summaryrefslogtreecommitdiffstats
path: root/sys/dev/ccd
diff options
context:
space:
mode:
Diffstat (limited to 'sys/dev/ccd')
-rw-r--r--sys/dev/ccd/ccd.c1014
1 files changed, 675 insertions, 339 deletions
diff --git a/sys/dev/ccd/ccd.c b/sys/dev/ccd/ccd.c
index c8d33a4..b70f8c2 100644
--- a/sys/dev/ccd/ccd.c
+++ b/sys/dev/ccd/ccd.c
@@ -1,3 +1,5 @@
+/* $FreeBSD$ */
+
/* $NetBSD: ccd.c,v 1.22 1995/12/08 19:13:26 thorpej Exp $ */
/*
@@ -85,77 +87,122 @@
* Moffett Field, CA 94035
*/
+#include "ccd.h"
+#if NCCD > 0
+
#include <sys/param.h>
#include <sys/systm.h>
+#include <sys/kernel.h>
+#include <sys/module.h>
#include <sys/proc.h>
-#include <sys/errno.h>
-#include <sys/dkstat.h>
#include <sys/buf.h>
#include <sys/malloc.h>
#include <sys/namei.h>
#include <sys/conf.h>
#include <sys/stat.h>
-#include <sys/ioctl.h>
+#include <sys/sysctl.h>
#include <sys/disklabel.h>
-#include <sys/device.h>
-#include <sys/disk.h>
-#include <sys/syslog.h>
+#include <ufs/ffs/fs.h>
+#include <sys/devicestat.h>
#include <sys/fcntl.h>
#include <sys/vnode.h>
-#include <dev/ccdvar.h>
+#include <sys/ccdvar.h>
+
+#include <vm/vm_zone.h>
#if defined(CCDDEBUG) && !defined(DEBUG)
#define DEBUG
#endif
#ifdef DEBUG
-int ccddebug = 0x00;
#define CCDB_FOLLOW 0x01
#define CCDB_INIT 0x02
#define CCDB_IO 0x04
#define CCDB_LABEL 0x08
#define CCDB_VNODE 0x10
+static int ccddebug = CCDB_FOLLOW | CCDB_INIT | CCDB_IO | CCDB_LABEL |
+ CCDB_VNODE;
+SYSCTL_INT(_debug, OID_AUTO, ccddebug, CTLFLAG_RW, &ccddebug, 0, "");
+#undef DEBUG
#endif
-#define ccdunit(x) DISKUNIT(x)
+#define ccdunit(x) dkunit(x)
+#define ccdpart(x) dkpart(x)
+
+/*
+ This is how mirroring works (only writes are special):
+
+ When initiating a write, ccdbuffer() returns two "struct ccdbuf *"s
+ linked together by the cb_mirror field. "cb_pflags &
+ CCDPF_MIRROR_DONE" is set to 0 on both of them.
+
+ When a component returns to ccdiodone(), it checks if "cb_pflags &
+ CCDPF_MIRROR_DONE" is set or not. If not, it sets the partner's
+ flag and returns. If it is, it means its partner has already
+ returned, so it will go to the regular cleanup.
+
+ */
struct ccdbuf {
struct buf cb_buf; /* new I/O buf */
struct buf *cb_obp; /* ptr. to original I/O buf */
+ struct ccdbuf *cb_freenext; /* free list link */
int cb_unit; /* target unit */
int cb_comp; /* target component */
+ int cb_pflags; /* mirror/parity status flag */
+ struct ccdbuf *cb_mirror; /* mirror counterpart */
};
-#define getccdbuf() \
- ((struct ccdbuf *)malloc(sizeof(struct ccdbuf), M_DEVBUF, M_WAITOK))
-#define putccdbuf(cbp) \
- free((caddr_t)(cbp), M_DEVBUF)
+/* bits in cb_pflags */
+#define CCDPF_MIRROR_DONE 1 /* if set, mirror counterpart is done */
#define CCDLABELDEV(dev) \
- (MAKEDISKDEV(major((dev)), ccdunit((dev)), RAW_PART))
-
-/* {b,c}devsw[] function prototypes */
-dev_type_open(ccdopen);
-dev_type_close(ccdclose);
-dev_type_strategy(ccdstrategy);
-dev_type_ioctl(ccdioctl);
-dev_type_read(ccdread);
-dev_type_write(ccdwrite);
+ (makedev(major((dev)), dkmakeminor(ccdunit((dev)), 0, RAW_PART)))
+
+static d_open_t ccdopen;
+static d_close_t ccdclose;
+static d_strategy_t ccdstrategy;
+static d_ioctl_t ccdioctl;
+static d_dump_t ccddump;
+static d_psize_t ccdsize;
+
+#define NCCDFREEHIWAT 16
+
+#define CDEV_MAJOR 74
+#define BDEV_MAJOR 21
+
+static struct cdevsw ccd_cdevsw = {
+ /* open */ ccdopen,
+ /* close */ ccdclose,
+ /* read */ physread,
+ /* write */ physwrite,
+ /* ioctl */ ccdioctl,
+ /* poll */ nopoll,
+ /* mmap */ nommap,
+ /* strategy */ ccdstrategy,
+ /* name */ "ccd",
+ /* maj */ CDEV_MAJOR,
+ /* dump */ ccddump,
+ /* psize */ ccdsize,
+ /* flags */ D_DISK,
+ /* bmaj */ BDEV_MAJOR
+};
-/* called by main() at boot time */
-void ccdattach __P((int));
+/* called during module initialization */
+static void ccdattach __P((void));
+static int ccd_modevent __P((module_t, int, void *));
/* called by biodone() at interrupt time */
-void ccdiodone __P((struct ccdbuf *cbp));
+static void ccdiodone __P((struct ccdbuf *cbp));
static void ccdstart __P((struct ccd_softc *, struct buf *));
static void ccdinterleave __P((struct ccd_softc *, int));
static void ccdintr __P((struct ccd_softc *, struct buf *));
static int ccdinit __P((struct ccddevice *, char **, struct proc *));
static int ccdlookup __P((char *, struct proc *p, struct vnode **));
-static struct ccdbuf *ccdbuffer __P((struct ccd_softc *, struct buf *,
- daddr_t, caddr_t, long));
+static void ccdbuffer __P((struct ccdbuf **ret, struct ccd_softc *,
+ struct buf *, daddr_t, caddr_t, long));
static void ccdgetdisklabel __P((dev_t));
static void ccdmakedisklabel __P((struct ccd_softc *));
static int ccdlock __P((struct ccd_softc *));
@@ -168,24 +215,98 @@ static void printiinfo __P((struct ccdiinfo *));
/* Non-private for the benefit of libkvm. */
struct ccd_softc *ccd_softc;
struct ccddevice *ccddevs;
-int numccd = 0;
+struct ccdbuf *ccdfreebufs;
+static int numccdfreebufs;
+static int numccd = 0;
/*
- * Called by main() during pseudo-device attachment. All we need
- * to do is allocate enough space for devices to be configured later.
+ * getccdbuf() - Allocate and zero a ccd buffer.
+ *
+ * This routine is called at splbio().
