diff options
Diffstat (limited to 'lib/libdevstat')
| -rw-r--r-- | lib/libdevstat/Makefile | 38 | ||||
| -rw-r--r-- | lib/libdevstat/devstat.3 | 795 | ||||
| -rw-r--r-- | lib/libdevstat/devstat.c | 1631 | ||||
| -rw-r--r-- | lib/libdevstat/devstat.h | 169 |
4 files changed, 2633 insertions, 0 deletions
diff --git a/lib/libdevstat/Makefile b/lib/libdevstat/Makefile new file mode 100644 index 0000000..aeb07b9 --- /dev/null +++ b/lib/libdevstat/Makefile @@ -0,0 +1,38 @@ +# $FreeBSD$ + +LIB= devstat +SHLIBDIR?= /lib +# Bump DEVSTAT_USER_API_VER in devstat.h every time this is incremented. +SHLIB_MAJOR= 7 +SRCS= devstat.c +INCS= devstat.h + +DPADD= ${LIBKVM} +LDADD= -lkvm + +MAN= devstat.3 + +MLINKS+=devstat.3 devstat_getnumdevs.3 +MLINKS+=devstat.3 devstat_getgeneration.3 +MLINKS+=devstat.3 devstat_getversion.3 +MLINKS+=devstat.3 devstat_checkversion.3 +MLINKS+=devstat.3 devstat_getdevs.3 +MLINKS+=devstat.3 devstat_selectdevs.3 +MLINKS+=devstat.3 devstat_buildmatch.3 +MLINKS+=devstat.3 devstat_compute_statistics.3 +MLINKS+=devstat.3 devstat_compute_etime.3 +MLINKS+=devstat.3 getnumdevs.3 +MLINKS+=devstat.3 getgeneration.3 +MLINKS+=devstat.3 getversion.3 +MLINKS+=devstat.3 checkversion.3 +MLINKS+=devstat.3 getdevs.3 +MLINKS+=devstat.3 selectdevs.3 +MLINKS+=devstat.3 buildmatch.3 +MLINKS+=devstat.3 compute_stats.3 +MLINKS+=devstat.3 compute_etime.3 + +CFLAGS+=-I${.CURDIR} + +WARNS?= 3 + +.include <bsd.lib.mk> diff --git a/lib/libdevstat/devstat.3 b/lib/libdevstat/devstat.3 new file mode 100644 index 0000000..fb82146 --- /dev/null +++ b/lib/libdevstat/devstat.3 @@ -0,0 +1,795 @@ +.\" +.\" Copyright (c) 1998, 1999, 2001 Kenneth D. Merry. +.\" All rights reserved. +.\" +.\" 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. The name of the author may not be used to endorse or promote products +.\" derived from this software without specific prior written permission. +.\" +.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``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 AUTHOR 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. +.\" +.\" $FreeBSD$ +.\" +.Dd March 18, 2003 +.Dt DEVSTAT 3 +.Os +.Sh NAME +.Nm devstat , +.Nm devstat_getnumdevs , +.Nm devstat_getgeneration , +.Nm devstat_getversion , +.Nm devstat_checkversion , +.Nm devstat_getdevs , +.Nm devstat_selectdevs , +.Nm devstat_buildmatch , +.Nm devstat_compute_statistics , +.Nm devstat_compute_etime +.Nd device statistics utility library +.Sh LIBRARY +.Lb libdevstat +.Sh SYNOPSIS +.In devstat.h +.Ft int +.Fn devstat_getnumdevs "kvm_t *kd" +.Ft long +.Fn devstat_getgeneration "kvm_t *kd" +.Ft int +.Fn devstat_getversion "kvm_t *kd" +.Ft int +.Fn devstat_checkversion "kvm_t *kd" +.Ft int +.Fn devstat_getdevs "kvm_t *kd" "struct statinfo *stats" +.Ft int +.Fo devstat_selectdevs +.Fa "struct device_selection **dev_select" +.Fa "int *num_selected" +.Fa "int *num_selections" +.Fa "long *select_generation" +.Fa "long current_generation" +.Fa "struct devstat *devices" +.Fa "int numdevs" +.Fa "struct devstat_match *matches" +.Fa "int num_matches" +.Fa "char **dev_selections" +.Fa "int num_dev_selections" +.Fa "devstat_select_mode select_mode" +.Fa "int maxshowdevs" +.Fa "int perf_select" +.Fc +.Ft int +.Fo devstat_buildmatch +.Fa "char *match_str" +.Fa "struct devstat_match **matches" +.Fa "int *num_matches" +.Fc +.Ft int +.Fo devstat_compute_statistics +.Fa "struct devstat *current" +.Fa "struct devstat *previous" +.Fa "long double etime" +.Fa "..." +.Fc +.Ft "long double" +.Fo devstat_compute_etime +.Fa "struct bintime *cur_time" +.Fa "struct bintime *prev_time" +.Fc +.Sh DESCRIPTION +The +.Nm +library is a library of helper functions for dealing with the kernel +.Xr devstat 9 +interface, which is accessible to users via +.Xr sysctl 3 +and +.Xr kvm 3 . +All functions that take a +.Vt "kvm_t *" +as first argument can be passed +.Dv NULL +instead of a kvm handle as this argument, +which causes the data to be read via +.Xr sysctl 3 . +Otherwise, it is read via +.Xr kvm 3 +using the supplied handle. +The +.Fn devstat_checkversion +function +should be called with each kvm handle that is going to be used (or with +.Dv NULL +if +.Xr sysctl 3 +is going to be used). +.Pp +The +.Fn devstat_getnumdevs +function +returns the number of devices registered with the +.Nm +subsystem in the kernel. +.Pp +The +.Fn devstat_getgeneration +function +returns the current generation of the +.Nm +list of devices in the kernel. +.Pp +The +.Fn devstat_getversion +function +returns the current kernel +.Nm +version. +.Pp +The +.Fn devstat_checkversion +function +checks the userland +.Nm +version against the kernel +.Nm +version. +If the two are identical, it returns zero. +Otherwise, it prints an appropriate error in +.Va devstat_errbuf +and returns \-1. +.Pp +The +.Fn devstat_getdevs +function +fetches the current list of devices and statistics into the supplied +.Vt statinfo +structure. +The +.Vt statinfo +structure can be found in +.In devstat.h : +.Bd -literal -offset indent +struct statinfo { + long cp_time[CPUSTATES]; + long tk_nin; + long tk_nout; + struct devinfo *dinfo; + long double snap_time; +}; +.Ed +.Pp +The +.Fn devstat_getdevs +function +expects the +.Vt statinfo +structure to be allocated, and it also expects the +.Va dinfo +subelement to be allocated and zeroed prior to the first invocation of +.Fn devstat_getdevs . +The +.Va dinfo +subelement is used to store state between calls, and should not be modified +after the first call to +.Fn devstat_getdevs . +The +.Va dinfo +subelement contains the following elements: +.Bd -literal -offset indent +struct devinfo { + struct devstat *devices; + u_int8_t *mem_ptr; + long generation; + int numdevs; +}; +.Ed +.Pp +The +.Va kern.devstat.all +.Xr sysctl 8 +variable contains an array of +.Nm +structures, but at the head of the array is the current +.Nm +generation. +The reason the generation is at the head of the buffer is so that userland +software accessing the +.Nm +statistics information can atomically get +both the statistics information and the corresponding generation number. +If client software were forced to get the generation number via a separate +.Xr sysctl 8 +variable (which is available for convenience), the list of devices could +change between the time the client gets the generation and the time the +client gets the device list. +.Pp +The +.Va mem_ptr +subelement of the +.Vt devinfo +structure is a pointer to memory that is allocated, and resized if +necessary, by +.Fn devstat_getdevs . +The devices subelement of the +.Vt devinfo +structure is basically a pointer to the beginning of the array of devstat +structures from the +.Va kern.devstat.all +.Xr sysctl 8 +variable (or the corresponding values read via +.Xr kvm 3 ) . +The generation subelement of the +.Vt devinfo +structure contains the corresponding generation number. +The +.Va numdevs +subelement of the +.Vt devinfo +structure contains the current +number of devices registered with the kernel +.Nm +subsystem. +.Pp +The +.Fn devstat_selectdevs +function +selects devices to display based upon a number of criteria: +.Bl -tag -width indent +.It specified devices +Specified devices are the first selection priority. +These are generally devices specified by name by the user e.g.\& +.Li da0 , da1 , cd0 . +.It match patterns +These are pattern matching expressions generated by +.Fn devstat_buildmatch +from user input. +.It performance +If performance mode is enabled, devices will be sorted based on the +.Va bytes +field in the +.Vt device_selection +structure passed in to +.Fn devstat_selectdevs . +The +.Va bytes +value currently must be maintained by the user. +In the future, this may be done for him in a +.Nm +library routine. +If no devices have been selected by name or by pattern, the performance +tracking code will select every device in the system, and sort them by +performance. +If devices have been selected by name or pattern, the performance tracking +code will honor those selections and will only sort among the selected +devices. +.It order in the devstat list +If the selection mode is set to +.Dv DS_SELECT_ADD , +and if there are still less +than +.Fa maxshowdevs +devices selected, +.Fn devstat_selectdevs +will automatically select up to +.Fa maxshowdevs +devices. +.El +.Pp +The +.Fn devstat_selectdevs +function +performs selections in four different modes: +.Bl -tag -width ".Dv DS_SELECT_ADDONLY" +.It Dv DS_SELECT_ADD +In +.Dq add +mode, +.Fn devstat_selectdevs +will select any unselected devices specified by name or matching pattern. +It will also select more devices, in devstat list order, until the number +of selected devices is equal to +.Fa maxshowdevs +or until all devices are +selected. +.It Dv DS_SELECT_ONLY +In +.Dq only +mode, +.Fn devstat_selectdevs +will clear all current selections, and will only select devices specified +by name or by matching pattern. +.It Dv DS_SELECT_REMOVE +In +.Dq remove +mode, +.Fn devstat_selectdevs +will remove devices specified by name or by matching