, down states + * only) and internally. + */ +int +set_sd_state(int sdno, enum sdstate state, enum setstateflags flags) +{ + struct sd *sd = &SD[sdno]; + int oldstate = sd->state; + int status = 1; /* status to return */ + + if (state == oldstate) + return 0; /* no change */ + + if (sd->state == sd_unallocated) /* no subdisk to do anything with, */ + return 0; + + if (sd->driveoffset < 0) { /* not allocated space */ + sd->state = sd_down; + if (state != sd_down) + return -1; + } else { /* space allocated */ + switch (state) { + case sd_down: + if ((!flags & setstate_force) /* but gently */ + &&(sd->plexno >= 0)) /* and we're attached to a plex, */ + return 0; /* don't do it */ + break; + + case sd_up: + if (DRIVE[sd->driveno].state != drive_up) /* can't bring the sd up if the drive isn't, */ + return 0; /* not even by force */ + switch (sd->state) { + case sd_obsolete: + case sd_down: /* been down, no data lost */ + if ((sd->plexno) /* we're associated with a plex */ + &&(((PLEX[sd->plexno].state < plex_firstup) /* and it's not up */ + ||(PLEX[sd->plexno].subdisks > 1)))) /* or it's the only one */ + break; + /* XXX Get this right: make sure that other plexes in + * the volume cover this address space, otherwise + * we make this one sd_up */ + sd->state = sd_reborn; /* here it is again */ + printf("vinum: subdisk %s is %s, not %s\n", sd->name, sd_state(sd->state), sd_state(state)); + status = -1; + break; + + case sd_init: /* brand new */ + if (flags & setstate_configuring) /* we're doing this while configuring */ + break; + sd->state = sd_empty; /* nothing in it */ + printf("vinum: subdisk %s is %s, not %s\n", sd->name, sd_state(sd->state), sd_state(state)); + status = -1; + break; + + case sd_initializing: + break; /* go on and do it */ + + case sd_empty: + if ((sd->plexno) /* we're associated with a plex */ + &&(((PLEX[sd->plexno].state < plex_firstup) /* and it's not up */ + ||(PLEX[sd->plexno].subdisks > 1)))) /* or it's the only one */ + break; + return 0; /* can't do it */ + + default: /* can't do it */ + /* There's no way to bring subdisks up directly from + * other states. First they need to be initialized + * or revived */ + return 0; + } + break; + + default: /* other ones, only internal with force */ + if (flags & setstate_force == 0) /* no force? What's this? */ + return 0; /* don't do it */ + } + } + sd->state = state; + printf("vinum: subdisk %s is %s\n", sd->name, sd_state(sd->state)); + if ((flags & setstate_norecurse) == 0) + set_plex_state(sd->plexno, plex_up, setstate_recursing); /* update plex state */ + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return status; +} + +/* Called from request routines when they find + * a subdisk which is not kosher. Decide whether + * it warrants changing the state. Return + * REQUEST_DOWN if we can't use the subdisk, + * REQUEST_OK if we can. */ +enum requeststatus +checksdstate(struct sd *sd, struct request *rq, daddr_t diskaddr, daddr_t diskend) +{ + struct plex *plex = &PLEX[sd->plexno]; + int writeop = (rq->bp->b_flags & B_READ) == 0; /* note if we're writing */ + + /* first, see if the plex wants to be accessed */ + switch (plex->state) { + case plex_reviving: + /* When writing, we'll write anything that starts + * up to the current revive pointer, but we'll + * only accept a read which finishes before the + * current revive pointer. + */ + if ((writeop && (diskaddr > plex->revived)) /* write starts after current revive pointer */ + ||((!writeop) && (diskend >= plex->revived))) { /* or read ends after current revive pointer */ + if (writeop) { /* writing to a consistent down disk */ + if (DRIVE[sd->driveno].state == drive_up) + set_sd_state(sd->sdno, sd_stale, setstate_force); /* it's not consistent now */ + else + set_sd_state(sd->sdno, sd_obsolete, setstate_force); /* it's not consistent now */ + } + return REQUEST_DOWN; /* that part of the plex is still down */ + } else if (diskend >= plex->revived) /* write finishes beyond revive pointer */ + rq->flags |= XFR_REVIVECONFLICT; /* note a potential conflict */ + /* FALLTHROUGH */ + + case plex_up: + case plex_degraded: + case plex_flaky: + /* We can access the plex: let's see + * how the subdisk feels */ + switch (sd->state) { + case sd_up: + return REQUEST_OK; + + case sd_reborn: + if (writeop) + return REQUEST_OK; /* always write to a reborn disk */ + /* Handle the mapping. We don't want to reject + * a read request to a reborn subdisk if that's + * all we have. XXX */ + return REQUEST_DOWN; + + case sd_down: + case sd_crashed: + if (writeop) { /* writing to a consistent down disk */ + if (DRIVE[sd->driveno].state == drive_up) + set_sd_state(sd->sdno, sd_stale, setstate_force); /* it's not consistent now */ + else + set_sd_state(sd->sdno, sd_obsolete, setstate_force); /* it's not consistent now */ + } + return REQUEST_DOWN; /* and it's down one way or another */ + + default: + return REQUEST_DOWN; + } + + default: + return REQUEST_DOWN; + } +} + +void +add_defective_region(struct plex *plex, off_t offset, size_t length) +{ +/* XXX get this ordered, and coalesce regions if necessary */ + if (++plex->defective_regions > plex->defective_region_count) + EXPAND(plex->defective_region, + struct plexregion, + plex->defective_region_count, + PLEX_REGION_TABLE_SIZE); + plex->defective_region[plex->defective_regions - 1].offset = offset; + plex->defective_region[plex->defective_regions - 1].length = length; +} + +void +add_unmapped_region(struct plex *plex, off_t offset, size_t length) +{ + if (++plex->unmapped_regions > plex->unmapped_region_count) + EXPAND(plex->unmapped_region, + struct plexregion, + plex->unmapped_region_count, + PLEX_REGION_TABLE_SIZE); + plex->unmapped_region[plex->unmapped_regions - 1].offset = offset; + plex->unmapped_region[plex->unmapped_regions - 1].length = length; +} + +/* Rebuild a plex free list and set state if + * we have a configuration error */ +void +rebuild_plex_unmappedlist(struct plex *plex) +{ + int sdno; + struct sd *sd; + int lastsdend = 0; /* end offset of last subdisk */ + + if (plex->unmapped_region != NULL) { /* we're going to rebuild it */ + Free(plex->unmapped_region); + plex->unmapped_region = NULL; + plex->unmapped_regions = 0; + plex->unmapped_region_count = 0; + } + if (plex->defective_region != NULL) { + Free(plex->defective_region); + plex->defective_region = NULL; + plex->defective_regions = 0; + plex->defective_region_count = 0; + } + for (sdno = 0; sdno < plex->subdisks; sdno++) { + sd = &SD[plex->sdnos[sdno]]; + if (sd->plexoffset < lastsdend) { /* overlap */ + printf("vinum: Plex %s, subdisk %s overlaps previous\n", plex->name, sd->name); + set_plex_state(plex->plexno, plex_down, setstate_force); /* don't allow that */ + } else if (sd->plexoffset > lastsdend) /* gap */ + add_unmapped_region(plex, lastsdend, sd->plexoffset - lastsdend); + else if (sd->state < sd_reborn) /* this part defective */ + add_defective_region(plex, sd->plexoffset, sd->sectors); + lastsdend = sd->plexoffset + sd->sectors; + } +} + +/* return a state map for the subdisks of a plex */ +enum sdstates +sdstatemap(struct plex *plex, int *sddowncount) +{ + int sdno; + enum sdstates statemap = 0; /* note the states we find */ + + *sddowncount = 0; /* no subdisks down yet */ + for (sdno = 0; sdno < plex->subdisks; sdno++) { + struct sd *sd = &SD[plex->sdnos[sdno]]; /* point to the subdisk */ + + switch (sd->state) { + case sd_empty: + statemap |= sd_emptystate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_init: + statemap |= sd_initstate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_down: + statemap |= sd_downstate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_crashed: + statemap |= sd_crashedstate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_obsolete: + statemap |= sd_obsolete; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_stale: + statemap |= sd_stalestate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_reborn: + statemap |= sd_rebornstate; + break; + + case sd_up: + statemap |= sd_upstate; + break; + + default: + statemap |= sd_otherstate; + break; + } + } + return statemap; +} + +/* determine the state of the volume relative to this plex */ +enum volplexstate +vpstate(struct plex *plex) +{ + struct volume *vol; + enum volplexstate state = volplex_onlyusdown; /* state to return */ + int plexno; + + if (plex->volno < 0) /* not associated with a volume */ + return volplex_onlyusdown; /* assume the worst */ + + vol = &VOL[plex->volno]; /* point to our volume */ + for (plexno = 0; plexno < vol->plexes; plexno++) { + if (&PLEX[vol->plex[plexno]] == plex) { /* us */ + if (PLEX[vol->plex[plexno]].state == plex_up) /* are we up? */ + state |= volplex_onlyus; /* yes */ + } else { + if (PLEX[vol->plex[plexno]].state == plex_up) /* not us */ + state |= volplex_otherup; /* and when they were up, they were up */ + else + state |= volplex_alldown; /* and when they were down, they were down */ + } + } + return state; /* and when they were only halfway up */ +} /* they were neither up nor down */ + +/* Check if all bits b are set in a */ +int allset(int a, int b); + +int +allset(int a, int b) +{ + return (a & b) == b; +} + +/* Update the state of a plex dependent on its subdisks. + * Also rebuild the unmapped_region and defective_region table */ +int +set_plex_state(int plexno, enum plexstate state, enum setstateflags flags) +{ + int sddowncount = 0; /* number of down subdisks */ + struct plex *plex = &PLEX[plexno]; /* point to our plex */ + enum plexstate oldstate = plex->state; + enum volplexstate vps = vpstate(plex); /* how do we compare with the other plexes? */ + enum sdstates statemap = sdstatemap(plex, &sddowncount); /* get a map of the subdisk states */ + + if ((flags & setstate_force) && (oldstate == state)) /* we're there already, */ + return 0; /* no change */ + + if (plex->state == plex_unallocated) /* no plex to do anything with, */ + return 0; + + switch (state) { + case plex_up: + if ((plex->state == plex_initializing) /* we're initializing */ + &&(statemap != sd_upstate)) /* but SDs aren't up yet */ + return 0; /* do nothing */ + + /* We don't really care what our state was before + * if we want to come up. We rely entirely on the + * state of our subdisks and our volume */ + switch (vps) { + case volplex_onlyusdown: + case volplex_alldown: /* another plex is down, and so are we */ + if (statemap == sd_upstate) { /* all subdisks ready for action */ + if ((plex->state == plex_init) /* we're brand spanking new */ + &&(VOL[plex->volno].flags & VF_CONFIG_SETUPSTATE)) { /* and we consider that up */ + /* Conceptually, an empty plex does not contain valid data, + * but normally we'll see this state when we have just + * created a plex, and it's either consistent from earlier, + * or we don't care about the previous contents (we're going + * to create a file system or use it for swap). + * + * We need to do this in one swell foop: on the next call + * we will no longer be just empty. + * + * We'll still come back to this function for the remaining + * plexes in the volume. They'll be up already, so that + * doesn't change anything, but it's not worth the additional + * code to stop doing it. */ + struct volume *vol = &VOL[plex->volno]; + int plexno; + + for (plexno = 0; plexno < vol->plexes; plexno++) + PLEX[vol->plex[plexno]].state = plex_up; + } + plex->state = plex_up; /* bring up up, anyway */ + } else + plex->state = plex_down; + break; + + case volplex_onlyusup: /* only we are up: others are down */ + case volplex_onlyus: /* we're up and alone */ + if ((statemap == sd_upstate) /* subdisks all up */ + ||(statemap == sd_emptystate)) /* or all empty */ + plex->state = plex_up; /* go for it */ + else if ((statemap & (sd_upstate | sd_reborn)) == statemap) /* all up or reborn, */ + plex->state = plex_flaky; + else if (statemap & (sd_upstate | sd_reborn)) /* some up or reborn, */ + plex->state = plex_degraded; /* so far no corruption */ + else + plex->state = plex_faulty; + break; + + case volplex_otherup: /* another plex is up */ + case volplex_otherupdown: /* other plexes are up and down */ + if ((statemap == sd_upstate) /* subdisks all up */ + ||(statemap == sd_emptystate) /* or all empty */ + ) { + /* Is the data in all subdisks valid? */ + if (statemap == statemap & (sd_downstate | sd_rebornstate | sd_upstate)) + break; /* yes, we can bring the plex up */ + plex->state = plex_reviving; /* we need reviving */ + return EAGAIN; + } else + plex->state = plex_faulty; /* still in error */ + break; + + case volplex_allup: /* all plexes are up */ + case volplex_someup: + if ((statemap & (sd_upstate | sd_reborn)) == statemap) /* all up or reborn, */ + break; /* no change */ + else + plex->state = plex_degraded; /* we're not all there */ + } + + if (plex->state != oldstate) + break; + return 0; /* no change */ + + case plex_down: /* want to take it down */ + if (((vps == volplex_onlyus) /* we're the only one up */ + ||(vps == volplex_onlyusup)) /* we're the only one up */ + &&(!(flags & setstate_force))) /* and we don't want to use force */ + return 0; /* can't do it */ + plex->state = state; /* do it */ + break; + + /* This is only requested by the driver. + * Trust ourselves */ + case plex_faulty: + plex->state = state; /* do it */ + break; + + case plex_initializing: + /* XXX consider what safeguards we need here */ + if ((flags & setstate_force) == 0) + return 0; + plex->state = state; /* do it */ + break; + + /* What's this? */ + default: + return 0; + } + printf("vinum: plex %s is %s\n", plex->name, plex_state(plex->state)); + /* Now see what we have left, and whether + * we're taking the volume down */ + if (plex->volno >= 0) { /* we have a volume */ + struct volume *vol = &VOL[plex->volno]; + + vps = vpstate(plex); /* get our combined state again */ + if ((flags & setstate_norecurse) == 0) { /* we can recurse */ + if ((vol->state == volume_up) + && (vps == volplex_alldown)) /* and we're all down */ + set_volume_state(plex->volno, volume_down, setstate_recursing); /* take our volume down */ + else if ((vol->state == volume_down) + && (vps & (volplex_otherup | volplex_onlyusup))) /* and at least one is up */ + set_volume_state(plex->volno, volume_up, setstate_recursing); /* bring our volume up */ + } + } + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return 1; +} + +/* Update the state of a plex dependent on its plexes. + * Also rebuild the unmapped_region and defective_region table */ +int +set_volume_state(int volno, enum volumestate state, enum setstateflags flags) +{ + int plexno; + enum plexstates { + plex_downstate = 1, /* found a plex which is down */ + plex_degradedstate = 2, /* found a plex which is halfway up */ + plex_upstate = 4 /* found a plex which is completely up */ + }; + + int plexstatemap = 0; /* note the states we find */ + struct volume *vol = &VOL[volno]; /* point to our volume */ + + if (vol->state == state) /* we're there already */ + return 0; /* no change */ + if (vol->state == volume_unallocated) /* no volume to do anything with, */ + return 0; + + for (plexno = 0; plexno < vol->plexes; plexno++) { + struct plex *plex = &PLEX[vol->plex[plexno]]; /* point to the plex */ + switch (plex->state) { + case plex_degraded: + case plex_flaky: + case plex_reviving: + plexstatemap |= plex_degradedstate; + break; + + case plex_up: + plexstatemap |= plex_upstate; + break; + + default: + plexstatemap |= plex_downstate; + break; + } + } + + if (state == volume_up) { /* want to come up */ + if (plexstatemap & plex_upstate) { /* we have a plex which is completely up */ + vol->state = volume_up; /* did it */ + printf("vinum: volume %s is %s\n", vol->name, volume_state(vol->state)); + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return 1; + } + /* Here we should check whether we have enough + * coverage for the complete volume. Writeme XXX */ + } else if (state == volume_down) { /* want to go down */ + if ((vol->opencount == 0) /* not open */ + ||(flags & setstate_force != 0)) { /* or we're forcing */ + vol->state = volume_down; + printf("vinum: volume %s is %s\n", vol->name, volume_state(vol->state)); + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return 1; + } + } + return 0; /* no change */ +} + +/* Start an object, in other words do what we can to get it up. + * This is called from vinumioctl (VINUMSTART). + * Return error indications via ioctl_reply + */ +void +start_object(struct vinum_ioctl_msg *data) +{ + int status; + int realstatus; /* what we really have */ + int objindex = data->index; /* data gets overwritten */ + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) data; /* format for returning replies */ + + switch (data->type) { + case drive_object: + status = set_drive_state(objindex, drive_up, setstate_none); + realstatus = DRIVE[objindex].state == drive_up; /* set status on whether we really did it */ + break; + + case sd_object: + status = set_sd_state(objindex, sd_up, setstate_none); /* set state */ + realstatus = SD[objindex].state == sd_up; /* set status on whether we really did it */ + break; + + case plex_object: + if (PLEX[objindex].state == plex_reviving) { /* reviving, */ + ioctl_reply->error = revive_block(objindex); /* revive another block */ + ioctl_reply->msg[0] = '\0'; /* no comment */ + return; + } + status = set_plex_state(objindex, plex_up, setstate_none); + realstatus = PLEX[objindex].state == plex_up; /* set status on whether we really did it */ + break; + + case volume_object: + status = set_volume_state(objindex, volume_up, setstate_none); + realstatus = VOL[objindex].state == volume_up; /* set status on whether we really did it */ + break; + + default: + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "Invalid object type"); + return; + } + /* There's no point in saying anything here: + * the userland program does it better */ + ioctl_reply->msg[0] = '\0'; + if (realstatus == 0) /* couldn't do it */ + ioctl_reply->error = EINVAL; + else + ioctl_reply->error = 0; +} + +/* Stop an object, in other words do what we can to get it down + * This is called from vinumioctl (VINUMSTOP). + * Return error indications via ioctl_reply. + */ +void +stop_object(struct vinum_ioctl_msg *data) +{ + int status = 1; + int objindex = data->index; /* save the number from change */ + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) data; /* format for returning replies */ + + switch (data->type) { + case drive_object: + status = set_drive_state(objindex, drive_down, data->force); + break; + + case sd_object: + status = set_sd_state(objindex, sd_down, data->force); + break; + + case plex_object: + status = set_plex_state(objindex, plex_down, data->force); + break; + + case volume_object: + status = set_volume_state(objindex, volume_down, data->force); + break; + + default: + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "Invalid object type"); + return; + } + ioctl_reply->msg[0] = '\0'; + if (status == 0) /* couldn't do it */ + ioctl_reply->error = EINVAL; + else + ioctl_reply->error = 0; +} + +/* VINUM_SETSTATE ioctl: set an object state + * msg is the message passed by the user */ +void +setstate(struct vinum_ioctl_msg *msg) +{ + int sdno; + struct sd *sd; + struct plex *plex; + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) msg; /* format for returning replies */ + + switch (msg->state) { + case object_down: + stop_object(msg); + break; + + case object_initializing: + switch (msg->type) { + case sd_object: + sd = &SD[msg->index]; + if ((msg->index >= vinum_conf.subdisks_used) + || (sd->state == sd_unallocated)) { + sprintf(ioctl_reply->msg, "Invalid subdisk %d", msg->index); + ioctl_reply->error = EFAULT; + return; + } + set_sd_state(msg->index, sd_initializing, msg->force); + if (sd->state != sd_initializing) { + strcpy(ioctl_reply->msg, "Can't set state"); + ioctl_reply->error = EINVAL; + } else + ioctl_reply->error = 0; + break; + + case plex_object: + plex = &PLEX[msg->index]; + if ((msg->index >= vinum_conf.plexes_used) + || (plex->state == plex_unallocated)) { + sprintf(ioctl_reply->msg, "Invalid subdisk %d", msg->index); + ioctl_reply->error = EFAULT; + return; + } + set_plex_state(msg->index, plex_initializing, msg->force); + if (plex->state != plex_initializing) { + strcpy(ioctl_reply->msg, "Can't set state"); + ioctl_reply->error = EINVAL; + } else { + ioctl_reply->error = 0; + for (sdno = 0; sdno < plex->subdisks; sdno++) { + sd = &SD[plex->sdnos[sdno]]; + set_sd_state(plex->sdnos[sdno], sd_initializing, msg->force); + if (sd->state != sd_initializing) { + strcpy(ioctl_reply->msg, "Can't set state"); + ioctl_reply->error = EINVAL; + break; + } + } + } + break; + + default: + strcpy(ioctl_reply->msg, "Invalid object"); + ioctl_reply->error = EINVAL; + } + break; + + case object_up: + start_object(msg); + } +} diff --git a/lkm/vinum/statetexts.h b/lkm/vinum/statetexts.h new file mode 100644 index 0000000..b7ce4eb --- /dev/null +++ b/lkm/vinum/statetexts.h @@ -0,0 +1,88 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: COPYRIGHT,v 1.1 1998/03/05 06:07:05 grog Exp grog $ + */ +/* Created by ./makestatetext on Tue 4 Aug 15:53:16 CST 1998. Do not edit */ + +/* Drive state texts */ +char *drivestatetext[] = +{ + "unallocated", + "uninit", + "down", + "coming_up", + "up", +}; + +/* Subdisk state texts */ +char *sdstatetext[] = +{ + "unallocated", + "uninit", + "init", + "initializing", + "empty", + "obsolete", + "stale", + "crashed", + "down", + "reborn", + "up", +}; + +/* Plex state texts */ +char *plexstatetext[] = +{ + "unallocated", + "init", + "faulty", + "down", + "reviving", + "initializing", + "corrupt", + "degraded", + "flaky", + "up", +}; + +/* Volume state texts */ +char *volstatetext[] = +{ + "unallocated", + "uninit", + "down", + "up", +}; diff --git a/lkm/vinum/util.c b/lkm/vinum/util.c new file mode 100644 index 0000000..5121404 --- /dev/null +++ b/lkm/vinum/util.c @@ -0,0 +1,211 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: util.c,v 1.7 1998/08/07 09:23:10 grog Exp grog $ + */ + +/* This file contains utility routines used both in kernel and user context */ + +#include "vinumhdr.h" +#include "statetexts.h" +#ifndef REALLYKERNEL +#include <stdio.h> +extern jmp_buf command_fail; /* return on a failed command */ +#endif + +static char numeric_state[32]; /* temporary buffer for ASCII conversions */ +#define STATECOUNT(x) (sizeof (x##statetext) / sizeof (char *)) +/* Return drive state as a string */ +char * +drive_state(enum drivestate state) +{ + if (((unsigned) state) >= STATECOUNT(drive)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return drivestatetext[state]; +} + +/* Return volume state as a string */ +char * +volume_state(enum volumestate state) +{ + if (((unsigned) state) >= STATECOUNT(vol)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return volstatetext[state]; +} + +/* Return plex state as a string */ +char * +plex_state(enum plexstate state) +{ + if (((unsigned) state) >= STATECOUNT(plex)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return plexstatetext[state]; +} + +/* Return plex organization as a string */ +char * +plex_org(enum plexorg org) +{ + switch (org) { + case plex_disorg: /* disorganized */ + return "disorg"; + break; + + case plex_concat: /* concatenated plex */ + return "concat"; + break; + + case plex_striped: /* striped plex */ + return "striped"; + break; + + + default: + sprintf(numeric_state, "Invalid org %d", (int) org); + return numeric_state; + } +} + +/* Return sd state as a string */ +char * +sd_state(enum sdstate state) +{ + if (((unsigned) state) >= STATECOUNT(sd)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return sdstatetext[state]; +} + +/* Now convert in the other direction */ +/* These are currently used only internally, + * so we don't do too much error checking */ +enum drivestate +DriveState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(drive); i++) + if (strcmp(text, drivestatetext[i]) == 0) /* found it */ + return (enum drivestate) i; + return -1; +} + +enum sdstate +SdState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(sd); i++) + if (strcmp(text, sdstatetext[i]) == 0) /* found it */ + return (enum sdstate) i; + return -1; +} + +enum plexstate +PlexState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(plex); i++) + if (strcmp(text, plexstatetext[i]) == 0) /* found it */ + return (enum plexstate) i; + return -1; +} + +enum volumestate +VolState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(vol); i++) + if (strcmp(text, volstatetext[i]) == 0) /* found it */ + return (enum volstate) i; + return -1; +} + +/* Take a number with an optional scale factor and convert + * it to a number of bytes. + * + * The scale factors are: + * + * b blocks (of 512 bytes) + * k kilobytes (1024 bytes) + * m megabytes (of 1024 * 1024 bytes) + * g gigabytes (of 1024 * 1024 * 1024 bytes) + */ +u_int64_t +sizespec(char *spec) +{ + u_int64_t size; + char *s; + + size = 0; + s = spec; + if ((*s >= '0') && (*s <= '9')) { /* it's numeric */ + while ((*s >= '0') && (*s <= '9')) /* it's numeric */ + size = size * 10 + *s++ - '0'; /* convert it */ + switch (*s) { + case '\0': + return size; + + case 'B': + case 'b': + return size * 512; + + case 'K': + case 'k': + return size * 1024; + + case 'M': + case 'm': + return size * 1024 * 1024; + + case 'G': + case 'g': + return size * 1024 * 1024 * 1024; + } + } +#ifdef REALLYKERNEL + throw_rude_remark(EINVAL, "Invalid length specification: %s", spec); +#else + fprintf(stderr, "Invalid length specification: %s", spec); + longjmp(command_fail, -1); +#endif + /* NOTREACHED */ + return -1; +} diff --git a/lkm/vinum/vinum.c b/lkm/vinum/vinum.c new file mode 100644 index 0000000..bffdf6d --- /dev/null +++ b/lkm/vinum/vinum.c @@ -0,0 +1,512 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinum.c,v 1.19 1998/08/13 05:24:02 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "sys/sysproto.h" /* for sync(2) */ +#ifdef DEBUG +#include <sys/reboot.h> +int debug = 0; +#endif + +/* pointer to ioctl p parameter, to save passing it around */ +struct proc *myproc; + +#if __FreeBSD__ < 3 +STATIC struct cdevsw vinum_cdevsw; +STATIC struct bdevsw vinum_bdevsw = +{ + vinumopen, vinumclose, vinumstrategy, vinumioctl, + vinumdump, vinumsize, 0, + "vinum", &vinum_cdevsw, -1 +}; +#else /* goodbye, bdevsw */ +STATIC struct cdevsw vinum_cdevsw = +{ + vinumopen, vinumclose, vinumread, vinumwrite, + vinumioctl, nostop, nullreset, nodevtotty, + seltrue, nommap, vinumstrategy, "vinum", + NULL, -1, vinumdump, vinumsize, + D_DISK, 0, -1 +}; +#endif + +/* Called by main() during pseudo-device attachment. */ +STATIC void vinumattach(void *); + +STATIC void vinumgetdisklabel(dev_t); +void vinum_scandisk(void); +int vinum_inactive(void); +void free_vinum(int); + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +extern jmp_buf command_fail; /* return here if config fails */ + +struct _vinum_conf vinum_conf; /* configuration information */ + +STATIC int vinum_devsw_installed = 0; + +/* + * Called by main() during pseudo-device attachment. All we need + * to do is allocate enough space for devices to be configured later, and + * add devsw entries. + */ +void +vinumattach(void *dummy) +{ + BROKEN_GDB; + char *buf; /* pointer to temporary buffer */ + struct _ioctl_reply *ioctl_reply; /* struct to return */ + struct uio uio; + struct iovec iovec; + + /* modload should prevent multiple loads, so this is worth a panic */ + if ((vinum_conf.flags & VF_LOADED) != NULL) + panic("vinum: already loaded"); + + printf("vinum: loaded\n"); + vinum_conf.flags |= VF_LOADED; /* we're loaded now */ + + /* We don't have a p pointer here, so take it from curproc */ + myproc = curproc; +#if __FreeBSD__ < 3 + bdevsw_add_generic(BDEV_MAJOR, CDEV_MAJOR, &vinum_bdevsw); +#else + cdevsw_add_generic(BDEV_MAJOR, CDEV_MAJOR, &vinum_cdevsw); +#endif +#ifdef DEVFS +#error DEVFS not finished yet +#endif + + uio.uio_iov = &iovec; + uio.uio_iovcnt = 1; /* just one buffer */ + uio.uio_offset = 0; /* start at the beginning */ + uio.uio_resid = 512; /* one sector */ + uio.uio_segflg = UIO_SYSSPACE; /* we're in system space */ + uio.uio_rw = UIO_READ; /* do we need this? */ + uio.uio_procp = curproc; /* do it for our own process */ + + iovec.iov_len = 512; + buf = (char *) Malloc(iovec.iov_len); /* get a buffer */ + CHECKALLOC(buf, "vinum: no memory\n"); /* can't get 512 bytes? */ + iovec.iov_base = buf; /* read into buf */ + + /* allocate space: drives... */ + DRIVE = (struct drive *) Malloc(sizeof(struct drive) * INITIAL_DRIVES); + CHECKALLOC(DRIVE, "vinum: no memory\n"); + vinum_conf.drives_allocated = INITIAL_DRIVES; /* number of drive slots allocated */ + vinum_conf.drives_used = 0; /* and number in use */ + + /* volumes, ... */ + VOL = (struct volume *) Malloc(sizeof(struct volume) * INITIAL_VOLUMES); + CHECKALLOC(VOL, "vinum: no memory\n"); + vinum_conf.volumes_allocated = INITIAL_VOLUMES; /* number of volume slots allocated */ + vinum_conf.volumes_used = 0; /* and number in use */ + + /* plexes, ... */ + PLEX = (struct plex *) Malloc(sizeof(struct plex) * INITIAL_PLEXES); + CHECKALLOC(PLEX, "vinum: no memory\n"); + vinum_conf.plexes_allocated = INITIAL_PLEXES; /* number of plex slots allocated */ + vinum_conf.plexes_used = 0; /* and number in use */ + + /* and subdisks */ + SD = (struct sd *) Malloc(sizeof(struct sd) * INITIAL_SUBDISKS); + CHECKALLOC(SD, "vinum: no memory\n"); + vinum_conf.subdisks_allocated = INITIAL_SUBDISKS; /* number of sd slots allocated */ + vinum_conf.subdisks_used = 0; /* and number in use */ + + ioctl_reply = NULL; /* no reply on longjmp */ +} + + +#ifdef ACTUALLY_LKM_NOT_KERNEL /* stuff for LKMs */ + +/* Check if we have anything open. If so, return 0 (not inactive), + * otherwise 1 (inactive) */ +int +vinum_inactive(void) +{ + BROKEN_GDB; + int i; + int can_do = 1; /* assume we can do it */ + + lock_config(); + for (i = 0; i < vinum_conf.volumes_used; i++) { + if (VOL[i].pid != NULL) { /* volume is open */ + can_do = 0; + break; + } + } + unlock_config(); + return can_do; +} + +/* Free all structures. + * If cleardrive is 0, save the configuration; otherwise + * remove the configuration from the drive. + * + * Before coming here, ensure that no volumes are open. + */ +void +free_vinum(int cleardrive) +{ + BROKEN_GDB; + int i; + + if (cleardrive) { + for (i = 0; i < vinum_conf.drives_used; i++) + remove_drive(i); /* remove the drive */ + } else { /* keep the config */ + save_config(); + if (DRIVE != NULL) { + for (i = 0; i < vinum_conf.drives_used; i++) + free_drive(&DRIVE[i]); /* close files and things */ + Free(DRIVE); + } + } + if (SD != NULL) + Free(SD); + if (PLEX != NULL) { + for (i = 0; i < vinum_conf.plexes_used; i++) { + struct plex *plex = &vinum_conf.plex[i]; + + if (plex->state != plex_unallocated) { /* we have real data there */ + if (plex->sdnos) + Free(plex->sdnos); + if (plex->unmapped_regions) + Free(plex->unmapped_region); + if (plex->defective_regions) + Free(plex->defective_region); + } + } + Free(PLEX); + } + if (VOL != NULL) + Free(VOL); + bzero(&vinum_conf, sizeof(vinum_conf)); +} + +MOD_MISC(vinum); + +/* + * Function called when loading the driver. + */ +STATIC int +vinum_load(struct lkm_table *lkmtp, int cmd) +{ + BROKEN_GDB; +/* Debugger ("vinum_load"); */ + vinumattach(NULL); + return 0; /* OK */ +} + +/* + * Function called when unloading the driver. + */ +STATIC int +vinum_unload(struct lkm_table *lkmtp, int cmd) +{ + BROKEN_GDB; + if (vinum_inactive()) { /* is anything open? */ + struct sync_args dummyarg = + {0}; +#if __FreeBSD__ < 3 + int retval; +#endif + + printf("vinum: unloaded\n"); +#if __FreeBSD__ < 3 + sync(curproc, &dummyarg, &retval); /* write out buffers */ +#else + sync(curproc, &dummyarg); /* write out buffers */ +#endif + free_vinum(0); /* no: clean up */ +#if __FreeBSD__ < 3 + bdevsw[BDEV_MAJOR] = NULL; /* clear bdevsw */ +#endif + cdevsw[CDEV_MAJOR] = NULL; /* and cdevsw */ + return 0; + } else + return EBUSY; +} + +/* + * Dispatcher function for the module (load/unload/stat). + */ +int +vinum_mod(struct lkm_table *lkmtp, int cmd, int ver) +{ + BROKEN_GDB; + MOD_DISPATCH(vinum, /* module name */ + lkmtp, /* LKM table */ + cmd, /* command */ + ver, + vinum_load, /* load with this function */ + vinum_unload, /* and unload with this */ + lkm_nullcmd); +} + +#else /* not LKM */ +#error "This driver must be compiled as a loadable kernel module" +#endif /* LKM */ + +/* ARGSUSED */ +/* Open a vinum object + * At the moment, we only open volumes and the + * super device. It's a nice concept to be + * able to open drives, subdisks and plexes, but + * I can't think what good it could be */ +int +vinumopen(dev_t dev, + int flags, + int fmt, + struct proc *p) +{ + BROKEN_GDB; + int s; /* spl */ + int error; + unsigned int index; + struct volume *vol; + struct plex *plex; + struct sd *sd; + struct devcode *device; + + device = (struct devcode *) &dev; + + error = 0; + /* First, decide what we're looking at */ + switch (device->type) { + case VINUM_VOLUME_TYPE: + index = VOLNO(dev); + if (index >= vinum_conf.volumes_used) + return ENXIO; /* no such device */ + vol = &VOL[index]; + + switch (vol->state) { + case volume_unallocated: + case volume_uninit: + return ENXIO; + + case volume_up: + s = splhigh(); /* quick lock */ + if (error) + return error; + if (vol->opencount == 0) + vol->openflags = flags; /* set our flags */ + vol->opencount++; + vol->pid = p->p_pid; /* and say who we are (do we need this? XXX) */ + splx(s); + return 0; + + case volume_down: + return EIO; + + default: + return EINVAL; + } + + case VINUM_PLEX_TYPE: + if (VOLNO(dev) >= vinum_conf.volumes_used) + return ENXIO; + index = PLEXNO(dev); /* get plex index in vinum_conf */ + if (index >= vinum_conf.plexes_used) + return ENXIO; /* no such device */ + plex = &PLEX[index]; + + switch (plex->state) { + case plex_unallocated: + return EINVAL; + + default: + s = splhigh(); + if (plex->pid /* it's open already */ + && (plex->pid != p->p_pid)) { /* and not by us, */ + splx(s); + return EBUSY; /* one at a time, please */ + } + plex->pid = p->p_pid; /* and say who we are (do we need this? XXX) */ + splx(s); + return 0; + } + + case VINUM_SD_TYPE: + if ((VOLNO(dev) >= vinum_conf.volumes_used) || /* no such volume */ + (PLEXNO(dev) >= vinum_conf.plexes_used)) /* or no such plex */ + return ENXIO; /* no such device */ + index = SDNO(dev); /* get the subdisk number */ + if (index >= vinum_conf.subdisks_used) + return ENXIO; /* no such device */ + sd = &SD[index]; + + /* Opening a subdisk is always a special operation, so we + * ignore the state as long as it represents a real subdisk */ + switch (sd->state) { + case sd_unallocated: + case sd_uninit: + return EINVAL; + + default: + s = splhigh(); + if (sd->pid /* it's open already */ + && (sd->pid != p->p_pid)) { /* and not by us, */ + splx(s); + return EBUSY; /* one at a time, please */ + } + sd->pid = p->p_pid; /* and say who we are (do we need this? XXX) */ + splx(s); + return 0; + } + + case VINUM_DRIVE_TYPE: + default: + return ENODEV; /* don't know what to do with these */ + + case VINUM_SUPERDEV_TYPE: + if (p->p_ucred->cr_uid == 0) { /* root calling, */ + vinum_conf.opencount++; /* one more opener */ + return 0; /* no worries opening super dev */ + } else + return EPERM; /* you can't do that! */ + } +} + +/* ARGSUSED */ +int +vinumclose(dev_t dev, + int flags, + int fmt, + struct proc *p) +{ + BROKEN_GDB; + unsigned int index; + struct volume *vol; + struct plex *plex; + struct sd *sd; + struct devcode *device = (struct devcode *) &dev; + + index = VOLNO(dev); + /* First, decide what we're looking at */ + switch (device->type) { + case VINUM_VOLUME_TYPE: + if (index >= vinum_conf.volumes_used) + return ENXIO; /* no such device */ + vol = &VOL[index]; + + switch (vol->state) { + case volume_unallocated: + case volume_uninit: + return ENXIO; + + case volume_up: + vol->opencount = 0; /* reset our flags */ + vol->pid = NULL; /* and forget who owned us */ + return 0; + + case volume_down: + return EIO; + + default: + return EINVAL; + } + + case VINUM_PLEX_TYPE: + if (VOLNO(dev) >= vinum_conf.volumes_used) + return ENXIO; + index = PLEXNO(dev); /* get plex index in vinum_conf */ + if (index >= vinum_conf.plexes_used) + return ENXIO; /* no such device */ + plex = &PLEX[index]; + plex->pid = 0; + return 0; + + case VINUM_SD_TYPE: + if ((VOLNO(dev) >= vinum_conf.volumes_used) || /* no such volume */ + (PLEXNO(dev) >= vinum_conf.plexes_used)) /* or no such plex */ + return ENXIO; /* no such device */ + index = SDNO(dev); /* get the subdisk number */ + if (index >= vinum_conf.subdisks_used) + return ENXIO; /* no such device */ + sd = &SD[index]; + sd->pid = 0; + return 0; + + case VINUM_SUPERDEV_TYPE: + if (p->p_ucred->cr_uid == 0) /* root calling, */ + vinum_conf.opencount--; /* one less opener */ + return 0; /* no worries closing super dev */ + + case VINUM_DRIVE_TYPE: + default: + return ENODEV; /* don't know what to do with these */ + } +} + +/* size routine */ +int +vinumsize(dev_t dev) +{ + BROKEN_GDB; + struct volume *vol; + int size; + +/* XXX This is bogus. We don't need to open + * a device to find its size */ + vol = &VOL[VOLNO(dev)]; + + if (vol->state == volume_up) + size = vol->size; + else + return 0; /* err on the size of conservatism */ + + + return size; +} + +int +vinumdump(dev_t dev) +{ + /* Not implemented. */ + return ENXIO; +} diff --git a/lkm/vinum/vinumext.h b/lkm/vinum/vinumext.h new file mode 100644 index 0000000..4b7d5a0 --- /dev/null +++ b/lkm/vinum/vinumext.h @@ -0,0 +1,214 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumext.h,v 1.14 1998/08/11 00:03:57 grog Exp grog $ + */ + +/* vinumext.h: external definitions */ + +extern struct _vinum_conf vinum_conf; /* configuration information */ + +#ifdef DEBUG +extern debug; /* debug flags */ +#endif + +#define CHECKALLOC(ptr, msg) \ + if (ptr == NULL) \ + { \ + printf (msg); \ + longjmp (command_fail, -1); \ + } +#ifndef KERNEL +struct vnode; +struct proc; +#endif + +#ifdef KERNEL +int give_sd_to_plex(int plexno, int sdno); +int give_plex_to_volume(int volno, int plexno); +int check_drive(char *); +enum drive_label_info read_drive_label(struct drive *drive); +int parse_config(char *, struct keywordset *); +int parse_user_config(char *cptr, struct keywordset *keyset); +u_int64_t sizespec(char *spec); +int volume_index(struct volume *volume); +int plex_index(struct plex *plex); +int sd_index(struct sd *sd); +int drive_index(struct drive *drive); +int my_plex(int volno, int plexno); +int my_sd(int plexno, int sdno); +int get_empty_drive(void); +int find_drive(const char *name, int create); +int find_drive_by_dev(const char *devname, int create); +int get_empty_sd(void); +int find_subdisk(const char *name, int create); +void free_sd(int sdno); +void free_volume(int volno); +int get_empty_plex(void); +int find_plex(const char *name, int create); +void free_plex(int plexno); +int get_empty_volume(void); +int find_volume(const char *name, int create); +void config_subdisk(void); +void config_plex(void); +void config_volume(void); +void config_drive(void); +void updateconfig(int); +void update_sd_config(int sdno, int kernelstate); +void update_plex_config(int plexno, int kernelstate); +void update_volume_config(int volno, int kernelstate); +void update_config(void); +void drive_io_done(struct buf *); +int save_config(void); +void write_config(char *, int); +int start_config(void); +void finish_config(int); +void remove(struct vinum_ioctl_msg *msg); +void remove_drive_entry(int driveno, int force, int recurse); +void remove_sd_entry(int sdno, int force, int recurse); +void remove_plex_entry(int plexno, int force, int recurse); +void remove_volume_entry(int volno, int force, int recurse); + +void checkernel(char *); +int open_drive(struct drive *, struct proc *); +void close_drive(struct drive *drive); +int driveio(struct drive *, void *, size_t, off_t, int); +/* #define read_drive(a, b, c, d) driveio (a, b, c, d, B_READ) + #define write_drive(a, b, c, d) driveio (a, b, c, d, B_WRITE) */ +int set_drive_parms(struct drive *drive); +int init_drive(struct drive *); +/* void throw_rude_remark (int, struct _ioctl_reply *, char *, ...); XXX */ +void throw_rude_remark(int, char *,...); + +int read_drive(struct drive *drive, void *buf, size_t length, off_t offset); +int write_drive(struct drive *drive, void *buf, size_t length, off_t offset); +void format_config(char *config, int len); +void checkkernel(char *op); +void free_drive(struct drive *drive); +void down_drive(struct drive *drive); +void remove_drive(int driveno); + +/* I/O */ +d_open_t vinumopen; +d_close_t vinumclose; +d_strategy_t vinumstrategy; +d_ioctl_t vinumioctl; +d_dump_t vinumdump; +d_psize_t vinumsize; +d_read_t vinumread; +d_write_t vinumwrite; + +int vinumstart(struct buf *bp, int reviveok); +int launch_requests(struct request *rq, int reviveok); + +/* XXX Do we need this? */ +int vinumpart(dev_t); + +/* Memory allocation */ +void vinum_meminfo(caddr_t data); +int vinum_mallocinfo(caddr_t data); + +void expand_table(void **, int, int); + +void add_defective_region(struct plex *plex, off_t offset, size_t length); +void add_unmapped_region(struct plex *plex, off_t offset, size_t length); +void rebuild_plex_unmappedlist(struct plex *plex); +struct request; +struct rqgroup *allocrqg(struct request *rq, int elements); +void deallocrqg(struct rqgroup *rqg); + +/* State transitions */ +int set_drive_state(int driveno, enum drivestate state, int force); +int set_sd_state(int sdno, enum sdstate state, enum setstateflags flags); +enum requeststatus checksdstate(struct sd *sd, struct request *rq, daddr_t diskaddr, daddr_t diskend); +int set_plex_state(int plexno, enum plexstate state, enum setstateflags flags); +int set_volume_state(int volumeno, enum volumestate state, enum setstateflags flags); +void get_volume_label(struct volume *vol, struct disklabel *lp); +int write_volume_label(int); +void start_object(struct vinum_ioctl_msg *); +void stop_object(struct vinum_ioctl_msg *); +void setstate(struct vinum_ioctl_msg *msg); +void vinum_label(int); +int vinum_writedisklabel(struct volume *, struct disklabel *); +int initsd(int); + +int restart_plex(int plexno); +int revive_block(int plexno); + +/* Auxiliary functions */ +enum sdstates sdstatemap(struct plex *plex, int *sddowncount); +enum volplexstate vpstate(struct plex *plex); +#endif + +enum keyword get_keyword(char *, struct keywordset *); +void listconfig(void); +char *drive_state(enum drivestate); +char *volume_state(enum volumestate); +char *plex_state(enum plexstate); +char *plex_org(enum plexorg); +char *sd_state(enum sdstate); +enum drivestate DriveState(char *text); +enum sdstate SdState(char *text); +enum plexstate PlexState(char *text); +enum volumestate VolState(char *text); +struct drive *validdrive(int driveno, struct _ioctl_reply *); +struct sd *validsd(int sdno, struct _ioctl_reply *); +struct plex *validplex(int plexno, struct _ioctl_reply *); +struct volume *validvol(int volno, struct _ioctl_reply *); +int tokenize(char *, char *[]); +void resetstats(struct vinum_ioctl_msg *msg); + +/* Locking */ +int lockvol(struct volume *vol); +void unlockvol(struct volume *vol); +int lockplex(struct plex *plex); +void unlockplex(struct plex *plex); +int lockrange(struct plex *plex, off_t first, off_t last); +void unlockrange(struct plex *plex, off_t first, off_t last); +int lock_config(void); +void unlock_config(void); + +#ifdef DEBUG +#define expandrq(prq) \ +{ \ + expand_table ((void **) &prq->rqe, \ + prq->requests * sizeof (struct rqelement), \ + (prq->requests + RQELTS) * sizeof (struct rqelement) ); \ + bzero (&prq->rqe [prq->requests], RQELTS * sizeof (struct rqelement)); \ + prq->rqcount += RQELTS; \ + } +#else +void expandrq(struct plexrq *); +#endif diff --git a/lkm/vinum/vinumhdr.h b/lkm/vinum/vinumhdr.h new file mode 100644 index 0000000..22a8561 --- /dev/null +++ b/lkm/vinum/vinumhdr.h @@ -0,0 +1,104 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + */ + +/* Header files used by all modules */ +/* $Id: vinumhdr.h,v 1.7 1998/08/07 04:41:18 grog Exp grog $ */ + +#ifdef KERNEL +#define REALLYKERNEL +#endif +#include <sys/param.h> +#ifdef REALLYKERNEL +#include <sys/systm.h> +#include <sys/kernel.h> +#endif +#ifdef DEVFS +#error "DEVFS code not complete yet" +#include <sys/devfsext.h> +#endif /*DEVFS */ +#include <sys/proc.h> +#include <sys/errno.h> +#include <sys/dkstat.h> +#include <sys/buf.h> +#include <sys/malloc.h> +#include <sys/uio.h> +#include <sys/namei.h> +#include <sys/conf.h> +#include <sys/stat.h> +#include <sys/disklabel.h> +#include <ufs/ffs/fs.h> +#include <sys/mount.h> +#include <sys/device.h> +#undef KERNEL /* XXX */ +#include <sys/disk.h> +#ifdef REALLYKERNEL +#define KERNEL +#endif +#include <sys/syslog.h> +#include <sys/fcntl.h> +#include <sys/vnode.h> +#include <sys/dkbad.h> +#include <setjmp.h> +#include <stdarg.h> +#include <vm/vm.h> +#ifdef USES_VM +/* XXX Do we need this? */ +#include <vm/vm_extern.h> +#include <vm/vm_kern.h> +#include <vm/vm_object.h> +#include <vm/vm_prot.h> +/* #include <vm/vm_page.h> */ +#include <sys/vmmeter.h> +/* #include <machine/pmap.h> */ +#include <machine/cputypes.h> +#endif /* USES_VM */ +#include <vinumvar.h> +#include <vinumio.h> +#include "vinumkw.h" +#include "vinumext.h" + +#undef Free /* defined in some funny net stuff */ +#ifdef REALLYKERNEL +#define Malloc(x) MMalloc ((x), __FILE__, __LINE__) /* show where we came from */ +#define Free(x) FFree ((x), __FILE__, __LINE__) /* show where we came from */ +caddr_t MMalloc (int size, char *, int); +void FFree (void *mem, char *, int); +#else +#define Malloc(x) malloc ((x)) /* just the size */ +#define Free(x) free ((x)) /* just the address */ +#endif + diff --git a/lkm/vinum/vinumio.h b/lkm/vinum/vinumio.h new file mode 100644 index 0000000..be79528 --- /dev/null +++ b/lkm/vinum/vinumio.h @@ -0,0 +1,132 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumio.h,v 1.10 1998/08/10 05:46:19 grog Exp grog $ + */ + +#define MAX_IOCTL_REPLY 256 +#define L 'F' /* ID letter of our ioctls */ +/* VINUM_CREATE returns a buffer of this kind */ +struct _ioctl_reply { + int error; + char msg[MAX_IOCTL_REPLY]; +}; + +/* ioctl requests */ +#define BUFSIZE 1024 /* size of buffer, including continuations */ +#define VINUM_CREATE _IOC(IOC_IN | IOC_OUT, L, 64, BUFSIZE) /* configure vinum */ +#define VINUM_GETCONFIG _IOR(L, 65, struct _vinum_conf) /* get global config */ +#define VINUM_DRIVECONFIG _IOWR(L, 66, struct drive) /* get drive config */ +#define VINUM_SDCONFIG _IOWR(L, 67, struct sd) /* get subdisk config */ +#define VINUM_PLEXCONFIG _IOWR(L, 68, struct plex) /* get plex config */ +#define VINUM_VOLCONFIG _IOWR(L, 69, struct volume) /* get volume config */ +#define VINUM_PLEXSDCONFIG _IOWR(L, 70, struct sd) /* get sd config for plex (plex, sdno) */ +#define VINUM_GETFREELIST _IOWR(L, 71, struct drive_freelist) /* get freelist element (drive, fe) */ +#define VINUM_SAVECONFIG _IOC(0, L, 72, 0) /* release locks, update, write config to disk */ +#define VINUM_RESETCONFIG _IOC(0, L, 73, 0) /* trash config on disk */ +#define VINUM_INIT _IOC(0, L, 74, 0) /* read config from disk */ +#ifdef DEBUG + +struct debuginfo { + int changeit; + int param; +}; + +#define VINUM_DEBUG _IOWR(L, 75, struct debuginfo) /* call the debugger from ioctl () */ +#endif + +enum objecttype { + drive_object, + sd_object, + plex_object, + volume_object, + invalid_object +}; + +/* Start an object. Pass two integers: + * msg [0] index in vinum_conf.<object> + * msg [1] type of object (see below) + * + * Return ioctl_reply + */ +#define VINUM_SETSTATE _IOC(IOC_IN | IOC_OUT, L, 76, MAX_IOCTL_REPLY) /* start an object */ + +/* The state to set with VINUM_SETSTATE. Since + * each object has a different set of states, we + * need to translate later */ +enum objectstate { + object_down, + object_initializing, + object_up +}; + +/* This structure is used for modifying objects + * (VINUM_SETSTATE, VINUM_REMOVE, VINUM_RESETSTATS, VINUM_ATTACH, + * VINUM_DETACH, VINUM_REPLACE + */ +struct vinum_ioctl_msg { + int index; + enum objecttype type; + enum objectstate state; /* state to set (VINUM_SETSTATE) */ + int force; /* do it even if it doesn't make sense */ + int recurse; /* recurse (VINUM_REMOVE) */ + int otherobject; /* superordinate object (attach), + * replacement object (replace) */ + int rename; /* rename object (attach) */ + int64_t offset; /* offset of subdisk (for attach) */ +}; + +#define VINUM_RELEASECONFIG _IOC(0, L, 77, 0) /* release locks and write config to disk */ +#define VINUM_STARTCONFIG _IOC(0, L, 78, 0) /* start a configuration operation */ +#define VINUM_MEMINFO _IOR(L, 79, struct meminfo) /* get memory usage summary */ +#define VINUM_MALLOCINFO _IOWR(L, 80, struct mc) /* get specific malloc information [i] */ +#define VINUM_LABEL _IOC(IOC_IN | IOC_OUT, L, 81, MAX_IOCTL_REPLY) /* label a volume */ +#define VINUM_INITSD _IOW(L, 82, int) /* initialize a subdisk */ +#define VINUM_REMOVE _IOC(IOC_IN | IOC_OUT, L, 83, MAX_IOCTL_REPLY) /* remove an object */ +#define VINUM_GETUNMAPPED _IOWR(L, 84, struct plexregion) /* get unmapped element (plex, re) */ +#define VINUM_GETDEFECTIVE _IOWR(L, 85, struct plexregion) /* get defective element (plex, re) */ +#define VINUM_RESETSTATS _IOC(IOC_IN | IOC_OUT, L, 86, MAX_IOCTL_REPLY) /* reset object stats */ +#define VINUM_ATTACH _IOC(IOC_IN | IOC_OUT, L, 87, MAX_IOCTL_REPLY) /* reset object stats */ +#define VINUM_DETACH _IOC(IOC_IN | IOC_OUT, L, 88, MAX_IOCTL_REPLY) /* reset object stats */ + +struct vinum_rename_msg { + int index; + int recurse; /* rename subordinate objects too */ + enum objecttype type; + char newname[MAXNAME]; /* new name to give to object */ +}; + +#define VINUM_RENAME _IOC(IOC_IN | IOC_OUT, L, 89, MAX_IOCTL_REPLY) /* reset object stats */ +#define VINUM_REPLACE _IOC(IOC_IN | IOC_OUT, L, 90, MAX_IOCTL_REPLY) /* reset object stats */ diff --git a/lkm/vinum/vinumioctl.c b/lkm/vinum/vinumioctl.c new file mode 100644 index 0000000..6dbe3c6 --- /dev/null +++ b/lkm/vinum/vinumioctl.c @@ -0,0 +1,787 @@ +/* XXX replace all the checks on object validity with + * calls to valid<object> */ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumioctl.c,v 1.1 1998/08/14 08:46:10 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "sys/sysproto.h" /* for sync(2) */ +#ifdef DEBUG +#include <sys/reboot.h> +#endif + +jmp_buf command_fail; /* return on a failed command */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +/* pointer to ioctl p parameter, to save passing it around */ +struct proc *myproc; + +int vinum_inactive(void); +void free_vinum(int); +void attachobject(struct vinum_ioctl_msg *); +void detachobject(struct vinum_ioctl_msg *); +void renameobject(struct vinum_rename_msg *); +void replaceobject(struct vinum_ioctl_msg *); + +/* ioctl routine */ +int +vinumioctl(dev_t dev, +#if __FreeBSD__ >= 3 + u_long cmd, +#else + int cmd, +#endif + caddr_t data, + int flag, + struct proc *p) +{ + BROKEN_GDB; + unsigned int objno; + int error = 0; + struct volume *vol; + unsigned int index; /* for transferring config info */ + unsigned int sdno; /* for transferring config info */ + int fe; /* free list element number */ + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) data; /* struct to return */ + + struct devcode *device = (struct devcode *) &dev; + + /* First, decide what we're looking at */ + switch (device->type) { + case VINUM_SUPERDEV_TYPE: + myproc = p; /* save pointer to process */ + ioctl_reply = (struct _ioctl_reply *) data; /* save the address to reply to */ + error = setjmp(command_fail); /* come back here on error */ + if (error) /* bombed out */ + return 0; /* the reply will contain meaningful info */ + switch (cmd) { + /* XXX #ifdef DEBUG */ + case VINUM_DEBUG: + boothowto |= RB_GDB; /* serial debug line */ + if (((struct debuginfo *) data)->changeit) /* change debug settings */ + debug = (((struct debuginfo *) data)->param); + else + Debugger("vinum debug"); + ioctl_reply = (struct _ioctl_reply *) data; /* reinstate the address to reply to */ + ioctl_reply->error = 0; + return 0; + /* XXX #endif */ + + case VINUM_CREATE: /* create a vinum object */ + error = lock_config(); /* get the config for us alone */ + if (error) /* can't do it, */ + return error; /* give up */ + error = setjmp(command_fail); /* come back here on error */ + if (error == 0) { /* first time, */ + parse_user_config((char *) data, &keyword_set); /* update the config */ + ioctl_reply->error = 0; /* no error if we make it here */ + } else if (ioctl_reply->error == 0) { /* longjmp, but no error status */ + ioctl_reply->error = EINVAL; /* note that something's up */ + ioctl_reply->msg[0] = '\0'; /* no message? */ + } + unlock_config(); + return 0; /* must be 0 to return the real error info */ + + case VINUM_GETCONFIG: /* get the configuration information */ + bcopy(&vinum_conf, data, sizeof(vinum_conf)); + return 0; + + /* start configuring the subsystem */ + case VINUM_STARTCONFIG: + return start_config(); /* just lock it */ + + /* Move the individual parts of the config to user space. + + * Specify the index of the object in the first word of data, + * and return the object there + */ + case VINUM_DRIVECONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.drives_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&DRIVE[index], data, sizeof(struct drive)); /* copy the config item out */ + return 0; + + case VINUM_SDCONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.subdisks_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&SD[index], data, sizeof(struct sd)); /* copy the config item out */ + return 0; + + case VINUM_PLEXCONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.plexes_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&PLEX[index], data, sizeof(struct plex)); /* copy the config item out */ + return 0; + + case VINUM_VOLCONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.volumes_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&VOL[index], data, sizeof(struct volume)); /* copy the config item out */ + return 0; + + case VINUM_PLEXSDCONFIG: + index = *(int *) data; /* get the plex index */ + sdno = ((int *) data)[1]; /* and the sd index */ + if ((index >= (unsigned) vinum_conf.plexes_used) /* plex doesn't exist */ + ||(sdno >= PLEX[index].subdisks)) /* or it doesn't have this many subdisks */ + return EFAULT; /* bang */ + bcopy(&SD[PLEX[index].sdnos[sdno]], /* copy the config item out */ + data, + sizeof(struct sd)); + return 0; + + case VINUM_SAVECONFIG: + if (VFLAGS & VF_CONFIGURING) { /* must be us, the others are asleep */ + finish_config(1); /* finish the configuration and update it */ + error = save_config(); /* save configuration to disk */ + } else + error = EINVAL; /* queue up for this one, please */ + return error; + + case VINUM_RELEASECONFIG: /* release the config */ + if (VFLAGS & VF_CONFIGURING) { /* must be us, the others are asleep */ + finish_config(0); /* finish the configuration, don't change it */ + error = save_config(); /* save configuration to disk */ + } else + error = EINVAL; /* release what config? */ + return error; + + case VINUM_INIT: + ioctl_reply = (struct _ioctl_reply *) data; /* reinstate the address to reply to */ + ioctl_reply->error = 0; + return 0; + + case VINUM_RESETCONFIG: + if (vinum_inactive() && (vinum_conf.opencount < 2)) { /* if we're not active */ + /* Note the open count. We may be called from v, so we'll be open. + * Keep the count so we don't underflow */ + int oc = vinum_conf.opencount; + free_vinum(1); /* clean up everything */ + printf("vinum: CONFIGURATION OBLITERATED\n"); + vinum_conf.opencount = oc; + ioctl_reply = (struct _ioctl_reply *) data; /* reinstate the address to reply to */ + ioctl_reply->error = 0; + return 0; + } + return EBUSY; + + case VINUM_SETSTATE: + setstate((struct vinum_ioctl_msg *) data); /* set an object state */ + return 0; + + case VINUM_MEMINFO: + vinum_meminfo(data); + return 0; + + case VINUM_MALLOCINFO: + return vinum_mallocinfo(data); + + case VINUM_LABEL: /* label a volume */ + ioctl_reply->error = write_volume_label(*(int *) data); /* index of the volume to label */ + ioctl_reply->msg[0] = '\0'; /* no message */ + return 0; + + case VINUM_REMOVE: + remove((struct vinum_ioctl_msg *) data); /* remove an object */ + return 0; + + case VINUM_GETFREELIST: /* get a drive free list element */ + index = *(int *) data; /* get the drive index */ + fe = ((int *) data)[1]; /* and the free list element */ + if ((index >= (unsigned) vinum_conf.drives_used) /* plex doesn't exist */ + ||(DRIVE[index].state == drive_unallocated)) + return ENODEV; + if (fe >= DRIVE[index].freelist_entries) /* no such entry */ + return ENOENT; + bcopy(&DRIVE[index].freelist[fe], + data, + sizeof(struct drive_freelist)); + return 0; + + case VINUM_GETDEFECTIVE: /* get a plex defective area element */ + index = *(int *) data; /* get the plex index */ + fe = ((int *) data)[1]; /* and the region number */ + if ((index >= (unsigned) vinum_conf.plexes_used) /* plex doesn't exist */ + ||(PLEX[index].state == plex_unallocated)) + return ENODEV; + if (fe >= PLEX[index].defective_regions) /* no such entry */ + return ENOENT; + bcopy(&PLEX[index].defective_region[fe], + data, + sizeof(struct plexregion)); + return 0; + + case VINUM_GETUNMAPPED: /* get a plex unmapped area element */ + index = *(int *) data; /* get the plex index */ + fe = ((int *) data)[1]; /* and the region number */ + if ((index >= (unsigned) vinum_conf.plexes_used) /* plex doesn't exist */ + ||(PLEX[index].state == plex_unallocated)) + return ENODEV; + if (fe >= PLEX[index].unmapped_regions) /* no such entry */ + return ENOENT; + bcopy(&PLEX[index].unmapped_region[fe], + data, + sizeof(struct plexregion)); + return 0; + + case VINUM_RESETSTATS: + resetstats((struct vinum_ioctl_msg *) data); /* reset object stats */ + return 0; + + /* attach an object to a superordinate object */ + case VINUM_ATTACH: + attachobject((struct vinum_ioctl_msg *) data); + return 0; + + /* detach an object from a superordinate object */ + case VINUM_DETACH: + detachobject((struct vinum_ioctl_msg *) data); + return 0; + + /* rename an object */ + case VINUM_RENAME: + renameobject((struct vinum_rename_msg *) data); + return 0; + + /* replace an object */ + case VINUM_REPLACE: + replaceobject((struct vinum_ioctl_msg *) data); + return 0; + + default: + /* FALLTHROUGH */ + } + + default: +#if __FreeBSD__>=3 + printf("vinumioctl: type %d, sd %d, plex %d, major %x, volume %d, command %lx\n", + device->type, + device->sd, + device->plex, + device->major, + device->volume, + cmd); /* XXX */ + +#else + printf("vinumioctl: type %d, sd %d, plex %d, major %x, volume %d, command %x\n", + device->type, + device->sd, + device->plex, + device->major, + device->volume, + cmd); /* XXX */ + +#endif + return EINVAL; + + case VINUM_DRIVE_TYPE: + case VINUM_PLEX_TYPE: + return EAGAIN; /* try again next week */ + + case VINUM_SD_TYPE: + objno = SDNO(dev); + + switch (cmd) { + case VINUM_INITSD: /* initialize subdisk */ + return initsd(objno); + + default: + return EINVAL; + } + break; + + case VINUM_VOLUME_TYPE: + objno = VOLNO(dev); + + if ((unsigned) objno >= (unsigned) vinum_conf.volumes_used) /* not a valid volume */ + return ENXIO; + vol = &VOL[objno]; + if (vol->state != volume_up) /* not up, */ + return EIO; /* I/O error */ + + switch (cmd) { + case DIOCGDINFO: /* get disk label */ + get_volume_label(vol, (struct disklabel *) data); + break; + + /* Care! DIOCGPART returns *pointers* to + * the caller, so we need to store this crap as well. + * And yes, we need it. */ + case DIOCGPART: /* get partition information */ + get_volume_label(vol, &vol->label); + ((struct partinfo *) data)->disklab = &vol->label; + ((struct partinfo *) data)->part = &vol->label.d_partitions[0]; + break; + + /* We don't have this stuff on hardware, + * so just pretend to do it so that + * utilities don't get upset. */ + case DIOCWDINFO: /* write partition info */ + case DIOCSDINFO: /* set partition info */ + return 0; /* not a titty */ + + case DIOCWLABEL: /* set or reset label writeable */ + if ((flag & FWRITE) == 0) /* not writeable? */ + return EACCES; /* no, die */ + if (*(int *) data != 0) /* set it? */ + vol->flags |= VF_WLABEL; /* yes */ + else + vol->flags &= ~VF_WLABEL; /* no, reset */ + break; + + default: + return ENOTTY; /* not my kind of ioctl */ + } + break; + } + return 0; /* XXX */ +} + +/* The following four functions check the supplied + * object index and return a pointer to the object + * if it exists. Otherwise they longjump out via + * throw_rude_remark */ +struct drive * +validdrive(int driveno, struct _ioctl_reply *reply) +{ + if ((driveno < vinum_conf.drives_used) + && (DRIVE[driveno].state != drive_unallocated)) + return &DRIVE[driveno]; + strcpy(reply->msg, "No such drive"); + reply->error = ENOENT; + return NULL; +} + +struct sd * +validsd(int sdno, struct _ioctl_reply *reply) +{ + if ((sdno < vinum_conf.subdisks_used) + && (SD[sdno].state != sd_unallocated)) + return &SD[sdno]; + strcpy(reply->msg, "No such subdisk"); + reply->error = ENOENT; + return NULL; +} + +struct plex * +validplex(int plexno, struct _ioctl_reply *reply) +{ + if ((plexno < vinum_conf.plexes_used) + && (PLEX[plexno].state != plex_unallocated)) + return &PLEX[plexno]; + strcpy(reply->msg, "No such plex"); + reply->error = ENOENT; + return NULL; +} + +struct volume * +validvol(int volno, struct _ioctl_reply *reply) +{ + if ((volno < vinum_conf.volumes_used) + && (VOL[volno].state != volume_unallocated)) + return &VOL[volno]; + strcpy(reply->msg, "No such volume"); + reply->error = ENOENT; + return NULL; +} + +/* reset an object's stats */ +void +resetstats(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + + switch (msg->type) { + case drive_object: + if (msg->index < vinum_conf.drives_used) { + struct drive *drive = &DRIVE[msg->index]; + if (drive->state != drive_unallocated) { + drive->reads = 0; /* number of reads on this drive */ + drive->writes = 0; /* number of writes on this drive */ + drive->bytes_read = 0; /* number of bytes read */ + drive->bytes_written = 0; /* number of bytes written */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + case sd_object: + if (msg->index < vinum_conf.subdisks_used) { + struct sd *sd = &SD[msg->index]; + if (sd->state != sd_unallocated) { + sd->reads = 0; /* number of reads on this subdisk */ + sd->writes = 0; /* number of writes on this subdisk */ + sd->bytes_read = 0; /* number of bytes read */ + sd->bytes_written = 0; /* number of bytes written */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + break; + + case plex_object: + if (msg->index < vinum_conf.plexes_used) { + struct plex *plex = &PLEX[msg->index]; + if (plex->state != plex_unallocated) { + plex->reads = 0; + plex->writes = 0; /* number of writes on this plex */ + plex->bytes_read = 0; /* number of bytes read */ + plex->bytes_written = 0; /* number of bytes written */ + plex->multiblock = 0; /* requests that needed more than one block */ + plex->multistripe = 0; /* requests that needed more than one stripe */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + break; + + case volume_object: + if (msg->index < vinum_conf.volumes_used) { + struct volume *vol = &VOL[msg->index]; + if (vol->state != volume_unallocated) { + vol->bytes_read = 0; /* number of bytes read */ + vol->bytes_written = 0; /* number of bytes written */ + vol->reads = 0; /* number of reads on this volume */ + vol->writes = 0; /* number of writes on this volume */ + vol->recovered_reads = 0; /* reads recovered from another plex */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + case invalid_object: /* can't get this */ + reply->error = EINVAL; + return; + } +} + +/* attach an object to a superior object */ +void +attachobject(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + struct sd *sd; + struct plex *plex; + struct volume *vol; + + switch (msg->type) { + case drive_object: /* you can't attach a drive to anything */ + case volume_object: /* nor a volume */ + case invalid_object: /* "this can't happen" */ + reply->error = EINVAL; + reply->msg[0] = '\0'; /* vinum(8) doesn't do this */ + return; + + case sd_object: + sd = validsd(msg->index, reply); + if (sd == NULL) /* not a valid subdisk */ + return; + plex = validplex(msg->otherobject, reply); + if (plex) { + if (sd->plexno >= 0) { /* already belong to a plex */ + reply->error = EBUSY; /* no message, the user should check */ + reply->msg[0] = '\0'; + return; + } + sd->plexoffset = msg->offset; /* this is where we want it */ + set_sd_state(sd->sdno, sd_stale, setstate_force); /* make sure it's stale */ + give_sd_to_plex(plex->plexno, sd->sdno); /* and give it to the plex */ + update_sd_config(sd->sdno, 0); + save_config(); + reply->error = 0; + } + break; + + case plex_object: + plex = validplex(msg->index, reply); /* get plex */ + if (plex == NULL) + return; + if (plex->organization != plex_concat) { /* can't attach to striped and raid-5 */ + reply->error = EINVAL; /* no message, the user should check */ + reply->msg[0] = '\0'; + return; + } + vol = validvol(msg->otherobject, reply); /* and volume information */ + if (vol) { + if ((vol->plexes == MAXPLEX) /* we have too many already */ + ||(plex->volno >= 0)) { /* or the plex has an owner */ + reply->error = EINVAL; /* no message, the user should check */ + reply->msg[0] = '\0'; + return; + } + set_plex_state(plex->plexno, plex_down, setstate_force); /* make sure it's down */ + give_plex_to_volume(msg->otherobject, msg->index); /* and give it to the volume */ + update_plex_config(plex->plexno, 0); + save_config(); + if (plex->state == plex_reviving) + reply->error = EAGAIN; /* need to revive it */ + else + reply->error = 0; + } + } +} + +/* detach an object from a superior object */ +void +detachobject(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + struct sd *sd; + struct plex *plex; + struct volume *vol; + int sdno; + int plexno; + + switch (msg->type) { + case drive_object: /* you can't attach a drive to anything */ + case volume_object: /* nor a volume */ + case invalid_object: /* "this can't happen" */ + reply->error = EINVAL; + reply->msg[0] = '\0'; /* vinum(8) doesn't do this */ + return; + + case sd_object: + sd = validsd(msg->index, reply); + if (sd == NULL) + return; + if (sd->plexno < 0) { /* doesn't belong to a plex */ + reply->error = ENOENT; + strcpy(reply->msg, "Subdisk is not attached"); + return; + } else { /* valid plex number */ + plex = &PLEX[sd->plexno]; + if ((!msg->force) /* don't force things */ + &&((plex->state == plex_up) /* and the plex is up */ + ||((plex->state == plex_flaky) && sd->state == sd_up))) { /* or flaky with this sd up */ + reply->error = EBUSY; /* we need this sd */ + reply->msg[0] = '\0'; + return; + } + sd->plexno = -1; /* anonymous sd */ + if (plex->subdisks == 1) { /* this was the only subdisk */ + Free(plex->sdnos); /* free the subdisk array */ + plex->sdnos = NULL; /* and note the fact */ + plex->subdisks_allocated = 0; /* no subdisk space */ + } else { + for (sdno = 0; sdno < plex->subdisks; sdno++) { + if (plex->sdnos[sdno] == msg->index) /* found our subdisk */ + break; + } + if (sdno < (plex->subdisks - 1)) /* not the last one, compact */ + bcopy(&plex->sdnos[sdno + 1], + &plex->sdnos[sdno], + (plex->subdisks - 1 - sdno) * sizeof(int)); + } + plex->subdisks--; + rebuild_plex_unmappedlist(plex); /* rebuild the unmapped list */ + if (!bcmp(plex->name, sd->name, strlen(plex->name))) { /* this subdisk is named after the plex */ + bcopy(sd->name, + &sd->name[3], + min(strlen(sd->name), MAXSDNAME - 3)); + bcopy("ex-", sd->name, 3); + sd->name[MAXSDNAME - 1] = '\0'; + } + update_plex_config(plex->plexno, 0); + if ((plex->organization == plex_striped) /* we've just mutilated our plex, */ + ||(plex->organization == plex_striped)) /* the data no longer matches */ + set_plex_state(plex->plexno, + plex_down, + setstate_force | setstate_configuring); + update_sd_config(sd->sdno, 0); + save_config(); + reply->error = 0; + } + return; + + case plex_object: + plex = validplex(msg->index, reply); /* get plex */ + if (plex == NULL) + return; + if (plex->volno >= 0) { + int volno = plex->volno; + + vol = &VOL[volno]; + if ((!msg->force) /* don't force things */ + &&((vol->state == volume_up) /* and the volume is up */ + &&(vol->plexes == 1))) { /* and this is the last plex */ + /* XXX As elsewhere, check whether we will lose + * mapping by removing this plex */ + reply->error = EBUSY; /* we need this plex */ + reply->msg[0] = '\0'; + return; + } + plex->volno = -1; /* anonymous plex */ + for (plexno = 0; plexno < vol->plexes; plexno++) { + if (vol->plex[plexno] == msg->index) /* found our plex */ + break; + } + if (plexno < (vol->plexes - 1)) /* not the last one, compact */ + bcopy(&vol[plexno + 1], &vol[plexno], (vol->plexes - 1 - plexno) * sizeof(int)); + vol->plexes--; + if (!bcmp(vol->name, plex->name, strlen(vol->name))) { /* this plex is named after the volume */ + /* First, check if the subdisks are the same */ + if (msg->recurse) { + int sdno; + + for (sdno = 0; sdno < plex->subdisks; sdno++) { + struct sd *sd = &SD[plex->sdnos[sdno]]; + + if (!bcmp(plex->name, sd->name, strlen(plex->name))) { /* subdisk is named after the plex */ + bcopy(sd->name, &sd->name[3], min(strlen(sd->name), MAXSDNAME - 3)); + bcopy("ex-", sd->name, 3); + sd->name[MAXSDNAME - 1] = '\0'; + } + } + } + bcopy(plex->name, &plex->name[3], min(strlen(plex->name), MAXPLEXNAME - 3)); + bcopy("ex-", plex->name, 3); + plex->name[MAXPLEXNAME - 1] = '\0'; + } + update_plex_config(plex->plexno, 0); + update_volume_config(volno, 