*/
+
+static __inline
+struct ccdbuf *
+getccdbuf(struct ccdbuf *cpy)
+{
+ struct ccdbuf *cbp;
+
+ /*
+ * Allocate from freelist or malloc as necessary
+ */
+ if ((cbp = ccdfreebufs) != NULL) {
+ ccdfreebufs = cbp->cb_freenext;
+ --numccdfreebufs;
+ } else {
+ cbp = malloc(sizeof(struct ccdbuf), M_DEVBUF, M_WAITOK);
+ }
+
+ /*
+ * Used by mirroring code
+ */
+ if (cpy)
+ bcopy(cpy, cbp, sizeof(struct ccdbuf));
+ else
+ bzero(cbp, sizeof(struct ccdbuf));
+
+ /*
+ * independant struct buf initialization
+ */
+ LIST_INIT(&cbp->cb_buf.b_dep);
+ BUF_LOCKINIT(&cbp->cb_buf);
+ BUF_LOCK(&cbp->cb_buf, LK_EXCLUSIVE);
+ BUF_KERNPROC(&cbp->cb_buf);
+
+ return(cbp);
+}
+
+/*
+ * putccdbuf() - Free a ccd buffer.
+ *
+ * This routine is called at splbio().
+ */
+
+static __inline
void
-ccdattach(num)
- int num;
+putccdbuf(struct ccdbuf *cbp)
{
- int i;
+ BUF_UNLOCK(&cbp->cb_buf);
+ BUF_LOCKFREE(&cbp->cb_buf);
+
+ if (numccdfreebufs < NCCDFREEHIWAT) {
+ cbp->cb_freenext = ccdfreebufs;
+ ccdfreebufs = cbp;
+ ++numccdfreebufs;
+ } else {
+ free((caddr_t)cbp, M_DEVBUF);
+ }
+}
+
- if (num <= 0) {
-#ifdef DIAGNOSTIC
- panic("ccdattach: count <= 0");
+/*
+ * Number of blocks to untouched in front of a component partition.
+ * This is to avoid violating its disklabel area when it starts at the
+ * beginning of the slice.
+ */
+#if !defined(CCD_OFFSET)
+#define CCD_OFFSET 16
#endif
- return;
- }
+
+/*
+ * 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.
+ */
+static void
+ccdattach()
+{
+ int i;
+ int num = NCCD;
+
+ if (num > 1)
+ printf("ccd0-%d: Concatenated disk drivers\n", num-1);
+ else
+ printf("ccd0: Concatenated disk driver\n");
ccd_softc = (struct ccd_softc *)malloc(num * sizeof(struct ccd_softc),
M_DEVBUF, M_NOWAIT);
@@ -203,38 +324,61 @@ ccdattach(num)
bzero(ccd_softc, num * sizeof(struct ccd_softc));
bzero(ccddevs, num * sizeof(struct ccddevice));
+ cdevsw_add(&ccd_cdevsw);
/* XXX: is this necessary? */
for (i = 0; i < numccd; ++i)
ccddevs[i].ccd_dk = -1;
}
static int
+ccd_modevent(mod, type, data)
+ module_t mod;
+ int type;
+ void *data;
+{
+ int error = 0;
+
+ switch (type) {
+ case MOD_LOAD:
+ ccdattach();
+ break;
+
+ case MOD_UNLOAD:
+ printf("ccd0: Unload not supported!\n");
+ error = EOPNOTSUPP;
+ break;
+
+ default: /* MOD_SHUTDOWN etc */
+ break;
+ }
+ return (error);
+}
+
+DEV_MODULE(ccd, ccd_modevent, NULL);
+
+static int
ccdinit(ccd, cpaths, p)
struct ccddevice *ccd;
char **cpaths;
struct proc *p;
{
- register struct ccd_softc *cs = &ccd_softc[ccd->ccd_unit];
- register struct ccdcinfo *ci;
- register size_t size;
- register int ix;
+ struct ccd_softc *cs = &ccd_softc[ccd->ccd_unit];
+ struct ccdcinfo *ci = NULL; /* XXX */
+ size_t size;
+ int ix;
struct vnode *vp;
- struct vattr va;
size_t minsize;
int maxsecsize;
struct partinfo dpart;
struct ccdgeom *ccg = &cs->sc_geom;
char tmppath[MAXPATHLEN];
- int error;
+ int error = 0;
#ifdef DEBUG
if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
printf("ccdinit: unit %d\n", ccd->ccd_unit);
#endif
-#ifdef WORKING_DISK_STATISTICS /* XXX !! */
- cs->sc_dk = ccd->ccd_dk;
-#endif
cs->sc_size = 0;
cs->sc_ileave = ccd->ccd_interleave;
cs->sc_nccdisks = ccd->ccd_ndev;
@@ -258,71 +402,51 @@ ccdinit(ccd, cpaths, p)
* Copy in the pathname of the component.
*/
bzero(tmppath, sizeof(tmppath)); /* sanity */
- if (error = copyinstr(cpaths[ix], tmppath,
- MAXPATHLEN, &ci->ci_pathlen)) {
+ if ((error = copyinstr(cpaths[ix], tmppath,
+ MAXPATHLEN, &ci->ci_pathlen)) != 0) {
#ifdef DEBUG
if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
printf("ccd%d: can't copy path, error = %d\n",
ccd->ccd_unit, error);
#endif
- free(cs->sc_cinfo, M_DEVBUF);
- return (error);
+ goto fail;
}
ci->ci_path = malloc(ci->ci_pathlen, M_DEVBUF, M_WAITOK);
bcopy(tmppath, ci->ci_path, ci->ci_pathlen);
- /*
- * XXX: Cache the component's dev_t.
- */
- if (error = VOP_GETATTR(vp, &va, p->p_ucred, p)) {
-#ifdef DEBUG
- if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
- printf("ccd%d: %s: getattr failed %s = %d\n",
- ccd->ccd_unit, ci->ci_path,
- "error", error);
-#endif
- free(ci->ci_path, M_DEVBUF);
- free(cs->sc_cinfo, M_DEVBUF);
- return (error);
- }
- ci->ci_dev = va.va_rdev;
+ ci->ci_dev = vn_todev(vp);
/*
* Get partition information for the component.
*/
- if (error = VOP_IOCTL(vp, DIOCGPART, (caddr_t)&dpart,
- FREAD, p->p_ucred, p)) {
+ if ((error = VOP_IOCTL(vp, DIOCGPART, (caddr_t)&dpart,
+ FREAD, p->p_ucred, p)) != 0) {
#ifdef DEBUG
if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
printf("ccd%d: %s: ioctl failed, error = %d\n",
ccd->ccd_unit, ci->ci_path, error);
#endif
- free(ci->ci_path, M_DEVBUF);
- free(cs->sc_cinfo, M_DEVBUF);
- return (error);
+ goto fail;
}
if (dpart.part->p_fstype == FS_BSDFFS) {
maxsecsize =
((dpart.disklab->d_secsize > maxsecsize) ?
dpart.disklab->d_secsize : maxsecsize);
- size = dpart.part->p_size;
+ size = dpart.part->p_size - CCD_OFFSET;
} else {
#ifdef DEBUG
if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
printf("ccd%d: %s: incorrect partition type\n",
ccd->ccd_unit, ci->ci_path);
#endif
- free(ci->ci_path, M_DEVBUF);
- free(cs->sc_cinfo, M_DEVBUF);
- return (EFTYPE);
+ error = EFTYPE;
+ goto fail;
}
/*
* Calculate the size, truncating to an interleave
* boundary if necessary.