pattern. +It will not select any additional devices. +.It Dv DS_SELECT_ADDONLY +In +.Dq "add only" +mode, +.Fn devstat_selectdevs +will select any unselected devices specified by name or matching pattern. +In this respect it is identical to +.Dq add +mode. +It will not, however, select any devices other than those specified. +.El +.Pp +In all selection modes, +.Fn devstat_selectdevs +will not select any more than +.Fa maxshowdevs +devices. +One exception to this is when you are in +.Dq top +mode and no devices have been selected. +In this case, +.Fn devstat_selectdevs +will select every device in the system. +Client programs must pay attention to selection order when deciding whether +to pay attention to a particular device. +This may be the wrong behavior, and probably requires additional thought. +.Pp +The +.Fn devstat_selectdevs +function +handles allocation and resizing of the +.Fa dev_select +structure passed in +by the client. +The +.Fn devstat_selectdevs +function +uses the +.Fa numdevs +and +.Fa current_generation +fields to track the +current +.Nm +generation and number of devices. +If +.Fa num_selections +is not the same +as +.Fa numdevs +or if +.Fa select_generation +is not the same as +.Fa current_generation , +.Fn devstat_selectdevs +will resize the selection list as necessary, and re-initialize the +selection array. +.Pp +The +.Fn devstat_buildmatch +function +takes a comma separated match string and compiles it into a +.Vt devstat_match +structure that is understood by +.Fn devstat_selectdevs . +Match strings have the following format: +.Pp +.D1 Ar device , Ns Ar type , Ns Ar if +.Pp +The +.Fn devstat_buildmatch +function +takes care of allocating and reallocating the match list as necessary. +Currently known match types include: +.Bl -tag -width indent +.It device type: +.Bl -tag -width ".Li enclosure" -compact +.It Li da +Direct Access devices +.It Li sa +Sequential Access devices +.It Li printer +Printers +.It Li proc +Processor devices +.It Li worm +Write Once Read Multiple devices +.It Li cd +CD devices +.It Li scanner +Scanner devices +.It Li optical +Optical Memory devices +.It Li changer +Medium Changer devices +.It Li comm +Communication devices +.It Li array +Storage Array devices +.It Li enclosure +Enclosure Services devices +.It Li floppy +Floppy devices +.El +.It interface: +.Bl -tag -width ".Li enclosure" -compact +.It Li IDE +Integrated Drive Electronics devices +.It Li SCSI +Small Computer System Interface devices +.It Li other +Any other device interface +.El +.It passthrough: +.Bl -tag -width ".Li enclosure" -compact +.It Li pass +Passthrough devices +.El +.El +.Pp +The +.Fn devstat_compute_statistics +function provides complete statistics calculation. +There are four arguments for which values +.Em must +be supplied: +.Fa current , +.Fa previous , +.Fa etime , +and the terminating argument for the varargs list, +.Dv DSM_NONE . +For most applications, the user will want to supply valid +.Vt devstat +structures for both +.Fa current +and +.Fa previous . +In some instances, for instance when calculating statistics since system +boot, the user may pass in a +.Dv NULL +pointer for the +.Fa previous +argument. +In that case, +.Fn devstat_compute_statistics +will use the total stats in the +.Fa current +structure to calculate statistics over +.Fa etime . +For each statistics to be calculated, the user should supply the proper +enumerated type (listed below), and a variable of the indicated type. +All statistics are either integer values, for which a +.Vt u_int64_t +is used, +or floating point, for which a +.Vt "long double" +is used. +The statistics that may be calculated are: +.Bl -tag -width ".Dv DSM_TRANSFERS_PER_SECOND_OTHER" +.It Dv DSM_NONE +type: N/A +.Pp +This +.Em must +be the last argument passed to +.Fn devstat_compute_statistics . +It is an argument list terminator. +.It Dv DSM_TOTAL_BYTES +type: +.Vt "u_int64_t *" +.Pp +The total number of bytes transferred between the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_TOTAL_BYTES_READ +.It Dv DSM_TOTAL_BYTES_WRITE +.It Dv DSM_TOTAL_BYTES_FREE +type: +.Vt "u_int64_t *" +.Pp +The total number of bytes in transactions of the specified type +between the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_TOTAL_TRANSFERS +type: +.Vt "u_int64_t *" +.Pp +The total number of transfers between the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_TOTAL_TRANSFERS_OTHER +.It Dv DSM_TOTAL_TRANSFERS_READ +.It Dv DSM_TOTAL_TRANSFERS_WRITE +.It Dv DSM_TOTAL_TRANSFERS_FREE +type: +.Vt "u_int64_t *" +.Pp +The total number of transactions of the specified type between +the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_TOTAL_BLOCKS +type: +.Vt "u_int64_t *" +.Pp +The total number of blocks transferred between the acquisition of +.Fa previous +and +.Fa current . +This number is in terms of the blocksize reported by the device. +If no blocksize has been reported (i.e., the block size is 0), a default +blocksize of 512 bytes will be used in the calculation. +.It Dv DSM_TOTAL_BLOCKS_READ +.It Dv DSM_TOTAL_BLOCKS_WRITE +.It Dv DSM_TOTAL_BLOCKS_FREE +type: +.Vt "u_int64_t *" +.Pp +The total number of blocks of the specified type between the acquisition of +.Fa previous +and +.Fa current . +This number is in terms of the blocksize reported by the device. +If no blocksize has been reported (i.e., the block size is 0), a default +blocksize of 512 bytes will be used in the calculation. +.It Dv DSM_KB_PER_TRANSFER +type: +.Vt "long double *" +.Pp +The average number of kilobytes per transfer between the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_KB_PER_TRANSFER_READ +.It Dv DSM_KB_PER_TRANSFER_WRITE +.It Dv DSM_KB_PER_TRANSFER_FREE +type: +.Vt "long double *" +.Pp +The average number of kilobytes in the specified type transaction between +the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_TRANSFERS_PER_SECOND +type: +.Vt "long double *" +.Pp +The average number of transfers per second between the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_TRANSFERS_PER_SECOND_OTHER +.It Dv DSM_TRANSFERS_PER_SECOND_READ +.It Dv DSM_TRANSFERS_PER_SECOND_WRITE +.It Dv DSM_TRANSFERS_PER_SECOND_FREE +type: +.Vt "long double *" +.Pp +The average number of transactions of the specified type per second +between the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_MB_PER_SECOND +type: +.Vt "long double *" +.Pp +The average number of megabytes transferred per second between the +acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_MB_PER_SECOND_READ +.It Dv DSM_MB_PER_SECOND_WRITE +.It Dv DSM_MB_PER_SECOND_FREE +type: +.Vt "long double *" +.Pp +The average number of megabytes per second in the specified type of +transaction between the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_BLOCKS_PER_SECOND +type: +.Vt "long double *" +.Pp +The average number of blocks transferred per second between the acquisition of +.Fa previous +and +.Fa current . +This number is in terms of the blocksize reported by the device. +If no blocksize has been reported (i.e., the block size is 0), a default +blocksize of 512 bytes will be used in the calculation. +.It Dv DSM_BLOCKS_PER_SECOND_READ +.It Dv DSM_BLOCKS_PER_SECOND_WRITE +.It Dv DSM_BLOCKS_PER_SECOND_FREE +type: +.Vt "long double *" +.Pp +The average number of blocks per second in the specified type of transaction +between the acquisition of +.Fa previous +and +.Fa current . +This number is in terms of the blocksize reported by the device. +If no blocksize has been reported (i.e., the block size is 0), a default +blocksize of 512 bytes will be used in the calculation. +.It Dv DSM_MS_PER_TRANSACTION +type: +.Vt "long double *" +.Pp +The average duration of transactions between the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_MS_PER_TRANSACTION_OTHER +.It Dv DSM_MS_PER_TRANSACTION_READ +.It Dv DSM_MS_PER_TRANSACTION_WRITE +.It Dv DSM_MS_PER_TRANSACTION_FREE +type: +.Vt "long double *" +.Pp +The average duration of transactions of the specified type between the +acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_BUSY_PCT +type: +.Vt "long double *" +.Pp +The percentage of time the device had one or more transactions outstanding +between the acquisition of +.Fa previous +and +.Fa current . +.It Dv DSM_QUEUE_LENGTH +type: +.Vt "u_int64_t *" +.Pp +The number of not yet completed transactions at the time when +.Fa current +was acquired. +.It Dv DSM_SKIP +type: N/A +.Pp +If you do not need a result from +.Fn devstat_compute_statistics , +just put +.Dv DSM_SKIP +as first (type) parameter and +.Dv NULL +as second parameter. +This can be useful in scenarios where the statistics to be calculated +are determined at run time. +.El +.Pp +The +.Fn devstat_compute_etime +function +provides an easy way to find the difference in seconds between two +.Vt bintime +structures. +This is most commonly used in conjunction with the time recorded by the +.Fn devstat_getdevs +function (in +.Vt "struct statinfo" ) +each time it fetches the current +.Nm +list. +.Sh RETURN VALUES +The +.Fn devstat_getnumdevs , +.Fn devstat_getgeneration , +and +.Fn devstat_getversion +function +return the indicated sysctl variable, or \-1 if there is an error +fetching the variable. +.Pp +The +.Fn devstat_checkversion +function +returns 0 if the kernel and userland +.Nm +versions match. +If they do not match, it returns \-1. +.Pp +The +.Fn devstat_getdevs +and +.Fn devstat_selectdevs +functions +return \-1 in case of an error, 0 if there is no error, and 1 if the device +list or selected devices have changed. +A return value of 1 from +.Fn devstat_getdevs +is usually a hint to re-run +.Fn devstat_selectdevs +because the device list has changed. +.Pp +The +.Fn devstat_buildmatch +function returns \-1 for error, and 0 if there is no error. +.Pp +The +.Fn devstat_compute_etime +function +returns the computed elapsed time. +.Pp +The +.Fn devstat_compute_statistics +function returns \-1 for error, and 0 for success. +.Pp +If an error is returned from one of the +.Nm +library functions, the reason for the error is generally printed in +the global string +.Va devstat_errbuf +which is +.Dv DEVSTAT_ERRBUF_SIZE +characters long. +.Sh SEE ALSO +.Xr systat 1 , +.Xr kvm 3 , +.Xr sysctl 3 , +.Xr iostat 8 , +.Xr rpc.rstatd 8 , +.Xr sysctl 8 , +.Xr vmstat 8 , +.Xr devstat 9 +.Sh HISTORY +The +.Nm +statistics system first appeared in +.Fx 3.0 . +The new interface (the functions prefixed with +.Li devstat_ ) +first appeared in +.Fx 5.0 . +.Sh AUTHORS +.An Kenneth Merry Aq ken@FreeBSD.org +.Sh BUGS +There should probably be an interface to de-allocate memory allocated by +.Fn devstat_getdevs , +.Fn devstat_selectdevs , +and +.Fn devstat_buildmatch . +.Pp +The +.Fn devstat_selectdevs +function +should probably not select more than +.Fa maxshowdevs +devices in +.Dq top +mode when no devices have been selected previously. +.Pp +There should probably be functions to perform the statistics buffer +swapping that goes on in most of the clients of this library. +.Pp +The +.Vt statinfo +and +.Vt devinfo +structures should probably be cleaned up and thought out a little more. diff --git a/lib/libdevstat/devstat.c b/lib/libdevstat/devstat.c new file mode 100644 index 0000000..d0ba704 --- /dev/null +++ b/lib/libdevstat/devstat.c @@ -0,0 +1,1631 @@ +/* + * Copyright (c) 1997, 1998 Kenneth D. Merry. + * All rights reserved. + * + * 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. The name of the author may not be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``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 AUTHOR 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. + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +#include <sys/types.h> +#include <sys/sysctl.h> +#include <sys/errno.h> +#include <sys/resource.h> +#include <sys/queue.h> + +#include <ctype.h> +#include <err.h> +#include <fcntl.h> +#include <limits.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <stdarg.h> +#include <kvm.h> +#include <nlist.h> + +#include "devstat.h" + +int +compute_stats(struct devstat *current, struct devstat *previous, + long double etime, u_int64_t *total_bytes, + u_int64_t *total_transfers, u_int64_t *total_blocks, + long double *kb_per_transfer, long double *transfers_per_second, + long double *mb_per_second, long double *blocks_per_second, + long double *ms_per_transaction); + +typedef enum { + DEVSTAT_ARG_NOTYPE, + DEVSTAT_ARG_UINT64, + DEVSTAT_ARG_LD, + DEVSTAT_ARG_SKIP +} devstat_arg_type; + +char devstat_errbuf[DEVSTAT_ERRBUF_SIZE]; + +/* + * Table to match descriptive strings with device types. These are in + * order from most common to least common to speed search time. + */ +struct devstat_match_table match_table[] = { + {"da", DEVSTAT_TYPE_DIRECT, DEVSTAT_MATCH_TYPE}, + {"cd", DEVSTAT_TYPE_CDROM, DEVSTAT_MATCH_TYPE}, + {"scsi", DEVSTAT_TYPE_IF_SCSI, DEVSTAT_MATCH_IF}, + {"ide", DEVSTAT_TYPE_IF_IDE, DEVSTAT_MATCH_IF}, + {"other", DEVSTAT_TYPE_IF_OTHER, DEVSTAT_MATCH_IF}, + {"worm", DEVSTAT_TYPE_WORM, DEVSTAT_MATCH_TYPE}, + {"sa", DEVSTAT_TYPE_SEQUENTIAL,DEVSTAT_MATCH_TYPE}, + {"pass", DEVSTAT_TYPE_PASS, DEVSTAT_MATCH_PASS}, + {"optical", DEVSTAT_TYPE_OPTICAL, DEVSTAT_MATCH_TYPE}, + {"array", DEVSTAT_TYPE_STORARRAY, DEVSTAT_MATCH_TYPE}, + {"changer", DEVSTAT_TYPE_CHANGER, DEVSTAT_MATCH_TYPE}, + {"scanner", DEVSTAT_TYPE_SCANNER, DEVSTAT_MATCH_TYPE}, + {"printer", DEVSTAT_TYPE_PRINTER, DEVSTAT_MATCH_TYPE}, + {"floppy", DEVSTAT_TYPE_FLOPPY, DEVSTAT_MATCH_TYPE}, + {"proc", DEVSTAT_TYPE_PROCESSOR, DEVSTAT_MATCH_TYPE}, + {"comm", DEVSTAT_TYPE_COMM, DEVSTAT_MATCH_TYPE}, + {"enclosure", DEVSTAT_TYPE_ENCLOSURE, DEVSTAT_MATCH_TYPE}, + {NULL, 0, 0} +}; + +struct devstat_args { + devstat_metric metric; + devstat_arg_type argtype; +} devstat_arg_list[] = { + { DSM_NONE, DEVSTAT_ARG_NOTYPE }, + { DSM_TOTAL_BYTES, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_BYTES_READ, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_BYTES_WRITE, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_TRANSFERS, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_TRANSFERS_READ, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_TRANSFERS_WRITE, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_TRANSFERS_OTHER, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_BLOCKS, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_BLOCKS_READ, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_BLOCKS_WRITE, DEVSTAT_ARG_UINT64 }, + { DSM_KB_PER_TRANSFER, DEVSTAT_ARG_LD }, + { DSM_KB_PER_TRANSFER_READ, DEVSTAT_ARG_LD }, + { DSM_KB_PER_TRANSFER_WRITE, DEVSTAT_ARG_LD }, + { DSM_TRANSFERS_PER_SECOND, DEVSTAT_ARG_LD }, + { DSM_TRANSFERS_PER_SECOND_READ, DEVSTAT_ARG_LD }, + { DSM_TRANSFERS_PER_SECOND_WRITE, DEVSTAT_ARG_LD }, + { DSM_TRANSFERS_PER_SECOND_OTHER, DEVSTAT_ARG_LD }, + { DSM_MB_PER_SECOND, DEVSTAT_ARG_LD }, + { DSM_MB_PER_SECOND_READ, DEVSTAT_ARG_LD }, + { DSM_MB_PER_SECOND_WRITE, DEVSTAT_ARG_LD }, + { DSM_BLOCKS_PER_SECOND, DEVSTAT_ARG_LD }, + { DSM_BLOCKS_PER_SECOND_READ, DEVSTAT_ARG_LD }, + { DSM_BLOCKS_PER_SECOND_WRITE, DEVSTAT_ARG_LD }, + { DSM_MS_PER_TRANSACTION, DEVSTAT_ARG_LD }, + { DSM_MS_PER_TRANSACTION_READ, DEVSTAT_ARG_LD }, + { DSM_MS_PER_TRANSACTION_WRITE, DEVSTAT_ARG_LD }, + { DSM_SKIP, DEVSTAT_ARG_SKIP }, + { DSM_TOTAL_BYTES_FREE, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_TRANSFERS_FREE, DEVSTAT_ARG_UINT64 }, + { DSM_TOTAL_BLOCKS_FREE, DEVSTAT_ARG_UINT64 }, + { DSM_KB_PER_TRANSFER_FREE, DEVSTAT_ARG_LD }, + { DSM_MB_PER_SECOND_FREE, DEVSTAT_ARG_LD }, + { DSM_TRANSFERS_PER_SECOND_FREE, DEVSTAT_ARG_LD }, + { DSM_BLOCKS_PER_SECOND_FREE, DEVSTAT_ARG_LD }, + { DSM_MS_PER_TRANSACTION_OTHER, DEVSTAT_ARG_LD }, + { DSM_MS_PER_TRANSACTION_FREE, DEVSTAT_ARG_LD }, + { DSM_BUSY_PCT, DEVSTAT_ARG_LD }, + { DSM_QUEUE_LENGTH, DEVSTAT_ARG_UINT64 }, +}; + +static const char *namelist[] = { +#define X_NUMDEVS 0 + "_devstat_num_devs", +#define X_GENERATION 1 + "_devstat_generation", +#define X_VERSION 2 + "_devstat_version", +#define X_DEVICE_STATQ 3 + "_device_statq", +#define X_END 4 +}; + +/* + * Local function declarations. + */ +static int compare_select(const void *arg1, const void *arg2); +static int readkmem(kvm_t *kd, unsigned long addr, void *buf, size_t nbytes); +static int readkmem_nl(kvm_t *kd, const char *name, void *buf, size_t nbytes); +static char *get_devstat_kvm(kvm_t *kd); + +#define KREADNL(kd, var, val) \ + readkmem_nl(kd, namelist[var], &val, sizeof(val)) + +int +devstat_getnumdevs(kvm_t *kd) +{ + size_t numdevsize; + int numdevs; + + numdevsize = sizeof(int); + + /* + * Find out how many devices we have in the system. + */ + if (kd == NULL) { + if (sysctlbyname("kern.devstat.numdevs", &numdevs, + &numdevsize, NULL, 0) == -1) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: error getting number of devices\n" + "%s: %s", __func__, __func__, + strerror(errno)); + return(-1); + } else + return(numdevs); + } else { + + if (KREADNL(kd, X_NUMDEVS, numdevs) == -1) + return(-1); + else + return(numdevs); + } +} + +/* + * This is an easy way to get the generation number, but the generation is + * supplied in a more atmoic manner by the kern.devstat.all sysctl. + * Because this generation sysctl is separate from the statistics sysctl, + * the device list and the generation could change between the time that + * this function is called and the device list is retreived. + */ +long +devstat_getgeneration(kvm_t *kd) +{ + size_t gensize; + long generation; + + gensize = sizeof(long); + + /* + * Get the current generation number. + */ + if (kd == NULL) { + if (sysctlbyname("kern.devstat.generation", &generation, + &gensize, NULL, 0) == -1) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: error getting devstat generation\n%s: %s", + __func__, __func__, strerror(errno)); + return(-1); + } else + return(generation); + } else { + if (KREADNL(kd, X_GENERATION, generation) == -1) + return(-1); + else + return(generation); + } +} + +/* + * Get the current devstat version. The return value of this function + * should be compared