15 static int last_majdev, last_mindev;
16 static struct disk_util *last_du;
18 static struct fio_mutex *disk_util_mutex;
20 FLIST_HEAD(disk_list);
22 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
23 int majdev, int mindev, char *path);
25 static void disk_util_free(struct disk_util *du)
30 while (!flist_empty(&du->slaves)) {
31 struct disk_util *slave;
33 slave = flist_first_entry(&du->slaves, struct disk_util, slavelist);
34 flist_del(&slave->slavelist);
38 fio_mutex_remove(du->lock);
42 static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
45 unsigned long long sectors[2];
51 dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
53 f = fopen(du->path, "r");
57 p = fgets(line, sizeof(line), f);
63 dprint(FD_DISKUTIL, "%s: %s", du->path, p);
65 ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n",
67 &dus->s.merges[0], §ors[0],
68 &dus->s.ticks[0], &dus->s.ios[1],
69 &dus->s.merges[1], §ors[1],
70 &dus->s.ticks[1], &in_flight,
72 &dus->s.time_in_queue);
74 dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
75 dus->s.sectors[0] = sectors[0];
76 dus->s.sectors[1] = sectors[1];
80 static void update_io_tick_disk(struct disk_util *du)
82 struct disk_util_stat __dus, *dus, *ldus;
87 if (get_io_ticks(du, &__dus))
93 dus->s.sectors[0] += (__dus.s.sectors[0] - ldus->s.sectors[0]);
94 dus->s.sectors[1] += (__dus.s.sectors[1] - ldus->s.sectors[1]);
95 dus->s.ios[0] += (__dus.s.ios[0] - ldus->s.ios[0]);
96 dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]);
97 dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]);
98 dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]);
99 dus->s.ticks[0] += (__dus.s.ticks[0] - ldus->s.ticks[0]);
100 dus->s.ticks[1] += (__dus.s.ticks[1] - ldus->s.ticks[1]);
101 dus->s.io_ticks += (__dus.s.io_ticks - ldus->s.io_ticks);
102 dus->s.time_in_queue += (__dus.s.time_in_queue - ldus->s.time_in_queue);
104 fio_gettime(&t, NULL);
105 dus->s.msec += mtime_since(&du->time, &t);
106 memcpy(&du->time, &t, sizeof(t));
107 memcpy(&ldus->s, &__dus.s, sizeof(__dus.s));
110 int update_io_ticks(void)
112 struct flist_head *entry;
113 struct disk_util *du;
116 dprint(FD_DISKUTIL, "update io ticks\n");
118 fio_mutex_down(disk_util_mutex);
120 if (!disk_util_exit) {
121 flist_for_each(entry, &disk_list) {
122 du = flist_entry(entry, struct disk_util, list);
123 update_io_tick_disk(du);
128 fio_mutex_up(disk_util_mutex);
132 static struct disk_util *disk_util_exists(int major, int minor)
134 struct flist_head *entry;
135 struct disk_util *du;
137 fio_mutex_down(disk_util_mutex);
139 flist_for_each(entry, &disk_list) {
140 du = flist_entry(entry, struct disk_util, list);
142 if (major == du->major && minor == du->minor) {
143 fio_mutex_up(disk_util_mutex);
148 fio_mutex_up(disk_util_mutex);
152 static int get_device_numbers(char *file_name, int *maj, int *min)
156 char tempname[PATH_MAX], *p;
158 if (!lstat(file_name, &st)) {
159 if (S_ISBLK(st.st_mode)) {
160 majdev = major(st.st_rdev);
161 mindev = minor(st.st_rdev);
162 } else if (S_ISCHR(st.st_mode)) {
163 majdev = major(st.st_rdev);
164 mindev = minor(st.st_rdev);
165 if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
167 } else if (S_ISFIFO(st.st_mode))
170 majdev = major(st.st_dev);
171 mindev = minor(st.st_dev);
175 * must be a file, open "." in that path
177 strncpy(tempname, file_name, PATH_MAX - 1);
178 p = dirname(tempname);
180 perror("disk util stat");
184 majdev = major(st.st_dev);
185 mindev = minor(st.st_dev);
194 static int read_block_dev_entry(char *path, int *maj, int *min)
199 f = fopen(path, "r");
205 p = fgets(line, sizeof(line), f);
211 if (sscanf(p, "%u:%u", maj, min) != 2)
217 static void find_add_disk_slaves(struct thread_data *td, char *path,
218 struct disk_util *masterdu)
220 DIR *dirhandle = NULL;
221 struct dirent *dirent = NULL;
222 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
223 struct disk_util *slavedu = NULL;
227 sprintf(slavesdir, "%s/%s", path, "slaves");
228 dirhandle = opendir(slavesdir);
232 while ((dirent = readdir(dirhandle)) != NULL) {
233 if (!strcmp(dirent->d_name, ".") ||
234 !strcmp(dirent->d_name, ".."))