0); + save_config(); + reply->error = 0; + } else { + reply->error = ENOENT; + strcpy(reply->msg, "Plex is not attached"); + } + } +} + +void +renameobject(struct vinum_rename_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + struct drive *drive; + struct sd *sd; + struct plex *plex; + struct volume *vol; + + switch (msg->type) { + case drive_object: /* you can't attach a drive to anything */ + if (find_drive(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + drive = validdrive(msg->index, reply); + if (drive) { + bcopy(msg->newname, drive->label.name, MAXDRIVENAME); + save_config(); + reply->error = 0; + } + return; + + case sd_object: /* you can't attach a subdisk to anything */ + if (find_subdisk(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + sd = validsd(msg->index, reply); + if (sd) { + bcopy(msg->newname, sd->name, MAXSDNAME); + update_sd_config(sd->sdno, 0); + save_config(); + reply->error = 0; + } + return; + + case plex_object: /* you can't attach a plex to anything */ + if (find_plex(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + plex = validplex(msg->index, reply); + if (plex) { + bcopy(msg->newname, plex->name, MAXPLEXNAME); + update_plex_config(plex->plexno, 0); + save_config(); + reply->error = 0; + } + return; + + case volume_object: /* you can't attach a volume to anything */ + if (find_volume(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + vol = validvol(msg->index, reply); + if (vol) { + bcopy(msg->newname, vol->name, MAXVOLNAME); + update_volume_config(msg->index, 0); + save_config(); + reply->error = 0; + } + return; + + case invalid_object: + reply->error = EINVAL; + reply->msg[0] = '\0'; + } +} + +/* Replace one object with another */ +void +replaceobject(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + + reply->error = ENODEV; /* until I know how to do this */ + strcpy(reply->msg, "replace not implemented yet"); +/* save_config (); */ +} diff --git a/lkm/vinum/vinumkw.h b/lkm/vinum/vinumkw.h new file mode 100644 index 0000000..1a81f37 --- /dev/null +++ b/lkm/vinum/vinumkw.h @@ -0,0 +1,120 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumkw.h,v 1.7 1998/08/07 02:35:51 grog Exp grog $ + */ + +/* Command keywords that vinum knows. These include both user-level + * and kernel-level stuff */ + +/* Our complete vocabulary. The names of the commands are + * the same as the identifier without the kw_ at the beginning + * (i.e. kw_create defines the "create" keyword). Preprocessor + * magic in parser.c does the rest. */ +enum keyword { + kw_create, + kw_modify, + kw_list, + kw_l = kw_list, + kw_ld, /* list drive */ + kw_ls, /* list subdisk */ + kw_lp, /* list plex */ + kw_lv, /* list volume */ + kw_set, + kw_rm, + kw_start, + kw_stop, + kw_drive, + kw_sd, + kw_subdisk = kw_sd, + kw_plex, + kw_volume, + kw_vol = kw_volume, + kw_read, + kw_readpol, + kw_org, + kw_name, + kw_concat, + kw_striped, + kw_raid5, + kw_driveoffset, + kw_plexoffset, + kw_len, + kw_length = kw_len, + kw_state, + kw_setupstate, + kw_d, /* flag names */ + kw_f, + kw_r, + kw_s, + kw_v, + kw_round, /* round robin */ + kw_prefer, /* prefer plex */ + kw_device, + kw_init, + kw_label, + kw_resetconfig, + kw_writethrough, + kw_writeback, + kw_raw, + kw_resetstats, + kw_attach, + kw_detach, + kw_rename, + kw_printconfig, + kw_replace, + kw_detached, +#ifdef DEBUG + kw_debug, /* go into debugger */ + kw_info, +#endif + kw_invalid_keyword = -1 +}; + +struct _keywords { + char *name; + enum keyword keyword; +}; + +struct keywordset { + int size; + struct _keywords *k; +}; + +extern struct _keywords keywords[]; +extern struct _keywords flag_keywords[]; + +extern struct keywordset keyword_set; +extern struct keywordset flag_set; diff --git a/lkm/vinum/vinumstate.h b/lkm/vinum/vinumstate.h new file mode 100644 index 0000000..0b07590 --- /dev/null +++ b/lkm/vinum/vinumstate.h @@ -0,0 +1,213 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumstate.h,v 1.11 1998/08/04 06:22:49 grog Exp grog $ + */ + +/* This file gets read by makestatetext to create text files + * with the names of the states, so don't change the file + * format */ + +enum volumestate { + volume_unallocated, + /* present but unused. Must be 0 */ + + volume_uninit, + /* mentioned elsewhere but not defined */ + + volume_down, + + /* The volume is up and functional, but not all plexes may be available */ + volume_up, + volume_laststate = volume_up /* last value, for table dimensions */ +}; + +enum plexstate { + /* An empty entry, not a plex at all. */ + plex_unallocated, + + /* The plex has been allocated, but there configuration + * is not complete */ + plex_init, + + /* A plex which has gone completely down because of + * I/O errors. */ + plex_faulty, + + /* A plex which has been taken down by the + * administrator. */ + plex_down, + + /* A plex which is currently being brought up after + * being not up. This involves copying data from + * another plex */ + plex_reviving, + + /* A plex which is being initialized */ + plex_initializing, + + /* *** The remaining states represent plexes which are + * at least partially up. Keep these separate so that + * they can be checked more easily. */ + + /* A plex entry which is at least partially up. Not + * all subdisks are available, and an inconsistency + * has occurred. If no other plex is uncorrupted, + * the volume is no longer consistent. */ + plex_corrupt, + + plex_firstup = plex_corrupt, /* first "up" state */ + + /* A plex entry which is at least partially up. Not + * all subdisks are available, but so far no + * inconsistency has occurred (this will change with + * the first write to the address space occupied by + * a defective subdisk). A RAID 5 plex with one subdisk + * down will remain degraded even after a write */ + plex_degraded, + + /* A plex which is really up, but which has a reborn + * subdisk which we don't completely trust, and + * which we don't want to read if we can avoid it */ + plex_flaky, + + /* A plex entry which is completely up. All subdisks + * are up. */ + plex_up, + + plex_laststate = plex_up /* last value, for table dimensions */ +}; + +/* subdisk states */ +enum sdstate { + /* An empty entry, not a subdisk at all. */ + sd_unallocated, + + /* A subdisk entry which has not been created + * completely. Some fields may be empty. + */ + sd_uninit, + + /* A subdisk entry which has been created completely. + * All fields are correct, but the disk hasn't + * been updated. + */ + sd_init, + + /* A subdisk entry which has been created completely and + * which is currently being initialized */ + sd_initializing, + + /* A subdisk entry which has been created completely. + * All fields are correct, and the disk has been + * updated, but there is no data on the disk. + */ + sd_empty, + + /* *** The following states represent invalid data */ + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down, and as a result updates have been + * missed. + */ + sd_obsolete, + + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down, updates have been lost, and then + * the drive came up again. + */ + sd_stale, + + /* *** The following states represent valid, inaccessible data */ + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down. No attempt has been made to write + * to the subdisk since the crash. + */ + sd_crashed, + + /* A subdisk entry which was up, which contained + * valid data, and which was taken down by the + * administrator. The data is valid. */ + sd_down, + + /* *** The following states represent accessible subdisks + * with valid data */ + + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down and up again. No updates were lost, + * but it is possible that the subdisk has been + * damaged. We won't read from this subdisk if we + * have a choice. If this is the only subdisk which + * covers this address space in the plex, we set its + * state to sd_up under these circumstances, so this + * status implies that there is another subdisk to + * fulfil the request. + */ + sd_reborn, + + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data is valid. + */ + sd_up, + + sd_laststate = sd_up /* last value, for table dimensions */ +}; + +enum drivestate { + drive_unallocated, + /* present but unused. Must be 0 */ + + drive_uninit, + /* just mentioned in some other config entry */ + + drive_down, + /* not accessible */ + + drive_coming_up, + /* in the process of being brought up */ + + drive_up, + /* up and running */ + + drive_laststate = drive_up /* last value, for table dimensions */ +}; diff --git a/lkm/vinum/vinumvar.h b/lkm/vinum/vinumvar.h new file mode 100644 index 0000000..ca54f76 --- /dev/null +++ b/lkm/vinum/vinumvar.h @@ -0,0 +1,510 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumvar.h,v 1.15 1998/08/14 06:36:41 grog Exp grog $ + */ + +/* XXX gdb can't find our global pointers, so use this kludge to + * point to them locally. Remove after testing */ +#define BROKEN_GDB struct _vinum_conf *VC = &vinum_conf + +#include <sys/time.h> +#include "vinumstate.h" +/* Some configuration maxima. They're an enum because + * we can't define global constants. Sorry about that. + * + * These aren't as bad as they look: most of them + * are soft limits. Only the MAXCONFIG parameter is set in stone + */ + +enum constants { + VINUM_HEADER = 512, /* size of header on disk */ + MAXCONFIGLINE = 1024, /* maximum size of a single config line */ + /* XXX Do we still need this? */ + MINVINUMSLICE = 1048576, /* minimum size of a slice */ + + CDEV_MAJOR = 91, /* major number for character device */ + BDEV_MAJOR = 25, /* and block device */ + + ROUND_ROBIN_READPOL = -1, /* round robin read policy */ + + /* type field in minor number */ + VINUM_VOLUME_TYPE = 0, + VINUM_PLEX_TYPE = 1, + VINUM_SD_TYPE = 2, + VINUM_DRIVE_TYPE = 3, + VINUM_SUPERDEV_TYPE = 4, /* super device. */ + + /* Shifts for the individual fields in the device */ + VINUM_TYPE_SHIFT = 28, + VINUM_VOL_SHIFT = 0, + VINUM_PLEX_SHIFT = 16, + VINUM_SD_SHIFT = 20, + VINUM_VOL_WIDTH = 8, + VINUM_PLEX_WIDTH = 3, + VINUM_SD_WIDTH = 8, + MAJORDEV_SHIFT = 8, + + +/* Create a block device number */ +#define VINUMBDEV(v,p,s,t) ((BDEV_MAJOR << MAJORDEV_SHIFT) \ + | (v << VINUM_VOL_SHIFT) \ + | (p << VINUM_PLEX_SHIFT) \ + | (s << VINUM_SD_SHIFT) \ + | (t << VINUM_TYPE_SHIFT) ) + +/* And a character device number */ +#define VINUMCDEV(v,p,s,t) ((CDEV_MAJOR << MAJORDEV_SHIFT) \ + | (v << VINUM_VOL_SHIFT) \ + | (p << VINUM_PLEX_SHIFT) \ + | (s << VINUM_SD_SHIFT) \ + | (t << VINUM_TYPE_SHIFT) ) + +/* extract device type */ +#define DEVTYPE(x) ((x >> VINUM_TYPE_SHIFT) & 7) + +/* extract volume number */ +#define VOLNO(x) (x & ((1 << VINUM_VOL_WIDTH) - 1)) + +/* extract plex number */ +#define PLEXNO(x) (VOL [VOLNO (x)].plex [(x >> VINUM_PLEX_SHIFT) & ((1 << VINUM_PLEX_WIDTH) - 1)]) + +/* extract subdisk number */ +#define SDNO(x) (PLEX [PLEXNO (x)].sdnos [(x >> VINUM_SD_SHIFT) & ((1 << VINUM_SD_WIDTH) - 1)]) + +/* extract drive number */ +#define DRIVENO(x) (SD [SDNO (x)].driveno) + + VINUM_SUPERDEV = VINUMBDEV(0, 0, 0, VINUM_SUPERDEV_TYPE), /* superdevice number */ + +/* the number of object entries to cater for initially, and also the + * value by which they are incremented. It doesn't take long + * to extend them, so theoretically we could start with 1 of each, but + * it's untidy to allocate such small areas. These values are + * probably too small. + */ + + INITIAL_DRIVES = 4, + INITIAL_VOLUMES = 4, + INITIAL_PLEXES = 8, + INITIAL_SUBDISKS = 16, + INITIAL_SUBDISKS_IN_PLEX = 4, /* number of subdisks to allocate to a plex */ + INITIAL_SUBDISKS_IN_DRIVE = 4, /* number of subdisks to allocate to a drive */ + INITIAL_DRIVE_FREELIST = 16, /* number of entries in drive freelist */ + PLEX_REGION_TABLE_SIZE = 8, /* number of entries in plex region tables */ + INITIAL_LOCKS = 8, /* number of locks to allocate to a volume */ + DEFAULT_REVIVE_BLOCKSIZE = 32768, /* size of block to transfer in one op */ +}; + +/* device numbers */ + +/* + * 31 30 28 27 20 19 18 16 15 8 7 0 + * |-----------------------------------------------------------------------------------------------| + * |X | Type | Subdisk number | X| Plex | Major number | volume number | + * |-----------------------------------------------------------------------------------------------| + * + * 0x2 03 1 19 06 + */ +struct devcode { +/* CARE. These fields assume a big-endian word. On a + * little-endian system, they're the wrong way around */ + unsigned volume:8; /* up to 256 volumes */ + unsigned major:8; /* this is where the major number fits */ + unsigned plex:3; /* up to 8 plexes per volume */ + unsigned unused:1; /* up for grabs */ + unsigned sd:8; /* up to 256 subdisks per plex */ + unsigned type:3; /* type of object */ + /* type field + VINUM_VOLUME = 0, + VINUM_PLEX = 1, + VINUM_SUBDISK = 2, + VINUM_DRIVE = 3, + VINUM_SUPERDEV = 4, */ + unsigned signbit:1; /* to make 32 bits */ +}; + +#define VINUM_DIR "/dev/vinum" +#define VINUM_RDIR "/dev/rvinum" +#define VINUM_SUPERDEV_NAME VINUM_DIR"/control" +#define MAXDRIVENAME 32 /* maximum length of a device name */ +#define MAXSDNAME 64 /* maximum length of a subdisk name */ +#define MAXPLEXNAME 64 /* maximum length of a plex name */ +#define MAXVOLNAME 64 /* maximum length of a volume name */ +#define MAXNAME 64 /* maximum length of any name */ +#define MAXVOLPLEX 8 /* maximum number of plexes in a volume */ + +/* Flags for all objects. Most of them only apply to + * specific objects, but we have space for all in any + * 32 bit flags word. */ +enum objflags { + VF_LOCKED = 1, /* somebody has locked access to this object */ + VF_LOCKING = 2, /* we want access to this object */ + VF_WRITETHROUGH = 8, /* volume: write through */ + VF_INITED = 0x10, /* unit has been initialized */ + VF_WLABEL = 0x20, /* label area is writable */ + VF_LABELLING = 0x40, /* unit is currently being labelled */ + VF_WANTED = 0x80, /* someone is waiting to obtain a lock */ + VF_RAW = 0x100, /* raw volume (no file system) */ + VF_LOADED = 0x200, /* module is loaded */ + VF_CONFIGURING = 0x400, /* somebody is changing the config */ + VF_WILL_CONFIGURE = 0x800, /* somebody wants to change the config */ + VF_CONFIG_INCOMPLETE = 0x1000, /* haven't finished changing the config */ + VF_CONFIG_SETUPSTATE = 0x2000, /* set a volume up if all plexes are empty */ + VF_READING_CONFIG = 0x4000, /* we're reading config database from disk */ + VF_KERNELOP = 0x8000, /* we're performing ops from kernel space */ +}; + +/* Global configuration information for the vinum subsystem */ +struct _vinum_conf { + /* Pointers to vinum structures */ + struct drive *drive; + struct sd *sd; + struct plex *plex; + struct volume *volume; + + /* the number allocated */ + int drives_allocated; + int subdisks_allocated; + int plexes_allocated; + int volumes_allocated; + + /* and the number currently in use */ + int drives_used; + int subdisks_used; + int plexes_used; + int volumes_used; + + int flags; + int opencount; /* number of times we've been opened */ +#if DEBUG + int lastrq; + struct buf *lastbuf; +#endif +}; + +/* Use these defines to simplify code */ +#define DRIVE vinum_conf.drive +#define SD vinum_conf.sd +#define PLEX vinum_conf.plex +#define VOL vinum_conf.volume +#define VFLAGS vinum_conf.flags + +/* Slice header + + * Vinum drives start with this structure: + * + * Sector + * |--------------------------------------| + * | PDP-11 memorial boot block | 0 + * |--------------------------------------| + * | Disk label, maybe | 1 + * |--------------------------------------| + * | Slice definition (vinum_hdr) | 2 + * |--------------------------------------| + * | | + * | Configuration info, first copy | 3 + * | | + * |--------------------------------------| + * | | + * | Configuration info, second copy | 3 + size of config + * | | + * |--------------------------------------| + */ + +/* Sizes and offsets of our information */ +enum { + VINUM_LABEL_OFFSET = 4096, /* offset of vinum label */ + VINUMHEADERLEN = 512, /* size of vinum label */ + VINUM_CONFIG_OFFSET = 4608, /* offset of first config copy */ + MAXCONFIG = 65536, /* and size of config copy */ + DATASTART = (MAXCONFIG * 2 + VINUM_CONFIG_OFFSET) / DEV_BSIZE /* this is where the data starts */ +}; + +/* hostname is 256 bytes long, but we don't need to shlep + * multiple copies in vinum. We use the host name just + * to identify this system, and 32 bytes should be ample + * for that purpose */ +#define VINUMHOSTNAMELEN 32 + +struct vinum_label { + char sysname[VINUMHOSTNAMELEN]; /* system name at time of creation */ + char name[MAXDRIVENAME]; /* our name of the drive */ + struct timeval date_of_birth; /* the time it was created */ + struct timeval last_update; /* and the time of last update */ + off_t drive_size; /* total size in bytes of the drive. + * This value includes the headers */ +}; + +struct vinum_hdr { + long long magic; /* we're long on magic numbers */ + /* XXX Get these right for big-endian */ +#define VINUM_MAGIC 22322600044678729LL /* should be this */ +#define VINUM_NOMAGIC 22322600044678990LL /* becomes this after obliteration */ + int config_length; /* size in bytes of each copy of the + * configuration info. + * This must be a multiple of the sector size. */ + + struct vinum_label label; /* unique label */ +}; + +/* Information returned from read_drive_label */ +enum drive_label_info { + DL_CANT_OPEN, /* invalid partition */ + DL_NOT_OURS, /* valid partition, but no vinum label */ + DL_DELETED_LABEL, /* valid partition, deleted label found */ + DL_WRONG_DRIVE, /* drive name doesn't match */ + DL_OURS /* valid partition and label found */ +}; + +/*** Drive definitions ***/ +/* A drive corresponds to a disk slice. We use a different term to show + * the difference in usage: it doesn't have to be a slice, and could + * theroretically be a complete, unpartitioned disk */ + +struct drive { + enum drivestate state; /* current state */ + int subdisks_allocated; /* number of entries in sd */ + int subdisks_used; /* and the number used */ + int blocksize; /* size of fs blocks */ + u_int64_t sectors_available; /* number of sectors still available */ + int secsperblock; + int lasterror; /* last error on drive */ + int driveno; /* index of drive in vinum_conf */ + int opencount; /* number of up subdisks */ + u_int64_t reads; /* number of reads on this drive */ + u_int64_t writes; /* number of writes on this drive */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + dev_t dev; /* and device number */ + char devicename[MAXDRIVENAME]; /* name of the slice it's on */ + struct vnode *vp; /* vnode pointer */ + struct proc *p; + struct vinum_label label; /* and the label information */ + struct partinfo partinfo; /* partition information */ + int freelist_size; /* number of entries alloced in free list */ + int freelist_entries; /* number of entries used in free list */ + struct drive_freelist { /* sorted list of free space on drive */ + u_int64_t offset; + long sectors; + } *freelist; +}; + +/*** Subdisk definitions ***/ + +struct sd { + enum sdstate state; /* state */ + /* offsets in blocks */ + int64_t driveoffset; /* offset on drive */ + int64_t plexoffset; /* offset in plex */ + u_int64_t sectors; /* and length in sectors */ + int plexno; /* index of plex, if it belongs */ + int driveno; /* index of the drive on which it is located */ + int sdno; /* our index in vinum_conf */ + int pid; /* pid of process which opened us */ + u_int64_t reads; /* number of reads on this subdisk */ + u_int64_t writes; /* number of writes on this subdisk */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + char name[MAXSDNAME]; /* name of subdisk */ +}; + +/*** Plex definitions ***/ + +/* kinds of plex organization */ +enum plexorg { + plex_disorg, /* disorganized */ + plex_concat, /* concatenated plex */ + plex_striped, /* striped plex */ + plex_raid5 /* RAID5 plex */ +}; + +/* Region in plex (either defective or unmapped) */ +struct plexregion { + u_int64_t offset; /* start of region */ + u_int64_t length; /* length */ +}; + +struct plex { + enum plexorg organization; /* Plex organization */ + enum plexstate state; /* and current state */ + u_int64_t length; /* total length of plex (max offset) */ + int flags; + int stripesize; /* size of stripe or raid band, in sectors */ + int subdisks; /* number of associated subdisks */ + int subdisks_allocated; /* number of subdisks allocated space for */ + int *sdnos; /* list of component subdisks */ + int plexno; /* index of plex in vinum_conf */ + int volno; /* index of volume */ + int volplexno; /* number of plex in volume */ + int pid; /* pid of process which opened us */ + /* Lock information */ + int locks; /* number of locks used */ + int alloclocks; /* number of locks allocated */ + struct rangelock *lock; /* ranges of locked addresses */ + /* Statistics */ + u_int64_t reads; /* number of reads on this plex */ + u_int64_t writes; /* number of writes on this plex */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + u_int64_t multiblock; /* requests that needed more than one block */ + u_int64_t multistripe; /* requests that needed more than one stripe */ + /* revive parameters */ + u_int64_t revived; /* block number of current revive request */ + int revive_blocksize; /* revive block size (bytes) */ + int revive_interval; /* and time to wait between transfers */ + struct request *waitlist; /* list of requests waiting on revive op */ + /* geometry control */ + int defective_regions; /* number of regions which are defective */ + int defective_region_count; /* number of entries in defective_region */ + struct plexregion *defective_region; /* list of offset/length pairs: defective sds */ + int unmapped_regions; /* number of regions which are missing */ + int unmapped_region_count; /* number of entries in unmapped_region */ + struct plexregion *unmapped_region; /* list of offset/length pairs: missing sds */ + char name[MAXPLEXNAME]; /* name of plex */ +}; + +/*** Volume definitions ***/ + +#define MAXPLEX 8 /* maximum number of plexes */ + + +struct volume { + enum volumestate state; /* current state */ + int plexes; /* number of plexes */ + int preferred_plex; /* plex to read from, -1 for round-robin */ + int last_plex_read; /* index of plex used for last read, + * for round-robin */ + dev_t devno; /* device number */ + int flags; /* status and configuration flags */ + int opencount; /* number of opens (all the same process) */ + int openflags; /* flags supplied to last open(2) */ + u_int64_t size; /* size of volume */ + int disk; /* disk index */ + int blocksize; /* logical block size */ + int active; /* number of outstanding requests active */ + int subops; /* and the number of suboperations */ + pid_t pid; /* pid of locker */ + /* Statistics */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + u_int64_t reads; /* number of reads on this volume */ + u_int64_t writes; /* number of writes on this volume */ + u_int64_t recovered_reads; /* reads recovered from another plex */ + /* Unlike subdisks in the plex, space for the plex pointers is static */ + int plex[MAXPLEX]; /* index of plexes */ + char name[MAXVOLNAME]; /* name of volume */ + struct disklabel label; /* for DIOCGPART */ +}; + +/* Table expansion. Expand table, which contains oldcount + * entries of type element, by increment entries, and change + * oldcount accordingly */ +#define EXPAND(table, element, oldcount, increment) \ +{ \ + expand_table ((void **) &table, \ + oldcount * sizeof (element), \ + (oldcount + increment) * sizeof (element) ); \ + oldcount += increment; \ + } + +/* Information on vinum's memory usage */ +struct meminfo { + int mallocs; /* number of malloced blocks */ + int total_malloced; /* total amount malloced */ + int highwater; /* maximum number of mallocs */ + struct mc *malloced; /* pointer to kernel table */ +}; + +struct mc { + int seq; + int size; + short line; + short flags; +#define ALLOC_KVA 1 /* allocated via kva calls */ + int *databuf; /* really vm_object_t */ + caddr_t address; + char file[16]; +}; + +/* These enums are used by the state transition + * routines. They're in bit map format: + * + * Bit 0: Other plexes in the volume are down + * Bit 1: Other plexes in the volume are up + * Bit 2: The current plex is up + * Maybe they should be local to + * state.c */ +enum volplexstate { + volplex_onlyusdown = 0, /* we're the only plex, and we're down */ + volplex_alldown, /* 1: another plex is down, and so are we */ + volplex_otherup, /* 2: another plex is up */ + volplex_otherupdown, /* other plexes are up and down */ + volplex_onlyus, /* 4: we're up and alone */ + volplex_onlyusup, /* only we are up, others are down */ + volplex_allup, /* all plexes are up */ + volplex_someup /* some plexes are up, including us */ +}; + +/* state map for plex */ +enum sdstates { + sd_emptystate = 1, + sd_downstate = 2, /* found an SD which is down */ + sd_crashedstate = 4, /* found an SD which is crashed */ + sd_obsoletestate = 8, /* found an SD which is obsolete */ + sd_stalestate = 16, /* found an SD which is stale */ + sd_rebornstate = 32, /* found an SD which is reborn */ + sd_upstate = 64, /* found an SD which is up */ + sd_initstate = 128, /* found an SD which is init */ + sd_otherstate = 256 /* found an SD in some other state */ +}; + +/* This is really just a parameter to pass to + * set_<foo>_state, but since it needs to be known + * in the external definitions, we need to define + * it here */ +enum setstateflags { + setstate_none = 0, /* no flags */ + setstate_force = 1, /* force the state change */ + setstate_configuring = 2, /* we're currently configuring, don't save */ + setstate_recursing = 4, /* we're called from another setstate function */ + setstate_norecurse = 8 /* don't call other setstate functions */ +}; + +#ifdef DEBUG +/* Debugging stuff */ +#define DEBUG_ADDRESSES 1 +#define DEBUG_NUMOUTPUT 2 +#endif diff --git a/sys/dev/vinum/COPYRIGHT b/sys/dev/vinum/COPYRIGHT new file mode 100644 index 0000000..4ee95f2 --- /dev/null +++ b/sys/dev/vinum/COPYRIGHT @@ -0,0 +1,37 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: COPYRIGHT,v 1.1 1998/03/05 06:07:05 grog Exp grog $ + */ diff --git a/sys/dev/vinum/Makefile b/sys/dev/vinum/Makefile new file mode 100644 index 0000000..4b88182 --- /dev/null +++ b/sys/dev/vinum/Makefile @@ -0,0 +1,26 @@ +# $Id: Makefile.lkm.lite,v 1.2 1998/08/13 06:07:29 grog Exp grog $ + +.PATH: ${.CURDIR}/../../sys/dev/ccd +KMOD= vinum_mod +SRCS= vinum.c vinum.h vnode_if.h parser.c config.c io.c util.c vinumhdr.h request.h \ + state.c memory.c request.c lock.c vinumext.h vinumio.h vinumkw.h \ + vinumstate.h vinumvar.h revive.c vinumioctl.c interrupt.c +NOMAN= +PSEUDO_LKM= +CFLAGS = -I. -O -g -I/usr/include/machine -DDEBUG -Wall -Wno-unused -Wno-parentheses + +CLEANFILES+= vinum.h vnode_if.h vnode_if.c + +all: + +# We don't need this, but the Makefile wants it +vinum.h: + touch $@ + +state.h: maketabs vinumstate.h + ./maketabs >state.h + +maketabs: maketabs.c + ${CC} -g -o maketabs maketabs.c + +.include <bsd.kmod.mk> diff --git a/sys/dev/vinum/makestatetext b/sys/dev/vinum/makestatetext new file mode 100755 index 0000000..ca5eb2d --- /dev/null +++ b/sys/dev/vinum/makestatetext @@ -0,0 +1,40 @@ +#!/bin/sh +# Make statetexts.h from vinumstate.h +# $Id: makestatetext,v 1.4 1998/03/13 05:36:16 grog Exp grog $ +infile=vinumstate.h +ofile=statetexts.h +cat <COPYRIGHT > $ofile + +echo >>$ofile "/* Created by $0 on" `date`. "Do not edit */" +echo >>$ofile +echo >>$ofile "/* Drive state texts */" +echo >>$ofile "char *drivestatetext [] = + { " +egrep -e 'drive_[A-z0-9]*,' <$infile | grep -v = | sed 's: *drive_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; + +/* Subdisk state texts */ +char *sdstatetext [] = + { +FOO +egrep -e 'sd_[A-z0-9]*,' $infile | grep -v = | sed 's: *sd_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; + +/* Plex state texts */ +char *plexstatetext [] = + { +FOO +egrep -e 'plex_[A-z0-9]*,' $infile | grep -v = | sed 's: *plex_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; + +/* Volume state texts */ +char *volstatetext [] = + { +FOO +egrep -e 'volume_[A-z0-9]*,' $infile | grep -v = | sed 's: *volume_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; +FOO diff --git a/sys/dev/vinum/request.h b/sys/dev/vinum/request.h new file mode 100644 index 0000000..b4beccc --- /dev/null +++ b/sys/dev/vinum/request.h @@ -0,0 +1,159 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: request.h,v 1.10 1998/08/03 07:15:26 grog Exp grog $ + */ + +/* Information needed to set up a transfer */ + +/* struct buf is surprisingly big (about 300 + * bytes), and it's part of the request, so this + * value is really important. Most requests + * don't need more than 2 subrequests per + * plex. The table is automatically extended if + * this value is too small. */ +#define RQELTS 2 /* default of 2 requests per transfer */ + +enum xferinfo { + XFR_NORMAL_READ = 1, + XFR_NORMAL_WRITE = 2, /* write request in normal mode */ + XFR_RECOVERY_READ = 4, + XFR_DEGRADED_WRITE = 8, + XFR_PARITYLESS_WRITE = 0x10, + XFR_NO_PARITY_STRIPE = 0x20, /* parity stripe is not available */ + XFR_DATA_BLOCK = 0x40, /* data block in request */ + XFR_PARITY_BLOCK = 0x80, /* parity block in request */ + XFR_BAD_SUBDISK = 0x100, /* this subdisk is dead */ + XFR_MALLOCED = 0x200, /* this buffer is malloced */ +#if DEBUG + XFR_PHASE2 = 0x800, /* documentation only: 2nd phase write */ +#endif + XFR_REVIVECONFLICT = 0x1000, /* possible conflict with a revive operation */ + /* operations that need a parity block */ + XFR_PARITYOP = (XFR_NORMAL_WRITE | XFR_RECOVERY_READ | XFR_DEGRADED_WRITE), + /* operations that use the group parameters */ + XFR_GROUPOP = (XFR_DEGRADED_WRITE | XFR_RECOVERY_READ), + /* operations that that use the data parameters */ + XFR_DATAOP = (XFR_NORMAL_READ | XFR_NORMAL_WRITE | XFR_PARITYLESS_WRITE), + /* operations requiring read before write */ + XFR_RBW = (XFR_NORMAL_WRITE | XFR_DEGRADED_WRITE), + /* operations that need a malloced buffer */ + XFR_NEEDS_MALLOC = (XFR_NORMAL_WRITE | XFR_RECOVERY_READ | XFR_DEGRADED_WRITE) +}; + +/* Describe one low-level request, part + * of a high-level request. This is an + * extended struct buf buffer, and the first + * element *must* be a struct buf. We pass this structure + * to the I/O routines instead of a struct buf in oder + * to be able to locate the high-level request when it + * completes. + * + * All offsets and lengths are in "blocks", i.e. sectors */ +struct rqelement { + struct buf b; /* buf structure */ + struct rqgroup *rqg; /* pointer to our group */ + /* Information about the transfer */ + daddr_t sdoffset; /* offset in subdisk */ + int useroffset; /* offset in user buffer of normal data */ + /* dataoffset and datalen refer to "individual" + * data transfers (normal read, parityless write) + * and also degraded write. + * + * groupoffset and grouplen refer to the other + * "group" operations (normal write, recovery read) + * Both the offsets are relative to the start of the + * local buffer */ + int dataoffset; /* offset in buffer of the normal data */ + int groupoffset; /* offset in buffer of group data */ + short datalen; /* length of normal data (sectors) */ + short grouplen; /* length of group data (sectors) */ + short buflen; /* total buffer length to allocate */ + short flags; /* really enum xferinfo (see above) */ + /* Ways to find other components */ + short sdno; /* subdisk number */ + short driveno; /* drive number */ +}; + +/* A group of requests built to satisfy a certain + * component of a user request */ +struct rqgroup { + struct rqgroup *next; /* pointer to next group */ + struct request *rq; /* pointer to the request */ + short count; /* number of requests in this group */ + short active; /* and number active */ + short plexno; /* index of plex */ + int badsdno; /* index of bad subdisk or -1 */ + enum xferinfo flags; /* description of transfer */ + struct rqelement rqe[0]; /* and the elements of this request */ +}; + +/* Describe one high-level request and the + * work we have to do to satisfy it */ +struct request { + struct buf *bp; /* pointer to the high-level request */ + int flags; + union { + int volno; /* volume index */ + int plexno; /* or plex index */ + } volplex; + int error; /* current error indication */ + short isplex; /* set if this is a plex request */ + short active; /* number of subrequests still active */ + struct rqgroup *rqg; /* pointer to the first group of requests */ + struct rqgroup *lrqg; /* and to the first group of requests */ + struct request *next; /* link of waiting requests */ +}; + +/* Extended buffer header for subdisk I/O. Includes + * a pointer to the user I/O request. */ +struct sdbuf { + struct buf b; /* our buffer */ + struct buf *bp; /* and pointer to parent */ + short driveno; /* drive index */ + short sdno; /* and subdisk index */ +}; + +/* Values returned by rqe and friends. + * Be careful with these: they are in order of increasing + * seriousness. Some routines check for > REQUEST_RECOVERED + * to indicate a completely failed request. */ +enum requeststatus { + REQUEST_OK, /* request built OK */ + REQUEST_RECOVERED, /* request OK, but involves RAID5 recovery */ + REQUEST_EOF, /* request failed: outside plex */ + REQUEST_DOWN, /* request failed: subdisk down */ + REQUEST_ENOMEM /* ran out of memory */ +}; diff --git a/sys/dev/vinum/statetexts.h b/sys/dev/vinum/statetexts.h new file mode 100644 index 0000000..b7ce4eb --- /dev/null +++ b/sys/dev/vinum/statetexts.h @@ -0,0 +1,88 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: COPYRIGHT,v 1.1 1998/03/05 06:07:05 grog Exp grog $ + */ +/* Created by ./makestatetext on Tue 4 Aug 15:53:16 CST 1998. Do not edit */ + +/* Drive state texts */ +char *drivestatetext[] = +{ + "unallocated", + "uninit", + "down", + "coming_up", + "up", +}; + +/* Subdisk state texts */ +char *sdstatetext[] = +{ + "unallocated", + "uninit", + "init", + "initializing", + "empty", + "obsolete", + "stale", + "crashed", + "down", + "reborn", + "up", +}; + +/* Plex state texts */ +char *plexstatetext[] = +{ + "unallocated", + "init", + "faulty", + "down", + "reviving", + "initializing", + "corrupt", + "degraded", + "flaky", + "up", +}; + +/* Volume state texts */ +char *volstatetext[] = +{ + "unallocated", + "uninit", + "down", + "up", +}; diff --git a/sys/dev/vinum/vinum.c b/sys/dev/vinum/vinum.c new file mode 100644 index 0000000..bffdf6d --- /dev/null +++ b/sys/dev/vinum/vinum.c @@ -0,0 +1,512 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinum.c,v 1.19 1998/08/13 05:24:02 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "sys/sysproto.h" /* for sync(2) */ +#ifdef DEBUG +#include <sys/reboot.h> +int debug = 0; +#endif + +/* pointer to ioctl p parameter, to save passing it around */ +struct proc *myproc; + +#if __FreeBSD__ < 3 +STATIC struct cdevsw vinum_cdevsw; +STATIC struct bdevsw vinum_bdevsw = +{ + vinumopen, vinumclose, vinumstrategy, vinumioctl, + vinumdump, vinumsize, 0, + "vinum", &vinum_cdevsw, -1 +}; +#else /* goodbye, bdevsw */ +STATIC struct cdevsw vinum_cdevsw = +{ + vinumopen, vinumclose, vinumread, vinumwrite, + vinumioctl, nostop, nullreset, nodevtotty, + seltrue, nommap, vinumstrategy, "vinum", + NULL, -1, vinumdump, vinumsize, + D_DISK, 0, -1 +}; +#endif + +/* Called by main() during pseudo-device attachment. */ +STATIC void vinumattach(void *); + +STATIC void vinumgetdisklabel(dev_t); +void vinum_scandisk(void); +int vinum_inactive(void); +void free_vinum(int); + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +extern jmp_buf command_fail; /* return here if config fails */ + +struct _vinum_conf vinum_conf; /* configuration information */ + +STATIC int vinum_devsw_installed = 0; + +/* + * Called by main() during pseudo-device attachment. All we need + * to do is allocate enough space for devices to be configured later, and + * add devsw entries. + */ +void +vinumattach(void *dummy) +{ + BROKEN_GDB; + char *buf; /* pointer to temporary buffer */ + struct _ioctl_reply *ioctl_reply; /* struct to return */ + struct uio uio; + struct iovec iovec; + + /* modload should prevent multiple loads, so this is worth a panic */ + if ((vinum_conf.flags & VF_LOADED) != NULL) + panic("vinum: already loaded"); + + printf("vinum: loaded\n"); + vinum_conf.flags |= VF_LOADED; /* we're loaded now */ + + /* We don't have a p pointer here, so take it from curproc */ + myproc = curproc; +#if __FreeBSD__ < 3 + bdevsw_add_generic(BDEV_MAJOR, CDEV_MAJOR, &vinum_bdevsw); +#else + cdevsw_add_generic(BDEV_MAJOR, CDEV_MAJOR, &vinum_cdevsw); +#endif +#ifdef DEVFS +#error DEVFS not finished yet +#endif + + uio.uio_iov = &iovec; + uio.uio_iovcnt = 1; /* just one buffer */ + uio.uio_offset = 0; /* start at the beginning */ + uio.uio_resid = 512; /* one sector */ + uio.uio_segflg = UIO_SYSSPACE; /* we're in system space */ + uio.uio_rw = UIO_READ; /* do we need this? */ + uio.uio_procp = curproc; /* do it for our own process */ + + iovec.iov_len = 512; + buf = (char *) Malloc(iovec.iov_len); /* get a buffer */ + CHECKALLOC(buf, "vinum: no memory\n"); /* can't get 512 bytes? */ + iovec.iov_base = buf; /* read into buf */ + + /* allocate space: drives... */ + DRIVE = (struct drive *) Malloc(sizeof(struct drive) * INITIAL_DRIVES); + CHECKALLOC(DRIVE, "vinum: no memory\n"); + vinum_conf.drives_allocated = INITIAL_DRIVES; /* number of drive slots allocated */ + vinum_conf.drives_used = 0; /* and number in use */ + + /* volumes, ... */ + VOL = (struct volume *) Malloc(sizeof(struct volume) * INITIAL_VOLUMES); + CHECKALLOC(VOL, "vinum: no memory\n"); + vinum_conf.volumes_allocated = INITIAL_VOLUMES; /* number of volume slots allocated */ + vinum_conf.volumes_used = 0; /* and number in use */ + + /* plexes, ... */ + PLEX = (struct plex *) Malloc(sizeof(struct plex) * INITIAL_PLEXES); + CHECKALLOC(PLEX, "vinum: no memory\n"); + vinum_conf.plexes_allocated = INITIAL_PLEXES; /* number of plex slots allocated */ + vinum_conf.plexes_used = 0; /* and number in use */ + + /* and subdisks */ + SD = (struct sd *) Malloc(sizeof(struct sd) * INITIAL_SUBDISKS); + CHECKALLOC(SD, "vinum: no memory\n"); + vinum_conf.subdisks_allocated = INITIAL_SUBDISKS; /* number of sd slots allocated */ + vinum_conf.subdisks_used = 0; /* and number in use */ + + ioctl_reply = NULL; /* no reply on longjmp */ +} + + +#ifdef ACTUALLY_LKM_NOT_KERNEL /* stuff for LKMs */ + +/* Check if we have anything open. If so, return 0 (not inactive), + * otherwise 1 (inactive) */ +int +vinum_inactive(void) +{ + BROKEN_GDB; + int i; + int can_do = 1; /* assume we can do it */ + + lock_config(); + for (i = 0; i < vinum_conf.volumes_used; i++) { + if (VOL[i].pid != NULL) { /* volume is open */ + can_do = 0; + break; + } + } + unlock_config(); + return can_do; +} + +/* Free all structures. + * If cleardrive is 0, save the configuration; otherwise + * remove the configuration from the drive. + * + * Before coming here, ensure that no volumes are open. + */ +void +free_vinum(int cleardrive) +{ + BROKEN_GDB; + int i; + + if (cleardrive) { + for (i = 0; i < vinum_conf.drives_used; i++) + remove_drive(i); /* remove the drive */ + } else { /* keep the config */ + save_config(); + if (DRIVE != NULL) { + for (i = 0; i < vinum_conf.drives_used; i++) + free_drive(&DRIVE[i]); /* close files and things */ + Free(DRIVE); + } + } + if (SD != NULL) + Free(SD); + if (PLEX != NULL) { + for (i = 0; i < vinum_conf.plexes_used; i++) { + struct plex *plex = &vinum_conf.plex[i]; + + if (plex->state != plex_unallocated) { /* we have real data there */ + if (plex->sdnos) + Free(plex->sdnos); + if (plex->unmapped_regions) + Free(plex->unmapped_region); + if (plex->defective_regions) + Free(plex->defective_region); + } + } + Free(PLEX); + } + if (VOL != NULL) + Free(VOL); + bzero(&vinum_conf, sizeof(vinum_conf)); +} + +MOD_MISC(vinum); + +/* + * Function called when loading the driver. + */ +STATIC int +vinum_load(struct lkm_table *lkmtp, int cmd) +{ + BROKEN_GDB; +/* Debugger ("vinum_load"); */ + vinumattach(NULL); + return 0; /* OK */ +} + +/* + * Function called when unloading the driver. + */ +STATIC int +vinum_unload(struct lkm_table *lkmtp, int cmd) +{ + BROKEN_GDB; + if (vinum_inactive()) { /* is anything open? */ + struct sync_args dummyarg = + {0}; +#if __FreeBSD__ < 3 + int retval; +#endif + + printf("vinum: unloaded\n"); +#if __FreeBSD__ < 3 + sync(curproc, &dummyarg, &retval); /* write out buffers */ +#else + sync(curproc, &dummyarg); /* write out buffers */ +#endif + free_vinum(0); /* no: clean up */ +#if __FreeBSD__ < 3 + bdevsw[BDEV_MAJOR] = NULL; /* clear bdevsw */ +#endif + cdevsw[CDEV_MAJOR] = NULL; /* and cdevsw */ + return 0; + } else + return EBUSY; +} + +/* + * Dispatcher function for the module (load/unload/stat). + */ +int +vinum_mod(struct lkm_table *lkmtp, int cmd, int ver) +{ + BROKEN_GDB; + MOD_DISPATCH(vinum, /* module name */ + lkmtp, /* LKM table */ + cmd, /* command */ + ver, + vinum_load, /* load with this function */ + vinum_unload, /* and unload with this */ + lkm_nullcmd); +} + +#else /* not LKM */ +#error "This driver must be compiled as a loadable kernel module" +#endif /* LKM */ + +/* ARGSUSED */ +/* Open a vinum object + * At the moment, we only open volumes and the + * super device. It's a nice concept to be + * able to open drives, subdisks and plexes, but + * I can't think what good it could be */ +int +vinumopen(dev_t dev, + int flags, + int fmt, + struct proc *p) +{ + BROKEN_GDB; + int s; /* spl */ + int error; + unsigned int index; + struct volume *vol; + struct plex *plex; + struct sd *sd; + struct devcode *device; + + device = (struct devcode *) &dev; + + error = 0; + /* First, decide what we're looking at */ + switch (device->type) { + case VINUM_VOLUME_TYPE: + index = VOLNO(dev); + if (index >= vinum_conf.volumes_used) + return ENXIO; /* no such device */ + vol = &VOL[index]; + + switch (vol->state) { + case volume_unallocated: + case volume_uninit: + return ENXIO; + + case volume_up: + s = splhigh(); /* quick lock */ + if (error) + return error; + if (vol->opencount == 0) + vol->openflags = flags; /* set our flags */ + vol->opencount++; + vol->pid = p->p_pid; /* and say who we are (do we need this? XXX) */ + splx(s); + return 0; + + case volume_down: + return EIO; + + default: + return EINVAL; + } + + case VINUM_PLEX_TYPE: + if (VOLNO(dev) >= vinum_conf.volumes_used) + return ENXIO; + index = PLEXNO(dev); /* get plex index in vinum_conf */ + if (index >= vinum_conf.plexes_used) + return ENXIO; /* no such device */ + plex = &PLEX[index]; + + switch (plex->state) { + case plex_unallocated: + return EINVAL; + + default: + s = splhigh(); + if (plex->pid /* it's open already */ + && (plex->pid != p->p_pid)) { /* and not by us, */ + splx(s); + return EBUSY; /* one at a time, please */ + } + plex->pid = p->p_pid; /* and say who we are (do we need this? XXX) */ + splx(s); + return 0; + } + + case VINUM_SD_TYPE: + if ((VOLNO(dev) >= vinum_conf.volumes_used) || /* no such volume */ + (PLEXNO(dev) >= vinum_conf.plexes_used)) /* or no such plex */ + return ENXIO; /* no such device */ + index = SDNO(dev); /* get the subdisk number */ + if (index >= vinum_conf.subdisks_used) + return ENXIO; /* no such device */ + sd = &SD[index]; + + /* Opening a subdisk is always a special operation, so we + * ignore the state as long as it represents a real subdisk */ + switch (sd->state) { + case sd_unallocated: + case sd_uninit: + return EINVAL; + + default: + s = splhigh(); + if (sd->pid /* it's open already */ + && (sd->pid != p->p_pid)) { /* and not by us, */ + splx(s); + return EBUSY; /* one at a time, please */ + } + sd->pid = p->p_pid; /* and say who we are (do we need this? XXX) */ + splx(s); + return 0; + } + + case VINUM_DRIVE_TYPE: + default: + return ENODEV; /* don't know what to do with these */ + + case VINUM_SUPERDEV_TYPE: + if (p->p_ucred->cr_uid == 0) { /* root calling, */ + vinum_conf.opencount++; /* one more opener */ + return 0; /* no worries opening super dev */ + } else + return EPERM; /* you can't do that! */ + } +} + +/* ARGSUSED */ +int +vinumclose(dev_t dev, + int flags, + int fmt, + struct proc *p) +{ + BROKEN_GDB; + unsigned int index; + struct volume *vol; + struct plex *plex; + struct sd *sd; + struct devcode *device = (struct devcode *) &dev; + + index = VOLNO(dev); + /* First, decide what we're looking at */ + switch (device->type) { + case VINUM_VOLUME_TYPE: + if (index >= vinum_conf.volumes_used) + return ENXIO; /* no such device */ + vol = &VOL[index]; + + switch (vol->state) { + case volume_unallocated: + case volume_uninit: + return ENXIO; + + case volume_up: + vol->opencount = 0; /* reset our flags */ + vol->pid = NULL; /* and forget who owned us */ + return 0; + + case volume_down: + return EIO; + + default: + return EINVAL; + } + + case VINUM_PLEX_TYPE: + if (VOLNO(dev) >= vinum_conf.volumes_used) + return ENXIO; + index = PLEXNO(dev); /* get plex index in vinum_conf */ + if (index >= vinum_conf.plexes_used) + return ENXIO; /* no such device */ + plex = &PLEX[index]; + plex->pid = 0; + return 0; + + case VINUM_SD_TYPE: + if ((VOLNO(dev) >= vinum_conf.volumes_used) || /* no such volume */ + (PLEXNO(dev) >= vinum_conf.plexes_used)) /* or no such plex */ + return ENXIO; /* no such device */ + index = SDNO(dev); /* get the subdisk number */ + if (index >= vinum_conf.subdisks_used) + return ENXIO; /* no such device */ + sd = &SD[index]; + sd->pid = 0; + return 0; + + case VINUM_SUPERDEV_TYPE: + if (p->p_ucred->cr_uid == 0) /* root calling, */ + vinum_conf.opencount--; /* one less opener */ + return 0; /* no worries closing super dev */ + + case VINUM_DRIVE_TYPE: + default: + return ENODEV; /* don't know what to do with these */ + } +} + +/* size routine */ +int +vinumsize(dev_t dev) +{ + BROKEN_GDB; + struct volume *vol; + int size; + +/* XXX This is bogus. We don't need to open + * a device to find its size */ + vol = &VOL[VOLNO(dev)]; + + if (vol->state == volume_up) + size = vol->size; + else + return 0; /* err on the size of conservatism */ + + + return size; +} + +int +vinumdump(dev_t dev) +{ + /* Not implemented. */ + return ENXIO; +} diff --git a/sys/dev/vinum/vinumconfig.c b/sys/dev/vinum/vinumconfig.c new file mode 100644 index 0000000..fd31c5e --- /dev/null +++ b/sys/dev/vinum/vinumconfig.c @@ -0,0 +1,1712 @@ +/* To do: + + * Don't store drive configuration on the config DB: read each drive's header + * to decide where it is. + * + * Accept any old crap in the config_<foo> functions, and complain when + * we try to bring it up. + * + * When trying to bring volumes up, check that the complete address range + * is covered. + */ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: config.c,v 1.17 1998/08/14 04:49:26 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#define REALLYKERNEL +#include "vinumhdr.h" + +extern jmp_buf command_fail; /* return on a failed command */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +void longjmp(jmp_buf, int); +#endif + +#define MAXTOKEN 64 /* maximum number of tokens in a line */ + +/* We can afford the luxury of global variables here, + * since start_config ensures that these functions + * are single-threaded. */ + +/* These are indices in vinum_conf of the last-mentioned of each kind of object */ +static int current_drive = -1; /* note the last drive we mention, for + * some defaults */ +static int current_plex = -1; /* and the same for the last plex */ +static int current_volume = -1; /* and the last volme */ +static struct _ioctl_reply *ioctl_reply; /* struct to return via ioctl */ + + +/* These values are used by most of these routines, so set them as globals */ +static char *token[MAXTOKEN]; /* pointers to individual tokens */ +static int tokens; /* number of tokens */ + +#define TOCONS 0x01 +#define TOTTY 0x02 +#define TOLOG 0x04 + +struct putchar_arg { + int flags; + struct tty *tty; +}; + +#define MSG_MAX 1024 /* maximum length of a formatted message */ +/* Format an error message and return to the user in the reply. + * CARE: This routine is designed to be called only from the + * configuration routines, so it assumes it's the owner of + * the configuration lock, and unlocks it on exit */ +void +throw_rude_remark(int error, char *msg,...) +{ + BROKEN_GDB; + int retval; + va_list ap; + char *text; + static int finishing; /* don't recurse */ + int was_finishing; + + va_start(ap, msg); + if ((ioctl_reply != NULL) /* we're called from the user */ + &&(!(vinum_conf.flags & VF_KERNELOP))) { /* and we're not doing kernel things: return msg */ + /* XXX We can't just format to ioctl_reply, since it + * may contain our input parameters */ + text = Malloc(MSG_MAX); + if (text == NULL) { + printf("vinum: can't allocate error message buffer"); + printf("vinum: "); + vprintf(msg, ap); /* print to the console */ + printf("\n"); + } else { + retval = kvprintf(msg, NULL, (void *) text, 10, ap); + text[retval] = '\0'; /* delimit */ + strcpy(ioctl_reply->msg, text); + ioctl_reply->error = error; /* first byte is the error number */ + Free(text); + } + } else { + printf("vinum: "); + vprintf(msg, ap); /* print to the console */ + printf("\n"); + } + va_end(ap); + + if (vinum_conf.flags & VF_READING_CONFIG) /* go through to the bitter end, */ + return; + /* We have a problem here: we want to unlock the + * configuration, which implies tidying up, but + * if we find an error while tidying up, we could + * recurse for ever. Use this kludge to only try + * once */ + was_finishing = finishing; + finishing = 1; + finish_config(was_finishing); /* unlock anything we may be holding */ + finishing = was_finishing; + longjmp(command_fail, error); +} + +/* Function declarations */ +int atoi(char *); /* no atoi in the kernel */ + +/* Minimal version of atoi */ +int +atoi(char *s) +{ /* no atoi in the kernel */ + BROKEN_GDB; + int r = 0; + int sign = 1; + + while (((*s >= '0') && (*s <= '9')) || (*s == '-')) { + if (*s == '-') + sign = -sign; + else + r = r * 10 + (*s - '0'); + } + return r; +} + +/* Find index of volume in vinum_conf. Return the index + * if found, or -1 if not */ +int +volume_index(struct volume *vol) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.volumes_used; i++) + if (&VOL[i] == vol) + return i; + return -1; +} + +/* Find index of plex in vinum_conf. Return the index + * if found, or -1 if not */ +int +plex_index(struct plex *plex) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.plexes_used; i++) + if (&PLEX[i] == plex) + return i; + return -1; +} + +/* Find index of subdisk in vinum_conf. Return the index + * if found, or -1 if not */ +int +sd_index(struct sd *sd) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.subdisks_used; i++) + if (&SD[i] == sd) + return i; + return -1; +} + +/* Find index of drive in vinum_conf. Return the index + * if found, or -1 if not */ +int +drive_index(struct drive *drive) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.drives_used; i++) + if (&DRIVE[i] == drive) + return i; + return -1; +} + +/* Check a volume to see if the plex is already assigned to it. + * Return index in volume->plex, or -1 if not assigned */ +int +my_plex(int volno, int plexno) +{ + BROKEN_GDB; + int i; + struct volume *vol; + + vol = &VOL[volno]; /* point to volno */ + for (i = 0; i < vol->plexes; i++) + if (vol->plex[i] == plexno) + return i; + return -1; /* not found */ +} + +/* Check a plex to see if the subdisk is already assigned to it. + * Return index in plex->sd, or -1 if not assigned */ +int +my_sd(int plexno, int sdno) +{ + BROKEN_GDB; + int i; + struct plex *plex; + + plex = &PLEX[plexno]; + for (i = 0; i < plex->subdisks; i++) + if (plex->sdnos[i] == sdno) + return i; + return -1; /* not found */ +} + +/* Check that this operation is being done in the kernel. + * longjmp out if not. op the name of the operation. */ +void +checkkernel(char *op) +{ + BROKEN_GDB; + if (vinum_conf.flags & VF_KERNELOP == 0) + throw_rude_remark(EPERM, "Can't perform '%s' from user space", op); +} + +/* Add plex to the volume if possible */ +int +give_plex_to_volume(int volno, int plexno) +{ + BROKEN_GDB; + struct volume *vol; + + /* XXX It's not an error for the plex to already + * belong to the volume, but we need to check a + * number of things to make sure it's done right. + * Some day. */ + if (my_plex(volno, plexno) >= 0) + return plexno; /* that's it */ + + vol = &VOL[volno]; /* point to volume */ + if (vol->plexes == MAXPLEX) /* all plexes allocated */ + throw_rude_remark(ENOSPC, + "Too many plexes for volume %s", + vol->name); + vol->plex[vol->plexes] = plexno; /* this one */ + vol->plexes++; /* add another plex */ + PLEX[plexno].volno = volno; /* note the number of our volume */ + + return vol->plexes - 1; /* and return its index */ +} + +/* Add subdisk to a plex if possible */ +int +give_sd_to_plex(int plexno, int sdno) +{ + BROKEN_GDB; + int i; + struct plex *plex; + struct sd *sd; + + /* XXX It's not an error for the sd to already + * belong to the plex, but we need to check a + * number of things to make sure it's done right. + * Some day. */ + i = my_sd(plexno, sdno); + if (i >= 0) /* does it already belong to us? */ + return i; /* that's it */ + + plex = &PLEX[plexno]; /* point to the plex */ + sd = &SD[sdno]; /* and the subdisk */ + + /* Do we have an offset? Otherwise put it after the last one */ + if (sd->plexoffset < 0) { /* no offset specified */ + if (plex->subdisks > 0) { + struct sd *lastsd = &SD[plex->sdnos[plex->subdisks - 1]]; /* last subdisk */ + sd->plexoffset = lastsd->sectors + lastsd->plexoffset; /* take it */ + } else /* first subdisk */ + sd->plexoffset = 0; /* start at the beginning */ + } + plex->subdisks++; /* another entry */ + if (plex->subdisks >= plex->subdisks_allocated) /* need more space */ + EXPAND(plex->sdnos, int, plex->subdisks_allocated, INITIAL_SUBDISKS_IN_PLEX); + + /* XXX I'm not sure this makes any sense + * for anything except concatenated plexes, + * and it comes up with the wrong answer for + * RAID-5 plexes, but it's currently needed + * for the calculations. We'll adjust for + * RAID-5 in config_plex */ + if ((sd->sectors + sd->plexoffset) > plex->length) { /* gone beyond the end of the plex */ + plex->length = sd->sectors + sd->plexoffset; /* adjust the length */ + + if ((plex->volno >= 0) /* we have a volume */ + &&(plex->length > VOL[plex->volno].size)) /* and we're now the longest plex */ + VOL[plex->volno].size = plex->length; /* increase the size of the volume */ + } + /* We need to check that the subdisks don't overlap, + * but we can't do that until a point where we *must* + * know the size of all the subdisks. That's not + * here. But we need to sort them by offset */ + for (i = 0; i < plex->subdisks - 1; i++) { + if (sd->plexoffset < SD[plex->sdnos[i]].plexoffset) { /* it fits before this one */ + /* First move any remaining subdisks by one */ + int j; + + for (j = plex->subdisks - 1; j > i; j--) /* move up one at a time */ + plex->sdnos[j] = plex->sdnos[j - 1]; + plex->sdnos[i] = sdno; + return i; + } + } + + /* The plex doesn't have any subdisk with a larger + * offset. Insert it */ + plex->sdnos[i] = sdno; + return i; +} + +/* Add a subdisk to drive if possible. The pointer to the drive + * must already be stored in the sd structure, but the drive + * doesn't know about the subdisk yet. */ +static void +give_sd_to_drive(int sdno) +{ + BROKEN_GDB; + struct sd *sd; /* pointer to subdisk */ + struct drive *drive; /* and drive */ + int fe; /* index in free list */ + + sd = &SD[sdno]; /* point to sd */ + drive = &DRIVE[sd->driveno]; /* and drive */ + + if (drive->state != drive_up) /* not up */ + throw_rude_remark(EIO, "Drive %s is not accessible", drive->label.name); + else if (sd->sectors > drive->sectors_available) { /* too big, */ + sd->driveoffset = -1; /* don't be confusing */ + throw_rude_remark(ENOSPC, "No space for %s on %s", sd->name, drive->label.name); + } + drive->subdisks_used++; /* one more subdisk */ + + /* no offset specified, find one */ + if (sd->driveoffset < 0) { + for (fe = 0; fe < drive->freelist_entries; fe++) { + if (drive->freelist[fe].sectors >= sd->sectors) { /* it'll fit here */ + sd->driveoffset = drive->freelist[fe].offset; + if (sd->sectors == drive->freelist[fe].sectors) { /* used up the entire entry */ + if (fe < (drive->freelist_entries - 1)) /* not the last one, */ + bcopy(&drive->freelist[fe + 1], + &drive->freelist[fe], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries--; /* one less entry */ + } else { + drive->freelist[fe].sectors -= sd->sectors; /* this much less space */ + drive->freelist[fe].offset += sd->sectors; /* this much further on */ + } + drive->sectors_available -= sd->sectors; /* and note how much less space we have */ + break; + } + } + if (fe == drive->freelist_entries) + /* Didn't find anything. Although the drive has + * enough space, it's too fragmented */ + { + sd->driveoffset = -1; /* don't be confusing */ + throw_rude_remark(ENOSPC, "No space for %s on %s", sd->name, drive->label.name); + } + } else { /* specific offset */ + /* For a specific offset to work, the space must be + * entirely in a single freelist entry. Look for it. */ + u_int64_t sdend = sd->driveoffset + sd->sectors; /* end of our subdisk */ + for (fe = 0; fe < drive->freelist_entries; fe++) { + u_int64_t dend = drive->freelist[fe].offset + drive->freelist[fe].sectors; /* end of entry */ + if (dend >= sdend) { /* fits before here */ + if (drive->freelist[fe].offset > sd->driveoffset) /* starts after the beginning of sd area */ + throw_rude_remark(ENOSPC, + "No space for subdisk %s on drive %s at offset %qd\n", + sd->name, + drive->label.name); + + /* We've found the space, and we can allocate it. + * We don't need to say that to the subdisk, which + * already knows about it. We need to tell it to + * the free list, though. We have four possibilities: + * + * 1. The subdisk exactly eats up the entry. That's the + * same as above. + * 2. The subdisk starts at the beginning and leaves space + * at the end. + * 3. The subdisk starts after the beginning and leaves + * space at the end as well: we end up with another + * fragment. + * 4. The subdisk leaves space at the beginning and finishes + * at the end. + */ + drive->sectors_available -= sd->sectors; /* note how much less space we have */ + if (sd->driveoffset == drive->freelist[fe].offset) { /* 1 or 2 */ + if (sd->sectors == drive->freelist[fe].sectors) { /* 1: used up the entire entry */ + if (fe < (drive->freelist_entries - 1)) /* not the last one, */ + bcopy(&drive->freelist[fe + 1], + &drive->freelist[fe], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries--; /* one less entry */ + } else { /* 2: space at the end */ + drive->freelist[fe].sectors -= sd->sectors; /* this much less space */ + drive->freelist[fe].offset += sd->sectors; /* this much further on */ + } + } else { /* 3 or 4 */ + drive->freelist[fe].sectors = sd->driveoffset - drive->freelist[fe].offset; + if (dend > sdend) { /* 3: space at the end as well */ + if (fe < (drive->freelist_entries - 1)) /* not the last one */ + bcopy(&drive->freelist[fe], /* move the rest down */ + &drive->freelist[fe + 1], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries++; /* one less entry */ + drive->freelist[fe + 1].offset = sdend; /* second entry starts after sd */ + drive->freelist[fe + 1].sectors = dend - sdend; /* and is this long */ + } + } + break; + } + } + } + drive->opencount++; /* one more subdisk attached */ +} + +/* Get an empty drive entry from the drive table */ +int +get_empty_drive(void) +{ + BROKEN_GDB; + int driveno; + struct drive *drive; + + /* first see if we have one which has been deallocated */ + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + if (DRIVE[driveno].state == drive_unallocated) /* bingo */ + break; + } + + if (driveno >= vinum_conf.drives_used) + /* Couldn't find a deallocated drive. Allocate a new one */ + { + vinum_conf.drives_used++; + if (vinum_conf.drives_used > vinum_conf.drives_allocated) /* we've used all our allocation */ + EXPAND(DRIVE, struct drive, vinum_conf.drives_allocated, INITIAL_DRIVES); + } + /* got a drive entry. Make it pretty */ + drive = &DRIVE[driveno]; + bzero(drive, sizeof(struct drive)); + drive->driveno = driveno; /* put number in structure */ + return driveno; /* return the index */ +} + +/* Find the named drive in vinum_conf.drive, return a pointer + * return the index in vinum_conf.drive. + * Don't mark the drive as allocated (XXX MP) + * If create != 0, create an entry if it doesn't exist + */ +/* XXX check if we have it open from attach */ +int +find_drive(const char *name, int create) +{ + BROKEN_GDB; + int driveno; + struct drive *drive; + + if (name != NULL) { + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + drive = &DRIVE[driveno]; /* point to drive */ + if ((drive->label.name[0] != '\0') /* it has a name */ + &&(strcmp(drive->label.name, name) == 0)) /* and it's this one: found */ + return driveno; + } + } + /* the drive isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + driveno = get_empty_drive(); + drive = &DRIVE[driveno]; + if (name != NULL) + bcopy(name, /* put in its name */ + drive->label.name, + min(sizeof(drive->label.name), + strlen(name))); + drive->state = drive_uninit; /* in use, nothing worthwhile there */ + return driveno; /* return the index */ +} + +/* Find a drive given its device name. + * devname must be valid. + * Otherwise the same as find_drive above */ +int +find_drive_by_dev(const char *devname, int create) +{ + BROKEN_GDB; + int driveno; + struct drive *drive; + + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + drive = &DRIVE[driveno]; /* point to drive */ + if ((drive->label.name[0] != '\0') /* it has a name */ + &&(strcmp(drive->label.name, devname) == 0)) /* and it's this one: found */ + return driveno; + } + + /* the drive isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + driveno = get_empty_drive(); + drive = &DRIVE[driveno]; + bcopy(devname, /* put in its name */ + drive->devicename, + min(sizeof(drive->devicename), + strlen(devname))); + drive->state = drive_uninit; /* in use, nothing worthwhile there */ + return driveno; /* return the index */ +} + +/* Find an empty subdisk in the subdisk table */ +int +get_empty_sd(void) +{ + BROKEN_GDB; + int sdno; + struct sd *sd; + + /* first see if we have one which has been deallocated */ + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) { + if (SD[sdno].state == sd_unallocated) /* bingo */ + break; + } + + if (sdno >= vinum_conf.subdisks_used) { /* No unused sd found. Allocate a new one */ + vinum_conf.subdisks_used++; + if (vinum_conf.subdisks_used > vinum_conf.subdisks_allocated) + EXPAND(SD, struct sd, vinum_conf.subdisks_allocated, INITIAL_SUBDISKS); + } + /* initialize some things */ + sd = &SD[sdno]; /* point to it */ + bzero(sd, sizeof(struct sd)); /* initialize */ + sd->plexno = -1; /* no plex */ + sd->driveno = -1; /* and no drive */ + sd->plexoffset = -1; /* and no offsets */ + sd->driveoffset = -1; + return sdno; /* return the index */ +} + +/* return a drive to the free pool */ +void +free_drive(struct drive *drive) +{ + BROKEN_GDB; + if (drive->vp != NULL) /* device open */ + vn_close(drive->vp, FREAD | FWRITE, FSCRED, drive->p); + bzero(drive, sizeof(struct drive)); /* this also sets drive_unallocated */ +} + +/* Find the named subdisk in vinum_conf.sd. + + * If create != 0, create an entry if it doesn't exist + * + * Return index in vinum_conf.sd + */ +int +find_subdisk(const char *name, int create) +{ + BROKEN_GDB; + int sdno; + struct sd *sd; + + for (sdno = 0; sdno < vinum_conf.subdisks_allocated; sdno++) { + if (strcmp(SD[sdno].name, name) == 0) /* found it */ + return sdno; + } + + /* the subdisk isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + /* Allocate one and insert the name */ + sdno = get_empty_sd(); + sd = &SD[sdno]; + bcopy(name, sd->name, min(sizeof(sd->name), strlen(name))); /* put in its name */ + return sdno; /* return the pointer */ +} + +/* Free an allocated sd entry + * This performs memory management only. remove() + * is responsible for checking relationships. + */ +void +free_sd(int sdno) +{ + BROKEN_GDB; + struct sd *sd; + struct drive *drive; + int fe; /* free list entry */ + u_int64_t sdend; /* end of our subdisk */ + u_int64_t dend; /* end of our freelist entry */ + + sd = &SD[sdno]; + if ((sd->driveno >= 0) /* we have a drive, */ + &&(sd->sectors > 0)) { /* and some space on it */ + drive = &DRIVE[sd->driveno]; + sdend = sd->driveoffset + sd->sectors; /* end of our subdisk */ + + /* Look for where to return the sd address space */ + for (fe = 0; + (fe < drive->freelist_entries) && (drive->freelist[fe].offset < sd->driveoffset); + fe++); + /* Now we are pointing to the last entry, the first + * with a higher offset than the subdisk, or both. */ + if ((fe > 1) /* not the first entry */ + &&((fe == drive->freelist_entries) /* gone past the end */ + ||(drive->freelist[fe].offset > sd->driveoffset))) /* or past the block were looking for */ + fe--; /* point to the block before */ + dend = drive->freelist[fe].offset + drive->freelist[fe].sectors; /* end of the entry */ + + /* At this point, we are pointing to the correct + * place in the free list. A number of possibilities + * exist: + * + * 1. The block to be freed immediately follows + * the block to which we are pointing. Just + * enlarge it. + * 2. The block to be freed starts at the end of + * the current block and ends at the beginning + * of the following block. Merge the three + * areas into a single block. + * 3. The block to be freed starts after the end + * of the block and ends before the start of + * the following block. Create a new free block. + * 4. The block to be freed starts after the end + * of the block, but ends at the start of the + * following block. Enlarge the following block + * downwards. + * + */ + if (sd->driveoffset == dend) { /* it starts after the end of this block */ + if ((fe < drive->freelist_entries - 1) /* we're not the last block in the free list */ + &&(sdend == drive->freelist[fe + 1].offset)) { /* and the subdisk ends at the start of the + * next block */ + drive->freelist[fe].sectors = drive->freelist[fe + 1].sectors; /* 2: merge all three blocks */ + if (fe < drive->freelist_entries - 2) /* still more blocks after next */ + bcopy(&drive->freelist[fe + 2], /* move down one */ + &drive->freelist[fe + 1], + (drive->freelist_entries - 2 - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries--; /* one less entry in the free list */ + } else /* 1: just enlarge this block */ + drive->freelist[fe].sectors += sd->sectors; + } else { + if (sd->driveoffset > dend) /* it starts after this block */ + fe++; /* so look at the next block */ + if ((fe < drive->freelist_entries) /* we're not the last block in the free list */ + &&(sdend == drive->freelist[fe].offset)) { /* and the subdisk ends at the start of + * this block: case 4 */ + drive->freelist[fe].offset = sd->driveoffset; /* it starts where the sd was */ + drive->freelist[fe].sectors += sd->sectors; /* and it's this much bigger */ + } else { /* case 3: non-contiguous */ + if (fe < drive->freelist_entries) /* not after the last block, */ + bcopy(&drive->freelist[fe], /* move the rest up one entry */ + &drive->freelist[fe + 1], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries++; /* one less entry */ + drive->freelist[fe].offset = sd->driveoffset; /* this entry represents the sd */ + drive->freelist[fe].sectors = sd->sectors; + } + } + drive->opencount--; /* one less subdisk attached */ + } + bzero(sd, sizeof(struct sd)); /* and clear it out */ + sd->state = sd_unallocated; +} + +/* Find an empty plex in the plex table */ +int +get_empty_plex(void) +{ + BROKEN_GDB; + int plexno; + struct plex *plex; /* if we allocate one */ + + /* first see if we have one which has been deallocated */ + for (plexno = 0; plexno < vinum_conf.plexes_used; plexno++) { + if (PLEX[plexno].state == plex_unallocated) /* bingo */ + break; /* and get out of here */ + } + + if (plexno >= vinum_conf.plexes_used) { + /* Couldn't find a deallocated plex. Allocate a new one */ + vinum_conf.plexes_used++; + if (vinum_conf.plexes_used > vinum_conf.plexes_allocated) + EXPAND(PLEX, struct plex, vinum_conf.plexes_allocated, INITIAL_PLEXES); + } + /* Found a plex. Give it an sd structure */ + plex = &PLEX[plexno]; /* this one is ours */ + bzero(plex, sizeof(struct plex)); /* polish it up */ + plex->sdnos = (int *) Malloc(sizeof(int) * INITIAL_SUBDISKS_IN_PLEX); /* allocate sd table */ + CHECKALLOC(plex->sdnos, "vinum: Can't allocate plex subdisk table"); + bzero(plex->sdnos, (sizeof(int) * INITIAL_SUBDISKS_IN_PLEX)); /* do we need this? */ + plex->subdisks = 0; /* no subdisks in use */ + plex->subdisks_allocated = INITIAL_SUBDISKS_IN_PLEX; /* and we have space for this many */ + plex->organization = plex_disorg; /* and it's not organized */ + plex->volno = -1; /* no volume yet */ + return plexno; /* return the index */ +} + +/* Find the named plex in vinum_conf.plex + + * If create != 0, create an entry if it doesn't exist + * return index in vinum_conf.plex + */ +int +find_plex(const char *name, int create) +{ + BROKEN_GDB; + int plexno; + struct plex *plex; + + for (plexno = 0; plexno < vinum_conf.plexes_allocated; plexno++) { + if (strcmp(PLEX[plexno].name, name) == 0) /* found it */ + return plexno; + } + + /* the plex isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + /* Allocate one and insert the name */ + plexno = get_empty_plex(); + plex = &PLEX[plexno]; /* point to it */ + bcopy(name, plex->name, min(sizeof(plex->name), strlen(name))); /* put in its name */ + return plexno; /* return the pointer */ +} + +/* Free an allocated plex entry + * and its