*/
- if (size < 0)
- size = 0;
if (cs->sc_ileave > 1)
size -= size % cs->sc_ileave;
@@ -333,9 +457,8 @@ ccdinit(ccd, cpaths, p)
printf("ccd%d: %s: size == 0\n",
ccd->ccd_unit, ci->ci_path);
#endif
- free(ci->ci_path, M_DEVBUF);
- free(cs->sc_cinfo, M_DEVBUF);
- return (ENODEV);
+ error = ENODEV;
+ goto fail;
}
if (minsize == 0 || size < minsize)
@@ -355,20 +478,52 @@ ccdinit(ccd, cpaths, p)
printf("ccd%d: interleave must be at least %d\n",
ccd->ccd_unit, (maxsecsize / DEV_BSIZE));
#endif
- free(ci->ci_path, M_DEVBUF);
- free(cs->sc_cinfo, M_DEVBUF);
- return (EINVAL);
+ error = EINVAL;
+ goto fail;
}
/*
* If uniform interleave is desired set all sizes to that of
- * the smallest component.
+ * the smallest component. This will guarentee that a single
+ * interleave table is generated.
+ *
+ * Lost space must be taken into account when calculating the
+ * overall size. Half the space is lost when CCDF_MIRROR is
+ * specified. One disk is lost when CCDF_PARITY is specified.
*/
if (ccd->ccd_flags & CCDF_UNIFORM) {
for (ci = cs->sc_cinfo;
- ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
+ ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) {
ci->ci_size = minsize;
- cs->sc_size = cs->sc_nccdisks * minsize;
+ }
+ if (ccd->ccd_flags & CCDF_MIRROR) {
+ /*
+ * Check to see if an even number of components
+ * have been specified. The interleave must also
+ * be non-zero in order for us to be able to
+ * guarentee the topology.
+ */
+ if (cs->sc_nccdisks % 2) {
+ printf("ccd%d: mirroring requires an even number of disks\n", ccd->ccd_unit );
+ error = EINVAL;
+ goto fail;
+ }
+ if (cs->sc_ileave == 0) {
+ printf("ccd%d: an interleave must be specified when mirroring\n", ccd->ccd_unit);
+ error = EINVAL;
+ goto fail;
+ }
+ cs->sc_size = (cs->sc_nccdisks/2) * minsize;
+ } else if (ccd->ccd_flags & CCDF_PARITY) {
+ cs->sc_size = (cs->sc_nccdisks-1) * minsize;
+ } else {
+ if (cs->sc_ileave == 0) {
+ printf("ccd%d: an interleave must be specified when using parity\n", ccd->ccd_unit);
+ error = EINVAL;
+ goto fail;
+ }
+ cs->sc_size = cs->sc_nccdisks * minsize;
+ }
}
/*
@@ -380,39 +535,53 @@ ccdinit(ccd, cpaths, p)
* Create pseudo-geometry based on 1MB cylinders. It's
* pretty close.
*/
- ccg->ccg_secsize = DEV_BSIZE;
+ ccg->ccg_secsize = maxsecsize;
ccg->ccg_ntracks = 1;
- ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
+ ccg->ccg_nsectors = 1024 * 1024 / ccg->ccg_secsize;
ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
-#ifdef WORKING_DISK_STATISTICS /* XXX !! */
- if (ccd->ccd_dk >= 0)
- dk_wpms[ccd->ccd_dk] = 32 * (60 * DEV_BSIZE / 2); /* XXX */
-#endif
+ /*
+ * Add an devstat entry for this device.
+ */
+ devstat_add_entry(&cs->device_stats, "ccd", ccd->ccd_unit,
+ ccg->ccg_secsize, DEVSTAT_ALL_SUPPORTED,
+ DEVSTAT_TYPE_STORARRAY |DEVSTAT_TYPE_IF_OTHER,
+ DEVSTAT_PRIORITY_ARRAY);
cs->sc_flags |= CCDF_INITED;
cs->sc_cflags = ccd->ccd_flags; /* So we can find out later... */
cs->sc_unit = ccd->ccd_unit;
return (0);
+fail:
+ while (ci > cs->sc_cinfo) {
+ ci--;
+ free(ci->ci_path, M_DEVBUF);
+ }
+ free(cs->sc_cinfo, M_DEVBUF);
+ return (error);
}
static void
ccdinterleave(cs, unit)
- register struct ccd_softc *cs;
+ struct ccd_softc *cs;
int unit;
{
- register struct ccdcinfo *ci, *smallci;
- register struct ccdiinfo *ii;
- register daddr_t bn, lbn;
- register int ix;
+ struct ccdcinfo *ci, *smallci;
+ struct ccdiinfo *ii;
+ daddr_t bn, lbn;
+ int ix;
u_long size;
#ifdef DEBUG
if (ccddebug & CCDB_INIT)
printf("ccdinterleave(%x): ileave %d\n", cs, cs->sc_ileave);
#endif
+
/*
- * Allocate an interleave table.
+ * Allocate an interleave table. The worst case occurs when each
+ * of N disks is of a different size, resulting in N interleave
+ * tables.
+ *
* Chances are this is too big, but we don't care.
*/
size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
@@ -422,6 +591,8 @@ ccdinterleave(cs, unit)
/*
* Trivial case: no interleave (actually interleave of disk size).
* Each table entry represents a single component in its entirety.
+ *
+ * An interleave of 0 may not be used with a mirror or parity setup.
*/
if (cs->sc_ileave == 0) {
bn = 0;
@@ -451,7 +622,10 @@ ccdinterleave(cs, unit)
size = 0;
bn = lbn = 0;
for (ii = cs->sc_itable; ; ii++) {
- /* Allocate space for ii_index. */
+ /*
+ * Allocate space for ii_index. We might allocate more then
+ * we use.
+ */
ii->ii_index = malloc((sizeof(int) * cs->sc_nccdisks),
M_DEVBUF, M_WAITOK);
@@ -459,12 +633,14 @@ ccdinterleave(cs, unit)
* Locate the smallest of the remaining components
*/
smallci = NULL;
- for (ci = cs->sc_cinfo;
- ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
+ for (ci = cs->sc_cinfo; ci < &cs->sc_cinfo[cs->sc_nccdisks];
+ ci++) {
if (ci->ci_size > size &&
(smallci == NULL ||
- ci->ci_size < smallci->ci_size))
+ ci->ci_size < smallci->ci_size)) {
smallci = ci;
+ }
+ }
/*
* Nobody left, all done
@@ -475,9 +651,15 @@ ccdinterleave(cs, unit)
}
/*
- * Record starting logical block and component offset
+ * Record starting logical block using an sc_ileave blocksize.
*/
ii->ii_startblk = bn / cs->sc_ileave;
+
+ /*
+ * Record starting comopnent block using an sc_ileave
+ * blocksize. This value is relative to the beginning of
+ * a component disk.
+ */
ii->ii_startoff = lbn;
/*
@@ -485,10 +667,12 @@ ccdinterleave(cs, unit)
* and record their indices.
