shaba/lvm2
1
0
Fork 0
mirror of git://sourceware.org/git/lvm2.git synced 2024-10-05 03:49:50 +03:00
Code Releases Activity
1b68841605
lvm2/libdm/libdm-stats.c
Zdenek Kabelac 39b7d1ba8f cleanup: typos in comments 
Collection of typos in code comments.
Should have no runtime effect.
2024-08-30 16:51:15 +02:00

5189 lines
128 KiB
C
Raw Blame History

<
/*
* Copyright (C) 2016 Red Hat, Inc. All rights reserved.
*
* _stats_get_extents_for_file() based in part on filefrag_fiemap() from
* e2fsprogs/misc/filefrag.c. Copyright 2003 by Theodore Ts'o.
*
* This file is part of the device-mapper userspace tools.
*
* This copyrighted material is made available to anyone wishing to use,
* modify, copy, or redistribute it subject to the terms and conditions
* of the GNU Lesser General Public License v.2.1.
*
* You should have received a copy of the GNU Lesser General Public License
* along with this program; if not, write to the Free Software Foundation,
* Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "libdm/misc/dmlib.h"
#include "libdm/misc/kdev_t.h"
#include "math.h" /* log10() */
#include <sys/sysmacros.h>
#include <sys/ioctl.h>
#include <sys/vfs.h> /* fstatfs */
#ifdef __linux__
#include <linux/fs.h> /* FS_IOC_FIEMAP */
#endif
#ifdef HAVE_LINUX_FIEMAP_H
#include <linux/fiemap.h> /* fiemap */
#endif
#ifdef HAVE_LINUX_MAGIC_H
#include <linux/magic.h> /* BTRFS_SUPER_MAGIC */
#endif
#define DM_STATS_REGION_NOT_PRESENT UINT64_MAX
#define DM_STATS_GROUP_NOT_PRESENT DM_STATS_GROUP_NONE
#define NSEC_PER_USEC 1000L
#define NSEC_PER_MSEC 1000000L
#define NSEC_PER_SEC 1000000000L
#define PRECISE_ARG "precise_timestamps"
#define HISTOGRAM_ARG "histogram:"
#define STATS_ROW_BUF_LEN 4096
#define STATS_MSG_BUF_LEN 1024
#define STATS_FIE_BUF_LEN 2048
#define SECTOR_SHIFT 9L
/* Histogram bin */
struct dm_histogram_bin {
uint64_t upper; /* Upper bound on this bin. */
uint64_t count; /* Count value for this bin. */
};
struct dm_histogram {
/* The stats handle this histogram belongs to. */
const struct dm_stats *dms;
/* The region this histogram belongs to. */
const struct dm_stats_region *region;
uint64_t sum; /* Sum of histogram bin counts. */
int nr_bins; /* Number of histogram bins assigned. */
struct dm_histogram_bin bins[];
};
/*
* See Documentation/device-mapper/statistics.txt for full descriptions
* of the device-mapper statistics counter fields.
*/
struct dm_stats_counters {
uint64_t reads; /* Num reads completed */
uint64_t reads_merged; /* Num reads merged */
uint64_t read_sectors; /* Num sectors read */
uint64_t read_nsecs; /* Num milliseconds spent reading */
uint64_t writes; /* Num writes completed */
uint64_t writes_merged; /* Num writes merged */
uint64_t write_sectors; /* Num sectors written */
uint64_t write_nsecs; /* Num milliseconds spent writing */
uint64_t io_in_progress; /* Num I/Os currently in progress */
uint64_t io_nsecs; /* Num milliseconds spent doing I/Os */
uint64_t weighted_io_nsecs; /* Weighted num milliseconds doing I/Os */
uint64_t total_read_nsecs; /* Total time spent reading in milliseconds */
uint64_t total_write_nsecs; /* Total time spent writing in milliseconds */
struct dm_histogram *histogram; /* Histogram. */
};
struct dm_stats_region {
uint64_t region_id; /* as returned by @stats_list */
uint64_t group_id;
uint64_t start;
uint64_t len;
uint64_t step;
char *program_id;
char *aux_data;
uint64_t timescale; /* precise_timestamps is per-region */
struct dm_histogram *bounds; /* histogram configuration */
struct dm_histogram *histogram; /* aggregate cache */
struct dm_stats_counters *counters;
};
struct dm_stats_group {
uint64_t group_id;
const char *alias;
dm_bitset_t regions;
struct dm_histogram *histogram;
};
struct dm_stats {
/* device binding */
int bind_major; /* device major that this dm_stats object is bound to */
int bind_minor; /* device minor that this dm_stats object is bound to */
char *bind_name; /* device-mapper device name */
char *bind_uuid; /* device-mapper UUID */
char *program_id; /* default program_id for this handle */
const char *name; /* cached device_name used for reporting */
struct dm_pool *mem; /* memory pool for region and counter tables */
struct dm_pool *hist_mem; /* separate pool for histogram tables */
struct dm_pool *group_mem; /* separate pool for group tables */
uint64_t nr_regions; /* total number of present regions */
uint64_t max_region; /* size of the regions table */
uint64_t interval_ns; /* sampling interval in nanoseconds */
uint64_t timescale; /* default sample value multiplier */
int precise; /* use precise_timestamps when creating regions */
struct dm_stats_region *regions;
struct dm_stats_group *groups;
/* statistics cursor */
uint64_t walk_flags; /* walk control flags */
uint64_t cur_flags;
uint64_t cur_group;
uint64_t cur_region;
uint64_t cur_area;
};
static char *_stats_escape_aux_data(const char *aux_data)
{
size_t aux_data_len = strlen(aux_data);
char *escaped = dm_malloc((3 * aux_data_len + 1) * sizeof(char));
size_t index = 0, i;
if (!escaped) {
log_error("Could not allocate memory for escaped "
"aux_data string.");
return NULL;
}
for (i = 0; i < aux_data_len; i++) {
if (aux_data[i] == ' ') {
escaped[index++] = '\\';
escaped[index++] = ' ';
} else if (aux_data[i] == '\\') {
escaped[index++] = '\\';
escaped[index++] = '\\';
} else if (aux_data[i] == '\t') {
escaped[index++] = '\\';
escaped[index++] = '\t';
} else {
escaped[index++] = aux_data[i];
}
}
escaped[index] = '\0';
return escaped;
}
#define PROC_SELF_COMM "/proc/self/comm"
static char *_program_id_from_proc(void)
{
FILE *comm = NULL;
char buf[STATS_ROW_BUF_LEN];
if (!(comm = fopen(PROC_SELF_COMM, "r")))
return_NULL;
if (!fgets(buf, sizeof(buf), comm)) {
log_error("Could not read from %s", PROC_SELF_COMM);
if (fclose(comm))
stack;
return NULL;
}
if (fclose(comm))
stack;
return dm_strdup(buf);
}
static uint64_t _nr_areas(uint64_t len, uint64_t step)
{
/* Default is one area. */
if (!len || !step)
return 1;
/*
* drivers/md/dm-stats.c::message_stats_create()
* A region may be sub-divided into areas with their own counters.
* Any partial area at the end of the region is treated as an
* additional complete area.
*/
return (len + step - 1) / step;
}
static uint64_t _nr_areas_region(struct dm_stats_region *region)
{
return _nr_areas(region->len, region->step);
}
struct dm_stats *dm_stats_create(const char *program_id)
{
size_t hist_hint = sizeof(struct dm_histogram_bin);
size_t group_hint = sizeof(struct dm_stats_group);
struct dm_stats *dms = NULL;
if (!(dms = dm_zalloc(sizeof(*dms))))
return_NULL;
/* FIXME: better hint. */
if (!(dms->mem = dm_pool_create("stats_pool", 4096))) {
dm_free(dms);
return_NULL;
}
if (!(dms->hist_mem = dm_pool_create("histogram_pool", hist_hint)))
goto_bad;
if (!(dms->group_mem = dm_pool_create("group_pool", group_hint)))
goto_bad;
if (!program_id || !strlen(program_id))
dms->program_id = _program_id_from_proc();
else
dms->program_id = dm_strdup(program_id);
if (!dms->program_id) {
log_error("Could not allocate memory for program_id");
goto bad;
}
dms->bind_major = -1;
dms->bind_minor = -1;
dms->bind_name = NULL;
dms->bind_uuid = NULL;
dms->name = NULL;
/* by default all regions use msec precision */
dms->timescale = NSEC_PER_MSEC;
dms->precise = 0;
dms->nr_regions = DM_STATS_REGION_NOT_PRESENT;
dms->max_region = DM_STATS_REGION_NOT_PRESENT;
dms->regions = NULL;
/* maintain compatibility with earlier walk version */
dms->walk_flags = dms->cur_flags = DM_STATS_WALK_DEFAULT;
return dms;
bad:
dm_pool_destroy(dms->mem);
if (dms->hist_mem)
dm_pool_destroy(dms->hist_mem);
if (dms->group_mem)
dm_pool_destroy(dms->group_mem);
dm_free(dms);
return NULL;
}
/*
* Test whether the stats region pointed to by region is present.
*/
static int _stats_region_present(const struct dm_stats_region *region)
{
return !(region->region_id == DM_STATS_REGION_NOT_PRESENT);
}
/*
* Test whether the stats group pointed to by group is present.
*/
static int _stats_group_present(const struct dm_stats_group *group)
{
return !(group->group_id == DM_STATS_GROUP_NOT_PRESENT);
}
/*
* Test whether a stats group id is present.
*/
static int _stats_group_id_present(const struct dm_stats *dms, uint64_t id)
{
struct dm_stats_group *group = NULL;
if (id == DM_STATS_GROUP_NOT_PRESENT)
return 0;
if (!dms)
return_0;
if (!dms->regions)
return 0;
if (id > dms->max_region)
return 0;
group = &dms->groups[id];
return _stats_group_present(group);
}
/*
* Test whether the given region_id is a member of any group.
*/
static uint64_t _stats_region_is_grouped(const struct dm_stats* dms,
uint64_t region_id)
{
uint64_t group_id;
if (region_id == DM_STATS_GROUP_NOT_PRESENT)
return 0;
if (!_stats_region_present(&dms->regions[region_id]))
return 0;
group_id = dms->regions[region_id].group_id;
return group_id != DM_STATS_GROUP_NOT_PRESENT;
}
static void _stats_histograms_destroy(struct dm_pool *mem,
struct dm_stats_region *region)
{
/* Unpopulated handle. */
if (!region->counters)
return;
/*
* Free everything in the pool back to the first histogram.