associated memory areas */ +void +free_plex(int plexno) +{ + BROKEN_GDB; + struct plex *plex; + + plex = &PLEX[plexno]; + if (plex->sdnos) + Free(plex->sdnos); + if (plex->lock) + Free(plex->lock); + if (plex->defective_region) + Free(plex->defective_region); + if (plex->unmapped_region) + Free(plex->unmapped_region); + bzero(plex, sizeof(struct plex)); /* and clear it out */ + plex->state = plex_unallocated; +} + +/* Find an empty volume in the volume table */ +int +get_empty_volume(void) +{ + BROKEN_GDB; + int volno; + struct volume *vol; + + /* first see if we have one which has been deallocated */ + for (volno = 0; volno < vinum_conf.volumes_used; volno++) { + if (VOL[volno].state == volume_unallocated) /* bingo */ + break; + } + + if (volno >= vinum_conf.volumes_used) + /* Couldn't find a deallocated volume. Allocate a new one */ + { + vinum_conf.volumes_used++; + if (vinum_conf.volumes_used > vinum_conf.volumes_allocated) + EXPAND(VOL, struct volume, vinum_conf.volumes_allocated, INITIAL_VOLUMES); + } + /* Now initialize fields */ + vol = &VOL[volno]; + bzero(vol, sizeof(struct volume)); + vol->preferred_plex = -1; /* default to round robin */ + vol->preferred_plex = ROUND_ROBIN_READPOL; /* round robin */ + + return volno; /* return the index */ +} + +/* Find the named volume in vinum_conf.volume. + + * If create != 0, create an entry if it doesn't exist + * return the index in vinum_conf + */ +int +find_volume(const char *name, int create) +{ + BROKEN_GDB; + int volno; + struct volume *vol; + + for (volno = 0; volno < vinum_conf.volumes_used; volno++) { + if (strcmp(VOL[volno].name, name) == 0) /* found it */ + return volno; + } + + /* the volume isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + /* Allocate one and insert the name */ + volno = get_empty_volume(); + vol = &VOL[volno]; + bcopy(name, vol->name, min(sizeof(vol->name), strlen(name))); /* put in its name */ + vol->blocksize = DEV_BSIZE; /* block size of this volume */ + return volno; /* return the pointer */ +} + +/* Free an allocated volume entry + * and its associated memory areas */ +void +free_volume(int volno) +{ + BROKEN_GDB; + struct volume *vol; + + vol = &VOL[volno]; + bzero(vol, sizeof(struct volume)); /* and clear it out */ + vol->state = volume_unallocated; +} + +/* Handle a drive definition. We store the information in the global variable + * drive, so we don't need to allocate. + * + * If we find an error, print a message and return + */ +void +config_drive(void) +{ + BROKEN_GDB; + enum drive_label_info partition_status; /* info about the partition */ + int parameter; + int driveno; /* index of drive in vinum_conf */ + struct drive *drive; /* and pointer to it */ + + if (tokens < 2) /* not enough tokens */ + throw_rude_remark(EINVAL, "Drive has no name"); + driveno = find_drive(token[1], 1); /* allocate a drive to initialize */ + drive = &DRIVE[driveno]; /* and get a pointer */ + + if (drive->state != drive_uninit) { /* we already know this drive */ + /* XXX Check which definition is more up-to-date. Give + * preference for the definition on its own drive */ + return; /* XXX */ + } + for (parameter = 2; parameter < tokens; parameter++) { /* look at the other tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_device: + parameter++; + if (drive->devicename[0] != '\0') { /* we know this drive... */ + if (strcmp(drive->devicename, token[parameter])) /* different name */ + close_drive(drive); /* close it if it's open */ + else /* no change */ + break; + } + bcopy(token[parameter], /* insert device information */ + drive->devicename, + min(sizeof(drive->devicename), + strlen(token[parameter]))); + /* open the device and get the configuration */ + partition_status = read_drive_label(drive); + if (partition_status == DL_CANT_OPEN) { /* not our kind */ + close_drive(drive); + if (drive->lasterror == EFTYPE) /* wrong kind of partition */ + throw_rude_remark(drive->lasterror, + "Drive %s has invalid partition type", + drive->label.name); + else /* I/O error of some kind */ + throw_rude_remark(drive->lasterror, + "Can't initialize drive %s", + drive->label.name); + } else if (partition_status == DL_WRONG_DRIVE) { /* valid drive, not ours */ + close_drive(drive); + throw_rude_remark(drive->lasterror, + "Incorrect drive name %s specified for drive %s", + token[1], + drive->label.name); + } + break; + + case kw_state: + checkkernel(token[++parameter]); /* must be a kernel user */ + drive->state = DriveState(token[parameter]); /* set the state */ + break; + + default: + close_drive(drive); + throw_rude_remark(EINVAL, + "Drive %s, invalid keyword: %s", + token[1], + token[parameter]); + } + } + + if (drive->devicename[0] == '\0') + throw_rude_remark(EINVAL, "No device name for %s", drive->label.name); + +} + +/* Handle a subdisk definition. We store the information in the global variable + * sd, so we don't need to allocate. + * + * If we find an error, print a message and return + */ +void +config_subdisk(void) +{ + BROKEN_GDB; + int parameter; + int sdno; /* index of sd in vinum_conf */ + struct sd *sd; /* and pointer to it */ + u_int64_t size; + int sectors; /* sector offset value */ + int detached = 0; /* set to 1 if this is a detached subdisk */ + int sdindex = -1; /* index in plexes subdisk table */ + + sdno = get_empty_sd(); /* allocate an SD to initialize */ + sd = &SD[sdno]; /* and get a pointer */ + for (parameter = 1; parameter < tokens; parameter++) { /* look at the other tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_detached: + detached = 1; + break; + + case kw_plexoffset: + size = sizespec(token[++parameter]); + if ((size % DEV_BSIZE) != 0) + throw_rude_remark(EINVAL, "sd %s, bad plex offset alignment: %qd", sd->name, size); + else + sd->plexoffset = size / DEV_BSIZE; + break; + + case kw_driveoffset: + size = sizespec(token[++parameter]); + if ((size % DEV_BSIZE) != 0) + throw_rude_remark(EINVAL, "sd %s, bad drive offset alignment: %qd", sd->name, size); + else + sd->driveoffset = size / DEV_BSIZE; + break; + + case kw_name: + ++parameter; + bcopy(token[parameter], + sd->name, + min(sizeof(sd->name), strlen(token[parameter]))); + break; + + case kw_len: + size = sizespec(token[++parameter]); + if ((size % DEV_BSIZE) != 0) + throw_rude_remark(EINVAL, "sd %s, length %d not multiple of sector size", sd->name, size); + else + sd->sectors = size / DEV_BSIZE; + break; + + case kw_drive: + sd->driveno = find_drive(token[++parameter], 1); /* insert drive information */ + break; + + case kw_plex: + sd->plexno = find_plex(token[++parameter], 1); /* insert plex information */ + break; + + case kw_state: + checkkernel(token[++parameter]); /* must be a kernel user */ + sd->state = SdState(token[parameter]); /* set the state */ + break; + + default: + throw_rude_remark(EINVAL, "sd %s, invalid keyword: %s", sd->name, token[parameter]); + } + } + + /* Check we have a drive name */ + if (sd->driveno < 0) { /* didn't specify a drive */ + sd->driveno = current_drive; /* set to the current drive */ + if (sd->driveno < 0) /* no current drive? */ + throw_rude_remark(EINVAL, "Subdisk %s is not associated with a drive", sd->name); + } + /* Check for a plex name */ + if ((sd->plexno < 0) /* didn't specify a plex */ + &&(!detached)) /* and didn't say not to, */ + sd->plexno = current_plex; /* set to the current plex */ + + if (sd->plexno >= 0) + sdindex = give_sd_to_plex(sd->plexno, sdno); /* now tell the plex that it has this sd */ + + sd->sdno = sdno; /* point to our entry in the table */ + + /* Does the subdisk have a name? If not, give it one */ + if (sd->name[0] == '\0') { /* no name */ + char sdsuffix[8]; /* form sd name suffix here */ + + /* Do we have a plex name? */ + if (sdindex >= 0) /* we have a plex */ + strcpy(sd->name, PLEX[sd->plexno].name); /* take it from there */ + else /* no way */ + throw_rude_remark(EINVAL, "Unnamed sd is not associated with a plex"); + sprintf(sdsuffix, ".s%d", sdindex); /* form the suffix */ + strcat(sd->name, sdsuffix); /* and add it to the name */ + } + /* do we have complete info for this subdisk? */ + if (sd->sectors == 0) + throw_rude_remark(EINVAL, "sd %s has no length spec", sd->name); + + if (sd->state == sd_unallocated) /* no state decided, */ + sd->state = sd_init; /* at least we're in the game */ + + /* register the subdisk with the drive. This action + * will have the side effect of setting the offset if + * we haven't specified one, and causing an error + * message if it overlaps with another subdisk. */ + give_sd_to_drive(sdno); +} + +/* Handle a plex definition. + * If we find an error, print a message, deallocate the nascent plex, and return + */ +void +config_plex(void) +{ + BROKEN_GDB; + int parameter; + int plexno; /* index of plex in vinum_conf */ + struct plex *plex; /* and pointer to it */ + int pindex = MAXPLEX; /* index in volume's plex list */ + int detached = 0; /* don't give it to a volume */ + + current_plex = -1; /* forget the previous plex */ + plexno = get_empty_plex(); /* allocate a plex */ + plex = &PLEX[plexno]; /* and point to it */ + plex->plexno = plexno; /* and back to the config */ + for (parameter = 1; parameter < tokens; parameter++) { /* look at the other tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_detached: + detached = 1; + break; + + case kw_name: + { + int namedplexno; + + namedplexno = find_plex(token[++parameter], 0); /* find an existing plex with this name */ + if (namedplexno >= 0) + throw_rude_remark(EINVAL, "Duplicate plex %s", token[parameter]); + } + bcopy(token[parameter], /* put in the name */ + plex->name, + min(MAXPLEXNAME, strlen(token[parameter]))); + break; + + case kw_org: /* plex organization */ + switch (get_keyword(token[++parameter], &keyword_set)) { + case kw_concat: + plex->organization = plex_concat; + break; + + case kw_striped: + { + int stripesize = sizespec(token[++parameter]); + + plex->organization = plex_striped; + if (stripesize % DEV_BSIZE != 0) /* not a multiple of block size, */ + throw_rude_remark(EINVAL, "plex %s: stripe size %d not a multiple of sector size", + plex->name, + stripesize); + else + plex->stripesize = stripesize / DEV_BSIZE; + break; + } + + + default: + throw_rude_remark(EINVAL, "Invalid plex organization"); + } + if (((plex->organization == plex_striped) + ) + && (plex->stripesize == 0)) /* didn't specify a valid stripe size */ + throw_rude_remark(EINVAL, "Need a stripe size parameter"); + break; + + case kw_volume: + plex->volno = find_volume(token[++parameter], 1); /* insert a pointer to the volume */ + break; + + case kw_sd: /* add a subdisk */ + { + int sdno; + + sdno = find_subdisk(token[++parameter], 1); /* find a subdisk */ + SD[sdno].plexoffset = sizespec(token[++parameter]); /* get the offset */ + give_sd_to_plex(plexno, sdno); /* and insert it there */ + break; + } + + case kw_state: + checkkernel(token[++parameter]); /* only for kernel use */ + plex->state = PlexState(token[parameter]); /* set the state */ + break; + + default: + throw_rude_remark(EINVAL, "plex %s, invalid keyword: %s", + plex->name, + token[parameter]); + } + } + + if ((plex->volno < 0) /* we don't have a volume */ + &&(!detached)) /* and we wouldn't object */ + plex->volno = current_volume; + + if (plex->volno >= 0) + pindex = give_plex_to_volume(plex->volno, plexno); /* Now tell the volume that it has this plex */ + + /* Does the plex have a name? If not, give it one */ + if (plex->name[0] == '\0') { /* no name */ + char plexsuffix[8]; /* form plex name suffix here */ + /* Do we have a volume name? */ + if (plex->volno >= 0) /* we have a volume */ + strcpy(plex->name, /* take it from there */ + VOL[plex->volno].name); + else /* no way */ + throw_rude_remark(EINVAL, "Unnamed plex is not associated with a volume"); + sprintf(plexsuffix, ".p%d", pindex); /* form the suffix */ + strcat(plex->name, plexsuffix); /* and add it to the name */ + } + /* Note the last plex we configured */ + current_plex = plexno; + if (plex->state == plex_unallocated) /* we haven't changed the state, */ + plex->state = plex_init; /* we're initialized now */ +} + +/* Handle a volume definition. + * If we find an error, print a message, deallocate the nascent volume, and return + */ +void +config_volume(void) +{ + BROKEN_GDB; + int parameter; + int volno; + struct volume *vol; /* collect volume info here */ + int i; + + if (tokens < 2) /* not enough tokens */ + throw_rude_remark(EINVAL, "Volume has no name"); + current_volume = -1; /* forget the previous volume */ + volno = find_volume(token[1], 1); /* allocate a volume to initialize */ + vol = &VOL[volno]; /* and get a pointer */ + + for (parameter = 2; parameter < tokens; parameter++) { /* look at all tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_plex: + { + int plexno; /* index of this plex */ + + plexno = find_plex(token[++parameter], 1); /* find a plex */ + if (plexno < 0) /* couldn't */ + break; /* we've already had an error message */ + plexno = my_plex(volno, plexno); /* does it already belong to us? */ + if (plexno > 0) /* yes, shouldn't get it again */ + throw_rude_remark(EINVAL, + "Plex %s already belongs to volume %s", + token[parameter], + vol->name); + else if (++vol->plexes > 8) /* another entry */ + throw_rude_remark(EINVAL, + "Too many plexes for volume %s", + vol->name); + vol->plex[vol->plexes - 1] = plexno; + } + break; + + case kw_readpol: + switch (get_keyword(token[++parameter], &keyword_set)) { /* decide what to do */ + case kw_round: + vol->preferred_plex = ROUND_ROBIN_READPOL; /* default */ + break; + + case kw_prefer: + { + int myplexno; /* index of this plex */ + + myplexno = find_plex(token[++parameter], 1); /* find a plex */ + if (myplexno < 0) /* couldn't */ + break; /* we've already had an error message */ + myplexno = my_plex(volno, myplexno); /* does it already belong to us? */ + if (myplexno > 0) /* yes */ + vol->preferred_plex = myplexno; /* just note the index */ + else if (++vol->plexes > 8) /* another entry */ + throw_rude_remark(EINVAL, "Too many plexes"); + else { /* space for the new plex */ + vol->plex[vol->plexes - 1] = myplexno; /* add it to our list */ + vol->preferred_plex = vol->plexes - 1; /* and note the index */ + } + } + break; + + default: + throw_rude_remark(EINVAL, "Invalid read policy"); + } + + case kw_setupstate: + vol->flags |= VF_CONFIG_SETUPSTATE; /* set the volume up later on */ + break; + + case kw_state: + checkkernel(token[++parameter]); /* must be a kernel user */ + vol->state = VolState(token[parameter]); /* set the state */ + break; + + /* XXX experimental ideas. These are not + * documented, and will not be until I + * decide they're worth keeping */ + case kw_writethrough: /* set writethrough mode */ + vol->flags |= VF_WRITETHROUGH; + break; + + case kw_writeback: /* set writeback mode */ + vol->flags &= ~VF_WRITETHROUGH; + break; + + case kw_raw: + vol->flags |= VF_RAW; /* raw volume (no label) */ + break; + + default: + throw_rude_remark(EINVAL, "volume %s, invalid keyword: %s", + vol->name, + token[parameter]); + } + } + + current_volume = volno; /* note last referred volume */ + vol->devno = VINUMBDEV(volno, 0, 0, VINUM_VOLUME_TYPE); /* also note device number */ + + /* Before we can actually use the volume, we need + * a volume label. We could start to fake one here, + * but it will be a lot easier when we have some + * to copy from the drives, so defer it until we + * set up the configuration. XXX */ + if (vol->state == volume_unallocated) + vol->state = volume_down; /* now ready to bring up at the end */ + + /* Find out how big our volume is */ + for (i = 0; i < vol->plexes; i++) + vol->size = max(vol->size, PLEX[vol->plex[i]].length); +} + +/* Parse a config entry. CARE! This destroys the original contents of the + * config entry, which we don't really need after this. More specifically, it + * places \0 characters at the end of each token. + * + * Return 0 if all is well, otherwise EINVAL */ +int +parse_config(char *cptr, struct keywordset *keyset) +{ + BROKEN_GDB; + int status; + + status = 0; /* until proven otherwise */ + tokens = tokenize(cptr, token); /* chop up into tokens */ + + if (tokens <= 0) /* screwed up or empty line */ + return tokens; /* give up */ + + if (token[0][0] == '#') /* comment line */ + return 0; + + switch (get_keyword(token[0], keyset)) { /* decide what to do */ + case kw_read: /* read config from a specified drive */ + vinum_conf.flags |= VF_KERNELOP | VF_READING_CONFIG; /* kernel operation: reading config */ + status = check_drive(token[1]); /* check the drive info */ + vinum_conf.flags &= ~(VF_KERNELOP | VF_READING_CONFIG); + if (status != 0) { + char *msg = "Can't read configuration from %s"; + if (status == ENODEV) + msg = "No vinum configuration on %s"; + throw_rude_remark(status, msg, token[1]); + } + updateconfig(VF_KERNELOP); /* update from kernel space */ + break; + + case kw_drive: + config_drive(); + break; + + case kw_subdisk: + config_subdisk(); + break; + + case kw_plex: + config_plex(); + break; + + case kw_volume: + config_volume(); + break; + + /* Anything else is invalid in this context */ + default: + throw_rude_remark(EINVAL, /* should we die? */ + "Invalid configuration information: %s", + token[0]); + } + return status; +} + +/* parse a line handed in from userland via ioctl. + * This differs only by the error reporting mechanism: + * we return the error indication in the reply to the + * ioctl, so we need to set a global static pointer in + * this file. This technique works because we have + * ensured that configuration is performed in a single- + * threaded manner */ +int +parse_user_config(char *cptr, struct keywordset *keyset) +{ + BROKEN_GDB; + int status; + + ioctl_reply = (struct _ioctl_reply *) cptr; + status = parse_config(cptr, keyset); + ioctl_reply = NULL; /* don't do this again */ + return status; +} + +/* Remove an object */ +void +remove(struct vinum_ioctl_msg *msg) +{ + struct vinum_ioctl_msg message = *msg; /* make a copy to hand on */ + + ioctl_reply = (struct _ioctl_reply *) msg; /* reinstate the address to reply to */ + ioctl_reply->error = 0; /* no error, */ + ioctl_reply->msg[0] = '\0'; /* no message */ + + switch (message.type) { + case drive_object: + remove_drive_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + case sd_object: + remove_sd_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + case plex_object: + remove_plex_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + case volume_object: + remove_volume_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + default: + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "Invalid object type"); + } +} + +/* Remove a drive. */ +void +remove_drive_entry(int driveno, int force, int recurse) +{ + struct drive *drive = &DRIVE[driveno]; + + if ((driveno > vinum_conf.drives_used) /* not a valid drive */ + ||(drive->state == drive_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such drive"); + } else if (drive->opencount > 0) { /* we have subdisks */ + if (force) { /* do it at any cost */ + int sdno; + struct vinum_ioctl_msg sdmsg; + + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) { + if ((SD[sdno].state != sd_unallocated) /* subdisk is allocated */ + &&(SD[sdno].driveno == driveno)) { /* and it belongs to this drive */ + sdmsg.type = sd_object; + sdmsg.recurse = 1; + sdmsg.force = force; + remove(&sdmsg); /* remove the subdisk by force */ + } + } + remove_drive(driveno); /* now remove it */ + } else + ioctl_reply->error = EBUSY; /* can't do that */ + } else + remove_drive(driveno); /* just remove it */ +} + +/* remove a subdisk */ +void +remove_sd_entry(int sdno, int force, int recurse) +{ + struct sd *sd = &SD[sdno]; + + if ((sdno > vinum_conf.subdisks_used) /* not a valid sd */ + ||(sd->state == sd_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such subdisk"); + } else if (sd->plexno >= 0) { /* we have a plex */ + if (force) { /* do it at any cost */ + struct plex *plex = &PLEX[sd->plexno]; /* point to our plex */ + int mysdno; + + for (mysdno = 0; /* look for ourselves */ + mysdno < plex->subdisks && &SD[plex->sdnos[mysdno]] != sd; + mysdno++); + if (mysdno == plex->subdisks) /* didn't find it */ + throw_rude_remark(ENOENT, "plex %s does not contain subdisk %s", plex->name, sd->name); + if (mysdno < (plex->subdisks - 1)) /* not the last subdisk */ + bcopy(&plex->sdnos[mysdno + 1], + &plex->sdnos[mysdno], + (plex->subdisks - 1 - mysdno) * sizeof(int)); + plex->subdisks--; + /* removing a subdisk from a striped or + * RAID-5 plex really tears the hell out + * of the structure, and it needs to be + * reinitialized */ + if (plex->organization != plex_concat) /* not concatenated, */ + set_plex_state(plex->plexno, plex_faulty, setstate_force); /* need to reinitialize */ + rebuild_plex_unmappedlist(plex); /* and see what remains */ + free_sd(sdno); + } else + ioctl_reply->error = EBUSY; /* can't do that */ + } else + free_sd(sdno); +} + +/* remove a plex */ +void +remove_plex_entry(int plexno, int force, int recurse) +{ + struct plex *plex = &PLEX[plexno]; + int sdno; + + if ((plexno > vinum_conf.plexes_used) /* not a valid plex */ + ||(plex->state == plex_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such plex"); + } else if (plex->pid) { /* we're open */ + ioctl_reply->error = EBUSY; /* no getting around that */ + return; + } + if (plex->subdisks) { + if (force) { /* do it anyway */ + if (recurse) { /* remove all below */ + for (sdno = 0; sdno < plex->subdisks; sdno++) + free_sd(plex->sdnos[sdno]); /* free all subdisks */ + } else { /* just tear them out */ + for (sdno = 0; sdno < plex->subdisks; sdno++) + SD[plex->sdnos[sdno]].plexno = -1; /* no plex any more */ + } + } else { /* can't do it without force */ + ioctl_reply->error = EBUSY; /* can't do that */ + return; + } + } + if (plex->volno >= 0) { /* we are part of a volume */ + /* XXX This should be more intelligent. We should + * be able to remove a plex as long as the volume + * does not lose any data, which is normally the + * case when it has more than one plex. To do it + * right we must compare the completeness of the + * mapping of all the plexes in the volume */ + if (force) { /* do it at any cost */ + struct volume *vol = &VOL[plex->volno]; + int myplexno; + + for (myplexno = 0; myplexno < vol->plexes; myplexno++) + if (vol->plex[myplexno] == plexno) /* found it */ + break; + if (myplexno == vol->plexes) /* didn't find it. Huh? */ + throw_rude_remark(ENOENT, "volume %s does not contain plex %s", vol->name, plex->name); + if (myplexno < (vol->plexes - 1)) /* not the last plex in the list */ + bcopy(&vol->plex[myplexno + 1], &vol->plex[myplexno], vol->plexes - 1 - myplexno); + vol->plexes--; + } else { + ioctl_reply->error = EBUSY; /* can't do that */ + return; + } + } + free_plex(plexno); +} + +/* remove a volume */ +void +remove_volume_entry(int volno, int force, int recurse) +{ + struct volume *vol = &VOL[volno]; + int plexno; + + if ((volno > vinum_conf.volumes_used) /* not a valid volume */ + ||(vol->state == volume_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such volume"); + } else if (vol->opencount) /* we're open */ + ioctl_reply->error = EBUSY; /* no getting around that */ + else if (vol->plexes) { + if (recurse && force) { /* remove all below */ + struct vinum_ioctl_msg plexmsg; + + plexmsg.type = plex_object; + plexmsg.recurse = 1; + plexmsg.force = force; + for (plexno = 0; plexno < vol->plexes; plexno++) { + plexmsg.index = vol->plex[plexno]; /* plex number */ + remove(&plexmsg); + } + free_volume(volno); + } else + ioctl_reply->error = EBUSY; /* can't do that */ + } else + free_volume(volno); +} + +void +update_sd_config(int sdno, int kernelstate) +{ + if (!kernelstate) + set_sd_state(sdno, sd_up, setstate_configuring | setstate_norecurse); +} + +void +update_plex_config(int plexno, int kernelstate) +{ + int error = 0; + int size; + int sdno; + struct plex *plex = &PLEX[plexno]; + enum plexstate state = plex_up; /* state we want the plex in */ + + /* XXX Insert checks here for sparse plexes and volumes */ + + /* Check that our subdisks make sense. For + * striped and RAID5 plexes, we need at least + * two subdisks, and they must all be the same + * size */ + if (((plex->organization == plex_striped) + ) + && (plex->subdisks < 2)) { + error = 1; + printf("vinum: plex %s does not have at least 2 subdisks\n", plex->name); + if (!kernelstate) + set_plex_state(plexno, plex_down, setstate_force | setstate_configuring | setstate_norecurse); + } + size = 0; + for (sdno = 0; sdno < plex->subdisks; sdno++) { + if (((plex->organization == plex_striped) + ) + && (sdno > 0) + && (SD[plex->sdnos[sdno]].sectors != SD[plex->sdnos[sdno - 1]].sectors)) { + error = 1; + printf("vinum: plex %s must have equal sized subdisks\n", plex->name); + set_plex_state(plexno, plex_down, setstate_force | setstate_configuring | setstate_norecurse); + } + size += SD[plex->sdnos[sdno]].sectors; + } + + if (plex->subdisks) { /* plex has subdisks, calculate size */ + rebuild_plex_unmappedlist(plex); /* rebuild the unmapped list first */ + + plex->length = size; + } else { /* no subdisks, */ + plex->length = 0; /* no size */ + state = plex_down; /* take it down */ + } + if (!(kernelstate || error)) + set_plex_state(plexno, state, setstate_none | setstate_configuring | setstate_norecurse); +} + +void +update_volume_config(int volno, int kernelstate) +{ + struct volume *vol = &VOL[volno]; + struct plex *plex; + int plexno; + + if (vol->state != volume_unallocated) + /* Recalculate the size of the volume */ + { + vol->size = 0; + for (plexno = 0; plexno < vol->plexes; plexno++) { + plex = &PLEX[vol->plex[plexno]]; + vol->size = max(plex->length, vol->size); /* maximum size */ + plex->volplexno = plexno; /* note it in the plex */ + } + } + if (!kernelstate) /* try to bring it up */ + set_volume_state(volno, volume_up, setstate_configuring | setstate_norecurse); +} + +/* Update the global configuration. + * kernelstate is != 0 if we're reading in a config + * from disk. In this case, we don't try to + * bring the devices up, though we will bring + * them down if there's some error which got + * missed when writing to disk. + */ +void +updateconfig(int kernelstate) +{ + BROKEN_GDB; + int sdno; + int plexno; + int volno; + struct volume *vol; + struct plex *plex; + + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) + update_sd_config(sdno, kernelstate); + + for (plexno = 0; plexno < vinum_conf.plexes_used; plexno++) + update_plex_config(plexno, kernelstate); + + for (volno = 0; volno < vinum_conf.volumes_used; volno++) + update_volume_config(volno, kernelstate); + save_config(); +} + +/* Start manual changes to the configuration and lock out + * others who may wish to do so. + * XXX why do we need this and lock_config too? */ +int +start_config(void) +{ + int error; + + while ((vinum_conf.flags & VF_CONFIGURING) != 0) { + vinum_conf.flags |= VF_WILL_CONFIGURE; + if ((error = tsleep(&vinum_conf, PRIBIO | PCATCH, "vincfg", 0)) != 0) + return error; + } + /* We need two flags here: VF_CONFIGURING + * tells other processes to hold off (this + * function), and VF_CONFIG_INCOMPLETE + * tells the state change routines not to + * propagate incrememntal state changes */ + vinum_conf.flags |= VF_CONFIGURING | VF_CONFIG_INCOMPLETE; + current_drive = -1; /* reset the defaults */ + current_plex = -1; /* and the same for the last plex */ + current_volume = -1; /* and the last volme */ + return 0; +} + +/* Update the config if update is 1, and unlock + * it. We won't update the configuration if we + * are called in a recursive loop via throw_rude_remark. + */ +void +finish_config(int update) +{ + vinum_conf.flags &= ~VF_CONFIG_INCOMPLETE; /* we've finished our config */ + if (update) + updateconfig(0); /* so update things */ + else + updateconfig(1); /* do some updates only */ + vinum_conf.flags &= ~VF_CONFIGURING; /* and now other people can take a turn */ + if ((vinum_conf.flags & VF_WILL_CONFIGURE) != 0) { + vinum_conf.flags &= ~VF_WILL_CONFIGURE; + wakeup(&vinum_conf); + } +} diff --git a/sys/dev/vinum/vinumext.h b/sys/dev/vinum/vinumext.h new file mode 100644 index 0000000..4b7d5a0 --- /dev/null +++ b/sys/dev/vinum/vinumext.h @@ -0,0 +1,214 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumext.h,v 1.14 1998/08/11 00:03:57 grog Exp grog $ + */ + +/* vinumext.h: external definitions */ + +extern struct _vinum_conf vinum_conf; /* configuration information */ + +#ifdef DEBUG +extern debug; /* debug flags */ +#endif + +#define CHECKALLOC(ptr, msg) \ + if (ptr == NULL) \ + { \ + printf (msg); \ + longjmp (command_fail, -1); \ + } +#ifndef KERNEL +struct vnode; +struct proc; +#endif + +#ifdef KERNEL +int give_sd_to_plex(int plexno, int sdno); +int give_plex_to_volume(int volno, int plexno); +int check_drive(char *); +enum drive_label_info read_drive_label(struct drive *drive); +int parse_config(char *, struct keywordset *); +int parse_user_config(char *cptr, struct keywordset *keyset); +u_int64_t sizespec(char *spec); +int volume_index(struct volume *volume); +int plex_index(struct plex *plex); +int sd_index(struct sd *sd); +int drive_index(struct drive *drive); +int my_plex(int volno, int plexno); +int my_sd(int plexno, int sdno); +int get_empty_drive(void); +int find_drive(const char *name, int create); +int find_drive_by_dev(const char *devname, int create); +int get_empty_sd(void); +int find_subdisk(const char *name, int create); +void free_sd(int sdno); +void free_volume(int volno); +int get_empty_plex(void); +int find_plex(const char *name, int create); +void free_plex(int plexno); +int get_empty_volume(void); +int find_volume(const char *name, int create); +void config_subdisk(void); +void config_plex(void); +void config_volume(void); +void config_drive(void); +void updateconfig(int); +void update_sd_config(int sdno, int kernelstate); +void update_plex_config(int plexno, int kernelstate); +void update_volume_config(int volno, int kernelstate); +void update_config(void); +void drive_io_done(struct buf *); +int save_config(void); +void write_config(char *, int); +int start_config(void); +void finish_config(int); +void remove(struct vinum_ioctl_msg *msg); +void remove_drive_entry(int driveno, int force, int recurse); +void remove_sd_entry(int sdno, int force, int recurse); +void remove_plex_entry(int plexno, int force, int recurse); +void remove_volume_entry(int volno, int force, int recurse); + +void checkernel(char *); +int open_drive(struct drive *, struct proc *); +void close_drive(struct drive *drive); +int driveio(struct drive *, void *, size_t, off_t, int); +/* #define read_drive(a, b, c, d) driveio (a, b, c, d, B_READ) + #define write_drive(a, b, c, d) driveio (a, b, c, d, B_WRITE) */ +int set_drive_parms(struct drive *drive); +int init_drive(struct drive *); +/* void throw_rude_remark (int, struct _ioctl_reply *, char *, ...); XXX */ +void throw_rude_remark(int, char *,...); + +int read_drive(struct drive *drive, void *buf, size_t length, off_t offset); +int write_drive(struct drive *drive, void *buf, size_t length, off_t offset); +void format_config(char *config, int len); +void checkkernel(char *op); +void free_drive(struct drive *drive); +void down_drive(struct drive *drive); +void remove_drive(int driveno); + +/* I/O */ +d_open_t vinumopen; +d_close_t vinumclose; +d_strategy_t vinumstrategy; +d_ioctl_t vinumioctl; +d_dump_t vinumdump; +d_psize_t vinumsize; +d_read_t vinumread; +d_write_t vinumwrite; + +int vinumstart(struct buf *bp, int reviveok); +int launch_requests(struct request *rq, int reviveok); + +/* XXX Do we need this? */ +int vinumpart(dev_t); + +/* Memory allocation */ +void vinum_meminfo(caddr_t data); +int vinum_mallocinfo(caddr_t data); + +void expand_table(void **, int, int); + +void add_defective_region(struct plex *plex, off_t offset, size_t length); +void add_unmapped_region(struct plex *plex, off_t offset, size_t length); +void rebuild_plex_unmappedlist(struct plex *plex); +struct request; +struct rqgroup *allocrqg(struct request *rq, int elements); +void deallocrqg(struct rqgroup *rqg); + +/* State transitions */ +int set_drive_state(int driveno, enum drivestate state, int force); +int set_sd_state(int sdno, enum sdstate state, enum setstateflags flags); +enum requeststatus checksdstate(struct sd *sd, struct request *rq, daddr_t diskaddr, daddr_t diskend); +int set_plex_state(int plexno, enum plexstate state, enum setstateflags flags); +int set_volume_state(int volumeno, enum volumestate state, enum setstateflags flags); +void get_volume_label(struct volume *vol, struct disklabel *lp); +int write_volume_label(int); +void start_object(struct vinum_ioctl_msg *); +void stop_object(struct vinum_ioctl_msg *); +void setstate(struct vinum_ioctl_msg *msg); +void vinum_label(int); +int vinum_writedisklabel(struct volume *, struct disklabel *); +int initsd(int); + +int restart_plex(int plexno); +int revive_block(int plexno); + +/* Auxiliary functions */ +enum sdstates sdstatemap(struct plex *plex, int *sddowncount); +enum volplexstate vpstate(struct plex *plex); +#endif + +enum keyword get_keyword(char *, struct keywordset *); +void listconfig(void); +char *drive_state(enum drivestate); +char *volume_state(enum volumestate); +char *plex_state(enum plexstate); +char *plex_org(enum plexorg); +char *sd_state(enum sdstate); +enum drivestate DriveState(char *text); +enum sdstate SdState(char *text); +enum plexstate PlexState(char *text); +enum volumestate VolState(char *text); +struct drive *validdrive(int driveno, struct _ioctl_reply *); +struct sd *validsd(int sdno, struct _ioctl_reply *); +struct plex *validplex(int plexno, struct _ioctl_reply *); +struct volume *validvol(int volno, struct _ioctl_reply *); +int tokenize(char *, char *[]); +void resetstats(struct vinum_ioctl_msg *msg); + +/* Locking */ +int lockvol(struct volume *vol); +void unlockvol(struct volume *vol); +int lockplex(struct plex *plex); +void unlockplex(struct plex *plex); +int lockrange(struct plex *plex, off_t first, off_t last); +void unlockrange(struct plex *plex, off_t first, off_t last); +int lock_config(void); +void unlock_config(void); + +#ifdef DEBUG +#define expandrq(prq) \ +{ \ + expand_table ((void **) &prq->rqe, \ + prq->requests * sizeof (struct rqelement), \ + (prq->requests + RQELTS) * sizeof (struct rqelement) ); \ + bzero (&prq->rqe [prq->requests], RQELTS * sizeof (struct rqelement)); \ + prq->rqcount += RQELTS; \ + } +#else +void expandrq(struct plexrq *); +#endif diff --git a/sys/dev/vinum/vinumhdr.h b/sys/dev/vinum/vinumhdr.h new file mode 100644 index 0000000..22a8561 --- /dev/null +++ b/sys/dev/vinum/vinumhdr.h @@ -0,0 +1,104 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + */ + +/* Header files used by all modules */ +/* $Id: vinumhdr.h,v 1.7 1998/08/07 04:41:18 grog Exp grog $ */ + +#ifdef KERNEL +#define REALLYKERNEL +#endif +#include <sys/param.h> +#ifdef REALLYKERNEL +#include <sys/systm.h> +#include <sys/kernel.h> +#endif +#ifdef DEVFS +#error "DEVFS code not complete yet" +#include <sys/devfsext.h> +#endif /*DEVFS */ +#include <sys/proc.h> +#include <sys/errno.h> +#include <sys/dkstat.h> +#include <sys/buf.h> +#include <sys/malloc.h> +#include <sys/uio.h> +#include <sys/namei.h> +#include <sys/conf.h> +#include <sys/stat.h> +#include <sys/disklabel.h> +#include <ufs/ffs/fs.h> +#include <sys/mount.h> +#include <sys/device.h> +#undef KERNEL /* XXX */ +#include <sys/disk.h> +#ifdef REALLYKERNEL +#define KERNEL +#endif +#include <sys/syslog.h> +#include <sys/fcntl.h> +#include <sys/vnode.h> +#include <sys/dkbad.h> +#include <setjmp.h> +#include <stdarg.h> +#include <vm/vm.h> +#ifdef USES_VM +/* XXX Do we need this? */ +#include <vm/vm_extern.h> +#include <vm/vm_kern.h> +#include <vm/vm_object.h> +#include <vm/vm_prot.h> +/* #include <vm/vm_page.h> */ +#include <sys/vmmeter.h> +/* #include <machine/pmap.h> */ +#include <machine/cputypes.h> +#endif /* USES_VM */ +#include <vinumvar.h> +#include <vinumio.h> +#include "vinumkw.h" +#include "vinumext.h" + +#undef Free /* defined in some funny net stuff */ +#ifdef REALLYKERNEL +#define Malloc(x) MMalloc ((x), __FILE__, __LINE__) /* show where we came from */ +#define Free(x) FFree ((x), __FILE__, __LINE__) /* show where we came from */ +caddr_t MMalloc (int size, char *, int); +void FFree (void *mem, char *, int); +#else +#define Malloc(x) malloc ((x)) /* just the size */ +#define Free(x) free ((x)) /* just the address */ +#endif + diff --git a/sys/dev/vinum/vinuminterrupt.c b/sys/dev/vinum/vinuminterrupt.c new file mode 100644 index 0000000..e7eb034 --- /dev/null +++ b/sys/dev/vinum/vinuminterrupt.c @@ -0,0 +1,190 @@ +/* interrupt.c: bottom half of the driver */ + +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: interrupt.c,v 1.1 1998/08/13 06:12:27 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" +#include <miscfs/specfs/specdev.h> +#include <sys/resourcevar.h> + +void complete_raid5_write(struct rqelement *); +void freerq(struct request *rq); +void free_rqg(struct rqgroup *rqg); +void complete_rqe(struct buf *bp); +void sdio_done(struct buf *bp); + +/* Take a completed buffer, transfer the data back if + * it's a read, and complete the high-level request + * if this is the last subrequest. + * + * The bp parameter is in fact a struct rqelement, which + * includes a couple of extras at the end. + */ +void +complete_rqe(struct buf *bp) +{ + BROKEN_GDB; + struct rqelement *rqe; + struct request *rq; + struct rqgroup *rqg; + struct buf *ubp; /* user buffer */ + + rqe = (struct rqelement *) bp; /* point to the element element that completed */ + rqg = rqe->rqg; /* and the request group */ + rq = rqg->rq; /* and the complete request */ + + if ((bp->b_flags & B_ERROR) != 0) { /* transfer in error */ + if (bp->b_error != 0) /* did it return a number? */ + rq->error = bp->b_error; /* yes, put it in. */ + else if (rq->error == 0) /* no: do we have one already? */ + rq->error = EIO; /* no: catchall "I/O error" */ + if (rq->error == EIO) /* I/O error, */ + set_sd_state(rqe->sdno, sd_crashed, setstate_force); /* take the subdisk down */ + } + /* Now update the statistics */ + if (bp->b_flags & B_READ) { /* read operation */ + DRIVE[rqe->driveno].reads++; + DRIVE[rqe->driveno].bytes_read += bp->b_bcount; + SD[rqe->sdno].reads++; + SD[rqe->sdno].bytes_read += bp->b_bcount; + PLEX[rqe->rqg->plexno].reads++; + PLEX[rqe->rqg->plexno].bytes_read += bp->b_bcount; + } else { /* write operation */ + DRIVE[rqe->driveno].writes++; + DRIVE[rqe->driveno].bytes_written += bp->b_bcount; + SD[rqe->sdno].writes++; + SD[rqe->sdno].bytes_written += bp->b_bcount; + PLEX[rqe->rqg->plexno].writes++; + PLEX[rqe->rqg->plexno].bytes_written += bp->b_bcount; + } + ubp = rq->bp; /* user buffer */ + rqg->active--; /* one less request active */ + if (rqg->active == 0) /* request group finished, */ + rq->active--; /* one less */ + if (rq->active == 0) { /* request finished, */ +#if DEBUG + if (debug & 4) { + if (ubp->b_resid != 0) /* still something to transfer? */ + Debugger("resid"); + + { + int i; + for (i = 0; i < ubp->b_bcount; i += 512) /* XXX debug */ + if (((char *) ubp->b_data)[i] != '<') { /* and not what we expected */ + printf("At 0x%x (offset 0x%x): '%c' (0x%x)\n", + (int) (&((char *) ubp->b_data)[i]), + i, + ((char *) ubp->b_data)[i], + ((char *) ubp->b_data)[i]); + Debugger("complete_request checksum"); + } + } + } +#endif + + if (rq->error) { /* did we have an error? */ + ubp->b_flags |= B_ERROR; /* yes, propagate to user */ + ubp->b_error = rq->error; + } else + ubp->b_resid = 0; /* completed our transfer */ + if (rq->isplex == 0) /* volume request, */ + VOL[rq->volplex.volno].active--; /* another request finished */ + biodone(ubp); /* top level buffer completed */ + freerq(rq); /* return the request storage */ + } +} + + +/* Free a request block and anything hanging off it */ +void +freerq(struct request *rq) +{ + BROKEN_GDB; + struct rqgroup *rqg; + struct rqgroup *nrqg; /* next in chain */ + int rqno; + + for (rqg = rq->rqg; rqg != NULL; rqg = nrqg) { /* through the whole request chain */ + for (rqno = 0; rqno < rqg->count; rqno++) + if ((rqg->rqe[rqno].flags & XFR_MALLOCED) /* data buffer was malloced, */ + &&rqg->rqe[rqno].b.b_data) /* and the allocation succeeded */ + Free(rqg->rqe[rqno].b.b_data); /* free it */ + nrqg = rqg->next; /* note the next one */ + Free(rqg); /* and free this one */ + } + Free(rq); /* free the request itself */ +} + +void +free_rqg(struct rqgroup *rqg) +{ + if ((rqg->flags & XFR_GROUPOP) /* RAID 5 request */ + &&(rqg->rqe) /* got a buffer structure */ + &&(rqg->rqe->b.b_data)) /* and it has a buffer allocated */ + Free(rqg->rqe->b.b_data); /* free it */ +} + +/* I/O on subdisk completed */ +void +sdio_done(struct buf *bp) +{ + struct sdbuf *sbp; + + sbp = (struct sdbuf *) bp; + if (sbp->b.b_flags & B_ERROR) { /* had an error */ + bp->b_flags |= B_ERROR; + bp->b_error = sbp->b.b_error; + } + bp->b_resid = sbp->b.b_resid; + biodone(sbp->bp); /* complete the caller's I/O */ + /* Now update the statistics */ + if (bp->b_flags & B_READ) { /* read operation */ + DRIVE[sbp->driveno].reads++; + DRIVE[sbp->driveno].bytes_read += bp->b_bcount; + SD[sbp->sdno].reads++; + SD[sbp->sdno].bytes_read += bp->b_bcount; + } else { /* write operation */ + DRIVE[sbp->driveno].writes++; + DRIVE[sbp->driveno].bytes_written += bp->b_bcount; + SD[sbp->sdno].writes++; + SD[sbp->sdno].bytes_written += bp->b_bcount; + } + Free(sbp); +} diff --git a/sys/dev/vinum/vinumio.c b/sys/dev/vinum/vinumio.c new file mode 100644 index 0000000..a16a33a --- /dev/null +++ b/sys/dev/vinum/vinumio.c @@ -0,0 +1,886 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: io.c,v 1.16 1998/08/10 23:47:21 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#if __FreeBSD__ < 3 /* this is in sys/disklabel.h in 3.0 and on */ +#define DTYPE_VINUM 12 /* vinum volume */ +#endif + +#define REALLYKERNEL +#include "vinumhdr.h" +#include <miscfs/specfs/specdev.h> + +extern jmp_buf command_fail; /* return on a failed command */ +struct _ioctl_reply *ioctl_reply; /* data pointer, for returning error messages */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +/* pointer to ioctl p parameter, to save passing it around */ +extern struct proc *myproc; + +/* Open the device associated with the drive, and set drive's vp */ +int +open_drive(struct drive *drive, struct proc *p) +{ + BROKEN_GDB; + struct nameidata nd; + struct vattr va; + int error; + + if (drive->devicename[0] == '\0') /* no device name */ + sprintf(drive->devicename, "/dev/%s", drive->label.name); /* get it from the drive name */ + NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, drive->devicename, p); + error = vn_open(&nd, FREAD | FWRITE, 0); /* open the device */ + if (error != 0) { /* can't open? */ + set_drive_state(drive->driveno, drive_down, 1); + drive->lasterror = error; + printf("vinum open_drive %s: failed with error %d\n", drive->devicename, error); /* XXX */ + return error; + } + drive->vp = nd.ni_vp; + drive->p = p; + + if (drive->vp->v_usecount > 1) { /* already in use? */ +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); + drive->lasterror = EBUSY; + printf("vinum open_drive %s: Drive in use\n", drive->devicename); /* XXX */ + return EBUSY; + } + error = VOP_GETATTR(drive->vp, &va, NOCRED, p); + if (error) { +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); + drive->lasterror = error; + printf("vinum open_drive %s: GETAATTR returns error %d\n", drive->devicename, error); /* XXX */ + return error; + } + drive->dev = va.va_rdev; /* device */ + + if (va.va_type != VBLK) { /* only consider block devices */ +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); /* this also closes the drive */ + drive->lasterror = ENOTBLK; + printf("vinum open_drive %s: Not a block device\n", drive->devicename); /* XXX */ + return ENOTBLK; + } + drive->vp->v_numoutput = 0; +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + return 0; +} + +/* Set some variables in the drive struct + * in more convenient form. Return error indication */ +int +set_drive_parms(struct drive *drive) +{ + drive->blocksize = BLKDEV_IOSIZE; /* XXX do we need this? */ + drive->secsperblock = drive->blocksize /* number of sectors per block */ + / drive->partinfo.disklab->d_secsize; + + /* Now update the label part */ + bcopy(hostname, drive->label.sysname, VINUMHOSTNAMELEN); /* put in host name */ +#if __FreeBSD__ >= 3 + getmicrotime(&drive->label.date_of_birth); /* and current time */ +#else + drive->label.date_of_birth = time; /* and current time */ +#endif + drive->label.drive_size = ((u_int64_t) drive->partinfo.part->p_size) /* size of the drive in bytes */ + *((u_int64_t) drive->partinfo.disklab->d_secsize); + + /* number of sectors available for subdisks */ + drive->sectors_available = drive->label.drive_size / DEV_BSIZE - DATASTART; + + /* XXX Bug in 3.0 as of January 1998: you can open + * non-existent slices. They have a length of 0 */ + if (drive->label.drive_size < MINVINUMSLICE) { /* too small to worry about */ + set_drive_state(drive->driveno, drive_down, 1); + printf("vinum open_drive %s: Drive too small\n", drive->devicename); /* XXX */ + drive->lasterror = ENOSPC; + return ENOSPC; + } + drive->freelist_size = INITIAL_DRIVE_FREELIST; /* initial number of entries */ + drive->freelist = (struct drive_freelist *) + Malloc(INITIAL_DRIVE_FREELIST * sizeof(struct drive_freelist)); + if (drive->freelist == NULL) /* can't malloc, dammit */ + return ENOSPC; + drive->freelist_entries = 1; /* just (almost) the complete drive */ + drive->freelist[0].offset = DATASTART; /* starts here */ + drive->freelist[0].sectors = (drive->label.drive_size >> DEV_BSHIFT) - DATASTART; /* and it's this long */ + set_drive_state(drive->driveno, drive_up, 1); /* our drive is accessible */ + return 0; +} + +/* Initialize a drive: open the device and add device + * information */ +int +init_drive(struct drive *drive) +{ + BROKEN_GDB; + int error; + + if (drive->devicename[0] == '\0') { /* no device name yet, default to drive name */ + drive->lasterror = EINVAL; + printf("vinum: Can't open drive without drive name\n"); /* XXX */ + return EINVAL; + } + error = open_drive(drive, myproc); /* open the drive */ + if (error) + return error; + + error = VOP_IOCTL(drive->vp, /* get the partition information */ + DIOCGPART, + (caddr_t) & drive->partinfo, + FREAD, + NOCRED, + myproc); + if (error) { + printf("vinum open_drive %s: Can't get partition information, error %d\n", + drive->devicename, + error); /* XXX */ + close_drive(drive); + drive->lasterror = error; + set_drive_state(drive->driveno, drive_down, 1); + return error; + } + if (drive->partinfo.part->p_fstype != 0) { /* not plain */ + drive->lasterror = EFTYPE; + printf("vinum open_drive %s: Wrong partition type for vinum\n", drive->devicename); /* XXX */ + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); + return EFTYPE; + } + return set_drive_parms(drive); /* set various odds and ends */ +} + +/* Close a drive if it's open. No errors */ +void +close_drive(struct drive *drive) +{ + if (drive->vp) { + vn_close(drive->vp, FREAD | FWRITE, NOCRED, drive->p); + drive->vp = NULL; + } +} + +/* Remove drive from the configuration. + * Caller must ensure that it isn't active + */ +void +remove_drive(int driveno) +{ + BROKEN_GDB; + struct drive *drive = &vinum_conf.drive[driveno]; + long long int nomagic = VINUM_NOMAGIC; /* no magic number */ + + write_drive(drive, /* obliterate the magic, but leave a hint */ + (char *) &nomagic, + 8, + VINUM_LABEL_OFFSET); + close_drive(drive); /* and close it */ + drive->state = drive_unallocated; /* and forget everything we knew about it */ + save_config(); /* and save the updated configuration */ +} + +/* Transfer drive data. Usually called from one of these defines; + + * #define read_drive(a, b, c, d) driveio (a, b, c, d, B_READ) + * #define write_drive(a, b, c, d) driveio (a, b, c, d, B_WRITE) + * + * Return error number + */ +int +driveio(struct drive *drive, void *buf, size_t length, off_t offset, int flag) +{ + BROKEN_GDB; + int error; + struct buf *bp; + int spl; + + error = 0; + + /* Get a buffer */ + bp = (struct buf *) Malloc(sizeof(struct buf)); /* get a buffer */ + CHECKALLOC(bp, "Can't allocate memory"); + + bzero(&buf, sizeof(buf)); + bp->b_flags = B_BUSY | flag; /* tell us when it's done */ + bp->b_iodone = drive_io_done; /* here */ + bp->b_proc = myproc; /* process */ + bp->b_dev = drive->vp->v_un.vu_specinfo->si_rdev; /* device */ + if (offset & (drive->partinfo.disklab->d_secsize - 1)) /* not on a block boundary */ + bp->b_blkno = offset / drive->partinfo.disklab->d_secsize; /* block number */ + bp->b_data = buf; + bp->b_vp = drive->vp; /* vnode */ + bp->b_bcount = length; + bp->b_bufsize = length; + + (*bdevsw[major(bp->b_dev)]->d_strategy) (bp); /* initiate the transfer */ + + spl = splbio(); + while ((bp->b_flags & B_DONE) == 0) { + bp->b_flags |= B_CALL; /* wake me again */ + tsleep((caddr_t) bp, PRIBIO, "driveio", 0); /* and wait for it to complete */ + } + splx(spl); + if (bp->b_flags & B_ERROR) /* didn't work */ + error = bp->b_error; /* get the error return */ + Free(bp); /* then return the buffer */ + return error; +} + +/* Read data from a drive + + * Return error number + */ +int +read_drive(struct drive *drive, void *buf, size_t length, off_t offset) +{ + BROKEN_GDB; + int error; + struct buf *bp; + daddr_t nextbn; + long bscale; + + struct uio uio; + struct iovec iov; + daddr_t blocknum; /* block number */ + int blockoff; /* offset in block */ + int count; /* amount to transfer */ + + iov.iov_base = buf; + iov.iov_len = length; + + uio.uio_iov = &iov; + uio.uio_iovcnt = length; + uio.uio_offset = offset; + uio.uio_resid = length; + uio.uio_segflg = UIO_SYSSPACE; + uio.uio_rw = UIO_READ; + uio.uio_procp = myproc; + + bscale = btodb(drive->blocksize); /* mask off offset from block number */ + do { + blocknum = btodb(uio.uio_offset) & ~(bscale - 1); /* get the block number */ + blockoff = uio.uio_offset % drive->blocksize; /* offset in block */ + count = min((unsigned) (drive->blocksize - blockoff), /* amount to transfer in this block */ + uio.uio_resid); + + /* XXX Check this. I think the test is wrong */ + if (drive->vp->v_lastr + bscale == blocknum) { /* did our last read finish in this block? */ + nextbn = blocknum + bscale; /* note the end of the transfer */ + error = breadn(drive->vp, /* and read with read-ahead */ + blocknum, + (int) drive->blocksize, + &nextbn, + (int *) &drive->blocksize, + 1, + NOCRED, + &bp); + } else /* random read: just read this block */ + error = bread(drive->vp, blocknum, (int) drive->blocksize, NOCRED, &bp); + drive->vp->v_lastr = blocknum; /* note the last block we read */ + count = min(count, drive->blocksize - bp->b_resid); + if (error) { + brelse(bp); + return error; + } + error = uiomove((char *) bp->b_data + blockoff, count, &uio); /* move the data */ + brelse(bp); + } + while (error == 0 && uio.uio_resid > 0 && count != 0); + return error; +} + +/* Write data to a drive + + * Return error number + */ +int +write_drive(struct drive *drive, void *buf, size_t length, off_t offset) +{ + BROKEN_GDB; + int error; + struct buf *bp; + struct uio uio; + struct iovec iov; + daddr_t blocknum; /* block number */ + int blockoff; /* offset in block */ + int count; /* amount to transfer */ + int blockshift; + + if (drive->state == drive_down) /* currently down */ + return 0; /* ignore */ + if (drive->vp == NULL) { + drive->lasterror = ENODEV; + return ENODEV; /* not configured yet */ + } + iov.iov_base = buf; + iov.iov_len = length; + + uio.uio_iov = &iov; + uio.uio_iovcnt = length; + uio.uio_offset = offset; + uio.uio_resid = length; + uio.uio_segflg = UIO_SYSSPACE; + uio.uio_rw = UIO_WRITE; + uio.uio_procp = myproc; + + error = 0; + blockshift = btodb(drive->blocksize) - 1; /* amount to shift block number + * to get sector number */ + do { + blocknum = btodb(uio.uio_offset) & ~blockshift; /* get the block number */ + blockoff = uio.uio_offset % drive->blocksize; /* offset in block */ + count = min((unsigned) (drive->blocksize - blockoff), /* amount to transfer in this block */ + uio.uio_resid); + if (count == drive->blocksize) /* the whole block */ + bp = getblk(drive->vp, blocknum, drive->blocksize, 0, 0); /* just get it */ + else /* partial block: */ + error = bread(drive->vp, /* read it first */ + blocknum, + drive->blocksize, + NOCRED, + &bp); + count = min(count, drive->blocksize - bp->b_resid); /* how much will we transfer now? */ + if (error == 0) + error = uiomove((char *) bp->b_data + blockoff, /* move the data to the block */ + count, + &uio); + if (error) { + brelse(bp); + drive->lasterror = error; + switch (error) { + case EIO: + set_drive_state(drive->driveno, drive_down, 1); + break; + + /* XXX Add other possibilities here */ + default: + } + return error; + } + if (count + blockoff == drive->blocksize) + /* The transfer goes to the end of the block. There's + * no need to wait for any more data to arrive. */ + bawrite(bp); /* start the write now */ + else + bdwrite(bp); /* do a delayed write */ + } + while (error == 0 && uio.uio_resid > 0 && count != 0); + if (error) + drive->lasterror = error; + return error; /* OK */ +} + +/* Wake up on completion */ +void +drive_io_done(struct buf *bp) +{ + BROKEN_GDB; + wakeup((caddr_t) bp); /* Wachet auf! */ + bp->b_flags &= ~B_CALL; /* don't do this again */ +} + +/* Check a drive for a vinum header. If found, + * update the drive information. We come here + * with a partially populated drive structure + * which includes the device name. + * + * Return information on what we found + */ +enum drive_label_info +read_drive_label(struct drive *drive) +{ + BROKEN_GDB; + int error; + int result; /* result of our search */ + struct vinum_hdr *vhdr; /* and as header */ + + error = init_drive(drive); /* find the drive */ + if (error) /* find the drive */ + return DL_CANT_OPEN; /* not ours */ + + vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* allocate buffers */ + CHECKALLOC(vhdr, "Can't allocate memory"); + + error = read_drive(drive, (void *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); + if (vhdr->magic == VINUM_MAGIC) { /* ours! */ + if (drive->label.name[0] /* we have a name for this drive */ + &&(strcmp(drive->label.name, vhdr->label.name))) { /* but it doesn't match the real name */ + drive->lasterror = EINVAL; + result = DL_WRONG_DRIVE; /* it's the wrong drive */ + } else { + set_drive_parms(drive); /* and set other parameters */ + result = DL_OURS; + } + /* We copy the drive anyway so that we have + * the correct name in the drive info. This + * may not be the name specified */ + drive->label = vhdr->label; /* put in the label information */ + } else if (vhdr->magic == VINUM_NOMAGIC) /* was ours, but we gave it away */ + result = DL_DELETED_LABEL; + else + result = DL_NOT_OURS; /* we could have it, but we don't yet */ + Free(vhdr); /* that's all. */ + return result; +} + +/* Check a drive for a vinum header. If found, + * read configuration information from the drive and + * incorporate the data into the configuration. + * + * Return error number + */ +int +check_drive(char *drivename) +{ + BROKEN_GDB; + int error; + struct nameidata nd; /* mount point credentials */ + char *config_text; /* read the config info from disk into here */ + volatile char *cptr; /* pointer into config information */ + char *eptr; /* end pointer into config information */ + int driveno; + struct drive *drive; + char *config_line; /* copy the config line to */ + + driveno = find_drive_by_dev(drivename, 1); /* doesn't exist, create it */ + drive = &vinum_conf.drive[driveno]; /* and get a pointer */ + strcpy(drive->devicename, drivename); /* put in device name */ + + if (read_drive_label(drive) == DL_OURS) { /* ours! */ + config_text = (char *) Malloc(MAXCONFIG * 2); /* allocate buffers */ + CHECKALLOC(config_text, "Can't allocate memory"); + config_line = (char *) Malloc(MAXCONFIGLINE * 2); /* allocate buffers */ + CHECKALLOC(config_line, "Can't allocate memory"); + + /* Read in both copies of the configuration information */ + error = read_drive(drive, config_text, MAXCONFIG * 2, VINUM_CONFIG_OFFSET); + + if (error != 0) { + printf("vinum: Can't read device %s, error %d\n", drive->devicename, error); + Free(config_text); + Free(config_line); + free_drive(drive); /* give it back */ + return error; + } + /* XXX At this point, check that the two copies are the same, and do something useful if not. + * In particular, consider which is newer, and what this means for the integrity of the + * data on the drive */ + + /* Parse the configuration, and add it to the global configuration */ + for (cptr = config_text; *cptr != '\0';) { /* love this style(9) */ + volatile int parse_status; /* return value from parse_config */ + + for (eptr = config_line; (*cptr != '\n') && (*cptr != '\0');) /* until the end of the line */ + *eptr++ = *cptr++; + *eptr = '\0'; /* and delimit */ + if (setjmp(command_fail) == 0) { /* come back here on error and continue */ + parse_status = parse_config(config_line, &keyword_set); /* parse the config line */ + if (parse_status < 0) { /* error in config */ + /* This config should have been parsed in user + * space. If we run into problems here, something + * serious is afoot. Complain and let the user + * snarf the config to see what's wrong */ + printf("vinum: Config error on drive %s, aborting integration\n", nd.ni_dirp); + Free(config_text); + Free(config_line); + free_drive(drive); /* give it back */ + return EINVAL; + } + } + while (*cptr == '\n') + cptr++; /* skip to next line */ + } + Free(config_text); + if ((vinum_conf.flags & VF_READING_CONFIG) == 0) /* not reading config */ + updateconfig(0); /* update object states */ + printf("vinum: read configuration from %s\n", drivename); + return 0; /* it all worked */ + } else { /* no vinum label found */ + if (drive->lasterror) { + set_drive_state(drive->driveno, drive_down, 1); + return drive->lasterror; + } else + return ENODEV; /* not our device */ + } +} + +/* Kludge: kernel printf doesn't handle longs correctly XXX */ +static char *lltoa(long long l, char *s); +static char *sappend(char *txt, char *s); + +static char * +lltoa(long long l, char *s) +{ + if (l < 0) { + *s++ = '-'; + l = -l; + } + if (l > 9) { + s = lltoa(l / 10, s); + l %= 10; + } + *s++ = l + '0'; + return s; +} + +static char * +sappend(char *txt, char *s) +{ + while (*s++ = *txt++); + return s - 1; +} + +/* Format the configuration in text form into the buffer + * at config. Don't go beyond len bytes + * XXX this stinks. Fix soon. */ +void +format_config(char *config, int len) +{ + BROKEN_GDB; + int i; + int j; + char *s = config; + + bzero(config, len); + + /* First write the drive configuration */ + for (i = 0; i < vinum_conf.drives_used; i++) { + struct drive *drive; + + drive = &vinum_conf.drive[i]; + if (drive->state != drive_unallocated) { + sprintf(s, + "drive %s state %s device %s\n", + drive->label.name, + drive_state(drive->state), + drive->devicename); + while (*s) + s++; /* find the end */ + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } + + /* Then the volume configuration */ + for (i = 0; i < vinum_conf.volumes_used; i++) { + struct volume *vol; + + vol = &vinum_conf.volume[i]; + if (vol->state != volume_unallocated) { + if (vol->preferred_plex >= 0) /* preferences, */ + sprintf(s, + "volume %s state %s readpol prefer %s", + vol->name, + volume_state(vol->state), + vinum_conf.plex[vol->preferred_plex].name); + else /* default round-robin */ + sprintf(s, + "volume %s state %s", + vol->name, + volume_state(vol->state)); + while (*s) + s++; /* find the end */ + s = sappend("\n", s); + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } + + /* Then the plex configuration */ + for (i = 0; i < vinum_conf.plexes_used; i++) { + struct plex *plex; + + plex = &vinum_conf.plex[i]; + if (plex->state != plex_unallocated) { + sprintf(s, "plex name %s state %s org %s ", + plex->name, + plex_state(plex->state), + plex_org(plex->organization)); + while (*s) + s++; /* find the end */ + if ((plex->organization == plex_striped) + ) { + sprintf(s, "%db ", (int) plex->stripesize); + while (*s) + s++; /* find the end */ + } + if (plex->volno >= 0) /* we have a volume */ + sprintf(s, "vol %s ", vinum_conf.volume[plex->volno].name); + while (*s) + s++; /* find the end */ + for (j = 0; j < plex->subdisks; j++) { + sprintf(s, " sd %s", vinum_conf.sd[plex->sdnos[j]].name); + } + s = sappend("\n", s); + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } + + /* And finally the subdisk configuration */ + for (i = 0; i < vinum_conf.subdisks_used; i++) { + struct sd *sd = &vinum_conf.sd[i]; /* XXX */ + if (vinum_conf.sd[i].state != sd_unallocated) { + sprintf(s, + "sd name %s drive %s plex %s state %s len ", + sd->name, + vinum_conf.drive[sd->driveno].label.name, + vinum_conf.plex[sd->plexno].name, + sd_state(sd->state)); + while (*s) + s++; /* find the end */ + s = lltoa(sd->sectors, s); + s = sappend("b driveoffset ", s); + s = lltoa(sd->driveoffset, s); + s = sappend("b plexoffset ", s); + s = lltoa(sd->plexoffset, s); + s = sappend("b\n", s); + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } +} + +/* Write the configuration to all vinum slices */ +int +save_config(void) +{ + BROKEN_GDB; + int error; + int written_config; /* set when we firstnwrite the config to disk */ + int driveno; + struct drive *drive; /* point to current drive info */ + struct vinum_hdr *vhdr; /* and as header */ + char *config; /* point to config data */ + int wlabel_on; /* to set writing label on/off */ + + /* don't save the configuration while we're still working on it */ + if (vinum_conf.flags & VF_CONFIGURING) + return 0; + written_config = 0; /* no config written yet */ + /* Build a volume header */ + vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* get space for the config data */ + CHECKALLOC(vhdr, "Can't allocate config data"); + vhdr->magic = VINUM_MAGIC; /* magic number */ + vhdr->config_length = MAXCONFIG; /* length of following config info */ + + config = Malloc(MAXCONFIG); /* get space for the config data */ + CHECKALLOC(config, "Can't allocate config data"); + + format_config(config, MAXCONFIG); + error = 0; /* no errors yet */ + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + drive = &vinum_conf.drive[driveno]; /* point to drive */ + + if (drive->state != drive_down) { +#if (__FreeBSD__ >= 3) + getmicrotime(&drive->label.last_update); /* time of last update is now */ +#else + drive->label.last_update = time; /* time of last update is now */ +#endif + bcopy((char *) &drive->label, /* and the label info from the drive structure */ + (char *) &vhdr->label, + sizeof(vhdr->label)); + if ((drive->state != drive_unallocated) + && (drive->state != drive_uninit)) { + wlabel_on = 1; /* enable writing the label */ + error = VOP_IOCTL(drive->vp, /* make the label writeable */ + DIOCWLABEL, + (caddr_t) & wlabel_on, + FWRITE, + NOCRED, + myproc); + if (error == 0) + error = write_drive(drive, vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); + if (error == 0) + error = write_drive(drive, config, MAXCONFIG, VINUM_CONFIG_OFFSET); + wlabel_on = 0; /* enable writing the label */ + VOP_IOCTL(drive->vp, /* make the label non-writeable again */ + DIOCWLABEL, + (caddr_t) & wlabel_on, + FWRITE, + NOCRED, + myproc); + if (error) { + printf("vinum: Can't write config to %s, error %d\n", drive->devicename, error); + set_drive_state(drive->driveno, drive_down, 1); + } else + written_config = 1; /* we've written it on at least one drive */ + } + } + } + Free(vhdr); + Free(config); + return written_config == 0; /* return 1 if we failed to write config */ +} + +/* Disk labels are a mess. The correct way to access them + * is with the DIOC[GSW]DINFO ioctls, but some programs, such + * as newfs, access the disk directly, so we have to write + * things there. We do this only on request. If a user + * request tries to read it directly, we fake up one on the fly. + */ + +/* get_volume_label returns a label structure to lp, which + * is allocated by the caller */ +void +get_volume_label(struct volume *vol, struct disklabel *lp) +{ + bzero(lp, sizeof(struct disklabel)); + + strncpy(lp->d_typename, "vinum", sizeof(lp->d_typename)); + lp->d_type = DTYPE_VINUM; + strncpy(lp->d_packname, vol->name, min(sizeof(lp->d_packname), sizeof(vol->name))); + lp->d_rpm = 14400 * vol->plexes; /* to keep them guessing */ + lp->d_interleave = 1; + lp->d_flags = 0; + + /* Fitting unto the vine, a vinum has a single + * track with all its sectors */ + lp->d_secsize = DEV_BSIZE; /* bytes per sector */ + lp->d_nsectors = vol->size; /* data sectors per track */ + lp->d_ntracks = 1; /* tracks per cylinder */ + lp->d_ncylinders = 1; /* data cylinders per unit */ + lp->d_secpercyl = vol->size; /* data sectors per cylinder */ + lp->d_secperunit = vol->size; /* data sectors per unit */ + + lp->d_bbsize = BBSIZE; + lp->d_sbsize = SBSIZE; + + lp->d_magic = DISKMAGIC; + lp->d_magic2 = DISKMAGIC; + + /* Set up partitions a, b and c to be identical + * and the size of the volume. a is UFS, b is + * swap, c is nothing */ + lp->d_partitions[0].p_size = vol->size; + lp->d_partitions[0].p_fsize = 1024; + lp->d_partitions[0].p_fstype = FS_BSDFFS; /* FreeBSD File System :-) */ + lp->d_partitions[0].p_fsize = 1024; /* FS fragment size */ + lp->d_partitions[0].p_frag = 8; /* and fragments per block */ + lp->d_partitions[SWAP_PART].p_size = vol->size; + lp->d_partitions[SWAP_PART].p_fstype = FS_SWAP; /* swap partition */ + lp->d_partitions[LABEL_PART].p_size = vol->size; + lp->d_npartitions = LABEL_PART + 1; + strncpy(lp->d_packname, vol->name, min(sizeof(lp->d_packname), sizeof(vol->name))); + lp->d_checksum = dkcksum(lp); +} + +int +write_volume_label(int volno) +{ + struct disklabel *lp; + struct buf *bp; + struct disklabel *dlp; + struct volume *vol; + int error; + + lp = (struct disklabel *) Malloc((sizeof(struct disklabel) + (DEV_BSIZE - 1)) & (DEV_BSIZE - 1)); + if (lp == 0) + return ENOMEM; + + if ((unsigned) (volno) >= (unsigned) vinum_conf.volumes_used) /* invalid volume */ + return ENOENT; + + vol = &VOL[volno]; /* volume in question */ + if (vol->state == volume_unallocated) /* nothing there */ + return ENOENT; + + get_volume_label(vol, lp); /* get the label */ + + /* Now write to disk. This code is derived from the + * system writedisklabel (), which does silly things + * like reading the label and refusing to write + * unless it's already there. */ + bp = geteblk((int) lp->d_secsize); /* get a buffer */ + bp->b_dev = minor(vol->devno) | (CDEV_MAJOR << MAJORDEV_SHIFT); /* our own raw volume */ + bp->b_blkno = LABELSECTOR * ((int) lp->d_secsize / DEV_BSIZE); + bp->b_bcount = lp->d_secsize; + bzero(bp->b_data, lp->d_secsize); + dlp = (struct disklabel *) bp->b_data; + *dlp = *lp; + bp->b_flags &= ~B_INVAL; + bp->b_flags |= B_BUSY | B_WRITE; + vinumstrategy(bp); /* write it out */ + error = biowait(bp); + bp->b_flags |= B_INVAL | B_AGE; + brelse(bp); + return error; +} + +/* Initialize a subdisk */ +int +initsd(int sdno) +{ + return 0; +} diff --git a/sys/dev/vinum/vinumio.h b/sys/dev/vinum/vinumio.h new file mode 100644 index 0000000..be79528 --- /dev/null +++ b/sys/dev/vinum/vinumio.h @@ -0,0 +1,132 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumio.h,v 1.10 1998/08/10 05:46:19 grog Exp grog $ + */ + +#define MAX_IOCTL_REPLY 256 +#define L 'F' /* ID letter of our ioctls */ +/* VINUM_CREATE returns a buffer of this kind */ +struct _ioctl_reply { + int error; + char msg[MAX_IOCTL_REPLY]; +}; + +/* ioctl requests */ +#define BUFSIZE 1024 /* size of buffer, including continuations */ +#define VINUM_CREATE _IOC(IOC_IN | IOC_OUT, L, 64, BUFSIZE) /* configure vinum */ +#define VINUM_GETCONFIG _IOR(L, 65, struct _vinum_conf) /* get global config */ +#define VINUM_DRIVECONFIG _IOWR(L, 66, struct drive) /* get drive config */ +#define VINUM_SDCONFIG _IOWR(L, 67, struct sd) /* get subdisk config */ +#define VINUM_PLEXCONFIG _IOWR(L, 68, struct plex) /* get plex config */ +#define VINUM_VOLCONFIG _IOWR(L, 69, struct volume) /* get volume config */ +#define VINUM_PLEXSDCONFIG _IOWR(L, 70, struct sd) /* get sd config for plex (plex, sdno) */ +#define VINUM_GETFREELIST _IOWR(L, 71, struct drive_freelist) /* get freelist element (drive, fe) */ +#define VINUM_SAVECONFIG _IOC(0, L, 72, 0) /* release locks, update, write config to disk */ +#define VINUM_RESETCONFIG _IOC(0, L, 73, 0) /* trash config on disk */ +#define VINUM_INIT _IOC(0, L, 74, 0) /* read config from disk */ +#ifdef DEBUG + +struct debuginfo { + int changeit; + int param; +}; + +#define VINUM_DEBUG _IOWR(L, 75, struct debuginfo) /* call the debugger from ioctl () */ +#endif + +enum objecttype { + drive_object, + sd_object, + plex_object, + volume_object, + invalid_object +}; + +/* Start an object. Pass two integers: + * msg [0] index in vinum_conf.