*/
ix = 0;
- for (ci = cs->sc_cinfo;
- ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
- if (ci->ci_size >= smallci->ci_size)
+ for (ci = cs->sc_cinfo;
+ ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++) {
+ if (ci->ci_size >= smallci->ci_size) {
ii->ii_index[ix++] = ci - cs->sc_cinfo;
+ }
+ }
ii->ii_ndisk = ix;
bn += ix * (smallci->ci_size - size);
lbn = smallci->ci_size / cs->sc_ileave;
@@ -501,7 +685,7 @@ ccdinterleave(cs, unit)
}
/* ARGSUSED */
-int
+static int
ccdopen(dev, flags, fmt, p)
dev_t dev;
int flags, fmt;
@@ -520,12 +704,12 @@ ccdopen(dev, flags, fmt, p)
return (ENXIO);
cs = &ccd_softc[unit];
- if (error = ccdlock(cs))
+ if ((error = ccdlock(cs)) != 0)
return (error);
- lp = &cs->sc_dkdev.dk_label;
+ lp = &cs->sc_label;
- part = DISKPART(dev);
+ part = ccdpart(dev);
pmask = (1 << part);
/*
@@ -533,11 +717,11 @@ ccdopen(dev, flags, fmt, p)
* open partitions. If not, then it's safe to update
* the in-core disklabel.
*/
- if ((cs->sc_flags & CCDF_INITED) && (cs->sc_dkdev.dk_openmask == 0))
+ if ((cs->sc_flags & CCDF_INITED) && (cs->sc_openmask == 0))
ccdgetdisklabel(dev);
/* Check that the partition exists. */
- if (part != RAW_PART && ((part > lp->d_npartitions) ||
+ if (part != RAW_PART && ((part >= lp->d_npartitions) ||
(lp->d_partitions[part].p_fstype == FS_UNUSED))) {
error = ENXIO;
goto done;
@@ -546,15 +730,15 @@ ccdopen(dev, flags, fmt, p)
/* Prevent our unit from being unconfigured while open. */
switch (fmt) {
case S_IFCHR:
- cs->sc_dkdev.dk_copenmask |= pmask;
+ cs->sc_copenmask |= pmask;
break;
case S_IFBLK:
- cs->sc_dkdev.dk_bopenmask |= pmask;
+ cs->sc_bopenmask |= pmask;
break;
}
- cs->sc_dkdev.dk_openmask =
- cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
+ cs->sc_openmask =
+ cs->sc_copenmask | cs->sc_bopenmask;
done:
ccdunlock(cs);
@@ -562,7 +746,7 @@ ccdopen(dev, flags, fmt, p)
}
/* ARGSUSED */
-int
+static int
ccdclose(dev, flags, fmt, p)
dev_t dev;
int flags, fmt;
@@ -581,36 +765,35 @@ ccdclose(dev, flags, fmt, p)
return (ENXIO);
cs = &ccd_softc[unit];
- if (error = ccdlock(cs))
+ if ((error = ccdlock(cs)) != 0)
return (error);
- part = DISKPART(dev);
+ part = ccdpart(dev);
/* ...that much closer to allowing unconfiguration... */
switch (fmt) {
case S_IFCHR:
- cs->sc_dkdev.dk_copenmask &= ~(1 << part);
+ cs->sc_copenmask &= ~(1 << part);
break;
case S_IFBLK:
- cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
+ cs->sc_bopenmask &= ~(1 << part);
break;
}
- cs->sc_dkdev.dk_openmask =
- cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
+ cs->sc_openmask =
+ cs->sc_copenmask | cs->sc_bopenmask;
ccdunlock(cs);
return (0);
}
-void
+static void
ccdstrategy(bp)
- register struct buf *bp;
+ struct buf *bp;
{
- register int unit = ccdunit(bp->b_dev);
- register struct ccd_softc *cs = &ccd_softc[unit];
- register daddr_t bn;
- register int sz, s;
+ int unit = ccdunit(bp->b_dev);
+ struct ccd_softc *cs = &ccd_softc[unit];
+ int s;
int wlabel;
struct disklabel *lp;
@@ -628,16 +811,45 @@ ccdstrategy(bp)
if (bp->b_bcount == 0)
goto done;
- lp = &cs->sc_dkdev.dk_label;
+ lp = &cs->sc_label;
/*
* Do bounds checking and adjust transfer. If there's an
* error, the bounds check will flag that for us.
*/
wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
- if (DISKPART(bp->b_dev) != RAW_PART)
+ if (ccdpart(bp->b_dev) != RAW_PART) {
if (bounds_check_with_label(bp, lp, wlabel) <= 0)
goto done;
+ } else {
+ int pbn; /* in sc_secsize chunks */
+ long sz; /* in sc_secsize chunks */
+
+ pbn = bp->b_blkno / (cs->sc_geom.ccg_secsize / DEV_BSIZE);
+ sz = howmany(bp->b_bcount, cs->sc_geom.ccg_secsize);
+
+ /*
+ * If out of bounds return an error. If at the EOF point,
+ * simply read or write less.
+ */
+
+ if (pbn < 0 || pbn >= cs->sc_size) {
+ bp->b_resid = bp->b_bcount;
+ if (pbn != cs->sc_size) {
+ bp->b_error = EINVAL;
+ bp->b_flags |= B_ERROR | B_INVAL;
+ }
+ goto done;
+ }
+
+ /*
+ * If the request crosses EOF, truncate the request.
+ */
+ if (pbn + sz > cs->sc_size) {
+ bp->b_bcount = (cs->sc_size - pbn) *
+ cs->sc_geom.ccg_secsize;
+ }
+ }
bp->b_resid = bp->b_bcount;
@@ -654,11 +866,12 @@ done:
static void
ccdstart(cs, bp)
- register struct ccd_softc *cs;
- register struct buf *bp;
+ struct ccd_softc *cs;
+ struct buf *bp;
{
- register long bcount, rcount;
- struct ccdbuf *cbp;
+ long bcount, rcount;
+ struct ccdbuf *cbp[4];
+ /* XXX! : 2 reads and 2 writes for RAID 4/5 */
caddr_t addr;
daddr_t bn;
struct partition *pp;
@@ -668,24 +881,15 @@ ccdstart(cs, bp)
printf("ccdstart(%x, %x)\n", cs, bp);
#endif
-#ifdef WORKING_DISK_STATISTICS /* XXX !! */
- /*
- * Instrumentation (not very meaningful)
- */
- cs->sc_nactive++;
- if (cs->sc_dk >= 0) {
- dk_busy |= 1 << cs->sc_dk;
- dk_xfer[cs->sc_dk]++;
- dk_wds[cs->sc_dk] += bp->b_bcount >> 6;
- }
-#endif
+ /* Record the transaction start */
+ devstat_start_transaction(&cs->device_stats);
/*
* Translate the partition-relative block number to an absolute.
*/
bn = bp->b_blkno;
- if (DISKPART(bp->b_dev) != RAW_PART) {
- pp = &cs->sc_dkdev.dk_label.d_partitions[DISKPART(bp->b_dev)];
+ if (ccdpart(bp->b_dev) != RAW_PART) {
+ pp = &cs->sc_label.d_partitions[ccdpart(bp->b_dev)];
bn += pp->p_offset;
}
@@ -694,11 +898,47 @@ ccdstart(cs, bp)
*/
addr = bp->b_data;
for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
- cbp = ccdbuffer(cs, bp, bn, addr, bcount);
- rcount = cbp->cb_buf.b_bcount;
- if ((cbp->cb_buf.b_flags & B_READ) == 0)
- cbp->cb_buf.b_vp->v_numoutput++;
- VOP_STRATEGY(&cbp->cb_buf);
+ ccdbuffer(cbp, cs, bp, bn, addr, bcount);
+ rcount = cbp[0]->cb_buf.b_bcount;
+
+ if (cs->sc_cflags & CCDF_MIRROR) {
+ /*
+ * Mirroring. Writes go to both disks, reads are
+ * taken from whichever disk seems most appropriate.