237 sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, dirent->d_name);
238 /* Can we always assume that the slaves device entries
239 * are links to the real directories for the slave
242 linklen = readlink(temppath, slavepath, PATH_MAX - 1);
244 perror("readlink() for slave device.");
248 slavepath[linklen] = '\0';
250 sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
251 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
252 perror("Error getting slave device numbers.");
258 * See if this maj,min already exists
260 slavedu = disk_util_exists(majdev, mindev);
264 sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, slavepath);
265 __init_per_file_disk_util(td, majdev, mindev, temppath);
266 slavedu = disk_util_exists(majdev, mindev);
268 /* Should probably use an assert here. slavedu should
269 * always be present at this point. */
272 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
279 static struct disk_util *disk_util_add(struct thread_data *td, int majdev,
280 int mindev, char *path)
282 struct disk_util *du, *__du;
283 struct flist_head *entry;
286 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
288 du = smalloc(sizeof(*du));
290 log_err("fio: smalloc() pool exhausted\n");
294 memset(du, 0, sizeof(*du));
295 INIT_FLIST_HEAD(&du->list);
296 l = snprintf(du->path, sizeof(du->path), "%s/stat", path);
297 if (l < 0 || l >= sizeof(du->path)) {
298 log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n",
299 path, sizeof(du->path) - 1);
303 strncpy((char *) du->dus.name, basename(path), FIO_DU_NAME_SZ - 1);
304 du->sysfs_root = path;
307 INIT_FLIST_HEAD(&du->slavelist);
308 INIT_FLIST_HEAD(&du->slaves);
309 du->lock = fio_mutex_init(FIO_MUTEX_UNLOCKED);
312 fio_mutex_down(disk_util_mutex);
314 flist_for_each(entry, &disk_list) {
315 __du = flist_entry(entry, struct disk_util, list);
317 dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
319 if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) {
321 fio_mutex_up(disk_util_mutex);
326 dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
328 fio_gettime(&du->time, NULL);
329 get_io_ticks(du, &du->last_dus);
331 flist_add_tail(&du->list, &disk_list);
332 fio_mutex_up(disk_util_mutex);
334 find_add_disk_slaves(td, path, du);
338 static int check_dev_match(int majdev, int mindev, char *path)
342 if (read_block_dev_entry(path, &major, &minor))
345 if (majdev == major && mindev == minor)
351 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
362 while ((dir = readdir(D)) != NULL) {
365 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
368 sprintf(full_path, "%s%s%s", path, FIO_OS_PATH_SEPARATOR, dir->d_name);
370 if (!strcmp(dir->d_name, "dev")) {
371 if (!check_dev_match(majdev, mindev, full_path)) {
378 if (stat(full_path, &st) == -1) {
383 if (lstat(full_path, &st) == -1) {
389 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
392 found = find_block_dir(majdev, mindev, full_path, 0);
394 strcpy(path, full_path);
403 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
404 int majdev, int mindev,
412 * If there's a ../queue/ directory there, we are inside a partition.
413 * Check if that is the case and jump back. For loop/md/dm etc we
414 * are already in the right spot.