<object> + * msg [1] type of object (see below) + * + * Return ioctl_reply + */ +#define VINUM_SETSTATE _IOC(IOC_IN | IOC_OUT, L, 76, MAX_IOCTL_REPLY) /* start an object */ + +/* The state to set with VINUM_SETSTATE. Since + * each object has a different set of states, we + * need to translate later */ +enum objectstate { + object_down, + object_initializing, + object_up +}; + +/* This structure is used for modifying objects + * (VINUM_SETSTATE, VINUM_REMOVE, VINUM_RESETSTATS, VINUM_ATTACH, + * VINUM_DETACH, VINUM_REPLACE + */ +struct vinum_ioctl_msg { + int index; + enum objecttype type; + enum objectstate state; /* state to set (VINUM_SETSTATE) */ + int force; /* do it even if it doesn't make sense */ + int recurse; /* recurse (VINUM_REMOVE) */ + int otherobject; /* superordinate object (attach), + * replacement object (replace) */ + int rename; /* rename object (attach) */ + int64_t offset; /* offset of subdisk (for attach) */ +}; + +#define VINUM_RELEASECONFIG _IOC(0, L, 77, 0) /* release locks and write config to disk */ +#define VINUM_STARTCONFIG _IOC(0, L, 78, 0) /* start a configuration operation */ +#define VINUM_MEMINFO _IOR(L, 79, struct meminfo) /* get memory usage summary */ +#define VINUM_MALLOCINFO _IOWR(L, 80, struct mc) /* get specific malloc information [i] */ +#define VINUM_LABEL _IOC(IOC_IN | IOC_OUT, L, 81, MAX_IOCTL_REPLY) /* label a volume */ +#define VINUM_INITSD _IOW(L, 82, int) /* initialize a subdisk */ +#define VINUM_REMOVE _IOC(IOC_IN | IOC_OUT, L, 83, MAX_IOCTL_REPLY) /* remove an object */ +#define VINUM_GETUNMAPPED _IOWR(L, 84, struct plexregion) /* get unmapped element (plex, re) */ +#define VINUM_GETDEFECTIVE _IOWR(L, 85, struct plexregion) /* get defective element (plex, re) */ +#define VINUM_RESETSTATS _IOC(IOC_IN | IOC_OUT, L, 86, MAX_IOCTL_REPLY) /* reset object stats */ +#define VINUM_ATTACH _IOC(IOC_IN | IOC_OUT, L, 87, MAX_IOCTL_REPLY) /* reset object stats */ +#define VINUM_DETACH _IOC(IOC_IN | IOC_OUT, L, 88, MAX_IOCTL_REPLY) /* reset object stats */ + +struct vinum_rename_msg { + int index; + int recurse; /* rename subordinate objects too */ + enum objecttype type; + char newname[MAXNAME]; /* new name to give to object */ +}; + +#define VINUM_RENAME _IOC(IOC_IN | IOC_OUT, L, 89, MAX_IOCTL_REPLY) /* reset object stats */ +#define VINUM_REPLACE _IOC(IOC_IN | IOC_OUT, L, 90, MAX_IOCTL_REPLY) /* reset object stats */ diff --git a/sys/dev/vinum/vinumioctl.c b/sys/dev/vinum/vinumioctl.c new file mode 100644 index 0000000..6dbe3c6 --- /dev/null +++ b/sys/dev/vinum/vinumioctl.c @@ -0,0 +1,787 @@ +/* XXX replace all the checks on object validity with + * calls to valid<object> */ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumioctl.c,v 1.1 1998/08/14 08:46:10 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "sys/sysproto.h" /* for sync(2) */ +#ifdef DEBUG +#include <sys/reboot.h> +#endif + +jmp_buf command_fail; /* return on a failed command */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +/* pointer to ioctl p parameter, to save passing it around */ +struct proc *myproc; + +int vinum_inactive(void); +void free_vinum(int); +void attachobject(struct vinum_ioctl_msg *); +void detachobject(struct vinum_ioctl_msg *); +void renameobject(struct vinum_rename_msg *); +void replaceobject(struct vinum_ioctl_msg *); + +/* ioctl routine */ +int +vinumioctl(dev_t dev, +#if __FreeBSD__ >= 3 + u_long cmd, +#else + int cmd, +#endif + caddr_t data, + int flag, + struct proc *p) +{ + BROKEN_GDB; + unsigned int objno; + int error = 0; + struct volume *vol; + unsigned int index; /* for transferring config info */ + unsigned int sdno; /* for transferring config info */ + int fe; /* free list element number */ + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) data; /* struct to return */ + + struct devcode *device = (struct devcode *) &dev; + + /* First, decide what we're looking at */ + switch (device->type) { + case VINUM_SUPERDEV_TYPE: + myproc = p; /* save pointer to process */ + ioctl_reply = (struct _ioctl_reply *) data; /* save the address to reply to */ + error = setjmp(command_fail); /* come back here on error */ + if (error) /* bombed out */ + return 0; /* the reply will contain meaningful info */ + switch (cmd) { + /* XXX #ifdef DEBUG */ + case VINUM_DEBUG: + boothowto |= RB_GDB; /* serial debug line */ + if (((struct debuginfo *) data)->changeit) /* change debug settings */ + debug = (((struct debuginfo *) data)->param); + else + Debugger("vinum debug"); + ioctl_reply = (struct _ioctl_reply *) data; /* reinstate the address to reply to */ + ioctl_reply->error = 0; + return 0; + /* XXX #endif */ + + case VINUM_CREATE: /* create a vinum object */ + error = lock_config(); /* get the config for us alone */ + if (error) /* can't do it, */ + return error; /* give up */ + error = setjmp(command_fail); /* come back here on error */ + if (error == 0) { /* first time, */ + parse_user_config((char *) data, &keyword_set); /* update the config */ + ioctl_reply->error = 0; /* no error if we make it here */ + } else if (ioctl_reply->error == 0) { /* longjmp, but no error status */ + ioctl_reply->error = EINVAL; /* note that something's up */ + ioctl_reply->msg[0] = '\0'; /* no message? */ + } + unlock_config(); + return 0; /* must be 0 to return the real error info */ + + case VINUM_GETCONFIG: /* get the configuration information */ + bcopy(&vinum_conf, data, sizeof(vinum_conf)); + return 0; + + /* start configuring the subsystem */ + case VINUM_STARTCONFIG: + return start_config(); /* just lock it */ + + /* Move the individual parts of the config to user space. + + * Specify the index of the object in the first word of data, + * and return the object there + */ + case VINUM_DRIVECONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.drives_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&DRIVE[index], data, sizeof(struct drive)); /* copy the config item out */ + return 0; + + case VINUM_SDCONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.subdisks_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&SD[index], data, sizeof(struct sd)); /* copy the config item out */ + return 0; + + case VINUM_PLEXCONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.plexes_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&PLEX[index], data, sizeof(struct plex)); /* copy the config item out */ + return 0; + + case VINUM_VOLCONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.volumes_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&VOL[index], data, sizeof(struct volume)); /* copy the config item out */ + return 0; + + case VINUM_PLEXSDCONFIG: + index = *(int *) data; /* get the plex index */ + sdno = ((int *) data)[1]; /* and the sd index */ + if ((index >= (unsigned) vinum_conf.plexes_used) /* plex doesn't exist */ + ||(sdno >= PLEX[index].subdisks)) /* or it doesn't have this many subdisks */ + return EFAULT; /* bang */ + bcopy(&SD[PLEX[index].sdnos[sdno]], /* copy the config item out */ + data, + sizeof(struct sd)); + return 0; + + case VINUM_SAVECONFIG: + if (VFLAGS & VF_CONFIGURING) { /* must be us, the others are asleep */ + finish_config(1); /* finish the configuration and update it */ + error = save_config(); /* save configuration to disk */ + } else + error = EINVAL; /* queue up for this one, please */ + return error; + + case VINUM_RELEASECONFIG: /* release the config */ + if (VFLAGS & VF_CONFIGURING) { /* must be us, the others are asleep */ + finish_config(0); /* finish the configuration, don't change it */ + error = save_config(); /* save configuration to disk */ + } else + error = EINVAL; /* release what config? */ + return error; + + case VINUM_INIT: + ioctl_reply = (struct _ioctl_reply *) data; /* reinstate the address to reply to */ + ioctl_reply->error = 0; + return 0; + + case VINUM_RESETCONFIG: + if (vinum_inactive() && (vinum_conf.opencount < 2)) { /* if we're not active */ + /* Note the open count. We may be called from v, so we'll be open. + * Keep the count so we don't underflow */ + int oc = vinum_conf.opencount; + free_vinum(1); /* clean up everything */ + printf("vinum: CONFIGURATION OBLITERATED\n"); + vinum_conf.opencount = oc; + ioctl_reply = (struct _ioctl_reply *) data; /* reinstate the address to reply to */ + ioctl_reply->error = 0; + return 0; + } + return EBUSY; + + case VINUM_SETSTATE: + setstate((struct vinum_ioctl_msg *) data); /* set an object state */ + return 0; + + case VINUM_MEMINFO: + vinum_meminfo(data); + return 0; + + case VINUM_MALLOCINFO: + return vinum_mallocinfo(data); + + case VINUM_LABEL: /* label a volume */ + ioctl_reply->error = write_volume_label(*(int *) data); /* index of the volume to label */ + ioctl_reply->msg[0] = '\0'; /* no message */ + return 0; + + case VINUM_REMOVE: + remove((struct vinum_ioctl_msg *) data); /* remove an object */ + return 0; + + case VINUM_GETFREELIST: /* get a drive free list element */ + index = *(int *) data; /* get the drive index */ + fe = ((int *) data)[1]; /* and the free list element */ + if ((index >= (unsigned) vinum_conf.drives_used) /* plex doesn't exist */ + ||(DRIVE[index].state == drive_unallocated)) + return ENODEV; + if (fe >= DRIVE[index].freelist_entries) /* no such entry */ + return ENOENT; + bcopy(&DRIVE[index].freelist[fe], + data, + sizeof(struct drive_freelist)); + return 0; + + case VINUM_GETDEFECTIVE: /* get a plex defective area element */ + index = *(int *) data; /* get the plex index */ + fe = ((int *) data)[1]; /* and the region number */ + if ((index >= (unsigned) vinum_conf.plexes_used) /* plex doesn't exist */ + ||(PLEX[index].state == plex_unallocated)) + return ENODEV; + if (fe >= PLEX[index].defective_regions) /* no such entry */ + return ENOENT; + bcopy(&PLEX[index].defective_region[fe], + data, + sizeof(struct plexregion)); + return 0; + + case VINUM_GETUNMAPPED: /* get a plex unmapped area element */ + index = *(int *) data; /* get the plex index */ + fe = ((int *) data)[1]; /* and the region number */ + if ((index >= (unsigned) vinum_conf.plexes_used) /* plex doesn't exist */ + ||(PLEX[index].state == plex_unallocated)) + return ENODEV; + if (fe >= PLEX[index].unmapped_regions) /* no such entry */ + return ENOENT; + bcopy(&PLEX[index].unmapped_region[fe], + data, + sizeof(struct plexregion)); + return 0; + + case VINUM_RESETSTATS: + resetstats((struct vinum_ioctl_msg *) data); /* reset object stats */ + return 0; + + /* attach an object to a superordinate object */ + case VINUM_ATTACH: + attachobject((struct vinum_ioctl_msg *) data); + return 0; + + /* detach an object from a superordinate object */ + case VINUM_DETACH: + detachobject((struct vinum_ioctl_msg *) data); + return 0; + + /* rename an object */ + case VINUM_RENAME: + renameobject((struct vinum_rename_msg *) data); + return 0; + + /* replace an object */ + case VINUM_REPLACE: + replaceobject((struct vinum_ioctl_msg *) data); + return 0; + + default: + /* FALLTHROUGH */ + } + + default: +#if __FreeBSD__>=3 + printf("vinumioctl: type %d, sd %d, plex %d, major %x, volume %d, command %lx\n", + device->type, + device->sd, + device->plex, + device->major, + device->volume, + cmd); /* XXX */ + +#else + printf("vinumioctl: type %d, sd %d, plex %d, major %x, volume %d, command %x\n", + device->type, + device->sd, + device->plex, + device->major, + device->volume, + cmd); /* XXX */ + +#endif + return EINVAL; + + case VINUM_DRIVE_TYPE: + case VINUM_PLEX_TYPE: + return EAGAIN; /* try again next week */ + + case VINUM_SD_TYPE: + objno = SDNO(dev); + + switch (cmd) { + case VINUM_INITSD: /* initialize subdisk */ + return initsd(objno); + + default: + return EINVAL; + } + break; + + case VINUM_VOLUME_TYPE: + objno = VOLNO(dev); + + if ((unsigned) objno >= (unsigned) vinum_conf.volumes_used) /* not a valid volume */ + return ENXIO; + vol = &VOL[objno]; + if (vol->state != volume_up) /* not up, */ + return EIO; /* I/O error */ + + switch (cmd) { + case DIOCGDINFO: /* get disk label */ + get_volume_label(vol, (struct disklabel *) data); + break; + + /* Care! DIOCGPART returns *pointers* to + * the caller, so we need to store this crap as well. + * And yes, we need it. */ + case DIOCGPART: /* get partition information */ + get_volume_label(vol, &vol->label); + ((struct partinfo *) data)->disklab = &vol->label; + ((struct partinfo *) data)->part = &vol->label.d_partitions[0]; + break; + + /* We don't have this stuff on hardware, + * so just pretend to do it so that + * utilities don't get upset. */ + case DIOCWDINFO: /* write partition info */ + case DIOCSDINFO: /* set partition info */ + return 0; /* not a titty */ + + case DIOCWLABEL: /* set or reset label writeable */ + if ((flag & FWRITE) == 0) /* not writeable? */ + return EACCES; /* no, die */ + if (*(int *) data != 0) /* set it? */ + vol->flags |= VF_WLABEL; /* yes */ + else + vol->flags &= ~VF_WLABEL; /* no, reset */ + break; + + default: + return ENOTTY; /* not my kind of ioctl */ + } + break; + } + return 0; /* XXX */ +} + +/* The following four functions check the supplied + * object index and return a pointer to the object + * if it exists. Otherwise they longjump out via + * throw_rude_remark */ +struct drive * +validdrive(int driveno, struct _ioctl_reply *reply) +{ + if ((driveno < vinum_conf.drives_used) + && (DRIVE[driveno].state != drive_unallocated)) + return &DRIVE[driveno]; + strcpy(reply->msg, "No such drive"); + reply->error = ENOENT; + return NULL; +} + +struct sd * +validsd(int sdno, struct _ioctl_reply *reply) +{ + if ((sdno < vinum_conf.subdisks_used) + && (SD[sdno].state != sd_unallocated)) + return &SD[sdno]; + strcpy(reply->msg, "No such subdisk"); + reply->error = ENOENT; + return NULL; +} + +struct plex * +validplex(int plexno, struct _ioctl_reply *reply) +{ + if ((plexno < vinum_conf.plexes_used) + && (PLEX[plexno].state != plex_unallocated)) + return &PLEX[plexno]; + strcpy(reply->msg, "No such plex"); + reply->error = ENOENT; + return NULL; +} + +struct volume * +validvol(int volno, struct _ioctl_reply *reply) +{ + if ((volno < vinum_conf.volumes_used) + && (VOL[volno].state != volume_unallocated)) + return &VOL[volno]; + strcpy(reply->msg, "No such volume"); + reply->error = ENOENT; + return NULL; +} + +/* reset an object's stats */ +void +resetstats(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + + switch (msg->type) { + case drive_object: + if (msg->index < vinum_conf.drives_used) { + struct drive *drive = &DRIVE[msg->index]; + if (drive->state != drive_unallocated) { + drive->reads = 0; /* number of reads on this drive */ + drive->writes = 0; /* number of writes on this drive */ + drive->bytes_read = 0; /* number of bytes read */ + drive->bytes_written = 0; /* number of bytes written */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + case sd_object: + if (msg->index < vinum_conf.subdisks_used) { + struct sd *sd = &SD[msg->index]; + if (sd->state != sd_unallocated) { + sd->reads = 0; /* number of reads on this subdisk */ + sd->writes = 0; /* number of writes on this subdisk */ + sd->bytes_read = 0; /* number of bytes read */ + sd->bytes_written = 0; /* number of bytes written */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + break; + + case plex_object: + if (msg->index < vinum_conf.plexes_used) { + struct plex *plex = &PLEX[msg->index]; + if (plex->state != plex_unallocated) { + plex->reads = 0; + plex->writes = 0; /* number of writes on this plex */ + plex->bytes_read = 0; /* number of bytes read */ + plex->bytes_written = 0; /* number of bytes written */ + plex->multiblock = 0; /* requests that needed more than one block */ + plex->multistripe = 0; /* requests that needed more than one stripe */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + break; + + case volume_object: + if (msg->index < vinum_conf.volumes_used) { + struct volume *vol = &VOL[msg->index]; + if (vol->state != volume_unallocated) { + vol->bytes_read = 0; /* number of bytes read */ + vol->bytes_written = 0; /* number of bytes written */ + vol->reads = 0; /* number of reads on this volume */ + vol->writes = 0; /* number of writes on this volume */ + vol->recovered_reads = 0; /* reads recovered from another plex */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + case invalid_object: /* can't get this */ + reply->error = EINVAL; + return; + } +} + +/* attach an object to a superior object */ +void +attachobject(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + struct sd *sd; + struct plex *plex; + struct volume *vol; + + switch (msg->type) { + case drive_object: /* you can't attach a drive to anything */ + case volume_object: /* nor a volume */ + case invalid_object: /* "this can't happen" */ + reply->error = EINVAL; + reply->msg[0] = '\0'; /* vinum(8) doesn't do this */ + return; + + case sd_object: + sd = validsd(msg->index, reply); + if (sd == NULL) /* not a valid subdisk */ + return; + plex = validplex(msg->otherobject, reply); + if (plex) { + if (sd->plexno >= 0) { /* already belong to a plex */ + reply->error = EBUSY; /* no message, the user should check */ + reply->msg[0] = '\0'; + return; + } + sd->plexoffset = msg->offset; /* this is where we want it */ + set_sd_state(sd->sdno, sd_stale, setstate_force); /* make sure it's stale */ + give_sd_to_plex(plex->plexno, sd->sdno); /* and give it to the plex */ + update_sd_config(sd->sdno, 0); + save_config(); + reply->error = 0; + } + break; + + case plex_object: + plex = validplex(msg->index, reply); /* get plex */ + if (plex == NULL) + return; + if (plex->organization != plex_concat) { /* can't attach to striped and raid-5 */ + reply->error = EINVAL; /* no message, the user should check */ + reply->msg[0] = '\0'; + return; + } + vol = validvol(msg->otherobject, reply); /* and volume information */ + if (vol) { + if ((vol->plexes == MAXPLEX) /* we have too many already */ + ||(plex->volno >= 0)) { /* or the plex has an owner */ + reply->error = EINVAL; /* no message, the user should check */ + reply->msg[0] = '\0'; + return; + } + set_plex_state(plex->plexno, plex_down, setstate_force); /* make sure it's down */ + give_plex_to_volume(msg->otherobject, msg->index); /* and give it to the volume */ + update_plex_config(plex->plexno, 0); + save_config(); + if (plex->state == plex_reviving) + reply->error = EAGAIN; /* need to revive it */ + else + reply->error = 0; + } + } +} + +/* detach an object from a superior object */ +void +detachobject(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + struct sd *sd; + struct plex *plex; + struct volume *vol; + int sdno; + int plexno; + + switch (msg->type) { + case drive_object: /* you can't attach a drive to anything */ + case volume_object: /* nor a volume */ + case invalid_object: /* "this can't happen" */ + reply->error = EINVAL; + reply->msg[0] = '\0'; /* vinum(8) doesn't do this */ + return; + + case sd_object: + sd = validsd(msg->index, reply); + if (sd == NULL) + return; + if (sd->plexno < 0) { /* doesn't belong to a plex */ + reply->error = ENOENT; + strcpy(reply->msg, "Subdisk is not attached"); + return; + } else { /* valid plex number */ + plex = &PLEX[sd->plexno]; + if ((!msg->force) /* don't force things */ + &&((plex->state == plex_up) /* and the plex is up */ + ||((plex->state == plex_flaky) && sd->state == sd_up))) { /* or flaky with this sd up */ + reply->error = EBUSY; /* we need this sd */ + reply->msg[0] = '\0'; + return; + } + sd->plexno = -1; /* anonymous sd */ + if (plex->subdisks == 1) { /* this was the only subdisk */ + Free(plex->sdnos); /* free the subdisk array */ + plex->sdnos = NULL; /* and note the fact */ + plex->subdisks_allocated = 0; /* no subdisk space */ + } else { + for (sdno = 0; sdno < plex->subdisks; sdno++) { + if (plex->sdnos[sdno] == msg->index) /* found our subdisk */ + break; + } + if (sdno < (plex->subdisks - 1)) /* not the last one, compact */ + bcopy(&plex->sdnos[sdno + 1], + &plex->sdnos[sdno], + (plex->subdisks - 1 - sdno) * sizeof(int)); + } + plex->subdisks--; + rebuild_plex_unmappedlist(plex); /* rebuild the unmapped list */ + if (!bcmp(plex->name, sd->name, strlen(plex->name))) { /* this subdisk is named after the plex */ + bcopy(sd->name, + &sd->name[3], + min(strlen(sd->name), MAXSDNAME - 3)); + bcopy("ex-", sd->name, 3); + sd->name[MAXSDNAME - 1] = '\0'; + } + update_plex_config(plex->plexno, 0); + if ((plex->organization == plex_striped) /* we've just mutilated our plex, */ + ||(plex->organization == plex_striped)) /* the data no longer matches */ + set_plex_state(plex->plexno, + plex_down, + setstate_force | setstate_configuring); + update_sd_config(sd->sdno, 0); + save_config(); + reply->error = 0; + } + return; + + case plex_object: + plex = validplex(msg->index, reply); /* get plex */ + if (plex == NULL) + return; + if (plex->volno >= 0) { + int volno = plex->volno; + + vol = &VOL[volno]; + if ((!msg->force) /* don't force things */ + &&((vol->state == volume_up) /* and the volume is up */ + &&(vol->plexes == 1))) { /* and this is the last plex */ + /* XXX As elsewhere, check whether we will lose + * mapping by removing this plex */ + reply->error = EBUSY; /* we need this plex */ + reply->msg[0] = '\0'; + return; + } + plex->volno = -1; /* anonymous plex */ + for (plexno = 0; plexno < vol->plexes; plexno++) { + if (vol->plex[plexno] == msg->index) /* found our plex */ + break; + } + if (plexno < (vol->plexes - 1)) /* not the last one, compact */ + bcopy(&vol[plexno + 1], &vol[plexno], (vol->plexes - 1 - plexno) * sizeof(int)); + vol->plexes--; + if (!bcmp(vol->name, plex->name, strlen(vol->name))) { /* this plex is named after the volume */ + /* First, check if the subdisks are the same */ + if (msg->recurse) { + int sdno; + + for (sdno = 0; sdno < plex->subdisks; sdno++) { + struct sd *sd = &SD[plex->sdnos[sdno]]; + + if (!bcmp(plex->name, sd->name, strlen(plex->name))) { /* subdisk is named after the plex */ + bcopy(sd->name, &sd->name[3], min(strlen(sd->name), MAXSDNAME - 3)); + bcopy("ex-", sd->name, 3); + sd->name[MAXSDNAME - 1] = '\0'; + } + } + } + bcopy(plex->name, &plex->name[3], min(strlen(plex->name), MAXPLEXNAME - 3)); + bcopy("ex-", plex->name, 3); + plex->name[MAXPLEXNAME - 1] = '\0'; + } + update_plex_config(plex->plexno, 0); + update_volume_config(volno, 0); + save_config(); + reply->error = 0; + } else { + reply->error = ENOENT; + strcpy(reply->msg, "Plex is not attached"); + } + } +} + +void +renameobject(struct vinum_rename_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + struct drive *drive; + struct sd *sd; + struct plex *plex; + struct volume *vol; + + switch (msg->type) { + case drive_object: /* you can't attach a drive to anything */ + if (find_drive(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + drive = validdrive(msg->index, reply); + if (drive) { + bcopy(msg->newname, drive->label.name, MAXDRIVENAME); + save_config(); + reply->error = 0; + } + return; + + case sd_object: /* you can't attach a subdisk to anything */ + if (find_subdisk(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + sd = validsd(msg->index, reply); + if (sd) { + bcopy(msg->newname, sd->name, MAXSDNAME); + update_sd_config(sd->sdno, 0); + save_config(); + reply->error = 0; + } + return; + + case plex_object: /* you can't attach a plex to anything */ + if (find_plex(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + plex = validplex(msg->index, reply); + if (plex) { + bcopy(msg->newname, plex->name, MAXPLEXNAME); + update_plex_config(plex->plexno, 0); + save_config(); + reply->error = 0; + } + return; + + case volume_object: /* you can't attach a volume to anything */ + if (find_volume(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + vol = validvol(msg->index, reply); + if (vol) { + bcopy(msg->newname, vol->name, MAXVOLNAME); + update_volume_config(msg->index, 0); + save_config(); + reply->error = 0; + } + return; + + case invalid_object: + reply->error = EINVAL; + reply->msg[0] = '\0'; + } +} + +/* Replace one object with another */ +void +replaceobject(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + + reply->error = ENODEV; /* until I know how to do this */ + strcpy(reply->msg, "replace not implemented yet"); +/* save_config (); */ +} diff --git a/sys/dev/vinum/vinumkw.h b/sys/dev/vinum/vinumkw.h new file mode 100644 index 0000000..1a81f37 --- /dev/null +++ b/sys/dev/vinum/vinumkw.h @@ -0,0 +1,120 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumkw.h,v 1.7 1998/08/07 02:35:51 grog Exp grog $ + */ + +/* Command keywords that vinum knows. These include both user-level + * and kernel-level stuff */ + +/* Our complete vocabulary. The names of the commands are + * the same as the identifier without the kw_ at the beginning + * (i.e. kw_create defines the "create" keyword). Preprocessor + * magic in parser.c does the rest. */ +enum keyword { + kw_create, + kw_modify, + kw_list, + kw_l = kw_list, + kw_ld, /* list drive */ + kw_ls, /* list subdisk */ + kw_lp, /* list plex */ + kw_lv, /* list volume */ + kw_set, + kw_rm, + kw_start, + kw_stop, + kw_drive, + kw_sd, + kw_subdisk = kw_sd, + kw_plex, + kw_volume, + kw_vol = kw_volume, + kw_read, + kw_readpol, + kw_org, + kw_name, + kw_concat, + kw_striped, + kw_raid5, + kw_driveoffset, + kw_plexoffset, + kw_len, + kw_length = kw_len, + kw_state, + kw_setupstate, + kw_d, /* flag names */ + kw_f, + kw_r, + kw_s, + kw_v, + kw_round, /* round robin */ + kw_prefer, /* prefer plex */ + kw_device, + kw_init, + kw_label, + kw_resetconfig, + kw_writethrough, + kw_writeback, + kw_raw, + kw_resetstats, + kw_attach, + kw_detach, + kw_rename, + kw_printconfig, + kw_replace, + kw_detached, +#ifdef DEBUG + kw_debug, /* go into debugger */ + kw_info, +#endif + kw_invalid_keyword = -1 +}; + +struct _keywords { + char *name; + enum keyword keyword; +}; + +struct keywordset { + int size; + struct _keywords *k; +}; + +extern struct _keywords keywords[]; +extern struct _keywords flag_keywords[]; + +extern struct keywordset keyword_set; +extern struct keywordset flag_set; diff --git a/sys/dev/vinum/vinumlock.c b/sys/dev/vinum/vinumlock.c new file mode 100644 index 0000000..ed69649 --- /dev/null +++ b/sys/dev/vinum/vinumlock.c @@ -0,0 +1,137 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: lock.c,v 1.6 1998/07/28 06:32:57 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" + +/* Lock routines. Currently, we lock either an individual volume + * or the global configuration. I don't think tsleep and + * wakeup are SMP safe. FIXME XXX */ + +/* Lock a volume, wait if it's in use */ +int +lockvol(struct volume *vol) +{ + int error; + + while ((vol->flags & VF_LOCKED) != 0) { + vol->flags |= VF_LOCKING; + /* It would seem to make more sense to sleep on + * the address 'vol'. Unfortuntaly we can't + * guarantee that this address won't change due to + * table expansion. The address we choose won't change. */ + if ((error = tsleep(&vinum_conf.volume + vol->devno, + PRIBIO | PCATCH, + "volock", + 0)) != 0) + return error; + } + vol->flags |= VF_LOCKED; + return 0; +} + +/* Unlock a volume and let the next one at it */ +void +unlockvol(struct volume *vol) +{ + vol->flags &= ~VF_LOCKED; + if ((vol->flags & VF_LOCKING) != 0) { + vol->flags &= ~VF_LOCKING; + wakeup(&vinum_conf.volume + vol->devno); + } +} + +/* Lock a plex, wait if it's in use */ +int +lockplex(struct plex *plex) +{ + int error; + + while ((plex->flags & VF_LOCKED) != 0) { + plex->flags |= VF_LOCKING; + /* It would seem to make more sense to sleep on + * the address 'plex'. Unfortuntaly we can't + * guarantee that this address won't change due to + * table expansion. The address we choose won't change. */ + if ((error = tsleep(&vinum_conf.plex + plex->sdnos[0], + PRIBIO | PCATCH, + "plexlk", + 0)) != 0) + return error; + } + plex->flags |= VF_LOCKED; + return 0; +} + +/* Unlock a plex and let the next one at it */ +void +unlockplex(struct plex *plex) +{ + plex->flags &= ~VF_LOCKED; + if ((plex->flags & VF_LOCKING) != 0) { + plex->flags &= ~VF_LOCKING; + wakeup(&vinum_conf.plex + plex->plexno); + } +} + + +/* Get a lock for the global config, wait if it's not available */ +int +lock_config(void) +{ + int error; + + while ((vinum_conf.flags & VF_LOCKED) != 0) { + vinum_conf.flags |= VF_LOCKING; + if ((error = tsleep(&vinum_conf, PRIBIO | PCATCH, "vincfg", 0)) != 0) + return error; + } + vinum_conf.flags |= VF_LOCKED; + return 0; +} + +/* Unlock and wake up any waiters */ +void +unlock_config(void) +{ + vinum_conf.flags &= ~VF_LOCKED; + if ((vinum_conf.flags & VF_LOCKING) != 0) { + vinum_conf.flags &= ~VF_LOCKING; + wakeup(&vinum_conf); + } +} diff --git a/sys/dev/vinum/vinummemory.c b/sys/dev/vinum/vinummemory.c new file mode 100644 index 0000000..5dee671 --- /dev/null +++ b/sys/dev/vinum/vinummemory.c @@ -0,0 +1,186 @@ + +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: memory.c,v 1.16 1998/08/08 04:43:22 grog Exp grog $ + */ + +#define REALLYKERNEL +#define USES_VM +#include "vinumhdr.h" + +extern jmp_buf command_fail; /* return on a failed command */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +void freedatabuf(struct mc *me); +caddr_t allocdatabuf(struct mc *me); + +void +expand_table(void **table, int oldsize, int newsize) +{ + if (newsize > oldsize) { + int *temp; + + temp = (int *) Malloc(newsize); /* allocate a new table */ + CHECKALLOC(temp, "vinum: Can't expand table\n"); + if (*table != NULL) { /* already something there, */ + bcopy((char *) *table, (char *) temp, oldsize); /* copy it to the old table */ + Free(*table); + } + *table = temp; + } +} + +#ifndef DEBUG +/* increase the size of a request block */ +void +expandrq(struct plexrq *prq) +{ + expand_table((void **) &prq->rqe, + prq->requests * sizeof(struct rqelement), + (prq->requests + RQELTS) * sizeof(struct rqelement)); + bzero(&prq->rqe[prq->requests], RQELTS * sizeof(struct rqelement)); /* clear the new part */ + prq->rqcount += RQELTS; +} + +#endif + +#if DEBUG /* XXX debug */ +#define MALLOCENTRIES 16384 +int malloccount = 0; +int highwater = 0; /* highest index ever allocated */ +static struct mc malloced[MALLOCENTRIES]; + +static total_malloced; + +caddr_t +MMalloc(int size, char *file, int line) +{ + caddr_t result; + int i; + static int seq = 0; + int s; + struct mc me; /* information to pass to allocdatabuf */ + + if (malloccount >= MALLOCENTRIES) { /* too many */ + printf("vinum: can't allocate table space to trace memory allocation"); + return 0; /* can't continue */ + } + result = malloc(size, M_DEVBUF, M_WAITOK); /* use malloc for smaller and irregular stuff */ + if (result == NULL) + printf("vinum: can't allocate %d bytes from %s:%d\n", size, file, line); + else { + me.flags = 0; /* allocation via malloc */ + s = splhigh(); + for (i = 0; i < malloccount; i++) { + if (((result + size) > malloced[i].address) + && (result < malloced[i].address + malloced[i].size)) /* overlap */ + Debugger("Malloc overlap"); + } + if (result) { + i = malloccount++; + total_malloced += size; + malloced[i].address = result; + malloced[i].size = size; + malloced[i].line = line; + malloced[i].seq = seq++; + malloced[i].flags = me.flags; + malloced[i].databuf = me.databuf; /* only used with kva alloc */ + bcopy(file, malloced[i].file, min(strlen(file) + 1, 16)); + } + if (malloccount > highwater) + highwater = malloccount; + splx(s); + } + return result; +} + +void +FFree(void *mem, char *file, int line) +{ + int i; + int s; + + s = splhigh(); + for (i = 0; i < malloccount; i++) { + if ((caddr_t) mem == malloced[i].address) { /* found it */ + bzero(mem, malloced[i].size); /* XXX */ + free(mem, M_DEVBUF); + malloccount--; + total_malloced -= malloced[i].size; + if (i < malloccount) /* more coming after */ + bcopy(&malloced[i + 1], &malloced[i], (malloccount - i) * sizeof(struct mc)); + splx(s); + return; + } + } + splx(s); + printf("Freeing unallocated data at 0x%08x from %s, line %d\n", (int) mem, file, line); + Debugger("Free"); +} + +void +vinum_meminfo(caddr_t data) +{ + struct meminfo *m = (struct meminfo *) data; + + m->mallocs = malloccount; + m->total_malloced = total_malloced; + m->malloced = malloced; + m->highwater = highwater; +} + +int +vinum_mallocinfo(caddr_t data) +{ + struct mc *m = (struct mc *) data; + unsigned int ent = *(int *) data; /* 1st word is index */ + + if (ent >= malloccount) + return ENOENT; + m->address = malloced[ent].address; + m->size = malloced[ent].size; + m->line = malloced[ent].line; + m->seq = malloced[ent].seq; + bcopy(malloced[ent].file, m->file, 16); + return 0; +} + +#endif diff --git a/sys/dev/vinum/vinumparser.c b/sys/dev/vinum/vinumparser.c new file mode 100644 index 0000000..9e7f791 --- /dev/null +++ b/sys/dev/vinum/vinumparser.c @@ -0,0 +1,206 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: parser.c,v 1.11 1998/08/10 08:50:42 grog Exp grog $ + */ + +/* This file contains the parser for the configuration routines. It's used + * both in the kernel and in the user interface program, thus the separate file. */ + +/* Go through a text and split up into text tokens. These are either non-blank + * sequences, or any sequence (except \0) enclosed in ' or ". Embedded ' or + * " characters may be escaped by \, which otherwise has no special meaning. + * + * Delimit by following with a \0, and return pointers to the starts at token []. + * Return the number of tokens found as the return value. + * + * This method has the restriction that a closing " or ' must be followed by + * grey space. + * + * Error conditions are end of line before end of quote, or no space after + * a closing quote. In this case, tokenize() returns -1. */ + +#include <sys/param.h> +#ifdef KERNEL +#undef KERNEL /* XXX */ +#define REALLYKERNEL +#else +#include <ctype.h> +#include <errno.h> +#include <fcntl.h> +#endif +/* All this mess for a single struct definition */ +#include <sys/uio.h> +#include <sys/namei.h> +#include <sys/disklabel.h> +#include <sys/mount.h> +#include <sys/device.h> +#include <sys/disk.h> +#include "sys/buf.h" + +#include <vinumvar.h> +#include "vinumkw.h" +#include "vinumio.h" +#include "vinumext.h" + +#ifdef REALLYKERNEL +#define isspace(c) ((c == ' ') || (c == '\t')) /* check for white space */ +#else /* get it from the headers */ +#include <ctype.h> +#endif + +/* enum keyword is defined in vinumvar.h */ + +#define keypair(x) { #x, kw_##x } /* create pair "foo", kw_foo */ +#define flagkeypair(x) { "-"#x, kw_##x } /* create pair "-foo", kw_foo */ +#define KEYWORDSET(x) {sizeof (x) / sizeof (struct _keywords), x} + +/* Normal keywords. These are all the words that vinum knows. */ +struct _keywords keywords[] = +{keypair(drive), + keypair(sd), + keypair(subdisk), + keypair(plex), + keypair(volume), + keypair(vol), + keypair(setupstate), + keypair(readpol), + keypair(org), + keypair(name), + keypair(writethrough), + keypair(writeback), + keypair(raw), + keypair(device), + keypair(concat), + keypair(raid5), + keypair(striped), + keypair(plexoffset), + keypair(driveoffset), + keypair(length), + keypair(len), + keypair(state), + keypair(round), + keypair(prefer), + keypair(rename), + keypair(detached), +#ifndef KERNEL /* for vinum(8) only */ +#ifdef DEBUG + keypair(debug), +#endif + keypair(attach), + keypair(detach), + keypair(printconfig), + keypair(replace), + keypair(create), + keypair(read), + keypair(modify), + keypair(list), + keypair(l), + keypair(ld), + keypair(ls), + keypair(lp), + keypair(lv), + keypair(info), + keypair(set), + keypair(rm), + keypair(init), + keypair(label), + keypair(resetconfig), + keypair(start), + keypair(stop), + keypair(resetstats) +#endif +}; +struct keywordset keyword_set = KEYWORDSET(keywords); + +#ifndef KERNEL +struct _keywords flag_keywords[] = +{flagkeypair(f), + flagkeypair(d), + flagkeypair(v), + flagkeypair(s), + flagkeypair(r) +}; +struct keywordset flag_set = KEYWORDSET(flag_keywords); + +#endif + +int +tokenize(char *cptr, char *token[]) +{ + char delim; /* delimiter for searching for the partner */ + int tokennr; /* index of this token */ + tokennr = 0; /* none found yet */ + + for (;;) { + while (isspace(*cptr)) + cptr++; /* skip initial white space */ + if ((*cptr == '\0') || (*cptr == '\n') || (*cptr == '#')) /* end of line */ + return tokennr; /* return number of tokens found */ + delim = *cptr; + token[tokennr] = cptr; /* point to it */ + tokennr++; /* one more */ + /* XXX this is broken. It leaves superfluous \\ characters in the text */ + if ((delim == '\'') || (delim == '"')) { /* delimitered */ + for (;;) { + cptr++; + if ((*cptr == delim) && (cptr[-1] != '\\')) { /* found the partner */ + cptr++; /* move on past */ + if (!isspace(*cptr)) /* error, no space after closing quote */ + return -1; + *cptr++ = '\0'; /* delimit */ + } else if ((*cptr == '\0') || (*cptr == '\n')) /* end of line */ + return -1; + } + } else { /* not quoted */ + while ((*cptr != '\0') && (!isspace(*cptr)) && (*cptr != '\n')) + cptr++; + if (*cptr != '\0') /* not end of the line, */ + *cptr++ = '\0'; /* delimit and move to the next */ + } + } +} + +/* Find a keyword and return an index */ +enum keyword +get_keyword(char *name, struct keywordset *keywordset) +{ + int i; + struct _keywords *keywords = keywordset->k; /* point to the keywords */ + for (i = 0; i < keywordset->size; i++) + if (!strcmp(name, keywords[i].name)) + return (enum keyword) keywords[i].keyword; + return kw_invalid_keyword; +} diff --git a/sys/dev/vinum/vinumrequest.c b/sys/dev/vinum/vinumrequest.c new file mode 100644 index 0000000..589eb3f --- /dev/null +++ b/sys/dev/vinum/vinumrequest.c @@ -0,0 +1,882 @@ +/* XXX to do: + + * Decide where we need splbio () + */ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: request.c,v 1.17 1998/08/13 06:04:47 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" +#include <miscfs/specfs/specdev.h> +#include <sys/resourcevar.h> + +/* pointer to ioctl p parameter, to save passing it around */ +extern struct proc *myproc; + +enum requeststatus bre(struct request *rq, + int plexno, + daddr_t * diskstart, + daddr_t diskend); +enum requeststatus bre5(struct request *rq, + int plexno, + daddr_t * diskstart, + daddr_t diskend); +enum requeststatus build_read_request(struct request *rq, int volplexno); +enum requeststatus build_write_request(struct request *rq); +enum requeststatus build_rq_buffer(struct rqelement *rqe, struct plex *plex); +void freerq(struct request *rq); +void free_rqg(struct rqgroup *rqg); +int find_alternate_sd(struct request *rq); +int check_range_covered(struct request *); +void complete_rqe(struct buf *bp); +void complete_raid5_write(struct rqelement *); +int abortrequest(struct request *rq, int error); +void sdio(struct buf *bp); +void sdio_done(struct buf *bp); +int vinum_bounds_check(struct buf *bp, struct volume *vol); +caddr_t allocdatabuf(struct rqelement *rqe); +void freedatabuf(struct rqelement *rqe); + +void +vinumstrategy(struct buf *bp) +{ + BROKEN_GDB; + int volno; + struct volume *vol = NULL; + int s; + struct devcode *device = (struct devcode *) &bp->b_dev; /* decode device number */ + enum requeststatus status; + + switch (device->type) { + case VINUM_SD_TYPE: + sdio(bp); + return; + + /* In fact, vinum doesn't handle drives: they're + * handled directly by the disk drivers */ + case VINUM_DRIVE_TYPE: + default: + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + biodone(bp); + return; + + case VINUM_VOLUME_TYPE: /* volume I/O */ + volno = VOLNO(bp->b_dev); + vol = &VOL[volno]; + if (vol->state != volume_up) { /* can't access this volume */ + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + biodone(bp); + return; + } + if (vinum_bounds_check(bp, vol) <= 0) { /* don't like them bounds */ + biodone(bp); /* have nothing to do with this */ + return; + } + /* FALLTHROUGH */ + /* Plex I/O is pretty much the same as volume I/O + * for a single plex. Indicate this by passing a NULL + * pointer (set above) for the volume */ + case VINUM_PLEX_TYPE: + bp->b_resid = bp->b_bcount; /* transfer everything */ + vinumstart(bp, 0); + return; + } +} + +/* Start a transfer. Return -1 on error, + * 0 if OK, 1 if we need to retry. + * Parameter reviveok is set when doing + * transfers for revives: it allows transfers to + * be started immediately when a revive is in + * progress. During revive, normal transfers + * are queued if they share address space with + * a currently active revive operation. */ +int +vinumstart(struct buf *bp, int reviveok) +{ + BROKEN_GDB; + int plexno; + int maxplex; /* maximum number of plexes to handle */ + struct volume *vol; + struct rqgroup *rqg; /* current plex's requests */ + struct rqelement *rqe; /* individual element */ + struct request *rq; /* build up our request here */ + int rqno; /* index in request list */ + enum requeststatus status; + + /* XXX In these routines, we're assuming that + * we will always be called with bp->b_bcount + * which is a multiple of the sector size. This + * is a reasonable assumption, since we are only + * called from system routines. Should we check + * anyway? */ + + if ((bp->b_bcount % DEV_BSIZE) != 0) { /* bad length */ + bp->b_error = EINVAL; /* invalid size */ + bp->b_flags |= B_ERROR; + biodone(bp); + return -1; + } + rq = (struct request *) Malloc(sizeof(struct request)); /* allocate a request struct */ + if (rq == NULL) { /* can't do it */ + bp->b_error = ENOMEM; /* can't get memory */ + bp->b_flags |= B_ERROR; + biodone(bp); + return -1; + } + bzero(rq, sizeof(struct request)); + + /* Note the volume ID. This can be NULL, which + * the request building functions use as an + * indication for single plex I/O */ + rq->bp = bp; /* and the user buffer struct */ + + if (DEVTYPE(bp->b_dev) == VINUM_VOLUME_TYPE) { /* it's a volume, */ + rq->volplex.volno = VOLNO(bp->b_dev); /* get the volume number */ + vol = &VOL[rq->volplex.volno]; /* and point to it */ + vol->active++; /* one more active request */ + maxplex = vol->plexes; /* consider all its plexes */ + } else { + vol = NULL; /* no volume */ + rq->volplex.plexno = PLEXNO(bp->b_dev); /* point to the plex */ + rq->isplex = 1; /* note that it's a plex */ + maxplex = 1; /* just the one plex */ + } + + if (bp->b_flags & B_READ) { + /* This is a read request. Decide + * which plex to read from. + * + * There's a potential race condition here, + * since we're not locked, and we could end + * up multiply incrementing the round-robin + * counter. This doesn't have any serious + * effects, however. */ + if (vol != NULL) { + vol->reads++; + vol->bytes_read += bp->b_bcount; + plexno = vol->preferred_plex; /* get the plex to use */ + if (plexno < 0) { /* round robin */ + plexno = vol->last_plex_read; + vol->last_plex_read++; + if (vol->last_plex_read == vol->plexes) /* got the the end? */ + vol->last_plex_read = 0; /* wrap around */ + } + status = build_read_request(rq, plexno); /* build a request */ + } else { + daddr_t diskaddr = bp->b_blkno; /* start offset of transfer */ + status = bre(rq, /* build a request list */ + rq->volplex.plexno, + &diskaddr, + diskaddr + (bp->b_bcount / DEV_BSIZE)); + } + + if ((status > REQUEST_RECOVERED) /* can't satisfy it */ + ||(bp->b_flags & B_DONE)) { /* XXX shouldn't get this without bad status */ + if (status == REQUEST_DOWN) { /* not enough subdisks */ + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + } + biodone(bp); + freerq(rq); + return -1; + } + return launch_requests(rq, reviveok); /* now start the requests if we can */ + } else + /* This is a write operation. We write to all + * plexes. If this is a RAID 5 plex, we must also + * update the parity stripe. */ + { + if (vol != NULL) { + vol->writes++; + vol->bytes_written += bp->b_bcount; + status = build_write_request(rq); /* Not all the subdisks are up */ + } else { /* plex I/O */ + daddr_t diskstart; + + diskstart = bp->b_blkno; /* start offset of transfer */ + status = bre(rq, + PLEXNO(bp->b_dev), + &diskstart, + bp->b_blkno + (bp->b_bcount / DEV_BSIZE)); /* build requests for the plex */ + } + if ((status > REQUEST_RECOVERED) /* can't satisfy it */ + ||(bp->b_flags & B_DONE)) { /* XXX shouldn't get this without bad status */ + if (status == REQUEST_DOWN) { /* not enough subdisks */ + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + } + if ((bp->b_flags & B_DONE) == 0) + biodone(bp); + freerq(rq); + return -1; + } + return launch_requests(rq, reviveok); /* start the requests */ + } +} + +/* Call the low-level strategy routines to + * perform the requests in a struct request */ +int +launch_requests(struct request *rq, int reviveok) +{ + struct rqgroup *rqg; + int rqno; /* loop index */ + struct rqelement *rqe; /* current element */ + int s; + + /* First find out whether we're reviving, and the + * request contains a conflict. If so, we hang + * the request off plex->waitlist of the first + * plex we find which is reviving */ + if ((rq->flags & XFR_REVIVECONFLICT) /* possible revive conflict */ + &&(!reviveok)) { /* and we don't want to do it now, */ + struct volume *vol = &VOL[VOLNO(rq->bp->b_dev)]; + struct plex *plex; + int plexno; + + for (plexno = 0; plexno < vol->plexes; plexno++) { /* find the reviving plex */ + plex = &PLEX[vol->plex[plexno]]; + if (plex->state == plex_reviving) /* found it */ + break; + } + if (plexno < vol->plexes) { /* found it? */ + struct request *waitlist = plex->waitlist; /* point to the waiting list */ + + while (waitlist->next != NULL) /* find the end */ + waitlist = waitlist->next; + waitlist->next = rq; /* hook our request there */ + return 0; /* and get out of here */ + } else /* bad vinum, bad */ + printf("vinum: can't find reviving plex for volume %s\n", vol->name); + } + rq->active = 0; /* nothing yet */ + /* XXX This is probably due to a bug */ + if (rq->rqg == NULL) { /* no request */ + abortrequest(rq, EINVAL); + return -1; + } +#if DEBUG + if (debug & DEBUG_ADDRESSES) + printf("Request: %x\nWrite dev 0x%x, offset 0x%x, length %ld\n", + (u_int) rq, + rq->bp->b_dev, + rq->bp->b_blkno, + rq->bp->b_bcount); /* XXX */ + vinum_conf.lastrq = (int) rq; + vinum_conf.lastbuf = rq->bp; +#endif + for (rqg = rq->rqg; rqg != NULL; rqg = rqg->next) { /* through the whole request chain */ + rqg->active = rqg->count; /* they're all active */ + rq->active++; /* one more active request group */ + for (rqno = 0; rqno < rqg->count; rqno++) { + rqe = &rqg->rqe[rqno]; + if (rqe->flags & XFR_BAD_SUBDISK) /* this subdisk is bad, */ + rqg->active--; /* one less active request */ + else { + struct drive *drive = &DRIVE[rqe->driveno]; /* drive to access */ + if ((rqe->b.b_flags & B_READ) == 0) + rqe->b.b_vp->v_numoutput++; /* one more output going */ +#if DEBUG + if (debug & DEBUG_ADDRESSES) + printf(" %s dev 0x%x, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n", + rqe->b.b_flags & B_READ ? "Read" : "Write", + rqe->b.b_dev, + rqe->sdno, + (u_int) (rqe->b.b_blkno - SD[rqe->sdno].driveoffset), + rqe->b.b_blkno, + rqe->b.b_bcount); /* XXX */ + if (debug & DEBUG_NUMOUTPUT) + printf(" vinumstart sd %d numoutput %ld\n", + rqe->sdno, + rqe->b.b_vp->v_numoutput); +#endif + /* fire off the request */ + s = splbio(); + (*bdevsw[major(rqe->b.b_dev)]->d_strategy) (&rqe->b); + splx(s); + } + /* XXX Do we need caching? Think about this more */ + } + } + return 0; +} + +/* define the low-level requests needed to perform a + * high-level I/O operation for a specific plex 'plexno'. + * + * Return 0 if all subdisks involved in the request are up, 1 if some + * subdisks are not up, and -1 if the request is at least partially + * outside the bounds of the subdisks. + * + * Modify the pointer *diskstart to point to the end address. On + * read, return on the first bad subdisk, so that the caller + * (build_read_request) can try alternatives. + * + * On entry to this routine, the rqg structures are not assigned. The + * assignment is performed by expandrq(). Strictly speaking, the + * elements rqe->sdno of all entries should be set to -1, since 0 + * (from bzero) is a valid subdisk number. We avoid this problem by + * initializing the ones we use, and not looking at the others (index + * >= rqg->requests). + */ +enum requeststatus +bre(struct request *rq, + int plexno, + daddr_t * diskaddr, + daddr_t diskend) +{ + BROKEN_GDB; + int sdno; + struct sd *sd; + struct rqgroup *rqg; + struct buf *bp; /* user's bp */ + struct plex *plex; + enum requeststatus status; /* return value */ + daddr_t plexoffset; /* offset of transfer in plex */ + daddr_t stripebase; /* base address of stripe (1st subdisk) */ + daddr_t stripeoffset; /* offset in stripe */ + daddr_t blockoffset; /* offset in stripe on subdisk */ + struct rqelement *rqe; /* point to this request information */ + daddr_t diskstart = *diskaddr; /* remember where this transfer starts */ + + bp = rq->bp; /* buffer pointer */ + status = REQUEST_OK; /* return value: OK until proven otherwise */ + plex = &PLEX[plexno]; /* point to the plex */ + + switch (plex->organization) { + case plex_concat: + for (sdno = 0; sdno < plex->subdisks; sdno++) { + sd = &SD[plex->sdnos[sdno]]; + if ((*diskaddr < (sd->plexoffset + sd->sectors)) /* The request starts before the end of this */ + &&(diskend > sd->plexoffset)) { /* subdisk and ends after the start of this sd */ + if ((sd->state != sd_up) || (plex->state != plex_up)) { + enum requeststatus s; + + s = checksdstate(sd, rq, *diskaddr, diskend); /* do we need to change state? */ + if (s) /* give up? */ + return s; /* yup */ + } + rqg = allocrqg(rq, 1); /* space for the request */ + if (rqg == NULL) { /* malloc failed */ + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; + } + rqg->plexno = plexno; + + rqe = &rqg->rqe[0]; /* point to the element */ + rqe->rqg = rqg; /* group */ + rqe->sdno = sd->sdno; /* put in the subdisk number */ + plexoffset = max(sd->plexoffset, *diskaddr); /* start offset in plex */ + rqe->sdoffset = plexoffset - sd->plexoffset; /* start offset in subdisk */ + rqe->useroffset = plexoffset - diskstart; /* start offset in user buffer */ + rqe->dataoffset = 0; + rqe->datalen = min(diskend - *diskaddr, /* number of sectors to transfer in this sd */ + sd->sectors - rqe->sdoffset); + rqe->groupoffset = 0; /* no groups for concatenated plexes */ + rqe->grouplen = 0; + rqe->buflen = rqe->datalen; /* buffer length is data buffer length */ + rqe->flags = 0; + rqe->driveno = sd->driveno; + *diskaddr += rqe->datalen; /* bump the address */ + if (build_rq_buffer(rqe, plex)) { /* build the buffer */ + deallocrqg(rqg); + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; /* can't do it */ + } + } + if (*diskaddr > diskend) /* we're finished, */ + break; /* get out of here */ + } + break; + + case plex_striped: + { + while (*diskaddr < diskend) { /* until we get it all sorted out */ + /* The offset of the start address from + * the start of the stripe */ + stripeoffset = *diskaddr % (plex->stripesize * plex->subdisks); + + /* The plex-relative address of the + * start of the stripe */ + stripebase = *diskaddr - stripeoffset; + + /* The number of the subdisk in which + * the start is located */ + sdno = stripeoffset / plex->stripesize; + + /* The offset from the beginning of the stripe + * on this subdisk */ + blockoffset = stripeoffset % plex->stripesize; + + sd = &SD[plex->sdnos[sdno]]; /* the subdisk in question */ + if ((sd->state != sd_up) || (plex->state != plex_up)) { + enum requeststatus s; + + s = checksdstate(sd, rq, *diskaddr, diskend); /* do we need to change state? */ + if (s) /* give up? */ + return s; /* yup */ + } + rqg = allocrqg(rq, 1); /* space for the request */ + if (rqg == NULL) { /* malloc failed */ + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; + } + rqg->plexno = plexno; + + rqe = &rqg->rqe[0]; /* point to the element */ + rqe->rqg = rqg; + rqe->sdoffset = stripebase / plex->subdisks + blockoffset; /* start offset in this subdisk */ + rqe->useroffset = *diskaddr - diskstart; /* The offset of the start in the user buffer */ + rqe->dataoffset = 0; + rqe->datalen = min(diskend - *diskaddr, /* the amount remaining to transfer */ + plex->stripesize - blockoffset); /* and the amount left in this stripe */ + rqe->groupoffset = 0; /* no groups for striped plexes */ + rqe->grouplen = 0; + rqe->buflen = rqe->datalen; /* buffer length is data buffer length */ + rqe->flags = 0; + rqe->sdno = sd->sdno; /* put in the subdisk number */ + rqe->driveno = sd->driveno; + + if (rqe->sdoffset >= sd->sectors) { /* starts beyond the end of the subdisk? */ + deallocrqg(rqg); + return REQUEST_EOF; + } else if (rqe->sdoffset + rqe->datalen > sd->sectors) /* ends beyond the end of the subdisk? */ + rqe->datalen = sd->sectors - rqe->sdoffset; /* yes, truncate */ + + if (build_rq_buffer(rqe, plex)) { /* build the buffer */ + deallocrqg(rqg); + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; /* can't do it */ + } + *diskaddr += rqe->datalen; /* look at the remainder */ + if (*diskaddr < diskend) { /* didn't finish the request on this stripe */ + plex->multiblock++; /* count another one */ + if (sdno == plex->subdisks - 1) /* last subdisk, */ + plex->multistripe++; /* another stripe as well */ + } + } + } + break; + + + default: + printf("vinum: invalid plex type in bre"); + } + + return status; +} + +/* Build up a request structure for reading volumes. + * This function is not needed for plex reads, since there's + * no recovery if a plex read can't be satisified. */ +enum requeststatus +build_read_request(struct request *rq, /* request */ + int plexindex) +{ /* index in the volume's plex table */ + BROKEN_GDB; + struct buf *bp; + daddr_t startaddr; /* offset of previous part of transfer */ + daddr_t diskaddr; /* offset of current part of transfer */ + daddr_t diskend; /* and end offset of transfer */ + int plexno; /* plex index in vinum_conf */ + struct rqgroup *rqg; /* point to the request we're working on */ + struct volume *vol; /* volume in question */ + off_t oldstart; /* note where we started */ + int recovered = 0; /* set if we recover a read */ + enum requeststatus status = REQUEST_OK; + + bp = rq->bp; /* buffer pointer */ + diskaddr = bp->b_blkno; /* start offset of transfer */ + diskend = diskaddr + (bp->b_bcount / DEV_BSIZE); /* and end offset of transfer */ + rqg = &rq->rqg[plexindex]; /* plex request */ + vol = &VOL[rq->volplex.volno]; /* point to volume */ + + while (diskaddr < diskend) { /* build up request components */ + startaddr = diskaddr; + status = bre(rq, vol->plex[plexindex], &diskaddr, diskend); /* build up a request */ + switch (status) { + case REQUEST_OK: + continue; + + case REQUEST_RECOVERED: + recovered = 1; + break; + + case REQUEST_EOF: + case REQUEST_ENOMEM: + return status; + + /* if we get here, we have either had a failure or + * a RAID 5 recovery. We don't want to use the + * recovery, because it's expensive, so first we + * check if we have alternatives */ + case REQUEST_DOWN: /* can't access the plex */ + if (vol != NULL) { /* and this is volume I/O */ + /* Try to satisfy the request + * from another plex */ + for (plexno = 0; plexno < vol->plexes; plexno++) { + diskaddr = startaddr; /* start at the beginning again */ + oldstart = startaddr; /* and note where that was */ + if (plexno != plexindex) { /* don't try this plex again */ + bre(rq, vol->plex[plexno], &diskaddr, diskend); /* try a request */ + if (diskaddr > oldstart) { /* we satisfied another part */ + recovered = 1; /* we recovered from the problem */ + status = REQUEST_OK; /* don't complain about it */ + break; + } + } + if (plexno == (vol->plexes - 1)) /* couldn't satisfy the request */ + return REQUEST_DOWN; /* failed */ + } + } else + return REQUEST_DOWN; /* bad luck */ + } + if (recovered) + vol->recovered_reads += recovered; /* adjust our recovery count */ + } + return status; +} + +/* Build up a request structure for writes. + * Return 0 if all subdisks involved in the request are up, 1 if some + * subdisks are not up, and -1 if the request is at least partially + * outside the bounds of the subdisks. */ +enum requeststatus +build_write_request(struct request *rq) +{ /* request */ + BROKEN_GDB; + struct buf *bp; + daddr_t diskstart; /* offset of current part of transfer */ + daddr_t diskend; /* and end offset of transfer */ + int plexno; /* plex index in vinum_conf */ + struct volume *vol; /* volume in question */ + enum requeststatus status; + + bp = rq->bp; /* buffer pointer */ + vol = &VOL[rq->volplex.volno]; /* point to volume */ + diskend = bp->b_blkno + (bp->b_bcount / DEV_BSIZE); /* end offset of transfer */ + status = REQUEST_OK; + for (plexno = 0; plexno < vol->plexes; plexno++) { + diskstart = bp->b_blkno; /* start offset of transfer */ + status = min(status, bre(rq, /* build requests for the plex */ + vol->plex[plexno], + &diskstart, + diskend)); + } + return status; +} + +/* Fill in the struct buf part of a request element. */ +enum requeststatus +build_rq_buffer(struct rqelement *rqe, struct plex *plex) +{ + BROKEN_GDB; + struct sd *sd; /* point to subdisk */ + struct volume *vol; + struct buf *bp; + struct buf *ubp; /* user (high level) buffer header */ + + vol = &VOL[rqe->rqg->rq->volplex.volno]; + sd = &SD[rqe->sdno]; /* point to subdisk */ + bp = &rqe->b; + ubp = rqe->rqg->rq->bp; /* pointer to user buffer header */ + + /* Initialize the buf struct */ + bzero(&rqe->b, sizeof(struct buf)); + bp->b_proc = ubp->b_proc; /* process pointer */ + bp->b_flags = ubp->b_flags & (B_NOCACHE | B_READ | B_ASYNC); /* copy these flags from user bp */ + bp->b_flags |= B_CALL | B_BUSY; /* inform us when it's done */ + if (plex->state == plex_reviving) + bp->b_flags |= B_ORDERED; /* keep request order if we're reviving */ + bp->b_iodone = complete_rqe; /* by calling us here */ + bp->b_dev = DRIVE[rqe->driveno].dev; /* drive device */ + bp->b_blkno = rqe->sdoffset + sd->driveoffset; /* start address */ + bp->b_bcount = rqe->buflen << DEV_BSHIFT; /* number of bytes to transfer */ + bp->b_resid = bp->b_bcount; /* and it's still all waiting */ + bp->b_bufsize = bp->b_bcount; /* and buffer size */ + bp->b_vp = DRIVE[rqe->driveno].vp; /* drive vnode */ + bp->b_rcred = FSCRED; /* we have the file system credentials */ + bp->b_wcred = FSCRED; /* we have the file system credentials */ + + if (rqe->flags & XFR_MALLOCED) { /* this operation requires a malloced buffer */ + bp->b_data = Malloc(bp->b_bcount); /* get a buffer to put it in */ + if (bp->b_data == NULL) { /* failed */ + Debugger("XXX"); + abortrequest(rqe->rqg->rq, ENOMEM); + return REQUEST_ENOMEM; /* no memory */ + } + } else + /* Point directly to user buffer data. This means + * that we don't need to do anything when we have + * finished the transfer */ + bp->b_data = ubp->b_data + rqe->useroffset * DEV_BSIZE; + return 0; +} +/* Abort a request: free resources and complete the + * user request with the specified error */ +int +abortrequest(struct request *rq, int error) +{ + struct buf *bp = rq->bp; /* user buffer */ + + bp->b_flags |= B_ERROR; + bp->b_error = error; + freerq(rq); /* free everything we're doing */ + biodone(bp); + return error; /* and give up */ +} + +/* Check that our transfer will cover the + * complete address space of the user request. + * + * Return 1 if it can, otherwise 0 */ +int +check_range_covered(struct request *rq) +{ + /* XXX */ + return 1; +} + +/* Perform I/O on a subdisk */ +void +sdio(struct buf *bp) +{ + int s; /* spl */ + struct sd *sd; + struct sdbuf *sbp; + daddr_t endoffset; + struct drive *drive; + + sd = &SD[SDNO(bp->b_dev)]; /* point to the subdisk */ + drive = &DRIVE[sd->driveno]; + + if (drive->state != drive_up) { /* XXX until we get the states fixed */ + set_sd_state(SDNO(bp->b_dev), sd_obsolete, setstate_force); + bp->b_flags |= B_ERROR; + bp->b_error = EIO; + biodone(bp); + return; + } + /* XXX decide which states we will really accept here. up + * implies it could be involved with a plex, in which + * case we don't want to dick with it */ + if ((sd->state != sd_up) + && (sd->state != sd_initializing) + && (sd->state != sd_reborn)) { /* we can't access it */ + bp->b_flags |= B_ERROR; + bp->b_flags = EIO; + if (bp->b_flags & B_BUSY) /* XXX why isn't this always the case? */ + biodone(bp); + return; + } + /* Get a buffer */ + sbp = (struct sdbuf *) Malloc(sizeof(struct sdbuf)); + if (sbp == NULL) { + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return; + } + bcopy(bp, &sbp->b, sizeof(struct buf)); /* start with the user's buffer */ + sbp->b.b_flags |= B_CALL; /* tell us when it's done */ + sbp->b.b_iodone = sdio_done; /* here */ + sbp->b.b_dev = DRIVE[sd->driveno].dev; /* device */ + sbp->b.b_vp = DRIVE[sd->driveno].vp; /* vnode */ + sbp->b.b_blkno += sd->driveoffset; + sbp->bp = bp; /* note the address of the original header */ + sbp->sdno = sd->sdno; /* note for statistics */ + sbp->driveno = sd->driveno; + endoffset = bp->b_blkno + sbp->b.b_bcount / DEV_BSIZE; /* final sector offset */ + if (endoffset > sd->sectors) { /* beyond the end */ + sbp->b.b_bcount -= (endoffset - sd->sectors) * DEV_BSIZE; /* trim */ + if (sbp->b.b_bcount <= 0) { /* nothing to transfer */ + bp->b_resid = bp->b_bcount; /* nothing transferred */ + /* XXX Grrr. This doesn't seem to work. Return + * an error after all */ + bp->b_flags |= B_ERROR; + bp->b_error = ENOSPC; + biodone(bp); + Free(sbp); + return; + } + } + if ((sbp->b.b_flags & B_READ) == 0) /* write */ + sbp->b.b_vp->v_numoutput++; /* one more output going */ +#if DEBUG + if (debug & DEBUG_ADDRESSES) + printf(" %s dev 0x%x, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n", + sbp->b.b_flags & B_READ ? "Read" : "Write", + sbp->b.b_dev, + sbp->sdno, + (u_int) (sbp->b.b_blkno - SD[sbp->sdno].driveoffset), + (int) sbp->b.b_blkno, + sbp->b.b_bcount); /* XXX */ + if (debug & DEBUG_NUMOUTPUT) + printf(" vinumstart sd %d numoutput %ld\n", + sbp->sdno, + sbp->b.b_vp->v_numoutput); +#endif + s = splbio(); + (*bdevsw[major(sbp->b.b_dev)]->d_strategy) (&sbp->b); + splx(s); +} + +/* Simplified version of bounds_check_with_label + * Determine the size of the transfer, and make sure it is + * within the boundaries of the partition. Adjust transfer + * if needed, and signal errors or early completion. + * + * Volumes are simpler than disk slices: they only contain + * one component (though we call them a, b and c to make + * system utilities happy), and they always take up the + * complete space of the "partition". + * + * I'm still not happy with this: why should the label be + * protected? If it weren't so damned difficult to write + * one in the first pleace (because it's protected), it wouldn't + * be a problem. + */ +int +vinum_bounds_check(struct buf *bp, struct volume *vol) +{ + int maxsize = vol->size; /* size of the partition (sectors) */ + int size = (bp->b_bcount + DEV_BSIZE - 1) >> DEV_BSHIFT; /* size of this request (sectors) */ + + /* Would this transfer overwrite the disk label? */ + if (bp->b_blkno <= LABELSECTOR /* starts before or at the label */ +#if LABELSECTOR != 0 + && bp->b_blkno + size > LABELSECTOR /* and finishes after */ +#endif + && (!(vol->flags & VF_RAW)) /* and it's not raw */ + &&major(bp->b_dev) == BDEV_MAJOR /* and it's the block device */ + && (bp->b_flags & B_READ) == 0 /* and it's a write */ + && (!vol->flags & (VF_WLABEL | VF_LABELLING))) { /* and we're not allowed to write the label */ + bp->b_error = EROFS; /* read-only */ + bp->b_flags |= B_ERROR; + return -1; + } + if (size == 0) /* no transfer specified, */ + return 0; /* treat as EOF */ + /* beyond partition? */ + if (bp->b_blkno < 0 /* negative start */ + || bp->b_blkno + size > maxsize) { /* or goes beyond the end of the partition */ + /* if exactly at end of disk, return an EOF */ + if (bp->b_blkno == maxsize) { + bp->b_resid = bp->b_bcount; + return 0; + } + /* or truncate if part of it fits */ + size = maxsize - bp->b_blkno; + if (size <= 0) { /* nothing to transfer */ + bp->b_error = EINVAL; + bp->b_flags |= B_ERROR; + return -1; + } + bp->b_bcount = size << DEV_BSHIFT; + } + bp->b_pblkno = bp->b_blkno; + return 1; +} + +/* Allocate a request group and hook + * it in in the list for rq */ +struct rqgroup * +allocrqg(struct request *rq, int elements) +{ + struct rqgroup *rqg; /* the one we're going to allocate */ + int size = sizeof(struct rqgroup) + elements * sizeof(struct rqelement); + + rqg = (struct rqgroup *) Malloc(size); + if (rqg != NULL) { /* malloc OK, */ + if (rq->rqg) /* we already have requests */ + rq->lrqg->next = rqg; /* hang it off the end */ + else /* first request */ + rq->rqg = rqg; /* at the start */ + rq->lrqg = rqg; /* this one is the last in the list */ + + bzero(rqg, size); /* no old junk */ + rqg->rq = rq; /* point back to the parent request */ + rqg->count = elements; /* number of requests in the group */ + } else + Debugger("XXX"); + return rqg; +} + +/* Deallocate a request group out of a chain. We do + * this by linear search: the chain is short, this + * almost never happens, and currently it can only + * happen to the first member of the chain. */ +void +deallocrqg(struct rqgroup *rqg) +{ + struct rqgroup *rqgc = rqg->rq->rqg; /* point to the request chain */ + + if (rqg->rq->rqg == rqg) /* we're first in line */ + rqg->rq->rqg = rqg->next; /* unhook ourselves */ + else { + while (rqgc->next != rqg) /* find the group */ + rqgc = rqgc->next; + rqgc->next = rqg->next; + } + Free(rqgc); +} + +/* Character device interface */ +int +vinumread(dev_t dev, struct uio *uio, int ioflag) +{ + return (physio(vinumstrategy, NULL, dev, 1, minphys, uio)); +} + +int +vinumwrite(dev_t dev, struct uio *uio, int ioflag) +{ + return (physio(vinumstrategy, NULL, dev, 0, minphys, uio)); +} diff --git a/sys/dev/vinum/vinumrevive.c b/sys/dev/vinum/vinumrevive.c new file mode 100644 index 0000000..2f8117a --- /dev/null +++ b/sys/dev/vinum/vinumrevive.c @@ -0,0 +1,128 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: revive.c,v 1.1 1998/08/14 06:16:59 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" + +/* revive a block of a plex. Return an error + * indication. EAGAIN means successful copy, but + * that more blocks remain to be copied. + * XXX We should specify a block size here. At the moment, + * just take a default value. FIXME */ +int +revive_block(int plexno) +{ + struct plex *plex = &PLEX[plexno]; + struct buf *bp; + int error = EAGAIN; + int size; + int s; /* priority level */ + + if (plex->revive_blocksize == 0) { + if (plex->stripesize != 0) /* we're striped, don't revive more than */ + plex->revive_blocksize = min(DEFAULT_REVIVE_BLOCKSIZE, plex->stripesize); /* one block at a time */ + else + plex->revive_blocksize = DEFAULT_REVIVE_BLOCKSIZE; + } + size = min(plex->revive_blocksize, plex->length - plex->revived) << DEV_BSHIFT; + + s = splbio(); + /* Get a buffer */ + bp = geteblk(size); + if (bp == NULL) { + splx(s); + return ENOMEM; + } + if (bp->b_qindex != 0) /* on a queue, */ + bremfree(bp); /* remove it */ + splx(s); + + /* Amount to transfer: block size, unless it + * would overlap the end */ + bp->b_bufsize = size; + bp->b_bcount = bp->b_bufsize; + bp->b_resid = 0x0; + bp->b_blkno = plex->revived; /* we've got this far */ + + /* XXX what about reviving anonymous plexes? */ + + /* First, read the data from the volume. We don't + * care which plex, that's bre's job */ + bp->b_dev = VINUMBDEV(plex->volno, 0, 0, VINUM_VOLUME_TYPE); /* create the device number */ + bp->b_flags = B_BUSY | B_READ; + vinumstart(bp, 1); + biowait(bp); + if (bp->b_flags & B_ERROR) + error = bp->b_error; + else + /* Now write to the plex */ + { + s = splbio(); + if (bp->b_qindex != 0) /* on a queue, */ + bremfree(bp); /* remove it */ + splx(s); + bp->b_dev = VINUMBDEV(plex->volno, plex->volplexno, 0, VINUM_PLEX_TYPE); /* create the device number */ + + bp->b_flags = B_BUSY; /* make this a write */ + bp->b_resid = 0x0; + vinumstart(bp, 1); + biowait(bp); + if (bp->b_flags & B_ERROR) + error = bp->b_error; + else { + plex->revived += bp->b_bcount >> DEV_BSHIFT; /* moved this much further down */ + if (plex->revived >= plex->length) { /* finished */ + plex->revived = 0; + plex->state = plex_up; /* do we need to do more? */ + if (plex->volno >= 0) /* we have a volume, */ + set_volume_state(plex->volno, volume_up, 0); + printf("vinum: plex %s is %s\n", plex->name, plex_state(plex->state)); + save_config(); /* and save the updated configuration */ + error = 0; /* we're done */ + } + } + while (plex->waitlist) { /* we have waiting requests */ + launch_requests(plex->waitlist, 1); /* do them now */ + plex->waitlist = plex->waitlist->next; /* and move on to the next */ + } + } + if (bp->b_qindex == 0) /* not on a queue, */ + brelse(bp); /* is this kosher? */ + return error; +} diff --git a/sys/dev/vinum/vinumstate.c b/sys/dev/vinum/vinumstate.c new file mode 100644 index 0000000..2ce2ed0 --- /dev/null +++ b/sys/dev/vinum/vinumstate.c @@ -0,0 +1,755 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: state.c,v 2.6 1998/08/19 08:04:47 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" + +/* Update drive state */ +/* Return 1 if the state changes, otherwise 0 */ +int +set_drive_state(int driveno, enum drivestate state, int flags) +{ + struct drive *drive = &DRIVE[driveno]; + int oldstate = drive->state; + int sdno; + + if (drive->state == drive_unallocated) /* no drive to do anything with, */ + return 0; + + if (state != oldstate) { /* don't change it if it's not different */ + if (state == drive_down) { /* the drive's going down */ + if (flags || (drive->opencount == 0)) { /* we can do it */ + close_drive(drive); + drive->state = state; + printf("vinum: drive %s is %s\n", drive->label.name, drive_state(drive->state)); + } else + return 0; /* don't do it */ + } + drive->state = state; /* set the state */ + if (((drive->state == drive_up) + || ((drive->state == drive_coming_up))) + && (drive->vp == NULL)) /* should be open, but we're not */ + init_drive(drive); /* which changes the state again */ + if ((state != oldstate) /* state has changed */ + &&((flags & setstate_norecurse) == 0)) { /* and we want to recurse, */ + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) { /* find this drive's subdisks */ + if (SD[sdno].driveno == driveno) /* belongs to this drive */ + set_sd_state(sdno, sd_down, setstate_force | setstate_recursing); /* take it down */ + } + save_config(); /* and save the updated configuration */ + return 1; + } + } + return 0; +} + +/* Try to set the subdisk state. Return 1 if state changed to + * what we wanted, -1 if it changed to something else, and 0 + * if no change. + * + * This routine is called both from the user (up, down states + * only) and internally. + */ +int +set_sd_state(int sdno, enum sdstate state, enum setstateflags flags) +{ + struct sd *sd = &SD[sdno]; + int oldstate = sd->state; + int status = 1; /* status to return */ + + if (state == oldstate) + return 0; /* no change */ + + if (sd->state == sd_unallocated) /* no subdisk to do anything with, */ + return 0; + + if (sd->driveoffset < 0) { /* not allocated space */ + sd->state = sd_down; + if (state != sd_down) + return -1; + } else { /* space allocated */ + switch (state) { + case sd_down: + if ((!flags & setstate_force) /* but gently */ + &&(sd->plexno >= 0)) /* and we're attached to a plex, */ + return 0; /* don't do it */ + break; + + case sd_up: + if (DRIVE[sd->driveno].state != drive_up) /* can't bring the sd up if the drive isn't, */ + return 0; /* not even by force */ + switch (sd->state) { + case sd_obsolete: + case sd_down: /* been down, no data lost */ + if ((sd->plexno) /* we're associated with a plex */ + &&(((PLEX[sd->plexno].state < plex_firstup) /* and it's not up */ + ||(PLEX[sd->plexno].subdisks > 1)))) /* or it's the only one */ + break; + /* XXX Get this right: make sure that other plexes in + * the volume cover this address space, otherwise + * we make this one sd_up */ + sd->state = sd_reborn; /* here it is again */ + printf("vinum: subdisk %s is %s, not %s\n", sd->name, sd_state(sd->state), sd_state(state)); + status = -1; + break; + + case sd_init: /* brand new */ + if (flags & setstate_configuring) /* we're doing this while configuring */ + break; + sd->state = sd_empty; /* nothing in it */ + printf("vinum: subdisk %s is %s, not %s\n", sd->name, sd_state(sd->state), sd_state(state)); + status = -1; + break; + + case sd_initializing: + break; /* go on and do it */ + + case sd_empty: + if ((sd->plexno) /* we're associated with a plex */ + &&(((PLEX[sd->plexno].state < plex_firstup) /* and it's not up */ + ||(PLEX[sd->plexno].subdisks > 1)))) /* or it's the only one */ + break; + return 0; /* can't do it */ + + default: /* can't do it */ + /* There's no way to bring subdisks up directly from + * other states. First they need to be initialized + * or revived */ + return 0; + } + break; + + default: /* other ones, only internal with force */ + if (flags & setstate_force == 0) /* no force? What's this? */ + return 0; /* don't do it */ + } + } + sd->state = state; + printf("vinum: subdisk %s is %s\n", sd->name, sd_state(sd->state)); + if ((flags & setstate_norecurse) == 0) + set_plex_state(sd->plexno, plex_up, setstate_recursing); /* update plex state */ + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return status; +} + +/* Called from request routines when they find + * a subdisk which is not kosher. Decide whether + * it warrants changing the state. Return + * REQUEST_DOWN if we can't use the subdisk, + * REQUEST_OK if we can. */ +enum requeststatus +checksdstate(struct sd *sd, struct request *rq, daddr_t diskaddr, daddr_t diskend) +{ + struct plex *plex = &PLEX[sd->plexno]; + int writeop = (rq->bp->b_flags & B_READ) == 0; /* note if we're writing */ + + /* first, see if the plex wants to be accessed */ + switch (plex->state) { + case plex_reviving: + /* When writing, we'll write anything that starts + * up to the current revive pointer, but we'll + * only accept a read which finishes before the + * current revive pointer. + */ + if ((writeop && (diskaddr > plex->revived)) /* write starts after current revive pointer */ + ||((!writeop) && (diskend >= plex->revived))) { /* or read ends after current revive pointer */ + if (writeop) { /* writing to a consistent down disk */ + if (DRIVE[sd->driveno].state == drive_up) + set_sd_state(sd->sdno, sd_stale, setstate_force); /* it's not consistent now */ + else + set_sd_state(sd->sdno, sd_obsolete, setstate_force); /* it's not consistent now */ + } + return REQUEST_DOWN; /* that part of the plex is still down */ + } else if (diskend >= plex->revived) /* write finishes beyond revive pointer */ + rq->flags |= XFR_REVIVECONFLICT; /* note a potential conflict */ + /* FALLTHROUGH */ + + case plex_up: + case plex_degraded: + case plex_flaky: + /* We can access the plex: let's see + * how the subdisk feels */ + switch (sd->state) { + case sd_up: + return REQUEST_OK; + + case sd_reborn: + if (writeop) + return REQUEST_OK; /* always write to a reborn disk */ + /* Handle the mapping. We don't want to reject + * a read request to a reborn subdisk if that's + * all we have. XXX */ + return REQUEST_DOWN; + + case sd_down: + case sd_crashed: + if (writeop) { /* writing to a consistent down disk */ + if (DRIVE[sd->driveno].state == drive_up) + set_sd_state(sd->sdno, sd_stale, setstate_force); /* it's not consistent now */ + else + set_sd_state(sd->sdno, sd_obsolete, setstate_force); /* it's not consistent now */ + } + return REQUEST_DOWN; /* and it's down one way or another */ + + default: + return REQUEST_DOWN; + } + + default: + return REQUEST_DOWN; + } +} + +void +add_defective_region(struct plex *plex, off_t offset, size_t length) +{ +/* XXX get this ordered, and coalesce regions if necessary */ + if (++plex->defective_regions > plex->defective_region_count) + EXPAND(plex->defective_region, + struct plexregion, + plex->defective_region_count, + PLEX_REGION_TABLE_SIZE); + plex->defective_region[plex->defective_regions - 1].offset = offset; + plex->defective_region[plex->defective_regions - 1].length = length; +} + +void +add_unmapped_region(struct plex *plex, off_t offset, size_t length) +{ + if (++plex->unmapped_regions > plex->unmapped_region_count) + EXPAND(plex->unmapped_region, + struct plexregion, + plex->unmapped_region_count, + PLEX_REGION_TABLE_SIZE); + plex->unmapped_region[plex->unmapped_regions - 1].offset = offset; + plex->unmapped_region[plex->unmapped_regions - 1].length = length; +} + +/* Rebuild a plex free list and set state if + * we have a configuration error */ +void +rebuild_plex_unmappedlist(struct plex *plex) +{ + int sdno; + struct sd *sd; + int lastsdend = 0; /* end offset of last subdisk */ + + if (plex->unmapped_region != NULL) { /* we're going to rebuild it */ + Free(plex->unmapped_region); + plex->unmapped_region = NULL; + plex->unmapped_regions = 0; + plex->unmapped_region_count = 0; + } + if (plex->defective_region != NULL) { + Free(plex->defective_region); + plex->defective_region = NULL; + plex->defective_regions = 0; + plex->defective_region_count = 0; + } + for (sdno = 0; sdno < plex->subdisks; sdno++) { + sd = &SD[plex->sdnos[sdno]]; + if (sd->plexoffset < lastsdend) { /* overlap */ + printf("vinum: Plex %s, subdisk %s overlaps previous\n", plex->name, sd->name); + set_plex_state(plex->plexno, plex_down, setstate_force); /* don't allow that */ + } else if (sd->plexoffset > lastsdend) /* gap */ + add_unmapped_region(plex, lastsdend, sd->plexoffset - lastsdend); + else if (sd->state < sd_reborn) /* this part defective */ + add_defective_region(plex, sd->plexoffset, sd->sectors); + lastsdend = sd->plexoffset + sd->sectors; + } +} + +/* return a state map for the subdisks of a plex */ +enum sdstates +sdstatemap(struct plex *plex, int *sddowncount) +{ + int sdno; + enum sdstates statemap = 0; /* note the states we find */ + + *sddowncount = 0; /* no subdisks down yet */ + for (sdno = 0; sdno < plex->subdisks; sdno++) { + struct sd *sd = &SD[plex->sdnos[sdno]]; /* point to the subdisk */ + + switch (sd->state) { + case sd_empty: + statemap |= sd_emptystate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_init: + statemap |= sd_initstate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_down: + statemap |= sd_downstate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_crashed: + statemap |= sd_crashedstate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_obsolete: + statemap |= sd_obsolete; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_stale: + statemap |= sd_stalestate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_reborn: + statemap |= sd_rebornstate; + break; + + case sd_up: + statemap |= sd_upstate; + break; + + default: + statemap |= sd_otherstate; + break; + } + } + return statemap; +} + +/* determine the state of the volume relative to this plex */ +enum volplexstate +vpstate(struct plex *plex) +{ + struct volume *vol; + enum volplexstate state = volplex_onlyusdown; /* state to return */ + int plexno; + + if (plex->volno < 0) /* not associated with a volume */ + return volplex_onlyusdown; /* assume the worst */ + + vol = &VOL[plex->volno]; /* point to our volume */ + for (plexno = 0; plexno < vol->plexes; plexno++) { + if (&PLEX[vol->plex[plexno]] == plex) { /* us */ + if (PLEX[vol->plex[plexno]].state == plex_up) /* are we up? */ + state |= volplex_onlyus; /* yes */ + } else { + if (PLEX[vol->plex[plexno]].state == plex_up) /* not us */ + state |= volplex_otherup; /* and when they were up, they were up */ + else + state |= volplex_alldown; /* and when they were down, they were down */ + } + } + return state; /* and when they were only halfway up */ +} /* they were neither up nor down */ + +/* Check if all bits b are set in a */ +int allset(int a, int b); + +int +allset(int a, int b) +{ + return (a & b) == b; +} + +/* Update the state of a plex dependent on its subdisks. + * Also rebuild the unmapped_region and defective_region table */ +int +set_plex_state(int plexno, enum plexstate state, enum setstateflags flags) +{ + int sddowncount = 0; /* number of down subdisks */ + struct plex *plex = &PLEX[plexno]; /* point to our plex */ + enum plexstate oldstate = plex->state; + enum volplexstate vps = vpstate(plex); /* how do we compare with the other plexes? */ + enum sdstates statemap = sdstatemap(plex, &sddowncount); /* get a map of the subdisk states */ + + if ((flags & setstate_force) && (oldstate == state)) /* we're there already, */ + return 0; /* no change */ + + if (plex->state == plex_unallocated) /* no plex to do anything with, */ + return 0; + + switch (state) { + case plex_up: + if ((plex->state == plex_initializing) /* we're initializing */ + &&(statemap != sd_upstate)) /* but SDs aren't up yet */ + return 0; /* do nothing */ + + /* We don't really care what our state was before + * if we want to come up. We rely entirely on the + * state of our subdisks and our volume */ + switch (vps) { + case volplex_onlyusdown: + case volplex_alldown: /* another plex is down, and so are we */ + if (statemap == sd_upstate) { /* all subdisks ready for action */ + if ((plex->state == plex_init) /* we're brand spanking new */ + &&(VOL[plex->volno].flags & VF_CONFIG_SETUPSTATE)) { /* and we consider that up */ + /* Conceptually, an empty plex does not contain valid data, + * but normally we'll see this state when we have just + * created a plex, and it's either consistent from earlier, + * or we don't care about the previous contents (we're going + * to create a file system or use it for swap). + * + * We need to do this in one swell foop: on the next call + * we will no longer be just empty. + * + * We'll still come back to this function for the remaining + * plexes in the volume. They'll be up already, so that + * doesn't change anything, but it's not worth the additional + * code to stop doing it. */ + struct volume *vol = &VOL[plex->volno]; + int plexno; + + for (plexno = 0; plexno < vol->plexes; plexno++) + PLEX[vol->plex[plexno]].state = plex_up; + } + plex->state = plex_up; /* bring up up, anyway */ + } else + plex->state = plex_down; + break; + + case volplex_onlyusup: /* only we are up: others are down */ + case volplex_onlyus: /* we're up and alone */ + if ((statemap == sd_upstate) /* subdisks all up */ + ||(statemap == sd_emptystate)) /* or all empty */ + plex->state = plex_up; /* go for it */ + else if ((statemap & (sd_upstate | sd_reborn)) == statemap) /* all up or reborn, */ + plex->state = plex_flaky; + else if (statemap & (sd_upstate | sd_reborn)) /* some up or reborn, */ + plex->state = plex_degraded; /* so far no corruption */ + else + plex->state = plex_faulty; + break; + + case volplex_otherup: /* another plex is up */ + case volplex_otherupdown: /* other plexes are up and down */ + if ((statemap == sd_upstate) /* subdisks all up */ + ||(statemap == sd_emptystate) /* or all empty */ + ) { + /* Is the data in all subdisks valid? */ + if (statemap == statemap & (sd_downstate | sd_rebornstate | sd_upstate)) + break; /* yes, we can bring the plex up */ + plex->state = plex_reviving; /* we need reviving */ + return EAGAIN; + } else + plex->state = plex_faulty; /* still in error */ + break; + + case volplex_allup: /* all plexes are up */ + case volplex_someup: + if ((statemap & (sd_upstate | sd_reborn)) == statemap) /* all up or reborn, */ + break; /* no change */ + else + plex->state = plex_degraded; /* we're not all there */ + } + + if (plex->state != oldstate) + break; + return 0; /* no change */ + + case plex_down: /* want to take it down */ + if (((vps == volplex_onlyus) /* we're the only one up */ + ||(vps == volplex_onlyusup)) /* we're the only one up */ + &&(!(flags & setstate_force))) /* and we don't want to use force */ + return 0; /* can't do it */ + plex->state = state; /* do it */ + break; + + /* This is only requested by the driver. + * Trust ourselves */ + case plex_faulty: + plex->state = state; /* do it */ + break; + + case plex_initializing: + /* XXX consider what safeguards we need here */ + if ((flags & setstate_force) == 0) + return 0; + plex->state = state; /* do it */ + break; + + /* What's this? */ + default: + return 0; + } + printf("vinum: plex %s is %s\n", plex->name, plex_state(plex->state)); + /* Now see what we have left, and whether + * we're taking the volume down */ + if (plex->volno >= 0) { /* we have a volume */ + struct volume *vol = &VOL[plex->volno]; + + vps = vpstate(plex); /* get our combined state again */ + if ((flags & setstate_norecurse) == 0) { /* we can recurse */ + if ((vol->state == volume_up) + && (vps == volplex_alldown)) /* and we're all down */ + set_volume_state(plex->volno, volume_down, setstate_recursing); /* take our volume down */ + else if ((vol->state == volume_down) + && (vps & (volplex_otherup | volplex_onlyusup))) /* and at least one is up */ + set_volume_state(plex->volno, volume_up, setstate_recursing); /* bring our volume up */ + } + } + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return 1; +} + +/* Update the state of a plex dependent on its plexes. + * Also rebuild the unmapped_region and defective_region table */ +int +set_volume_state(int volno, enum volumestate state, enum setstateflags flags) +{ + int plexno; + enum plexstates { + plex_downstate = 1, /* found a plex which is down */ + plex_degradedstate = 2, /* found a plex which is halfway up */ + plex_upstate = 4 /* found a plex which is completely up */ + }; + + int plexstatemap = 0; /* note the states we find */ + struct volume *vol = &VOL[volno]; /* point to our volume */ + + if (vol->state == state) /* we're there already */ + return 0; /* no change */ + if (vol->state == volume_unallocated) /* no volume to do anything with, */ + return 0; + + for (plexno = 0; plexno < vol->plexes; plexno++) { + struct plex *plex = &PLEX[vol->plex[plexno]]; /* point to the plex */ + switch (plex->state) { + case plex_degraded: + case plex_flaky: + case plex_reviving: + plexstatemap |= plex_degradedstate; + break; + + case plex_up: + plexstatemap |= plex_upstate; + break; + + default: + plexstatemap |= plex_downstate; + break; + } + } + + if (state == volume_up) { /* want to come up */ + if (plexstatemap & plex_upstate) { /* we have a plex which is completely up */ + vol->state = volume_up; /* did it */ + printf("vinum: volume %s is %s\n", vol->name, volume_state(vol->state)); + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return 1; + } + /* Here we should check whether we have enough + * coverage for the complete volume. Writeme XXX */ + } else if (state == volume_down) { /* want to go down */ + if ((vol->opencount == 0) /* not open */ + ||(flags & setstate_force != 0)) { /* or we're forcing */ + vol->state = volume_down; + printf("vinum: volume %s is %s\n", vol->name, volume_state(vol->state)); + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return 1; + } + } + return 0; /* no change */ +} + +/* Start an object, in other words do what we can to get it up. + * This is called from vinumioctl (VINUMSTART). + * Return error indications via ioctl_reply + */ +void +start_object(struct vinum_ioctl_msg *data) +{ + int status; + int realstatus; /* what we really have */ + int objindex = data->index; /* data gets overwritten */ + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) data; /* format for returning replies */ + + switch (data->type) { + case drive_object: + status = set_drive_state(objindex, drive_up, setstate_none); + realstatus = DRIVE[objindex].state == drive_up; /* set status on whether we really did it */ + break; + + case sd_object: + status = set_sd_state(objindex, sd_up, setstate_none); /* set state */ + realstatus = SD[objindex].state == sd_up; /* set status on whether we really did it */ + break; + + case plex_object: + if (PLEX[objindex].state == plex_reviving) { /* reviving, */ + ioctl_reply->error = revive_block(objindex); /* revive another block */ + ioctl_reply->msg[0] = '\0'; /* no comment */ + return; + } + status = set_plex_state(objindex, plex_up, setstate_none); + realstatus = PLEX[objindex].state == plex_up; /* set status on whether we really did it */ + break; + + case volume_object: + status = set_volume_state(objindex, volume_up, setstate_none); + realstatus = VOL[objindex].state == volume_up; /* set status on whether we really did it */ + break; + + default: + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "Invalid object type"); + return; + } + /* There's no point in saying anything here: + * the userland program does it better */ + ioctl_reply->msg[0] = '\0'; + if (realstatus == 0) /* couldn't do it */ + ioctl_reply->error = EINVAL; + else + ioctl_reply->error = 0; +} + +/* Stop an object, in other words do what we can to get it down + * This is called from vinumioctl (VINUMSTOP). + * Return error indications via ioctl_reply. + */ +void +stop_object(struct vinum_ioctl_msg *data) +{ + int status = 1; + int objindex = data->index; /* save the number from change */ + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) data; /* format for returning replies */ + + switch (data->type) { + case drive_object: + status = set_drive_state(objindex, drive_down, data->force); + break; + + case sd_object: + status = set_sd_state(objindex, sd_down, data->force); + break; + + case plex_object: + status = set_plex_state(objindex, plex_down, data->force); + break; + + case volume_object: + status = set_volume_state(objindex, volume_down, data->force); + break; + + default: + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "Invalid object type"); + return; + } + ioctl_reply->msg[0] = '\0'; + if (status == 0) /* couldn't do it */ + ioctl_reply->error = EINVAL; + else + ioctl_reply->error = 0; +} + +/* VINUM_SETSTATE ioctl: set an object state + * msg is the message passed by the user */ +void +setstate(struct vinum_ioctl_msg *msg) +{ + int sdno; + struct sd *sd; + struct plex *plex; + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) msg; /* format for returning replies */ + + switch (msg->state) { + case object_down: + stop_object(msg); + break; + + case object_initializing: + switch (msg->type) { + case sd_object: + sd = &SD[msg->index]; + if ((msg->index >= vinum_conf.subdisks_used) + || (sd->state == sd_unallocated)) { + sprintf(ioctl_reply->msg, "Invalid subdisk %d", msg->index); + ioctl_reply->error = EFAULT; + return; + } + set_sd_state(msg->index, sd_initializing, msg->force); + if (sd->state != sd_initializing) { + strcpy(ioctl_reply->msg, "Can't set state"); + ioctl_reply->error = EINVAL; + } else + ioctl_reply->error = 0; + break; + + case plex_object: + plex = &PLEX[msg->index]; + if ((msg->index >= vinum_conf.plexes_used) + || (plex->state == plex_unallocated)) { + sprintf(ioctl_reply->msg, "Invalid subdisk %d", msg->index); + ioctl_reply->error = EFAULT; + return; + } + set_plex_state(msg->index, plex_initializing, msg->force); + if (plex->state != plex_initializing) { + strcpy(ioctl_reply->msg, "Can't set state"); + ioctl_reply->error = EINVAL; + } else { + ioctl_reply->error = 0; + for (sdno = 0; sdno < plex->subdisks; sdno++) { + sd = &SD[plex->sdnos[sdno]]; + set_sd_state(plex->sdnos[sdno], sd_initializing, msg->force); + if (sd->state != sd_initializing) { + strcpy(ioctl_reply->msg, "Can't set state"); + ioctl_reply->error = EINVAL; + break; + } + } + } + break; + + default: + strcpy(ioctl_reply->msg, "Invalid object"); + ioctl_reply->error = EINVAL; + } + break; + + case object_up: + start_object(msg); + } +} diff --git a/sys/dev/vinum/vinumstate.h b/sys/dev/vinum/vinumstate.h new file mode 100644 index 0000000..0b07590 --- /dev/null +++ b/sys/dev/vinum/vinumstate.h @@ -0,0 +1,213 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumstate.h,v 1.11 1998/08/04 06:22:49 grog Exp grog $ + */ + +/* This file gets read by makestatetext to create text files + * with the names of the states, so don't change the file + * format */ + +enum volumestate { + volume_unallocated, + /* present but unused. Must be 0 */ + + volume_uninit, + /* mentioned elsewhere but not defined */ + + volume_down, + + /* The volume is up and functional, but not all plexes may be available */ + volume_up, + volume_laststate = volume_up /* last value, for table dimensions */ +}; + +enum plexstate { + /* An empty entry, not a plex at all. */ + plex_unallocated, + + /* The plex has been allocated, but there configuration + * is not complete */ + plex_init, + + /* A plex which has gone completely down because of + * I/O errors. */ + plex_faulty, + + /* A plex which has been taken down by the + * administrator. */ + plex_down, + + /* A plex which is currently being brought up after + * being not up. This involves copying data from + * another plex */ + plex_reviving, + + /* A plex which is being initialized */ + plex_initializing, + + /* *** The remaining states represent plexes which are + * at least partially up. Keep these separate so that + * they can be checked more easily. */ + + /* A plex entry which is at least partially up. Not + * all subdisks are available, and an inconsistency + * has occurred. If no other plex is uncorrupted, + * the volume is no longer consistent. */ + plex_corrupt, + + plex_firstup = plex_corrupt, /* first "up" state */ + + /* A plex entry which is at least partially up. Not + * all subdisks are available, but so far no + * inconsistency has occurred (this will change with + * the first write to the address space occupied by + * a defective subdisk). A RAID 5 plex with one subdisk + * down will remain degraded even after a write */ + plex_degraded, + + /* A plex which is really up, but which has a reborn + * subdisk which we don't completely trust, and + * which we don't want to read if we can avoid it */ + plex_flaky, + + /* A plex entry which is completely up. All subdisks + * are up. */ + plex_up, + + plex_laststate = plex_up /* last value, for table dimensions */ +}; + +/* subdisk states */ +enum sdstate { + /* An empty entry, not a subdisk at all. */ + sd_unallocated, + + /* A subdisk entry which has not been created + * completely. Some fields may be empty. + */ + sd_uninit, + + /* A subdisk entry which has been created completely. + * All fields are correct, but the disk hasn't + * been updated. + */ + sd_init, + + /* A subdisk entry which has been created completely and + * which is currently being initialized */ + sd_initializing, + + /* A subdisk entry which has been created completely. + * All fields are correct, and the disk has been + * updated, but there is no data on the disk. + */ + sd_empty, + + /* *** The following states represent invalid data */ + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down, and as a result updates have been + * missed. + */ + sd_obsolete, + + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down, updates have been lost, and then + * the drive came up again. + */ + sd_stale, + + /* *** The following states represent valid, inaccessible data */ + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down. No attempt has been made to write + * to the subdisk since the crash. + */ + sd_crashed, + + /* A subdisk entry which was up, which contained + * valid data, and which was taken down by the + * administrator. The data is valid. */ + sd_down, + + /* *** The following states represent accessible subdisks + * with valid data */ + + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down and up again. No updates were lost, + * but it is possible that the subdisk has been + * damaged. We won't read from this subdisk if we + * have a choice. If this is the only subdisk which + * covers this address space in the plex, we set its + * state to sd_up under these circumstances, so this + * status implies that there is another subdisk to + * fulfil the request. + */ + sd_reborn, + + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data is valid. + */ + sd_up, + + sd_laststate = sd_up /* last value, for table dimensions */ +}; + +enum drivestate { + drive_unallocated, + /* present but unused. Must be 0 */ + + drive_uninit, + /* just mentioned in some other config entry */ + + drive_down, + /* not accessible */ + + drive_coming_up, + /* in the process of being brought up */ + + drive_up, + /* up and running */ + + drive_laststate = drive_up /* last value, for table dimensions */ +}; diff --git a/sys/dev/vinum/vinumutil.c b/sys/dev/vinum/vinumutil.c new file mode 100644 index 0000000..5121404 --- /dev/null +++ b/sys/dev/vinum/vinumutil.c @@ -0,0 +1,211 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: util.c,v 1.7 1998/08/07 09:23:10 grog Exp grog $ + */ + +/* This file contains utility routines used both in kernel and user context */ + +#include "vinumhdr.h" +#include "statetexts.h" +#ifndef REALLYKERNEL +#include <stdio.h> +extern jmp_buf command_fail; /* return on a failed command */ +#endif + +static char numeric_state[32]; /* temporary buffer for ASCII conversions */ +#define STATECOUNT(x) (sizeof (x##statetext) / sizeof (char *)) +/* Return drive state as a string */ +char * +drive_state(enum drivestate state) +{ + if (((unsigned) state) >= STATECOUNT(drive)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return drivestatetext[state]; +} + +/* Return volume state as a string */ +char * +volume_state(enum volumestate state) +{ + if (((unsigned) state) >= STATECOUNT(vol)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return volstatetext[state]; +} + +/* Return plex state as a string */ +char * +plex_state(enum plexstate state) +{ + if (((unsigned) state) >= STATECOUNT(plex)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return plexstatetext[state]; +} + +/* Return plex organization as a string */ +char * +plex_org(enum plexorg org) +{ + switch (org) { + case plex_disorg: /* disorganized */ + return "disorg"; + break; + + case plex_concat: /* concatenated plex */ + return "concat"; + break; + + case plex_striped: /* striped plex */ + return "striped"; + break; + + + default: + sprintf(numeric_state, "Invalid org %d", (int) org); + return numeric_state; + } +} + +/* Return sd state as a string */ +char * +sd_state(enum sdstate state) +{ + if (((unsigned) state) >= STATECOUNT(sd)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return sdstatetext[state]; +} + +/* Now convert in the other direction */ +/* These are currently used only internally, + * so we don't do too much error checking */ +enum drivestate +DriveState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(drive); i++) + if (strcmp(text, drivestatetext[i]) == 0) /* found it */ + return (enum drivestate) i; + return -1; +} + +enum sdstate +SdState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(sd); i++) + if (strcmp(text, sdstatetext[i]) == 0) /* found it */ + return (enum sdstate) i; + return -1; +} + +enum plexstate +PlexState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(plex); i++) + if (strcmp(text, plexstatetext[i]) == 0) /* found it */ + return (enum plexstate) i; + return -1; +} + +enum volumestate +VolState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(vol); i++) + if (strcmp(text, volstatetext[i]) == 0) /* found it */ + return (enum volstate) i; + return -1; +} + +/* Take a number with an optional scale factor and convert + * it to a number of bytes. + * + * The scale factors are: + * + * b blocks (of 512 bytes) + * k kilobytes (1024 bytes) + * m megabytes (of 1024 * 1024 bytes) + * g gigabytes (of 1024 * 1024 * 1024 bytes) + */ +u_int64_t +sizespec(char *spec) +{ + u_int64_t size; + char *s; + + size = 0; + s = spec; + if ((*s >= '0') && (*s <= '9')) { /* it's numeric */ + while ((*s >= '0') && (*s <= '9')) /* it's numeric */ + size = size * 10 + *s++ - '0'; /* convert it */ + switch (*s) { + case '\0': + return size; + + case 'B': + case 'b': + return size * 512; + + case 'K': + case 'k': + return size * 1024; + + case 'M': + case 'm': + return size * 1024 * 1024; + + case 'G': + case 'g': + return size * 1024 * 1024 * 1024; + } + } +#ifdef REALLYKERNEL + throw_rude_remark(EINVAL, "Invalid length specification: %s", spec); +#else + fprintf(stderr, "Invalid length specification: %s", spec); + longjmp(command_fail, -1); +#endif + /* NOTREACHED */ + return -1; +} diff --git a/sys/dev/vinum/vinumvar.h b/sys/dev/vinum/vinumvar.h new file mode 100644 index 0000000..ca54f76 --- /dev/null +++ b/sys/dev/vinum/vinumvar.h @@ -0,0 +1,510 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumvar.h,v 1.15 1998/08/14 06:36:41 grog Exp grog $ + */ + +/* XXX gdb can't find our global pointers, so use this kludge to + * point to them locally. Remove after testing */ +#define BROKEN_GDB struct _vinum_conf *VC = &vinum_conf + +#include <sys/time.h> +#include "vinumstate.h" +/* Some configuration maxima. They're an enum because + * we can't define global constants. Sorry about that. + * + * These aren't as bad as they look: most of them + * are soft limits. Only the MAXCONFIG parameter is set in stone + */ + +enum constants { + VINUM_HEADER = 512, /* size of header on disk */ + MAXCONFIGLINE = 1024, /* maximum size of a single config line */ + /* XXX Do we still need this? */ + MINVINUMSLICE = 1048576, /* minimum size of a slice */ + + CDEV_MAJOR = 91, /* major number for character device */ + BDEV_MAJOR = 25, /* and block device */ + + ROUND_ROBIN_READPOL = -1, /* round robin read policy */ + + /* type field in minor number */ + VINUM_VOLUME_TYPE = 0, + VINUM_PLEX_TYPE = 1, + VINUM_SD_TYPE = 2, + VINUM_DRIVE_TYPE = 3, + VINUM_SUPERDEV_TYPE = 4, /* super device. */ + + /* Shifts for the individual fields in the device */ + VINUM_TYPE_SHIFT = 28, + VINUM_VOL_SHIFT = 0, + VINUM_PLEX_SHIFT = 16, + VINUM_SD_SHIFT = 20, + VINUM_VOL_WIDTH = 8, + VINUM_PLEX_WIDTH = 3, + VINUM_SD_WIDTH = 8, + MAJORDEV_SHIFT = 8, + + +/* Create a block device number */ +#define VINUMBDEV(v,p,s,t) ((BDEV_MAJOR << MAJORDEV_SHIFT) \ + | (v << VINUM_VOL_SHIFT) \ + | (p << VINUM_PLEX_SHIFT) \ + | (s << VINUM_SD_SHIFT) \ + | (t << VINUM_TYPE_SHIFT) ) + +/* And a character device number */ +#define VINUMCDEV(v,p,s,t) ((CDEV_MAJOR << MAJORDEV_SHIFT) \ + | (v << VINUM_VOL_SHIFT) \ + | (p << VINUM_PLEX_SHIFT) \ + | (s << VINUM_SD_SHIFT) \ + | (t << VINUM_TYPE_SHIFT) ) + +/* extract device type */ +#define DEVTYPE(x) ((x >> VINUM_TYPE_SHIFT) & 7) + +/* extract volume number */ +#define VOLNO(x) (x & ((1 << VINUM_VOL_WIDTH) - 1)) + +/* extract plex number */ +#define PLEXNO(x) (VOL [VOLNO (x)].plex [(x >> VINUM_PLEX_SHIFT) & ((1 << VINUM_PLEX_WIDTH) - 1)]) + +/* extract subdisk number */ +#define SDNO(x) (PLEX [PLEXNO (x)].sdnos [(x >> VINUM_SD_SHIFT) & ((1 << VINUM_SD_WIDTH) - 1)]) + +/* extract drive number */ +#define DRIVENO(x) (SD [SDNO (x)].driveno) + + VINUM_SUPERDEV = VINUMBDEV(0, 0, 0, VINUM_SUPERDEV_TYPE), /* superdevice number */ + +/* the number of object entries to cater for initially, and also the + * value by which they are incremented. It doesn't take long + * to extend them, so theoretically we could start with 1 of each, but + * it's untidy to allocate such small areas. These values are + * probably too small. + */ + + INITIAL_DRIVES = 4, + INITIAL_VOLUMES = 4, + INITIAL_PLEXES = 8, + INITIAL_SUBDISKS = 16, + INITIAL_SUBDISKS_IN_PLEX = 4, /* number of subdisks to allocate to a plex */ + INITIAL_SUBDISKS_IN_DRIVE = 4, /* number of subdisks to allocate to a drive */ + INITIAL_DRIVE_FREELIST = 16, /* number of entries in drive freelist */ + PLEX_REGION_TABLE_SIZE = 8, /* number of entries in plex region tables */ + INITIAL_LOCKS = 8, /* number of locks to allocate to a volume */ + DEFAULT_REVIVE_BLOCKSIZE = 32768, /* size of block to transfer in one op */ +}; + +/* device numbers */ + +/* + * 31 30 28 27 20 19 18 16 15 8 7 0 + * |-----------------------------------------------------------------------------------------------| + * |X | Type | Subdisk number | X| Plex | Major number | volume number | + * |-----------------------------------------------------------------------------------------------| + * + * 0x2 03 1 19 06 + */ +struct devcode { +/* CARE. These fields assume a big-endian word. On a + * little-endian system, they're the wrong way around */ + unsigned volume:8; /* up to 256 volumes */ + unsigned major:8; /* this is where the major number fits */ + unsigned plex:3; /* up to 8 plexes per volume */ + unsigned unused:1; /* up for grabs */ + unsigned sd:8; /* up to 256 subdisks per plex */ + unsigned type:3; /* type of object */ + /* type field + VINUM_VOLUME = 0, + VINUM_PLEX = 1, + VINUM_SUBDISK = 2, + VINUM_DRIVE = 3, + VINUM_SUPERDEV = 4, */ + unsigned signbit:1; /* to make 32 bits */ +}; + +#define VINUM_DIR "/dev/vinum" +#define VINUM_RDIR "/dev/rvinum" +#define VINUM_SUPERDEV_NAME VINUM_DIR"/control" +#define MAXDRIVENAME 32 /* maximum length of a device name */ +#define MAXSDNAME 64 /* maximum length of a subdisk name */ +#define MAXPLEXNAME 64 /* maximum length of a plex name */ +#define MAXVOLNAME 64 /* maximum length of a volume name */ +#define MAXNAME 64 /* maximum length of any name */ +#define MAXVOLPLEX 8 /* maximum number of plexes in a volume */ + +/* Flags for all objects. Most of them only apply to + * specific objects, but we have space for all in any + * 32 bit flags word. */ +enum objflags { + VF_LOCKED = 1, /* somebody has locked access to this object */ + VF_LOCKING = 2, /* we want access to this object */ + VF_WRITETHROUGH = 8, /* volume: write through */ + VF_INITED = 0x10, /* unit has been initialized */ + VF_WLABEL = 0x20, /* label area is writable */ + VF_LABELLING = 0x40, /* unit is currently being labelled */ + VF_WANTED = 0x80, /* someone is waiting to obtain a lock */ + VF_RAW = 0x100, /* raw volume (no file system) */ + VF_LOADED = 0x200, /* module is loaded */ + VF_CONFIGURING = 0x400, /* somebody is changing the config */ + VF_WILL_CONFIGURE = 0x800, /* somebody wants to change the config */ + VF_CONFIG_INCOMPLETE = 0x1000, /* haven't finished changing the config */ + VF_CONFIG_SETUPSTATE = 0x2000, /* set a volume up if all plexes are empty */ + VF_READING_CONFIG = 0x4000, /* we're reading config database from disk */ + VF_KERNELOP = 0x8000, /* we're performing ops from kernel space */ +}; + +/* Global configuration information for the vinum subsystem */ +struct _vinum_conf { + /* Pointers to vinum structures */ + struct drive *drive; + struct sd *sd; + struct plex *plex; + struct volume *volume; + + /* the number allocated */ + int drives_allocated; + int subdisks_allocated; + int plexes_allocated; + int volumes_allocated; + + /* and the number currently in use */ + int drives_used; + int subdisks_used; + int plexes_used; + int volumes_used; + + int flags; + int opencount; /* number of times we've been opened */ +#if DEBUG + int lastrq; + struct buf *lastbuf; +#endif +}; + +/* Use these defines to simplify code */ +#define DRIVE vinum_conf.drive +#define SD vinum_conf.sd +#define PLEX vinum_conf.plex +#define VOL vinum_conf.volume +#define VFLAGS vinum_conf.flags + +/* Slice header + + * Vinum drives start with this structure: + * + * Sector + * |--------------------------------------| + * | PDP-11 memorial boot block | 0 + * |--------------------------------------| + * | Disk label, maybe | 1 + * |--------------------------------------| + * | Slice definition (vinum_hdr) | 2 + * |--------------------------------------| + * | | + * | Configuration info, first copy | 3 + * | | + * |--------------------------------------| + * | | + * | Configuration info, second copy | 3 + size of config + * | | + * |--------------------------------------| + */ + +/* Sizes and offsets of our information */ +enum { + VINUM_LABEL_OFFSET = 4096, /* offset of vinum label */ + VINUMHEADERLEN = 512, /* size of vinum label */ + VINUM_CONFIG_OFFSET = 4608, /* offset of first config copy */ + MAXCONFIG = 65536, /* and size of config copy */ + DATASTART = (MAXCONFIG * 2 + VINUM_CONFIG_OFFSET) / DEV_BSIZE /* this is where the data starts */ +}; + +/* hostname is 256 bytes long, but we don't need to shlep + * multiple copies in vinum. We use the host name just + * to identify this system, and 32 bytes should be ample + * for that purpose */ +#define VINUMHOSTNAMELEN 32 + +struct vinum_label { + char sysname[VINUMHOSTNAMELEN]; /* system name at time of creation */ + char name[MAXDRIVENAME]; /* our name of the drive */ + struct timeval date_of_birth; /* the time it was created */ + struct timeval last_update; /* and the time of last update */ + off_t drive_size; /* total size in bytes of the drive. + * This value includes the headers */ +}; + +struct vinum_hdr { + long long magic; /* we're long on magic numbers */ + /* XXX Get these right for big-endian */ +#define VINUM_MAGIC 22322600044678729LL /* should be this */ +#define VINUM_NOMAGIC 22322600044678990LL /* becomes this after obliteration */ + int config_length; /* size in bytes of each copy of the + * configuration info. + * This must be a multiple of the sector size. */ + + struct vinum_label label; /* unique label */ +}; + +/* Information returned from read_drive_label */ +enum drive_label_info { + DL_CANT_OPEN, /* invalid partition */ + DL_NOT_OURS, /* valid partition, but no vinum label */ + DL_DELETED_LABEL, /* valid partition, deleted label found */ + DL_WRONG_DRIVE, /* drive name doesn't match */ + DL_OURS /* valid partition and label found */ +}; + +/*** Drive definitions ***/ +/* A drive corresponds to a disk slice. We use a different term to show + * the difference in usage: it doesn't have to be a slice, and could + * theroretically be a complete, unpartitioned disk */ + +struct drive { + enum drivestate state; /* current state */ + int subdisks_allocated; /* number of entries in sd */ + int subdisks_used; /* and the number used */ + int blocksize; /* size of fs blocks */ + u_int64_t sectors_available; /* number of sectors still available */ + int secsperblock; + int lasterror; /* last error on drive */ + int driveno; /* index of drive in vinum_conf */ + int opencount; /* number of up subdisks */ + u_int64_t reads; /* number of reads on this drive */ + u_int64_t writes; /* number of writes on this drive */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + dev_t dev; /* and device number */ + char devicename[MAXDRIVENAME]; /* name of the slice it's on */ + struct vnode *vp; /* vnode pointer */ + struct proc *p; + struct vinum_label label; /* and the label information */ + struct partinfo partinfo; /* partition information */ + int freelist_size; /* number of entries alloced in free list */ + int freelist_entries; /* number of entries used in free list */ + struct drive_freelist { /* sorted list of free space on drive */ + u_int64_t offset; + long sectors; + } *freelist; +}; + +/*** Subdisk definitions ***/ + +struct sd { + enum sdstate state; /* state */ + /* offsets in blocks */ + int64_t driveoffset; /* offset on drive */ + int64_t plexoffset; /* offset in plex */ + u_int64_t sectors; /* and length in sectors */ + int plexno; /* index of plex, if it belongs */ + int driveno; /* index of the drive on which it is located */ + int sdno; /* our index in vinum_conf */ + int pid; /* pid of process which opened us */ + u_int64_t reads; /* number of reads on this subdisk */ + u_int64_t writes; /* number of writes on this subdisk */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + char name[MAXSDNAME]; /* name of subdisk */ +}; + +/*** Plex definitions ***/ + +/* kinds of plex organization */ +enum plexorg { + plex_disorg, /* disorganized */ + plex_concat, /* concatenated plex */ + plex_striped, /* striped plex */ + plex_raid5 /* RAID5 plex */ +}; + +/* Region in plex (either defective or unmapped) */ +struct plexregion { + u_int64_t offset; /* start of region */ + u_int64_t length; /* length */ +}; + +struct plex { + enum plexorg organization; /* Plex organization */ + enum plexstate state; /* and current state */ + u_int64_t length; /* total length of plex (max offset) */ + int flags; + int stripesize; /* size of stripe or raid band, in sectors */ + int subdisks; /* number of associated subdisks */ + int subdisks_allocated; /* number of subdisks allocated space for */ + int *sdnos; /* list of component subdisks */ + int plexno; /* index of plex in vinum_conf */ + int volno; /* index of volume */ + int volplexno; /* number of plex in volume */ + int pid; /* pid of process which opened us */ + /* Lock information */ + int locks; /* number of locks used */ + int alloclocks; /* number of locks allocated */ + struct rangelock *lock; /* ranges of locked addresses */ + /* Statistics */ + u_int64_t reads; /* number of reads on this plex */ + u_int64_t writes; /* number of writes on this plex */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + u_int64_t multiblock; /* requests that needed more than one block */ + u_int64_t multistripe; /* requests that needed more than one stripe */ + /* revive parameters */ + u_int64_t revived; /* block number of current revive request */ + int revive_blocksize; /* revive block size (bytes) */ + int revive_interval; /* and time to wait between transfers */ + struct request *waitlist; /* list of requests waiting on revive op */ + /* geometry control */ + int defective_regions; /* number of regions which are defective */ + int defective_region_count; /* number of entries in defective_region */ + struct plexregion *defective_region; /* list of offset/length pairs: defective sds */ + int unmapped_regions; /* number of regions which are missing */ + int unmapped_region_count; /* number of entries in unmapped_region */ + struct plexregion *unmapped_region; /* list of offset/length pairs: missing sds */ + char name[MAXPLEXNAME]; /* name of plex */ +}; + +/*** Volume definitions ***/ + +#define MAXPLEX 8 /* maximum number of plexes */ + + +struct volume { + enum volumestate state; /* current state */ + int plexes; /* number of plexes */ + int preferred_plex; /* plex to read from, -1 for round-robin */ + int last_plex_read; /* index of plex used for last read, + * for round-robin */ + dev_t devno; /* device number */ + int flags; /* status and configuration flags */ + int opencount; /* number of opens (all the same process) */ + int openflags; /* flags supplied to last open(2) */ + u_int64_t size; /* size of volume */ + int disk; /* disk index */ + int blocksize; /* logical block size */ + int active; /* number of outstanding requests active */ + int subops; /* and the number of suboperations */ + pid_t pid; /* pid of locker */ + /* Statistics */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + u_int64_t reads; /* number of reads on this volume */ + u_int64_t writes; /* number of writes on this volume */ + u_int64_t recovered_reads; /* reads recovered from another plex */ + /* Unlike subdisks in the plex, space for the plex pointers is static */ + int plex[MAXPLEX]; /* index of plexes */ + char name[MAXVOLNAME]; /* name of volume */ + struct disklabel label; /* for DIOCGPART */ +}; + +/* Table expansion. Expand table, which contains oldcount + * entries of type element, by increment entries, and change + * oldcount accordingly */ +#define EXPAND(table, element, oldcount, increment) \ +{ \ + expand_table ((void **) &table, \ + oldcount * sizeof (element), \ + (oldcount + increment) * sizeof (element) ); \ + oldcount += increment; \ + } + +/* Information on vinum's memory usage */ +struct meminfo { + int mallocs; /* number of malloced blocks */ + int total_malloced; /* total amount malloced */ + int highwater; /* maximum number of mallocs */ + struct mc *malloced; /* pointer to kernel table */ +}; + +struct mc { + int seq; + int size; + short line; + short flags; +#define ALLOC_KVA 1 /* allocated via kva calls */ + int *databuf; /* really vm_object_t */ + caddr_t address; + char file[16]; +}; + +/* These enums are used by the state transition + * routines. They're in bit map format: + * + * Bit 0: Other plexes in the volume are down + * Bit 1: Other plexes in the volume are up + * Bit 2: The current plex is up + * Maybe they should be local to + * state.c */ +enum volplexstate { + volplex_onlyusdown = 0, /* we're the only plex, and we're down */ + volplex_alldown, /* 1: another plex is down, and so are we */ + volplex_otherup, /* 2: another plex is up */ + volplex_otherupdown, /* other plexes are up and down */ + volplex_onlyus, /* 4: we're up and alone */ + volplex_onlyusup, /* only we are up, others are down */ + volplex_allup, /* all plexes are up */ + volplex_someup /* some plexes are up, including us */ +}; + +/* state map for plex */ +enum sdstates { + sd_emptystate = 1, + sd_downstate = 2, /* found an SD which is down */ + sd_crashedstate = 4, /* found an SD which is crashed */ + sd_obsoletestate = 8, /* found an SD which is obsolete */ + sd_stalestate = 16, /* found an SD which is stale */ + sd_rebornstate = 32, /* found an SD which is reborn */ + sd_upstate = 64, /* found an SD which is up */ + sd_initstate = 128, /* found an SD which is init */ + sd_otherstate = 256 /* found an SD in some other state */ +}; + +/* This is really just a parameter to pass to + * set_<foo>_state, but since it needs to be known + * in the external definitions, we need to define + * it here */ +enum setstateflags { + setstate_none = 0, /* no flags */ + setstate_force = 1, /* force the state change */ + setstate_configuring = 2, /* we're currently configuring, don't save */ + setstate_recursing = 4, /* we're called from another setstate function */ + setstate_norecurse = 8 /* don't call other setstate functions */ +}; + +#ifdef DEBUG +/* Debugging stuff */ +#define DEBUG_ADDRESSES 1 +#define DEBUG_NUMOUTPUT 2 +#endif diff --git a/sys/modules/vinum/COPYRIGHT b/sys/modules/vinum/COPYRIGHT new file mode 100644 index 0000000..4ee95f2 --- /dev/null +++ b/sys/modules/vinum/COPYRIGHT @@ -0,0 +1,37 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: COPYRIGHT,v 1.1 1998/03/05 06:07:05 grog Exp grog $ + */ diff --git a/sys/modules/vinum/Makefile b/sys/modules/vinum/Makefile new file mode 100644 index 0000000..4b88182 --- /dev/null +++ b/sys/modules/vinum/Makefile @@ -0,0 +1,26 @@ +# $Id: Makefile.lkm.lite,v 1.2 1998/08/13 06:07:29 grog Exp grog $ + +.PATH: ${.CURDIR}/../../sys/dev/ccd +KMOD= vinum_mod +SRCS= vinum.c vinum.h vnode_if.h parser.c config.c io.c util.c vinumhdr.h request.h \ + state.c memory.c request.c lock.c vinumext.h vinumio.h vinumkw.h \ + vinumstate.h vinumvar.h revive.c vinumioctl.c interrupt.c +NOMAN= +PSEUDO_LKM= +CFLAGS = -I. -O -g -I/usr/include/machine -DDEBUG -Wall -Wno-unused -Wno-parentheses + +CLEANFILES+= vinum.h vnode_if.h vnode_if.c + +all: + +# We don't need this, but the Makefile wants it +vinum.h: + touch $@ + +state.h: maketabs vinumstate.h + ./maketabs >state.h + +maketabs: maketabs.c + ${CC} -g -o maketabs maketabs.c + +.include <bsd.kmod.mk> diff --git a/sys/modules/vinum/config.c b/sys/modules/vinum/config.c new file mode 100644 index 0000000..fd31c5e --- /dev/null +++ b/sys/modules/vinum/config.c @@ -0,0 +1,1712 @@ +/* To do: + + * Don't store drive configuration on the config DB: read each drive's header + * to decide where it is. + * + * Accept any old crap in the config_<foo> functions, and complain when + * we try to bring it up. + * + * When trying to bring volumes up, check that the complete address range + * is covered. + */ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: config.c,v 1.17 1998/08/14 04:49:26 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#define REALLYKERNEL +#include "vinumhdr.h" + +extern jmp_buf command_fail; /* return on a failed command */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +void longjmp(jmp_buf, int); +#endif + +#define MAXTOKEN 64 /* maximum number of tokens in a line */ + +/* We can afford the luxury of global variables here, + * since start_config ensures that these functions + * are single-threaded. */ + +/* These are indices in vinum_conf of the last-mentioned of each kind of object */ +static int current_drive = -1; /* note the last drive we mention, for + * some defaults */ +static int current_plex = -1; /* and the same for the last plex */ +static int current_volume = -1; /* and the last volme */ +static struct _ioctl_reply *ioctl_reply; /* struct to return via ioctl */ + + +/* These values are used by most of these routines, so set them as globals */ +static char *token[MAXTOKEN]; /* pointers to individual tokens */ +static int tokens; /* number of tokens */ + +#define TOCONS 0x01 +#define TOTTY 0x02 +#define TOLOG 0x04 + +struct putchar_arg { + int flags; + struct tty *tty; +}; + +#define MSG_MAX 1024 /* maximum length of a formatted message */ +/* Format an error message and return to the user in the reply. + * CARE: This routine is designed to be called only from the + * configuration routines, so it assumes it's the owner of + * the configuration lock, and unlocks it on exit */ +void +throw_rude_remark(int error, char *msg,...) +{ + BROKEN_GDB; + int retval; + va_list ap; + char *text; + static int finishing; /* don't recurse */ + int was_finishing; + + va_start(ap, msg); + if ((ioctl_reply != NULL) /* we're called from the user */ + &&(!(vinum_conf.flags & VF_KERNELOP))) { /* and we're not doing kernel things: return msg */ + /* XXX We can't just format to ioctl_reply, since it + * may contain our input parameters */ + text = Malloc(MSG_MAX); + if (text == NULL) { + printf("vinum: can't allocate error message buffer"); + printf("vinum: "); + vprintf(msg, ap); /* print to the console */ + printf("\n"); + } else { + retval = kvprintf(msg, NULL, (void *) text, 10, ap); + text[retval] = '\0'; /* delimit */ + strcpy(ioctl_reply->msg, text); + ioctl_reply->error = error; /* first byte is the error number */ + Free(text); + } + } else { + printf("vinum: "); + vprintf(msg, ap); /* print to the console */ + printf("\n"); + } + va_end(ap); + + if (vinum_conf.flags & VF_READING_CONFIG) /* go through to the bitter end, */ + return; + /* We have a problem here: we want to unlock the + * configuration, which implies tidying up, but + * if we find an error while tidying up, we could + * recurse for ever. Use this kludge to only try + * once */ + was_finishing = finishing; + finishing = 1; + finish_config(was_finishing); /* unlock anything we may be holding */ + finishing = was_finishing; + longjmp(command_fail, error); +} + +/* Function declarations */ +int atoi(char *); /* no atoi in the kernel */ + +/* Minimal version of atoi */ +int +atoi(char *s) +{ /* no atoi in the kernel */ + BROKEN_GDB; + int r = 0; + int sign = 1; + + while (((*s >= '0') && (*s <= '9')) || (*s == '-')) { + if (*s == '-') + sign = -sign; + else + r = r * 10 + (*s - '0'); + } + return r; +} + +/* Find index of volume in vinum_conf. Return the index + * if found, or -1 if not */ +int +volume_index(struct volume *vol) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.volumes_used; i++) + if (&VOL[i] == vol) + return i; + return -1; +} + +/* Find index of plex in vinum_conf. Return the index + * if found, or -1 if not */ +int +plex_index(struct plex *plex) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.plexes_used; i++) + if (&PLEX[i] == plex) + return i; + return -1; +} + +/* Find index of subdisk in vinum_conf. Return the index + * if found, or -1 if not */ +int +sd_index(struct sd *sd) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.subdisks_used; i++) + if (&SD[i] == sd) + return i; + return -1; +} + +/* Find index of drive in vinum_conf. Return the index + * if found, or -1 if not */ +int +drive_index(struct drive *drive) +{ + BROKEN_GDB; + int i; + + for (i = 0; i < vinum_conf.drives_used; i++) + if (&DRIVE[i] == drive) + return i; + return -1; +} + +/* Check a volume to see if the plex is already assigned to it. + * Return index in volume->plex, or -1 if not assigned */ +int +my_plex(int volno, int plexno) +{ + BROKEN_GDB; + int i; + struct volume *vol; + + vol = &VOL[volno]; /* point to volno */ + for (i = 0; i < vol->plexes; i++) + if (vol->plex[i] == plexno) + return i; + return -1; /* not found */ +} + +/* Check a plex to see if the subdisk is already assigned to it. + * Return index in plex->sd, or -1 if not assigned */ +int +my_sd(int plexno, int sdno) +{ + BROKEN_GDB; + int i; + struct plex *plex; + + plex = &PLEX[plexno]; + for (i = 0; i < plex->subdisks; i++) + if (plex->sdnos[i] == sdno) + return i; + return -1; /* not found */ +} + +/* Check that this operation is being done in the kernel. + * longjmp out if not. op the name of the operation. */ +void +checkkernel(char *op) +{ + BROKEN_GDB; + if (vinum_conf.flags & VF_KERNELOP == 0) + throw_rude_remark(EPERM, "Can't perform '%s' from user space", op); +} + +/* Add plex to the volume if possible */ +int +give_plex_to_volume(int volno, int plexno) +{ + BROKEN_GDB; + struct volume *vol; + + /* XXX It's not an error for the plex to already + * belong to the volume, but we need to check a + * number of things to make sure it's done right. + * Some day. */ + if (my_plex(volno, plexno) >= 0) + return plexno; /* that's it */ + + vol = &VOL[volno]; /* point to volume */ + if (vol->plexes == MAXPLEX) /* all plexes allocated */ + throw_rude_remark(ENOSPC, + "Too many plexes for volume %s", + vol->name); + vol->plex[vol->plexes] = plexno; /* this one */ + vol->plexes++; /* add another plex */ + PLEX[plexno].volno = volno; /* note the number of our volume */ + + return vol->plexes - 1; /* and return its index */ +} + +/* Add subdisk to a plex if possible */ +int +give_sd_to_plex(int plexno, int sdno) +{ + BROKEN_GDB; + int i; + struct plex *plex; + struct sd *sd; + + /* XXX It's not an error for the sd to already + * belong to the plex, but we need to check a + * number of things to make sure it's done right. + * Some day. */ + i = my_sd(plexno, sdno); + if (i >= 0) /* does it already belong to us? */ + return i; /* that's it */ + + plex = &PLEX[plexno]; /* point to the plex */ + sd = &SD[sdno]; /* and the subdisk */ + + /* Do we have an offset? Otherwise put it after the last one */ + if (sd->plexoffset < 0) { /* no offset specified */ + if (plex->subdisks > 0) { + struct sd *lastsd = &SD[plex->sdnos[plex->subdisks - 1]]; /* last subdisk */ + sd->plexoffset = lastsd->sectors + lastsd->plexoffset; /* take it */ + } else /* first subdisk */ + sd->plexoffset = 0; /* start at the beginning */ + } + plex->subdisks++; /* another entry */ + if (plex->subdisks >= plex->subdisks_allocated) /* need more space */ + EXPAND(plex->sdnos, int, plex->subdisks_allocated, INITIAL_SUBDISKS_IN_PLEX); + + /* XXX I'm not sure this makes any sense + * for anything except concatenated plexes, + * and it comes up with the wrong answer for + * RAID-5 plexes, but it's currently needed + * for the calculations. We'll adjust for + * RAID-5 in config_plex */ + if ((sd->sectors + sd->plexoffset) > plex->length) { /* gone beyond the end of the plex */ + plex->length = sd->sectors + sd->plexoffset; /* adjust the length */ + + if ((plex->volno >= 0) /* we have a volume */ + &&(plex->length > VOL[plex->volno].size)) /* and we're now the longest plex */ + VOL[plex->volno].size = plex->length; /* increase the size of the volume */ + } + /* We need to check that the subdisks don't overlap, + * but we can't do that until a point where we *must* + * know the size of all the subdisks. That's not + * here. But we need to sort them by offset */ + for (i = 0; i < plex->subdisks - 1; i++) { + if (sd->plexoffset < SD[plex->sdnos[i]].plexoffset) { /* it fits before this one */ + /* First move any remaining subdisks by one */ + int j; + + for (j = plex->subdisks - 1; j > i; j--) /* move up one at a time */ + plex->sdnos[j] = plex->sdnos[j - 1]; + plex->sdnos[i] = sdno; + return i; + } + } + + /* The plex doesn't have any subdisk with a larger + * offset. Insert it */ + plex->sdnos[i] = sdno; + return i; +} + +/* Add a subdisk to drive if possible. The pointer to the drive + * must already be stored in the sd structure, but the drive + * doesn't know about the subdisk yet. */ +static void +give_sd_to_drive(int sdno) +{ + BROKEN_GDB; + struct sd *sd; /* pointer to subdisk */ + struct drive *drive; /* and drive */ + int fe; /* index in free list */ + + sd = &SD[sdno]; /* point to sd */ + drive = &DRIVE[sd->driveno]; /* and drive */ + + if (drive->state != drive_up) /* not up */ + throw_rude_remark(EIO, "Drive %s is not accessible", drive->label.name); + else if (sd->sectors > drive->sectors_available) { /* too big, */ + sd->driveoffset = -1; /* don't be confusing */ + throw_rude_remark(ENOSPC, "No space for %s on %s", sd->name, drive->label.name); + } + drive->subdisks_used++; /* one more subdisk */ + + /* no offset specified, find one */ + if (sd->driveoffset < 0) { + for (fe = 0; fe < drive->freelist_entries; fe++) { + if (drive->freelist[fe].sectors >= sd->sectors) { /* it'll fit here */ + sd->driveoffset = drive->freelist[fe].offset; + if (sd->sectors == drive->freelist[fe].sectors) { /* used up the entire entry */ + if (fe < (drive->freelist_entries - 1)) /* not the last one, */ + bcopy(&drive->freelist[fe + 1], + &drive->freelist[fe], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries--; /* one less entry */ + } else { + drive->freelist[fe].sectors -= sd->sectors; /* this much less space */ + drive->freelist[fe].offset += sd->sectors; /* this much further on */ + } + drive->sectors_available -= sd->sectors; /* and note how much less space we have */ + break; + } + } + if (fe == drive->freelist_entries) + /* Didn't find anything. Although the drive has + * enough space, it's too fragmented */ + { + sd->driveoffset = -1; /* don't be confusing */ + throw_rude_remark(ENOSPC, "No space for %s on %s", sd->name, drive->label.name); + } + } else { /* specific offset */ + /* For a specific offset to work, the space must be + * entirely in a single freelist entry. Look for it. */ + u_int64_t sdend = sd->driveoffset + sd->sectors; /* end of our subdisk */ + for (fe = 0; fe < drive->freelist_entries; fe++) { + u_int64_t dend = drive->freelist[fe].offset + drive->freelist[fe].sectors; /* end of entry */ + if (dend >= sdend) { /* fits before here */ + if (drive->freelist[fe].offset > sd->driveoffset) /* starts after the beginning of sd area */ + throw_rude_remark(ENOSPC, + "No space for subdisk %s on drive %s at offset %qd\n", + sd->name, + drive->label.name); + + /* We've found the space, and we can allocate it. + * We don't need to say that to the subdisk, which + * already knows about it. We need to tell it to + * the free list, though. We have four possibilities: + * + * 1. The subdisk exactly eats up the entry. That's the + * same as above. + * 2. The subdisk starts at the beginning and leaves space + * at the end. + * 3. The subdisk starts after the beginning and leaves + * space at the end as well: we end up with another + * fragment. + * 4. The subdisk leaves space at the beginning and finishes + * at the end. + */ + drive->sectors_available -= sd->sectors; /* note how much less space we have */ + if (sd->driveoffset == drive->freelist[fe].offset) { /* 1 or 2 */ + if (sd->sectors == drive->freelist[fe].sectors) { /* 1: used up the entire entry */ + if (fe < (drive->freelist_entries - 1)) /* not the last one, */ + bcopy(&drive->freelist[fe + 1], + &drive->freelist[fe], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries--; /* one less entry */ + } else { /* 2: space at the end */ + drive->freelist[fe].sectors -= sd->sectors; /* this much less space */ + drive->freelist[fe].offset += sd->sectors; /* this much further on */ + } + } else { /* 3 or 4 */ + drive->freelist[fe].sectors = sd->driveoffset - drive->freelist[fe].offset; + if (dend > sdend) { /* 3: space at the end as well */ + if (fe < (drive->freelist_entries - 1)) /* not the last one */ + bcopy(&drive->freelist[fe], /* move the rest down */ + &drive->freelist[fe + 1], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries++; /* one less entry */ + drive->freelist[fe + 1].offset = sdend; /* second entry starts after sd */ + drive->freelist[fe + 1].sectors = dend - sdend; /* and is this long */ + } + } + break; + } + } + } + drive->opencount++; /* one more subdisk attached */ +} + +/* Get an empty drive entry from the drive table */ +int +get_empty_drive(void) +{ + BROKEN_GDB; + int driveno; + struct drive *drive; + + /* first see if we have one which has been deallocated */ + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + if (DRIVE[driveno].state == drive_unallocated) /* bingo */ + break; + } + + if (driveno >= vinum_conf.drives_used) + /* Couldn't find a deallocated drive. Allocate a new one */ + { + vinum_conf.drives_used++; + if (vinum_conf.drives_used > vinum_conf.drives_allocated) /* we've used all our allocation */ + EXPAND(DRIVE, struct drive, vinum_conf.drives_allocated, INITIAL_DRIVES); + } + /* got a drive entry. Make it pretty */ + drive = &DRIVE[driveno]; + bzero(drive, sizeof(struct drive)); + drive->driveno = driveno; /* put number in structure */ + return driveno; /* return the index */ +} + +/* Find the named drive in vinum_conf.drive, return a pointer + * return the index in vinum_conf.drive. + * Don't mark the drive as allocated (XXX MP) + * If create != 0, create an entry if it doesn't exist + */ +/* XXX check if we have it open from attach */ +int +find_drive(const char *name, int create) +{ + BROKEN_GDB; + int driveno; + struct drive *drive; + + if (name != NULL) { + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + drive = &DRIVE[driveno]; /* point to drive */ + if ((drive->label.name[0] != '\0') /* it has a name */ + &&(strcmp(drive->label.name, name) == 0)) /* and it's this one: found */ + return driveno; + } + } + /* the drive isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + driveno = get_empty_drive(); + drive = &DRIVE[driveno]; + if (name != NULL) + bcopy(name, /* put in its name */ + drive->label.name, + min(sizeof(drive->label.name), + strlen(name))); + drive->state = drive_uninit; /* in use, nothing worthwhile there */ + return driveno; /* return the index */ +} + +/* Find a drive given its device name. + * devname must be valid. + * Otherwise the same as find_drive above */ +int +find_drive_by_dev(const char *devname, int create) +{ + BROKEN_GDB; + int driveno; + struct drive *drive; + + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + drive = &DRIVE[driveno]; /* point to drive */ + if ((drive->label.name[0] != '\0') /* it has a name */ + &&(strcmp(drive->label.name, devname) == 0)) /* and it's this one: found */ + return driveno; + } + + /* the drive isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + driveno = get_empty_drive(); + drive = &DRIVE[driveno]; + bcopy(devname, /* put in its name */ + drive->devicename, + min(sizeof(drive->devicename), + strlen(devname))); + drive->state = drive_uninit; /* in use, nothing worthwhile there */ + return driveno; /* return the index */ +} + +/* Find an empty subdisk in the subdisk table */ +int +get_empty_sd(void) +{ + BROKEN_GDB; + int sdno; + struct sd *sd; + + /* first see if we have one which has been deallocated */ + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) { + if (SD[sdno].state == sd_unallocated) /* bingo */ + break; + } + + if (sdno >= vinum_conf.subdisks_used) { /* No unused sd found. Allocate a new one */ + vinum_conf.subdisks_used++; + if (vinum_conf.subdisks_used > vinum_conf.subdisks_allocated) + EXPAND(SD, struct sd, vinum_conf.subdisks_allocated, INITIAL_SUBDISKS); + } + /* initialize some things */ + sd = &SD[sdno]; /* point to it */ + bzero(sd, sizeof(struct sd)); /* initialize */ + sd->plexno = -1; /* no plex */ + sd->driveno = -1; /* and no drive */ + sd->plexoffset = -1; /* and no offsets */ + sd->driveoffset = -1; + return sdno; /* return the index */ +} + +/* return a drive to the free pool */ +void +free_drive(struct drive *drive) +{ + BROKEN_GDB; + if (drive->vp != NULL) /* device open */ + vn_close(drive->vp, FREAD | FWRITE, FSCRED, drive->p); + bzero(drive, sizeof(struct drive)); /* this also sets drive_unallocated */ +} + +/* Find the named subdisk in vinum_conf.sd. + + * If create != 0, create an entry if it doesn't exist + * + * Return index in vinum_conf.sd + */ +int +find_subdisk(const char *name, int create) +{ + BROKEN_GDB; + int sdno; + struct sd *sd; + + for (sdno = 0; sdno < vinum_conf.subdisks_allocated; sdno++) { + if (strcmp(SD[sdno].name, name) == 0) /* found it */ + return sdno; + } + + /* the subdisk isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + /* Allocate one and insert the name */ + sdno = get_empty_sd(); + sd = &SD[sdno]; + bcopy(name, sd->name, min(sizeof(sd->name), strlen(name))); /* put in its name */ + return sdno; /* return the pointer */ +} + +/* Free an allocated sd entry + * This performs memory management only. remove() + * is responsible for checking relationships. + */ +void +free_sd(int sdno) +{ + BROKEN_GDB; + struct sd *sd; + struct drive *drive; + int fe; /* free list entry */ + u_int64_t sdend; /* end of our subdisk */ + u_int64_t dend; /* end of our freelist entry */ + + sd = &SD[sdno]; + if ((sd->driveno >= 0) /* we have a drive, */ + &&(sd->sectors > 0)) { /* and some space on it */ + drive = &DRIVE[sd->driveno]; + sdend = sd->driveoffset + sd->sectors; /* end of our subdisk */ + + /* Look for where to return the sd address space */ + for (fe = 0; + (fe < drive->freelist_entries) && (drive->freelist[fe].offset < sd->driveoffset); + fe++); + /* Now we are pointing to the last entry, the first + * with a higher offset than the subdisk, or both. */ + if ((fe > 1) /* not the first entry */ + &&((fe == drive->freelist_entries) /* gone past the end */ + ||(drive->freelist[fe].offset > sd->driveoffset))) /* or past the block were looking for */ + fe--; /* point to the block before */ + dend = drive->freelist[fe].offset + drive->freelist[fe].sectors; /* end of the entry */ + + /* At this point, we are pointing to the correct + * place in the free list. A number of possibilities + * exist: + * + * 1. The block to be freed immediately follows + * the block to which we are pointing. Just + * enlarge it. + * 2. The block to be freed starts at the end of + * the current block and ends at the beginning + * of the following block. Merge the three + * areas into a single block. + * 3. The block to be freed starts after the end + * of the block and ends before the start of + * the following block. Create a new free block. + * 4. The block to be freed starts after the end + * of the block, but ends at the start of the + * following block. Enlarge the following block + * downwards. + * + */ + if (sd->driveoffset == dend) { /* it starts after the end of this block */ + if ((fe < drive->freelist_entries - 1) /* we're not the last block in the free list */ + &&(sdend == drive->freelist[fe + 1].offset)) { /* and the subdisk ends at the start of the + * next block */ + drive->freelist[fe].sectors = drive->freelist[fe + 1].sectors; /* 2: merge all three blocks */ + if (fe < drive->freelist_entries - 2) /* still more blocks after next */ + bcopy(&drive->freelist[fe + 2], /* move down one */ + &drive->freelist[fe + 1], + (drive->freelist_entries - 2 - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries--; /* one less entry in the free list */ + } else /* 1: just enlarge this block */ + drive->freelist[fe].sectors += sd->sectors; + } else { + if (sd->driveoffset > dend) /* it starts after this block */ + fe++; /* so look at the next block */ + if ((fe < drive->freelist_entries) /* we're not the last block in the free list */ + &&(sdend == drive->freelist[fe].offset)) { /* and the subdisk ends at the start of + * this block: case 4 */ + drive->freelist[fe].offset = sd->driveoffset; /* it starts where the sd was */ + drive->freelist[fe].sectors += sd->sectors; /* and it's this much bigger */ + } else { /* case 3: non-contiguous */ + if (fe < drive->freelist_entries) /* not after the last block, */ + bcopy(&drive->freelist[fe], /* move the rest up one entry */ + &drive->freelist[fe + 1], + (drive->freelist_entries - fe) * sizeof(struct drive_freelist)); + drive->freelist_entries++; /* one less entry */ + drive->freelist[fe].offset = sd->driveoffset; /* this entry represents the sd */ + drive->freelist[fe].sectors = sd->sectors; + } + } + drive->opencount--; /* one less subdisk attached */ + } + bzero(sd, sizeof(struct sd)); /* and clear it out */ + sd->state = sd_unallocated; +} + +/* Find an empty plex in the plex table */ +int +get_empty_plex(void) +{ + BROKEN_GDB; + int plexno; + struct plex *plex; /* if we allocate one */ + + /* first see if we have one which has been deallocated */ + for (plexno = 0; plexno < vinum_conf.plexes_used; plexno++) { + if (PLEX[plexno].state == plex_unallocated) /* bingo */ + break; /* and get out of here */ + } + + if (plexno >= vinum_conf.plexes_used) { + /* Couldn't find a deallocated plex. Allocate a new one */ + vinum_conf.plexes_used++; + if (vinum_conf.plexes_used > vinum_conf.plexes_allocated) + EXPAND(PLEX, struct plex, vinum_conf.plexes_allocated, INITIAL_PLEXES); + } + /* Found a plex. Give it an sd structure */ + plex = &PLEX[plexno]; /* this one is ours */ + bzero(plex, sizeof(struct plex)); /* polish it up */ + plex->sdnos = (int *) Malloc(sizeof(int) * INITIAL_SUBDISKS_IN_PLEX); /* allocate sd table */ + CHECKALLOC(plex->sdnos, "vinum: Can't allocate plex subdisk table"); + bzero(plex->sdnos, (sizeof(int) * INITIAL_SUBDISKS_IN_PLEX)); /* do we need this? */ + plex->subdisks = 0; /* no subdisks in use */ + plex->subdisks_allocated = INITIAL_SUBDISKS_IN_PLEX; /* and we have space for this many */ + plex->organization = plex_disorg; /* and it's not organized */ + plex->volno = -1; /* no volume yet */ + return plexno; /* return the index */ +} + +/* Find the named plex in vinum_conf.plex + + * If create != 0, create an entry if it doesn't exist + * return index in vinum_conf.plex + */ +int +find_plex(const char *name, int create) +{ + BROKEN_GDB; + int plexno; + struct plex *plex; + + for (plexno = 0; plexno < vinum_conf.plexes_allocated; plexno++) { + if (strcmp(PLEX[plexno].name, name) == 0) /* found it */ + return plexno; + } + + /* the plex isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + /* Allocate one and insert the name */ + plexno = get_empty_plex(); + plex = &PLEX[plexno]; /* point to it */ + bcopy(name, plex->name, min(sizeof(plex->name), strlen(name))); /* put in its name */ + return plexno; /* return the pointer */ +} + +/* Free an allocated plex entry + * and its associated memory areas */ +void +free_plex(int plexno) +{ + BROKEN_GDB; + struct plex *plex; + + plex = &PLEX[plexno]; + if (plex->sdnos) + Free(plex->sdnos); + if (plex->lock) + Free(plex->lock); + if (plex->defective_region) + Free(plex->defective_region); + if (plex->unmapped_region) + Free(plex->unmapped_region); + bzero(plex, sizeof(struct plex)); /* and clear it out */ + plex->state = plex_unallocated; +} + +/* Find an empty volume in the volume table */ +int +get_empty_volume(void) +{ + BROKEN_GDB; + int volno; + struct volume *vol; + + /* first see if we have one which has been deallocated */ + for (volno = 0; volno < vinum_conf.volumes_used; volno++) { + if (VOL[volno].state == volume_unallocated) /* bingo */ + break; + } + + if (volno >= vinum_conf.volumes_used) + /* Couldn't find a deallocated volume. Allocate a new one */ + { + vinum_conf.volumes_used++; + if (vinum_conf.volumes_used > vinum_conf.volumes_allocated) + EXPAND(VOL, struct volume, vinum_conf.volumes_allocated, INITIAL_VOLUMES); + } + /* Now initialize fields */ + vol = &VOL[volno]; + bzero(vol, sizeof(struct volume)); + vol->preferred_plex = -1; /* default to round robin */ + vol->preferred_plex = ROUND_ROBIN_READPOL; /* round robin */ + + return volno; /* return the index */ +} + +/* Find the named volume in vinum_conf.volume. + + * If create != 0, create an entry if it doesn't exist + * return the index in vinum_conf + */ +int +find_volume(const char *name, int create) +{ + BROKEN_GDB; + int volno; + struct volume *vol; + + for (volno = 0; volno < vinum_conf.volumes_used; volno++) { + if (strcmp(VOL[volno].name, name) == 0) /* found it */ + return volno; + } + + /* the volume isn't in the list. Add it if he wants */ + if (create == 0) /* don't want to create */ + return -1; /* give up */ + + /* Allocate one and insert the name */ + volno = get_empty_volume(); + vol = &VOL[volno]; + bcopy(name, vol->name, min(sizeof(vol->name), strlen(name))); /* put in its name */ + vol->blocksize = DEV_BSIZE; /* block size of this volume */ + return volno; /* return the pointer */ +} + +/* Free an allocated volume entry + * and its associated memory areas */ +void +free_volume(int volno) +{ + BROKEN_GDB; + struct volume *vol; + + vol = &VOL[volno]; + bzero(vol, sizeof(struct volume)); /* and clear it out */ + vol->state = volume_unallocated; +} + +/* Handle a drive definition. We store the information in the global variable + * drive, so we don't need to allocate. + * + * If we find an error, print a message and return + */ +void +config_drive(void) +{ + BROKEN_GDB; + enum drive_label_info partition_status; /* info about the partition */ + int parameter; + int driveno; /* index of drive in vinum_conf */ + struct drive *drive; /* and pointer to it */ + + if (tokens < 2) /* not enough tokens */ + throw_rude_remark(EINVAL, "Drive has no name"); + driveno = find_drive(token[1], 1); /* allocate a drive to initialize */ + drive = &DRIVE[driveno]; /* and get a pointer */ + + if (drive->state != drive_uninit) { /* we already know this drive */ + /* XXX Check which definition is more up-to-date. Give + * preference for the definition on its own drive */ + return; /* XXX */ + } + for (parameter = 2; parameter < tokens; parameter++) { /* look at the other tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_device: + parameter++; + if (drive->devicename[0] != '\0') { /* we know this drive... */ + if (strcmp(drive->devicename, token[parameter])) /* different name */ + close_drive(drive); /* close it if it's open */ + else /* no change */ + break; + } + bcopy(token[parameter], /* insert device information */ + drive->devicename, + min(sizeof(drive->devicename), + strlen(token[parameter]))); + /* open the device and get the configuration */ + partition_status = read_drive_label(drive); + if (partition_status == DL_CANT_OPEN) { /* not our kind */ + close_drive(drive); + if (drive->lasterror == EFTYPE) /* wrong kind of partition */ + throw_rude_remark(drive->lasterror, + "Drive %s has invalid partition type", + drive->label.name); + else /* I/O error of some kind */ + throw_rude_remark(drive->lasterror, + "Can't initialize drive %s", + drive->label.name); + } else if (partition_status == DL_WRONG_DRIVE) { /* valid drive, not ours */ + close_drive(drive); + throw_rude_remark(drive->lasterror, + "Incorrect drive name %s specified for drive %s", + token[1], + drive->label.name); + } + break; + + case kw_state: + checkkernel(token[++parameter]); /* must be a kernel user */ + drive->state = DriveState(token[parameter]); /* set the state */ + break; + + default: + close_drive(drive); + throw_rude_remark(EINVAL, + "Drive %s, invalid keyword: %s", + token[1], + token[parameter]); + } + } + + if (drive->devicename[0] == '\0') + throw_rude_remark(EINVAL, "No device name for %s", drive->label.name); + +} + +/* Handle a subdisk definition. We store the information in the global variable + * sd, so we don't need to allocate. + * + * If we find an error, print a message and return + */ +void +config_subdisk(void) +{ + BROKEN_GDB; + int parameter; + int sdno; /* index of sd in vinum_conf */ + struct sd *sd; /* and pointer to it */ + u_int64_t size; + int sectors; /* sector offset value */ + int detached = 0; /* set to 1 if this is a detached subdisk */ + int sdindex = -1; /* index in plexes subdisk table */ + + sdno = get_empty_sd(); /* allocate an SD to initialize */ + sd = &SD[sdno]; /* and get a pointer */ + for (parameter = 1; parameter < tokens; parameter++) { /* look at the other tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_detached: + detached = 1; + break; + + case kw_plexoffset: + size = sizespec(token[++parameter]); + if ((size % DEV_BSIZE) != 0) + throw_rude_remark(EINVAL, "sd %s, bad plex offset alignment: %qd", sd->name, size); + else + sd->plexoffset = size / DEV_BSIZE; + break; + + case kw_driveoffset: + size = sizespec(token[++parameter]); + if ((size % DEV_BSIZE) != 0) + throw_rude_remark(EINVAL, "sd %s, bad drive offset alignment: %qd", sd->name, size); + else + sd->driveoffset = size / DEV_BSIZE; + break; + + case kw_name: + ++parameter; + bcopy(token[parameter], + sd->name, + min(sizeof(sd->name), strlen(token[parameter]))); + break; + + case kw_len: + size = sizespec(token[++parameter]); + if ((size % DEV_BSIZE) != 0) + throw_rude_remark(EINVAL, "sd %s, length %d not multiple of sector size", sd->name, size); + else + sd->sectors = size / DEV_BSIZE; + break; + + case kw_drive: + sd->driveno = find_drive(token[++parameter], 1); /* insert drive information */ + break; + + case kw_plex: + sd->plexno = find_plex(token[++parameter], 1); /* insert plex information */ + break; + + case kw_state: + checkkernel(token[++parameter]); /* must be a kernel user */ + sd->state = SdState(token[parameter]); /* set the state */ + break; + + default: + throw_rude_remark(EINVAL, "sd %s, invalid keyword: %s", sd->name, token[parameter]); + } + } + + /* Check we have a drive name */ + if (sd->driveno < 0) { /* didn't specify a drive */ + sd->driveno = current_drive; /* set to the current drive */ + if (sd->driveno < 0) /* no current drive? */ + throw_rude_remark(EINVAL, "Subdisk %s is not associated with a drive", sd->name); + } + /* Check for a plex name */ + if ((sd->plexno < 0) /* didn't specify a plex */ + &&(!detached)) /* and didn't say not to, */ + sd->plexno = current_plex; /* set to the current plex */ + + if (sd->plexno >= 0) + sdindex = give_sd_to_plex(sd->plexno, sdno); /* now tell the plex that it has this sd */ + + sd->sdno = sdno; /* point to our entry in the table */ + + /* Does the subdisk have a name? If not, give it one */ + if (sd->name[0] == '\0') { /* no name */ + char sdsuffix[8]; /* form sd name suffix here */ + + /* Do we have a plex name? */ + if (sdindex >= 0) /* we have a plex */ + strcpy(sd->name, PLEX[sd->plexno].name); /* take it from there */ + else /* no way */ + throw_rude_remark(EINVAL, "Unnamed sd is not associated with a plex"); + sprintf(sdsuffix, ".s%d", sdindex); /* form the suffix */ + strcat(sd->name, sdsuffix); /* and add it to the name */ + } + /* do we have complete info for this subdisk? */ + if (sd->sectors == 0) + throw_rude_remark(EINVAL, "sd %s has no length spec", sd->name); + + if (sd->state == sd_unallocated) /* no state decided, */ + sd->state = sd_init; /* at least we're in the game */ + + /* register the subdisk with the drive. This action + * will have the side effect of setting the offset if + * we haven't specified one, and causing an error + * message if it overlaps with another subdisk. */ + give_sd_to_drive(sdno); +} + +/* Handle a plex definition. + * If we find an error, print a message, deallocate the nascent plex, and return + */ +void +config_plex(void) +{ + BROKEN_GDB; + int parameter; + int plexno; /* index of plex in vinum_conf */ + struct plex *plex; /* and pointer to it */ + int pindex = MAXPLEX; /* index in volume's plex list */ + int detached = 0; /* don't give it to a volume */ + + current_plex = -1; /* forget the previous plex */ + plexno = get_empty_plex(); /* allocate a plex */ + plex = &PLEX[plexno]; /* and point to it */ + plex->plexno = plexno; /* and back to the config */ + for (parameter = 1; parameter < tokens; parameter++) { /* look at the other tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_detached: + detached = 1; + break; + + case kw_name: + { + int namedplexno; + + namedplexno = find_plex(token[++parameter], 0); /* find an existing plex with this name */ + if (namedplexno >= 0) + throw_rude_remark(EINVAL, "Duplicate plex %s", token[parameter]); + } + bcopy(token[parameter], /* put in the name */ + plex->name, + min(MAXPLEXNAME, strlen(token[parameter]))); + break; + + case kw_org: /* plex organization */ + switch (get_keyword(token[++parameter], &keyword_set)) { + case kw_concat: + plex->organization = plex_concat; + break; + + case kw_striped: + { + int stripesize = sizespec(token[++parameter]); + + plex->organization = plex_striped; + if (stripesize % DEV_BSIZE != 0) /* not a multiple of block size, */ + throw_rude_remark(EINVAL, "plex %s: stripe size %d not a multiple of sector size", + plex->name, + stripesize); + else + plex->stripesize = stripesize / DEV_BSIZE; + break; + } + + + default: + throw_rude_remark(EINVAL, "Invalid plex organization"); + } + if (((plex->organization == plex_striped) + ) + && (plex->stripesize == 0)) /* didn't specify a valid stripe size */ + throw_rude_remark(EINVAL, "Need a stripe size parameter"); + break; + + case kw_volume: + plex->volno = find_volume(token[++parameter], 1); /* insert a pointer to the volume */ + break; + + case kw_sd: /* add a subdisk */ + { + int sdno; + + sdno = find_subdisk(token[++parameter], 1); /* find a subdisk */ + SD[sdno].plexoffset = sizespec(token[++parameter]); /* get the offset */ + give_sd_to_plex(plexno, sdno); /* and insert it there */ + break; + } + + case kw_state: + checkkernel(token[++parameter]); /* only for kernel use */ + plex->state = PlexState(token[parameter]); /* set the state */ + break; + + default: + throw_rude_remark(EINVAL, "plex %s, invalid keyword: %s", + plex->name, + token[parameter]); + } + } + + if ((plex->volno < 0) /* we don't have a volume */ + &&(!detached)) /* and we wouldn't object */ + plex->volno = current_volume; + + if (plex->volno >= 0) + pindex = give_plex_to_volume(plex->volno, plexno); /* Now tell the volume that it has this plex */ + + /* Does the plex have a name? If not, give it one */ + if (plex->name[0] == '\0') { /* no name */ + char plexsuffix[8]; /* form plex name suffix here */ + /* Do we have a volume name? */ + if (plex->volno >= 0) /* we have a volume */ + strcpy(plex->name, /* take it from there */ + VOL[plex->volno].name); + else /* no way */ + throw_rude_remark(EINVAL, "Unnamed plex is not associated with a volume"); + sprintf(plexsuffix, ".p%d", pindex); /* form the suffix */ + strcat(plex->name, plexsuffix); /* and add it to the name */ + } + /* Note the last plex we configured */ + current_plex = plexno; + if (plex->state == plex_unallocated) /* we haven't changed the state, */ + plex->state = plex_init; /* we're initialized now */ +} + +/* Handle a volume definition. + * If we find an error, print a message, deallocate the nascent volume, and return + */ +void +config_volume(void) +{ + BROKEN_GDB; + int parameter; + int volno; + struct volume *vol; /* collect volume info here */ + int i; + + if (tokens < 2) /* not enough tokens */ + throw_rude_remark(EINVAL, "Volume has no name"); + current_volume = -1; /* forget the previous volume */ + volno = find_volume(token[1], 1); /* allocate a volume to initialize */ + vol = &VOL[volno]; /* and get a pointer */ + + for (parameter = 2; parameter < tokens; parameter++) { /* look at all tokens */ + switch (get_keyword(token[parameter], &keyword_set)) { + case kw_plex: + { + int plexno; /* index of this plex */ + + plexno = find_plex(token[++parameter], 1); /* find a plex */ + if (plexno < 0) /* couldn't */ + break; /* we've already had an error message */ + plexno = my_plex(volno, plexno); /* does it already belong to us? */ + if (plexno > 0) /* yes, shouldn't get it again */ + throw_rude_remark(EINVAL, + "Plex %s already belongs to volume %s", + token[parameter], + vol->name); + else if (++vol->plexes > 8) /* another entry */ + throw_rude_remark(EINVAL, + "Too many plexes for volume %s", + vol->name); + vol->plex[vol->plexes - 1] = plexno; + } + break; + + case kw_readpol: + switch (get_keyword(token[++parameter], &keyword_set)) { /* decide what to do */ + case kw_round: + vol->preferred_plex = ROUND_ROBIN_READPOL; /* default */ + break; + + case kw_prefer: + { + int myplexno; /* index of this plex */ + + myplexno = find_plex(token[++parameter], 1); /* find a plex */ + if (myplexno < 0) /* couldn't */ + break; /* we've already had an error message */ + myplexno = my_plex(volno, myplexno); /* does it already belong to us? */ + if (myplexno > 0) /* yes */ + vol->preferred_plex = myplexno; /* just note the index */ + else if (++vol->plexes > 8) /* another entry */ + throw_rude_remark(EINVAL, "Too many plexes"); + else { /* space for the new plex */ + vol->plex[vol->plexes - 1] = myplexno; /* add it to our list */ + vol->preferred_plex = vol->plexes - 1; /* and note the index */ + } + } + break; + + default: + throw_rude_remark(EINVAL, "Invalid read policy"); + } + + case kw_setupstate: + vol->flags |= VF_CONFIG_SETUPSTATE; /* set the volume up later on */ + break; + + case kw_state: + checkkernel(token[++parameter]); /* must be a kernel user */ + vol->state = VolState(token[parameter]); /* set the state */ + break; + + /* XXX experimental ideas. These are not + * documented, and will not be until I + * decide they're worth keeping */ + case kw_writethrough: /* set writethrough mode */ + vol->flags |= VF_WRITETHROUGH; + break; + + case kw_writeback: /* set writeback mode */ + vol->flags &= ~VF_WRITETHROUGH; + break; + + case kw_raw: + vol->flags |= VF_RAW; /* raw volume (no label) */ + break; + + default: + throw_rude_remark(EINVAL, "volume %s, invalid keyword: %s", + vol->name, + token[parameter]); + } + } + + current_volume = volno; /* note last referred volume */ + vol->devno = VINUMBDEV(volno, 0, 0, VINUM_VOLUME_TYPE); /* also note device number */ + + /* Before we can actually use the volume, we need + * a volume label. We could start to fake one here, + * but it will be a lot easier when we have some + * to copy from the drives, so defer it until we + * set up the configuration. XXX */ + if (vol->state == volume_unallocated) + vol->state = volume_down; /* now ready to bring up at the end */ + + /* Find out how big our volume is */ + for (i = 0; i < vol->plexes; i++) + vol->size = max(vol->size, PLEX[vol->plex[i]].length); +} + +/* Parse a config entry. CARE! This destroys the original contents of the + * config entry, which we don't really need after this. More specifically, it + * places \0 characters at the end of each token. + * + * Return 0 if all is well, otherwise EINVAL */ +int +parse_config(char *cptr, struct keywordset *keyset) +{ + BROKEN_GDB; + int status; + + status = 0; /* until proven otherwise */ + tokens = tokenize(cptr, token); /* chop up into tokens */ + + if (tokens <= 0) /* screwed up or empty line */ + return tokens; /* give up */ + + if (token[0][0] == '#') /* comment line */ + return 0; + + switch (get_keyword(token[0], keyset)) { /* decide what to do */ + case kw_read: /* read config from a specified drive */ + vinum_conf.flags |= VF_KERNELOP | VF_READING_CONFIG; /* kernel operation: reading config */ + status = check_drive(token[1]); /* check the drive info */ + vinum_conf.flags &= ~(VF_KERNELOP | VF_READING_CONFIG); + if (status != 0) { + char *msg = "Can't read configuration from %s"; + if (status == ENODEV) + msg = "No vinum configuration on %s"; + throw_rude_remark(status, msg, token[1]); + } + updateconfig(VF_KERNELOP); /* update from kernel space */ + break; + + case kw_drive: + config_drive(); + break; + + case kw_subdisk: + config_subdisk(); + break; + + case kw_plex: + config_plex(); + break; + + case kw_volume: + config_volume(); + break; + + /* Anything else is invalid in this context */ + default: + throw_rude_remark(EINVAL, /* should we die? */ + "Invalid configuration information: %s", + token[0]); + } + return status; +} + +/* parse a line handed in from userland via ioctl. + * This differs only by the error reporting mechanism: + * we return the error indication in the reply to the + * ioctl, so we need to set a global static pointer in + * this file. This technique works because we have + * ensured that configuration is performed in a single- + * threaded manner */ +int +parse_user_config(char *cptr, struct keywordset *keyset) +{ + BROKEN_GDB; + int status; + + ioctl_reply = (struct _ioctl_reply *) cptr; + status = parse_config(cptr, keyset); + ioctl_reply = NULL; /* don't do this again */ + return status; +} + +/* Remove an object */ +void +remove(struct vinum_ioctl_msg *msg) +{ + struct vinum_ioctl_msg message = *msg; /* make a copy to hand on */ + + ioctl_reply = (struct _ioctl_reply *) msg; /* reinstate the address to reply to */ + ioctl_reply->error = 0; /* no error, */ + ioctl_reply->msg[0] = '\0'; /* no message */ + + switch (message.type) { + case drive_object: + remove_drive_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + case sd_object: + remove_sd_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + case plex_object: + remove_plex_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + case volume_object: + remove_volume_entry(message.index, message.force, message.recurse); + updateconfig(0); + return; + + default: + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "Invalid object type"); + } +} + +/* Remove a drive. */ +void +remove_drive_entry(int driveno, int force, int recurse) +{ + struct drive *drive = &DRIVE[driveno]; + + if ((driveno > vinum_conf.drives_used) /* not a valid drive */ + ||(drive->state == drive_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such drive"); + } else if (drive->opencount > 0) { /* we have subdisks */ + if (force) { /* do it at any cost */ + int sdno; + struct vinum_ioctl_msg sdmsg; + + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) { + if ((SD[sdno].state != sd_unallocated) /* subdisk is allocated */ + &&(SD[sdno].driveno == driveno)) { /* and it belongs to this drive */ + sdmsg.type = sd_object; + sdmsg.recurse = 1; + sdmsg.force = force; + remove(&sdmsg); /* remove the subdisk by force */ + } + } + remove_drive(driveno); /* now remove it */ + } else + ioctl_reply->error = EBUSY; /* can't do that */ + } else + remove_drive(driveno); /* just remove it */ +} + +/* remove a subdisk */ +void +remove_sd_entry(int sdno, int force, int recurse) +{ + struct sd *sd = &SD[sdno]; + + if ((sdno > vinum_conf.subdisks_used) /* not a valid sd */ + ||(sd->state == sd_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such subdisk"); + } else if (sd->plexno >= 0) { /* we have a plex */ + if (force) { /* do it at any cost */ + struct plex *plex = &PLEX[sd->plexno]; /* point to our plex */ + int mysdno; + + for (mysdno = 0; /* look for ourselves */ + mysdno < plex->subdisks && &SD[plex->sdnos[mysdno]] != sd; + mysdno++); + if (mysdno == plex->subdisks) /* didn't find it */ + throw_rude_remark(ENOENT, "plex %s does not contain subdisk %s", plex->name, sd->name); + if (mysdno < (plex->subdisks - 1)) /* not the last subdisk */ + bcopy(&plex->sdnos[mysdno + 1], + &plex->sdnos[mysdno], + (plex->subdisks - 1 - mysdno) * sizeof(int)); + plex->subdisks--; + /* removing a subdisk from a striped or + * RAID-5 plex really tears the hell out + * of the structure, and it needs to be + * reinitialized */ + if (plex->organization != plex_concat) /* not concatenated, */ + set_plex_state(plex->plexno, plex_faulty, setstate_force); /* need to reinitialize */ + rebuild_plex_unmappedlist(plex); /* and see what remains */ + free_sd(sdno); + } else + ioctl_reply->error = EBUSY; /* can't do that */ + } else + free_sd(sdno); +} + +/* remove a plex */ +void +remove_plex_entry(int plexno, int force, int recurse) +{ + struct plex *plex = &PLEX[plexno]; + int sdno; + + if ((plexno > vinum_conf.plexes_used) /* not a valid plex */ + ||(plex->state == plex_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such plex"); + } else if (plex->pid) { /* we're open */ + ioctl_reply->error = EBUSY; /* no getting around that */ + return; + } + if (plex->subdisks) { + if (force) { /* do it anyway */ + if (recurse) { /* remove all below */ + for (sdno = 0; sdno < plex->subdisks; sdno++) + free_sd(plex->sdnos[sdno]); /* free all subdisks */ + } else { /* just tear them out */ + for (sdno = 0; sdno < plex->subdisks; sdno++) + SD[plex->sdnos[sdno]].plexno = -1; /* no plex any more */ + } + } else { /* can't do it without force */ + ioctl_reply->error = EBUSY; /* can't do that */ + return; + } + } + if (plex->volno >= 0) { /* we are part of a volume */ + /* XXX This should be more intelligent. We should + * be able to remove a plex as long as the volume + * does not lose any data, which is normally the + * case when it has more than one plex. To do it + * right we must compare the completeness of the + * mapping of all the plexes in the volume */ + if (force) { /* do it at any cost */ + struct volume *vol = &VOL[plex->volno]; + int myplexno; + + for (myplexno = 0; myplexno < vol->plexes; myplexno++) + if (vol->plex[myplexno] == plexno) /* found it */ + break; + if (myplexno == vol->plexes) /* didn't find it. Huh? */ + throw_rude_remark(ENOENT, "volume %s does not contain plex %s", vol->name, plex->name); + if (myplexno < (vol->plexes - 1)) /* not the last plex in the list */ + bcopy(&vol->plex[myplexno + 1], &vol->plex[myplexno], vol->plexes - 1 - myplexno); + vol->plexes--; + } else { + ioctl_reply->error = EBUSY; /* can't do that */ + return; + } + } + free_plex(plexno); +} + +/* remove a volume */ +void +remove_volume_entry(int volno, int force, int recurse) +{ + struct volume *vol = &VOL[volno]; + int plexno; + + if ((volno > vinum_conf.volumes_used) /* not a valid volume */ + ||(vol->state == volume_unallocated)) { /* or nothing there */ + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "No such volume"); + } else if (vol->opencount) /* we're open */ + ioctl_reply->error = EBUSY; /* no getting around that */ + else if (vol->plexes) { + if (recurse && force) { /* remove all below */ + struct vinum_ioctl_msg plexmsg; + + plexmsg.type = plex_object; + plexmsg.recurse = 1; + plexmsg.force = force; + for (plexno = 0; plexno < vol->plexes; plexno++) { + plexmsg.index = vol->plex[plexno]; /* plex number */ + remove(&plexmsg); + } + free_volume(volno); + } else + ioctl_reply->error = EBUSY; /* can't do that */ + } else + free_volume(volno); +} + +void +update_sd_config(int sdno, int kernelstate) +{ + if (!kernelstate) + set_sd_state(sdno, sd_up, setstate_configuring | setstate_norecurse); +} + +void +update_plex_config(int plexno, int kernelstate) +{ + int error = 0; + int size; + int sdno; + struct plex *plex = &PLEX[plexno]; + enum plexstate state = plex_up; /* state we want the plex in */ + + /* XXX Insert checks here for sparse plexes and volumes */ + + /* Check that our subdisks make sense. For + * striped and RAID5 plexes, we need at least + * two subdisks, and they must all be the same + * size */ + if (((plex->organization == plex_striped) + ) + && (plex->subdisks < 2)) { + error = 1; + printf("vinum: plex %s does not have at least 2 subdisks\n", plex->name); + if (!kernelstate) + set_plex_state(plexno, plex_down, setstate_force | setstate_configuring | setstate_norecurse); + } + size = 0; + for (sdno = 0; sdno < plex->subdisks; sdno++) { + if (((plex->organization == plex_striped) + ) + && (sdno > 0) + && (SD[plex->sdnos[sdno]].sectors != SD[plex->sdnos[sdno - 1]].sectors)) { + error = 1; + printf("vinum: plex %s must have equal sized subdisks\n", plex->name); + set_plex_state(plexno, plex_down, setstate_force | setstate_configuring | setstate_norecurse); + } + size += SD[plex->sdnos[sdno]].sectors; + } + + if (plex->subdisks) { /* plex has subdisks, calculate size */ + rebuild_plex_unmappedlist(plex); /* rebuild the unmapped list first */ + + plex->length = size; + } else { /* no subdisks, */ + plex->length = 0; /* no size */ + state = plex_down; /* take it down */ + } + if (!(kernelstate || error)) + set_plex_state(plexno, state, setstate_none | setstate_configuring | setstate_norecurse); +} + +void +update_volume_config(int volno, int kernelstate) +{ + struct volume *vol = &VOL[volno]; + struct plex *plex; + int plexno; + + if (vol->state != volume_unallocated) + /* Recalculate the size of the volume */ + { + vol->size = 0; + for (plexno = 0; plexno < vol->plexes; plexno++) { + plex = &PLEX[vol->plex[plexno]]; + vol->size = max(plex->length, vol->size); /* maximum size */ + plex->volplexno = plexno; /* note it in the plex */ + } + } + if (!kernelstate) /* try to bring it up */ + set_volume_state(volno, volume_up, setstate_configuring | setstate_norecurse); +} + +/* Update the global configuration. + * kernelstate is != 0 if we're reading in a config + * from disk. In this case, we don't try to + * bring the devices up, though we will bring + * them down if there's some error which got + * missed when writing to disk. + */ +void +updateconfig(int kernelstate) +{ + BROKEN_GDB; + int sdno; + int plexno; + int volno; + struct volume *vol; + struct plex *plex; + + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) + update_sd_config(sdno, kernelstate); + + for (plexno = 0; plexno < vinum_conf.plexes_used; plexno++) + update_plex_config(plexno, kernelstate); + + for (volno = 0; volno < vinum_conf.volumes_used; volno++) + update_volume_config(volno, kernelstate); + save_config(); +} + +/* Start manual changes to the configuration and lock out + * others who may wish to do so. + * XXX why do we need this and lock_config too? */ +int +start_config(void) +{ + int error; + + while ((vinum_conf.flags & VF_CONFIGURING) != 0) { + vinum_conf.flags |= VF_WILL_CONFIGURE; + if ((error = tsleep(&vinum_conf, PRIBIO | PCATCH, "vincfg", 0)) != 0) + return error; + } + /* We need two flags here: VF_CONFIGURING + * tells other processes to hold off (this + * function), and VF_CONFIG_INCOMPLETE + * tells the state change routines not to + * propagate incrememntal state changes */ + vinum_conf.flags |= VF_CONFIGURING | VF_CONFIG_INCOMPLETE; + current_drive = -1; /* reset the defaults */ + current_plex = -1; /* and the same for the last plex */ + current_volume = -1; /* and the last volme */ + return 0; +} + +/* Update the config if update is 1, and unlock + * it. We won't update the configuration if we + * are called in a recursive loop via throw_rude_remark. + */ +void +finish_config(int update) +{ + vinum_conf.flags &= ~VF_CONFIG_INCOMPLETE; /* we've finished our config */ + if (update) + updateconfig(0); /* so update things */ + else + updateconfig(1); /* do some updates only */ + vinum_conf.flags &= ~VF_CONFIGURING; /* and now other people can take a turn */ + if ((vinum_conf.flags & VF_WILL_CONFIGURE) != 0) { + vinum_conf.flags &= ~VF_WILL_CONFIGURE; + wakeup(&vinum_conf); + } +} diff --git a/sys/modules/vinum/interrupt.c b/sys/modules/vinum/interrupt.c new file mode 100644 index 0000000..e7eb034 --- /dev/null +++ b/sys/modules/vinum/interrupt.c @@ -0,0 +1,190 @@ +/* interrupt.c: bottom half of the driver */ + +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: interrupt.c,v 1.1 1998/08/13 06:12:27 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" +#include <miscfs/specfs/specdev.h> +#include <sys/resourcevar.h> + +void complete_raid5_write(struct rqelement *); +void freerq(struct request *rq); +void free_rqg(struct rqgroup *rqg); +void complete_rqe(struct buf *bp); +void sdio_done(struct buf *bp); + +/* Take a completed buffer, transfer the data back if + * it's a read, and complete the high-level request + * if this is the last subrequest. + * + * The bp parameter is in fact a struct rqelement, which + * includes a couple of extras at the end. + */ +void +complete_rqe(struct buf *bp) +{ + BROKEN_GDB; + struct rqelement *rqe; + struct request *rq; + struct rqgroup *rqg; + struct buf *ubp; /* user buffer */ + + rqe = (struct rqelement *) bp; /* point to the element element that completed */ + rqg = rqe->rqg; /* and the request group */ + rq = rqg->rq; /* and the complete request */ + + if ((bp->b_flags & B_ERROR) != 0) { /* transfer in error */ + if (bp->b_error != 0) /* did it return a number? */ + rq->error = bp->b_error; /* yes, put it in. */ + else if (rq->error == 0) /* no: do we have one already? */ + rq->error = EIO; /* no: catchall "I/O error" */ + if (rq->error == EIO) /* I/O error, */ + set_sd_state(rqe->sdno, sd_crashed, setstate_force); /* take the subdisk down */ + } + /* Now update the statistics */ + if (bp->b_flags & B_READ) { /* read operation */ + DRIVE[rqe->driveno].reads++; + DRIVE[rqe->driveno].bytes_read += bp->b_bcount; + SD[rqe->sdno].reads++; + SD[rqe->sdno].bytes_read += bp->b_bcount; + PLEX[rqe->rqg->plexno].reads++; + PLEX[rqe->rqg->plexno].bytes_read += bp->b_bcount; + } else { /* write operation */ + DRIVE[rqe->driveno].writes++; + DRIVE[rqe->driveno].bytes_written += bp->b_bcount; + SD[rqe->sdno].writes++; + SD[rqe->sdno].bytes_written += bp->b_bcount; + PLEX[rqe->rqg->plexno].writes++; + PLEX[rqe->rqg->plexno].bytes_written += bp->b_bcount; + } + ubp = rq->bp; /* user buffer */ + rqg->active--; /* one less request active */ + if (rqg->active == 0) /* request group finished, */ + rq->active--; /* one less */ + if (rq->active == 0) { /* request finished, */ +#if DEBUG + if (debug & 4) { + if (ubp->b_resid != 0) /* still something to transfer? */ + Debugger("resid"); + + { + int i; + for (i = 0; i < ubp->b_bcount; i += 512) /* XXX debug */ + if (((char *) ubp->b_data)[i] != '<') { /* and not what we expected */ + printf("At 0x%x (offset 0x%x): '%c' (0x%x)\n", + (int) (&((char *) ubp->b_data)[i]), + i, + ((char *) ubp->b_data)[i], + ((char *) ubp->b_data)[i]); + Debugger("complete_request checksum"); + } + } + } +#endif + + if (rq->error) { /* did we have an error? */ + ubp->b_flags |= B_ERROR; /* yes, propagate to user */ + ubp->b_error = rq->error; + } else + ubp->b_resid = 0; /* completed our transfer */ + if (rq->isplex == 0) /* volume request, */ + VOL[rq->volplex.volno].active--; /* another request finished */ + biodone(ubp); /* top level buffer completed */ + freerq(rq); /* return the request storage */ + } +} + + +/* Free a request block and anything hanging off it */ +void +freerq(struct request *rq) +{ + BROKEN_GDB; + struct rqgroup *rqg; + struct rqgroup *nrqg; /* next in chain */ + int rqno; + + for (rqg = rq->rqg; rqg != NULL; rqg = nrqg) { /* through the whole request chain */ + for (rqno = 0; rqno < rqg->count; rqno++) + if ((rqg->rqe[rqno].flags & XFR_MALLOCED) /* data buffer was malloced, */ + &&rqg->rqe[rqno].b.b_data) /* and the allocation succeeded */ + Free(rqg->rqe[rqno].b.b_data); /* free it */ + nrqg = rqg->next; /* note the next one */ + Free(rqg); /* and free this one */ + } + Free(rq); /* free the request itself */ +} + +void +free_rqg(struct rqgroup *rqg) +{ + if ((rqg->flags & XFR_GROUPOP) /* RAID 5 request */ + &&(rqg->rqe) /* got a buffer structure */ + &&(rqg->rqe->b.b_data)) /* and it has a buffer allocated */ + Free(rqg->rqe->b.b_data); /* free it */ +} + +/* I/O on subdisk completed */ +void +sdio_done(struct buf *bp) +{ + struct sdbuf *sbp; + + sbp = (struct sdbuf *) bp; + if (sbp->b.b_flags & B_ERROR) { /* had an error */ + bp->b_flags |= B_ERROR; + bp->b_error = sbp->b.b_error; + } + bp->b_resid = sbp->b.b_resid; + biodone(sbp->bp); /* complete the caller's I/O */ + /* Now update the statistics */ + if (bp->b_flags & B_READ) { /* read operation */ + DRIVE[sbp->driveno].reads++; + DRIVE[sbp->driveno].bytes_read += bp->b_bcount; + SD[sbp->sdno].reads++; + SD[sbp->sdno].bytes_read += bp->b_bcount; + } else { /* write operation */ + DRIVE[sbp->driveno].writes++; + DRIVE[sbp->driveno].bytes_written += bp->b_bcount; + SD[sbp->sdno].writes++; + SD[sbp->sdno].bytes_written += bp->b_bcount; + } + Free(sbp); +} diff --git a/sys/modules/vinum/io.c b/sys/modules/vinum/io.c new file mode 100644 index 0000000..a16a33a --- /dev/null +++ b/sys/modules/vinum/io.c @@ -0,0 +1,886 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: io.c,v 1.16 1998/08/10 23:47:21 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#if __FreeBSD__ < 3 /* this is in sys/disklabel.h in 3.0 and on */ +#define DTYPE_VINUM 12 /* vinum volume */ +#endif + +#define REALLYKERNEL +#include "vinumhdr.h" +#include <miscfs/specfs/specdev.h> + +extern jmp_buf command_fail; /* return on a failed command */ +struct _ioctl_reply *ioctl_reply; /* data pointer, for returning error messages */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +/* pointer to ioctl p parameter, to save passing it around */ +extern struct proc *myproc; + +/* Open the device associated with the drive, and set drive's vp */ +int +open_drive(struct drive *drive, struct proc *p) +{ + BROKEN_GDB; + struct nameidata nd; + struct vattr va; + int error; + + if (drive->devicename[0] == '\0') /* no device name */ + sprintf(drive->devicename, "/dev/%s", drive->label.name); /* get it from the drive name */ + NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, drive->devicename, p); + error = vn_open(&nd, FREAD | FWRITE, 0); /* open the device */ + if (error != 0) { /* can't open? */ + set_drive_state(drive->driveno, drive_down, 1); + drive->lasterror = error; + printf("vinum open_drive %s: failed with error %d\n", drive->devicename, error); /* XXX */ + return error; + } + drive->vp = nd.ni_vp; + drive->p = p; + + if (drive->vp->v_usecount > 1) { /* already in use? */ +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); + drive->lasterror = EBUSY; + printf("vinum open_drive %s: Drive in use\n", drive->devicename); /* XXX */ + return EBUSY; + } + error = VOP_GETATTR(drive->vp, &va, NOCRED, p); + if (error) { +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); + drive->lasterror = error; + printf("vinum open_drive %s: GETAATTR returns error %d\n", drive->devicename, error); /* XXX */ + return error; + } + drive->dev = va.va_rdev; /* device */ + + if (va.va_type != VBLK) { /* only consider block devices */ +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); /* this also closes the drive */ + drive->lasterror = ENOTBLK; + printf("vinum open_drive %s: Not a block device\n", drive->devicename); /* XXX */ + return ENOTBLK; + } + drive->vp->v_numoutput = 0; +#if __FreeBSD__ == 2 /* pre-4.4BSD Lite/2 parameters */ + VOP_UNLOCK(drive->vp); +#else + VOP_UNLOCK(drive->vp, 0, p); +#endif + return 0; +} + +/* Set some variables in the drive struct + * in more convenient form. Return error indication */ +int +set_drive_parms(struct drive *drive) +{ + drive->blocksize = BLKDEV_IOSIZE; /* XXX do we need this? */ + drive->secsperblock = drive->blocksize /* number of sectors per block */ + / drive->partinfo.disklab->d_secsize; + + /* Now update the label part */ + bcopy(hostname, drive->label.sysname, VINUMHOSTNAMELEN); /* put in host name */ +#if __FreeBSD__ >= 3 + getmicrotime(&drive->label.date_of_birth); /* and current time */ +#else + drive->label.date_of_birth = time; /* and current time */ +#endif + drive->label.drive_size = ((u_int64_t) drive->partinfo.part->p_size) /* size of the drive in bytes */ + *((u_int64_t) drive->partinfo.disklab->d_secsize); + + /* number of sectors available for subdisks */ + drive->sectors_available = drive->label.drive_size / DEV_BSIZE - DATASTART; + + /* XXX Bug in 3.0 as of January 1998: you can open + * non-existent slices. They have a length of 0 */ + if (drive->label.drive_size < MINVINUMSLICE) { /* too small to worry about */ + set_drive_state(drive->driveno, drive_down, 1); + printf("vinum open_drive %s: Drive too small\n", drive->devicename); /* XXX */ + drive->lasterror = ENOSPC; + return ENOSPC; + } + drive->freelist_size = INITIAL_DRIVE_FREELIST; /* initial number of entries */ + drive->freelist = (struct drive_freelist *) + Malloc(INITIAL_DRIVE_FREELIST * sizeof(struct drive_freelist)); + if (drive->freelist == NULL) /* can't malloc, dammit */ + return ENOSPC; + drive->freelist_entries = 1; /* just (almost) the complete drive */ + drive->freelist[0].offset = DATASTART; /* starts here */ + drive->freelist[0].sectors = (drive->label.drive_size >> DEV_BSHIFT) - DATASTART; /* and it's this long */ + set_drive_state(drive->driveno, drive_up, 1); /* our drive is accessible */ + return 0; +} + +/* Initialize a drive: open the device and add device + * information */ +int +init_drive(struct drive *drive) +{ + BROKEN_GDB; + int error; + + if (drive->devicename[0] == '\0') { /* no device name yet, default to drive name */ + drive->lasterror = EINVAL; + printf("vinum: Can't open drive without drive name\n"); /* XXX */ + return EINVAL; + } + error = open_drive(drive, myproc); /* open the drive */ + if (error) + return error; + + error = VOP_IOCTL(drive->vp, /* get the partition information */ + DIOCGPART, + (caddr_t) & drive->partinfo, + FREAD, + NOCRED, + myproc); + if (error) { + printf("vinum open_drive %s: Can't get partition information, error %d\n", + drive->devicename, + error); /* XXX */ + close_drive(drive); + drive->lasterror = error; + set_drive_state(drive->driveno, drive_down, 1); + return error; + } + if (drive->partinfo.part->p_fstype != 0) { /* not plain */ + drive->lasterror = EFTYPE; + printf("vinum open_drive %s: Wrong partition type for vinum\n", drive->devicename); /* XXX */ + close_drive(drive); + set_drive_state(drive->driveno, drive_down, 1); + return EFTYPE; + } + return set_drive_parms(drive); /* set various odds and ends */ +} + +/* Close a drive if it's open. No errors */ +void +close_drive(struct drive *drive) +{ + if (drive->vp) { + vn_close(drive->vp, FREAD | FWRITE, NOCRED, drive->p); + drive->vp = NULL; + } +} + +/* Remove drive from the configuration. + * Caller must ensure that it isn't active + */ +void +remove_drive(int driveno) +{ + BROKEN_GDB; + struct drive *drive = &vinum_conf.drive[driveno]; + long long int nomagic = VINUM_NOMAGIC; /* no magic number */ + + write_drive(drive, /* obliterate the magic, but leave a hint */ + (char *) &nomagic, + 8, + VINUM_LABEL_OFFSET); + close_drive(drive); /* and close it */ + drive->state = drive_unallocated; /* and forget everything we knew about it */ + save_config(); /* and save the updated configuration */ +} + +/* Transfer drive data. Usually called from one of these defines; + + * #define read_drive(a, b, c, d) driveio (a, b, c, d, B_READ) + * #define write_drive(a, b, c, d) driveio (a, b, c, d, B_WRITE) + * + * Return error number + */ +int +driveio(struct drive *drive, void *buf, size_t length, off_t offset, int flag) +{ + BROKEN_GDB; + int error; + struct buf *bp; + int spl; + + error = 0; + + /* Get a buffer */ + bp = (struct buf *) Malloc(sizeof(struct buf)); /* get a buffer */ + CHECKALLOC(bp, "Can't allocate memory"); + + bzero(&buf, sizeof(buf)); + bp->b_flags = B_BUSY | flag; /* tell us when it's done */ + bp->b_iodone = drive_io_done; /* here */ + bp->b_proc = myproc; /* process */ + bp->b_dev = drive->vp->v_un.vu_specinfo->si_rdev; /* device */ + if (offset & (drive->partinfo.disklab->d_secsize - 1)) /* not on a block boundary */ + bp->b_blkno = offset / drive->partinfo.disklab->d_secsize; /* block number */ + bp->b_data = buf; + bp->b_vp = drive->vp; /* vnode */ + bp->b_bcount = length; + bp->b_bufsize = length; + + (*bdevsw[major(bp->b_dev)]->d_strategy) (bp); /* initiate the transfer */ + + spl = splbio(); + while ((bp->b_flags & B_DONE) == 0) { + bp->b_flags |= B_CALL; /* wake me again */ + tsleep((caddr_t) bp, PRIBIO, "driveio", 0); /* and wait for it to complete */ + } + splx(spl); + if (bp->b_flags & B_ERROR) /* didn't work */ + error = bp->b_error; /* get the error return */ + Free(bp); /* then return the buffer */ + return error; +} + +/* Read data from a drive + + * Return error number + */ +int +read_drive(struct drive *drive, void *buf, size_t length, off_t offset) +{ + BROKEN_GDB; + int error; + struct buf *bp; + daddr_t nextbn; + long bscale; + + struct uio uio; + struct iovec iov; + daddr_t blocknum; /* block number */ + int blockoff; /* offset in block */ + int count; /* amount to transfer */ + + iov.iov_base = buf; + iov.iov_len = length; + + uio.uio_iov = &iov; + uio.uio_iovcnt = length; + uio.uio_offset = offset; + uio.uio_resid = length; + uio.uio_segflg = UIO_SYSSPACE; + uio.uio_rw = UIO_READ; + uio.uio_procp = myproc; + + bscale = btodb(drive->blocksize); /* mask off offset from block number */ + do { + blocknum = btodb(uio.uio_offset) & ~(bscale - 1); /* get the block number */ + blockoff = uio.uio_offset % drive->blocksize; /* offset in block */ + count = min((unsigned) (drive->blocksize - blockoff), /* amount to transfer in this block */ + uio.uio_resid); + + /* XXX Check this. I think the test is wrong */ + if (drive->vp->v_lastr + bscale == blocknum) { /* did our last read finish in this block? */ + nextbn = blocknum + bscale; /* note the end of the transfer */ + error = breadn(drive->vp, /* and read with read-ahead */ + blocknum, + (int) drive->blocksize, + &nextbn, + (int *) &drive->blocksize, + 1, + NOCRED, + &bp); + } else /* random read: just read this block */ + error = bread(drive->vp, blocknum, (int) drive->blocksize, NOCRED, &bp); + drive->vp->v_lastr = blocknum; /* note the last block we read */ + count = min(count, drive->blocksize - bp->b_resid); + if (error) { + brelse(bp); + return error; + } + error = uiomove((char *) bp->b_data + blockoff, count, &uio); /* move the data */ + brelse(bp); + } + while (error == 0 && uio.uio_resid > 0 && count != 0); + return error; +} + +/* Write data to a drive + + * Return error number + */ +int +write_drive(struct drive *drive, void *buf, size_t length, off_t offset) +{ + BROKEN_GDB; + int error; + struct buf *bp; + struct uio uio; + struct iovec iov; + daddr_t blocknum; /* block number */ + int blockoff; /* offset in block */ + int count; /* amount to transfer */ + int blockshift; + + if (drive->state == drive_down) /* currently down */ + return 0; /* ignore */ + if (drive->vp == NULL) { + drive->lasterror = ENODEV; + return ENODEV; /* not configured yet */ + } + iov.iov_base = buf; + iov.iov_len = length; + + uio.uio_iov = &iov; + uio.uio_iovcnt = length; + uio.uio_offset = offset; + uio.uio_resid = length; + uio.uio_segflg = UIO_SYSSPACE; + uio.uio_rw = UIO_WRITE; + uio.uio_procp = myproc; + + error = 0; + blockshift = btodb(drive->blocksize) - 1; /* amount to shift block number + * to get sector number */ + do { + blocknum = btodb(uio.uio_offset) & ~blockshift; /* get the block number */ + blockoff = uio.uio_offset % drive->blocksize; /* offset in block */ + count = min((unsigned) (drive->blocksize - blockoff), /* amount to transfer in this block */ + uio.uio_resid); + if (count == drive->blocksize) /* the whole block */ + bp = getblk(drive->vp, blocknum, drive->blocksize, 0, 0); /* just get it */ + else /* partial block: */ + error = bread(drive->vp, /* read it first */ + blocknum, + drive->blocksize, + NOCRED, + &bp); + count = min(count, drive->blocksize - bp->b_resid); /* how much will we transfer now? */ + if (error == 0) + error = uiomove((char *) bp->b_data + blockoff, /* move the data to the block */ + count, + &uio); + if (error) { + brelse(bp); + drive->lasterror = error; + switch (error) { + case EIO: + set_drive_state(drive->driveno, drive_down, 1); + break; + + /* XXX Add other possibilities here */ + default: + } + return error; + } + if (count + blockoff == drive->blocksize) + /* The transfer goes to the end of the block. There's + * no need to wait for any more data to arrive. */ + bawrite(bp); /* start the write now */ + else + bdwrite(bp); /* do a delayed write */ + } + while (error == 0 && uio.uio_resid > 0 && count != 0); + if (error) + drive->lasterror = error; + return error; /* OK */ +} + +/* Wake up on completion */ +void +drive_io_done(struct buf *bp) +{ + BROKEN_GDB; + wakeup((caddr_t) bp); /* Wachet auf! */ + bp->b_flags &= ~B_CALL; /* don't do this again */ +} + +/* Check a drive for a vinum header. If found, + * update the drive information. We come here + * with a partially populated drive structure + * which includes the device name. + * + * Return information on what we found + */ +enum drive_label_info +read_drive_label(struct drive *drive) +{ + BROKEN_GDB; + int error; + int result; /* result of our search */ + struct vinum_hdr *vhdr; /* and as header */ + + error = init_drive(drive); /* find the drive */ + if (error) /* find the drive */ + return DL_CANT_OPEN; /* not ours */ + + vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* allocate buffers */ + CHECKALLOC(vhdr, "Can't allocate memory"); + + error = read_drive(drive, (void *) vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); + if (vhdr->magic == VINUM_MAGIC) { /* ours! */ + if (drive->label.name[0] /* we have a name for this drive */ + &&(strcmp(drive->label.name, vhdr->label.name))) { /* but it doesn't match the real name */ + drive->lasterror = EINVAL; + result = DL_WRONG_DRIVE; /* it's the wrong drive */ + } else { + set_drive_parms(drive); /* and set other parameters */ + result = DL_OURS; + } + /* We copy the drive anyway so that we have + * the correct name in the drive info. This + * may not be the name specified */ + drive->label = vhdr->label; /* put in the label information */ + } else if (vhdr->magic == VINUM_NOMAGIC) /* was ours, but we gave it away */ + result = DL_DELETED_LABEL; + else + result = DL_NOT_OURS; /* we could have it, but we don't yet */ + Free(vhdr); /* that's all. */ + return result; +} + +/* Check a drive for a vinum header. If found, + * read configuration information from the drive and + * incorporate the data into the configuration. + * + * Return error number + */ +int +check_drive(char *drivename) +{ + BROKEN_GDB; + int error; + struct nameidata nd; /* mount point credentials */ + char *config_text; /* read the config info from disk into here */ + volatile char *cptr; /* pointer into config information */ + char *eptr; /* end pointer into config information */ + int driveno; + struct drive *drive; + char *config_line; /* copy the config line to */ + + driveno = find_drive_by_dev(drivename, 1); /* doesn't exist, create it */ + drive = &vinum_conf.drive[driveno]; /* and get a pointer */ + strcpy(drive->devicename, drivename); /* put in device name */ + + if (read_drive_label(drive) == DL_OURS) { /* ours! */ + config_text = (char *) Malloc(MAXCONFIG * 2); /* allocate buffers */ + CHECKALLOC(config_text, "Can't allocate memory"); + config_line = (char *) Malloc(MAXCONFIGLINE * 2); /* allocate buffers */ + CHECKALLOC(config_line, "Can't allocate memory"); + + /* Read in both copies of the configuration information */ + error = read_drive(drive, config_text, MAXCONFIG * 2, VINUM_CONFIG_OFFSET); + + if (error != 0) { + printf("vinum: Can't read device %s, error %d\n", drive->devicename, error); + Free(config_text); + Free(config_line); + free_drive(drive); /* give it back */ + return error; + } + /* XXX At this point, check that the two copies are the same, and do something useful if not. + * In particular, consider which is newer, and what this means for the integrity of the + * data on the drive */ + + /* Parse the configuration, and add it to the global configuration */ + for (cptr = config_text; *cptr != '\0';) { /* love this style(9) */ + volatile int parse_status; /* return value from parse_config */ + + for (eptr = config_line; (*cptr != '\n') && (*cptr != '\0');) /* until the end of the line */ + *eptr++ = *cptr++; + *eptr = '\0'; /* and delimit */ + if (setjmp(command_fail) == 0) { /* come back here on error and continue */ + parse_status = parse_config(config_line, &keyword_set); /* parse the config line */ + if (parse_status < 0) { /* error in config */ + /* This config should have been parsed in user + * space. If we run into problems here, something + * serious is afoot. Complain and let the user + * snarf the config to see what's wrong */ + printf("vinum: Config error on drive %s, aborting integration\n", nd.ni_dirp); + Free(config_text); + Free(config_line); + free_drive(drive); /* give it back */ + return EINVAL; + } + } + while (*cptr == '\n') + cptr++; /* skip to next line */ + } + Free(config_text); + if ((vinum_conf.flags & VF_READING_CONFIG) == 0) /* not reading config */ + updateconfig(0); /* update object states */ + printf("vinum: read configuration from %s\n", drivename); + return 0; /* it all worked */ + } else { /* no vinum label found */ + if (drive->lasterror) { + set_drive_state(drive->driveno, drive_down, 1); + return drive->lasterror; + } else + return ENODEV; /* not our device */ + } +} + +/* Kludge: kernel printf doesn't handle longs correctly XXX */ +static char *lltoa(long long l, char *s); +static char *sappend(char *txt, char *s); + +static char * +lltoa(long long l, char *s) +{ + if (l < 0) { + *s++ = '-'; + l = -l; + } + if (l > 9) { + s = lltoa(l / 10, s); + l %= 10; + } + *s++ = l + '0'; + return s; +} + +static char * +sappend(char *txt, char *s) +{ + while (*s++ = *txt++); + return s - 1; +} + +/* Format the configuration in text form into the buffer + * at config. Don't go beyond len bytes + * XXX this stinks. Fix soon. */ +void +format_config(char *config, int len) +{ + BROKEN_GDB; + int i; + int j; + char *s = config; + + bzero(config, len); + + /* First write the drive configuration */ + for (i = 0; i < vinum_conf.drives_used; i++) { + struct drive *drive; + + drive = &vinum_conf.drive[i]; + if (drive->state != drive_unallocated) { + sprintf(s, + "drive %s state %s device %s\n", + drive->label.name, + drive_state(drive->state), + drive->devicename); + while (*s) + s++; /* find the end */ + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } + + /* Then the volume configuration */ + for (i = 0; i < vinum_conf.volumes_used; i++) { + struct volume *vol; + + vol = &vinum_conf.volume[i]; + if (vol->state != volume_unallocated) { + if (vol->preferred_plex >= 0) /* preferences, */ + sprintf(s, + "volume %s state %s readpol prefer %s", + vol->name, + volume_state(vol->state), + vinum_conf.plex[vol->preferred_plex].name); + else /* default round-robin */ + sprintf(s, + "volume %s state %s", + vol->name, + volume_state(vol->state)); + while (*s) + s++; /* find the end */ + s = sappend("\n", s); + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } + + /* Then the plex configuration */ + for (i = 0; i < vinum_conf.plexes_used; i++) { + struct plex *plex; + + plex = &vinum_conf.plex[i]; + if (plex->state != plex_unallocated) { + sprintf(s, "plex name %s state %s org %s ", + plex->name, + plex_state(plex->state), + plex_org(plex->organization)); + while (*s) + s++; /* find the end */ + if ((plex->organization == plex_striped) + ) { + sprintf(s, "%db ", (int) plex->stripesize); + while (*s) + s++; /* find the end */ + } + if (plex->volno >= 0) /* we have a volume */ + sprintf(s, "vol %s ", vinum_conf.volume[plex->volno].name); + while (*s) + s++; /* find the end */ + for (j = 0; j < plex->subdisks; j++) { + sprintf(s, " sd %s", vinum_conf.sd[plex->sdnos[j]].name); + } + s = sappend("\n", s); + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } + + /* And finally the subdisk configuration */ + for (i = 0; i < vinum_conf.subdisks_used; i++) { + struct sd *sd = &vinum_conf.sd[i]; /* XXX */ + if (vinum_conf.sd[i].state != sd_unallocated) { + sprintf(s, + "sd name %s drive %s plex %s state %s len ", + sd->name, + vinum_conf.drive[sd->driveno].label.name, + vinum_conf.plex[sd->plexno].name, + sd_state(sd->state)); + while (*s) + s++; /* find the end */ + s = lltoa(sd->sectors, s); + s = sappend("b driveoffset ", s); + s = lltoa(sd->driveoffset, s); + s = sappend("b plexoffset ", s); + s = lltoa(sd->plexoffset, s); + s = sappend("b\n", s); + if (s > &config[len - 80]) { + printf("vinum: configuration data overflow\n"); + return; + } + } + } +} + +/* Write the configuration to all vinum slices */ +int +save_config(void) +{ + BROKEN_GDB; + int error; + int written_config; /* set when we firstnwrite the config to disk */ + int driveno; + struct drive *drive; /* point to current drive info */ + struct vinum_hdr *vhdr; /* and as header */ + char *config; /* point to config data */ + int wlabel_on; /* to set writing label on/off */ + + /* don't save the configuration while we're still working on it */ + if (vinum_conf.flags & VF_CONFIGURING) + return 0; + written_config = 0; /* no config written yet */ + /* Build a volume header */ + vhdr = (struct vinum_hdr *) Malloc(VINUMHEADERLEN); /* get space for the config data */ + CHECKALLOC(vhdr, "Can't allocate config data"); + vhdr->magic = VINUM_MAGIC; /* magic number */ + vhdr->config_length = MAXCONFIG; /* length of following config info */ + + config = Malloc(MAXCONFIG); /* get space for the config data */ + CHECKALLOC(config, "Can't allocate config data"); + + format_config(config, MAXCONFIG); + error = 0; /* no errors yet */ + for (driveno = 0; driveno < vinum_conf.drives_used; driveno++) { + drive = &vinum_conf.drive[driveno]; /* point to drive */ + + if (drive->state != drive_down) { +#if (__FreeBSD__ >= 3) + getmicrotime(&drive->label.last_update); /* time of last update is now */ +#else + drive->label.last_update = time; /* time of last update is now */ +#endif + bcopy((char *) &drive->label, /* and the label info from the drive structure */ + (char *) &vhdr->label, + sizeof(vhdr->label)); + if ((drive->state != drive_unallocated) + && (drive->state != drive_uninit)) { + wlabel_on = 1; /* enable writing the label */ + error = VOP_IOCTL(drive->vp, /* make the label writeable */ + DIOCWLABEL, + (caddr_t) & wlabel_on, + FWRITE, + NOCRED, + myproc); + if (error == 0) + error = write_drive(drive, vhdr, VINUMHEADERLEN, VINUM_LABEL_OFFSET); + if (error == 0) + error = write_drive(drive, config, MAXCONFIG, VINUM_CONFIG_OFFSET); + wlabel_on = 0; /* enable writing the label */ + VOP_IOCTL(drive->vp, /* make the label non-writeable again */ + DIOCWLABEL, + (caddr_t) & wlabel_on, + FWRITE, + NOCRED, + myproc); + if (error) { + printf("vinum: Can't write config to %s, error %d\n", drive->devicename, error); + set_drive_state(drive->driveno, drive_down, 1); + } else + written_config = 1; /* we've written it on at least one drive */ + } + } + } + Free(vhdr); + Free(config); + return written_config == 0; /* return 1 if we failed to write config */ +} + +/* Disk labels are a mess. The correct way to access them + * is with the DIOC[GSW]DINFO ioctls, but some programs, such + * as newfs, access the disk directly, so we have to write + * things there. We do this only on request. If a user + * request tries to read it directly, we fake up one on the fly. + */ + +/* get_volume_label returns a label structure to lp, which + * is allocated by the caller */ +void +get_volume_label(struct volume *vol, struct disklabel *lp) +{ + bzero(lp, sizeof(struct disklabel)); + + strncpy(lp->d_typename, "vinum", sizeof(lp->d_typename)); + lp->d_type = DTYPE_VINUM; + strncpy(lp->d_packname, vol->name, min(sizeof(lp->d_packname), sizeof(vol->name))); + lp->d_rpm = 14400 * vol->plexes; /* to keep them guessing */ + lp->d_interleave = 1; + lp->d_flags = 0; + + /* Fitting unto the vine, a vinum has a single + * track with all its sectors */ + lp->d_secsize = DEV_BSIZE; /* bytes per sector */ + lp->d_nsectors = vol->size; /* data sectors per track */ + lp->d_ntracks = 1; /* tracks per cylinder */ + lp->d_ncylinders = 1; /* data cylinders per unit */ + lp->d_secpercyl = vol->size; /* data sectors per cylinder */ + lp->d_secperunit = vol->size; /* data sectors per unit */ + + lp->d_bbsize = BBSIZE; + lp->d_sbsize = SBSIZE; + + lp->d_magic = DISKMAGIC; + lp->d_magic2 = DISKMAGIC; + + /* Set up partitions a, b and c to be identical + * and the size of the volume. a is UFS, b is + * swap, c is nothing */ + lp->d_partitions[0].p_size = vol->size; + lp->d_partitions[0].p_fsize = 1024; + lp->d_partitions[0].p_fstype = FS_BSDFFS; /* FreeBSD File System :-) */ + lp->d_partitions[0].p_fsize = 1024; /* FS fragment size */ + lp->d_partitions[0].p_frag = 8; /* and fragments per block */ + lp->d_partitions[SWAP_PART].p_size = vol->size; + lp->d_partitions[SWAP_PART].p_fstype = FS_SWAP; /* swap partition */ + lp->d_partitions[LABEL_PART].p_size = vol->size; + lp->d_npartitions = LABEL_PART + 1; + strncpy(lp->d_packname, vol->name, min(sizeof(lp->d_packname), sizeof(vol->name))); + lp->d_checksum = dkcksum(lp); +} + +int +write_volume_label(int volno) +{ + struct disklabel *lp; + struct buf *bp; + struct disklabel *dlp; + struct volume *vol; + int error; + + lp = (struct disklabel *) Malloc((sizeof(struct disklabel) + (DEV_BSIZE - 1)) & (DEV_BSIZE - 1)); + if (lp == 0) + return ENOMEM; + + if ((unsigned) (volno) >= (unsigned) vinum_conf.volumes_used) /* invalid volume */ + return ENOENT; + + vol = &VOL[volno]; /* volume in question */ + if (vol->state == volume_unallocated) /* nothing there */ + return ENOENT; + + get_volume_label(vol, lp); /* get the label */ + + /* Now write to disk. This code is derived from the + * system writedisklabel (), which does silly things + * like reading the label and refusing to write + * unless it's already there. */ + bp = geteblk((int) lp->d_secsize); /* get a buffer */ + bp->b_dev = minor(vol->devno) | (CDEV_MAJOR << MAJORDEV_SHIFT); /* our own raw volume */ + bp->b_blkno = LABELSECTOR * ((int) lp->d_secsize / DEV_BSIZE); + bp->b_bcount = lp->d_secsize; + bzero(bp->b_data, lp->d_secsize); + dlp = (struct disklabel *) bp->b_data; + *dlp = *lp; + bp->b_flags &= ~B_INVAL; + bp->b_flags |= B_BUSY | B_WRITE; + vinumstrategy(bp); /* write it out */ + error = biowait(bp); + bp->b_flags |= B_INVAL | B_AGE; + brelse(bp); + return error; +} + +/* Initialize a subdisk */ +int +initsd(int sdno) +{ + return 0; +} diff --git a/sys/modules/vinum/lock.c b/sys/modules/vinum/lock.c new file mode 100644 index 0000000..ed69649 --- /dev/null +++ b/sys/modules/vinum/lock.c @@ -0,0 +1,137 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: lock.c,v 1.6 1998/07/28 06:32:57 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" + +/* Lock routines. Currently, we lock either an individual volume + * or the global configuration. I don't think tsleep and + * wakeup are SMP safe. FIXME XXX */ + +/* Lock a volume, wait if it's in use */ +int +lockvol(struct volume *vol) +{ + int error; + + while ((vol->flags & VF_LOCKED) != 0) { + vol->flags |= VF_LOCKING; + /* It would seem to make more sense to sleep on + * the address 'vol'. Unfortuntaly we can't + * guarantee that this address won't change due to + * table expansion. The address we choose won't change. */ + if ((error = tsleep(&vinum_conf.volume + vol->devno, + PRIBIO | PCATCH, + "volock", + 0)) != 0) + return error; + } + vol->flags |= VF_LOCKED; + return 0; +} + +/* Unlock a volume and let the next one at it */ +void +unlockvol(struct volume *vol) +{ + vol->flags &= ~VF_LOCKED; + if ((vol->flags & VF_LOCKING) != 0) { + vol->flags &= ~VF_LOCKING; + wakeup(&vinum_conf.volume + vol->devno); + } +} + +/* Lock a plex, wait if it's in use */ +int +lockplex(struct plex *plex) +{ + int error; + + while ((plex->flags & VF_LOCKED) != 0) { + plex->flags |= VF_LOCKING; + /* It would seem to make more sense to sleep on + * the address 'plex'. Unfortuntaly we can't + * guarantee that this address won't change due to + * table expansion. The address we choose won't change. */ + if ((error = tsleep(&vinum_conf.plex + plex->sdnos[0], + PRIBIO | PCATCH, + "plexlk", + 0)) != 0) + return error; + } + plex->flags |= VF_LOCKED; + return 0; +} + +/* Unlock a plex and let the next one at it */ +void +unlockplex(struct plex *plex) +{ + plex->flags &= ~VF_LOCKED; + if ((plex->flags & VF_LOCKING) != 0) { + plex->flags &= ~VF_LOCKING; + wakeup(&vinum_conf.plex + plex->plexno); + } +} + + +/* Get a lock for the global config, wait if it's not available */ +int +lock_config(void) +{ + int error; + + while ((vinum_conf.flags & VF_LOCKED) != 0) { + vinum_conf.flags |= VF_LOCKING; + if ((error = tsleep(&vinum_conf, PRIBIO | PCATCH, "vincfg", 0)) != 0) + return error; + } + vinum_conf.flags |= VF_LOCKED; + return 0; +} + +/* Unlock and wake up any waiters */ +void +unlock_config(void) +{ + vinum_conf.flags &= ~VF_LOCKED; + if ((vinum_conf.flags & VF_LOCKING) != 0) { + vinum_conf.flags &= ~VF_LOCKING; + wakeup(&vinum_conf); + } +} diff --git a/sys/modules/vinum/makestatetext b/sys/modules/vinum/makestatetext new file mode 100755 index 0000000..ca5eb2d --- /dev/null +++ b/sys/modules/vinum/makestatetext @@ -0,0 +1,40 @@ +#!/bin/sh +# Make statetexts.h from vinumstate.h +# $Id: makestatetext,v 1.4 1998/03/13 05:36:16 grog Exp grog $ +infile=vinumstate.h +ofile=statetexts.h +cat <COPYRIGHT > $ofile + +echo >>$ofile "/* Created by $0 on" `date`. "Do not edit */" +echo >>$ofile +echo >>$ofile "/* Drive state texts */" +echo >>$ofile "char *drivestatetext [] = + { " +egrep -e 'drive_[A-z0-9]*,' <$infile | grep -v = | sed 's: *drive_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; + +/* Subdisk state texts */ +char *sdstatetext [] = + { +FOO +egrep -e 'sd_[A-z0-9]*,' $infile | grep -v = | sed 's: *sd_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; + +/* Plex state texts */ +char *plexstatetext [] = + { +FOO +egrep -e 'plex_[A-z0-9]*,' $infile | grep -v = | sed 's: *plex_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; + +/* Volume state texts */ +char *volstatetext [] = + { +FOO +egrep -e 'volume_[A-z0-9]*,' $infile | grep -v = | sed 's: *volume_\([^,]*\).