with DEVSTAT_VERSION, which is defined in + * sys/devicestat.h. This will enable userland programs to determine + * whether they are out of sync with the kernel. + */ +int +devstat_getversion(kvm_t *kd) +{ + size_t versize; + int version; + + versize = sizeof(int); + + /* + * Get the current devstat version. + */ + if (kd == NULL) { + if (sysctlbyname("kern.devstat.version", &version, &versize, + NULL, 0) == -1) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: error getting devstat version\n%s: %s", + __func__, __func__, strerror(errno)); + return(-1); + } else + return(version); + } else { + if (KREADNL(kd, X_VERSION, version) == -1) + return(-1); + else + return(version); + } +} + +/* + * Check the devstat version we know about against the devstat version the + * kernel knows about. If they don't match, print an error into the + * devstat error buffer, and return -1. If they match, return 0. + */ +int +devstat_checkversion(kvm_t *kd) +{ + int buflen, res, retval = 0, version; + + version = devstat_getversion(kd); + + if (version != DEVSTAT_VERSION) { + /* + * If getversion() returns an error (i.e. -1), then it + * has printed an error message in the buffer. Therefore, + * we need to add a \n to the end of that message before we + * print our own message in the buffer. + */ + if (version == -1) + buflen = strlen(devstat_errbuf); + else + buflen = 0; + + res = snprintf(devstat_errbuf + buflen, + DEVSTAT_ERRBUF_SIZE - buflen, + "%s%s: userland devstat version %d is not " + "the same as the kernel\n%s: devstat " + "version %d\n", version == -1 ? "\n" : "", + __func__, DEVSTAT_VERSION, __func__, version); + + if (res < 0) + devstat_errbuf[buflen] = '\0'; + + buflen = strlen(devstat_errbuf); + if (version < DEVSTAT_VERSION) + res = snprintf(devstat_errbuf + buflen, + DEVSTAT_ERRBUF_SIZE - buflen, + "%s: libdevstat newer than kernel\n", + __func__); + else + res = snprintf(devstat_errbuf + buflen, + DEVSTAT_ERRBUF_SIZE - buflen, + "%s: kernel newer than libdevstat\n", + __func__); + + if (res < 0) + devstat_errbuf[buflen] = '\0'; + + retval = -1; + } + + return(retval); +} + +/* + * Get the current list of devices and statistics, and the current + * generation number. + * + * Return values: + * -1 -- error + * 0 -- device list is unchanged + * 1 -- device list has changed + */ +int +devstat_getdevs(kvm_t *kd, struct statinfo *stats) +{ + int error; + size_t dssize; + int oldnumdevs; + long oldgeneration; + int retval = 0; + struct devinfo *dinfo; + struct timespec ts; + + dinfo = stats->dinfo; + + if (dinfo == NULL) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: stats->dinfo was NULL", __func__); + return(-1); + } + + oldnumdevs = dinfo->numdevs; + oldgeneration = dinfo->generation; + + clock_gettime(CLOCK_MONOTONIC, &ts); + stats->snap_time = ts.tv_sec + ts.tv_nsec * 1e-9; + + if (kd == NULL) { + /* If this is our first time through, mem_ptr will be null. */ + if (dinfo->mem_ptr == NULL) { + /* + * Get the number of devices. If it's negative, it's an + * error. Don't bother setting the error string, since + * getnumdevs() has already done that for us. + */ + if ((dinfo->numdevs = devstat_getnumdevs(kd)) < 0) + return(-1); + + /* + * The kern.devstat.all sysctl returns the current + * generation number, as well as all the devices. + * So we need four bytes more. + */ + dssize = (dinfo->numdevs * sizeof(struct devstat)) + + sizeof(long); + dinfo->mem_ptr = (u_int8_t *)malloc(dssize); + } else + dssize = (dinfo->numdevs * sizeof(struct devstat)) + + sizeof(long); + + /* + * Request all of the devices. We only really allow for one + * ENOMEM failure. It would, of course, be possible to just go + * in a loop and keep reallocing the device structure until we + * don't get ENOMEM back. I'm not sure it's worth it, though. + * If devices are being added to the system that quickly, maybe + * the user can just wait until all devices are added. + */ + for (;;) { + error = sysctlbyname("kern.devstat.all", + dinfo->mem_ptr, + &dssize, NULL, 0); + if (error != -1 || errno != EBUSY) + break; + } + if (error == -1) { + /* + * If we get ENOMEM back, that means that there are + * more devices now, so we need to allocate more + * space for the device array. + */ + if (errno == ENOMEM) { + /* + * No need to set the error string here, + * devstat_getnumdevs() will do that if it fails. + */ + if ((dinfo->numdevs = devstat_getnumdevs(kd)) < 0) + return(-1); + + dssize = (dinfo->numdevs * + sizeof(struct devstat)) + sizeof(long); + dinfo->mem_ptr = (u_int8_t *) + realloc(dinfo->mem_ptr, dssize); + if ((error = sysctlbyname("kern.devstat.all", + dinfo->mem_ptr, &dssize, NULL, 0)) == -1) { + snprintf(devstat_errbuf, + sizeof(devstat_errbuf), + "%s: error getting device " + "stats\n%s: %s", __func__, + __func__, strerror(errno)); + return(-1); + } + } else { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: error getting device stats\n" + "%s: %s", __func__, __func__, + strerror(errno)); + return(-1); + } + } + + } else { + /* + * This is of course non-atomic, but since we are working + * on a core dump, the generation is unlikely to change + */ + if ((dinfo->numdevs = devstat_getnumdevs(kd)) == -1) + return(-1); + if ((dinfo->mem_ptr = (u_int8_t *)get_devstat_kvm(kd)) == NULL) + return(-1); + } + /* + * The sysctl spits out the generation as the first four bytes, + * then all of the device statistics structures. + */ + dinfo->generation = *(long *)dinfo->mem_ptr; + + /* + * If the generation has changed, and if the current number of + * devices is not the same as the number of devices recorded in the + * devinfo structure, it is likely that the device list has shrunk. + * The reason that it is likely that the device list has shrunk in + * this case is that if the device list has grown, the sysctl above + * will return an ENOMEM error, and we will reset the number of + * devices and reallocate the device array. If the second sysctl + * fails, we will return an error and therefore never get to this + * point. If the device list has shrunk, the sysctl will not + * return an error since we have more space allocated than is + * necessary. So, in the shrinkage case, we catch it here and + * reallocate the array so that we don't use any more space than is + * necessary. + */ + if (oldgeneration != dinfo->generation) { + if (devstat_getnumdevs(kd) != dinfo->numdevs) { + if ((dinfo->numdevs = devstat_getnumdevs(kd)) < 0) + return(-1); + dssize = (dinfo->numdevs * sizeof(struct devstat)) + + sizeof(long); + dinfo->mem_ptr = (u_int8_t *)realloc(dinfo->mem_ptr, + dssize); + } + retval = 1; + } + + dinfo->devices = (struct devstat *)(dinfo->mem_ptr + sizeof(long)); + + return(retval); +} + +/* + * selectdevs(): + * + * Devices are selected/deselected based upon the following criteria: + * - devices specified by the user on the command line + * - devices matching any device type expressions given on the command line + * - devices with the highest I/O, if 'top' mode is enabled + * - the first n unselected devices in the device list, if maxshowdevs + * devices haven't already been selected and if the user has not + * specified any devices on the command line and if we're in "add" mode. + * + * Input parameters: + * - device selection list (dev_select) + * - current number of devices selected (num_selected) + * - total number of devices in the selection list (num_selections) + * - devstat generation as of the last time selectdevs() was called + * (select_generation) + * - current devstat generation (current_generation) + * - current list of devices and statistics (devices) + * - number of devices in the current device list (numdevs) + * - compiled version of the command line device type arguments (matches) + * - This is optional. If the number of devices is 0, this will be ignored. + * - The matching code pays attention to the current selection mode. So + * if you pass in a matching expression, it will be evaluated based + * upon the selection mode that is passed in. See below for details. + * - number of device type matching expressions (num_matches) + * - Set to 0 to disable the matching code. + * - list of devices specified on the command line by the user (dev_selections) + * - number of devices selected on the command line by the user + * (num_dev_selections) + * - Our selection mode. There are four different selection modes: + * - add mode. (DS_SELECT_ADD) Any devices matching devices explicitly + * selected by the user or devices matching a pattern given by the + * user will be selected in addition to devices that are already + * selected. Additional devices will be selected, up to maxshowdevs + * number of devices. + * - only mode. (DS_SELECT_ONLY) Only devices matching devices + * explicitly given by the user or devices matching a pattern + * given by the user will be selected. No other devices will be + * selected. + * - addonly mode. (DS_SELECT_ADDONLY) This is similar to add and + * only. Basically, this will not de-select any devices that are + * current selected, as