+ *
+ * We attempt to localize reads to the disk whos arm
+ * is nearest the read request. We ignore seeks due
+ * to writes when making this determination and we
+ * also try to avoid hogging.
+ */
+ if ((cbp[0]->cb_buf.b_flags & B_READ) == 0) {
+ cbp[0]->cb_buf.b_vp->v_numoutput++;
+ cbp[1]->cb_buf.b_vp->v_numoutput++;
+ VOP_STRATEGY(cbp[0]->cb_buf.b_vp,
+ &cbp[0]->cb_buf);
+ VOP_STRATEGY(cbp[1]->cb_buf.b_vp,
+ &cbp[1]->cb_buf);
+ } else {
+ int pick = cs->sc_pick;
+ daddr_t range = cs->sc_size / 16;
+
+ if (bn < cs->sc_blk[pick] - range ||
+ bn > cs->sc_blk[pick] + range
+ ) {
+ cs->sc_pick = pick = 1 - pick;
+ }
+ cs->sc_blk[pick] = bn + btodb(rcount);
+ VOP_STRATEGY(cbp[pick]->cb_buf.b_vp,
+ &cbp[pick]->cb_buf);
+ }
+ } else {
+ /*
+ * Not mirroring
+ */
+ if ((cbp[0]->cb_buf.b_flags & B_READ) == 0)
+ cbp[0]->cb_buf.b_vp->v_numoutput++;
+ VOP_STRATEGY(cbp[0]->cb_buf.b_vp, &cbp[0]->cb_buf);
+ }
bn += btodb(rcount);
addr += rcount;
}
@@ -707,17 +947,19 @@ ccdstart(cs, bp)
/*
* Build a component buffer header.
*/
-static struct ccdbuf *
-ccdbuffer(cs, bp, bn, addr, bcount)
- register struct ccd_softc *cs;
+static void
+ccdbuffer(cb, cs, bp, bn, addr, bcount)
+ struct ccdbuf **cb;
+ struct ccd_softc *cs;
struct buf *bp;
daddr_t bn;
caddr_t addr;
long bcount;
{
- register struct ccdcinfo *ci;
- register struct ccdbuf *cbp;
- register daddr_t cbn, cboff;
+ struct ccdcinfo *ci, *ci2 = NULL; /* XXX */
+ struct ccdbuf *cbp;
+ daddr_t cbn, cboff;
+ off_t cbc;
#ifdef DEBUG
if (ccddebug & CCDB_IO)
@@ -730,59 +972,121 @@ ccdbuffer(cs, bp, bn, addr, bcount)
cbn = bn;
cboff = 0;
- /*
- * Serially concatenated
- */
if (cs->sc_ileave == 0) {
- register daddr_t sblk;
+ /*
+ * Serially concatenated and neither a mirror nor a parity
+ * config. This is a special case.
+ */
+ daddr_t sblk;
sblk = 0;
for (ci = cs->sc_cinfo; cbn >= sblk + ci->ci_size; ci++)
sblk += ci->ci_size;
cbn -= sblk;
- }
- /*
- * Interleaved
- */
- else {
- register struct ccdiinfo *ii;
+ } else {
+ struct ccdiinfo *ii;
int ccdisk, off;
- cboff = cbn % cs->sc_ileave;
- cbn /= cs->sc_ileave;
- for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
+ /*
+ * Calculate cbn, the logical superblock (sc_ileave chunks),
+ * and cboff, a normal block offset (DEV_BSIZE chunks) relative
+ * to cbn.
+ */
+ cboff = cbn % cs->sc_ileave; /* DEV_BSIZE gran */
+ cbn = cbn / cs->sc_ileave; /* DEV_BSIZE * ileave gran */
+
+ /*
+ * Figure out which interleave table to use.
+ */
+ for (ii = cs->sc_itable; ii->ii_ndisk; ii++) {
if (ii->ii_startblk > cbn)
break;
+ }
ii--;
+
+ /*
+ * off is the logical superblock relative to the beginning
+ * of this interleave block.
+ */
off = cbn - ii->ii_startblk;
+
+ /*
+ * We must calculate which disk component to use (ccdisk),
+ * and recalculate cbn to be the superblock relative to
+ * the beginning of the component. This is typically done by
+ * adding 'off' and ii->ii_startoff together. However, 'off'
+ * must typically be divided by the number of components in
+ * this interleave array to be properly convert it from a
+ * CCD-relative logical superblock number to a
+ * component-relative superblock number.
+ */
if (ii->ii_ndisk == 1) {
+ /*
+ * When we have just one disk, it can't be a mirror
+ * or a parity config.
+ */
ccdisk = ii->ii_index[0];
cbn = ii->ii_startoff + off;
} else {
- ccdisk = ii->ii_index[off % ii->ii_ndisk];
- cbn = ii->ii_startoff + off / ii->ii_ndisk;
+ if (cs->sc_cflags & CCDF_MIRROR) {
+ /*
+ * We have forced a uniform mapping, resulting
+ * in a single interleave array. We double
+ * up on the first half of the available
+ * components and our mirror is in the second
+ * half. This only works with a single
+ * interleave array because doubling up
+ * doubles the number of sectors, so there
+ * cannot be another interleave array because
+ * the next interleave array's calculations
+ * would be off.
+ */
+ int ndisk2 = ii->ii_ndisk / 2;
+ ccdisk = ii->ii_index[off % ndisk2];
+ cbn = ii->ii_startoff + off / ndisk2;
+ ci2 = &cs->sc_cinfo[ccdisk + ndisk2];
+ } else if (cs->sc_cflags & CCDF_PARITY) {
+ /*
+ * XXX not implemented yet
+ */
+ int ndisk2 = ii->ii_ndisk - 1;
+ ccdisk = ii->ii_index[off % ndisk2];
+ cbn = ii->ii_startoff + off / ndisk2;
+ if (cbn % ii->ii_ndisk <= ccdisk)
+ ccdisk++;
+ } else {
+ ccdisk = ii->ii_index[off % ii->ii_ndisk];
+ cbn = ii->ii_startoff + off / ii->ii_ndisk;
+ }
}
- cbn *= cs->sc_ileave;
+
ci = &cs->sc_cinfo[ccdisk];
+
+ /*
+ * Convert cbn from a superblock to a normal block so it
+ * can be used to calculate (along with cboff) the normal
+ * block index into this particular disk.
+ */
+ cbn *= cs->sc_ileave;
}
/*
* Fill in the component buf structure.
*/
- cbp = getccdbuf();
+ cbp = getccdbuf(NULL);
cbp->cb_buf.b_flags = bp->b_flags | B_CALL;
- cbp->cb_buf.b_iodone = (void (*)())ccdiodone;
- cbp->cb_buf.b_proc = bp->b_proc;
+ cbp->cb_buf.b_iodone = (void (*)(struct buf *))ccdiodone;
cbp->cb_buf.b_dev = ci->ci_dev; /* XXX */
- cbp->cb_buf.b_blkno = cbn + cboff;
+ cbp->cb_buf.b_blkno = cbn + cboff + CCD_OFFSET;
+ cbp->cb_buf.b_offset = dbtob(cbn + cboff + CCD_OFFSET);
cbp->cb_buf.b_data = addr;
cbp->cb_buf.b_vp = ci->ci_vp;
if (cs->sc_ileave == 0)
- cbp->cb_buf.b_bcount = dbtob(ci->ci_size - cbn);
+ cbc = dbtob((off_t)(ci->ci_size - cbn));
else
- cbp->cb_buf.b_bcount = dbtob(cs->sc_ileave - cboff);
- if (cbp->cb_buf.b_bcount > bcount)
- cbp->cb_buf.b_bcount = bcount;
+ cbc = dbtob((off_t)(cs->sc_ileave - cboff));
+ cbp->cb_buf.b_bcount = (cbc < bcount) ? cbc : bcount;
+ cbp->cb_buf.b_bufsize = cbp->cb_buf.b_bcount;
/*
* context for ccdiodone
@@ -797,15 +1101,32 @@ ccdbuffer(cs, bp, bn, addr, bcount)
ci->ci_dev, ci-cs->sc_cinfo, cbp, cbp->cb_buf.b_blkno,
cbp->cb_buf.b_data, cbp->cb_buf.b_bcount);
#endif
- return (cbp);
+ cb[0] = cbp;
+
+ /*
+ * Note: both I/O's setup when reading from mirror, but only one
+ * will be executed.
+ */
+ if (cs->sc_cflags & CCDF_MIRROR) {
+ /* mirror, setup second I/O */
+ cbp = getccdbuf(cb[0]);
+ cbp->cb_buf.b_dev = ci2->ci_dev;
+ cbp->cb_buf.b_vp = ci2->ci_vp;
+ cbp->cb_comp = ci2 - cs->sc_cinfo;
+ cb[1] = cbp;
+ /* link together the ccdbuf's and clear "mirror done" flag */
+ cb[0]->cb_mirror = cb[1];
+ cb[1]->cb_mirror = cb[0];
+ cb[0]->cb_pflags &= ~CCDPF_MIRROR_DONE;
+ cb[1]->cb_pflags &= ~CCDPF_MIRROR_DONE;
+ }
}
static void
ccdintr(cs, bp)
- register struct ccd_softc *cs;
- register struct buf *bp;
+ struct ccd_softc *cs;
+ struct buf *bp;
{
-
#ifdef DEBUG
if (ccddebug & CCDB_FOLLOW)
printf("ccdintr(%x, %x)\n", cs, bp);
@@ -813,18 +1134,9 @@ ccdintr(cs, bp)
/*
* Request is done for better or worse, wakeup the top half.
*/
-#ifdef WORKING_DISK_STATISTICS /* XXX !! */
- --cs->sc_nactive;
-#ifdef DIAGNOSTIC
- if (cs->sc_nactive < 0)
- panic("ccdintr: ccd%d: sc_nactive < 0", cs->sc_unit);
-#endif
-
- if (cs->sc_nactive == 0 && cs->sc_dk >= 0)
- dk_busy &= ~(1 << cs->sc_dk);
-#endif
if (bp->b_flags & B_ERROR)
bp->b_resid = bp->b_bcount;
+ devstat_end_transaction_buf(&cs->device_stats, bp);
biodone(bp);
}
@@ -833,12 +1145,12 @@ ccdintr(cs, bp)
* Mark the component as done and if all components are done,
* take a ccd interrupt.
*/
-void
+static void
ccdiodone(cbp)
struct ccdbuf *cbp;
{
- register struct buf *bp = cbp->cb_obp;
- register int unit = cbp->cb_unit;
+ struct buf *bp = cbp->cb_obp;
+ int unit = cbp->cb_unit;
int count, s;
s = splbio();
@@ -854,16 +1166,97 @@ ccdiodone(cbp)
cbp->cb_buf.b_bcount);
}
#endif
+ /*
+ * If an error occured, report it. If this is a mirrored
+ * configuration and the first of two possible reads, do not
+ * set the error in the bp yet because the second read may
+ * succeed.
+ */
if (cbp->cb_buf.b_flags & B_ERROR) {
- bp->b_flags |= B_ERROR;
- bp->b_error = cbp->cb_buf.b_error ? cbp->cb_buf.b_error : EIO;
-#ifdef DEBUG
- printf("ccd%d: error %d on component %d\n",
- unit, bp->b_error, cbp->cb_comp);
-#endif
+ const char *msg = "";
+
+ if ((ccd_softc[unit].sc_cflags & CCDF_MIRROR) &&
+ (cbp->cb_buf.b_flags & B_READ) &&
+ (cbp->cb_pflags & CCDPF_MIRROR_DONE) == 0) {
+ /*
+ * We will try our read on the other disk down
+ * below, also reverse the default pick so if we
+ * are doing a scan we do not keep hitting the
+ * bad disk first.
+ */
+ struct ccd_softc *cs = &ccd_softc[unit];
+
+ msg = ", trying other disk";
+ cs->sc_pick = 1 - cs->sc_pick;
+ cs->sc_blk[cs->sc_pick] = bp->b_blkno;
+ } else {
+ bp->b_flags |= B_ERROR;
+ bp->b_error = cbp->cb_buf.b_error ?
+ cbp->cb_buf.b_error : EIO;
+ }
+ printf("ccd%d: error %d on component %d block %d (ccd block %d)%s\n",
+ unit, bp->b_error, cbp->cb_comp,
+ (int)cbp->cb_buf.b_blkno, bp->b_blkno, msg);
+ }
+
+ /*
+ * Process mirror. If we are writing, I/O has been initiated on both
+ * buffers and we fall through only after both are finished.
+ *
+ * If we are reading only one I/O is initiated at a time. If an
+ * error occurs we initiate the second I/O and return, otherwise
+ * we free the second I/O without initiating it.
+ */
+
+ if (ccd_softc[unit].sc_cflags & CCDF_MIRROR) {
+ if ((cbp->cb_buf.b_flags & B_READ) == 0) {
+ /*
+ * When writing, handshake with the second buffer
+ * to determine when both are done. If both are not
+ * done, return here.
+ */
+ if ((cbp->cb_pflags & CCDPF_MIRROR_DONE) == 0) {
+ cbp->cb_mirror->cb_pflags |= CCDPF_MIRROR_DONE;
+ putccdbuf(cbp);
+ splx(s);
+ return;
+ }
+ } else {
+ /*
+ * When reading, either dispose of the second buffer
+ * or initiate I/O on the second buffer if an error
+ * occured with this one.
+ */
+ if ((cbp->cb_pflags & CCDPF_MIRROR_DONE) == 0) {
+ if (cbp->cb_buf.b_flags & B_ERROR) {
+ cbp->cb_mirror->cb_pflags |=
+ CCDPF_MIRROR_DONE;
+ VOP_STRATEGY(
+ cbp->cb_mirror->cb_buf.b_vp,
+ &cbp->cb_mirror->cb_buf
+ );
+ putccdbuf(cbp);
+ splx(s);
+ return;
+ } else {
+ putccdbuf(cbp->cb_mirror);
+ /* fall through */
+ }
+ }
+ }
}
- count = cbp->cb_buf.b_bcount;
+
+ /*
+ * use b_bufsize to determine how big the original request was rather
+ * then b_bcount, because b_bcount may have been truncated for EOF.