*: \"\1\",:' >>$ofile +cat <<FOO >> $ofile + }; +FOO diff --git a/sys/modules/vinum/memory.c b/sys/modules/vinum/memory.c new file mode 100644 index 0000000..5dee671 --- /dev/null +++ b/sys/modules/vinum/memory.c @@ -0,0 +1,186 @@ + +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: memory.c,v 1.16 1998/08/08 04:43:22 grog Exp grog $ + */ + +#define REALLYKERNEL +#define USES_VM +#include "vinumhdr.h" + +extern jmp_buf command_fail; /* return on a failed command */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +void freedatabuf(struct mc *me); +caddr_t allocdatabuf(struct mc *me); + +void +expand_table(void **table, int oldsize, int newsize) +{ + if (newsize > oldsize) { + int *temp; + + temp = (int *) Malloc(newsize); /* allocate a new table */ + CHECKALLOC(temp, "vinum: Can't expand table\n"); + if (*table != NULL) { /* already something there, */ + bcopy((char *) *table, (char *) temp, oldsize); /* copy it to the old table */ + Free(*table); + } + *table = temp; + } +} + +#ifndef DEBUG +/* increase the size of a request block */ +void +expandrq(struct plexrq *prq) +{ + expand_table((void **) &prq->rqe, + prq->requests * sizeof(struct rqelement), + (prq->requests + RQELTS) * sizeof(struct rqelement)); + bzero(&prq->rqe[prq->requests], RQELTS * sizeof(struct rqelement)); /* clear the new part */ + prq->rqcount += RQELTS; +} + +#endif + +#if DEBUG /* XXX debug */ +#define MALLOCENTRIES 16384 +int malloccount = 0; +int highwater = 0; /* highest index ever allocated */ +static struct mc malloced[MALLOCENTRIES]; + +static total_malloced; + +caddr_t +MMalloc(int size, char *file, int line) +{ + caddr_t result; + int i; + static int seq = 0; + int s; + struct mc me; /* information to pass to allocdatabuf */ + + if (malloccount >= MALLOCENTRIES) { /* too many */ + printf("vinum: can't allocate table space to trace memory allocation"); + return 0; /* can't continue */ + } + result = malloc(size, M_DEVBUF, M_WAITOK); /* use malloc for smaller and irregular stuff */ + if (result == NULL) + printf("vinum: can't allocate %d bytes from %s:%d\n", size, file, line); + else { + me.flags = 0; /* allocation via malloc */ + s = splhigh(); + for (i = 0; i < malloccount; i++) { + if (((result + size) > malloced[i].address) + && (result < malloced[i].address + malloced[i].size)) /* overlap */ + Debugger("Malloc overlap"); + } + if (result) { + i = malloccount++; + total_malloced += size; + malloced[i].address = result; + malloced[i].size = size; + malloced[i].line = line; + malloced[i].seq = seq++; + malloced[i].flags = me.flags; + malloced[i].databuf = me.databuf; /* only used with kva alloc */ + bcopy(file, malloced[i].file, min(strlen(file) + 1, 16)); + } + if (malloccount > highwater) + highwater = malloccount; + splx(s); + } + return result; +} + +void +FFree(void *mem, char *file, int line) +{ + int i; + int s; + + s = splhigh(); + for (i = 0; i < malloccount; i++) { + if ((caddr_t) mem == malloced[i].address) { /* found it */ + bzero(mem, malloced[i].size); /* XXX */ + free(mem, M_DEVBUF); + malloccount--; + total_malloced -= malloced[i].size; + if (i < malloccount) /* more coming after */ + bcopy(&malloced[i + 1], &malloced[i], (malloccount - i) * sizeof(struct mc)); + splx(s); + return; + } + } + splx(s); + printf("Freeing unallocated data at 0x%08x from %s, line %d\n", (int) mem, file, line); + Debugger("Free"); +} + +void +vinum_meminfo(caddr_t data) +{ + struct meminfo *m = (struct meminfo *) data; + + m->mallocs = malloccount; + m->total_malloced = total_malloced; + m->malloced = malloced; + m->highwater = highwater; +} + +int +vinum_mallocinfo(caddr_t data) +{ + struct mc *m = (struct mc *) data; + unsigned int ent = *(int *) data; /* 1st word is index */ + + if (ent >= malloccount) + return ENOENT; + m->address = malloced[ent].address; + m->size = malloced[ent].size; + m->line = malloced[ent].line; + m->seq = malloced[ent].seq; + bcopy(malloced[ent].file, m->file, 16); + return 0; +} + +#endif diff --git a/sys/modules/vinum/parser.c b/sys/modules/vinum/parser.c new file mode 100644 index 0000000..9e7f791 --- /dev/null +++ b/sys/modules/vinum/parser.c @@ -0,0 +1,206 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: parser.c,v 1.11 1998/08/10 08:50:42 grog Exp grog $ + */ + +/* This file contains the parser for the configuration routines. It's used + * both in the kernel and in the user interface program, thus the separate file. */ + +/* Go through a text and split up into text tokens. These are either non-blank + * sequences, or any sequence (except \0) enclosed in ' or ". Embedded ' or + * " characters may be escaped by \, which otherwise has no special meaning. + * + * Delimit by following with a \0, and return pointers to the starts at token []. + * Return the number of tokens found as the return value. + * + * This method has the restriction that a closing " or ' must be followed by + * grey space. + * + * Error conditions are end of line before end of quote, or no space after + * a closing quote. In this case, tokenize() returns -1. */ + +#include <sys/param.h> +#ifdef KERNEL +#undef KERNEL /* XXX */ +#define REALLYKERNEL +#else +#include <ctype.h> +#include <errno.h> +#include <fcntl.h> +#endif +/* All this mess for a single struct definition */ +#include <sys/uio.h> +#include <sys/namei.h> +#include <sys/disklabel.h> +#include <sys/mount.h> +#include <sys/device.h> +#include <sys/disk.h> +#include "sys/buf.h" + +#include <vinumvar.h> +#include "vinumkw.h" +#include "vinumio.h" +#include "vinumext.h" + +#ifdef REALLYKERNEL +#define isspace(c) ((c == ' ') || (c == '\t')) /* check for white space */ +#else /* get it from the headers */ +#include <ctype.h> +#endif + +/* enum keyword is defined in vinumvar.h */ + +#define keypair(x) { #x, kw_##x } /* create pair "foo", kw_foo */ +#define flagkeypair(x) { "-"#x, kw_##x } /* create pair "-foo", kw_foo */ +#define KEYWORDSET(x) {sizeof (x) / sizeof (struct _keywords), x} + +/* Normal keywords. These are all the words that vinum knows. */ +struct _keywords keywords[] = +{keypair(drive), + keypair(sd), + keypair(subdisk), + keypair(plex), + keypair(volume), + keypair(vol), + keypair(setupstate), + keypair(readpol), + keypair(org), + keypair(name), + keypair(writethrough), + keypair(writeback), + keypair(raw), + keypair(device), + keypair(concat), + keypair(raid5), + keypair(striped), + keypair(plexoffset), + keypair(driveoffset), + keypair(length), + keypair(len), + keypair(state), + keypair(round), + keypair(prefer), + keypair(rename), + keypair(detached), +#ifndef KERNEL /* for vinum(8) only */ +#ifdef DEBUG + keypair(debug), +#endif + keypair(attach), + keypair(detach), + keypair(printconfig), + keypair(replace), + keypair(create), + keypair(read), + keypair(modify), + keypair(list), + keypair(l), + keypair(ld), + keypair(ls), + keypair(lp), + keypair(lv), + keypair(info), + keypair(set), + keypair(rm), + keypair(init), + keypair(label), + keypair(resetconfig), + keypair(start), + keypair(stop), + keypair(resetstats) +#endif +}; +struct keywordset keyword_set = KEYWORDSET(keywords); + +#ifndef KERNEL +struct _keywords flag_keywords[] = +{flagkeypair(f), + flagkeypair(d), + flagkeypair(v), + flagkeypair(s), + flagkeypair(r) +}; +struct keywordset flag_set = KEYWORDSET(flag_keywords); + +#endif + +int +tokenize(char *cptr, char *token[]) +{ + char delim; /* delimiter for searching for the partner */ + int tokennr; /* index of this token */ + tokennr = 0; /* none found yet */ + + for (;;) { + while (isspace(*cptr)) + cptr++; /* skip initial white space */ + if ((*cptr == '\0') || (*cptr == '\n') || (*cptr == '#')) /* end of line */ + return tokennr; /* return number of tokens found */ + delim = *cptr; + token[tokennr] = cptr; /* point to it */ + tokennr++; /* one more */ + /* XXX this is broken. It leaves superfluous \\ characters in the text */ + if ((delim == '\'') || (delim == '"')) { /* delimitered */ + for (;;) { + cptr++; + if ((*cptr == delim) && (cptr[-1] != '\\')) { /* found the partner */ + cptr++; /* move on past */ + if (!isspace(*cptr)) /* error, no space after closing quote */ + return -1; + *cptr++ = '\0'; /* delimit */ + } else if ((*cptr == '\0') || (*cptr == '\n')) /* end of line */ + return -1; + } + } else { /* not quoted */ + while ((*cptr != '\0') && (!isspace(*cptr)) && (*cptr != '\n')) + cptr++; + if (*cptr != '\0') /* not end of the line, */ + *cptr++ = '\0'; /* delimit and move to the next */ + } + } +} + +/* Find a keyword and return an index */ +enum keyword +get_keyword(char *name, struct keywordset *keywordset) +{ + int i; + struct _keywords *keywords = keywordset->k; /* point to the keywords */ + for (i = 0; i < keywordset->size; i++) + if (!strcmp(name, keywords[i].name)) + return (enum keyword) keywords[i].keyword; + return kw_invalid_keyword; +} diff --git a/sys/modules/vinum/request.c b/sys/modules/vinum/request.c new file mode 100644 index 0000000..589eb3f --- /dev/null +++ b/sys/modules/vinum/request.c @@ -0,0 +1,882 @@ +/* XXX to do: + + * Decide where we need splbio () + */ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: request.c,v 1.17 1998/08/13 06:04:47 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" +#include <miscfs/specfs/specdev.h> +#include <sys/resourcevar.h> + +/* pointer to ioctl p parameter, to save passing it around */ +extern struct proc *myproc; + +enum requeststatus bre(struct request *rq, + int plexno, + daddr_t * diskstart, + daddr_t diskend); +enum requeststatus bre5(struct request *rq, + int plexno, + daddr_t * diskstart, + daddr_t diskend); +enum requeststatus build_read_request(struct request *rq, int volplexno); +enum requeststatus build_write_request(struct request *rq); +enum requeststatus build_rq_buffer(struct rqelement *rqe, struct plex *plex); +void freerq(struct request *rq); +void free_rqg(struct rqgroup *rqg); +int find_alternate_sd(struct request *rq); +int check_range_covered(struct request *); +void complete_rqe(struct buf *bp); +void complete_raid5_write(struct rqelement *); +int abortrequest(struct request *rq, int error); +void sdio(struct buf *bp); +void sdio_done(struct buf *bp); +int vinum_bounds_check(struct buf *bp, struct volume *vol); +caddr_t allocdatabuf(struct rqelement *rqe); +void freedatabuf(struct rqelement *rqe); + +void +vinumstrategy(struct buf *bp) +{ + BROKEN_GDB; + int volno; + struct volume *vol = NULL; + int s; + struct devcode *device = (struct devcode *) &bp->b_dev; /* decode device number */ + enum requeststatus status; + + switch (device->type) { + case VINUM_SD_TYPE: + sdio(bp); + return; + + /* In fact, vinum doesn't handle drives: they're + * handled directly by the disk drivers */ + case VINUM_DRIVE_TYPE: + default: + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + biodone(bp); + return; + + case VINUM_VOLUME_TYPE: /* volume I/O */ + volno = VOLNO(bp->b_dev); + vol = &VOL[volno]; + if (vol->state != volume_up) { /* can't access this volume */ + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + biodone(bp); + return; + } + if (vinum_bounds_check(bp, vol) <= 0) { /* don't like them bounds */ + biodone(bp); /* have nothing to do with this */ + return; + } + /* FALLTHROUGH */ + /* Plex I/O is pretty much the same as volume I/O + * for a single plex. Indicate this by passing a NULL + * pointer (set above) for the volume */ + case VINUM_PLEX_TYPE: + bp->b_resid = bp->b_bcount; /* transfer everything */ + vinumstart(bp, 0); + return; + } +} + +/* Start a transfer. Return -1 on error, + * 0 if OK, 1 if we need to retry. + * Parameter reviveok is set when doing + * transfers for revives: it allows transfers to + * be started immediately when a revive is in + * progress. During revive, normal transfers + * are queued if they share address space with + * a currently active revive operation. */ +int +vinumstart(struct buf *bp, int reviveok) +{ + BROKEN_GDB; + int plexno; + int maxplex; /* maximum number of plexes to handle */ + struct volume *vol; + struct rqgroup *rqg; /* current plex's requests */ + struct rqelement *rqe; /* individual element */ + struct request *rq; /* build up our request here */ + int rqno; /* index in request list */ + enum requeststatus status; + + /* XXX In these routines, we're assuming that + * we will always be called with bp->b_bcount + * which is a multiple of the sector size. This + * is a reasonable assumption, since we are only + * called from system routines. Should we check + * anyway? */ + + if ((bp->b_bcount % DEV_BSIZE) != 0) { /* bad length */ + bp->b_error = EINVAL; /* invalid size */ + bp->b_flags |= B_ERROR; + biodone(bp); + return -1; + } + rq = (struct request *) Malloc(sizeof(struct request)); /* allocate a request struct */ + if (rq == NULL) { /* can't do it */ + bp->b_error = ENOMEM; /* can't get memory */ + bp->b_flags |= B_ERROR; + biodone(bp); + return -1; + } + bzero(rq, sizeof(struct request)); + + /* Note the volume ID. This can be NULL, which + * the request building functions use as an + * indication for single plex I/O */ + rq->bp = bp; /* and the user buffer struct */ + + if (DEVTYPE(bp->b_dev) == VINUM_VOLUME_TYPE) { /* it's a volume, */ + rq->volplex.volno = VOLNO(bp->b_dev); /* get the volume number */ + vol = &VOL[rq->volplex.volno]; /* and point to it */ + vol->active++; /* one more active request */ + maxplex = vol->plexes; /* consider all its plexes */ + } else { + vol = NULL; /* no volume */ + rq->volplex.plexno = PLEXNO(bp->b_dev); /* point to the plex */ + rq->isplex = 1; /* note that it's a plex */ + maxplex = 1; /* just the one plex */ + } + + if (bp->b_flags & B_READ) { + /* This is a read request. Decide + * which plex to read from. + * + * There's a potential race condition here, + * since we're not locked, and we could end + * up multiply incrementing the round-robin + * counter. This doesn't have any serious + * effects, however. */ + if (vol != NULL) { + vol->reads++; + vol->bytes_read += bp->b_bcount; + plexno = vol->preferred_plex; /* get the plex to use */ + if (plexno < 0) { /* round robin */ + plexno = vol->last_plex_read; + vol->last_plex_read++; + if (vol->last_plex_read == vol->plexes) /* got the the end? */ + vol->last_plex_read = 0; /* wrap around */ + } + status = build_read_request(rq, plexno); /* build a request */ + } else { + daddr_t diskaddr = bp->b_blkno; /* start offset of transfer */ + status = bre(rq, /* build a request list */ + rq->volplex.plexno, + &diskaddr, + diskaddr + (bp->b_bcount / DEV_BSIZE)); + } + + if ((status > REQUEST_RECOVERED) /* can't satisfy it */ + ||(bp->b_flags & B_DONE)) { /* XXX shouldn't get this without bad status */ + if (status == REQUEST_DOWN) { /* not enough subdisks */ + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + } + biodone(bp); + freerq(rq); + return -1; + } + return launch_requests(rq, reviveok); /* now start the requests if we can */ + } else + /* This is a write operation. We write to all + * plexes. If this is a RAID 5 plex, we must also + * update the parity stripe. */ + { + if (vol != NULL) { + vol->writes++; + vol->bytes_written += bp->b_bcount; + status = build_write_request(rq); /* Not all the subdisks are up */ + } else { /* plex I/O */ + daddr_t diskstart; + + diskstart = bp->b_blkno; /* start offset of transfer */ + status = bre(rq, + PLEXNO(bp->b_dev), + &diskstart, + bp->b_blkno + (bp->b_bcount / DEV_BSIZE)); /* build requests for the plex */ + } + if ((status > REQUEST_RECOVERED) /* can't satisfy it */ + ||(bp->b_flags & B_DONE)) { /* XXX shouldn't get this without bad status */ + if (status == REQUEST_DOWN) { /* not enough subdisks */ + bp->b_error = EIO; /* I/O error */ + bp->b_flags |= B_ERROR; + } + if ((bp->b_flags & B_DONE) == 0) + biodone(bp); + freerq(rq); + return -1; + } + return launch_requests(rq, reviveok); /* start the requests */ + } +} + +/* Call the low-level strategy routines to + * perform the requests in a struct request */ +int +launch_requests(struct request *rq, int reviveok) +{ + struct rqgroup *rqg; + int rqno; /* loop index */ + struct rqelement *rqe; /* current element */ + int s; + + /* First find out whether we're reviving, and the + * request contains a conflict. If so, we hang + * the request off plex->waitlist of the first + * plex we find which is reviving */ + if ((rq->flags & XFR_REVIVECONFLICT) /* possible revive conflict */ + &&(!reviveok)) { /* and we don't want to do it now, */ + struct volume *vol = &VOL[VOLNO(rq->bp->b_dev)]; + struct plex *plex; + int plexno; + + for (plexno = 0; plexno < vol->plexes; plexno++) { /* find the reviving plex */ + plex = &PLEX[vol->plex[plexno]]; + if (plex->state == plex_reviving) /* found it */ + break; + } + if (plexno < vol->plexes) { /* found it? */ + struct request *waitlist = plex->waitlist; /* point to the waiting list */ + + while (waitlist->next != NULL) /* find the end */ + waitlist = waitlist->next; + waitlist->next = rq; /* hook our request there */ + return 0; /* and get out of here */ + } else /* bad vinum, bad */ + printf("vinum: can't find reviving plex for volume %s\n", vol->name); + } + rq->active = 0; /* nothing yet */ + /* XXX This is probably due to a bug */ + if (rq->rqg == NULL) { /* no request */ + abortrequest(rq, EINVAL); + return -1; + } +#if DEBUG + if (debug & DEBUG_ADDRESSES) + printf("Request: %x\nWrite dev 0x%x, offset 0x%x, length %ld\n", + (u_int) rq, + rq->bp->b_dev, + rq->bp->b_blkno, + rq->bp->b_bcount); /* XXX */ + vinum_conf.lastrq = (int) rq; + vinum_conf.lastbuf = rq->bp; +#endif + for (rqg = rq->rqg; rqg != NULL; rqg = rqg->next) { /* through the whole request chain */ + rqg->active = rqg->count; /* they're all active */ + rq->active++; /* one more active request group */ + for (rqno = 0; rqno < rqg->count; rqno++) { + rqe = &rqg->rqe[rqno]; + if (rqe->flags & XFR_BAD_SUBDISK) /* this subdisk is bad, */ + rqg->active--; /* one less active request */ + else { + struct drive *drive = &DRIVE[rqe->driveno]; /* drive to access */ + if ((rqe->b.b_flags & B_READ) == 0) + rqe->b.b_vp->v_numoutput++; /* one more output going */ +#if DEBUG + if (debug & DEBUG_ADDRESSES) + printf(" %s dev 0x%x, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n", + rqe->b.b_flags & B_READ ? "Read" : "Write", + rqe->b.b_dev, + rqe->sdno, + (u_int) (rqe->b.b_blkno - SD[rqe->sdno].driveoffset), + rqe->b.b_blkno, + rqe->b.b_bcount); /* XXX */ + if (debug & DEBUG_NUMOUTPUT) + printf(" vinumstart sd %d numoutput %ld\n", + rqe->sdno, + rqe->b.b_vp->v_numoutput); +#endif + /* fire off the request */ + s = splbio(); + (*bdevsw[major(rqe->b.b_dev)]->d_strategy) (&rqe->b); + splx(s); + } + /* XXX Do we need caching? Think about this more */ + } + } + return 0; +} + +/* define the low-level requests needed to perform a + * high-level I/O operation for a specific plex 'plexno'. + * + * Return 0 if all subdisks involved in the request are up, 1 if some + * subdisks are not up, and -1 if the request is at least partially + * outside the bounds of the subdisks. + * + * Modify the pointer *diskstart to point to the end address. On + * read, return on the first bad subdisk, so that the caller + * (build_read_request) can try alternatives. + * + * On entry to this routine, the rqg structures are not assigned. The + * assignment is performed by expandrq(). Strictly speaking, the + * elements rqe->sdno of all entries should be set to -1, since 0 + * (from bzero) is a valid subdisk number. We avoid this problem by + * initializing the ones we use, and not looking at the others (index + * >= rqg->requests). + */ +enum requeststatus +bre(struct request *rq, + int plexno, + daddr_t * diskaddr, + daddr_t diskend) +{ + BROKEN_GDB; + int sdno; + struct sd *sd; + struct rqgroup *rqg; + struct buf *bp; /* user's bp */ + struct plex *plex; + enum requeststatus status; /* return value */ + daddr_t plexoffset; /* offset of transfer in plex */ + daddr_t stripebase; /* base address of stripe (1st subdisk) */ + daddr_t stripeoffset; /* offset in stripe */ + daddr_t blockoffset; /* offset in stripe on subdisk */ + struct rqelement *rqe; /* point to this request information */ + daddr_t diskstart = *diskaddr; /* remember where this transfer starts */ + + bp = rq->bp; /* buffer pointer */ + status = REQUEST_OK; /* return value: OK until proven otherwise */ + plex = &PLEX[plexno]; /* point to the plex */ + + switch (plex->organization) { + case plex_concat: + for (sdno = 0; sdno < plex->subdisks; sdno++) { + sd = &SD[plex->sdnos[sdno]]; + if ((*diskaddr < (sd->plexoffset + sd->sectors)) /* The request starts before the end of this */ + &&(diskend > sd->plexoffset)) { /* subdisk and ends after the start of this sd */ + if ((sd->state != sd_up) || (plex->state != plex_up)) { + enum requeststatus s; + + s = checksdstate(sd, rq, *diskaddr, diskend); /* do we need to change state? */ + if (s) /* give up? */ + return s; /* yup */ + } + rqg = allocrqg(rq, 1); /* space for the request */ + if (rqg == NULL) { /* malloc failed */ + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; + } + rqg->plexno = plexno; + + rqe = &rqg->rqe[0]; /* point to the element */ + rqe->rqg = rqg; /* group */ + rqe->sdno = sd->sdno; /* put in the subdisk number */ + plexoffset = max(sd->plexoffset, *diskaddr); /* start offset in plex */ + rqe->sdoffset = plexoffset - sd->plexoffset; /* start offset in subdisk */ + rqe->useroffset = plexoffset - diskstart; /* start offset in user buffer */ + rqe->dataoffset = 0; + rqe->datalen = min(diskend - *diskaddr, /* number of sectors to transfer in this sd */ + sd->sectors - rqe->sdoffset); + rqe->groupoffset = 0; /* no groups for concatenated plexes */ + rqe->grouplen = 0; + rqe->buflen = rqe->datalen; /* buffer length is data buffer length */ + rqe->flags = 0; + rqe->driveno = sd->driveno; + *diskaddr += rqe->datalen; /* bump the address */ + if (build_rq_buffer(rqe, plex)) { /* build the buffer */ + deallocrqg(rqg); + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; /* can't do it */ + } + } + if (*diskaddr > diskend) /* we're finished, */ + break; /* get out of here */ + } + break; + + case plex_striped: + { + while (*diskaddr < diskend) { /* until we get it all sorted out */ + /* The offset of the start address from + * the start of the stripe */ + stripeoffset = *diskaddr % (plex->stripesize * plex->subdisks); + + /* The plex-relative address of the + * start of the stripe */ + stripebase = *diskaddr - stripeoffset; + + /* The number of the subdisk in which + * the start is located */ + sdno = stripeoffset / plex->stripesize; + + /* The offset from the beginning of the stripe + * on this subdisk */ + blockoffset = stripeoffset % plex->stripesize; + + sd = &SD[plex->sdnos[sdno]]; /* the subdisk in question */ + if ((sd->state != sd_up) || (plex->state != plex_up)) { + enum requeststatus s; + + s = checksdstate(sd, rq, *diskaddr, diskend); /* do we need to change state? */ + if (s) /* give up? */ + return s; /* yup */ + } + rqg = allocrqg(rq, 1); /* space for the request */ + if (rqg == NULL) { /* malloc failed */ + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; + } + rqg->plexno = plexno; + + rqe = &rqg->rqe[0]; /* point to the element */ + rqe->rqg = rqg; + rqe->sdoffset = stripebase / plex->subdisks + blockoffset; /* start offset in this subdisk */ + rqe->useroffset = *diskaddr - diskstart; /* The offset of the start in the user buffer */ + rqe->dataoffset = 0; + rqe->datalen = min(diskend - *diskaddr, /* the amount remaining to transfer */ + plex->stripesize - blockoffset); /* and the amount left in this stripe */ + rqe->groupoffset = 0; /* no groups for striped plexes */ + rqe->grouplen = 0; + rqe->buflen = rqe->datalen; /* buffer length is data buffer length */ + rqe->flags = 0; + rqe->sdno = sd->sdno; /* put in the subdisk number */ + rqe->driveno = sd->driveno; + + if (rqe->sdoffset >= sd->sectors) { /* starts beyond the end of the subdisk? */ + deallocrqg(rqg); + return REQUEST_EOF; + } else if (rqe->sdoffset + rqe->datalen > sd->sectors) /* ends beyond the end of the subdisk? */ + rqe->datalen = sd->sectors - rqe->sdoffset; /* yes, truncate */ + + if (build_rq_buffer(rqe, plex)) { /* build the buffer */ + deallocrqg(rqg); + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return REQUEST_ENOMEM; /* can't do it */ + } + *diskaddr += rqe->datalen; /* look at the remainder */ + if (*diskaddr < diskend) { /* didn't finish the request on this stripe */ + plex->multiblock++; /* count another one */ + if (sdno == plex->subdisks - 1) /* last subdisk, */ + plex->multistripe++; /* another stripe as well */ + } + } + } + break; + + + default: + printf("vinum: invalid plex type in bre"); + } + + return status; +} + +/* Build up a request structure for reading volumes. + * This function is not needed for plex reads, since there's + * no recovery if a plex read can't be satisified. */ +enum requeststatus +build_read_request(struct request *rq, /* request */ + int plexindex) +{ /* index in the volume's plex table */ + BROKEN_GDB; + struct buf *bp; + daddr_t startaddr; /* offset of previous part of transfer */ + daddr_t diskaddr; /* offset of current part of transfer */ + daddr_t diskend; /* and end offset of transfer */ + int plexno; /* plex index in vinum_conf */ + struct rqgroup *rqg; /* point to the request we're working on */ + struct volume *vol; /* volume in question */ + off_t oldstart; /* note where we started */ + int recovered = 0; /* set if we recover a read */ + enum requeststatus status = REQUEST_OK; + + bp = rq->bp; /* buffer pointer */ + diskaddr = bp->b_blkno; /* start offset of transfer */ + diskend = diskaddr + (bp->b_bcount / DEV_BSIZE); /* and end offset of transfer */ + rqg = &rq->rqg[plexindex]; /* plex request */ + vol = &VOL[rq->volplex.volno]; /* point to volume */ + + while (diskaddr < diskend) { /* build up request components */ + startaddr = diskaddr; + status = bre(rq, vol->plex[plexindex], &diskaddr, diskend); /* build up a request */ + switch (status) { + case REQUEST_OK: + continue; + + case REQUEST_RECOVERED: + recovered = 1; + break; + + case REQUEST_EOF: + case REQUEST_ENOMEM: + return status; + + /* if we get here, we have either had a failure or + * a RAID 5 recovery. We don't want to use the + * recovery, because it's expensive, so first we + * check if we have alternatives */ + case REQUEST_DOWN: /* can't access the plex */ + if (vol != NULL) { /* and this is volume I/O */ + /* Try to satisfy the request + * from another plex */ + for (plexno = 0; plexno < vol->plexes; plexno++) { + diskaddr = startaddr; /* start at the beginning again */ + oldstart = startaddr; /* and note where that was */ + if (plexno != plexindex) { /* don't try this plex again */ + bre(rq, vol->plex[plexno], &diskaddr, diskend); /* try a request */ + if (diskaddr > oldstart) { /* we satisfied another part */ + recovered = 1; /* we recovered from the problem */ + status = REQUEST_OK; /* don't complain about it */ + break; + } + } + if (plexno == (vol->plexes - 1)) /* couldn't satisfy the request */ + return REQUEST_DOWN; /* failed */ + } + } else + return REQUEST_DOWN; /* bad luck */ + } + if (recovered) + vol->recovered_reads += recovered; /* adjust our recovery count */ + } + return status; +} + +/* Build up a request structure for writes. + * Return 0 if all subdisks involved in the request are up, 1 if some + * subdisks are not up, and -1 if the request is at least partially + * outside the bounds of the subdisks. */ +enum requeststatus +build_write_request(struct request *rq) +{ /* request */ + BROKEN_GDB; + struct buf *bp; + daddr_t diskstart; /* offset of current part of transfer */ + daddr_t diskend; /* and end offset of transfer */ + int plexno; /* plex index in vinum_conf */ + struct volume *vol; /* volume in question */ + enum requeststatus status; + + bp = rq->bp; /* buffer pointer */ + vol = &VOL[rq->volplex.volno]; /* point to volume */ + diskend = bp->b_blkno + (bp->b_bcount / DEV_BSIZE); /* end offset of transfer */ + status = REQUEST_OK; + for (plexno = 0; plexno < vol->plexes; plexno++) { + diskstart = bp->b_blkno; /* start offset of transfer */ + status = min(status, bre(rq, /* build requests for the plex */ + vol->plex[plexno], + &diskstart, + diskend)); + } + return status; +} + +/* Fill in the struct buf part of a request element. */ +enum requeststatus +build_rq_buffer(struct rqelement *rqe, struct plex *plex) +{ + BROKEN_GDB; + struct sd *sd; /* point to subdisk */ + struct volume *vol; + struct buf *bp; + struct buf *ubp; /* user (high level) buffer header */ + + vol = &VOL[rqe->rqg->rq->volplex.volno]; + sd = &SD[rqe->sdno]; /* point to subdisk */ + bp = &rqe->b; + ubp = rqe->rqg->rq->bp; /* pointer to user buffer header */ + + /* Initialize the buf struct */ + bzero(&rqe->b, sizeof(struct buf)); + bp->b_proc = ubp->b_proc; /* process pointer */ + bp->b_flags = ubp->b_flags & (B_NOCACHE | B_READ | B_ASYNC); /* copy these flags from user bp */ + bp->b_flags |= B_CALL | B_BUSY; /* inform us when it's done */ + if (plex->state == plex_reviving) + bp->b_flags |= B_ORDERED; /* keep request order if we're reviving */ + bp->b_iodone = complete_rqe; /* by calling us here */ + bp->b_dev = DRIVE[rqe->driveno].dev; /* drive device */ + bp->b_blkno = rqe->sdoffset + sd->driveoffset; /* start address */ + bp->b_bcount = rqe->buflen << DEV_BSHIFT; /* number of bytes to transfer */ + bp->b_resid = bp->b_bcount; /* and it's still all waiting */ + bp->b_bufsize = bp->b_bcount; /* and buffer size */ + bp->b_vp = DRIVE[rqe->driveno].vp; /* drive vnode */ + bp->b_rcred = FSCRED; /* we have the file system credentials */ + bp->b_wcred = FSCRED; /* we have the file system credentials */ + + if (rqe->flags & XFR_MALLOCED) { /* this operation requires a malloced buffer */ + bp->b_data = Malloc(bp->b_bcount); /* get a buffer to put it in */ + if (bp->b_data == NULL) { /* failed */ + Debugger("XXX"); + abortrequest(rqe->rqg->rq, ENOMEM); + return REQUEST_ENOMEM; /* no memory */ + } + } else + /* Point directly to user buffer data. This means + * that we don't need to do anything when we have + * finished the transfer */ + bp->b_data = ubp->b_data + rqe->useroffset * DEV_BSIZE; + return 0; +} +/* Abort a request: free resources and complete the + * user request with the specified error */ +int +abortrequest(struct request *rq, int error) +{ + struct buf *bp = rq->bp; /* user buffer */ + + bp->b_flags |= B_ERROR; + bp->b_error = error; + freerq(rq); /* free everything we're doing */ + biodone(bp); + return error; /* and give up */ +} + +/* Check that our transfer will cover the + * complete address space of the user request. + * + * Return 1 if it can, otherwise 0 */ +int +check_range_covered(struct request *rq) +{ + /* XXX */ + return 1; +} + +/* Perform I/O on a subdisk */ +void +sdio(struct buf *bp) +{ + int s; /* spl */ + struct sd *sd; + struct sdbuf *sbp; + daddr_t endoffset; + struct drive *drive; + + sd = &SD[SDNO(bp->b_dev)]; /* point to the subdisk */ + drive = &DRIVE[sd->driveno]; + + if (drive->state != drive_up) { /* XXX until we get the states fixed */ + set_sd_state(SDNO(bp->b_dev), sd_obsolete, setstate_force); + bp->b_flags |= B_ERROR; + bp->b_error = EIO; + biodone(bp); + return; + } + /* XXX decide which states we will really accept here. up + * implies it could be involved with a plex, in which + * case we don't want to dick with it */ + if ((sd->state != sd_up) + && (sd->state != sd_initializing) + && (sd->state != sd_reborn)) { /* we can't access it */ + bp->b_flags |= B_ERROR; + bp->b_flags = EIO; + if (bp->b_flags & B_BUSY) /* XXX why isn't this always the case? */ + biodone(bp); + return; + } + /* Get a buffer */ + sbp = (struct sdbuf *) Malloc(sizeof(struct sdbuf)); + if (sbp == NULL) { + bp->b_flags |= B_ERROR; + bp->b_error = ENOMEM; + biodone(bp); + return; + } + bcopy(bp, &sbp->b, sizeof(struct buf)); /* start with the user's buffer */ + sbp->b.b_flags |= B_CALL; /* tell us when it's done */ + sbp->b.b_iodone = sdio_done; /* here */ + sbp->b.b_dev = DRIVE[sd->driveno].dev; /* device */ + sbp->b.b_vp = DRIVE[sd->driveno].vp; /* vnode */ + sbp->b.b_blkno += sd->driveoffset; + sbp->bp = bp; /* note the address of the original header */ + sbp->sdno = sd->sdno; /* note for statistics */ + sbp->driveno = sd->driveno; + endoffset = bp->b_blkno + sbp->b.b_bcount / DEV_BSIZE; /* final sector offset */ + if (endoffset > sd->sectors) { /* beyond the end */ + sbp->b.b_bcount -= (endoffset - sd->sectors) * DEV_BSIZE; /* trim */ + if (sbp->b.b_bcount <= 0) { /* nothing to transfer */ + bp->b_resid = bp->b_bcount; /* nothing transferred */ + /* XXX Grrr. This doesn't seem to work. Return + * an error after all */ + bp->b_flags |= B_ERROR; + bp->b_error = ENOSPC; + biodone(bp); + Free(sbp); + return; + } + } + if ((sbp->b.b_flags & B_READ) == 0) /* write */ + sbp->b.b_vp->v_numoutput++; /* one more output going */ +#if DEBUG + if (debug & DEBUG_ADDRESSES) + printf(" %s dev 0x%x, sd %d, offset 0x%x, devoffset 0x%x, length %ld\n", + sbp->b.b_flags & B_READ ? "Read" : "Write", + sbp->b.b_dev, + sbp->sdno, + (u_int) (sbp->b.b_blkno - SD[sbp->sdno].driveoffset), + (int) sbp->b.b_blkno, + sbp->b.b_bcount); /* XXX */ + if (debug & DEBUG_NUMOUTPUT) + printf(" vinumstart sd %d numoutput %ld\n", + sbp->sdno, + sbp->b.b_vp->v_numoutput); +#endif + s = splbio(); + (*bdevsw[major(sbp->b.b_dev)]->d_strategy) (&sbp->b); + splx(s); +} + +/* Simplified version of bounds_check_with_label + * Determine the size of the transfer, and make sure it is + * within the boundaries of the partition. Adjust transfer + * if needed, and signal errors or early completion. + * + * Volumes are simpler than disk slices: they only contain + * one component (though we call them a, b and c to make + * system utilities happy), and they always take up the + * complete space of the "partition". + * + * I'm still not happy with this: why should the label be + * protected? If it weren't so damned difficult to write + * one in the first pleace (because it's protected), it wouldn't + * be a problem. + */ +int +vinum_bounds_check(struct buf *bp, struct volume *vol) +{ + int maxsize = vol->size; /* size of the partition (sectors) */ + int size = (bp->b_bcount + DEV_BSIZE - 1) >> DEV_BSHIFT; /* size of this request (sectors) */ + + /* Would this transfer overwrite the disk label? */ + if (bp->b_blkno <= LABELSECTOR /* starts before or at the label */ +#if LABELSECTOR != 0 + && bp->b_blkno + size > LABELSECTOR /* and finishes after */ +#endif + && (!(vol->flags & VF_RAW)) /* and it's not raw */ + &&major(bp->b_dev) == BDEV_MAJOR /* and it's the block device */ + && (bp->b_flags & B_READ) == 0 /* and it's a write */ + && (!vol->flags & (VF_WLABEL | VF_LABELLING))) { /* and we're not allowed to write the label */ + bp->b_error = EROFS; /* read-only */ + bp->b_flags |= B_ERROR; + return -1; + } + if (size == 0) /* no transfer specified, */ + return 0; /* treat as EOF */ + /* beyond partition? */ + if (bp->b_blkno < 0 /* negative start */ + || bp->b_blkno + size > maxsize) { /* or goes beyond the end of the partition */ + /* if exactly at end of disk, return an EOF */ + if (bp->b_blkno == maxsize) { + bp->b_resid = bp->b_bcount; + return 0; + } + /* or truncate if part of it fits */ + size = maxsize - bp->b_blkno; + if (size <= 0) { /* nothing to transfer */ + bp->b_error = EINVAL; + bp->b_flags |= B_ERROR; + return -1; + } + bp->b_bcount = size << DEV_BSHIFT; + } + bp->b_pblkno = bp->b_blkno; + return 1; +} + +/* Allocate a request group and hook + * it in in the list for rq */ +struct rqgroup * +allocrqg(struct request *rq, int elements) +{ + struct rqgroup *rqg; /* the one we're going to allocate */ + int size = sizeof(struct rqgroup) + elements * sizeof(struct rqelement); + + rqg = (struct rqgroup *) Malloc(size); + if (rqg != NULL) { /* malloc OK, */ + if (rq->rqg) /* we already have requests */ + rq->lrqg->next = rqg; /* hang it off the end */ + else /* first request */ + rq->rqg = rqg; /* at the start */ + rq->lrqg = rqg; /* this one is the last in the list */ + + bzero(rqg, size); /* no old junk */ + rqg->rq = rq; /* point back to the parent request */ + rqg->count = elements; /* number of requests in the group */ + } else + Debugger("XXX"); + return rqg; +} + +/* Deallocate a request group out of a chain. We do + * this by linear search: the chain is short, this + * almost never happens, and currently it can only + * happen to the first member of the chain. */ +void +deallocrqg(struct rqgroup *rqg) +{ + struct rqgroup *rqgc = rqg->rq->rqg; /* point to the request chain */ + + if (rqg->rq->rqg == rqg) /* we're first in line */ + rqg->rq->rqg = rqg->next; /* unhook ourselves */ + else { + while (rqgc->next != rqg) /* find the group */ + rqgc = rqgc->next; + rqgc->next = rqg->next; + } + Free(rqgc); +} + +/* Character device interface */ +int +vinumread(dev_t dev, struct uio *uio, int ioflag) +{ + return (physio(vinumstrategy, NULL, dev, 1, minphys, uio)); +} + +int +vinumwrite(dev_t dev, struct uio *uio, int ioflag) +{ + return (physio(vinumstrategy, NULL, dev, 0, minphys, uio)); +} diff --git a/sys/modules/vinum/request.h b/sys/modules/vinum/request.h new file mode 100644 index 0000000..b4beccc --- /dev/null +++ b/sys/modules/vinum/request.h @@ -0,0 +1,159 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: request.h,v 1.10 1998/08/03 07:15:26 grog Exp grog $ + */ + +/* Information needed to set up a transfer */ + +/* struct buf is surprisingly big (about 300 + * bytes), and it's part of the request, so this + * value is really important. Most requests + * don't need more than 2 subrequests per + * plex. The table is automatically extended if + * this value is too small. */ +#define RQELTS 2 /* default of 2 requests per transfer */ + +enum xferinfo { + XFR_NORMAL_READ = 1, + XFR_NORMAL_WRITE = 2, /* write request in normal mode */ + XFR_RECOVERY_READ = 4, + XFR_DEGRADED_WRITE = 8, + XFR_PARITYLESS_WRITE = 0x10, + XFR_NO_PARITY_STRIPE = 0x20, /* parity stripe is not available */ + XFR_DATA_BLOCK = 0x40, /* data block in request */ + XFR_PARITY_BLOCK = 0x80, /* parity block in request */ + XFR_BAD_SUBDISK = 0x100, /* this subdisk is dead */ + XFR_MALLOCED = 0x200, /* this buffer is malloced */ +#if DEBUG + XFR_PHASE2 = 0x800, /* documentation only: 2nd phase write */ +#endif + XFR_REVIVECONFLICT = 0x1000, /* possible conflict with a revive operation */ + /* operations that need a parity block */ + XFR_PARITYOP = (XFR_NORMAL_WRITE | XFR_RECOVERY_READ | XFR_DEGRADED_WRITE), + /* operations that use the group parameters */ + XFR_GROUPOP = (XFR_DEGRADED_WRITE | XFR_RECOVERY_READ), + /* operations that that use the data parameters */ + XFR_DATAOP = (XFR_NORMAL_READ | XFR_NORMAL_WRITE | XFR_PARITYLESS_WRITE), + /* operations requiring read before write */ + XFR_RBW = (XFR_NORMAL_WRITE | XFR_DEGRADED_WRITE), + /* operations that need a malloced buffer */ + XFR_NEEDS_MALLOC = (XFR_NORMAL_WRITE | XFR_RECOVERY_READ | XFR_DEGRADED_WRITE) +}; + +/* Describe one low-level request, part + * of a high-level request. This is an + * extended struct buf buffer, and the first + * element *must* be a struct buf. We pass this structure + * to the I/O routines instead of a struct buf in oder + * to be able to locate the high-level request when it + * completes. + * + * All offsets and lengths are in "blocks", i.e. sectors */ +struct rqelement { + struct buf b; /* buf structure */ + struct rqgroup *rqg; /* pointer to our group */ + /* Information about the transfer */ + daddr_t sdoffset; /* offset in subdisk */ + int useroffset; /* offset in user buffer of normal data */ + /* dataoffset and datalen refer to "individual" + * data transfers (normal read, parityless write) + * and also degraded write. + * + * groupoffset and grouplen refer to the other + * "group" operations (normal write, recovery read) + * Both the offsets are relative to the start of the + * local buffer */ + int dataoffset; /* offset in buffer of the normal data */ + int groupoffset; /* offset in buffer of group data */ + short datalen; /* length of normal data (sectors) */ + short grouplen; /* length of group data (sectors) */ + short buflen; /* total buffer length to allocate */ + short flags; /* really enum xferinfo (see above) */ + /* Ways to find other components */ + short sdno; /* subdisk number */ + short driveno; /* drive number */ +}; + +/* A group of requests built to satisfy a certain + * component of a user request */ +struct rqgroup { + struct rqgroup *next; /* pointer to next group */ + struct request *rq; /* pointer to the request */ + short count; /* number of requests in this group */ + short active; /* and number active */ + short plexno; /* index of plex */ + int badsdno; /* index of bad subdisk or -1 */ + enum xferinfo flags; /* description of transfer */ + struct rqelement rqe[0]; /* and the elements of this request */ +}; + +/* Describe one high-level request and the + * work we have to do to satisfy it */ +struct request { + struct buf *bp; /* pointer to the high-level request */ + int flags; + union { + int volno; /* volume index */ + int plexno; /* or plex index */ + } volplex; + int error; /* current error indication */ + short isplex; /* set if this is a plex request */ + short active; /* number of subrequests still active */ + struct rqgroup *rqg; /* pointer to the first group of requests */ + struct rqgroup *lrqg; /* and to the first group of requests */ + struct request *next; /* link of waiting requests */ +}; + +/* Extended buffer header for subdisk I/O. Includes + * a pointer to the user I/O request. */ +struct sdbuf { + struct buf b; /* our buffer */ + struct buf *bp; /* and pointer to parent */ + short driveno; /* drive index */ + short sdno; /* and subdisk index */ +}; + +/* Values returned by rqe and friends. + * Be careful with these: they are in order of increasing + * seriousness. Some routines check for > REQUEST_RECOVERED + * to indicate a completely failed request. */ +enum requeststatus { + REQUEST_OK, /* request built OK */ + REQUEST_RECOVERED, /* request OK, but involves RAID5 recovery */ + REQUEST_EOF, /* request failed: outside plex */ + REQUEST_DOWN, /* request failed: subdisk down */ + REQUEST_ENOMEM /* ran out of memory */ +}; diff --git a/sys/modules/vinum/revive.c b/sys/modules/vinum/revive.c new file mode 100644 index 0000000..2f8117a --- /dev/null +++ b/sys/modules/vinum/revive.c @@ -0,0 +1,128 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: revive.c,v 1.1 1998/08/14 06:16:59 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" + +/* revive a block of a plex. Return an error + * indication. EAGAIN means successful copy, but + * that more blocks remain to be copied. + * XXX We should specify a block size here. At the moment, + * just take a default value. FIXME */ +int +revive_block(int plexno) +{ + struct plex *plex = &PLEX[plexno]; + struct buf *bp; + int error = EAGAIN; + int size; + int s; /* priority level */ + + if (plex->revive_blocksize == 0) { + if (plex->stripesize != 0) /* we're striped, don't revive more than */ + plex->revive_blocksize = min(DEFAULT_REVIVE_BLOCKSIZE, plex->stripesize); /* one block at a time */ + else + plex->revive_blocksize = DEFAULT_REVIVE_BLOCKSIZE; + } + size = min(plex->revive_blocksize, plex->length - plex->revived) << DEV_BSHIFT; + + s = splbio(); + /* Get a buffer */ + bp = geteblk(size); + if (bp == NULL) { + splx(s); + return ENOMEM; + } + if (bp->b_qindex != 0) /* on a queue, */ + bremfree(bp); /* remove it */ + splx(s); + + /* Amount to transfer: block size, unless it + * would overlap the end */ + bp->b_bufsize = size; + bp->b_bcount = bp->b_bufsize; + bp->b_resid = 0x0; + bp->b_blkno = plex->revived; /* we've got this far */ + + /* XXX what about reviving anonymous plexes? */ + + /* First, read the data from the volume. We don't + * care which plex, that's bre's job */ + bp->b_dev = VINUMBDEV(plex->volno, 0, 0, VINUM_VOLUME_TYPE); /* create the device number */ + bp->b_flags = B_BUSY | B_READ; + vinumstart(bp, 1); + biowait(bp); + if (bp->b_flags & B_ERROR) + error = bp->b_error; + else + /* Now write to the plex */ + { + s = splbio(); + if (bp->b_qindex != 0) /* on a queue, */ + bremfree(bp); /* remove it */ + splx(s); + bp->b_dev = VINUMBDEV(plex->volno, plex->volplexno, 0, VINUM_PLEX_TYPE); /* create the device number */ + + bp->b_flags = B_BUSY; /* make this a write */ + bp->b_resid = 0x0; + vinumstart(bp, 1); + biowait(bp); + if (bp->b_flags & B_ERROR) + error = bp->b_error; + else { + plex->revived += bp->b_bcount >> DEV_BSHIFT; /* moved this much further down */ + if (plex->revived >= plex->length) { /* finished */ + plex->revived = 0; + plex->state = plex_up; /* do we need to do more? */ + if (plex->volno >= 0) /* we have a volume, */ + set_volume_state(plex->volno, volume_up, 0); + printf("vinum: plex %s is %s\n", plex->name, plex_state(plex->state)); + save_config(); /* and save the updated configuration */ + error = 0; /* we're done */ + } + } + while (plex->waitlist) { /* we have waiting requests */ + launch_requests(plex->waitlist, 1); /* do them now */ + plex->waitlist = plex->waitlist->next; /* and move on to the next */ + } + } + if (bp->b_qindex == 0) /* not on a queue, */ + brelse(bp); /* is this kosher? */ + return error; +} diff --git a/sys/modules/vinum/state.c b/sys/modules/vinum/state.c new file mode 100644 index 0000000..2ce2ed0 --- /dev/null +++ b/sys/modules/vinum/state.c @@ -0,0 +1,755 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: state.c,v 2.6 1998/08/19 08:04:47 grog Exp grog $ + */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "request.h" + +/* Update drive state */ +/* Return 1 if the state changes, otherwise 0 */ +int +set_drive_state(int driveno, enum drivestate state, int flags) +{ + struct drive *drive = &DRIVE[driveno]; + int oldstate = drive->state; + int sdno; + + if (drive->state == drive_unallocated) /* no drive to do anything with, */ + return 0; + + if (state != oldstate) { /* don't change it if it's not different */ + if (state == drive_down) { /* the drive's going down */ + if (flags || (drive->opencount == 0)) { /* we can do it */ + close_drive(drive); + drive->state = state; + printf("vinum: drive %s is %s\n", drive->label.name, drive_state(drive->state)); + } else + return 0; /* don't do it */ + } + drive->state = state; /* set the state */ + if (((drive->state == drive_up) + || ((drive->state == drive_coming_up))) + && (drive->vp == NULL)) /* should be open, but we're not */ + init_drive(drive); /* which changes the state again */ + if ((state != oldstate) /* state has changed */ + &&((flags & setstate_norecurse) == 0)) { /* and we want to recurse, */ + for (sdno = 0; sdno < vinum_conf.subdisks_used; sdno++) { /* find this drive's subdisks */ + if (SD[sdno].driveno == driveno) /* belongs to this drive */ + set_sd_state(sdno, sd_down, setstate_force | setstate_recursing); /* take it down */ + } + save_config(); /* and save the updated configuration */ + return 1; + } + } + return 0; +} + +/* Try to set the subdisk state. Return 1 if state changed to + * what we wanted, -1 if it changed to something else, and 0 + * if no change. + * + * This routine is called both from the user (up, down states + * only) and internally. + */ +int +set_sd_state(int sdno, enum sdstate state, enum setstateflags flags) +{ + struct sd *sd = &SD[sdno]; + int oldstate = sd->state; + int status = 1; /* status to return */ + + if (state == oldstate) + return 0; /* no change */ + + if (sd->state == sd_unallocated) /* no subdisk to do anything with, */ + return 0; + + if (sd->driveoffset < 0) { /* not allocated space */ + sd->state = sd_down; + if (state != sd_down) + return -1; + } else { /* space allocated */ + switch (state) { + case sd_down: + if ((!flags & setstate_force) /* but gently */ + &&(sd->plexno >= 0)) /* and we're attached to a plex, */ + return 0; /* don't do it */ + break; + + case sd_up: + if (DRIVE[sd->driveno].state != drive_up) /* can't bring the sd up if the drive isn't, */ + return 0; /* not even by force */ + switch (sd->state) { + case sd_obsolete: + case sd_down: /* been down, no data lost */ + if ((sd->plexno) /* we're associated with a plex */ + &&(((PLEX[sd->plexno].state < plex_firstup) /* and it's not up */ + ||(PLEX[sd->plexno].subdisks > 1)))) /* or it's the only one */ + break; + /* XXX Get this right: make sure that other plexes in + * the volume cover this address space, otherwise + * we make this one sd_up */ + sd->state = sd_reborn; /* here it is again */ + printf("vinum: subdisk %s is %s, not %s\n", sd->name, sd_state(sd->state), sd_state(state)); + status = -1; + break; + + case sd_init: /* brand new */ + if (flags & setstate_configuring) /* we're doing this while configuring */ + break; + sd->state = sd_empty; /* nothing in it */ + printf("vinum: subdisk %s is %s, not %s\n", sd->name, sd_state(sd->state), sd_state(state)); + status = -1; + break; + + case sd_initializing: + break; /* go on and do it */ + + case sd_empty: + if ((sd->plexno) /* we're associated with a plex */ + &&(((PLEX[sd->plexno].state < plex_firstup) /* and it's not up */ + ||(PLEX[sd->plexno].subdisks > 1)))) /* or it's the only one */ + break; + return 0; /* can't do it */ + + default: /* can't do it */ + /* There's no way to bring subdisks up directly from + * other states. First they need to be initialized + * or revived */ + return 0; + } + break; + + default: /* other ones, only internal with force */ + if (flags & setstate_force == 0) /* no force? What's this? */ + return 0; /* don't do it */ + } + } + sd->state = state; + printf("vinum: subdisk %s is %s\n", sd->name, sd_state(sd->state)); + if ((flags & setstate_norecurse) == 0) + set_plex_state(sd->plexno, plex_up, setstate_recursing); /* update plex state */ + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return status; +} + +/* Called from request routines when they find + * a subdisk which is not kosher. Decide whether + * it warrants changing the state. Return + * REQUEST_DOWN if we can't use the subdisk, + * REQUEST_OK if we can. */ +enum requeststatus +checksdstate(struct sd *sd, struct request *rq, daddr_t diskaddr, daddr_t diskend) +{ + struct plex *plex = &PLEX[sd->plexno]; + int writeop = (rq->bp->b_flags & B_READ) == 0; /* note if we're writing */ + + /* first, see if the plex wants to be accessed */ + switch (plex->state) { + case plex_reviving: + /* When writing, we'll write anything that starts + * up to the current revive pointer, but we'll + * only accept a read which finishes before the + * current revive pointer. + */ + if ((writeop && (diskaddr > plex->revived)) /* write starts after current revive pointer */ + ||((!writeop) && (diskend >= plex->revived))) { /* or read ends after current revive pointer */ + if (writeop) { /* writing to a consistent down disk */ + if (DRIVE[sd->driveno].state == drive_up) + set_sd_state(sd->sdno, sd_stale, setstate_force); /* it's not consistent now */ + else + set_sd_state(sd->sdno, sd_obsolete, setstate_force); /* it's not consistent now */ + } + return REQUEST_DOWN; /* that part of the plex is still down */ + } else if (diskend >= plex->revived) /* write finishes beyond revive pointer */ + rq->flags |= XFR_REVIVECONFLICT; /* note a potential conflict */ + /* FALLTHROUGH */ + + case plex_up: + case plex_degraded: + case plex_flaky: + /* We can access the plex: let's see + * how the subdisk feels */ + switch (sd->state) { + case sd_up: + return REQUEST_OK; + + case sd_reborn: + if (writeop) + return REQUEST_OK; /* always write to a reborn disk */ + /* Handle the mapping. We don't want to reject + * a read request to a reborn subdisk if that's + * all we have. XXX */ + return REQUEST_DOWN; + + case sd_down: + case sd_crashed: + if (writeop) { /* writing to a consistent down disk */ + if (DRIVE[sd->driveno].state == drive_up) + set_sd_state(sd->sdno, sd_stale, setstate_force); /* it's not consistent now */ + else + set_sd_state(sd->sdno, sd_obsolete, setstate_force); /* it's not consistent now */ + } + return REQUEST_DOWN; /* and it's down one way or another */ + + default: + return REQUEST_DOWN; + } + + default: + return REQUEST_DOWN; + } +} + +void +add_defective_region(struct plex *plex, off_t offset, size_t length) +{ +/* XXX get this ordered, and coalesce regions if necessary */ + if (++plex->defective_regions > plex->defective_region_count) + EXPAND(plex->defective_region, + struct plexregion, + plex->defective_region_count, + PLEX_REGION_TABLE_SIZE); + plex->defective_region[plex->defective_regions - 1].offset = offset; + plex->defective_region[plex->defective_regions - 1].length = length; +} + +void +add_unmapped_region(struct plex *plex, off_t offset, size_t length) +{ + if (++plex->unmapped_regions > plex->unmapped_region_count) + EXPAND(plex->unmapped_region, + struct plexregion, + plex->unmapped_region_count, + PLEX_REGION_TABLE_SIZE); + plex->unmapped_region[plex->unmapped_regions - 1].offset = offset; + plex->unmapped_region[plex->unmapped_regions - 1].length = length; +} + +/* Rebuild a plex free list and set state if + * we have a configuration error */ +void +rebuild_plex_unmappedlist(struct plex *plex) +{ + int sdno; + struct sd *sd; + int lastsdend = 0; /* end offset of last subdisk */ + + if (plex->unmapped_region != NULL) { /* we're going to rebuild it */ + Free(plex->unmapped_region); + plex->unmapped_region = NULL; + plex->unmapped_regions = 0; + plex->unmapped_region_count = 0; + } + if (plex->defective_region != NULL) { + Free(plex->defective_region); + plex->defective_region = NULL; + plex->defective_regions = 0; + plex->defective_region_count = 0; + } + for (sdno = 0; sdno < plex->subdisks; sdno++) { + sd = &SD[plex->sdnos[sdno]]; + if (sd->plexoffset < lastsdend) { /* overlap */ + printf("vinum: Plex %s, subdisk %s overlaps previous\n", plex->name, sd->name); + set_plex_state(plex->plexno, plex_down, setstate_force); /* don't allow that */ + } else if (sd->plexoffset > lastsdend) /* gap */ + add_unmapped_region(plex, lastsdend, sd->plexoffset - lastsdend); + else if (sd->state < sd_reborn) /* this part defective */ + add_defective_region(plex, sd->plexoffset, sd->sectors); + lastsdend = sd->plexoffset + sd->sectors; + } +} + +/* return a state map for the subdisks of a plex */ +enum sdstates +sdstatemap(struct plex *plex, int *sddowncount) +{ + int sdno; + enum sdstates statemap = 0; /* note the states we find */ + + *sddowncount = 0; /* no subdisks down yet */ + for (sdno = 0; sdno < plex->subdisks; sdno++) { + struct sd *sd = &SD[plex->sdnos[sdno]]; /* point to the subdisk */ + + switch (sd->state) { + case sd_empty: + statemap |= sd_emptystate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_init: + statemap |= sd_initstate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_down: + statemap |= sd_downstate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_crashed: + statemap |= sd_crashedstate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_obsolete: + statemap |= sd_obsolete; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_stale: + statemap |= sd_stalestate; + (*sddowncount)++; /* another unusable subdisk */ + break; + + case sd_reborn: + statemap |= sd_rebornstate; + break; + + case sd_up: + statemap |= sd_upstate; + break; + + default: + statemap |= sd_otherstate; + break; + } + } + return statemap; +} + +/* determine the state of the volume relative to this plex */ +enum volplexstate +vpstate(struct plex *plex) +{ + struct volume *vol; + enum volplexstate state = volplex_onlyusdown; /* state to return */ + int plexno; + + if (plex->volno < 0) /* not associated with a volume */ + return volplex_onlyusdown; /* assume the worst */ + + vol = &VOL[plex->volno]; /* point to our volume */ + for (plexno = 0; plexno < vol->plexes; plexno++) { + if (&PLEX[vol->plex[plexno]] == plex) { /* us */ + if (PLEX[vol->plex[plexno]].state == plex_up) /* are we up? */ + state |= volplex_onlyus; /* yes */ + } else { + if (PLEX[vol->plex[plexno]].state == plex_up) /* not us */ + state |= volplex_otherup; /* and when they were up, they were up */ + else + state |= volplex_alldown; /* and when they were down, they were down */ + } + } + return state; /* and when they were only halfway up */ +} /* they were neither up nor down */ + +/* Check if all bits b are set in a */ +int allset(int a, int b); + +int +allset(int a, int b) +{ + return (a & b) == b; +} + +/* Update the state of a plex dependent on its subdisks. + * Also rebuild the unmapped_region and defective_region table */ +int +set_plex_state(int plexno, enum plexstate state, enum setstateflags flags) +{ + int sddowncount = 0; /* number of down subdisks */ + struct plex *plex = &PLEX[plexno]; /* point to our plex */ + enum plexstate oldstate = plex->state; + enum volplexstate vps = vpstate(plex); /* how do we compare with the other plexes? */ + enum sdstates statemap = sdstatemap(plex, &sddowncount); /* get a map of the subdisk states */ + + if ((flags & setstate_force) && (oldstate == state)) /* we're there already, */ + return 0; /* no change */ + + if (plex->state == plex_unallocated) /* no plex to do anything with, */ + return 0; + + switch (state) { + case plex_up: + if ((plex->state == plex_initializing) /* we're initializing */ + &&(statemap != sd_upstate)) /* but SDs aren't up yet */ + return 0; /* do nothing */ + + /* We don't really care what our state was before + * if we want to come up. We rely entirely on the + * state of our subdisks and our volume */ + switch (vps) { + case volplex_onlyusdown: + case volplex_alldown: /* another plex is down, and so are we */ + if (statemap == sd_upstate) { /* all subdisks ready for action */ + if ((plex->state == plex_init) /* we're brand spanking new */ + &&(VOL[plex->volno].flags & VF_CONFIG_SETUPSTATE)) { /* and we consider that up */ + /* Conceptually, an empty plex does not contain valid data, + * but normally we'll see this state when we have just + * created a plex, and it's either consistent from earlier, + * or we don't care about the previous contents (we're going + * to create a file system or use it for swap). + * + * We need to do this in one swell foop: on the next call + * we will no longer be just empty. + * + * We'll still come back to this function for the remaining + * plexes in the volume. They'll be up already, so that + * doesn't change anything, but it's not worth the additional + * code to stop doing it. */ + struct volume *vol = &VOL[plex->volno]; + int plexno; + + for (plexno = 0; plexno < vol->plexes; plexno++) + PLEX[vol->plex[plexno]].state = plex_up; + } + plex->state = plex_up; /* bring up up, anyway */ + } else + plex->state = plex_down; + break; + + case volplex_onlyusup: /* only we are up: others are down */ + case volplex_onlyus: /* we're up and alone */ + if ((statemap == sd_upstate) /* subdisks all up */ + ||(statemap == sd_emptystate)) /* or all empty */ + plex->state = plex_up; /* go for it */ + else if ((statemap & (sd_upstate | sd_reborn)) == statemap) /* all up or reborn, */ + plex->state = plex_flaky; + else if (statemap & (sd_upstate | sd_reborn)) /* some up or reborn, */ + plex->state = plex_degraded; /* so far no corruption */ + else + plex->state = plex_faulty; + break; + + case volplex_otherup: /* another plex is up */ + case volplex_otherupdown: /* other plexes are up and down */ + if ((statemap == sd_upstate) /* subdisks all up */ + ||(statemap == sd_emptystate) /* or all empty */ + ) { + /* Is the data in all subdisks valid? */ + if (statemap == statemap & (sd_downstate | sd_rebornstate | sd_upstate)) + break; /* yes, we can bring the plex up */ + plex->state = plex_reviving; /* we need reviving */ + return EAGAIN; + } else + plex->state = plex_faulty; /* still in error */ + break; + + case volplex_allup: /* all plexes are up */ + case volplex_someup: + if ((statemap & (sd_upstate | sd_reborn)) == statemap) /* all up or reborn, */ + break; /* no change */ + else + plex->state = plex_degraded; /* we're not all there */ + } + + if (plex->state != oldstate) + break; + return 0; /* no change */ + + case plex_down: /* want to take it down */ + if (((vps == volplex_onlyus) /* we're the only one up */ + ||(vps == volplex_onlyusup)) /* we're the only one up */ + &&(!(flags & setstate_force))) /* and we don't want to use force */ + return 0; /* can't do it */ + plex->state = state; /* do it */ + break; + + /* This is only requested by the driver. + * Trust ourselves */ + case plex_faulty: + plex->state = state; /* do it */ + break; + + case plex_initializing: + /* XXX consider what safeguards we need here */ + if ((flags & setstate_force) == 0) + return 0; + plex->state = state; /* do it */ + break; + + /* What's this? */ + default: + return 0; + } + printf("vinum: plex %s is %s\n", plex->name, plex_state(plex->state)); + /* Now see what we have left, and whether + * we're taking the volume down */ + if (plex->volno >= 0) { /* we have a volume */ + struct volume *vol = &VOL[plex->volno]; + + vps = vpstate(plex); /* get our combined state again */ + if ((flags & setstate_norecurse) == 0) { /* we can recurse */ + if ((vol->state == volume_up) + && (vps == volplex_alldown)) /* and we're all down */ + set_volume_state(plex->volno, volume_down, setstate_recursing); /* take our volume down */ + else if ((vol->state == volume_down) + && (vps & (volplex_otherup | volplex_onlyusup))) /* and at least one is up */ + set_volume_state(plex->volno, volume_up, setstate_recursing); /* bring our volume up */ + } + } + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return 1; +} + +/* Update the state of a plex dependent on its plexes. + * Also rebuild the unmapped_region and defective_region table */ +int +set_volume_state(int volno, enum volumestate state, enum setstateflags flags) +{ + int plexno; + enum plexstates { + plex_downstate = 1, /* found a plex which is down */ + plex_degradedstate = 2, /* found a plex which is halfway up */ + plex_upstate = 4 /* found a plex which is completely up */ + }; + + int plexstatemap = 0; /* note the states we find */ + struct volume *vol = &VOL[volno]; /* point to our volume */ + + if (vol->state == state) /* we're there already */ + return 0; /* no change */ + if (vol->state == volume_unallocated) /* no volume to do anything with, */ + return 0; + + for (plexno = 0; plexno < vol->plexes; plexno++) { + struct plex *plex = &PLEX[vol->plex[plexno]]; /* point to the plex */ + switch (plex->state) { + case plex_degraded: + case plex_flaky: + case plex_reviving: + plexstatemap |= plex_degradedstate; + break; + + case plex_up: + plexstatemap |= plex_upstate; + break; + + default: + plexstatemap |= plex_downstate; + break; + } + } + + if (state == volume_up) { /* want to come up */ + if (plexstatemap & plex_upstate) { /* we have a plex which is completely up */ + vol->state = volume_up; /* did it */ + printf("vinum: volume %s is %s\n", vol->name, volume_state(vol->state)); + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return 1; + } + /* Here we should check whether we have enough + * coverage for the complete volume. Writeme XXX */ + } else if (state == volume_down) { /* want to go down */ + if ((vol->opencount == 0) /* not open */ + ||(flags & setstate_force != 0)) { /* or we're forcing */ + vol->state = volume_down; + printf("vinum: volume %s is %s\n", vol->name, volume_state(vol->state)); + if ((flags & (setstate_configuring | setstate_recursing)) == 0) /* save config now */ + save_config(); + return 1; + } + } + return 0; /* no change */ +} + +/* Start an object, in other words do what we can to get it up. + * This is called from vinumioctl (VINUMSTART). + * Return error indications via ioctl_reply + */ +void +start_object(struct vinum_ioctl_msg *data) +{ + int status; + int realstatus; /* what we really have */ + int objindex = data->index; /* data gets overwritten */ + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) data; /* format for returning replies */ + + switch (data->type) { + case drive_object: + status = set_drive_state(objindex, drive_up, setstate_none); + realstatus = DRIVE[objindex].state == drive_up; /* set status on whether we really did it */ + break; + + case sd_object: + status = set_sd_state(objindex, sd_up, setstate_none); /* set state */ + realstatus = SD[objindex].state == sd_up; /* set status on whether we really did it */ + break; + + case plex_object: + if (PLEX[objindex].state == plex_reviving) { /* reviving, */ + ioctl_reply->error = revive_block(objindex); /* revive another block */ + ioctl_reply->msg[0] = '\0'; /* no comment */ + return; + } + status = set_plex_state(objindex, plex_up, setstate_none); + realstatus = PLEX[objindex].state == plex_up; /* set status on whether we really did it */ + break; + + case volume_object: + status = set_volume_state(objindex, volume_up, setstate_none); + realstatus = VOL[objindex].state == volume_up; /* set status on whether we really did it */ + break; + + default: + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "Invalid object type"); + return; + } + /* There's no point in saying anything here: + * the userland program does it better */ + ioctl_reply->msg[0] = '\0'; + if (realstatus == 0) /* couldn't do it */ + ioctl_reply->error = EINVAL; + else + ioctl_reply->error = 0; +} + +/* Stop an object, in other words do what we can to get it down + * This is called from vinumioctl (VINUMSTOP). + * Return error indications via ioctl_reply. + */ +void +stop_object(struct vinum_ioctl_msg *data) +{ + int status = 1; + int objindex = data->index; /* save the number from change */ + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) data; /* format for returning replies */ + + switch (data->type) { + case drive_object: + status = set_drive_state(objindex, drive_down, data->force); + break; + + case sd_object: + status = set_sd_state(objindex, sd_down, data->force); + break; + + case plex_object: + status = set_plex_state(objindex, plex_down, data->force); + break; + + case volume_object: + status = set_volume_state(objindex, volume_down, data->force); + break; + + default: + ioctl_reply->error = EINVAL; + strcpy(ioctl_reply->msg, "Invalid object type"); + return; + } + ioctl_reply->msg[0] = '\0'; + if (status == 0) /* couldn't do it */ + ioctl_reply->error = EINVAL; + else + ioctl_reply->error = 0; +} + +/* VINUM_SETSTATE ioctl: set an object state + * msg is the message passed by the user */ +void +setstate(struct vinum_ioctl_msg *msg) +{ + int sdno; + struct sd *sd; + struct plex *plex; + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) msg; /* format for returning replies */ + + switch (msg->state) { + case object_down: + stop_object(msg); + break; + + case object_initializing: + switch (msg->type) { + case sd_object: + sd = &SD[msg->index]; + if ((msg->index >= vinum_conf.subdisks_used) + || (sd->state == sd_unallocated)) { + sprintf(ioctl_reply->msg, "Invalid subdisk %d", msg->index); + ioctl_reply->error = EFAULT; + return; + } + set_sd_state(msg->index, sd_initializing, msg->force); + if (sd->state != sd_initializing) { + strcpy(ioctl_reply->msg, "Can't set state"); + ioctl_reply->error = EINVAL; + } else + ioctl_reply->error = 0; + break; + + case plex_object: + plex = &PLEX[msg->index]; + if ((msg->index >= vinum_conf.plexes_used) + || (plex->state == plex_unallocated)) { + sprintf(ioctl_reply->msg, "Invalid subdisk %d", msg->index); + ioctl_reply->error = EFAULT; + return; + } + set_plex_state(msg->index, plex_initializing, msg->force); + if (plex->state != plex_initializing) { + strcpy(ioctl_reply->msg, "Can't set state"); + ioctl_reply->error = EINVAL; + } else { + ioctl_reply->error = 0; + for (sdno = 0; sdno < plex->subdisks; sdno++) { + sd = &SD[plex->sdnos[sdno]]; + set_sd_state(plex->sdnos[sdno], sd_initializing, msg->force); + if (sd->state != sd_initializing) { + strcpy(ioctl_reply->msg, "Can't set state"); + ioctl_reply->error = EINVAL; + break; + } + } + } + break; + + default: + strcpy(ioctl_reply->msg, "Invalid object"); + ioctl_reply->error = EINVAL; + } + break; + + case object_up: + start_object(msg); + } +} diff --git a/sys/modules/vinum/statetexts.h b/sys/modules/vinum/statetexts.h new file mode 100644 index 0000000..b7ce4eb --- /dev/null +++ b/sys/modules/vinum/statetexts.h @@ -0,0 +1,88 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: COPYRIGHT,v 1.1 1998/03/05 06:07:05 grog Exp grog $ + */ +/* Created by ./makestatetext on Tue 4 Aug 15:53:16 CST 1998. Do not edit */ + +/* Drive state texts */ +char *drivestatetext[] = +{ + "unallocated", + "uninit", + "down", + "coming_up", + "up", +}; + +/* Subdisk state texts */ +char *sdstatetext[] = +{ + "unallocated", + "uninit", + "init", + "initializing", + "empty", + "obsolete", + "stale", + "crashed", + "down", + "reborn", + "up", +}; + +/* Plex state texts */ +char *plexstatetext[] = +{ + "unallocated", + "init", + "faulty", + "down", + "reviving", + "initializing", + "corrupt", + "degraded", + "flaky", + "up", +}; + +/* Volume state texts */ +char *volstatetext[] = +{ + "unallocated", + "uninit", + "down", + "up", +}; diff --git a/sys/modules/vinum/util.c b/sys/modules/vinum/util.c new file mode 100644 index 0000000..5121404 --- /dev/null +++ b/sys/modules/vinum/util.c @@ -0,0 +1,211 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: util.c,v 1.7 1998/08/07 09:23:10 grog Exp grog $ + */ + +/* This file contains utility routines used both in kernel and user context */ + +#include "vinumhdr.h" +#include "statetexts.h" +#ifndef REALLYKERNEL +#include <stdio.h> +extern jmp_buf command_fail; /* return on a failed command */ +#endif + +static char numeric_state[32]; /* temporary buffer for ASCII conversions */ +#define STATECOUNT(x) (sizeof (x##statetext) / sizeof (char *)) +/* Return drive state as a string */ +char * +drive_state(enum drivestate state) +{ + if (((unsigned) state) >= STATECOUNT(drive)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return drivestatetext[state]; +} + +/* Return volume state as a string */ +char * +volume_state(enum volumestate state) +{ + if (((unsigned) state) >= STATECOUNT(vol)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return volstatetext[state]; +} + +/* Return plex state as a string */ +char * +plex_state(enum plexstate state) +{ + if (((unsigned) state) >= STATECOUNT(plex)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return plexstatetext[state]; +} + +/* Return plex organization as a string */ +char * +plex_org(enum plexorg org) +{ + switch (org) { + case plex_disorg: /* disorganized */ + return "disorg"; + break; + + case plex_concat: /* concatenated plex */ + return "concat"; + break; + + case plex_striped: /* striped plex */ + return "striped"; + break; + + + default: + sprintf(numeric_state, "Invalid org %d", (int) org); + return numeric_state; + } +} + +/* Return sd state as a string */ +char * +sd_state(enum sdstate state) +{ + if (((unsigned) state) >= STATECOUNT(sd)) { + sprintf(numeric_state, "Invalid state %d", (int) state); + return numeric_state; + } else + return sdstatetext[state]; +} + +/* Now convert in the other direction */ +/* These are currently used only internally, + * so we don't do too much error checking */ +enum drivestate +DriveState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(drive); i++) + if (strcmp(text, drivestatetext[i]) == 0) /* found it */ + return (enum drivestate) i; + return -1; +} + +enum sdstate +SdState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(sd); i++) + if (strcmp(text, sdstatetext[i]) == 0) /* found it */ + return (enum sdstate) i; + return -1; +} + +enum plexstate +PlexState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(plex); i++) + if (strcmp(text, plexstatetext[i]) == 0) /* found it */ + return (enum plexstate) i; + return -1; +} + +enum volumestate +VolState(char *text) +{ + int i; + for (i = 0; i < STATECOUNT(vol); i++) + if (strcmp(text, volstatetext[i]) == 0) /* found it */ + return (enum volstate) i; + return -1; +} + +/* Take a number with an optional scale factor and convert + * it to a number of bytes. + * + * The scale factors are: + * + * b blocks (of 512 bytes) + * k kilobytes (1024 bytes) + * m megabytes (of 1024 * 1024 bytes) + * g gigabytes (of 1024 * 1024 * 1024 bytes) + */ +u_int64_t +sizespec(char *spec) +{ + u_int64_t size; + char *s; + + size = 0; + s = spec; + if ((*s >= '0') && (*s <= '9')) { /* it's numeric */ + while ((*s >= '0') && (*s <= '9')) /* it's numeric */ + size = size * 10 + *s++ - '0'; /* convert it */ + switch (*s) { + case '\0': + return size; + + case 'B': + case 'b': + return size * 512; + + case 'K': + case 'k': + return size * 1024; + + case 'M': + case 'm': + return size * 1024 * 1024; + + case 'G': + case 'g': + return size * 1024 * 1024 * 1024; + } + } +#ifdef REALLYKERNEL + throw_rude_remark(EINVAL, "Invalid length specification: %s", spec); +#else + fprintf(stderr, "Invalid length specification: %s", spec); + longjmp(command_fail, -1); +#endif + /* NOTREACHED */ + return -1; +} diff --git a/sys/modules/vinum/vinum.c b/sys/modules/vinum/vinum.c new file mode 100644 index 0000000..bffdf6d --- /dev/null +++ b/sys/modules/vinum/vinum.c @@ -0,0 +1,512 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinum.c,v 1.19 1998/08/13 05:24:02 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "sys/sysproto.h" /* for sync(2) */ +#ifdef DEBUG +#include <sys/reboot.h> +int debug = 0; +#endif + +/* pointer to ioctl p parameter, to save passing it around */ +struct proc *myproc; + +#if __FreeBSD__ < 3 +STATIC struct cdevsw vinum_cdevsw; +STATIC struct bdevsw vinum_bdevsw = +{ + vinumopen, vinumclose, vinumstrategy, vinumioctl, + vinumdump, vinumsize, 0, + "vinum", &vinum_cdevsw, -1 +}; +#else /* goodbye, bdevsw */ +STATIC struct cdevsw vinum_cdevsw = +{ + vinumopen, vinumclose, vinumread, vinumwrite, + vinumioctl, nostop, nullreset, nodevtotty, + seltrue, nommap, vinumstrategy, "vinum", + NULL, -1, vinumdump, vinumsize, + D_DISK, 0, -1 +}; +#endif + +/* Called by main() during pseudo-device attachment. */ +STATIC void vinumattach(void *); + +STATIC void vinumgetdisklabel(dev_t); +void vinum_scandisk(void); +int vinum_inactive(void); +void free_vinum(int); + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +extern jmp_buf command_fail; /* return here if config fails */ + +struct _vinum_conf vinum_conf; /* configuration information */ + +STATIC int vinum_devsw_installed = 0; + +/* + * Called by main() during pseudo-device attachment. All we need + * to do is allocate enough space for devices to be configured later, and + * add devsw entries. + */ +void +vinumattach(void *dummy) +{ + BROKEN_GDB; + char *buf; /* pointer to temporary buffer */ + struct _ioctl_reply *ioctl_reply; /* struct to return */ + struct uio uio; + struct iovec iovec; + + /* modload should prevent multiple loads, so this is worth a panic */ + if ((vinum_conf.flags & VF_LOADED) != NULL) + panic("vinum: already loaded"); + + printf("vinum: loaded\n"); + vinum_conf.flags |= VF_LOADED; /* we're loaded now */ + + /* We don't have a