only mode would, but it will also not + * gratuitously select up to maxshowdevs devices as add mode would. + * - remove mode. (DS_SELECT_REMOVE) Any devices matching devices + * explicitly selected by the user or devices matching a pattern + * given by the user will be de-selected. + * - maximum number of devices we can select (maxshowdevs) + * - flag indicating whether or not we're in 'top' mode (perf_select) + * + * Output data: + * - the device selection list may be modified and passed back out + * - the number of devices selected and the total number of items in the + * device selection list may be changed + * - the selection generation may be changed to match the current generation + * + * Return values: + * -1 -- error + * 0 -- selected devices are unchanged + * 1 -- selected devices changed + */ +int +devstat_selectdevs(struct device_selection **dev_select, int *num_selected, + int *num_selections, long *select_generation, + long current_generation, struct devstat *devices, + int numdevs, struct devstat_match *matches, int num_matches, + char **dev_selections, int num_dev_selections, + devstat_select_mode select_mode, int maxshowdevs, + int perf_select) +{ + int i, j, k; + int init_selections = 0, init_selected_var = 0; + struct device_selection *old_dev_select = NULL; + int old_num_selections = 0, old_num_selected; + int selection_number = 0; + int changed = 0, found = 0; + + if ((dev_select == NULL) || (devices == NULL) || (numdevs < 0)) + return(-1); + + /* + * We always want to make sure that we have as many dev_select + * entries as there are devices. + */ + /* + * In this case, we haven't selected devices before. + */ + if (*dev_select == NULL) { + *dev_select = (struct device_selection *)malloc(numdevs * + sizeof(struct device_selection)); + *select_generation = current_generation; + init_selections = 1; + changed = 1; + /* + * In this case, we have selected devices before, but the device + * list has changed since we last selected devices, so we need to + * either enlarge or reduce the size of the device selection list. + */ + } else if (*num_selections != numdevs) { + *dev_select = (struct device_selection *)realloc(*dev_select, + numdevs * sizeof(struct device_selection)); + *select_generation = current_generation; + init_selections = 1; + /* + * In this case, we've selected devices before, and the selection + * list is the same size as it was the last time, but the device + * list has changed. + */ + } else if (*select_generation < current_generation) { + *select_generation = current_generation; + init_selections = 1; + } + + /* + * If we're in "only" mode, we want to clear out the selected + * variable since we're going to select exactly what the user wants + * this time through. + */ + if (select_mode == DS_SELECT_ONLY) + init_selected_var = 1; + + /* + * In all cases, we want to back up the number of selected devices. + * It is a quick and accurate way to determine whether the selected + * devices have changed. + */ + old_num_selected = *num_selected; + + /* + * We want to make a backup of the current selection list if + * the list of devices has changed, or if we're in performance + * selection mode. In both cases, we don't want to make a backup + * if we already know for sure that the list will be different. + * This is certainly the case if this is our first time through the + * selection code. + */ + if (((init_selected_var != 0) || (init_selections != 0) + || (perf_select != 0)) && (changed == 0)){ + old_dev_select = (struct device_selection *)malloc( + *num_selections * sizeof(struct device_selection)); + old_num_selections = *num_selections; + bcopy(*dev_select, old_dev_select, + sizeof(struct device_selection) * *num_selections); + } + + if (init_selections != 0) { + bzero(*dev_select, sizeof(struct device_selection) * numdevs); + + for (i = 0; i < numdevs; i++) { + (*dev_select)[i].device_number = + devices[i].device_number; + strncpy((*dev_select)[i].device_name, + devices[i].device_name, + DEVSTAT_NAME_LEN); + (*dev_select)[i].device_name[DEVSTAT_NAME_LEN - 1]='\0'; + (*dev_select)[i].unit_number = devices[i].unit_number; + (*dev_select)[i].position = i; + } + *num_selections = numdevs; + } else if (init_selected_var != 0) { + for (i = 0; i < numdevs; i++) + (*dev_select)[i].selected = 0; + } + + /* we haven't gotten around to selecting anything yet.. */ + if ((select_mode == DS_SELECT_ONLY) || (init_selections != 0) + || (init_selected_var != 0)) + *num_selected = 0; + + /* + * Look through any devices the user specified on the command line + * and see if they match known devices. If so, select them. + */ + for (i = 0; (i < *num_selections) && (num_dev_selections > 0); i++) { + char tmpstr[80]; + + snprintf(tmpstr, sizeof(tmpstr), "%s%d", + (*dev_select)[i].device_name, + (*dev_select)[i].unit_number); + for (j = 0; j < num_dev_selections; j++) { + if (strcmp(tmpstr, dev_selections[j]) == 0) { + /* + * Here we do different things based on the + * mode we're in. If we're in add or + * addonly mode, we only select this device + * if it hasn't already been selected. + * Otherwise, we would be unnecessarily + * changing the selection order and + * incrementing the selection count. If + * we're in only mode, we unconditionally + * select this device, since in only mode + * any previous selections are erased and + * manually specified devices are the first + * ones to be selected. If we're in remove + * mode, we de-select the specified device and + * decrement the selection count. + */ + switch(select_mode) { + case DS_SELECT_ADD: + case DS_SELECT_ADDONLY: + if ((*dev_select)[i].selected) + break; + /* FALLTHROUGH */ + case DS_SELECT_ONLY: + (*dev_select)[i].selected = + ++selection_number; + (*num_selected)++; + break; + case DS_SELECT_REMOVE: + (*dev_select)[i].selected = 0; + (*num_selected)--; + /* + * This isn't passed back out, we + * just use it to keep track of + * how many devices we've removed. + */ + num_dev_selections--; + break; + } + break; + } + } + } + + /* + * Go through the user's device type expressions and select devices + * accordingly. We only do this if the number of devices already + * selected is less than the maximum number we can show. + */ + for (i = 0; (i < num_matches) && (*num_selected < maxshowdevs); i++) { + /* We should probably indicate some error here */ + if ((matches[i].match_fields == DEVSTAT_MATCH_NONE) + || (matches[i].num_match_categories <= 0)) + continue; + + for (j = 0; j < numdevs; j++) { + int num_match_categories; + + num_match_categories = matches[i].num_match_categories; + + /* + * Determine whether or not the current device + * matches the given matching expression. This if + * statement consists of three components: + * - the device type check + * - the device interface check + * - the passthrough check + * If a the matching test is successful, it + * decrements the number of matching categories, + * and if we've reached the last element that + * needed to be matched, the if statement succeeds. + * + */ + if ((((matches[i].match_fields & DEVSTAT_MATCH_TYPE)!=0) + && ((devices[j].device_type & DEVSTAT_TYPE_MASK) == + (matches[i].device_type & DEVSTAT_TYPE_MASK)) + &&(((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0) + || (((matches[i].match_fields & + DEVSTAT_MATCH_PASS) == 0) + && ((devices[j].device_type & + DEVSTAT_TYPE_PASS) == 0))) + && (--num_match_categories == 0)) + || (((matches[i].match_fields & DEVSTAT_MATCH_IF) != 0) + && ((devices[j].device_type & DEVSTAT_TYPE_IF_MASK) == + (matches[i].device_type & DEVSTAT_TYPE_IF_MASK)) + &&(((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0) + || (((matches[i].match_fields & + DEVSTAT_MATCH_PASS) == 0) + && ((devices[j].device_type & + DEVSTAT_TYPE_PASS) == 0))) + && (--num_match_categories == 0)) + || (((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0) + && ((devices[j].device_type & DEVSTAT_TYPE_PASS) != 0) + && (--num_match_categories == 0))) { + + /* + * This is probably a non-optimal solution + * to the problem that the devices in the + * device list will not be in the same + * order as the devices in the selection + * array. + */ + for (k = 0; k < numdevs; k++) { + if ((*dev_select)[k].position == j) { + found = 1; + break; + } + } + + /* + * There shouldn't be a case where a device + * in the device list is not in the + * selection list...but it could happen. + */ + if (found != 1) { + fprintf(stderr, "selectdevs: couldn't" + " find %s%d in selection " + "list\n", + devices[j].device_name, + devices[j].unit_number); + break; + } + + /* + * We do different things based upon the + * mode we're in. If we're in add or only + * mode, we go ahead and select this device + * if it hasn't already been selected. If + * it has already been selected, we leave + * it alone so we don't mess up the + * selection ordering. Manually specified + * devices have already been selected, and + * they have higher priority than pattern + * matched devices. If we're in remove + * mode, we de-select the given device and + * decrement