+ *
+ * XXX We check for an error, but we do not test the resid for an
+ * aligned EOF condition. This may result in character & block
+ * device access not recognizing EOF properly when read or written
+ * sequentially, but will not effect filesystems.
+ */
+ count = cbp->cb_buf.b_bufsize;
putccdbuf(cbp);
/*
@@ -877,65 +1270,7 @@ ccdiodone(cbp)
splx(s);
}
-/* ARGSUSED */
-int
-ccdread(dev, uio, flags)
- dev_t dev;
- struct uio *uio;
- int flags;
-{
- int unit = ccdunit(dev);
- struct ccd_softc *cs;
-
-#ifdef DEBUG
- if (ccddebug & CCDB_FOLLOW)
- printf("ccdread(%x, %x)\n", dev, uio);
-#endif
- if (unit >= numccd)
- return (ENXIO);
- cs = &ccd_softc[unit];
-
- if ((cs->sc_flags & CCDF_INITED) == 0)
- return (ENXIO);
-
- /*
- * XXX: It's not clear that using minphys() is completely safe,
- * in particular, for raw I/O. Underlying devices might have some
- * non-obvious limits, because of the copy to user-space.
- */
- return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
-}
-
-/* ARGSUSED */
-int
-ccdwrite(dev, uio, flags)
- dev_t dev;
- struct uio *uio;
- int flags;
-{
- int unit = ccdunit(dev);
- struct ccd_softc *cs;
-
-#ifdef DEBUG
- if (ccddebug & CCDB_FOLLOW)
- printf("ccdwrite(%x, %x)\n", dev, uio);
-#endif
- if (unit >= numccd)
- return (ENXIO);
- cs = &ccd_softc[unit];
-
- if ((cs->sc_flags & CCDF_INITED) == 0)
- return (ENXIO);
-
- /*
- * XXX: It's not clear that using minphys() is completely safe,
- * in particular, for raw I/O. Underlying devices might have some
- * non-obvious limits, because of the copy to user-space.
- */
- return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
-}
-
-int
+static int
ccdioctl(dev, cmd, data, flag, p)
dev_t dev;
u_long cmd;
@@ -951,9 +1286,6 @@ ccdioctl(dev, cmd, data, flag, p)
struct ccddevice ccd;
char **cpp;
struct vnode **vpp;
-#ifdef WORKING_DISK_STATISTICS /* XXX !! */
- extern int dkn;
-#endif
if (unit >= numccd)
return (ENXIO);
@@ -969,12 +1301,29 @@ ccdioctl(dev, cmd, data, flag, p)
if ((flag & FWRITE) == 0)
return (EBADF);
- if (error = ccdlock(cs))
+ if ((error = ccdlock(cs)) != 0)
return (error);
/* Fill in some important bits. */
ccd.ccd_unit = unit;
ccd.ccd_interleave = ccio->ccio_ileave;
+ if (ccd.ccd_interleave == 0 &&
+ ((ccio->ccio_flags & CCDF_MIRROR) ||
+ (ccio->ccio_flags & CCDF_PARITY))) {
+ printf("ccd%d: disabling mirror/parity, interleave is 0\n", unit);
+ ccio->ccio_flags &= ~(CCDF_MIRROR | CCDF_PARITY);
+ }
+ if ((ccio->ccio_flags & CCDF_MIRROR) &&
+ (ccio->ccio_flags & CCDF_PARITY)) {
+ printf("ccd%d: can't specify both mirror and parity, using mirror\n", unit);
+ ccio->ccio_flags &= ~CCDF_PARITY;
+ }
+ if ((ccio->ccio_flags & (CCDF_MIRROR | CCDF_PARITY)) &&
+ !(ccio->ccio_flags & CCDF_UNIFORM)) {
+ printf("ccd%d: mirror/parity forces uniform flag\n",
+ unit);
+ ccio->ccio_flags |= CCDF_UNIFORM;
+ }
ccd.ccd_flags = ccio->ccio_flags & CCDF_USERMASK;
/*
@@ -1007,7 +1356,7 @@ ccdioctl(dev, cmd, data, flag, p)
if (ccddebug & CCDB_INIT)
printf("ccdioctl: lookedup = %d\n", lookedup);
#endif
- if (error = ccdlookup(cpp[i], p, &vpp[i])) {
+ if ((error = ccdlookup(cpp[i], p, &vpp[i])) != 0) {
for (j = 0; j < lookedup; ++j)
(void)vn_close(vpp[j], FREAD|FWRITE,
p->p_ucred, p);
@@ -1022,27 +1371,10 @@ ccdioctl(dev, cmd, data, flag, p)
ccd.ccd_vpp = vpp;
ccd.ccd_ndev = ccio->ccio_ndisks;
-#ifdef WORKING_DISK_STATISTICS /* XXX !! */
- /*
- * Assign disk index first so that init routine
- * can use it (saves having the driver drag around
- * the ccddevice pointer just to set up the dk_*
- * info in the open routine).
- */
- if (dkn < DK_NDRIVE)
- ccd.ccd_dk = dkn++;
- else
- ccd.ccd_dk = -1;
-#endif
-
/*
* Initialize the ccd. Fills in the softc for us.
*/
- if (error = ccdinit(&ccd, cpp, p)) {
-#ifdef WORKING_DISK_STATISTICS /* XXX !! */
- if (ccd.ccd_dk >= 0)
- --dkn;
-#endif
+ if ((error = ccdinit(&ccd, cpp, p)) != 0) {
for (j = 0; j < lookedup; ++j)
(void)vn_close(vpp[j], FREAD|FWRITE,
p->p_ucred, p);
@@ -1073,7 +1405,7 @@ ccdioctl(dev, cmd, data, flag, p)
if ((flag & FWRITE) == 0)
return (EBADF);
- if (error = ccdlock(cs))
+ if ((error = ccdlock(cs)) != 0)
return (error);
/*
@@ -1081,11 +1413,11 @@ ccdioctl(dev, cmd, data, flag, p)
* or if both the character and block flavors of this
* partition are open.
*/
- part = DISKPART(dev);
+ part = ccdpart(dev);
pmask = (1 << part);
- if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
- ((cs->sc_dkdev.dk_bopenmask & pmask) &&
- (cs->sc_dkdev.dk_copenmask & pmask))) {
+ if ((cs->sc_openmask & ~pmask) ||
+ ((cs->sc_bopenmask & pmask) &&
+ (cs->sc_copenmask & pmask))) {
ccdunlock(cs);
return (EBUSY);
}
@@ -1128,6 +1460,11 @@ ccdioctl(dev, cmd, data, flag, p)
ccd.ccd_dk = -1;
bcopy(&ccd, &ccddevs[unit], sizeof(ccd));
+ /*
+ * And remove the devstat entry.