*/
if (region->counters[0].histogram)
dm_pool_free(mem, region->counters[0].histogram);
}
static void _stats_region_destroy(struct dm_stats_region *region)
{
if (!_stats_region_present(region))
return;
region->start = region->len = region->step = 0;
region->timescale = 0;
/*
* Don't free counters and histogram bounds here: they are
* dropped from the pool along with the corresponding
* regions table.
*
* The following objects are all allocated with dm_malloc.
*/
region->counters = NULL;
region->bounds = NULL;
dm_free(region->program_id);
region->program_id = NULL;
dm_free(region->aux_data);
region->aux_data = NULL;
region->region_id = DM_STATS_REGION_NOT_PRESENT;
}
static void _stats_regions_destroy(struct dm_stats *dms)
{
struct dm_pool *mem = dms->mem;
uint64_t i;
if (!dms->regions)
return;
/* walk backwards to obey pool order */
for (i = dms->max_region; (i != DM_STATS_REGION_NOT_PRESENT); i--) {
_stats_histograms_destroy(dms->hist_mem, &dms->regions[i]);
_stats_region_destroy(&dms->regions[i]);
}
dm_pool_free(mem, dms->regions);
dms->regions = NULL;
}
static void _stats_group_destroy(struct dm_stats_group *group)
{
if (!_stats_group_present(group))
return;
group->histogram = NULL;
if (group->alias) {
dm_free((char *) group->alias);
group->alias = NULL;
}
if (group->regions) {
dm_bitset_destroy(group->regions);
group->regions = NULL;
}
group->group_id = DM_STATS_GROUP_NOT_PRESENT;
}
static void _stats_groups_destroy(struct dm_stats *dms)
{
uint64_t i;
if (!dms->groups)
return;
for (i = dms->max_region; (i != DM_STATS_REGION_NOT_PRESENT); i--)
_stats_group_destroy(&dms->groups[i]);
dm_pool_free(dms->group_mem, dms->groups);
dms->groups = NULL;
}
static int _set_stats_device(struct dm_stats *dms, struct dm_task *dmt)
{
if (dms->bind_name)
return dm_task_set_name(dmt, dms->bind_name);
if (dms->bind_uuid)
return dm_task_set_uuid(dmt, dms->bind_uuid);
if (dms->bind_major > 0)
return dm_task_set_major(dmt, dms->bind_major)
&& dm_task_set_minor(dmt, dms->bind_minor);
return_0;
}
static int _stats_bound(const struct dm_stats *dms)
{
if (dms->bind_major > 0 || dms->bind_name || dms->bind_uuid)
return 1;
/* %p format specifier expects a void pointer. */
log_error("Stats handle at %p is not bound.", (const void *)dms);
return 0;
}
static void _stats_clear_binding(struct dm_stats *dms)
{
if (dms->bind_name)
dm_pool_free(dms->mem, dms->bind_name);
if (dms->bind_uuid)
dm_pool_free(dms->mem, dms->bind_uuid);
dm_free((char *) dms->name);
dms->bind_name = dms->bind_uuid = NULL;
dms->bind_major = dms->bind_minor = -1;
dms->name = NULL;
}
int dm_stats_bind_devno(struct dm_stats *dms, int major, int minor)
{
_stats_clear_binding(dms);
_stats_regions_destroy(dms);
_stats_groups_destroy(dms);
dms->bind_major = major;
dms->bind_minor = minor;
return 1;
}
int dm_stats_bind_name(struct dm_stats *dms, const char *name)
{
_stats_clear_binding(dms);
_stats_regions_destroy(dms);
_stats_groups_destroy(dms);
if (!(dms->bind_name = dm_pool_strdup(dms->mem, name)))
return_0;
return 1;
}
int dm_stats_bind_uuid(struct dm_stats *dms, const char *uuid)
{
_stats_clear_binding(dms);
_stats_regions_destroy(dms);
_stats_groups_destroy(dms);
if (!(dms->bind_uuid = dm_pool_strdup(dms->mem, uuid)))
return_0;
return 1;
}
int dm_stats_bind_from_fd(struct dm_stats *dms, int fd)
{
int major, minor;
struct stat buf;
if (fstat(fd, &buf)) {
log_error("fstat failed for fd %d.", fd);
return 0;
}
major = (int) MAJOR(buf.st_dev);
minor = (int) MINOR(buf.st_dev);
if (!dm_stats_bind_devno(dms, major, minor))
return_0;
return 1;
}
static int _stats_check_precise_timestamps(const struct dm_stats *dms)
{
/* Already checked? */
if (dms && dms->precise)
return 1;
return dm_message_supports_precise_timestamps();
}
int dm_stats_driver_supports_precise(void)
{
return _stats_check_precise_timestamps(NULL);
}
int dm_stats_driver_supports_histogram(void)
{
return _stats_check_precise_timestamps(NULL);
}
static int _fill_hist_arg(char *hist_arg, size_t hist_len, uint64_t scale,
struct dm_histogram *bounds)
{
int i, l, len = 0, nr_bins;
char *arg = hist_arg;
uint64_t value;
nr_bins = bounds->nr_bins;
for (i = 0; i < nr_bins; i++) {
value = bounds->bins[i].upper / scale;
if ((l = dm_snprintf(arg, hist_len - len, FMTu64"%s", value,
(i == (nr_bins - 1)) ? "" : ",")) < 0)
return_0;
len += l;
arg += l;
}
return 1;
}
static void *_get_hist_arg(struct dm_histogram *bounds, uint64_t scale,
size_t *len)
{
struct dm_histogram_bin *entry, *bins;
size_t hist_len = 1; /* terminating '\0' */
double value;
entry = bins = bounds->bins;
entry += bounds->nr_bins - 1;
while(entry >= bins) {
value = (double) (entry--)->upper;
/* Use lround to avoid size_t -> double cast warning. */
hist_len += 1 + (size_t) lround(log10(value / scale));
if (entry != bins)
hist_len++; /* ',' */
}
*len = hist_len;
return dm_zalloc(hist_len);
}
static char *_build_histogram_arg(struct dm_histogram *bounds, int *precise)
{
struct dm_histogram_bin *entry, *bins;
size_t hist_len;
char *hist_arg;
uint64_t scale;
entry = bins = bounds->bins;
/* Empty histogram is invalid. */
if (!bounds->nr_bins) {
log_error("Cannot format empty histogram description.");
return NULL;
}
/* Validate entries and set *precise if precision < 1ms. */
entry += bounds->nr_bins - 1;
while (entry >= bins) {
if (entry != bins) {
if (entry->upper < (entry - 1)->upper) {
log_error("Histogram boundaries must be in "
"order of increasing magnitude.");
return 0;
}
}
/*
* Only enable precise_timestamps automatically if any
* value in the histogram bounds uses precision < 1ms.
*/
if (((entry--)->upper % NSEC_PER_MSEC) && !*precise)
*precise = 1;
}
scale = (*precise) ? 1 : NSEC_PER_MSEC;
/* Calculate hist_len and allocate a character buffer. */
if (!(hist_arg = _get_hist_arg(bounds, scale, &hist_len))) {
log_error("Could not allocate memory for histogram argument.");
return 0;
}
/* Fill hist_arg with boundary strings. */
if (!_fill_hist_arg(hist_arg, hist_len, scale, bounds))
goto_bad;
return hist_arg;
bad:
log_error("Could not build histogram arguments.");
dm_free(hist_arg);
return NULL;
}
static struct dm_task *_stats_send_message(struct dm_stats *dms, char *msg)
{
struct dm_task *dmt;
if (!(dmt = dm_task_create(DM_DEVICE_TARGET_MSG)))
return_0;
if (!_set_stats_device(dms, dmt))
goto_bad;
if (!dm_task_set_message(dmt, msg))
goto_bad;
if (!dm_task_run(dmt))
goto_bad;
return dmt;
bad:
dm_task_destroy(dmt);
return NULL;
}
/*
* Cache the dm device_name for the device bound to dms.
*/
static int _stats_set_name_cache(struct dm_stats *dms)
{
struct dm_task *dmt;
if (dms->name)
return 1;
if (!(dmt = dm_task_create(DM_DEVICE_INFO)))
return_0;
if (!_set_stats_device(dms, dmt))
goto_bad;
if (!dm_task_run(dmt))
goto_bad;
if (!(dms->name = dm_strdup(dm_task_get_name(dmt))))
goto_bad;
dm_task_destroy(dmt);
return 1;
bad:
log_error("Could not retrieve device-mapper name for device.");
dm_task_destroy(dmt);
return 0;
}
/*
* update region group_id values
*/
static void _stats_update_groups(struct dm_stats *dms)
{
struct dm_stats_group *group;
uint64_t group_id, i;
for (group_id = 0; group_id < dms->max_region + 1; group_id++) {
if (!_stats_group_id_present(dms, group_id))
continue;
group = &dms->groups[group_id];
for (i = dm_bit_get_first(group->regions);
i != DM_STATS_GROUP_NOT_PRESENT;
i = dm_bit_get_next(group->regions, i))
dms->regions[i].group_id = group_id;
}
}
static void _check_group_regions_present(struct dm_stats *dms,
struct dm_stats_group *group)
{
dm_bitset_t regions = group->regions;
int64_t i, group_id;
group_id = i = dm_bit_get_first(regions);
for (; i > 0; i = dm_bit_get_next(regions, i))
if (!_stats_region_present(&dms->regions[i])) {
log_warn("Group descriptor " FMTd64 " contains "
"non-existent region_id " FMTd64 ".",
group_id, i);
dm_bit_clear(regions, i);
}
}
/*
* Parse a DMS_GROUP group descriptor embedded in a region's aux_data.
*
* DMS_GROUP="ALIAS:MEMBERS"
*
* ALIAS: group alias
* MEMBERS: list of group member region ids.