the selected count. + */ + switch(select_mode) { + case DS_SELECT_ADD: + case DS_SELECT_ADDONLY: + case DS_SELECT_ONLY: + if ((*dev_select)[k].selected != 0) + break; + (*dev_select)[k].selected = + ++selection_number; + (*num_selected)++; + break; + case DS_SELECT_REMOVE: + (*dev_select)[k].selected = 0; + (*num_selected)--; + break; + } + } + } + } + + /* + * Here we implement "top" mode. Devices are sorted in the + * selection array based on two criteria: whether or not they are + * selected (not selection number, just the fact that they are + * selected!) and the number of bytes in the "bytes" field of the + * selection structure. The bytes field generally must be kept up + * by the user. In the future, it may be maintained by library + * functions, but for now the user has to do the work. + * + * At first glance, it may seem wrong that we don't go through and + * select every device in the case where the user hasn't specified + * any devices or patterns. In fact, though, it won't make any + * difference in the device sorting. In that particular case (i.e. + * when we're in "add" or "only" mode, and the user hasn't + * specified anything) the first time through no devices will be + * selected, so the only criterion used to sort them will be their + * performance. The second time through, and every time thereafter, + * all devices will be selected, so again selection won't matter. + */ + if (perf_select != 0) { + + /* Sort the device array by throughput */ + qsort(*dev_select, *num_selections, + sizeof(struct device_selection), + compare_select); + + if (*num_selected == 0) { + /* + * Here we select every device in the array, if it + * isn't already selected. Because the 'selected' + * variable in the selection array entries contains + * the selection order, the devstats routine can show + * the devices that were selected first. + */ + for (i = 0; i < *num_selections; i++) { + if ((*dev_select)[i].selected == 0) { + (*dev_select)[i].selected = + ++selection_number; + (*num_selected)++; + } + } + } else { + selection_number = 0; + for (i = 0; i < *num_selections; i++) { + if ((*dev_select)[i].selected != 0) { + (*dev_select)[i].selected = + ++selection_number; + } + } + } + } + + /* + * If we're in the "add" selection mode and if we haven't already + * selected maxshowdevs number of devices, go through the array and + * select any unselected devices. If we're in "only" mode, we + * obviously don't want to select anything other than what the user + * specifies. If we're in "remove" mode, it probably isn't a good + * idea to go through and select any more devices, since we might + * end up selecting something that the user wants removed. Through + * more complicated logic, we could actually figure this out, but + * that would probably require combining this loop with the various + * selections loops above. + */ + if ((select_mode == DS_SELECT_ADD) && (*num_selected < maxshowdevs)) { + for (i = 0; i < *num_selections; i++) + if ((*dev_select)[i].selected == 0) { + (*dev_select)[i].selected = ++selection_number; + (*num_selected)++; + } + } + + /* + * Look at the number of devices that have been selected. If it + * has changed, set the changed variable. Otherwise, if we've + * made a backup of the selection list, compare it to the current + * selection list to see if the selected devices have changed. + */ + if ((changed == 0) && (old_num_selected != *num_selected)) + changed = 1; + else if ((changed == 0) && (old_dev_select != NULL)) { + /* + * Now we go through the selection list and we look at + * it three different ways. + */ + for (i = 0; (i < *num_selections) && (changed == 0) && + (i < old_num_selections); i++) { + /* + * If the device at index i in both the new and old + * selection arrays has the same device number and + * selection status, it hasn't changed. We + * continue on to the next index. + */ + if (((*dev_select)[i].device_number == + old_dev_select[i].device_number) + && ((*dev_select)[i].selected == + old_dev_select[i].selected)) + continue; + + /* + * Now, if we're still going through the if + * statement, the above test wasn't true. So we + * check here to see if the device at index i in + * the current array is the same as the device at + * index i in the old array. If it is, that means + * that its selection number has changed. Set + * changed to 1 and exit the loop. + */ + else if ((*dev_select)[i].device_number == + old_dev_select[i].device_number) { + changed = 1; + break; + } + /* + * If we get here, then the device at index i in + * the current array isn't the same device as the + * device at index i in the old array. + */ + else { + found = 0; + + /* + * Search through the old selection array + * looking for a device with the same + * device number as the device at index i + * in the current array. If the selection + * status is the same, then we mark it as + * found. If the selection status isn't + * the same, we break out of the loop. + * Since found isn't set, changed will be + * set to 1 below. + */ + for (j = 0; j < old_num_selections; j++) { + if (((*dev_select)[i].device_number == + old_dev_select[j].device_number) + && ((*dev_select)[i].selected == + old_dev_select[j].selected)){ + found = 1; + break; + } + else if ((*dev_select)[i].device_number + == old_dev_select[j].device_number) + break; + } + if (found == 0) + changed = 1; + } + } + } + if (old_dev_select != NULL) + free(old_dev_select); + + return(changed); +} + +/* + * Comparison routine for qsort() above. Note that the comparison here is + * backwards -- generally, it should return a value to indicate whether + * arg1 is <, =, or > arg2. Instead, it returns the opposite. The reason + * it returns the opposite is so that the selection array will be sorted in + * order of decreasing performance. We sort on two parameters. The first + * sort key is whether or not one or the other of the devices in question + * has been selected. If one of them has, and the other one has not, the + * selected device is automatically more important than the unselected + * device. If neither device is selected, we judge the devices based upon + * performance. + */ +static int +compare_select(const void *arg1, const void *arg2) +{ + if ((((const struct device_selection *)arg1)->selected) + && (((const struct device_selection *)arg2)->selected == 0)) + return(-1); + else if ((((const struct device_selection *)arg1)->selected == 0) + && (((const struct device_selection *)arg2)->selected)) + return(1); + else if (((const struct device_selection *)arg2)->bytes < + ((const struct device_selection *)arg1)->bytes) + return(-1); + else if (((const struct device_selection *)arg2)->bytes > + ((const struct device_selection *)arg1)->bytes) + return(1); + else + return(0); +} + +/* + * Take a string with the general format "arg1,arg2,arg3", and build a + * device matching expression from it. + */ +int +devstat_buildmatch(char *match_str, struct devstat_match **matches, + int *num_matches) +{ + char *tstr[5]; + char **tempstr; + int num_args; + int i, j; + + /* We can't do much without a string to parse */ + if (match_str == NULL) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: no match expression", __func__); + return(-1); + } + + /* + * Break the (comma delimited) input string out into separate strings. + */ + for (tempstr = tstr, num_args = 0; + (*tempstr = strsep(&match_str, ",")) != NULL && (num_args < 5); + num_args++) + if (**tempstr != '\0') + if (++tempstr >= &tstr[5]) + break; + + /* The user gave us too many type arguments */ + if (num_args > 3) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: too many type arguments", __func__); + return(-1); + } + + /* + * Since you can't realloc a pointer that hasn't been malloced + * first, we malloc first and then realloc. + */ + if (*num_matches == 0) + *matches = (struct devstat_match *)malloc( + sizeof(struct devstat_match)); + else + *matches = (struct devstat_match *)realloc(*matches, + sizeof(struct devstat_match) * (*num_matches + 1)); + + /* Make sure the current entry is clear */ + bzero(&matches[0][*num_matches], sizeof(struct devstat_match)); + + /* + * Step through the arguments the user gave us and build a device + * matching expression from them. + */ + for (i = 0; i < num_args; i++) { + char *tempstr2, *tempstr3; + + /* + * Get rid of leading white space. + */ + tempstr2 = tstr[i]; + while (isspace(*tempstr2) && (*tempstr2 != '\0')) + tempstr2++; + + /* + * Get rid of trailing white space. + */ + tempstr3 = &tempstr2[strlen(tempstr2) - 1]; + + while ((*tempstr3 != '\0') && (tempstr3 > tempstr2) + && (isspace(*tempstr3))) { + *tempstr3 = '\0'; + tempstr3--; + } + + /* + * Go through the match table comparing the user's + * arguments to known device types, interfaces, etc. + */ + for (j = 0; match_table[j].match_str != NULL; j++) { + /* + * We do case-insensitive matching, in case someone + * wants to enter "SCSI" instead of "scsi" or + * something like that. Only compare as many + * characters as are in the string in the match + * table. This should help if someone tries to use + * a super-long match expression. + */ + if (strncasecmp(tempstr2, match_table[j].match_str, + strlen(match_table[j].match_str)) == 0) { + /* + * Make sure the user hasn't specified two + * items of the same type, like "da" and + * "cd". One device cannot be both. + */ + if (((*matches)[*num_matches].match_fields & + match_table[j].match_field) != 0) { + snprintf(devstat_errbuf, + sizeof(devstat_errbuf), + "%s: cannot have more than " + "one match item in a single " + "category", __func__); + return(-1); + } + /* + * If we've gotten this far, we have a + * winner. Set the appropriate fields in + * the match entry. + */ + (*matches)[*num_matches].match_fields |= + match_table[j].match_field; + (*matches)[*num_matches].device_type |= + match_table[j].type; + (*matches)[*num_matches].num_match_categories++; + break; + } + } + /* + * We should have found a match in the above for loop. If + * not, that means the user entered an invalid device type + * or interface. + */ + if ((*matches)[*num_matches].num_match_categories != (i + 1)) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: unknown match item \"%s\"", __func__, + tstr[i]); + return(-1); + } + } + + (*num_matches)++; + + return(0); +} + +/* + * Compute a number of device statistics. Only one field is mandatory, and + * that is "current". Everything else is optional. The caller passes in + * pointers to variables to hold the various statistics he desires. If he + * doesn't want a particular staistic, he should pass in a NULL pointer. + * Return values: + * 0 -- success + * -1 -- failure + */ +int +compute_stats(struct devstat *current, struct devstat *previous, + long double etime, u_int64_t *total_bytes, + u_int64_t *total_transfers, u_int64_t *total_blocks, + long double *kb_per_transfer, long double *transfers_per_second, + long double *mb_per_second, long double *blocks_per_second, + long double *ms_per_transaction) +{ + return(devstat_compute_statistics(current, previous, etime, + total_bytes ? DSM_TOTAL_BYTES : DSM_SKIP, + total_bytes, + total_transfers ? DSM_TOTAL_TRANSFERS : DSM_SKIP, + total_transfers, + total_blocks ? DSM_TOTAL_BLOCKS : DSM_SKIP, + total_blocks, + kb_per_transfer ? DSM_KB_PER_TRANSFER : DSM_SKIP, + kb_per_transfer, + transfers_per_second ? DSM_TRANSFERS_PER_SECOND : DSM_SKIP, + transfers_per_second, + mb_per_second ? DSM_MB_PER_SECOND : DSM_SKIP, + mb_per_second, + blocks_per_second ? DSM_BLOCKS_PER_SECOND : DSM_SKIP, + blocks_per_second, + ms_per_transaction ? DSM_MS_PER_TRANSACTION : DSM_SKIP, + ms_per_transaction, + DSM_NONE)); +} + + +/* This is 1/2^64 */ +#define BINTIME_SCALE 5.42101086242752217003726400434970855712890625e-20 + +long double +devstat_compute_etime(struct bintime *cur_time, struct bintime *prev_time) +{ + long double etime; + + etime = cur_time->sec; + etime += cur_time->frac * BINTIME_SCALE; + if (prev_time != NULL) { + etime -= prev_time->sec; + etime -= prev_time->frac * BINTIME_SCALE; + } + return(etime); +} + +#define DELTA(field, index) \ + (current->field[(index)] - (previous ? previous->field[(index)] : 0)) + +#define DELTA_T(field) \ + devstat_compute_etime(¤t->field, \ + (previous ? &previous->field : NULL)) + +int +devstat_compute_statistics(struct devstat *current, struct devstat *previous, + long double etime, ...) +{ + u_int64_t totalbytes, totalbytesread, totalbyteswrite, totalbytesfree; + u_int64_t totaltransfers, totaltransfersread, totaltransferswrite; + u_int64_t totaltransfersother, totalblocks, totalblocksread; + u_int64_t totalblockswrite, totaltransfersfree, totalblocksfree; + va_list ap; + devstat_metric metric; + u_int64_t *destu64; + long double *destld; + int retval, i; + + retval = 0; + + /* + * current is the only mandatory field. + */ + if (current == NULL) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: current stats structure was NULL", __func__); + return(-1); + } + + totalbytesread = DELTA(bytes, DEVSTAT_READ); + totalbyteswrite = DELTA(bytes, DEVSTAT_WRITE); + totalbytesfree = DELTA(bytes, DEVSTAT_FREE); + totalbytes = totalbytesread + totalbyteswrite + totalbytesfree; + + totaltransfersread = DELTA(operations, DEVSTAT_READ); + totaltransferswrite = DELTA(operations, DEVSTAT_WRITE); + totaltransfersother = DELTA(operations, DEVSTAT_NO_DATA); + totaltransfersfree = DELTA(operations, DEVSTAT_FREE); + totaltransfers = totaltransfersread + totaltransferswrite + + totaltransfersother + totaltransfersfree; + + totalblocks = totalbytes; + totalblocksread = totalbytesread; + totalblockswrite = totalbyteswrite; + totalblocksfree = totalbytesfree; + + if (current->block_size > 0) { + totalblocks /= current->block_size; + totalblocksread /= current->block_size; + totalblockswrite /= current->block_size; + totalblocksfree /= current->block_size; + } else { + totalblocks /= 512; + totalblocksread /= 512; + totalblockswrite /= 512; + totalblocksfree /= 512; + } + + va_start(ap, etime); + + while ((metric = (devstat_metric)va_arg(ap, devstat_metric)) != 0) { + + if (metric == DSM_NONE) + break; + + if (metric >= DSM_MAX) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: metric %d is out of range", __func__, + metric); + retval = -1; + goto bailout; + } + + switch (devstat_arg_list[metric].argtype) { + case DEVSTAT_ARG_UINT64: + destu64 = (u_int64_t *)va_arg(ap, u_int64_t *); + break; + case DEVSTAT_ARG_LD: + destld = (long double *)va_arg(ap, long double *); + break; + case DEVSTAT_ARG_SKIP: + destld = (long double *)va_arg(ap, long double *); + break; + default: + retval = -1; + goto bailout; + break; /* NOTREACHED */ + } + + if (devstat_arg_list[metric].argtype == DEVSTAT_ARG_SKIP) + continue; + + switch (metric) { + case DSM_TOTAL_BYTES: + *destu64 = totalbytes; + break; + case DSM_TOTAL_BYTES_READ: + *destu64 = totalbytesread; + break; + case DSM_TOTAL_BYTES_WRITE: + *destu64 = totalbyteswrite; + break; + case DSM_TOTAL_BYTES_FREE: + *destu64 = totalbytesfree; + break; + case DSM_TOTAL_TRANSFERS: + *destu64 = totaltransfers; + break; + case DSM_TOTAL_TRANSFERS_READ: + *destu64 = totaltransfersread; + break; + case DSM_TOTAL_TRANSFERS_WRITE: + *destu64 = totaltransferswrite; + break; + case DSM_TOTAL_TRANSFERS_FREE: + *destu64 = totaltransfersfree; + break; + case DSM_TOTAL_TRANSFERS_OTHER: + *destu64 = totaltransfersother; + break; + case DSM_TOTAL_BLOCKS: + *destu64 = totalblocks; + break; + case DSM_TOTAL_BLOCKS_READ: + *destu64 = totalblocksread; + break; + case DSM_TOTAL_BLOCKS_WRITE: + *destu64 = totalblockswrite; + break; + case DSM_TOTAL_BLOCKS_FREE: + *destu64 = totalblocksfree; + break; + case DSM_KB_PER_TRANSFER: + *destld = totalbytes; + *destld /= 1024; + if (totaltransfers > 0) + *destld /= totaltransfers; + else + *destld = 0.0; + break; + case DSM_KB_PER_TRANSFER_READ: + *destld = totalbytesread; + *destld /= 1024; + if (totaltransfersread > 0) + *destld /= totaltransfersread; + else + *destld = 0.0; + break; + case DSM_KB_PER_TRANSFER_WRITE: + *destld = totalbyteswrite; + *destld /= 1024; + if (totaltransferswrite > 0) + *destld /= totaltransferswrite; + else + *destld = 0.0; + break; + case DSM_KB_PER_TRANSFER_FREE: + *destld = totalbytesfree; + *destld /= 1024; + if (totaltransfersfree > 0) + *destld /= totaltransfersfree; + else + *destld = 0.0; + break; + case DSM_TRANSFERS_PER_SECOND: + if (etime > 0.0) { + *destld = totaltransfers; + *destld /= etime; + } else + *destld = 0.0; + break; + case DSM_TRANSFERS_PER_SECOND_READ: + if (etime > 0.0) { + *destld = totaltransfersread; + *destld /= etime; + } else + *destld = 0.0; + break; + case DSM_TRANSFERS_PER_SECOND_WRITE: + if (etime > 0.0) { + *destld = totaltransferswrite; + *destld /= etime; + } else + *destld = 0.0; + break; + case DSM_TRANSFERS_PER_SECOND_FREE: + if (etime > 0.0) { + *destld = totaltransfersfree; + *destld /= etime; + } else + *destld = 0.0; + break; + case DSM_TRANSFERS_PER_SECOND_OTHER: + if (etime > 0.0) { + *destld = totaltransfersother; + *destld /= etime; + } else + *destld = 0.0; + break; + case DSM_MB_PER_SECOND: + *destld = totalbytes; + *destld /= 1024 * 1024; + if (etime > 0.0) + *destld /= etime; + else + *destld = 0.0; + break; + case DSM_MB_PER_SECOND_READ: + *destld = totalbytesread; + *destld /= 1024 * 1024; + if (etime > 0.0) + *destld /= etime; + else + *destld = 0.0; + break; + case DSM_MB_PER_SECOND_WRITE: + *destld = totalbyteswrite; + *destld /= 1024 * 1024; + if (etime > 0.0) + *destld /= etime; + else + *destld = 0.0; + break; + case DSM_MB_PER_SECOND_FREE: + *destld = totalbytesfree; + *destld /= 1024 * 1024; + if (etime > 0.0) + *destld /= etime; + else + *destld = 0.0; + break; + case DSM_BLOCKS_PER_SECOND: + *destld = totalblocks; + if (etime > 0.0) + *destld /= etime; + else + *destld = 0.0; + break; + case DSM_BLOCKS_PER_SECOND_READ: + *destld = totalblocksread; + if (etime > 0.0) + *destld /= etime; + else + *destld = 0.0; + break; + case DSM_BLOCKS_PER_SECOND_WRITE: + *destld = totalblockswrite; + if (etime > 0.0) + *destld /= etime; + else + *destld = 0.0; + break; + case DSM_BLOCKS_PER_SECOND_FREE: + *destld = totalblocksfree; + if (etime > 0.0) + *destld /= etime; + else + *destld = 0.0; + break; + /* + * This calculation is somewhat bogus. It simply divides + * the elapsed time by the total number of transactions + * completed. While that does give the caller a good + * picture of the average rate of transaction completion, + * it doesn't necessarily give the caller a good view of + * how long transactions took to complete on average. + * Those two numbers will be different for a device that + * can handle more than one transaction at a time. e.g. + * SCSI disks doing tagged queueing. + * + * The only way to accurately determine the real average + * time per transaction would be to compute and store the + * time on a per-transaction basis. That currently isn't + * done in the kernel, and would only be desireable if it + * could be implemented in a somewhat non-intrusive and high + * performance way. + */ + case DSM_MS_PER_TRANSACTION: + if (totaltransfers > 0) { + *destld = 0; + for (i = 0; i < DEVSTAT_N_TRANS_FLAGS; i++) + *destld += DELTA_T(duration[i]); + *destld /= totaltransfers; + *destld *= 1000; + } else + *destld = 0.0; + break; + /* + * As above, these next two really only give the average + * rate of completion for read and write transactions, not + * the average time the transaction took to complete. + */ + case DSM_MS_PER_TRANSACTION_READ: + if (totaltransfersread > 0) { + *destld = DELTA_T(duration[DEVSTAT_READ]); + *destld /= totaltransfersread; + *destld *= 1000; + } else + *destld = 0.0; + break; + case DSM_MS_PER_TRANSACTION_WRITE: + if (totaltransferswrite > 0) { + *destld = DELTA_T(duration[DEVSTAT_WRITE]); + *destld /= totaltransferswrite; + *destld *= 1000; + } else + *destld = 0.0; + break; + case DSM_MS_PER_TRANSACTION_FREE: + if (totaltransfersfree > 0) { + *destld = DELTA_T(duration[DEVSTAT_FREE]); + *destld /= totaltransfersfree; + *destld *= 1000; + } else + *destld = 0.0; + break; + case DSM_MS_PER_TRANSACTION_OTHER: + if (totaltransfersother > 0) { + *destld = DELTA_T(duration[DEVSTAT_NO_DATA]); + *destld /= totaltransfersother; + *destld *= 1000; + } else + *destld = 0.0; + break; + case DSM_BUSY_PCT: + *destld = DELTA_T(busy_time); + if (*destld < 0) + *destld = 0; + *destld /= etime; + *destld *= 100; + if (*destld < 0) + *destld = 0; + break; + case DSM_QUEUE_LENGTH: + *destu64 = current->start_count - current->end_count; + break; +/* + * XXX: comment out the default block to see if any case's are missing. + */ +#if 1 + default: + /* + * This shouldn't happen, since we should have + * caught any out of range metrics at the top of + * the loop. + */ + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: unknown metric %d", __func__, metric); + retval = -1; + goto bailout; + break; /* NOTREACHED */ +#endif + } + } + +bailout: + + va_end(ap); + return(retval); +} + +static int +readkmem(kvm_t *kd, unsigned long addr, void *buf, size_t nbytes) +{ + + if (kvm_read(kd, addr, buf, nbytes) == -1) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: error reading value (kvm_read): %s", __func__, + kvm_geterr(kd)); + return(-1); + } + return(0); +} + +static int +readkmem_nl(kvm_t *kd, const char *name, void *buf, size_t nbytes) +{ + struct nlist nl[2]; + + nl[0].n_name = (char *)name; + nl[1].n_name = NULL; + + if (kvm_nlist(kd, nl) == -1) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: error getting name list (kvm_nlist): %s", + __func__, kvm_geterr(kd)); + return(-1); + } + return(readkmem(kd, nl[0].n_value, buf, nbytes)); +} + +/* + * This duplicates the functionality of the kernel sysctl handler for poking + * through crash dumps. + */ +static char * +get_devstat_kvm(kvm_t *kd) +{ + int i, wp; + long gen; + struct devstat *nds; + struct devstat ds; + struct devstatlist dhead; + int num_devs; + char *rv = NULL; + + if ((num_devs = devstat_getnumdevs(kd)) <= 0) + return(NULL); + if (KREADNL(kd, X_DEVICE_STATQ, dhead) == -1) + return(NULL); + + nds = STAILQ_FIRST(&dhead); + + if ((rv = malloc(sizeof(gen))) == NULL) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: out of memory (initial malloc failed)", + __func__); + return(NULL); + } + gen = devstat_getgeneration(kd); + memcpy(rv, &gen, sizeof(gen)); + wp = sizeof(gen); + /* + * Now push out all the devices. + */ + for (i = 0; (nds != NULL) && (i < num_devs); + nds = STAILQ_NEXT(nds, dev_links), i++) { + if (readkmem(kd, (long)nds, &ds, sizeof(ds)) == -1) { + free(rv); + return(NULL); + } + nds = &ds; + rv = (char *)reallocf(rv, sizeof(gen) + + sizeof(ds) * (i + 1)); + if (rv == NULL) { + snprintf(devstat_errbuf, sizeof(devstat_errbuf), + "%s: out of memory (malloc failed)", + __func__); + return(NULL); + } + memcpy(rv + wp, &ds, sizeof(ds)); + wp += sizeof(ds); + } + return(rv); +} diff --git a/lib/libdevstat/devstat.h b/lib/libdevstat/devstat.h new file mode 100644 index 0000000..7717cb1 --- /dev/null +++ b/lib/libdevstat/devstat.h @@ -0,0 +1,169 @@ +/* + * Copyright (c) 1997, 1998 Kenneth D. Merry. + * All rights reserved. + * + * 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. The name of the author may not be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``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 AUTHOR 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. + * + * $FreeBSD$ + */ + +#ifndef _DEVSTAT_H +#define _DEVSTAT_H +#include <sys/cdefs.h> +#include <sys/devicestat.h> +#include <sys/resource.h> + +#include <kvm.h> + +/* + * Bumped every time we change the userland API. Hopefully this doesn't + * happen very often! This should be bumped every time we have to + * increment SHLIB_MAJOR in the libdevstat Makefile (for non-backwards + * compatible API changes) and should also be bumped every time we make + * backwards-compatible API changes, so application writers have a way to + * determine when a particular feature is available. + */ +#define DEVSTAT_USER_API_VER 6 + +#define DEVSTAT_ERRBUF_SIZE 2048 /* size of the devstat library error string */ + +extern char devstat_errbuf[]; + +typedef enum { + DEVSTAT_MATCH_NONE = 0x00, + DEVSTAT_MATCH_TYPE = 0x01, + DEVSTAT_MATCH_IF = 0x02, + DEVSTAT_MATCH_PASS = 0x04 +} devstat_match_flags; + +typedef enum { + DSM_NONE, + DSM_TOTAL_BYTES, + DSM_TOTAL_BYTES_READ, + DSM_TOTAL_BYTES_WRITE, + DSM_TOTAL_TRANSFERS, + DSM_TOTAL_TRANSFERS_READ, + DSM_TOTAL_TRANSFERS_WRITE, + DSM_TOTAL_TRANSFERS_OTHER, + DSM_TOTAL_BLOCKS, + DSM_TOTAL_BLOCKS_READ, + DSM_TOTAL_BLOCKS_WRITE, + DSM_KB_PER_TRANSFER, + DSM_KB_PER_TRANSFER_READ, + DSM_KB_PER_TRANSFER_WRITE, + DSM_TRANSFERS_PER_SECOND, + DSM_TRANSFERS_PER_SECOND_READ, + DSM_TRANSFERS_PER_SECOND_WRITE, + DSM_TRANSFERS_PER_SECOND_OTHER, + DSM_MB_PER_SECOND, + DSM_MB_PER_SECOND_READ, + DSM_MB_PER_SECOND_WRITE, + DSM_BLOCKS_PER_SECOND, + DSM_BLOCKS_PER_SECOND_READ, + DSM_BLOCKS_PER_SECOND_WRITE, + DSM_MS_PER_TRANSACTION, + DSM_MS_PER_TRANSACTION_READ, + DSM_MS_PER_TRANSACTION_WRITE, + DSM_SKIP, + DSM_TOTAL_BYTES_FREE, + DSM_TOTAL_TRANSFERS_FREE, + DSM_TOTAL_BLOCKS_FREE, + DSM_KB_PER_TRANSFER_FREE, + DSM_MB_PER_SECOND_FREE, + DSM_TRANSFERS_PER_SECOND_FREE, + DSM_BLOCKS_PER_SECOND_FREE, + DSM_MS_PER_TRANSACTION_OTHER, + DSM_MS_PER_TRANSACTION_FREE, + DSM_BUSY_PCT, + DSM_QUEUE_LENGTH, + DSM_MAX +} devstat_metric; + +struct devstat_match { + devstat_match_flags match_fields; + devstat_type_flags device_type; + int num_match_categories; +}; + +struct devstat_match_table { + const char * match_str; + devstat_type_flags type; + devstat_match_flags match_field; +}; + +struct device_selection { + u_int32_t device_number; + char device_name[DEVSTAT_NAME_LEN]; + int unit_number; + int selected; + u_int64_t bytes; + int position; +}; + +struct devinfo { + struct devstat *devices; + u_int8_t *mem_ptr; + long generation; + int numdevs; +}; + +struct statinfo { + long cp_time[CPUSTATES]; + long tk_nin; + long tk_nout; + struct devinfo *dinfo; + long double snap_time; +}; + +typedef enum { + DS_SELECT_ADD, + DS_SELECT_ONLY, + DS_SELECT_REMOVE, + DS_SELECT_ADDONLY +} devstat_select_mode; + +__BEGIN_DECLS + +int devstat_getnumdevs(kvm_t *kd); +long devstat_getgeneration(kvm_t *kd); +int devstat_getversion(kvm_t *kd); +int devstat_checkversion(kvm_t *kd); +int devstat_getdevs(kvm_t *kd, struct statinfo *stats); +int devstat_selectdevs(struct device_selection **dev_select, int *num_selected, + int *num_selections, long *select_generation, + long current_generation, struct devstat *devices, + int numdevs, struct devstat_match *matches, + int num_matches, char **dev_selections, + int num_dev_selections, devstat_select_mode select_mode, + int maxshowdevs, int perf_select); +int devstat_buildmatch(char *match_str, struct devstat_match **matches, + int *num_matches); +int devstat_compute_statistics(struct devstat *current, + struct devstat *previous, + long double etime, ...); +long double devstat_compute_etime(struct bintime *cur_time, + struct bintime *prev_time); +__END_DECLS + +#endif /* _DEVSTAT_H */ |