+ */
+ devstat_remove_entry(&cs->device_stats);
+
/* This must be atomic. */
s = splhigh();
ccdunlock(cs);
@@ -1140,16 +1477,16 @@ ccdioctl(dev, cmd, data, flag, p)
if ((cs->sc_flags & CCDF_INITED) == 0)
return (ENXIO);
- *(struct disklabel *)data = cs->sc_dkdev.dk_label;
+ *(struct disklabel *)data = cs->sc_label;
break;
case DIOCGPART:
if ((cs->sc_flags & CCDF_INITED) == 0)
return (ENXIO);
- ((struct partinfo *)data)->disklab = &cs->sc_dkdev.dk_label;
+ ((struct partinfo *)data)->disklab = &cs->sc_label;
((struct partinfo *)data)->part =
- &cs->sc_dkdev.dk_label.d_partitions[DISKPART(dev)];
+ &cs->sc_label.d_partitions[ccdpart(dev)];
break;
case DIOCWDINFO:
@@ -1160,18 +1497,17 @@ ccdioctl(dev, cmd, data, flag, p)
if ((flag & FWRITE) == 0)
return (EBADF);
- if (error = ccdlock(cs))
+ if ((error = ccdlock(cs)) != 0)
return (error);
cs->sc_flags |= CCDF_LABELLING;
- error = setdisklabel(&cs->sc_dkdev.dk_label,
- (struct disklabel *)data, 0, &cs->sc_dkdev.dk_cpulabel);
+ error = setdisklabel(&cs->sc_label,
+ (struct disklabel *)data, 0);
if (error == 0) {
if (cmd == DIOCWDINFO)
error = writedisklabel(CCDLABELDEV(dev),
- ccdstrategy, &cs->sc_dkdev.dk_label,
- &cs->sc_dkdev.dk_cpulabel);
+ &cs->sc_label);
}
cs->sc_flags &= ~CCDF_LABELLING;
@@ -1201,7 +1537,7 @@ ccdioctl(dev, cmd, data, flag, p)
return (0);
}
-int
+static int
ccdsize(dev)
dev_t dev;
{
@@ -1212,15 +1548,15 @@ ccdsize(dev)
return (-1);
cs = &ccd_softc[ccdunit(dev)];
- part = DISKPART(dev);
+ part = ccdpart(dev);
if ((cs->sc_flags & CCDF_INITED) == 0)
return (-1);
- if (cs->sc_dkdev.dk_label.d_partitions[part].p_fstype != FS_SWAP)
+ if (cs->sc_label.d_partitions[part].p_fstype != FS_SWAP)
size = -1;
else
- size = cs->sc_dkdev.dk_label.d_partitions[part].p_size;
+ size = cs->sc_label.d_partitions[part].p_size;
if (ccdclose(dev, 0, S_IFBLK, curproc))
return (-1);
@@ -1228,12 +1564,9 @@ ccdsize(dev)
return (size);
}
-int
-ccddump(dev, blkno, va, size)
+static int
+ccddump(dev)
dev_t dev;
- daddr_t blkno;
- caddr_t va;
- size_t size;
{
/* Not implemented. */
@@ -1257,7 +1590,7 @@ ccdlookup(path, p, vpp)
int error;
NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, path, p);
- if (error = vn_open(&nd, FREAD|FWRITE, 0)) {
+ if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
#ifdef DEBUG
if (ccddebug & CCDB_FOLLOW|CCDB_INIT)
printf("ccdlookup: vn_open error = %d\n", error);
@@ -1267,26 +1600,13 @@ ccdlookup(path, p, vpp)
vp = nd.ni_vp;
if (vp->v_usecount > 1) {
- VOP_UNLOCK(vp);
- (void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
- return (EBUSY);
+ error = EBUSY;
+ goto bad;
}
- if (error = VOP_GETATTR(vp, &va, p->p_ucred, p)) {
-#ifdef DEBUG
- if (ccddebug & CCDB_FOLLOW|CCDB_INIT)
- printf("ccdlookup: getattr error = %d\n", error);
-#endif
- VOP_UNLOCK(vp);
- (void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
- return (error);
- }
-
- /* XXX: eventually we should handle VREG, too. */
- if (va.va_type != VBLK) {
- VOP_UNLOCK(vp);
- (void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
- return (ENOTBLK);
+ if (!vn_isdisk(vp)) {
+ error = ENOTBLK;
+ goto bad;
}
#ifdef DEBUG
@@ -1294,9 +1614,16 @@ ccdlookup(path, p, vpp)
vprint("ccdlookup: vnode info", vp);
#endif
- VOP_UNLOCK(vp);
+ VOP_UNLOCK(vp, 0, p);
+ NDFREE(&nd, NDF_ONLY_PNBUF);
*vpp = vp;
return (0);
+bad:
+ VOP_UNLOCK(vp, 0, p);
+ NDFREE(&nd, NDF_ONLY_PNBUF);
+ /* vn_close does vrele() for vp */
+ (void)vn_close(vp, FREAD|FWRITE, p->p_ucred, p);
+ return (error);
}
/*
@@ -1310,12 +1637,10 @@ ccdgetdisklabel(dev)
int unit = ccdunit(dev);
struct ccd_softc *cs = &ccd_softc[unit];
char *errstring;
- struct disklabel *lp = &cs->sc_dkdev.dk_label;
- struct cpu_disklabel *clp = &cs->sc_dkdev.dk_cpulabel;
+ struct disklabel *lp = &cs->sc_label;
struct ccdgeom *ccg = &cs->sc_geom;
bzero(lp, sizeof(*lp));
- bzero(clp, sizeof(*clp));
lp->d_secperunit = cs->sc_size;
lp->d_secsize = ccg->ccg_secsize;
@@ -1336,15 +1661,18 @@ ccdgetdisklabel(dev)
lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
lp->d_npartitions = RAW_PART + 1;
+ lp->d_bbsize = BBSIZE; /* XXX */
+ lp->d_sbsize = SBSIZE; /* XXX */
+
lp->d_magic = DISKMAGIC;
lp->d_magic2 = DISKMAGIC;
- lp->d_checksum = dkcksum(&cs->sc_dkdev.dk_label);
+ lp->d_checksum = dkcksum(&cs->sc_label);
/*
* Call the generic disklabel extraction routine.
*/
- if (errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
- &cs->sc_dkdev.dk_label, &cs->sc_dkdev.dk_cpulabel))
+ errstring = readdisklabel(CCDLABELDEV(dev), &cs->sc_label);
+ if (errstring != NULL)
ccdmakedisklabel(cs);
#ifdef DEBUG
@@ -1363,7 +1691,7 @@ static void
ccdmakedisklabel(cs)
struct ccd_softc *cs;
{
- struct disklabel *lp = &cs->sc_dkdev.dk_label;
+ struct disklabel *lp = &cs->sc_label;
/*
* For historical reasons, if there's no disklabel present
@@ -1415,7 +1743,7 @@ static void
printiinfo(ii)
struct ccdiinfo *ii;
{
- register int ix, i;
+ int ix, i;
for (ix = 0; ii->ii_ndisk; ix++, ii++) {
printf(" itab[%d]: #dk %d sblk %d soff %d",
@@ -1426,3 +1754,11 @@ printiinfo(ii)
}
}
#endif
+
+#endif /* NCCD > 0 */
+
+/* Local Variables: */
+/* c-argdecl-indent: 8 */
+/* c-continued-statement-offset: 8 */
+/* c-indent-level: 8 */
+/* End: */
OpenPOWER on IntegriCloud