*
*/
#define DMS_GROUP_TAG "DMS_GROUP="
#define DMS_GROUP_TAG_LEN (sizeof(DMS_GROUP_TAG) - 1)
#define DMS_GROUP_SEP ':'
#define DMS_AUX_SEP "#"
#define DMS_AUX_SEP_CHAR '#'
#define DMS_GROUP_QUOTE '"'
static int _parse_aux_data_group(struct dm_stats *dms,
struct dm_stats_region *region,
struct dm_stats_group *group)
{
char *alias, *c, *end;
dm_bitset_t regions;
memset(group, 0, sizeof(*group));
group->group_id = DM_STATS_GROUP_NOT_PRESENT;
/* find start of group tag */
c = strstr(region->aux_data, DMS_GROUP_TAG);
if (!c)
return 1; /* no group is not an error */
/* extract alias from quotes */
alias = c + strlen(DMS_GROUP_TAG) + 1;
c = strchr(c, DMS_GROUP_SEP);
if (!c) {
log_error("Found malformed group tag while reading aux_data");
return 0;
}
/* terminate alias and advance to members accounting for closing quote */
*(c - 1) = '\0';
c++;
log_debug("Read alias '%s' from aux_data", alias);
if (!c) {
log_error("Found malformed group descriptor while "
"reading aux_data, expected '%c'", DMS_GROUP_SEP);
return 0;
}
/* if user aux_data follows make sure we have a terminated
* string to pass to dm_bitset_parse_list().
*/
end = strstr(c, DMS_AUX_SEP);
if (!end)
end = c + strlen(c);
*(end++) = '\0';
if (!(regions = dm_bitset_parse_list(c, NULL, 0))) {
log_error("Could not parse member list while "
"reading group aux_data");
return 0;
}
group->group_id = dm_bit_get_first(regions);
if (group->group_id != region->region_id) {
log_error("Found invalid group descriptor in region " FMTu64
" aux_data.", region->region_id);
group->group_id = DM_STATS_GROUP_NOT_PRESENT;
goto bad;
}
group->regions = regions;
group->alias = NULL;
if (strlen(alias)) {
group->alias = dm_strdup(alias);
if (!group->alias) {
log_error("Could not allocate memory for group alias");
goto bad;
}
}
/* separate group tag from user aux_data */
if ((strlen(end) > 1) || strncmp(end, "-", 1))
c = dm_strdup(end);
else
c = dm_strdup("");
if (!c) {
log_error("Could not allocate memory for user aux_data");
goto bad_alias;
}
dm_free(region->aux_data);
region->aux_data = c;
log_debug("Found group_id " FMTu64 ": alias=\"%s\"", group->group_id,
(group->alias) ? group->alias : "");
return 1;
bad_alias:
dm_free((char *) group->alias);
bad:
dm_bitset_destroy(regions);
return 0;
}
/*
* Parse a histogram specification returned by the kernel in a
* @stats_list response.
*/
static int _stats_parse_histogram_spec(struct dm_stats *dms,
struct dm_stats_region *region,
const char *histogram)
{
const char valid_chars[] = "0123456789,";
uint64_t scale = region->timescale, this_val = 0;
struct dm_pool *mem = dms->hist_mem;
struct dm_histogram_bin cur;
struct dm_histogram hist = { 0 };
int nr_bins = 1;
const char *c, *v, *val_start;
char *p, *endptr = NULL;
/* Advance past "histogram:". */
histogram = strchr(histogram, ':');
if (!histogram) {
log_error("Could not parse histogram description.");
return 0;
}
histogram++;
/* @stats_list rows are newline terminated. */
if ((p = strchr(histogram, '\n')))
*p = '\0';
if (!dm_pool_begin_object(mem, sizeof(cur)))
return_0;
hist.nr_bins = 0; /* fix later */
hist.region = region;
hist.dms = dms;
if (!dm_pool_grow_object(mem, &hist, sizeof(hist)))
goto_bad;
c = histogram;
do {
for (v = valid_chars; *v; v++)
if (*c == *v)
break;
if (!*v) {
stack;
goto badchar;
}
if (*c == ',') {
log_error("Invalid histogram description: %s",
histogram);
goto bad;
} else {
val_start = c;
endptr = NULL;
errno = 0;
this_val = strtoull(val_start, &endptr, 10);
if (errno || !endptr) {
log_error("Could not parse histogram boundary.");
goto bad;
}
c = endptr; /* Advance to units, comma, or end. */
if (*c == ',')
c++;
else if (*c || (*c == ' ')) { /* Expected ',' or NULL. */
stack;
goto badchar;
}
if (*c == ',')
c++;
cur.upper = scale * this_val;
cur.count = 0;
if (!dm_pool_grow_object(mem, &cur, sizeof(cur)))
goto_bad;
nr_bins++;
}
} while (*c && (*c != ' '));
/* final upper bound. */
cur.upper = UINT64_MAX;
if (!dm_pool_grow_object(mem, &cur, sizeof(cur)))
goto_bad;
region->bounds = dm_pool_end_object(mem);
if (!region->bounds)
return_0;
region->bounds->nr_bins = nr_bins;
log_debug("Added region histogram spec with %d entries.", nr_bins);
return 1;
badchar:
log_error("Invalid character in histogram: '%c' (0x%x)", *c, *c);
bad:
dm_pool_abandon_object(mem);
return 0;
}
static int _stats_parse_string_data(char *string_data, char **program_id,
char **aux_data, char **stats_args)
{
char *p, *next_gap, *empty_string = (char *)"";
size_t len;
/*
* String data format:
* <program_id> <aux_data> [precise_timestamps] [histogram:n1,n2,n3,..]
*/
/* Remove trailing whitespace */
len = strlen(string_data);
if (len > 0 && (string_data)[len - 1] == '\n') {
(string_data)[len - 1] = '\0';
}
p = strchr(string_data, ' ');
*program_id = string_data;
if (!p) {
*aux_data = *stats_args = empty_string;
return 1;
}
*p = '\0';
p++;
if (strstr(p, DMS_GROUP_TAG)) {
*aux_data = p;
/* Skip over the group tag */
if ((next_gap = strchr(p, DMS_AUX_SEP_CHAR)))
next_gap = strchr(next_gap, ' ');
if (next_gap) {
*(next_gap++) = '\0';
*stats_args = next_gap++;
} else
*stats_args = empty_string;
} else {
next_gap = strchr(p, ' ');
if (next_gap) {
*next_gap = '\0';
*aux_data = p;
*stats_args = next_gap + 1;
} else {
*aux_data = p;
*stats_args = empty_string;
}
}
if (!strncmp(*program_id, "-", 1))
*program_id = empty_string;
if (!strncmp(*aux_data, "-", 1))
*aux_data = empty_string;
return 1;
}
static int _stats_parse_list_region(struct dm_stats *dms,
struct dm_stats_region *region, char *line)
{
char string_data[STATS_ROW_BUF_LEN] = { 0 };
char *p, *program_id, *aux_data, *stats_args;
int r;
/*
* Parse fixed fields, line format:
*
* <region_id>: <start_sector>+<length> <step> <string data>
*
* Maximum string data size is 4096 - 1 bytes.
*/
r = sscanf(line, FMTu64 ": " FMTu64 "+" FMTu64 " " FMTu64 " %4095c",
&region->region_id, &region->start, &region->len,
&region->step, string_data);
if (r != 5) {
return 0;
}
if (!_stats_parse_string_data(string_data, &program_id, &aux_data, &stats_args)) {
return_0;
}
region->timescale = strstr(stats_args, PRECISE_ARG) ? 1 : NSEC_PER_MSEC;
p = strstr(stats_args, HISTOGRAM_ARG);
if (p) {
if (!_stats_parse_histogram_spec(dms, region, p)) {
return_0;
}
} else {
region->bounds = NULL;
}
region->histogram = NULL;
region->group_id = DM_STATS_GROUP_NOT_PRESENT;
if (!(region->program_id = dm_strdup(program_id))) {
return_0;
}
if (!(region->aux_data = dm_strdup(aux_data))) {
dm_free(region->program_id);
return_0;
}
region->counters = NULL;
return 1;
}
static int _stats_parse_list(struct dm_stats *dms, const char *resp)
{
uint64_t max_region = 0, nr_regions = 0;
struct dm_stats_region cur, fill;
struct dm_stats_group cur_group;
struct dm_pool *mem = dms->mem, *group_mem = dms->group_mem;
char line[STATS_ROW_BUF_LEN];
FILE *list_rows;
if (!resp) {
log_error("Could not parse NULL @stats_list response.");
return 0;
}
_stats_regions_destroy(dms);
_stats_groups_destroy(dms);
/* no regions */
if (!strlen(resp)) {
dms->nr_regions = dms->max_region = 0;
dms->regions = NULL;
return 1;
}
/*
* dm_task_get_message_response() returns a 'const char *' but
* since fmemopen also permits "w" it expects a 'char *'.
*/
/* coverity[alloc_strlen] intentional */
if (!(list_rows = fmemopen((char *)resp, strlen(resp), "r")))
return_0;
/* begin region table */
if (!dm_pool_begin_object(mem, 1024))
goto_bad;
/* begin group table */
if (!dm_pool_begin_object(group_mem, 32))
goto_bad;
while(fgets(line, sizeof(line), list_rows)) {
cur_group.group_id = DM_STATS_GROUP_NOT_PRESENT;
cur_group.regions = NULL;
cur_group.alias = NULL;
if (!_stats_parse_list_region(dms, &cur, line))
goto_bad;
/* handle holes in the list of region_ids */
if (cur.region_id > max_region) {
memset(&fill, 0, sizeof(fill));
memset(&cur_group, 0, sizeof(cur_group));
fill.region_id = DM_STATS_REGION_NOT_PRESENT;
cur_group.group_id = DM_STATS_GROUP_NOT_PRESENT;
do {
if (!dm_pool_grow_object(mem, &fill, sizeof(fill)))
goto_bad;
if (!dm_pool_grow_object(group_mem, &cur_group,
sizeof(cur_group)))
goto_bad;
} while (max_region++ < (cur.region_id - 1));
}
if (cur.aux_data)
if (!_parse_aux_data_group(dms, &cur, &cur_group))
log_error("Failed to parse group descriptor "
"from region_id " FMTu64 " aux_data:"
"'%s'", cur.region_id, cur.aux_data);
/* continue */
if (!dm_pool_grow_object(mem, &cur, sizeof(cur)))
goto_bad;
if (!dm_pool_grow_object(group_mem, &cur_group,
sizeof(cur_group)))
goto_bad;
max_region++;
nr_regions++;
}
if (!nr_regions)
/* no region data read from @stats_list */
goto bad;
dms->nr_regions = nr_regions;
dms->max_region = max_region - 1;
dms->regions = dm_pool_end_object(mem);
dms->groups = dm_pool_end_object(group_mem);
dm_stats_foreach_group(dms)
_check_group_regions_present(dms, &dms->groups[dms->cur_group]);
_stats_update_groups(dms);
if (fclose(list_rows))
stack;
return 1;
bad:
if (fclose(list_rows))
stack;
dm_pool_abandon_object(mem);
dm_pool_abandon_object(group_mem);
return 0;
}
int dm_stats_list(struct dm_stats *dms, const char *program_id)
{
char msg[STATS_MSG_BUF_LEN];
struct dm_task *dmt;
int r;
if (!_stats_bound(dms))
return_0;
/* allow zero-length program_id for list */
if (!program_id)
program_id = dms->program_id;
if (!_stats_set_name_cache(dms))
return_0;
if (dms->regions)
_stats_regions_destroy(dms);
r = dm_snprintf(msg, sizeof(msg), "@stats_list %s", program_id);
if (r < 0) {
log_error("Failed to prepare stats message.");
return 0;
}
if (!(dmt = _stats_send_message(dms, msg)))
return_0;
if (!_stats_parse_list(dms, dm_task_get_message_response(dmt))) {
log_error("Could not parse @stats_list response.");
goto bad;
}
dm_task_destroy(dmt);
return 1;
bad:
dm_task_destroy(dmt);
return 0;
}
/*
* Parse histogram data returned from a @stats_print operation.