416 sprintf(tmp, "%s/../queue", path);
417 if (!stat(tmp, &st)) {
419 sprintf(tmp, "%s/queue", p);
420 if (stat(tmp, &st)) {
421 log_err("unknown sysfs layout\n");
424 strncpy(tmp, p, PATH_MAX - 1);
425 sprintf(path, "%s", tmp);
428 if (td->o.ioscheduler && !td->sysfs_root)
429 td->sysfs_root = strdup(path);
431 return disk_util_add(td, majdev, mindev, path);
434 static struct disk_util *init_per_file_disk_util(struct thread_data *td,
439 struct disk_util *du;
442 if (get_device_numbers(filename, &majdev, &mindev))
445 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
448 du = disk_util_exists(majdev, mindev);
450 if (td->o.ioscheduler && !td->sysfs_root)
451 td->sysfs_root = strdup(du->sysfs_root);
457 * for an fs without a device, we will repeatedly stat through
458 * sysfs which can take oodles of time for thousands of files. so
459 * cache the last lookup and compare with that before going through
462 if (mindev == last_mindev && majdev == last_majdev)
465 last_mindev = mindev;
466 last_majdev = majdev;
468 sprintf(foo, "/sys/block");
469 if (!find_block_dir(majdev, mindev, foo, 1))
472 return __init_per_file_disk_util(td, majdev, mindev, foo);
475 static struct disk_util *__init_disk_util(struct thread_data *td,
478 return init_per_file_disk_util(td, f->file_name);
481 void init_disk_util(struct thread_data *td)
486 if (!td->o.do_disk_util ||
487 (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL)))
490 for_each_file(td, f, i)
491 f->du = __init_disk_util(td, f);
494 static void show_agg_stats(struct disk_util_agg *agg, int terse)
496 if (!agg->slavecount)
500 log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
501 " aggrin_queue=%u, aggrutil=%3.2f%%",
502 agg->ios[0] / agg->slavecount,
503 agg->ios[1] / agg->slavecount,
504 agg->merges[0] / agg->slavecount,
505 agg->merges[1] / agg->slavecount,
506 agg->ticks[0] / agg->slavecount,
507 agg->ticks[1] / agg->slavecount,
508 agg->time_in_queue / agg->slavecount,
511 log_info(";slaves;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
512 agg->ios[0] / agg->slavecount,
513 agg->ios[1] / agg->slavecount,
514 agg->merges[0] / agg->slavecount,
515 agg->merges[1] / agg->slavecount,
516 agg->ticks[0] / agg->slavecount,
517 agg->ticks[1] / agg->slavecount,
518 agg->time_in_queue / agg->slavecount,
523 static void aggregate_slaves_stats(struct disk_util *masterdu)
525 struct disk_util_agg *agg = &masterdu->agg;
526 struct disk_util_stat *dus;
527 struct flist_head *entry;
528 struct disk_util *slavedu;
531 flist_for_each(entry, &masterdu->slaves) {
532 slavedu = flist_entry(entry, struct disk_util, slavelist);
534 agg->ios[0] += dus->s.ios[0];
535 agg->ios[1] += dus->s.ios[1];
536 agg->merges[0] += dus->s.merges[0];
537 agg->merges[1] += dus->s.merges[1];
538 agg->sectors[0] += dus->s.sectors[0];
539 agg->sectors[1] += dus->s.sectors[1];
540 agg->ticks[0] += dus->s.ticks[0];
541 agg->ticks[1] += dus->s.ticks[1];
542 agg->time_in_queue += dus->s.time_in_queue;
545 util = (double) (100 * dus->s.io_ticks / (double) slavedu->dus.s.msec);
546 /* System utilization is the utilization of the
547 * component with the highest utilization.
549 if (util > agg->max_util.u.f)
550 agg->max_util.u.f = util;
554 if (agg->max_util.u.f > 100.0)
555 agg->max_util.u.f = 100.0;
558 void disk_util_prune_entries(void)
560 fio_mutex_down(disk_util_mutex);
562 while (!flist_empty(&disk_list)) {
563 struct disk_util *du;
565 du = flist_first_entry(&disk_list, struct disk_util, list);
566 flist_del(&du->list);
570 last_majdev = last_mindev = -1;
571 fio_mutex_up(disk_util_mutex);
572 fio_mutex_remove(disk_util_mutex);
575 void print_disk_util(struct disk_util_stat *dus, struct disk_util_agg *agg,
581 util = (double) 100 * dus->s.io_ticks / (double) dus->s.msec;
589 log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
590 "in_queue=%u, util=%3.2f%%", dus->name,
591 dus->s.ios[0], dus->s.ios[1],
592 dus->s.merges[0], dus->s.merges[1],
593 dus->s.ticks[0], dus->s.ticks[1],
594 dus->s.time_in_queue, util);
596 log_info(";%s;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
597 dus->name, dus->s.ios[0],
598 dus->s.ios[1], dus->s.merges[0],
599 dus->s.merges[1], dus->s.ticks[0],
601 dus->s.time_in_queue, util);
605 * If the device has slaves, aggregate the stats for
606 * those slave devices also.