p pointer here, so take it from curproc */ + myproc = curproc; +#if __FreeBSD__ < 3 + bdevsw_add_generic(BDEV_MAJOR, CDEV_MAJOR, &vinum_bdevsw); +#else + cdevsw_add_generic(BDEV_MAJOR, CDEV_MAJOR, &vinum_cdevsw); +#endif +#ifdef DEVFS +#error DEVFS not finished yet +#endif + + uio.uio_iov = &iovec; + uio.uio_iovcnt = 1; /* just one buffer */ + uio.uio_offset = 0; /* start at the beginning */ + uio.uio_resid = 512; /* one sector */ + uio.uio_segflg = UIO_SYSSPACE; /* we're in system space */ + uio.uio_rw = UIO_READ; /* do we need this? */ + uio.uio_procp = curproc; /* do it for our own process */ + + iovec.iov_len = 512; + buf = (char *) Malloc(iovec.iov_len); /* get a buffer */ + CHECKALLOC(buf, "vinum: no memory\n"); /* can't get 512 bytes? */ + iovec.iov_base = buf; /* read into buf */ + + /* allocate space: drives... */ + DRIVE = (struct drive *) Malloc(sizeof(struct drive) * INITIAL_DRIVES); + CHECKALLOC(DRIVE, "vinum: no memory\n"); + vinum_conf.drives_allocated = INITIAL_DRIVES; /* number of drive slots allocated */ + vinum_conf.drives_used = 0; /* and number in use */ + + /* volumes, ... */ + VOL = (struct volume *) Malloc(sizeof(struct volume) * INITIAL_VOLUMES); + CHECKALLOC(VOL, "vinum: no memory\n"); + vinum_conf.volumes_allocated = INITIAL_VOLUMES; /* number of volume slots allocated */ + vinum_conf.volumes_used = 0; /* and number in use */ + + /* plexes, ... */ + PLEX = (struct plex *) Malloc(sizeof(struct plex) * INITIAL_PLEXES); + CHECKALLOC(PLEX, "vinum: no memory\n"); + vinum_conf.plexes_allocated = INITIAL_PLEXES; /* number of plex slots allocated */ + vinum_conf.plexes_used = 0; /* and number in use */ + + /* and subdisks */ + SD = (struct sd *) Malloc(sizeof(struct sd) * INITIAL_SUBDISKS); + CHECKALLOC(SD, "vinum: no memory\n"); + vinum_conf.subdisks_allocated = INITIAL_SUBDISKS; /* number of sd slots allocated */ + vinum_conf.subdisks_used = 0; /* and number in use */ + + ioctl_reply = NULL; /* no reply on longjmp */ +} + + +#ifdef ACTUALLY_LKM_NOT_KERNEL /* stuff for LKMs */ + +/* Check if we have anything open. If so, return 0 (not inactive), + * otherwise 1 (inactive) */ +int +vinum_inactive(void) +{ + BROKEN_GDB; + int i; + int can_do = 1; /* assume we can do it */ + + lock_config(); + for (i = 0; i < vinum_conf.volumes_used; i++) { + if (VOL[i].pid != NULL) { /* volume is open */ + can_do = 0; + break; + } + } + unlock_config(); + return can_do; +} + +/* Free all structures. + * If cleardrive is 0, save the configuration; otherwise + * remove the configuration from the drive. + * + * Before coming here, ensure that no volumes are open. + */ +void +free_vinum(int cleardrive) +{ + BROKEN_GDB; + int i; + + if (cleardrive) { + for (i = 0; i < vinum_conf.drives_used; i++) + remove_drive(i); /* remove the drive */ + } else { /* keep the config */ + save_config(); + if (DRIVE != NULL) { + for (i = 0; i < vinum_conf.drives_used; i++) + free_drive(&DRIVE[i]); /* close files and things */ + Free(DRIVE); + } + } + if (SD != NULL) + Free(SD); + if (PLEX != NULL) { + for (i = 0; i < vinum_conf.plexes_used; i++) { + struct plex *plex = &vinum_conf.plex[i]; + + if (plex->state != plex_unallocated) { /* we have real data there */ + if (plex->sdnos) + Free(plex->sdnos); + if (plex->unmapped_regions) + Free(plex->unmapped_region); + if (plex->defective_regions) + Free(plex->defective_region); + } + } + Free(PLEX); + } + if (VOL != NULL) + Free(VOL); + bzero(&vinum_conf, sizeof(vinum_conf)); +} + +MOD_MISC(vinum); + +/* + * Function called when loading the driver. + */ +STATIC int +vinum_load(struct lkm_table *lkmtp, int cmd) +{ + BROKEN_GDB; +/* Debugger ("vinum_load"); */ + vinumattach(NULL); + return 0; /* OK */ +} + +/* + * Function called when unloading the driver. + */ +STATIC int +vinum_unload(struct lkm_table *lkmtp, int cmd) +{ + BROKEN_GDB; + if (vinum_inactive()) { /* is anything open? */ + struct sync_args dummyarg = + {0}; +#if __FreeBSD__ < 3 + int retval; +#endif + + printf("vinum: unloaded\n"); +#if __FreeBSD__ < 3 + sync(curproc, &dummyarg, &retval); /* write out buffers */ +#else + sync(curproc, &dummyarg); /* write out buffers */ +#endif + free_vinum(0); /* no: clean up */ +#if __FreeBSD__ < 3 + bdevsw[BDEV_MAJOR] = NULL; /* clear bdevsw */ +#endif + cdevsw[CDEV_MAJOR] = NULL; /* and cdevsw */ + return 0; + } else + return EBUSY; +} + +/* + * Dispatcher function for the module (load/unload/stat). + */ +int +vinum_mod(struct lkm_table *lkmtp, int cmd, int ver) +{ + BROKEN_GDB; + MOD_DISPATCH(vinum, /* module name */ + lkmtp, /* LKM table */ + cmd, /* command */ + ver, + vinum_load, /* load with this function */ + vinum_unload, /* and unload with this */ + lkm_nullcmd); +} + +#else /* not LKM */ +#error "This driver must be compiled as a loadable kernel module" +#endif /* LKM */ + +/* ARGSUSED */ +/* Open a vinum object + * At the moment, we only open volumes and the + * super device. It's a nice concept to be + * able to open drives, subdisks and plexes, but + * I can't think what good it could be */ +int +vinumopen(dev_t dev, + int flags, + int fmt, + struct proc *p) +{ + BROKEN_GDB; + int s; /* spl */ + int error; + unsigned int index; + struct volume *vol; + struct plex *plex; + struct sd *sd; + struct devcode *device; + + device = (struct devcode *) &dev; + + error = 0; + /* First, decide what we're looking at */ + switch (device->type) { + case VINUM_VOLUME_TYPE: + index = VOLNO(dev); + if (index >= vinum_conf.volumes_used) + return ENXIO; /* no such device */ + vol = &VOL[index]; + + switch (vol->state) { + case volume_unallocated: + case volume_uninit: + return ENXIO; + + case volume_up: + s = splhigh(); /* quick lock */ + if (error) + return error; + if (vol->opencount == 0) + vol->openflags = flags; /* set our flags */ + vol->opencount++; + vol->pid = p->p_pid; /* and say who we are (do we need this? XXX) */ + splx(s); + return 0; + + case volume_down: + return EIO; + + default: + return EINVAL; + } + + case VINUM_PLEX_TYPE: + if (VOLNO(dev) >= vinum_conf.volumes_used) + return ENXIO; + index = PLEXNO(dev); /* get plex index in vinum_conf */ + if (index >= vinum_conf.plexes_used) + return ENXIO; /* no such device */ + plex = &PLEX[index]; + + switch (plex->state) { + case plex_unallocated: + return EINVAL; + + default: + s = splhigh(); + if (plex->pid /* it's open already */ + && (plex->pid != p->p_pid)) { /* and not by us, */ + splx(s); + return EBUSY; /* one at a time, please */ + } + plex->pid = p->p_pid; /* and say who we are (do we need this? XXX) */ + splx(s); + return 0; + } + + case VINUM_SD_TYPE: + if ((VOLNO(dev) >= vinum_conf.volumes_used) || /* no such volume */ + (PLEXNO(dev) >= vinum_conf.plexes_used)) /* or no such plex */ + return ENXIO; /* no such device */ + index = SDNO(dev); /* get the subdisk number */ + if (index >= vinum_conf.subdisks_used) + return ENXIO; /* no such device */ + sd = &SD[index]; + + /* Opening a subdisk is always a special operation, so we + * ignore the state as long as it represents a real subdisk */ + switch (sd->state) { + case sd_unallocated: + case sd_uninit: + return EINVAL; + + default: + s = splhigh(); + if (sd->pid /* it's open already */ + && (sd->pid != p->p_pid)) { /* and not by us, */ + splx(s); + return EBUSY; /* one at a time, please */ + } + sd->pid = p->p_pid; /* and say who we are (do we need this? XXX) */ + splx(s); + return 0; + } + + case VINUM_DRIVE_TYPE: + default: + return ENODEV; /* don't know what to do with these */ + + case VINUM_SUPERDEV_TYPE: + if (p->p_ucred->cr_uid == 0) { /* root calling, */ + vinum_conf.opencount++; /* one more opener */ + return 0; /* no worries opening super dev */ + } else + return EPERM; /* you can't do that! */ + } +} + +/* ARGSUSED */ +int +vinumclose(dev_t dev, + int flags, + int fmt, + struct proc *p) +{ + BROKEN_GDB; + unsigned int index; + struct volume *vol; + struct plex *plex; + struct sd *sd; + struct devcode *device = (struct devcode *) &dev; + + index = VOLNO(dev); + /* First, decide what we're looking at */ + switch (device->type) { + case VINUM_VOLUME_TYPE: + if (index >= vinum_conf.volumes_used) + return ENXIO; /* no such device */ + vol = &VOL[index]; + + switch (vol->state) { + case volume_unallocated: + case volume_uninit: + return ENXIO; + + case volume_up: + vol->opencount = 0; /* reset our flags */ + vol->pid = NULL; /* and forget who owned us */ + return 0; + + case volume_down: + return EIO; + + default: + return EINVAL; + } + + case VINUM_PLEX_TYPE: + if (VOLNO(dev) >= vinum_conf.volumes_used) + return ENXIO; + index = PLEXNO(dev); /* get plex index in vinum_conf */ + if (index >= vinum_conf.plexes_used) + return ENXIO; /* no such device */ + plex = &PLEX[index]; + plex->pid = 0; + return 0; + + case VINUM_SD_TYPE: + if ((VOLNO(dev) >= vinum_conf.volumes_used) || /* no such volume */ + (PLEXNO(dev) >= vinum_conf.plexes_used)) /* or no such plex */ + return ENXIO; /* no such device */ + index = SDNO(dev); /* get the subdisk number */ + if (index >= vinum_conf.subdisks_used) + return ENXIO; /* no such device */ + sd = &SD[index]; + sd->pid = 0; + return 0; + + case VINUM_SUPERDEV_TYPE: + if (p->p_ucred->cr_uid == 0) /* root calling, */ + vinum_conf.opencount--; /* one less opener */ + return 0; /* no worries closing super dev */ + + case VINUM_DRIVE_TYPE: + default: + return ENODEV; /* don't know what to do with these */ + } +} + +/* size routine */ +int +vinumsize(dev_t dev) +{ + BROKEN_GDB; + struct volume *vol; + int size; + +/* XXX This is bogus. We don't need to open + * a device to find its size */ + vol = &VOL[VOLNO(dev)]; + + if (vol->state == volume_up) + size = vol->size; + else + return 0; /* err on the size of conservatism */ + + + return size; +} + +int +vinumdump(dev_t dev) +{ + /* Not implemented. */ + return ENXIO; +} diff --git a/sys/modules/vinum/vinumext.h b/sys/modules/vinum/vinumext.h new file mode 100644 index 0000000..4b7d5a0 --- /dev/null +++ b/sys/modules/vinum/vinumext.h @@ -0,0 +1,214 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumext.h,v 1.14 1998/08/11 00:03:57 grog Exp grog $ + */ + +/* vinumext.h: external definitions */ + +extern struct _vinum_conf vinum_conf; /* configuration information */ + +#ifdef DEBUG +extern debug; /* debug flags */ +#endif + +#define CHECKALLOC(ptr, msg) \ + if (ptr == NULL) \ + { \ + printf (msg); \ + longjmp (command_fail, -1); \ + } +#ifndef KERNEL +struct vnode; +struct proc; +#endif + +#ifdef KERNEL +int give_sd_to_plex(int plexno, int sdno); +int give_plex_to_volume(int volno, int plexno); +int check_drive(char *); +enum drive_label_info read_drive_label(struct drive *drive); +int parse_config(char *, struct keywordset *); +int parse_user_config(char *cptr, struct keywordset *keyset); +u_int64_t sizespec(char *spec); +int volume_index(struct volume *volume); +int plex_index(struct plex *plex); +int sd_index(struct sd *sd); +int drive_index(struct drive *drive); +int my_plex(int volno, int plexno); +int my_sd(int plexno, int sdno); +int get_empty_drive(void); +int find_drive(const char *name, int create); +int find_drive_by_dev(const char *devname, int create); +int get_empty_sd(void); +int find_subdisk(const char *name, int create); +void free_sd(int sdno); +void free_volume(int volno); +int get_empty_plex(void); +int find_plex(const char *name, int create); +void free_plex(int plexno); +int get_empty_volume(void); +int find_volume(const char *name, int create); +void config_subdisk(void); +void config_plex(void); +void config_volume(void); +void config_drive(void); +void updateconfig(int); +void update_sd_config(int sdno, int kernelstate); +void update_plex_config(int plexno, int kernelstate); +void update_volume_config(int volno, int kernelstate); +void update_config(void); +void drive_io_done(struct buf *); +int save_config(void); +void write_config(char *, int); +int start_config(void); +void finish_config(int); +void remove(struct vinum_ioctl_msg *msg); +void remove_drive_entry(int driveno, int force, int recurse); +void remove_sd_entry(int sdno, int force, int recurse); +void remove_plex_entry(int plexno, int force, int recurse); +void remove_volume_entry(int volno, int force, int recurse); + +void checkernel(char *); +int open_drive(struct drive *, struct proc *); +void close_drive(struct drive *drive); +int driveio(struct drive *, void *, size_t, off_t, int); +/* #define read_drive(a, b, c, d) driveio (a, b, c, d, B_READ) + #define write_drive(a, b, c, d) driveio (a, b, c, d, B_WRITE) */ +int set_drive_parms(struct drive *drive); +int init_drive(struct drive *); +/* void throw_rude_remark (int, struct _ioctl_reply *, char *, ...); XXX */ +void throw_rude_remark(int, char *,...); + +int read_drive(struct drive *drive, void *buf, size_t length, off_t offset); +int write_drive(struct drive *drive, void *buf, size_t length, off_t offset); +void format_config(char *config, int len); +void checkkernel(char *op); +void free_drive(struct drive *drive); +void down_drive(struct drive *drive); +void remove_drive(int driveno); + +/* I/O */ +d_open_t vinumopen; +d_close_t vinumclose; +d_strategy_t vinumstrategy; +d_ioctl_t vinumioctl; +d_dump_t vinumdump; +d_psize_t vinumsize; +d_read_t vinumread; +d_write_t vinumwrite; + +int vinumstart(struct buf *bp, int reviveok); +int launch_requests(struct request *rq, int reviveok); + +/* XXX Do we need this? */ +int vinumpart(dev_t); + +/* Memory allocation */ +void vinum_meminfo(caddr_t data); +int vinum_mallocinfo(caddr_t data); + +void expand_table(void **, int, int); + +void add_defective_region(struct plex *plex, off_t offset, size_t length); +void add_unmapped_region(struct plex *plex, off_t offset, size_t length); +void rebuild_plex_unmappedlist(struct plex *plex); +struct request; +struct rqgroup *allocrqg(struct request *rq, int elements); +void deallocrqg(struct rqgroup *rqg); + +/* State transitions */ +int set_drive_state(int driveno, enum drivestate state, int force); +int set_sd_state(int sdno, enum sdstate state, enum setstateflags flags); +enum requeststatus checksdstate(struct sd *sd, struct request *rq, daddr_t diskaddr, daddr_t diskend); +int set_plex_state(int plexno, enum plexstate state, enum setstateflags flags); +int set_volume_state(int volumeno, enum volumestate state, enum setstateflags flags); +void get_volume_label(struct volume *vol, struct disklabel *lp); +int write_volume_label(int); +void start_object(struct vinum_ioctl_msg *); +void stop_object(struct vinum_ioctl_msg *); +void setstate(struct vinum_ioctl_msg *msg); +void vinum_label(int); +int vinum_writedisklabel(struct volume *, struct disklabel *); +int initsd(int); + +int restart_plex(int plexno); +int revive_block(int plexno); + +/* Auxiliary functions */ +enum sdstates sdstatemap(struct plex *plex, int *sddowncount); +enum volplexstate vpstate(struct plex *plex); +#endif + +enum keyword get_keyword(char *, struct keywordset *); +void listconfig(void); +char *drive_state(enum drivestate); +char *volume_state(enum volumestate); +char *plex_state(enum plexstate); +char *plex_org(enum plexorg); +char *sd_state(enum sdstate); +enum drivestate DriveState(char *text); +enum sdstate SdState(char *text); +enum plexstate PlexState(char *text); +enum volumestate VolState(char *text); +struct drive *validdrive(int driveno, struct _ioctl_reply *); +struct sd *validsd(int sdno, struct _ioctl_reply *); +struct plex *validplex(int plexno, struct _ioctl_reply *); +struct volume *validvol(int volno, struct _ioctl_reply *); +int tokenize(char *, char *[]); +void resetstats(struct vinum_ioctl_msg *msg); + +/* Locking */ +int lockvol(struct volume *vol); +void unlockvol(struct volume *vol); +int lockplex(struct plex *plex); +void unlockplex(struct plex *plex); +int lockrange(struct plex *plex, off_t first, off_t last); +void unlockrange(struct plex *plex, off_t first, off_t last); +int lock_config(void); +void unlock_config(void); + +#ifdef DEBUG +#define expandrq(prq) \ +{ \ + expand_table ((void **) &prq->rqe, \ + prq->requests * sizeof (struct rqelement), \ + (prq->requests + RQELTS) * sizeof (struct rqelement) ); \ + bzero (&prq->rqe [prq->requests], RQELTS * sizeof (struct rqelement)); \ + prq->rqcount += RQELTS; \ + } +#else +void expandrq(struct plexrq *); +#endif diff --git a/sys/modules/vinum/vinumhdr.h b/sys/modules/vinum/vinumhdr.h new file mode 100644 index 0000000..22a8561 --- /dev/null +++ b/sys/modules/vinum/vinumhdr.h @@ -0,0 +1,104 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + */ + +/* Header files used by all modules */ +/* $Id: vinumhdr.h,v 1.7 1998/08/07 04:41:18 grog Exp grog $ */ + +#ifdef KERNEL +#define REALLYKERNEL +#endif +#include <sys/param.h> +#ifdef REALLYKERNEL +#include <sys/systm.h> +#include <sys/kernel.h> +#endif +#ifdef DEVFS +#error "DEVFS code not complete yet" +#include <sys/devfsext.h> +#endif /*DEVFS */ +#include <sys/proc.h> +#include <sys/errno.h> +#include <sys/dkstat.h> +#include <sys/buf.h> +#include <sys/malloc.h> +#include <sys/uio.h> +#include <sys/namei.h> +#include <sys/conf.h> +#include <sys/stat.h> +#include <sys/disklabel.h> +#include <ufs/ffs/fs.h> +#include <sys/mount.h> +#include <sys/device.h> +#undef KERNEL /* XXX */ +#include <sys/disk.h> +#ifdef REALLYKERNEL +#define KERNEL +#endif +#include <sys/syslog.h> +#include <sys/fcntl.h> +#include <sys/vnode.h> +#include <sys/dkbad.h> +#include <setjmp.h> +#include <stdarg.h> +#include <vm/vm.h> +#ifdef USES_VM +/* XXX Do we need this? */ +#include <vm/vm_extern.h> +#include <vm/vm_kern.h> +#include <vm/vm_object.h> +#include <vm/vm_prot.h> +/* #include <vm/vm_page.h> */ +#include <sys/vmmeter.h> +/* #include <machine/pmap.h> */ +#include <machine/cputypes.h> +#endif /* USES_VM */ +#include <vinumvar.h> +#include <vinumio.h> +#include "vinumkw.h" +#include "vinumext.h" + +#undef Free /* defined in some funny net stuff */ +#ifdef REALLYKERNEL +#define Malloc(x) MMalloc ((x), __FILE__, __LINE__) /* show where we came from */ +#define Free(x) FFree ((x), __FILE__, __LINE__) /* show where we came from */ +caddr_t MMalloc (int size, char *, int); +void FFree (void *mem, char *, int); +#else +#define Malloc(x) malloc ((x)) /* just the size */ +#define Free(x) free ((x)) /* just the address */ +#endif + diff --git a/sys/modules/vinum/vinumio.h b/sys/modules/vinum/vinumio.h new file mode 100644 index 0000000..be79528 --- /dev/null +++ b/sys/modules/vinum/vinumio.h @@ -0,0 +1,132 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumio.h,v 1.10 1998/08/10 05:46:19 grog Exp grog $ + */ + +#define MAX_IOCTL_REPLY 256 +#define L 'F' /* ID letter of our ioctls */ +/* VINUM_CREATE returns a buffer of this kind */ +struct _ioctl_reply { + int error; + char msg[MAX_IOCTL_REPLY]; +}; + +/* ioctl requests */ +#define BUFSIZE 1024 /* size of buffer, including continuations */ +#define VINUM_CREATE _IOC(IOC_IN | IOC_OUT, L, 64, BUFSIZE) /* configure vinum */ +#define VINUM_GETCONFIG _IOR(L, 65, struct _vinum_conf) /* get global config */ +#define VINUM_DRIVECONFIG _IOWR(L, 66, struct drive) /* get drive config */ +#define VINUM_SDCONFIG _IOWR(L, 67, struct sd) /* get subdisk config */ +#define VINUM_PLEXCONFIG _IOWR(L, 68, struct plex) /* get plex config */ +#define VINUM_VOLCONFIG _IOWR(L, 69, struct volume) /* get volume config */ +#define VINUM_PLEXSDCONFIG _IOWR(L, 70, struct sd) /* get sd config for plex (plex, sdno) */ +#define VINUM_GETFREELIST _IOWR(L, 71, struct drive_freelist) /* get freelist element (drive, fe) */ +#define VINUM_SAVECONFIG _IOC(0, L, 72, 0) /* release locks, update, write config to disk */ +#define VINUM_RESETCONFIG _IOC(0, L, 73, 0) /* trash config on disk */ +#define VINUM_INIT _IOC(0, L, 74, 0) /* read config from disk */ +#ifdef DEBUG + +struct debuginfo { + int changeit; + int param; +}; + +#define VINUM_DEBUG _IOWR(L, 75, struct debuginfo) /* call the debugger from ioctl () */ +#endif + +enum objecttype { + drive_object, + sd_object, + plex_object, + volume_object, + invalid_object +}; + +/* Start an object. Pass two integers: + * msg [0] index in vinum_conf.<object> + * msg [1] type of object (see below) + * + * Return ioctl_reply + */ +#define VINUM_SETSTATE _IOC(IOC_IN | IOC_OUT, L, 76, MAX_IOCTL_REPLY) /* start an object */ + +/* The state to set with VINUM_SETSTATE. Since + * each object has a different set of states, we + * need to translate later */ +enum objectstate { + object_down, + object_initializing, + object_up +}; + +/* This structure is used for modifying objects + * (VINUM_SETSTATE, VINUM_REMOVE, VINUM_RESETSTATS, VINUM_ATTACH, + * VINUM_DETACH, VINUM_REPLACE + */ +struct vinum_ioctl_msg { + int index; + enum objecttype type; + enum objectstate state; /* state to set (VINUM_SETSTATE) */ + int force; /* do it even if it doesn't make sense */ + int recurse; /* recurse (VINUM_REMOVE) */ + int otherobject; /* superordinate object (attach), + * replacement object (replace) */ + int rename; /* rename object (attach) */ + int64_t offset; /* offset of subdisk (for attach) */ +}; + +#define VINUM_RELEASECONFIG _IOC(0, L, 77, 0) /* release locks and write config to disk */ +#define VINUM_STARTCONFIG _IOC(0, L, 78, 0) /* start a configuration operation */ +#define VINUM_MEMINFO _IOR(L, 79, struct meminfo) /* get memory usage summary */ +#define VINUM_MALLOCINFO _IOWR(L, 80, struct mc) /* get specific malloc information [i] */ +#define VINUM_LABEL _IOC(IOC_IN | IOC_OUT, L, 81, MAX_IOCTL_REPLY) /* label a volume */ +#define VINUM_INITSD _IOW(L, 82, int) /* initialize a subdisk */ +#define VINUM_REMOVE _IOC(IOC_IN | IOC_OUT, L, 83, MAX_IOCTL_REPLY) /* remove an object */ +#define VINUM_GETUNMAPPED _IOWR(L, 84, struct plexregion) /* get unmapped element (plex, re) */ +#define VINUM_GETDEFECTIVE _IOWR(L, 85, struct plexregion) /* get defective element (plex, re) */ +#define VINUM_RESETSTATS _IOC(IOC_IN | IOC_OUT, L, 86, MAX_IOCTL_REPLY) /* reset object stats */ +#define VINUM_ATTACH _IOC(IOC_IN | IOC_OUT, L, 87, MAX_IOCTL_REPLY) /* reset object stats */ +#define VINUM_DETACH _IOC(IOC_IN | IOC_OUT, L, 88, MAX_IOCTL_REPLY) /* reset object stats */ + +struct vinum_rename_msg { + int index; + int recurse; /* rename subordinate objects too */ + enum objecttype type; + char newname[MAXNAME]; /* new name to give to object */ +}; + +#define VINUM_RENAME _IOC(IOC_IN | IOC_OUT, L, 89, MAX_IOCTL_REPLY) /* reset object stats */ +#define VINUM_REPLACE _IOC(IOC_IN | IOC_OUT, L, 90, MAX_IOCTL_REPLY) /* reset object stats */ diff --git a/sys/modules/vinum/vinumioctl.c b/sys/modules/vinum/vinumioctl.c new file mode 100644 index 0000000..6dbe3c6 --- /dev/null +++ b/sys/modules/vinum/vinumioctl.c @@ -0,0 +1,787 @@ +/* XXX replace all the checks on object validity with + * calls to valid<object> */ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumioctl.c,v 1.1 1998/08/14 08:46:10 grog Exp grog $ + */ + +#define STATIC /* nothing while we're testing XXX */ + +#define REALLYKERNEL +#include "vinumhdr.h" +#include "sys/sysproto.h" /* for sync(2) */ +#ifdef DEBUG +#include <sys/reboot.h> +#endif + +jmp_buf command_fail; /* return on a failed command */ + +#if __FreeBSD__ >= 3 +/* Why aren't these declared anywhere? XXX */ +int setjmp(jmp_buf); +void longjmp(jmp_buf, int); +#endif + +/* pointer to ioctl p parameter, to save passing it around */ +struct proc *myproc; + +int vinum_inactive(void); +void free_vinum(int); +void attachobject(struct vinum_ioctl_msg *); +void detachobject(struct vinum_ioctl_msg *); +void renameobject(struct vinum_rename_msg *); +void replaceobject(struct vinum_ioctl_msg *); + +/* ioctl routine */ +int +vinumioctl(dev_t dev, +#if __FreeBSD__ >= 3 + u_long cmd, +#else + int cmd, +#endif + caddr_t data, + int flag, + struct proc *p) +{ + BROKEN_GDB; + unsigned int objno; + int error = 0; + struct volume *vol; + unsigned int index; /* for transferring config info */ + unsigned int sdno; /* for transferring config info */ + int fe; /* free list element number */ + struct _ioctl_reply *ioctl_reply = (struct _ioctl_reply *) data; /* struct to return */ + + struct devcode *device = (struct devcode *) &dev; + + /* First, decide what we're looking at */ + switch (device->type) { + case VINUM_SUPERDEV_TYPE: + myproc = p; /* save pointer to process */ + ioctl_reply = (struct _ioctl_reply *) data; /* save the address to reply to */ + error = setjmp(command_fail); /* come back here on error */ + if (error) /* bombed out */ + return 0; /* the reply will contain meaningful info */ + switch (cmd) { + /* XXX #ifdef DEBUG */ + case VINUM_DEBUG: + boothowto |= RB_GDB; /* serial debug line */ + if (((struct debuginfo *) data)->changeit) /* change debug settings */ + debug = (((struct debuginfo *) data)->param); + else + Debugger("vinum debug"); + ioctl_reply = (struct _ioctl_reply *) data; /* reinstate the address to reply to */ + ioctl_reply->error = 0; + return 0; + /* XXX #endif */ + + case VINUM_CREATE: /* create a vinum object */ + error = lock_config(); /* get the config for us alone */ + if (error) /* can't do it, */ + return error; /* give up */ + error = setjmp(command_fail); /* come back here on error */ + if (error == 0) { /* first time, */ + parse_user_config((char *) data, &keyword_set); /* update the config */ + ioctl_reply->error = 0; /* no error if we make it here */ + } else if (ioctl_reply->error == 0) { /* longjmp, but no error status */ + ioctl_reply->error = EINVAL; /* note that something's up */ + ioctl_reply->msg[0] = '\0'; /* no message? */ + } + unlock_config(); + return 0; /* must be 0 to return the real error info */ + + case VINUM_GETCONFIG: /* get the configuration information */ + bcopy(&vinum_conf, data, sizeof(vinum_conf)); + return 0; + + /* start configuring the subsystem */ + case VINUM_STARTCONFIG: + return start_config(); /* just lock it */ + + /* Move the individual parts of the config to user space. + + * Specify the index of the object in the first word of data, + * and return the object there + */ + case VINUM_DRIVECONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.drives_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&DRIVE[index], data, sizeof(struct drive)); /* copy the config item out */ + return 0; + + case VINUM_SDCONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.subdisks_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&SD[index], data, sizeof(struct sd)); /* copy the config item out */ + return 0; + + case VINUM_PLEXCONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.plexes_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&PLEX[index], data, sizeof(struct plex)); /* copy the config item out */ + return 0; + + case VINUM_VOLCONFIG: + index = *(int *) data; /* get the index */ + if (index >= (unsigned) vinum_conf.volumes_used) /* can't do it */ + return EFAULT; /* bang */ + bcopy(&VOL[index], data, sizeof(struct volume)); /* copy the config item out */ + return 0; + + case VINUM_PLEXSDCONFIG: + index = *(int *) data; /* get the plex index */ + sdno = ((int *) data)[1]; /* and the sd index */ + if ((index >= (unsigned) vinum_conf.plexes_used) /* plex doesn't exist */ + ||(sdno >= PLEX[index].subdisks)) /* or it doesn't have this many subdisks */ + return EFAULT; /* bang */ + bcopy(&SD[PLEX[index].sdnos[sdno]], /* copy the config item out */ + data, + sizeof(struct sd)); + return 0; + + case VINUM_SAVECONFIG: + if (VFLAGS & VF_CONFIGURING) { /* must be us, the others are asleep */ + finish_config(1); /* finish the configuration and update it */ + error = save_config(); /* save configuration to disk */ + } else + error = EINVAL; /* queue up for this one, please */ + return error; + + case VINUM_RELEASECONFIG: /* release the config */ + if (VFLAGS & VF_CONFIGURING) { /* must be us, the others are asleep */ + finish_config(0); /* finish the configuration, don't change it */ + error = save_config(); /* save configuration to disk */ + } else + error = EINVAL; /* release what config? */ + return error; + + case VINUM_INIT: + ioctl_reply = (struct _ioctl_reply *) data; /* reinstate the address to reply to */ + ioctl_reply->error = 0; + return 0; + + case VINUM_RESETCONFIG: + if (vinum_inactive() && (vinum_conf.opencount < 2)) { /* if we're not active */ + /* Note the open count. We may be called from v, so we'll be open. + * Keep the count so we don't underflow */ + int oc = vinum_conf.opencount; + free_vinum(1); /* clean up everything */ + printf("vinum: CONFIGURATION OBLITERATED\n"); + vinum_conf.opencount = oc; + ioctl_reply = (struct _ioctl_reply *) data; /* reinstate the address to reply to */ + ioctl_reply->error = 0; + return 0; + } + return EBUSY; + + case VINUM_SETSTATE: + setstate((struct vinum_ioctl_msg *) data); /* set an object state */ + return 0; + + case VINUM_MEMINFO: + vinum_meminfo(data); + return 0; + + case VINUM_MALLOCINFO: + return vinum_mallocinfo(data); + + case VINUM_LABEL: /* label a volume */ + ioctl_reply->error = write_volume_label(*(int *) data); /* index of the volume to label */ + ioctl_reply->msg[0] = '\0'; /* no message */ + return 0; + + case VINUM_REMOVE: + remove((struct vinum_ioctl_msg *) data); /* remove an object */ + return 0; + + case VINUM_GETFREELIST: /* get a drive free list element */ + index = *(int *) data; /* get the drive index */ + fe = ((int *) data)[1]; /* and the free list element */ + if ((index >= (unsigned) vinum_conf.drives_used) /* plex doesn't exist */ + ||(DRIVE[index].state == drive_unallocated)) + return ENODEV; + if (fe >= DRIVE[index].freelist_entries) /* no such entry */ + return ENOENT; + bcopy(&DRIVE[index].freelist[fe], + data, + sizeof(struct drive_freelist)); + return 0; + + case VINUM_GETDEFECTIVE: /* get a plex defective area element */ + index = *(int *) data; /* get the plex index */ + fe = ((int *) data)[1]; /* and the region number */ + if ((index >= (unsigned) vinum_conf.plexes_used) /* plex doesn't exist */ + ||(PLEX[index].state == plex_unallocated)) + return ENODEV; + if (fe >= PLEX[index].defective_regions) /* no such entry */ + return ENOENT; + bcopy(&PLEX[index].defective_region[fe], + data, + sizeof(struct plexregion)); + return 0; + + case VINUM_GETUNMAPPED: /* get a plex unmapped area element */ + index = *(int *) data; /* get the plex index */ + fe = ((int *) data)[1]; /* and the region number */ + if ((index >= (unsigned) vinum_conf.plexes_used) /* plex doesn't exist */ + ||(PLEX[index].state == plex_unallocated)) + return ENODEV; + if (fe >= PLEX[index].unmapped_regions) /* no such entry */ + return ENOENT; + bcopy(&PLEX[index].unmapped_region[fe], + data, + sizeof(struct plexregion)); + return 0; + + case VINUM_RESETSTATS: + resetstats((struct vinum_ioctl_msg *) data); /* reset object stats */ + return 0; + + /* attach an object to a superordinate object */ + case VINUM_ATTACH: + attachobject((struct vinum_ioctl_msg *) data); + return 0; + + /* detach an object from a superordinate object */ + case VINUM_DETACH: + detachobject((struct vinum_ioctl_msg *) data); + return 0; + + /* rename an object */ + case VINUM_RENAME: + renameobject((struct vinum_rename_msg *) data); + return 0; + + /* replace an object */ + case VINUM_REPLACE: + replaceobject((struct vinum_ioctl_msg *) data); + return 0; + + default: + /* FALLTHROUGH */ + } + + default: +#if __FreeBSD__>=3 + printf("vinumioctl: type %d, sd %d, plex %d, major %x, volume %d, command %lx\n", + device->type, + device->sd, + device->plex, + device->major, + device->volume, + cmd); /* XXX */ + +#else + printf("vinumioctl: type %d, sd %d, plex %d, major %x, volume %d, command %x\n", + device->type, + device->sd, + device->plex, + device->major, + device->volume, + cmd); /* XXX */ + +#endif + return EINVAL; + + case VINUM_DRIVE_TYPE: + case VINUM_PLEX_TYPE: + return EAGAIN; /* try again next week */ + + case VINUM_SD_TYPE: + objno = SDNO(dev); + + switch (cmd) { + case VINUM_INITSD: /* initialize subdisk */ + return initsd(objno); + + default: + return EINVAL; + } + break; + + case VINUM_VOLUME_TYPE: + objno = VOLNO(dev); + + if ((unsigned) objno >= (unsigned) vinum_conf.volumes_used) /* not a valid volume */ + return ENXIO; + vol = &VOL[objno]; + if (vol->state != volume_up) /* not up, */ + return EIO; /* I/O error */ + + switch (cmd) { + case DIOCGDINFO: /* get disk label */ + get_volume_label(vol, (struct disklabel *) data); + break; + + /* Care! DIOCGPART returns *pointers* to + * the caller, so we need to store this crap as well. + * And yes, we need it. */ + case DIOCGPART: /* get partition information */ + get_volume_label(vol, &vol->label); + ((struct partinfo *) data)->disklab = &vol->label; + ((struct partinfo *) data)->part = &vol->label.d_partitions[0]; + break; + + /* We don't have this stuff on hardware, + * so just pretend to do it so that + * utilities don't get upset. */ + case DIOCWDINFO: /* write partition info */ + case DIOCSDINFO: /* set partition info */ + return 0; /* not a titty */ + + case DIOCWLABEL: /* set or reset label writeable */ + if ((flag & FWRITE) == 0) /* not writeable? */ + return EACCES; /* no, die */ + if (*(int *) data != 0) /* set it? */ + vol->flags |= VF_WLABEL; /* yes */ + else + vol->flags &= ~VF_WLABEL; /* no, reset */ + break; + + default: + return ENOTTY; /* not my kind of ioctl */ + } + break; + } + return 0; /* XXX */ +} + +/* The following four functions check the supplied + * object index and return a pointer to the object + * if it exists. Otherwise they longjump out via + * throw_rude_remark */ +struct drive * +validdrive(int driveno, struct _ioctl_reply *reply) +{ + if ((driveno < vinum_conf.drives_used) + && (DRIVE[driveno].state != drive_unallocated)) + return &DRIVE[driveno]; + strcpy(reply->msg, "No such drive"); + reply->error = ENOENT; + return NULL; +} + +struct sd * +validsd(int sdno, struct _ioctl_reply *reply) +{ + if ((sdno < vinum_conf.subdisks_used) + && (SD[sdno].state != sd_unallocated)) + return &SD[sdno]; + strcpy(reply->msg, "No such subdisk"); + reply->error = ENOENT; + return NULL; +} + +struct plex * +validplex(int plexno, struct _ioctl_reply *reply) +{ + if ((plexno < vinum_conf.plexes_used) + && (PLEX[plexno].state != plex_unallocated)) + return &PLEX[plexno]; + strcpy(reply->msg, "No such plex"); + reply->error = ENOENT; + return NULL; +} + +struct volume * +validvol(int volno, struct _ioctl_reply *reply) +{ + if ((volno < vinum_conf.volumes_used) + && (VOL[volno].state != volume_unallocated)) + return &VOL[volno]; + strcpy(reply->msg, "No such volume"); + reply->error = ENOENT; + return NULL; +} + +/* reset an object's stats */ +void +resetstats(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + + switch (msg->type) { + case drive_object: + if (msg->index < vinum_conf.drives_used) { + struct drive *drive = &DRIVE[msg->index]; + if (drive->state != drive_unallocated) { + drive->reads = 0; /* number of reads on this drive */ + drive->writes = 0; /* number of writes on this drive */ + drive->bytes_read = 0; /* number of bytes read */ + drive->bytes_written = 0; /* number of bytes written */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + case sd_object: + if (msg->index < vinum_conf.subdisks_used) { + struct sd *sd = &SD[msg->index]; + if (sd->state != sd_unallocated) { + sd->reads = 0; /* number of reads on this subdisk */ + sd->writes = 0; /* number of writes on this subdisk */ + sd->bytes_read = 0; /* number of bytes read */ + sd->bytes_written = 0; /* number of bytes written */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + break; + + case plex_object: + if (msg->index < vinum_conf.plexes_used) { + struct plex *plex = &PLEX[msg->index]; + if (plex->state != plex_unallocated) { + plex->reads = 0; + plex->writes = 0; /* number of writes on this plex */ + plex->bytes_read = 0; /* number of bytes read */ + plex->bytes_written = 0; /* number of bytes written */ + plex->multiblock = 0; /* requests that needed more than one block */ + plex->multistripe = 0; /* requests that needed more than one stripe */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + break; + + case volume_object: + if (msg->index < vinum_conf.volumes_used) { + struct volume *vol = &VOL[msg->index]; + if (vol->state != volume_unallocated) { + vol->bytes_read = 0; /* number of bytes read */ + vol->bytes_written = 0; /* number of bytes written */ + vol->reads = 0; /* number of reads on this volume */ + vol->writes = 0; /* number of writes on this volume */ + vol->recovered_reads = 0; /* reads recovered from another plex */ + reply->error = 0; + return; + } + reply->error = EINVAL; + return; + } + case invalid_object: /* can't get this */ + reply->error = EINVAL; + return; + } +} + +/* attach an object to a superior object */ +void +attachobject(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + struct sd *sd; + struct plex *plex; + struct volume *vol; + + switch (msg->type) { + case drive_object: /* you can't attach a drive to anything */ + case volume_object: /* nor a volume */ + case invalid_object: /* "this can't happen" */ + reply->error = EINVAL; + reply->msg[0] = '\0'; /* vinum(8) doesn't do this */ + return; + + case sd_object: + sd = validsd(msg->index, reply); + if (sd == NULL) /* not a valid subdisk */ + return; + plex = validplex(msg->otherobject, reply); + if (plex) { + if (sd->plexno >= 0) { /* already belong to a plex */ + reply->error = EBUSY; /* no message, the user should check */ + reply->msg[0] = '\0'; + return; + } + sd->plexoffset = msg->offset; /* this is where we want it */ + set_sd_state(sd->sdno, sd_stale, setstate_force); /* make sure it's stale */ + give_sd_to_plex(plex->plexno, sd->sdno); /* and give it to the plex */ + update_sd_config(sd->sdno, 0); + save_config(); + reply->error = 0; + } + break; + + case plex_object: + plex = validplex(msg->index, reply); /* get plex */ + if (plex == NULL) + return; + if (plex->organization != plex_concat) { /* can't attach to striped and raid-5 */ + reply->error = EINVAL; /* no message, the user should check */ + reply->msg[0] = '\0'; + return; + } + vol = validvol(msg->otherobject, reply); /* and volume information */ + if (vol) { + if ((vol->plexes == MAXPLEX) /* we have too many already */ + ||(plex->volno >= 0)) { /* or the plex has an owner */ + reply->error = EINVAL; /* no message, the user should check */ + reply->msg[0] = '\0'; + return; + } + set_plex_state(plex->plexno, plex_down, setstate_force); /* make sure it's down */ + give_plex_to_volume(msg->otherobject, msg->index); /* and give it to the volume */ + update_plex_config(plex->plexno, 0); + save_config(); + if (plex->state == plex_reviving) + reply->error = EAGAIN; /* need to revive it */ + else + reply->error = 0; + } + } +} + +/* detach an object from a superior object */ +void +detachobject(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + struct sd *sd; + struct plex *plex; + struct volume *vol; + int sdno; + int plexno; + + switch (msg->type) { + case drive_object: /* you can't attach a drive to anything */ + case volume_object: /* nor a volume */ + case invalid_object: /* "this can't happen" */ + reply->error = EINVAL; + reply->msg[0] = '\0'; /* vinum(8) doesn't do this */ + return; + + case sd_object: + sd = validsd(msg->index, reply); + if (sd == NULL) + return; + if (sd->plexno < 0) { /* doesn't belong to a plex */ + reply->error = ENOENT; + strcpy(reply->msg, "Subdisk is not attached"); + return; + } else { /* valid plex number */ + plex = &PLEX[sd->plexno]; + if ((!msg->force) /* don't force things */ + &&((plex->state == plex_up) /* and the plex is up */ + ||((plex->state == plex_flaky) && sd->state == sd_up))) { /* or flaky with this sd up */ + reply->error = EBUSY; /* we need this sd */ + reply->msg[0] = '\0'; + return; + } + sd->plexno = -1; /* anonymous sd */ + if (plex->subdisks == 1) { /* this was the only subdisk */ + Free(plex->sdnos); /* free the subdisk array */ + plex->sdnos = NULL; /* and note the fact */ + plex->subdisks_allocated = 0; /* no subdisk space */ + } else { + for (sdno = 0; sdno < plex->subdisks; sdno++) { + if (plex->sdnos[sdno] == msg->index) /* found our subdisk */ + break; + } + if (sdno < (plex->subdisks - 1)) /* not the last one, compact */ + bcopy(&plex->sdnos[sdno + 1], + &plex->sdnos[sdno], + (plex->subdisks - 1 - sdno) * sizeof(int)); + } + plex->subdisks--; + rebuild_plex_unmappedlist(plex); /* rebuild the unmapped list */ + if (!bcmp(plex->name, sd->name, strlen(plex->name))) { /* this subdisk is named after the plex */ + bcopy(sd->name, + &sd->name[3], + min(strlen(sd->name), MAXSDNAME - 3)); + bcopy("ex-", sd->name, 3); + sd->name[MAXSDNAME - 1] = '\0'; + } + update_plex_config(plex->plexno, 0); + if ((plex->organization == plex_striped) /* we've just mutilated our plex, */ + ||(plex->organization == plex_striped)) /* the data no longer matches */ + set_plex_state(plex->plexno, + plex_down, + setstate_force | setstate_configuring); + update_sd_config(sd->sdno, 0); + save_config(); + reply->error = 0; + } + return; + + case plex_object: + plex = validplex(msg->index, reply); /* get plex */ + if (plex == NULL) + return; + if (plex->volno >= 0) { + int volno = plex->volno; + + vol = &VOL[volno]; + if ((!msg->force) /* don't force things */ + &&((vol->state == volume_up) /* and the volume is up */ + &&(vol->plexes == 1))) { /* and this is the last plex */ + /* XXX As elsewhere, check whether we will lose + * mapping by removing this plex */ + reply->error = EBUSY; /* we need this plex */ + reply->msg[0] = '\0'; + return; + } + plex->volno = -1; /* anonymous plex */ + for (plexno = 0; plexno < vol->plexes; plexno++) { + if (vol->plex[plexno] == msg->index) /* found our plex */ + break; + } + if (plexno < (vol->plexes - 1)) /* not the last one, compact */ + bcopy(&vol[plexno + 1], &vol[plexno], (vol->plexes - 1 - plexno) * sizeof(int)); + vol->plexes--; + if (!bcmp(vol->name, plex->name, strlen(vol->name))) { /* this plex is named after the volume */ + /* First, check if the subdisks are the same */ + if (msg->recurse) { + int sdno; + + for (sdno = 0; sdno < plex->subdisks; sdno++) { + struct sd *sd = &SD[plex->sdnos[sdno]]; + + if (!bcmp(plex->name, sd->name, strlen(plex->name))) { /* subdisk is named after the plex */ + bcopy(sd->name, &sd->name[3], min(strlen(sd->name), MAXSDNAME - 3)); + bcopy("ex-", sd->name, 3); + sd->name[MAXSDNAME - 1] = '\0'; + } + } + } + bcopy(plex->name, &plex->name[3], min(strlen(plex->name), MAXPLEXNAME - 3)); + bcopy("ex-", plex->name, 3); + plex->name[MAXPLEXNAME - 1] = '\0'; + } + update_plex_config(plex->plexno, 0); + update_volume_config(volno, 0); + save_config(); + reply->error = 0; + } else { + reply->error = ENOENT; + strcpy(reply->msg, "Plex is not attached"); + } + } +} + +void +renameobject(struct vinum_rename_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + struct drive *drive; + struct sd *sd; + struct plex *plex; + struct volume *vol; + + switch (msg->type) { + case drive_object: /* you can't attach a drive to anything */ + if (find_drive(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + drive = validdrive(msg->index, reply); + if (drive) { + bcopy(msg->newname, drive->label.name, MAXDRIVENAME); + save_config(); + reply->error = 0; + } + return; + + case sd_object: /* you can't attach a subdisk to anything */ + if (find_subdisk(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + sd = validsd(msg->index, reply); + if (sd) { + bcopy(msg->newname, sd->name, MAXSDNAME); + update_sd_config(sd->sdno, 0); + save_config(); + reply->error = 0; + } + return; + + case plex_object: /* you can't attach a plex to anything */ + if (find_plex(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + plex = validplex(msg->index, reply); + if (plex) { + bcopy(msg->newname, plex->name, MAXPLEXNAME); + update_plex_config(plex->plexno, 0); + save_config(); + reply->error = 0; + } + return; + + case volume_object: /* you can't attach a volume to anything */ + if (find_volume(msg->newname, 0) >= 0) { /* we have that name already, */ + reply->error = EEXIST; + reply->msg[0] = '\0'; + return; + } + vol = validvol(msg->index, reply); + if (vol) { + bcopy(msg->newname, vol->name, MAXVOLNAME); + update_volume_config(msg->index, 0); + save_config(); + reply->error = 0; + } + return; + + case invalid_object: + reply->error = EINVAL; + reply->msg[0] = '\0'; + } +} + +/* Replace one object with another */ +void +replaceobject(struct vinum_ioctl_msg *msg) +{ + struct _ioctl_reply *reply = (struct _ioctl_reply *) msg; + + reply->error = ENODEV; /* until I know how to do this */ + strcpy(reply->msg, "replace not implemented yet"); +/* save_config (); */ +} diff --git a/sys/modules/vinum/vinumkw.h b/sys/modules/vinum/vinumkw.h new file mode 100644 index 0000000..1a81f37 --- /dev/null +++ b/sys/modules/vinum/vinumkw.h @@ -0,0 +1,120 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumkw.h,v 1.7 1998/08/07 02:35:51 grog Exp grog $ + */ + +/* Command keywords that vinum knows. These include both user-level + * and kernel-level stuff */ + +/* Our complete vocabulary. The names of the commands are + * the same as the identifier without the kw_ at the beginning + * (i.e. kw_create defines the "create" keyword). Preprocessor + * magic in parser.c does the rest. */ +enum keyword { + kw_create, + kw_modify, + kw_list, + kw_l = kw_list, + kw_ld, /* list drive */ + kw_ls, /* list subdisk */ + kw_lp, /* list plex */ + kw_lv, /* list volume */ + kw_set, + kw_rm, + kw_start, + kw_stop, + kw_drive, + kw_sd, + kw_subdisk = kw_sd, + kw_plex, + kw_volume, + kw_vol = kw_volume, + kw_read, + kw_readpol, + kw_org, + kw_name, + kw_concat, + kw_striped, + kw_raid5, + kw_driveoffset, + kw_plexoffset, + kw_len, + kw_length = kw_len, + kw_state, + kw_setupstate, + kw_d, /* flag names */ + kw_f, + kw_r, + kw_s, + kw_v, + kw_round, /* round robin */ + kw_prefer, /* prefer plex */ + kw_device, + kw_init, + kw_label, + kw_resetconfig, + kw_writethrough, + kw_writeback, + kw_raw, + kw_resetstats, + kw_attach, + kw_detach, + kw_rename, + kw_printconfig, + kw_replace, + kw_detached, +#ifdef DEBUG + kw_debug, /* go into debugger */ + kw_info, +#endif + kw_invalid_keyword = -1 +}; + +struct _keywords { + char *name; + enum keyword keyword; +}; + +struct keywordset { + int size; + struct _keywords *k; +}; + +extern struct _keywords keywords[]; +extern struct _keywords flag_keywords[]; + +extern struct keywordset keyword_set; +extern struct keywordset flag_set; diff --git a/sys/modules/vinum/vinumstate.h b/sys/modules/vinum/vinumstate.h new file mode 100644 index 0000000..0b07590 --- /dev/null +++ b/sys/modules/vinum/vinumstate.h @@ -0,0 +1,213 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumstate.h,v 1.11 1998/08/04 06:22:49 grog Exp grog $ + */ + +/* This file gets read by makestatetext to create text files + * with the names of the states, so don't change the file + * format */ + +enum volumestate { + volume_unallocated, + /* present but unused. Must be 0 */ + + volume_uninit, + /* mentioned elsewhere but not defined */ + + volume_down, + + /* The volume is up and functional, but not all plexes may be available */ + volume_up, + volume_laststate = volume_up /* last value, for table dimensions */ +}; + +enum plexstate { + /* An empty entry, not a plex at all. */ + plex_unallocated, + + /* The plex has been allocated, but there configuration + * is not complete */ + plex_init, + + /* A plex which has gone completely down because of + * I/O errors. */ + plex_faulty, + + /* A plex which has been taken down by the + * administrator. */ + plex_down, + + /* A plex which is currently being brought up after + * being not up. This involves copying data from + * another plex */ + plex_reviving, + + /* A plex which is being initialized */ + plex_initializing, + + /* *** The remaining states represent plexes which are + * at least partially up. Keep these separate so that + * they can be checked more easily. */ + + /* A plex entry which is at least partially up. Not + * all subdisks are available, and an inconsistency + * has occurred. If no other plex is uncorrupted, + * the volume is no longer consistent. */ + plex_corrupt, + + plex_firstup = plex_corrupt, /* first "up" state */ + + /* A plex entry which is at least partially up. Not + * all subdisks are available, but so far no + * inconsistency has occurred (this will change with + * the first write to the address space occupied by + * a defective subdisk). A RAID 5 plex with one subdisk + * down will remain degraded even after a write */ + plex_degraded, + + /* A plex which is really up, but which has a reborn + * subdisk which we don't completely trust, and + * which we don't want to read if we can avoid it */ + plex_flaky, + + /* A plex entry which is completely up. All subdisks + * are up. */ + plex_up, + + plex_laststate = plex_up /* last value, for table dimensions */ +}; + +/* subdisk states */ +enum sdstate { + /* An empty entry, not a subdisk at all. */ + sd_unallocated, + + /* A subdisk entry which has not been created + * completely. Some fields may be empty. + */ + sd_uninit, + + /* A subdisk entry which has been created completely. + * All fields are correct, but the disk hasn't + * been updated. + */ + sd_init, + + /* A subdisk entry which has been created completely and + * which is currently being initialized */ + sd_initializing, + + /* A subdisk entry which has been created completely. + * All fields are correct, and the disk has been + * updated, but there is no data on the disk. + */ + sd_empty, + + /* *** The following states represent invalid data */ + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down, and as a result updates have been + * missed. + */ + sd_obsolete, + + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down, updates have been lost, and then + * the drive came up again. + */ + sd_stale, + + /* *** The following states represent valid, inaccessible data */ + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down. No attempt has been made to write + * to the subdisk since the crash. + */ + sd_crashed, + + /* A subdisk entry which was up, which contained + * valid data, and which was taken down by the + * administrator. The data is valid. */ + sd_down, + + /* *** The following states represent accessible subdisks + * with valid data */ + + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data was valid, but since then the drive + * has gone down and up again. No updates were lost, + * but it is possible that the subdisk has been + * damaged. We won't read from this subdisk if we + * have a choice. If this is the only subdisk which + * covers this address space in the plex, we set its + * state to sd_up under these circumstances, so this + * status implies that there is another subdisk to + * fulfil the request. + */ + sd_reborn, + + /* A subdisk entry which has been created completely. + * All fields are correct, the disk has been updated, + * and the data is valid. + */ + sd_up, + + sd_laststate = sd_up /* last value, for table dimensions */ +}; + +enum drivestate { + drive_unallocated, + /* present but unused. Must be 0 */ + + drive_uninit, + /* just mentioned in some other config entry */ + + drive_down, + /* not accessible */ + + drive_coming_up, + /* in the process of being brought up */ + + drive_up, + /* up and running */ + + drive_laststate = drive_up /* last value, for table dimensions */ +}; diff --git a/sys/modules/vinum/vinumvar.h b/sys/modules/vinum/vinumvar.h new file mode 100644 index 0000000..ca54f76 --- /dev/null +++ b/sys/modules/vinum/vinumvar.h @@ -0,0 +1,510 @@ +/*- + * Copyright (c) 1997, 1998 + * Nan Yang Computer Services Limited. All rights reserved. + * + * This software is distributed under the so-called ``Berkeley + * License'': + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by Nan Yang Computer + * Services Limited. + * 4. Neither the name of the Company nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * This software is provided ``as is'', and any express or implied + * warranties, including, but not limited to, the implied warranties of + * merchantability and fitness for a particular purpose are disclaimed. + * In no event shall the company or contributors be liable for any + * direct, indirect, incidental, special, exemplary, or consequential + * damages (including, but not limited to, procurement of substitute + * goods or services; loss of use, data, or profits; or business + * interruption) however caused and on any theory of liability, whether + * in contract, strict liability, or tort (including negligence or + * otherwise) arising in any way out of the use of this software, even if + * advised of the possibility of such damage. + * + * $Id: vinumvar.h,v 1.15 1998/08/14 06:36:41 grog Exp grog $ + */ + +/* XXX gdb can't find our global pointers, so use this kludge to + * point to them locally. Remove after testing */ +#define BROKEN_GDB struct _vinum_conf *VC = &vinum_conf + +#include <sys/time.h> +#include "vinumstate.h" +/* Some configuration maxima. They're an enum because + * we can't define global constants. Sorry about that. + * + * These aren't as bad as they look: most of them + * are soft limits. Only the MAXCONFIG parameter is set in stone + */ + +enum constants { + VINUM_HEADER = 512, /* size of header on disk */ + MAXCONFIGLINE = 1024, /* maximum size of a single config line */ + /* XXX Do we still need this? */ + MINVINUMSLICE = 1048576, /* minimum size of a slice */ + + CDEV_MAJOR = 91, /* major number for character device */ + BDEV_MAJOR = 25, /* and block device */ + + ROUND_ROBIN_READPOL = -1, /* round robin read policy */ + + /* type field in minor number */ + VINUM_VOLUME_TYPE = 0, + VINUM_PLEX_TYPE = 1, + VINUM_SD_TYPE = 2, + VINUM_DRIVE_TYPE = 3, + VINUM_SUPERDEV_TYPE = 4, /* super device. */ + + /* Shifts for the individual fields in the device */ + VINUM_TYPE_SHIFT = 28, + VINUM_VOL_SHIFT = 0, + VINUM_PLEX_SHIFT = 16, + VINUM_SD_SHIFT = 20, + VINUM_VOL_WIDTH = 8, + VINUM_PLEX_WIDTH = 3, + VINUM_SD_WIDTH = 8, + MAJORDEV_SHIFT = 8, + + +/* Create a block device number */ +#define VINUMBDEV(v,p,s,t) ((BDEV_MAJOR << MAJORDEV_SHIFT) \ + | (v << VINUM_VOL_SHIFT) \ + | (p << VINUM_PLEX_SHIFT) \ + | (s << VINUM_SD_SHIFT) \ + | (t << VINUM_TYPE_SHIFT) ) + +/* And a character device number */ +#define VINUMCDEV(v,p,s,t) ((CDEV_MAJOR << MAJORDEV_SHIFT) \ + | (v << VINUM_VOL_SHIFT) \ + | (p << VINUM_PLEX_SHIFT) \ + | (s << VINUM_SD_SHIFT) \ + | (t << VINUM_TYPE_SHIFT) ) + +/* extract device type */ +#define DEVTYPE(x) ((x >> VINUM_TYPE_SHIFT) & 7) + +/* extract volume number */ +#define VOLNO(x) (x & ((1 << VINUM_VOL_WIDTH) - 1)) + +/* extract plex number */ +#define PLEXNO(x) (VOL [VOLNO (x)].plex [(x >> VINUM_PLEX_SHIFT) & ((1 << VINUM_PLEX_WIDTH) - 1)]) + +/* extract subdisk number */ +#define SDNO(x) (PLEX [PLEXNO (x)].sdnos [(x >> VINUM_SD_SHIFT) & ((1 << VINUM_SD_WIDTH) - 1)]) + +/* extract drive number */ +#define DRIVENO(x) (SD [SDNO (x)].driveno) + + VINUM_SUPERDEV = VINUMBDEV(0, 0, 0, VINUM_SUPERDEV_TYPE), /* superdevice number */ + +/* the number of object entries to cater for initially, and also the + * value by which they are incremented. It doesn't take long + * to extend them, so theoretically we could start with 1 of each, but + * it's untidy to allocate such small areas. These values are + * probably too small. + */ + + INITIAL_DRIVES = 4, + INITIAL_VOLUMES = 4, + INITIAL_PLEXES = 8, + INITIAL_SUBDISKS = 16, + INITIAL_SUBDISKS_IN_PLEX = 4, /* number of subdisks to allocate to a plex */ + INITIAL_SUBDISKS_IN_DRIVE = 4, /* number of subdisks to allocate to a drive */ + INITIAL_DRIVE_FREELIST = 16, /* number of entries in drive freelist */ + PLEX_REGION_TABLE_SIZE = 8, /* number of entries in plex region tables */ + INITIAL_LOCKS = 8, /* number of locks to allocate to a volume */ + DEFAULT_REVIVE_BLOCKSIZE = 32768, /* size of block to transfer in one op */ +}; + +/* device numbers */ + +/* + * 31 30 28 27 20 19 18 16 15 8 7 0 + * |-----------------------------------------------------------------------------------------------| + * |X | Type | Subdisk number | X| Plex | Major number | volume number | + * |-----------------------------------------------------------------------------------------------| + * + * 0x2 03 1 19 06 + */ +struct devcode { +/* CARE. These fields assume a big-endian word. On a + * little-endian system, they're the wrong way around */ + unsigned volume:8; /* up to 256 volumes */ + unsigned major:8; /* this is where the major number fits */ + unsigned plex:3; /* up to 8 plexes per volume */ + unsigned unused:1; /* up for grabs */ + unsigned sd:8; /* up to 256 subdisks per plex */ + unsigned type:3; /* type of object */ + /* type field + VINUM_VOLUME = 0, + VINUM_PLEX = 1, + VINUM_SUBDISK = 2, + VINUM_DRIVE = 3, + VINUM_SUPERDEV = 4, */ + unsigned signbit:1; /* to make 32 bits */ +}; + +#define VINUM_DIR "/dev/vinum" +#define VINUM_RDIR "/dev/rvinum" +#define VINUM_SUPERDEV_NAME VINUM_DIR"/control" +#define MAXDRIVENAME 32 /* maximum length of a device name */ +#define MAXSDNAME 64 /* maximum length of a subdisk name */ +#define MAXPLEXNAME 64 /* maximum length of a plex name */ +#define MAXVOLNAME 64 /* maximum length of a volume name */ +#define MAXNAME 64 /* maximum length of any name */ +#define MAXVOLPLEX 8 /* maximum number of plexes in a volume */ + +/* Flags for all objects. Most of them only apply to + * specific objects, but we have space for all in any + * 32 bit flags word. */ +enum objflags { + VF_LOCKED = 1, /* somebody has locked access to this object */ + VF_LOCKING = 2, /* we want access to this object */ + VF_WRITETHROUGH = 8, /* volume: write through */ + VF_INITED = 0x10, /* unit has been initialized */ + VF_WLABEL = 0x20, /* label area is writable */ + VF_LABELLING = 0x40, /* unit is currently being labelled */ + VF_WANTED = 0x80, /* someone is waiting to obtain a lock */ + VF_RAW = 0x100, /* raw volume (no file system) */ + VF_LOADED = 0x200, /* module is loaded */ + VF_CONFIGURING = 0x400, /* somebody is changing the config */ + VF_WILL_CONFIGURE = 0x800, /* somebody wants to change the config */ + VF_CONFIG_INCOMPLETE = 0x1000, /* haven't finished changing the config */ + VF_CONFIG_SETUPSTATE = 0x2000, /* set a volume up if all plexes are empty */ + VF_READING_CONFIG = 0x4000, /* we're reading config database from disk */ + VF_KERNELOP = 0x8000, /* we're performing ops from kernel space */ +}; + +/* Global configuration information for the vinum subsystem */ +struct _vinum_conf { + /* Pointers to vinum structures */ + struct drive *drive; + struct sd *sd; + struct plex *plex; + struct volume *volume; + + /* the number allocated */ + int drives_allocated; + int subdisks_allocated; + int plexes_allocated; + int volumes_allocated; + + /* and the number currently in use */ + int drives_used; + int subdisks_used; + int plexes_used; + int volumes_used; + + int flags; + int opencount; /* number of times we've been opened */ +#if DEBUG + int lastrq; + struct buf *lastbuf; +#endif +}; + +/* Use these defines to simplify code */ +#define DRIVE vinum_conf.drive +#define SD vinum_conf.sd +#define PLEX vinum_conf.plex +#define VOL vinum_conf.volume +#define VFLAGS vinum_conf.flags + +/* Slice header + + * Vinum drives start with this structure: + * + * Sector + * |--------------------------------------| + * | PDP-11 memorial boot block | 0 + * |--------------------------------------| + * | Disk label, maybe | 1 + * |--------------------------------------| + * | Slice definition (vinum_hdr) | 2 + * |--------------------------------------| + * | | + * | Configuration info, first copy | 3 + * | | + * |--------------------------------------| + * | | + * | Configuration info, second copy | 3 + size of config + * | | + * |--------------------------------------| + */ + +/* Sizes and offsets of our information */ +enum { + VINUM_LABEL_OFFSET = 4096, /* offset of vinum label */ + VINUMHEADERLEN = 512, /* size of vinum label */ + VINUM_CONFIG_OFFSET = 4608, /* offset of first config copy */ + MAXCONFIG = 65536, /* and size of config copy */ + DATASTART = (MAXCONFIG * 2 + VINUM_CONFIG_OFFSET) / DEV_BSIZE /* this is where the data starts */ +}; + +/* hostname is 256 bytes long, but we don't need to shlep + * multiple copies in vinum. We use the host name just + * to identify this system, and 32 bytes should be ample + * for that purpose */ +#define VINUMHOSTNAMELEN 32 + +struct vinum_label { + char sysname[VINUMHOSTNAMELEN]; /* system name at time of creation */ + char name[MAXDRIVENAME]; /* our name of the drive */ + struct timeval date_of_birth; /* the time it was created */ + struct timeval last_update; /* and the time of last update */ + off_t drive_size; /* total size in bytes of the drive. + * This value includes the headers */ +}; + +struct vinum_hdr { + long long magic; /* we're long on magic numbers */ + /* XXX Get these right for big-endian */ +#define VINUM_MAGIC 22322600044678729LL /* should be this */ +#define VINUM_NOMAGIC 22322600044678990LL /* becomes this after obliteration */ + int config_length; /* size in bytes of each copy of the + * configuration info. + * This must be a multiple of the sector size. */ + + struct vinum_label label; /* unique label */ +}; + +/* Information returned from read_drive_label */ +enum drive_label_info { + DL_CANT_OPEN, /* invalid partition */ + DL_NOT_OURS, /* valid partition, but no vinum label */ + DL_DELETED_LABEL, /* valid partition, deleted label found */ + DL_WRONG_DRIVE, /* drive name doesn't match */ + DL_OURS /* valid partition and label found */ +}; + +/*** Drive definitions ***/ +/* A drive corresponds to a disk slice. We use a different term to show + * the difference in usage: it doesn't have to be a slice, and could + * theroretically be a complete, unpartitioned disk */ + +struct drive { + enum drivestate state; /* current state */ + int subdisks_allocated; /* number of entries in sd */ + int subdisks_used; /* and the number used */ + int blocksize; /* size of fs blocks */ + u_int64_t sectors_available; /* number of sectors still available */ + int secsperblock; + int lasterror; /* last error on drive */ + int driveno; /* index of drive in vinum_conf */ + int opencount; /* number of up subdisks */ + u_int64_t reads; /* number of reads on this drive */ + u_int64_t writes; /* number of writes on this drive */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + dev_t dev; /* and device number */ + char devicename[MAXDRIVENAME]; /* name of the slice it's on */ + struct vnode *vp; /* vnode pointer */ + struct proc *p; + struct vinum_label label; /* and the label information */ + struct partinfo partinfo; /* partition information */ + int freelist_size; /* number of entries alloced in free list */ + int freelist_entries; /* number of entries used in free list */ + struct drive_freelist { /* sorted list of free space on drive */ + u_int64_t offset; + long sectors; + } *freelist; +}; + +/*** Subdisk definitions ***/ + +struct sd { + enum sdstate state; /* state */ + /* offsets in blocks */ + int64_t driveoffset; /* offset on drive */ + int64_t plexoffset; /* offset in plex */ + u_int64_t sectors; /* and length in sectors */ + int plexno; /* index of plex, if it belongs */ + int driveno; /* index of the drive on which it is located */ + int sdno; /* our index in vinum_conf */ + int pid; /* pid of process which opened us */ + u_int64_t reads; /* number of reads on this subdisk */ + u_int64_t writes; /* number of writes on this subdisk */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + char name[MAXSDNAME]; /* name of subdisk */ +}; + +/*** Plex definitions ***/ + +/* kinds of plex organization */ +enum plexorg { + plex_disorg, /* disorganized */ + plex_concat, /* concatenated plex */ + plex_striped, /* striped plex */ + plex_raid5 /* RAID5 plex */ +}; + +/* Region in plex (either defective or unmapped) */ +struct plexregion { + u_int64_t offset; /* start of region */ + u_int64_t length; /* length */ +}; + +struct plex { + enum plexorg organization; /* Plex organization */ + enum plexstate state; /* and current state */ + u_int64_t length; /* total length of plex (max offset) */ + int flags; + int stripesize; /* size of stripe or raid band, in sectors */ + int subdisks; /* number of associated subdisks */ + int subdisks_allocated; /* number of subdisks allocated space for */ + int *sdnos; /* list of component subdisks */ + int plexno; /* index of plex in vinum_conf */ + int volno; /* index of volume */ + int volplexno; /* number of plex in volume */ + int pid; /* pid of process which opened us */ + /* Lock information */ + int locks; /* number of locks used */ + int alloclocks; /* number of locks allocated */ + struct rangelock *lock; /* ranges of locked addresses */ + /* Statistics */ + u_int64_t reads; /* number of reads on this plex */ + u_int64_t writes; /* number of writes on this plex */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + u_int64_t multiblock; /* requests that needed more than one block */ + u_int64_t multistripe; /* requests that needed more than one stripe */ + /* revive parameters */ + u_int64_t revived; /* block number of current revive request */ + int revive_blocksize; /* revive block size (bytes) */ + int revive_interval; /* and time to wait between transfers */ + struct request *waitlist; /* list of requests waiting on revive op */ + /* geometry control */ + int defective_regions; /* number of regions which are defective */ + int defective_region_count; /* number of entries in defective_region */ + struct plexregion *defective_region; /* list of offset/length pairs: defective sds */ + int unmapped_regions; /* number of regions which are missing */ + int unmapped_region_count; /* number of entries in unmapped_region */ + struct plexregion *unmapped_region; /* list of offset/length pairs: missing sds */ + char name[MAXPLEXNAME]; /* name of plex */ +}; + +/*** Volume definitions ***/ + +#define MAXPLEX 8 /* maximum number of plexes */ + + +struct volume { + enum volumestate state; /* current state */ + int plexes; /* number of plexes */ + int preferred_plex; /* plex to read from, -1 for round-robin */ + int last_plex_read; /* index of plex used for last read, + * for round-robin */ + dev_t devno; /* device number */ + int flags; /* status and configuration flags */ + int opencount; /* number of opens (all the same process) */ + int openflags; /* flags supplied to last open(2) */ + u_int64_t size; /* size of volume */ + int disk; /* disk index */ + int blocksize; /* logical block size */ + int active; /* number of outstanding requests active */ + int subops; /* and the number of suboperations */ + pid_t pid; /* pid of locker */ + /* Statistics */ + u_int64_t bytes_read; /* number of bytes read */ + u_int64_t bytes_written; /* number of bytes written */ + u_int64_t reads; /* number of reads on this volume */ + u_int64_t writes; /* number of writes on this volume */ + u_int64_t recovered_reads; /* reads recovered from another plex */ + /* Unlike subdisks in the plex, space for the plex pointers is static */ + int plex[MAXPLEX]; /* index of plexes */ + char name[MAXVOLNAME]; /* name of volume */ + struct disklabel label; /* for DIOCGPART */ +}; + +/* Table expansion. Expand table, which contains oldcount + * entries of type element, by increment entries, and change + * oldcount accordingly */ +#define EXPAND(table, element, oldcount, increment) \ +{ \ + expand_table ((void **) &table, \ + oldcount * sizeof (element), \ + (oldcount + increment) * sizeof (element) ); \ + oldcount += increment; \ + } + +/* Information on vinum's memory usage */ +struct meminfo { + int mallocs; /* number of malloced blocks */ + int total_malloced; /* total amount malloced */ + int highwater; /* maximum number of mallocs */ + struct mc *malloced; /* pointer to kernel table */ +}; + +struct mc { + int seq; + int size; + short line; + short flags; +#define ALLOC_KVA 1 /* allocated via kva calls */ + int *databuf; /* really vm_object_t */ + caddr_t address; + char file[16]; +}; + +/* These enums are used by the state transition + * routines. They're in bit map format: + * + * Bit 0: Other plexes in the volume are down + * Bit 1: Other plexes in the volume are up + * Bit 2: The current plex is up + * Maybe they should be local to + * state.c */ +enum volplexstate { + volplex_onlyusdown = 0, /* we're the only plex, and we're down */ + volplex_alldown, /* 1: another plex is down, and so are we */ + volplex_otherup, /* 2: another plex is up */ + volplex_otherupdown, /* other plexes are up and down */ + volplex_onlyus, /* 4: we're up and alone */ + volplex_onlyusup, /* only we are up, others are down */ + volplex_allup, /* all plexes are up */ + volplex_someup /* some plexes are up, including us */ +}; + +/* state map for plex */ +enum sdstates { + sd_emptystate = 1, + sd_downstate = 2, /* found an SD which is down */ + sd_crashedstate = 4, /* found an SD which is crashed */ + sd_obsoletestate = 8, /* found an SD which is obsolete */ + sd_stalestate = 16, /* found an SD which is stale */ + sd_rebornstate = 32, /* found an SD which is reborn */ + sd_upstate = 64, /* found an SD which is up */ + sd_initstate = 128, /* found an SD which is init */ + sd_otherstate = 256 /* found an SD in some other state */ +}; + +/* This is really just a parameter to pass to + * set_<foo>_state, but since it needs to be known + * in the external definitions, we need to define + * it here */ +enum setstateflags { + setstate_none = 0, /* no flags */ + setstate_force = 1, /* force the state change */ + setstate_configuring = 2, /* we're currently configuring, don't save */ + setstate_recursing = 4, /* we're called from another setstate function */ + setstate_norecurse = 8 /* don't call other setstate functions */ +}; + +#ifdef DEBUG +/* Debugging stuff */ +#define DEBUG_ADDRESSES 1 +#define DEBUG_NUMOUTPUT 2 +#endif |