*/
static int _stats_parse_histogram(struct dm_pool *mem, char *hist_str,
struct dm_histogram **histogram,
struct dm_stats_region *region)
{
const char valid_chars[] = "0123456789:";
struct dm_histogram *bounds = region->bounds;
struct dm_histogram hist = {
.nr_bins = region->bounds->nr_bins
};
const char *c, *v, *val_start;
struct dm_histogram_bin cur;
uint64_t sum = 0, this_val;
char *endptr = NULL;
int bin = 0;
c = hist_str;
if (!dm_pool_begin_object(mem, sizeof(cur)))
return_0;
if (!dm_pool_grow_object(mem, &hist, sizeof(hist)))
goto_bad;
do {
memset(&cur, 0, sizeof(cur));
for (v = valid_chars; *v; v++)
if (*c == *v)
break;
if (!*v)
goto badchar;
if (*c == ',')
goto badchar;
else {
val_start = c;
endptr = NULL;
errno = 0;
this_val = strtoull(val_start, &endptr, 10);
if (errno || !endptr) {
log_error("Could not parse histogram value.");
goto bad;
}
c = endptr; /* Advance to colon, or end. */
if (*c == ':')
c++;
else if (*c & (*c != '\n'))
/* Expected ':', '\n', or NULL. */
goto badchar;
if (*c == ':')
c++;
cur.upper = bounds->bins[bin].upper;
cur.count = this_val;
sum += this_val;
if (!dm_pool_grow_object(mem, &cur, sizeof(cur)))
goto_bad;
bin++;
}
} while (*c && (*c != '\n'));
log_debug("Added region histogram data with %d entries.", hist.nr_bins);
*histogram = dm_pool_end_object(mem);
(*histogram)->sum = sum;
return 1;
badchar:
log_error("Invalid character in histogram data: '%c' (0x%x)", *c, *c);
bad:
dm_pool_abandon_object(mem);
return 0;
}
static int _stats_parse_region(struct dm_stats *dms, const char *resp,
struct dm_stats_region *region,
uint64_t timescale)
{
struct dm_histogram *hist = NULL;
struct dm_pool *mem = dms->mem;
struct dm_stats_counters cur;
FILE *stats_rows = NULL;
uint64_t start = 0, len = 0;
char row[STATS_ROW_BUF_LEN];
int r;
if (!resp) {
log_error("Could not parse empty @stats_print response.");
return 0;
}
region->start = UINT64_MAX;
if (!dm_pool_begin_object(mem, 512))
goto_bad;
/*
* dm_task_get_message_response() returns a 'const char *' but
* since fmemopen also permits "w" it expects a 'char *'.
*/
/* coverity[alloc_strlen] intentional */
stats_rows = fmemopen((char *)resp, strlen(resp), "r");
if (!stats_rows)
goto_bad;
/*
* Output format for each step-sized area of a region:
*
* <start_sector>+<length> counters
*
* The first 11 counters have the same meaning as
* /sys/block/ * /stat or /proc/diskstats.
*
* Please refer to Documentation/iostats.txt for details.
*
* 1. the number of reads completed
* 2. the number of reads merged
* 3. the number of sectors read
* 4. the number of milliseconds spent reading
* 5. the number of writes completed
* 6. the number of writes merged
* 7. the number of sectors written
* 8. the number of milliseconds spent writing
* 9. the number of I/Os currently in progress
* 10. the number of milliseconds spent doing I/Os
* 11. the weighted number of milliseconds spent doing I/Os
*
* Additional counters:
* 12. the total time spent reading in milliseconds
* 13. the total time spent writing in milliseconds
*
*/
while (fgets(row, sizeof(row), stats_rows)) {
r = sscanf(row, FMTu64 "+" FMTu64 /* start+len */
/* reads */
FMTu64 " " FMTu64 " " FMTu64 " " FMTu64 " "
/* writes */
FMTu64 " " FMTu64 " " FMTu64 " " FMTu64 " "
/* in flight & io nsecs */
FMTu64 " " FMTu64 " " FMTu64 " "
/* tot read/write nsecs */
FMTu64 " " FMTu64, &start, &len,
&cur.reads, &cur.reads_merged, &cur.read_sectors,
&cur.read_nsecs,
&cur.writes, &cur.writes_merged, &cur.write_sectors,
&cur.write_nsecs,
&cur.io_in_progress,
&cur.io_nsecs, &cur.weighted_io_nsecs,
&cur.total_read_nsecs, &cur.total_write_nsecs);
if (r != 15) {
log_error("Could not parse @stats_print row.");
goto bad;
}
/* scale time values up if needed */
if (timescale != 1) {
cur.read_nsecs *= timescale;
cur.write_nsecs *= timescale;
cur.io_nsecs *= timescale;
cur.weighted_io_nsecs *= timescale;
cur.total_read_nsecs *= timescale;
cur.total_write_nsecs *= timescale;
}
if (region->bounds) {
/* Find first histogram separator. */
char *hist_str = strchr(row, ':');
if (!hist_str) {
log_error("Could not parse histogram value.");
goto bad;
}
/* Find space preceding histogram. */
while (hist_str && *(hist_str - 1) != ' ')
hist_str--;
/* Use a separate pool for histogram objects since we
* are growing the area table and each area's histogram
* table simultaneously.
*/
if (!_stats_parse_histogram(dms->hist_mem, hist_str,
&hist, region))
goto_bad;
hist->dms = dms;
hist->region = region;
}
cur.histogram = hist;
if (!dm_pool_grow_object(mem, &cur, sizeof(cur)))
goto_bad;
if (region->start == UINT64_MAX) {
region->start = start;
region->step = len; /* area size is always uniform. */
}
}
if (region->start == UINT64_MAX)
/* no area data read from @stats_print */
goto bad;
region->len = (start + len) - region->start;
region->timescale = timescale;
region->counters = dm_pool_end_object(mem);
if (fclose(stats_rows))
stack;
return 1;
bad:
if (stats_rows)
if (fclose(stats_rows))
stack;
dm_pool_abandon_object(mem);
return 0;
}
static void _stats_walk_next_present(const struct dm_stats *dms,
uint64_t *flags,
uint64_t *cur_r, uint64_t *cur_a,
uint64_t *cur_g)
{
struct dm_stats_region *cur = NULL;
/* start of walk: region loop advances *cur_r to 0. */
if (*cur_r != DM_STATS_REGION_NOT_PRESENT)
cur = &dms->regions[*cur_r];
/* within current region? */
if (cur && (*flags & DM_STATS_WALK_AREA)) {
if (++(*cur_a) < _nr_areas_region(cur))
return;
else
*cur_a = 0;
}
/* advance to next present, non-skipped region or end */
/* count can start as UINT64_MAX, probably rework to use post++ */
/* coverity[overflow_const] overflow is expected here */
while (++(*cur_r) <= dms->max_region) {
cur = &dms->regions[*cur_r];
if (!_stats_region_present(cur))
continue;
if ((*flags & DM_STATS_WALK_SKIP_SINGLE_AREA))
if (!(*flags & DM_STATS_WALK_AREA))
if (_nr_areas_region(cur) < 2)
continue;
/* matching region found */
break;
}
return;
}
static void _stats_walk_next(const struct dm_stats *dms, uint64_t *flags,
uint64_t *cur_r, uint64_t *cur_a, uint64_t *cur_g)
{
if (!dms || !dms->regions)
return;
if (*flags & DM_STATS_WALK_AREA) {
/* advance to next area, region, or end */
_stats_walk_next_present(dms, flags, cur_r, cur_a, cur_g);
return;
}
if (*flags & DM_STATS_WALK_REGION) {
/* enable region aggregation */
*cur_a = DM_STATS_WALK_REGION;
_stats_walk_next_present(dms, flags, cur_r, cur_a, cur_g);
return;
}
if (*flags & DM_STATS_WALK_GROUP) {
/* enable group aggregation */
*cur_r = *cur_a = DM_STATS_WALK_GROUP;
while (!_stats_group_id_present(dms, ++(*cur_g))
&& (*cur_g) < dms->max_region + 1)
; /* advance to next present group or end */
return;
}
log_error("stats_walk_next called with empty walk flags");
}
static void _group_walk_start(const struct dm_stats *dms, uint64_t *flags,
uint64_t *cur_r, uint64_t *cur_a, uint64_t *cur_g)
{
if (!(*flags & DM_STATS_WALK_GROUP))
return;
*cur_a = *cur_r = DM_STATS_WALK_GROUP;
*cur_g = 0;
/* advance to next present group or end */
while ((*cur_g) <= dms->max_region) {
if (_stats_region_is_grouped(dms, *cur_g))
break;
(*cur_g)++;
}
if (*cur_g > dms->max_region)
/* no groups to walk */
*flags &= ~DM_STATS_WALK_GROUP;
}
static void _stats_walk_start(const struct dm_stats *dms, uint64_t *flags,
uint64_t *cur_r, uint64_t *cur_a,
uint64_t *cur_g)
{
log_debug("starting stats walk with %s %s %s %s",
(*flags & DM_STATS_WALK_AREA) ? "AREA" : "",
(*flags & DM_STATS_WALK_REGION) ? "REGION" : "",
(*flags & DM_STATS_WALK_GROUP) ? "GROUP" : "",
(*flags & DM_STATS_WALK_SKIP_SINGLE_AREA) ? "SKIP" : "");
if (!dms->regions)
return;
if (!(*flags & (DM_STATS_WALK_AREA | DM_STATS_WALK_REGION))) {
_group_walk_start(dms, flags, cur_r, cur_a, cur_g);
return;
}
/* initialise cursor state */
*cur_a = 0;
*cur_r = DM_STATS_REGION_NOT_PRESENT;
*cur_g = DM_STATS_GROUP_NOT_PRESENT;
if (!(*flags & DM_STATS_WALK_AREA))
*cur_a = DM_STATS_WALK_REGION;
/* advance to first present, non-skipped region */
_stats_walk_next_present(dms, flags, cur_r, cur_a, cur_g);
}
#define DM_STATS_WALK_MASK (DM_STATS_WALK_AREA \
| DM_STATS_WALK_REGION \
| DM_STATS_WALK_GROUP \
| DM_STATS_WALK_SKIP_SINGLE_AREA)
int dm_stats_walk_init(struct dm_stats *dms, uint64_t flags)
{
if (!dms)
return_0;
if (flags & ~DM_STATS_WALK_MASK) {
log_error("Unknown value in walk flags: 0x" FMTx64,
(uint64_t) (flags & ~DM_STATS_WALK_MASK));
return 0;
}
dms->walk_flags = flags;
log_debug("dm_stats_walk_init: initialised flags to " FMTx64, flags);
return 1;
}
void dm_stats_walk_start(struct dm_stats *dms)
{
if (!dms || !dms->regions)
return;
dms->cur_flags = dms->walk_flags;
_stats_walk_start(dms, &dms->cur_flags,
&dms->cur_region, &dms->cur_area,
&dms->cur_group);
}
void dm_stats_walk_next(struct dm_stats *dms)
{
_stats_walk_next(dms, &dms->cur_flags,
&dms->cur_region, &dms->cur_area,
&dms->cur_group);
}
void dm_stats_walk_next_region(struct dm_stats *dms)
{
dms->cur_flags &= ~DM_STATS_WALK_AREA;
_stats_walk_next(dms, &dms->cur_flags,
&dms->cur_region, &dms->cur_area,
&dms->cur_group);
}
/*
* Return 1 if any regions remain that are present and not skipped
* by the current walk flags or 0 otherwise.
*/
static uint64_t _stats_walk_any_unskipped(const struct dm_stats *dms,
uint64_t *flags,
uint64_t *cur_r, uint64_t *cur_a)
{
struct dm_stats_region *region;
uint64_t i;
if (*cur_r > dms->max_region)
return 0;
for (i = *cur_r; i <= dms->max_region; i++) {
region = &dms->regions[i];
if (!_stats_region_present(region))
continue;
if ((*flags & DM_STATS_WALK_SKIP_SINGLE_AREA)
&& !(*flags & DM_STATS_WALK_AREA))
if (_nr_areas_region(region) < 2)
continue;
return 1;
}
return 0;
}
static void _stats_walk_end_areas(const struct dm_stats *dms, uint64_t *flags,
uint64_t *cur_r, uint64_t *cur_a,
uint64_t *cur_g)
{
int end = !_stats_walk_any_unskipped(dms, flags, cur_r, cur_a);
if (!(*flags & DM_STATS_WALK_AREA))
return;
if (!end)
return;
*flags &= ~DM_STATS_WALK_AREA;
if (*flags & DM_STATS_WALK_REGION) {
/* start region walk */
*cur_a = DM_STATS_WALK_REGION;
*cur_r = DM_STATS_REGION_NOT_PRESENT;
_stats_walk_next_present(dms, flags, cur_r, cur_a, cur_g);
if (!_stats_walk_any_unskipped(dms, flags, cur_r, cur_a)) {
/* no more regions */
*flags &= ~DM_STATS_WALK_REGION;
if (!(*flags & DM_STATS_WALK_GROUP))
*cur_r = dms->max_region;
}
}
if (*flags & DM_STATS_WALK_REGION)
return;
if (*flags & DM_STATS_WALK_GROUP)
_group_walk_start(dms, flags, cur_r, cur_a, cur_g);
}
static int _stats_walk_end(const struct dm_stats *dms, uint64_t *flags,
uint64_t *cur_r, uint64_t *cur_a, uint64_t *cur_g)
{
if (*flags & DM_STATS_WALK_AREA) {
_stats_walk_end_areas(dms, flags, cur_r, cur_a, cur_g);
goto out;
}
if (*flags & DM_STATS_WALK_REGION) {
if (!_stats_walk_any_unskipped(dms, flags, cur_r, cur_a)) {
*flags &= ~DM_STATS_WALK_REGION;
_group_walk_start(dms, flags, cur_r, cur_a, cur_g);
}
goto out;
}
if (*flags & DM_STATS_WALK_GROUP) {
if (*cur_g <= dms->max_region)
goto out;
*flags &= ~DM_STATS_WALK_GROUP;
}
out:
return !(*flags & ~DM_STATS_WALK_SKIP_SINGLE_AREA);
}
int dm_stats_walk_end(struct dm_stats *dms)
{
if (!dms)
return 1;
if (_stats_walk_end(dms, &dms->cur_flags,
&dms->cur_region, &dms->cur_area,
&dms->cur_group)) {
dms->cur_flags = dms->walk_flags;
return 1;
}
return 0;
}
dm_stats_obj_type_t dm_stats_object_type(const struct dm_stats *dms,
uint64_t region_id,
uint64_t area_id)
{
uint64_t group_id;
region_id = (region_id == DM_STATS_REGION_CURRENT)
? dms->cur_region : region_id ;
area_id = (area_id == DM_STATS_AREA_CURRENT)
? dms->cur_area : area_id ;
if (region_id == DM_STATS_REGION_NOT_PRESENT)
/* no region */
return DM_STATS_OBJECT_TYPE_NONE;
if (region_id & DM_STATS_WALK_GROUP) {
if (region_id == DM_STATS_WALK_GROUP)
/* indirect group_id from cursor */
group_id = dms->cur_group;
else
/* immediate group_id encoded in region_id */
group_id = region_id & ~DM_STATS_WALK_GROUP;
if (!_stats_group_id_present(dms, group_id))
return DM_STATS_OBJECT_TYPE_NONE;
return DM_STATS_OBJECT_TYPE_GROUP;
}
if (region_id > dms->max_region)
/* end of table */
return DM_STATS_OBJECT_TYPE_NONE;
if (area_id & DM_STATS_WALK_REGION)
/* aggregate region */
return DM_STATS_OBJECT_TYPE_REGION;
/* plain region_id and area_id */
return DM_STATS_OBJECT_TYPE_AREA;
}
dm_stats_obj_type_t dm_stats_current_object_type(const struct dm_stats *dms)
{
/* dm_stats_object_type will decode region/area */
return dm_stats_object_type(dms,
DM_STATS_REGION_CURRENT,
DM_STATS_AREA_CURRENT);
}
uint64_t dm_stats_get_region_nr_areas(const struct dm_stats *dms,
uint64_t region_id)
{
struct dm_stats_region *region = NULL;
/* groups or aggregate regions cannot be subdivided */
if (region_id & DM_STATS_WALK_GROUP)
return 1;
region = &dms->regions[region_id];
return _nr_areas_region(region);
}
uint64_t dm_stats_get_current_nr_areas(const struct dm_stats *dms)
{
/* groups or aggregate regions cannot be subdivided */
if (dms->cur_region & DM_STATS_WALK_GROUP)
return 1;
return dm_stats_get_region_nr_areas(dms, dms->cur_region);
}
uint64_t dm_stats_get_nr_areas(const struct dm_stats *dms)
{
uint64_t nr_areas = 0, flags = DM_STATS_WALK_AREA;
/* use a separate cursor */
uint64_t cur_region = 0, cur_area = 0, cur_group = 0;
/* no regions to visit? */
if (!dms->regions)
return 0;
flags = DM_STATS_WALK_AREA;
_stats_walk_start(dms, &flags, &cur_region, &cur_area, &cur_group);
do {
nr_areas += dm_stats_get_current_nr_areas(dms);
_stats_walk_next(dms, &flags,
&cur_region, &cur_area,
&cur_group);
} while (!_stats_walk_end(dms, &flags,
&cur_region, &cur_area,
&cur_group));
return nr_areas;
}
int dm_stats_group_present(const struct dm_stats *dms, uint64_t group_id)
{
return _stats_group_id_present(dms, group_id);
}
int dm_stats_get_region_nr_histogram_bins(const struct dm_stats *dms,
uint64_t region_id)
{
region_id = (region_id == DM_STATS_REGION_CURRENT)
? dms->cur_region : region_id ;
/* FIXME: support group histograms if all region bounds match */
if (region_id & DM_STATS_WALK_GROUP)
return 0;
if (!dms->regions[region_id].bounds)
return 0;
return dms->regions[region_id].bounds->nr_bins;
}
/*
* Fill buf with a list of set regions in the regions bitmap. Consecutive
* ranges of set region IDs are output using "M-N" range notation.
*
* The number of bytes consumed is returned or zero on error.
*/
static size_t _stats_group_tag_fill(const struct dm_stats *dms,
dm_bitset_t regions,
char *buf, size_t buflen)
{
int i, j, r, next, last = 0;
size_t used = 0;
last = dm_bit_get_last(regions);
i = dm_bit_get_first(regions);
for(; i >= 0; i = dm_bit_get_next(regions, i)) {
/* find range end */
j = i;
do
next = j + 1;
while ((j = dm_bit_get_next(regions, j)) == next);
/* set to last set bit */
j = next - 1;
/* handle range vs. single region */
if (i != j)
r = dm_snprintf(buf, buflen, FMTu64 "-" FMTu64 "%s",
(uint64_t) i, (uint64_t) j,
(j == last) ? "" : ",");
else
r = dm_snprintf(buf, buflen, FMTu64 "%s", (uint64_t) i,
(i == last) ? "" : ",");
if (r < 0)
goto_bad;
i = next; /* skip handled bits if in range */
buf += r;
used += r;
}
return used;
bad:
log_error("Could not format group list.");
return 0;
}
/*
* Calculate the space required to hold a string description of the group
* described by the regions bitset using comma separated list in range
* notation ("A,B,C,M-N").
*/
static size_t _stats_group_tag_len(const struct dm_stats *dms,
dm_bitset_t regions)
{
int64_t i, j, next, nr_regions = 0;
size_t buflen = 0, id_len = 0;
/* check region ids and find last set bit */
i = dm_bit_get_first(regions);
for (; i >= 0; i = dm_bit_get_next(regions, i)) {
/* length of region_id or range start in characters */
id_len = (i) ? 1 + (size_t) log10(i) : 1;
buflen += id_len;
j = i;
do
next = j + 1;
while ((j = dm_bit_get_next(regions, j)) == next);
/* set to last set bit */
j = next - 1;
nr_regions += j - i + 1;
/* handle range */
if (i != j) {
/* j is always > i, which is always >= 0 */
id_len = 1 + (size_t) log10(j);
buflen += id_len + 1; /* range end plus "-" */
}
buflen++;
i = next; /* skip bits if handling range */
}
return buflen;
}
/*
* Build a DMS_GROUP="..." tag for the group specified by group_id,
* to be stored in the corresponding region's aux_data field.
*/
static char *_build_group_tag(struct dm_stats *dms, uint64_t group_id)
{
char *aux_string, *buf;
dm_bitset_t regions;
const char *alias;
size_t buflen = 0;
int r;
regions = dms->groups[group_id].regions;
alias = dms->groups[group_id].alias;
buflen = _stats_group_tag_len(dms, regions);
if (!buflen)
return_0;
buflen += DMS_GROUP_TAG_LEN;
buflen += 1 + (alias ? strlen(alias) + 2 : 0); /* 'alias:' */
buf = aux_string = dm_malloc(buflen);
if (!buf) {
log_error("Could not allocate memory for aux_data string.");
return NULL;
}
if (!_dm_strncpy(buf, DMS_GROUP_TAG, DMS_GROUP_TAG_LEN + 1))
goto_bad;
buf += DMS_GROUP_TAG_LEN;
buflen -= DMS_GROUP_TAG_LEN;
if (alias)
r = dm_snprintf(buf, buflen, "\"%s\"%c", alias, DMS_GROUP_SEP);
else
r = dm_snprintf(buf, buflen, "%c", DMS_GROUP_SEP);
if (r < 0)
goto_bad;
buf += r;
buflen -= r;
r = _stats_group_tag_fill(dms, regions, buf, buflen);
if (!r)
goto_bad;
return aux_string;
bad:
log_error("Could not format group aux_data.");
dm_free(aux_string);
return NULL;
}
/*
* Store updated aux_data for a region. The aux_data is passed to the
* kernel using the @stats_set_aux message. Any required group tag is
* generated from the current group table and included in the message.
*/
static int _stats_set_aux(struct dm_stats *dms,
uint64_t region_id, const char *user_data)
{
char *group_tag = NULL, *group_tag_escaped = NULL;
struct dm_task *dmt = NULL;
char msg[STATS_MSG_BUF_LEN];
int r = 0;
/* group data required? */
if (_stats_group_id_present(dms, region_id)) {
group_tag = _build_group_tag(dms, region_id);
if (!group_tag) {
log_error("Could not build group descriptor for "
"region ID " FMTu64, region_id);
goto bad;
}
group_tag_escaped = _stats_escape_aux_data(group_tag);
if (!group_tag_escaped)
goto bad;
}
if (dm_snprintf(msg, sizeof(msg), "@stats_set_aux " FMTu64 " %s%s%s ",
region_id, (group_tag_escaped) ? group_tag_escaped : "",
(group_tag_escaped) ? DMS_AUX_SEP : "",
(strlen(user_data)) ? user_data : "-") < 0) {
log_error("Could not prepare @stats_set_aux message");
goto bad;
}
if (!(dmt = _stats_send_message(dms, msg)))
goto_bad;
/* no response to a @stats_set_aux message */
dm_task_destroy(dmt);
r = 1;
bad:
dm_free(group_tag_escaped);
dm_free(group_tag);
return r;
}
/*
* Maximum length of a "start+end" range string:
* Two 20 digit uint64_t, '+', and NULL.
*/
#define RANGE_LEN 42
static int _stats_create_region(struct dm_stats *dms, uint64_t *region_id,
uint64_t start, uint64_t len, int64_t step,
int precise, const char *hist_arg,
const char *program_id, const char *aux_data)
{
char msg[STATS_MSG_BUF_LEN], range[RANGE_LEN], *endptr = NULL;
const char *err = NULL;
const char *precise_str = PRECISE_ARG;
const char *resp, *opt_args = NULL;
char *aux_data_escaped = NULL;
struct dm_task *dmt = NULL;
int r = 0, nr_opt = 0;
if (!_stats_bound(dms))
return_0;
if (!program_id || !strlen(program_id))
program_id = dms->program_id;
if (start || len) {
if (dm_snprintf(range, sizeof(range), FMTu64 "+" FMTu64,
start, len) < 0) {
err ="range";
goto_bad;
}
}
if (precise < 0)
precise = dms->precise;
if (precise)
nr_opt++;
else
precise_str = "";
if (hist_arg)
nr_opt++;
else
hist_arg = "";
aux_data_escaped = _stats_escape_aux_data(aux_data);
if (!aux_data_escaped)
return_0;
if (nr_opt) {
if ((dm_asprintf((char **)&opt_args, "%d %s %s%s", nr_opt,
precise_str,
(strlen(hist_arg)) ? HISTOGRAM_ARG : "",
hist_arg)) < 0) {
err = PRECISE_ARG " option.";
goto_bad;
}
} else
opt_args = dm_strdup("");
if (dm_snprintf(msg, sizeof(msg), "@stats_create %s %s" FMTu64
" %s %s %s", (start || len) ? range : "-",
(step < 0) ? "/" : "",
(uint64_t)llabs(step),
opt_args, program_id, aux_data) < 0) {
err = "message";
goto_bad;
}
if (!(dmt = _stats_send_message(dms, msg)))
goto_out;
resp = dm_task_get_message_response(dmt);
if (!resp) {
log_error("Could not parse empty @stats_create response.");
goto out;
}
if (region_id) {
errno = 0;
*region_id = strtoull(resp, &endptr, 10);
if (errno || resp == endptr)
goto_out;
}
r = 1;
goto out;
bad:
log_error("Could not prepare @stats_create %s.", err);
out:
if (dmt)
dm_task_destroy(dmt);
dm_free((void *) opt_args);
dm_free(aux_data_escaped);
return r;
}
DM_EXPORT_NEW_SYMBOL(int, dm_stats_create_region, 1_02_107)
(struct dm_stats *dms, uint64_t *region_id,
uint64_t start, uint64_t len, int64_t step,
int precise, struct dm_histogram *bounds,
const char *program_id, const char *user_data)
{
char *hist_arg = NULL;
int r = 0;
/* Nanosecond counters and histograms both need precise_timestamps. */
if ((precise || bounds) && !_stats_check_precise_timestamps(dms))
return_0;
if (bounds) {
/* _build_histogram_arg enables precise if vals < 1ms. */
if (!(hist_arg = _build_histogram_arg(bounds, &precise)))
goto_out;
}
r = _stats_create_region(dms, region_id, start, len, step,
precise, hist_arg, program_id, user_data);
dm_free(hist_arg);
out:
return r;
}
static void _stats_clear_group_regions(struct dm_stats *dms, uint64_t group_id)
{
struct dm_stats_group *group;
uint64_t i;
group = &dms->groups[group_id];
for (i = dm_bit_get_first(group->regions);
i != DM_STATS_GROUP_NOT_PRESENT;
i = dm_bit_get_next(group->regions, i))
dms->regions[i].group_id = DM_STATS_GROUP_NOT_PRESENT;
}
static int _stats_remove_region_id_from_group(struct dm_stats *dms,
uint64_t region_id)
{
struct dm_stats_region *region = &dms->regions[region_id];
uint64_t group_id = region->group_id;
dm_bitset_t regions = dms->groups[group_id].regions;
if (!_stats_region_is_grouped(dms, region_id))
return_0;
dm_bit_clear(regions, region_id);
/* removing group leader? */
if (region_id == group_id) {
_stats_clear_group_regions(dms, group_id);
_stats_group_destroy(&dms->groups[group_id]);
}
return _stats_set_aux(dms, group_id, dms->regions[group_id].aux_data);
}
static int _stats_delete_region(struct dm_stats *dms, uint64_t region_id)
{
char msg[STATS_MSG_BUF_LEN];
struct dm_task *dmt;
if (_stats_region_is_grouped(dms, region_id))
if (!_stats_remove_region_id_from_group(dms, region_id)) {
log_error("Could not remove region ID " FMTu64 " from "
"group ID " FMTu64,
region_id, dms->regions[region_id].group_id);
return 0;
}
if (dm_snprintf(msg, sizeof(msg), "@stats_delete " FMTu64, region_id) < 0) {
log_error("Could not prepare @stats_delete message.");
return 0;
}
dmt = _stats_send_message(dms, msg);
if (!dmt)
return_0;
dm_task_destroy(dmt);
return 1;
}
int dm_stats_delete_region(struct dm_stats *dms, uint64_t region_id)
{
int listed = 0;
if (!_stats_bound(dms))
return_0;
/*
* To correctly delete a region, that may be part of a group, a
* listed handle is required, since the region may need to be
* removed from another region's group descriptor; earlier
* versions of the region deletion interface do not have this
* requirement since there are no dependencies between regions.
*
* Listing a previously unlisted handle has numerous
* side-effects on other calls and operations (e.g. stats
* walks), especially when returning to a function that depends
* on the state of the region table, or statistics cursor.
*
* To avoid changing the semantics of the API, and the need for
* a versioned symbol, maintain a flag indicating when a listing
* has been carried out, and drop the region table before
* returning.
*
* This ensures compatibility with programs compiled against
* earlier versions of libdm.
*/
if (!dms->regions && !(listed = dm_stats_list(dms, dms->program_id))) {
log_error("Could not obtain region list while deleting "
"region ID " FMTu64, region_id);
goto bad;
}
if (!dm_stats_get_nr_regions(dms)) {
log_error("Could not delete region ID " FMTu64 ": "
"no regions found", region_id);
goto bad;
}
/* includes invalid and special region_id values */
if (!dm_stats_region_present(dms, region_id)) {
log_error("Region ID " FMTu64 " does not exist", region_id);
goto bad;
}
if (!_stats_delete_region(dms, region_id))
goto bad;
if (!listed)
/* wipe region and mark as not present */
_stats_region_destroy(&dms->regions[region_id]);
else
/* return handle to prior state */
_stats_regions_destroy(dms);
return 1;
bad:
if (listed)
_stats_regions_destroy(dms);
return 0;
}
int dm_stats_clear_region(struct dm_stats *dms, uint64_t region_id)
{
char msg[STATS_MSG_BUF_LEN];
struct dm_task *dmt;
if (!_stats_bound(dms))
return_0;
if (dm_snprintf(msg, sizeof(msg), "@stats_clear " FMTu64, region_id) < 0) {
log_error("Could not prepare @stats_clear message.");
return 0;
}
dmt = _stats_send_message(dms, msg);
if (!dmt)
return_0;
dm_task_destroy(dmt);
return 1;
}
static struct dm_task *_stats_print_region(struct dm_stats *dms,
uint64_t region_id, unsigned start_line,
unsigned num_lines, unsigned clear)
{
/* @stats_print[_clear] <region_id> [<start_line> <num_lines>] */
char msg[STATS_MSG_BUF_LEN], lines[RANGE_LEN];
struct dm_task *dmt = NULL;
const char *err = NULL;
if (start_line || num_lines)
if (dm_snprintf(lines, sizeof(lines),
"%u %u", start_line, num_lines) < 0) {
err = "row specification";
goto_bad;
}
if (dm_snprintf(msg, sizeof(msg), "@stats_print%s " FMTu64 " %s",
(clear) ? "_clear" : "",
region_id, (start_line || num_lines) ? lines : "") < 0) {
err = "message";
goto_bad;
}
if (!(dmt = _stats_send_message(dms, msg)))
return_NULL;
return dmt;
bad:
log_error("Could not prepare @stats_print %s.", err);
return NULL;
}
char *dm_stats_print_region(struct dm_stats *dms, uint64_t region_id,
unsigned start_line, unsigned num_lines,
unsigned clear)
{
char *resp = NULL;
struct dm_task *dmt = NULL;
const char *response;
if (!_stats_bound(dms))
return_0;
/*
* FIXME: 'print' can be emulated for groups or aggregate regions
* by populating the handle and emitting aggregate counter data
* in the kernel print format.
*/
if (region_id == DM_STATS_WALK_GROUP)
return_0;
dmt = _stats_print_region(dms, region_id,
start_line, num_lines, clear);
if (!dmt)
return_0;
if (!(response = dm_task_get_message_response(dmt)))
goto_out;
if (!(resp = dm_pool_strdup(dms->mem, response)))
log_error("Could not allocate memory for response buffer.");
out:
dm_task_destroy(dmt);
return resp;
}
void dm_stats_buffer_destroy(struct dm_stats *dms, char *buffer)
{
dm_pool_free(dms->mem, buffer);
}
uint64_t dm_stats_get_nr_regions(const struct dm_stats *dms)
{
if (!dms)
return_0;
if (!dms->regions)
return 0;
return dms->nr_regions;
}
uint64_t dm_stats_get_nr_groups(const struct dm_stats *dms)
{
uint64_t group_id, nr_groups = 0;
if (!dms)
return_0;
/* no regions or groups? */
if (!dms->regions || !dms->groups)
return 0;
for (group_id = 0; group_id <= dms->max_region; group_id++)
if (dms->groups[group_id].group_id
!= DM_STATS_GROUP_NOT_PRESENT)
nr_groups++;
return nr_groups;
}
/**
* Test whether region_id is present in this set of stats data.
*/
int dm_stats_region_present(const struct dm_stats *dms, uint64_t region_id)
{
if (!dms->regions)
return_0;
if (region_id > dms->max_region)
return 0;
return _stats_region_present(&dms->regions[region_id]);
}
static int _dm_stats_populate_region(struct dm_stats *dms, uint64_t region_id,
const char *resp)
{
struct dm_stats_region *region = &dms->regions[region_id];
if (!_stats_bound(dms))
return_0;
if (!region) {
log_error("Cannot populate empty handle before dm_stats_list().");
return 0;
}
if (!_stats_parse_region(dms, resp, region, region->timescale)) {
log_error("Could not parse @stats_print message response.");
return 0;
}
region->region_id = region_id;
return 1;
}
int dm_stats_populate(struct dm_stats *dms, const char *program_id,
uint64_t region_id)
{
int all_regions = (region_id == DM_STATS_REGIONS_ALL);
struct dm_task *dmt = NULL; /* @stats_print task */
uint64_t saved_flags; /* saved walk flags */
const char *resp;
/*
* We are about do destroy and re-create the region table, so it
* is safe to use the cursor embedded in the stats handle: just
* save a copy of the current walk_flags to restore later.
*/
saved_flags = dms->walk_flags;
if (!_stats_bound(dms))
return_0;
if ((!all_regions) && (region_id & DM_STATS_WALK_GROUP)) {
log_error("Invalid region_id for dm_stats_populate: "
"DM_STATS_WALK_GROUP");
return 0;
}
/* allow zero-length program_id for populate */
if (!program_id)
program_id = dms->program_id;
if (all_regions && !dm_stats_list(dms, program_id)) {
log_error("Could not parse @stats_list response.");
goto bad;
} else if (!_stats_set_name_cache(dms)) {
goto_bad;
}
if (!dms->nr_regions) {
log_verbose("No stats regions registered: %s", dms->name);
return 0;
}
dms->walk_flags = DM_STATS_WALK_REGION;
dm_stats_walk_start(dms);
do {
region_id = (all_regions)
? dm_stats_get_current_region(dms) : region_id;
/* obtain all lines and clear counter values */
if (!(dmt = _stats_print_region(dms, region_id, 0, 0, 1)))
goto_bad;
resp = dm_task_get_message_response(dmt);
if (!_dm_stats_populate_region(dms, region_id, resp)) {
dm_task_destroy(dmt);
goto_bad;
}
dm_task_destroy(dmt);
dm_stats_walk_next(dms);
} while (all_regions && !dm_stats_walk_end(dms));
dms->walk_flags = saved_flags;
return 1;
bad:
dms->walk_flags = saved_flags;
_stats_regions_destroy(dms);
dms->regions = NULL;
return 0;
}
/**
* destroy a dm_stats object and all associated regions and counter sets.
*/
void dm_stats_destroy(struct dm_stats *dms)
{
if (!dms)
return;
_stats_regions_destroy(dms);
_stats_groups_destroy(dms);
_stats_clear_binding(dms);
dm_pool_destroy(dms->mem);
dm_pool_destroy(dms->hist_mem);
dm_pool_destroy(dms->group_mem);
dm_free(dms->program_id);
dm_free((char *) dms->name);
dm_free(dms);
}
/*
* Walk each area that is a member of region_id rid.
* i is a variable of type int that holds the current area_id.
*/
#define _foreach_region_area(dms, rid, i) \
for ((i) = 0; (i) < _nr_areas_region(&dms->regions[(rid)]); (i)++) \
/*
* Walk each region that is a member of group_id gid.
* i is a variable of type int that holds the current region_id.
*/
#define _foreach_group_region(dms, gid, i) \
for ((i) = dm_bit_get_first((dms)->groups[(gid)].regions); \
(i) != DM_STATS_GROUP_NOT_PRESENT; \
(i) = dm_bit_get_next((dms)->groups[(gid)].regions, (i))) \
/*
* Walk each region that is a member of group_id gid visiting each
* area within the region.
* i is a variable of type int that holds the current region_id.
* j is a variable of type int variable that holds the current area_id.
*/
#define _foreach_group_area(dms, gid, i, j) \
_foreach_group_region(dms, gid, i) \
_foreach_region_area(dms, i, j)
static uint64_t _stats_get_counter(const struct dm_stats *dms,
const struct dm_stats_counters *area,
dm_stats_counter_t counter)
{
switch(counter) {
case DM_STATS_READS_COUNT:
return area->reads;
case DM_STATS_READS_MERGED_COUNT:
return area->reads_merged;
case DM_STATS_READ_SECTORS_COUNT:
return area->read_sectors;
case DM_STATS_READ_NSECS:
return area->read_nsecs;
case DM_STATS_WRITES_COUNT:
return area->writes;
case DM_STATS_WRITES_MERGED_COUNT:
return area->writes_merged;
case DM_STATS_WRITE_SECTORS_COUNT:
return area->write_sectors;
case DM_STATS_WRITE_NSECS:
return area->write_nsecs;
case DM_STATS_IO_IN_PROGRESS_COUNT:
return area->io_in_progress;
case DM_STATS_IO_NSECS:
return area->io_nsecs;
case DM_STATS_WEIGHTED_IO_NSECS:
return area->weighted_io_nsecs;
case DM_STATS_TOTAL_READ_NSECS:
return area->total_read_nsecs;
case DM_STATS_TOTAL_WRITE_NSECS:
return area->total_write_nsecs;
case DM_STATS_NR_COUNTERS:
default:
log_error("Attempt to read invalid counter: %d", counter);
}
return 0;
}
uint64_t dm_stats_get_counter(const struct dm_stats *dms,
dm_stats_counter_t counter,
uint64_t region_id, uint64_t area_id)
{
uint64_t i, j, sum = 0; /* aggregation */
int sum_regions = 0;
struct dm_stats_region *region;
struct dm_stats_counters *area;
region_id = (region_id == DM_STATS_REGION_CURRENT)
? dms->cur_region : region_id ;
area_id = (area_id == DM_STATS_REGION_CURRENT)
? dms->cur_area : area_id ;
sum_regions = !!(region_id & DM_STATS_WALK_GROUP);
if (region_id == DM_STATS_WALK_GROUP)
/* group walk using the cursor */
region_id = dms->cur_group;
else if (region_id & DM_STATS_WALK_GROUP)
/* group walk using immediate group_id */
region_id &= ~DM_STATS_WALK_GROUP;
region = &dms->regions[region_id];
/*
* All statistics aggregation takes place here: aggregate metrics
* are calculated as normal using the aggregated counter values
* returned for the region or group specified.
*/
if (_stats_region_is_grouped(dms, region_id) && (sum_regions)) {
/* group */
if (area_id & DM_STATS_WALK_GROUP)
_foreach_group_area(dms, region->group_id, i, j) {
area = &dms->regions[i].counters[j];
sum += _stats_get_counter(dms, area, counter);
}
else
_foreach_group_region(dms, region->group_id, i) {
area = &dms->regions[i].counters[area_id];
sum += _stats_get_counter(dms, area, counter);
}
} else if (area_id == DM_STATS_WALK_REGION) {
/* aggregate region */
_foreach_region_area(dms, region_id, j) {
area = &dms->regions[region_id].counters[j];
sum += _stats_get_counter(dms, area, counter);
}
} else {
/* plain region / area */
area = &region->counters[area_id];
sum = _stats_get_counter(dms, area, counter);
}
return sum;
}
/*
* Methods for accessing named counter fields. All methods share the
* following naming scheme and prototype:
*
* uint64_t dm_stats_get_COUNTER(const struct dm_stats *, uint64_t, uint64_t)
*
* Where the two integer arguments are the region_id and area_id
* respectively.
*
* name is the name of the counter (lower case)
* counter is the part of the enum name following DM_STATS_ (upper case)
*/
#define MK_STATS_GET_COUNTER_FN(name, counter) \
uint64_t dm_stats_get_ ## name(const struct dm_stats *dms, \
uint64_t region_id, uint64_t area_id) \
{ \
return dm_stats_get_counter(dms, DM_STATS_ ## counter, \
region_id, area_id); \
}
MK_STATS_GET_COUNTER_FN(reads, READS_COUNT)
MK_STATS_GET_COUNTER_FN(reads_merged, READS_MERGED_COUNT)
MK_STATS_GET_COUNTER_FN(read_sectors, READ_SECTORS_COUNT)
MK_STATS_GET_COUNTER_FN(read_nsecs, READ_NSECS)
MK_STATS_GET_COUNTER_FN(writes, WRITES_COUNT)
MK_STATS_GET_COUNTER_FN(writes_merged, WRITES_MERGED_COUNT)
MK_STATS_GET_COUNTER_FN(write_sectors, WRITE_SECTORS_COUNT)
MK_STATS_GET_COUNTER_FN(write_nsecs, WRITE_NSECS)
MK_STATS_GET_COUNTER_FN(io_in_progress, IO_IN_PROGRESS_COUNT)
MK_STATS_GET_COUNTER_FN(io_nsecs, IO_NSECS)
MK_STATS_GET_COUNTER_FN(weighted_io_nsecs, WEIGHTED_IO_NSECS)
MK_STATS_GET_COUNTER_FN(total_read_nsecs, TOTAL_READ_NSECS)
MK_STATS_GET_COUNTER_FN(total_write_nsecs, TOTAL_WRITE_NSECS)
#undef MK_STATS_GET_COUNTER_FN
/*
* Floating point stats metric functions
*
* Called from dm_stats_get_metric() to calculate the value of
* the requested metric.
*
* int _metric_name(const struct dm_stats *dms,
* struct dm_stats_counters *c,
* double *value);
*
* Calculate a metric value from the counter data for the given
* identifiers and store it in the memory pointed to by value,
* applying group or region aggregation if enabled.
*
* Return one on success or zero on failure.
*
* To add a new metric:
*
* o Add a new name to the dm_stats_metric_t enum.
* o Create a _metric_fn() to calculate the new metric.
* o Add _metric_fn to the _metrics function table
* (entries in enum order).
* o Do not add a new named public function for the metric -
* users of new metrics are encouraged to convert to the enum
* based metric interface.
*
*/
static int _rd_merges_per_sec(const struct dm_stats *dms, double *rrqm,
uint64_t region_id, uint64_t area_id)
{
double mrgs;
mrgs = (double) dm_stats_get_counter(dms, DM_STATS_READS_MERGED_COUNT,
region_id, area_id);
*rrqm = mrgs / (double) dms->interval_ns;
return 1;
}
static int _wr_merges_per_sec(const struct dm_stats *dms, double *wrqm,
uint64_t region_id, uint64_t area_id)
{
double mrgs;
mrgs = (double) dm_stats_get_counter(dms, DM_STATS_WRITES_MERGED_COUNT,
region_id, area_id);
*wrqm = mrgs / (double) dms->interval_ns;
return 1;
}
static int _reads_per_sec(const struct dm_stats *dms, double *rd_s,
uint64_t region_id, uint64_t area_id)
{
double reads;
reads = (double) dm_stats_get_counter(dms, DM_STATS_READS_COUNT,
region_id, area_id);
*rd_s = (reads * NSEC_PER_SEC) / (double) dms->interval_ns;
return 1;
}
static int _writes_per_sec(const struct dm_stats *dms, double *wr_s,
uint64_t region_id, uint64_t area_id)
{
double writes;
writes = (double) dm_stats_get_counter(dms, DM_STATS_WRITES_COUNT,
region_id, area_id);
*wr_s = (writes * NSEC_PER_SEC) / (double) dms->interval_ns;
return 1;
}
static int _read_sectors_per_sec(const struct dm_stats *dms, double *rsec_s,
uint64_t region_id, uint64_t area_id)
{
double sect;
sect = (double) dm_stats_get_counter(dms, DM_STATS_READ_SECTORS_COUNT,
region_id, area_id);
*rsec_s = (sect * (double) NSEC_PER_SEC) / (double) dms->interval_ns;
return 1;
}
static int _write_sectors_per_sec(const struct dm_stats *dms, double *wsec_s,
uint64_t region_id, uint64_t area_id)
{
double sect;
sect = (double) dm_stats_get_counter(dms, DM_STATS_WRITE_SECTORS_COUNT,
region_id, area_id);
*wsec_s = (sect * (double) NSEC_PER_SEC) / (double) dms->interval_ns;
return 1;
}
static int _average_request_size(const struct dm_stats *dms, double *arqsz,
uint64_t region_id, uint64_t area_id)
{
double ios, sectors;
ios = (double) (dm_stats_get_counter(dms, DM_STATS_READS_COUNT,
region_id, area_id)
+ dm_stats_get_counter(dms, DM_STATS_WRITES_COUNT,
region_id, area_id));
sectors = (double) (dm_stats_get_counter(dms, DM_STATS_READ_SECTORS_COUNT,
region_id, area_id)
+ dm_stats_get_counter(dms, DM_STATS_WRITE_SECTORS_COUNT,
region_id, area_id));
if (ios > 0.0)
*arqsz = sectors / ios;
else
*arqsz = 0.0;
return 1;
}
static int _average_queue_size(const struct dm_stats *dms, double *qusz,
uint64_t region_id, uint64_t area_id)
{
double io_ticks;
io_ticks = (double) dm_stats_get_counter(dms, DM_STATS_WEIGHTED_IO_NSECS,
region_id, area_id);
if (io_ticks > 0.0)
*qusz = io_ticks / (double) dms->interval_ns;
else
*qusz = 0.0;
return 1;
}
static int _average_wait_time(const struct dm_stats *dms, double *await,
uint64_t region_id, uint64_t area_id)
{
uint64_t io_ticks, nr_ios;
io_ticks = dm_stats_get_counter(dms, DM_STATS_READ_NSECS,
region_id, area_id);
io_ticks += dm_stats_get_counter(dms, DM_STATS_WRITE_NSECS,
region_id, area_id);
nr_ios = dm_stats_get_counter(dms, DM_STATS_READS_COUNT,
region_id, area_id);
nr_ios += dm_stats_get_counter(dms, DM_STATS_WRITES_COUNT,
region_id, area_id);
if (nr_ios > 0)
*await = (double) io_ticks / (double) nr_ios;
else
*await = 0.0;
return 1;
}
static int _average_rd_wait_time(const struct dm_stats *dms, double *await,
uint64_t region_id, uint64_t area_id)
{
uint64_t rd_io_ticks, nr_rd_ios;
rd_io_ticks = dm_stats_get_counter(dms, DM_STATS_READ_NSECS,
region_id, area_id);
nr_rd_ios = dm_stats_get_counter(dms, DM_STATS_READS_COUNT,
region_id, area_id);
/*
* If rd_io_ticks is > 0 this should imply that nr_rd_ios is
* also > 0 (unless a kernel bug exists). Test for both here
* before using the IO count as a divisor (Coverity).
*/
if (rd_io_ticks > 0 && nr_rd_ios > 0)
*await = (double) rd_io_ticks / (double) nr_rd_ios;
else
*await = 0.0;
return 1;
}
static int _average_wr_wait_time(const struct dm_stats *dms, double *await,
uint64_t region_id, uint64_t area_id)
{
uint64_t wr_io_ticks, nr_wr_ios;
wr_io_ticks = dm_stats_get_counter(dms, DM_STATS_WRITE_NSECS,
region_id, area_id);
nr_wr_ios = dm_stats_get_counter(dms, DM_STATS_WRITES_COUNT,
region_id, area_id);
/*
* If wr_io_ticks is > 0 this should imply that nr_wr_ios is
* also > 0 (unless a kernel bug exists). Test for both here
* before using the IO count as a divisor (Coverity).
*/
if (wr_io_ticks > 0 && nr_wr_ios > 0)
*await = (double) wr_io_ticks / (double) nr_wr_ios;
else
*await = 0.0;
return 1;
}
static int _throughput(const struct dm_stats *dms, double *tput,
uint64_t region_id, uint64_t area_id)
{
uint64_t nr_ios;
nr_ios = dm_stats_get_counter(dms, DM_STATS_READS_COUNT,
region_id, area_id);
nr_ios += dm_stats_get_counter(dms, DM_STATS_WRITES_COUNT,
region_id, area_id);
*tput = ((double) NSEC_PER_SEC * (double) nr_ios)
/ (double) (dms->interval_ns);
return 1;
}
static int _utilization(const struct dm_stats *dms, double *util,
uint64_t region_id, uint64_t area_id)
{
uint64_t io_nsecs, interval_ns = dms->interval_ns;
/**