608 show_agg_stats(agg, terse);
614 void json_array_add_disk_util(struct disk_util_stat *dus,
615 struct disk_util_agg *agg, struct json_array *array)
617 struct json_object *obj;
621 util = (double) 100 * dus->s.io_ticks / (double) dus->s.msec;
625 obj = json_create_object();
626 json_array_add_value_object(array, obj);
628 json_object_add_value_string(obj, "name", dus->name);
629 json_object_add_value_int(obj, "read_ios", dus->s.ios[0]);
630 json_object_add_value_int(obj, "write_ios", dus->s.ios[1]);
631 json_object_add_value_int(obj, "read_merges", dus->s.merges[0]);
632 json_object_add_value_int(obj, "write_merges", dus->s.merges[1]);
633 json_object_add_value_int(obj, "read_ticks", dus->s.ticks[0]);
634 json_object_add_value_int(obj, "write_ticks", dus->s.ticks[1]);
635 json_object_add_value_int(obj, "in_queue", dus->s.time_in_queue);
636 json_object_add_value_float(obj, "util", util);
639 * If the device has slaves, aggregate the stats for
640 * those slave devices also.
642 if (!agg->slavecount)
644 json_object_add_value_int(obj, "aggr_read_ios",
645 agg->ios[0] / agg->slavecount);
646 json_object_add_value_int(obj, "aggr_write_ios",
647 agg->ios[1] / agg->slavecount);
648 json_object_add_value_int(obj, "aggr_read_merges",
649 agg->merges[0] / agg->slavecount);
650 json_object_add_value_int(obj, "aggr_write_merge",
651 agg->merges[1] / agg->slavecount);
652 json_object_add_value_int(obj, "aggr_read_ticks",
653 agg->ticks[0] / agg->slavecount);
654 json_object_add_value_int(obj, "aggr_write_ticks",
655 agg->ticks[1] / agg->slavecount);
656 json_object_add_value_int(obj, "aggr_in_queue",
657 agg->time_in_queue / agg->slavecount);
658 json_object_add_value_float(obj, "aggr_util", agg->max_util.u.f);
661 static void json_object_add_disk_utils(struct json_object *obj,
662 struct flist_head *head)
664 struct json_array *array = json_create_array();
665 struct flist_head *entry;
666 struct disk_util *du;
668 json_object_add_value_array(obj, "disk_util", array);
670 flist_for_each(entry, head) {
671 du = flist_entry(entry, struct disk_util, list);
673 aggregate_slaves_stats(du);
674 json_array_add_disk_util(&du->dus, &du->agg, array);
678 void show_disk_util(int terse, struct json_object *parent)
680 struct flist_head *entry;
681 struct disk_util *du;
683 fio_mutex_down(disk_util_mutex);
685 if (flist_empty(&disk_list)) {
686 fio_mutex_up(disk_util_mutex);
690 if (output_format == FIO_OUTPUT_JSON)
693 if (!terse && output_format != FIO_OUTPUT_JSON)
694 log_info("\nDisk stats (read/write):\n");
696 if (output_format == FIO_OUTPUT_JSON)
697 json_object_add_disk_utils(parent, &disk_list);
699 flist_for_each(entry, &disk_list) {
700 du = flist_entry(entry, struct disk_util, list);
702 aggregate_slaves_stats(du);
703 print_disk_util(&du->dus, &du->agg, terse);
706 fio_mutex_up(disk_util_mutex);
709 void setup_disk_util(void)
711 disk_util_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED);