14 static int last_majdev, last_mindev;
15 static struct disk_util *last_du;
17 static struct fio_mutex *disk_util_mutex;
19 FLIST_HEAD(disk_list);
21 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
22 int majdev, int mindev, char *path);
24 static void disk_util_free(struct disk_util *du)
29 while (!flist_empty(&du->slaves)) {
30 struct disk_util *slave;
32 slave = flist_entry(du->slaves.next, struct disk_util, slavelist);
33 flist_del(&slave->slavelist);
37 fio_mutex_remove(du->lock);
41 static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
44 unsigned long long sectors[2];
50 dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
52 f = fopen(du->path, "r");
56 p = fgets(line, sizeof(line), f);
62 dprint(FD_DISKUTIL, "%s: %s", du->path, p);
64 ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n",
66 &dus->s.merges[0], §ors[0],
67 &dus->s.ticks[0], &dus->s.ios[1],
68 &dus->s.merges[1], §ors[1],
69 &dus->s.ticks[1], &in_flight,
71 &dus->s.time_in_queue);
73 dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
74 dus->s.sectors[0] = sectors[0];
75 dus->s.sectors[1] = sectors[1];
79 static void update_io_tick_disk(struct disk_util *du)
81 struct disk_util_stat __dus, *dus, *ldus;
86 if (get_io_ticks(du, &__dus))
92 dus->s.sectors[0] += (__dus.s.sectors[0] - ldus->s.sectors[0]);
93 dus->s.sectors[1] += (__dus.s.sectors[1] - ldus->s.sectors[1]);
94 dus->s.ios[0] += (__dus.s.ios[0] - ldus->s.ios[0]);
95 dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]);
96 dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]);
97 dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]);
98 dus->s.ticks[0] += (__dus.s.ticks[0] - ldus->s.ticks[0]);
99 dus->s.ticks[1] += (__dus.s.ticks[1] - ldus->s.ticks[1]);
100 dus->s.io_ticks += (__dus.s.io_ticks - ldus->s.io_ticks);
101 dus->s.time_in_queue += (__dus.s.time_in_queue - ldus->s.time_in_queue);
103 fio_gettime(&t, NULL);
104 dus->s.msec += mtime_since(&du->time, &t);
105 memcpy(&du->time, &t, sizeof(t));
106 memcpy(&ldus->s, &__dus.s, sizeof(__dus.s));
109 int update_io_ticks(void)
111 struct flist_head *entry;
112 struct disk_util *du;
115 dprint(FD_DISKUTIL, "update io ticks\n");
117 fio_mutex_down(disk_util_mutex);
119 if (!disk_util_exit) {
120 flist_for_each(entry, &disk_list) {
121 du = flist_entry(entry, struct disk_util, list);
122 update_io_tick_disk(du);
127 fio_mutex_up(disk_util_mutex);
131 static struct disk_util *disk_util_exists(int major, int minor)
133 struct flist_head *entry;
134 struct disk_util *du;
136 fio_mutex_down(disk_util_mutex);
138 flist_for_each(entry, &disk_list) {
139 du = flist_entry(entry, struct disk_util, list);
141 if (major == du->major && minor == du->minor) {
142 fio_mutex_up(disk_util_mutex);
147 fio_mutex_up(disk_util_mutex);
151 static int get_device_numbers(char *file_name, int *maj, int *min)
155 char tempname[PATH_MAX], *p;
157 if (!lstat(file_name, &st)) {
158 if (S_ISBLK(st.st_mode)) {
159 majdev = major(st.st_rdev);
160 mindev = minor(st.st_rdev);
161 } else if (S_ISCHR(st.st_mode)) {
162 majdev = major(st.st_rdev);
163 mindev = minor(st.st_rdev);
164 if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
166 } else if (S_ISFIFO(st.st_mode))
169 majdev = major(st.st_dev);
170 mindev = minor(st.st_dev);
174 * must be a file, open "." in that path
176 strncpy(tempname, file_name, PATH_MAX - 1);
177 p = dirname(tempname);
179 perror("disk util stat");
183 majdev = major(st.st_dev);
184 mindev = minor(st.st_dev);
193 static int read_block_dev_entry(char *path, int *maj, int *min)
198 f = fopen(path, "r");
204 p = fgets(line, sizeof(line), f);
210 if (sscanf(p, "%u:%u", maj, min) != 2)
216 static void find_add_disk_slaves(struct thread_data *td, char *path,
217 struct disk_util *masterdu)
219 DIR *dirhandle = NULL;
220 struct dirent *dirent = NULL;
221 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
222 struct disk_util *slavedu = NULL;
226 sprintf(slavesdir, "%s/%s", path, "slaves");
227 dirhandle = opendir(slavesdir);
231 while ((dirent = readdir(dirhandle)) != NULL) {
232 if (!strcmp(dirent->d_name, ".") ||
233 !strcmp(dirent->d_name, ".."))
236 sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, dirent->d_name);
237 /* Can we always assume that the slaves device entries
238 * are links to the real directories for the slave
241 linklen = readlink(temppath, slavepath, PATH_MAX - 1);
243 perror("readlink() for slave device.");
247 slavepath[linklen] = '\0';
249 sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
250 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
251 perror("Error getting slave device numbers.");
257 * See if this maj,min already exists
259 slavedu = disk_util_exists(majdev, mindev);
263 sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, slavepath);
264 __init_per_file_disk_util(td, majdev, mindev, temppath);
265 slavedu = disk_util_exists(majdev, mindev);
267 /* Should probably use an assert here. slavedu should
268 * always be present at this point. */
271 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
278 static struct disk_util *disk_util_add(struct thread_data *td, int majdev,
279 int mindev, char *path)
281 struct disk_util *du, *__du;
282 struct flist_head *entry;
285 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
287 du = smalloc(sizeof(*du));
289 log_err("fio: smalloc() pool exhausted\n");
293 memset(du, 0, sizeof(*du));
294 INIT_FLIST_HEAD(&du->list);
295 l = snprintf(du->path, sizeof(du->path), "%s/stat", path);
296 if (l < 0 || l >= sizeof(du->path)) {
297 log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n",
298 path, sizeof(du->path) - 1);
302 strncpy((char *) du->dus.name, basename(path) - 1, FIO_DU_NAME_SZ);
303 du->sysfs_root = path;
306 INIT_FLIST_HEAD(&du->slavelist);
307 INIT_FLIST_HEAD(&du->slaves);
308 du->lock = fio_mutex_init(FIO_MUTEX_UNLOCKED);
311 fio_mutex_down(disk_util_mutex);
313 flist_for_each(entry, &disk_list) {
314 __du = flist_entry(entry, struct disk_util, list);
316 dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
318 if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) {
320 fio_mutex_up(disk_util_mutex);
325 dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
327 fio_gettime(&du->time, NULL);
328 get_io_ticks(du, &du->last_dus);
330 flist_add_tail(&du->list, &disk_list);
331 fio_mutex_up(disk_util_mutex);
333 find_add_disk_slaves(td, path, du);
337 static int check_dev_match(int majdev, int mindev, char *path)
341 if (read_block_dev_entry(path, &major, &minor))
344 if (majdev == major && mindev == minor)
350 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
361 while ((dir = readdir(D)) != NULL) {
364 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
367 sprintf(full_path, "%s%s%s", path, FIO_OS_PATH_SEPARATOR, dir->d_name);
369 if (!strcmp(dir->d_name, "dev")) {
370 if (!check_dev_match(majdev, mindev, full_path)) {
377 if (stat(full_path, &st) == -1) {
382 if (lstat(full_path, &st) == -1) {
388 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
391 found = find_block_dir(majdev, mindev, full_path, 0);
393 strcpy(path, full_path);
402 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
403 int majdev, int mindev,
411 * If there's a ../queue/ directory there, we are inside a partition.
412 * Check if that is the case and jump back. For loop/md/dm etc we
413 * are already in the right spot.
415 sprintf(tmp, "%s/../queue", path);
416 if (!stat(tmp, &st)) {
418 sprintf(tmp, "%s/queue", p);
419 if (stat(tmp, &st)) {
420 log_err("unknown sysfs layout\n");
423 strncpy(tmp, p, PATH_MAX - 1);
424 sprintf(path, "%s", tmp);
427 if (td->o.ioscheduler && !td->sysfs_root)
428 td->sysfs_root = strdup(path);
430 return disk_util_add(td, majdev, mindev, path);
433 static struct disk_util *init_per_file_disk_util(struct thread_data *td,
438 struct disk_util *du;
441 if (get_device_numbers(filename, &majdev, &mindev))
444 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
447 du = disk_util_exists(majdev, mindev);
449 if (td->o.ioscheduler && !td->sysfs_root)
450 td->sysfs_root = strdup(du->sysfs_root);
456 * for an fs without a device, we will repeatedly stat through
457 * sysfs which can take oodles of time for thousands of files. so
458 * cache the last lookup and compare with that before going through
461 if (mindev == last_mindev && majdev == last_majdev)
464 last_mindev = mindev;
465 last_majdev = majdev;
467 sprintf(foo, "/sys/block");
468 if (!find_block_dir(majdev, mindev, foo, 1))
471 return __init_per_file_disk_util(td, majdev, mindev, foo);
474 static struct disk_util *__init_disk_util(struct thread_data *td,
477 return init_per_file_disk_util(td, f->file_name);
480 void init_disk_util(struct thread_data *td)
485 if (!td->o.do_disk_util ||
486 (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL)))
489 for_each_file(td, f, i)
490 f->du = __init_disk_util(td, f);
493 static void show_agg_stats(struct disk_util_agg *agg, int terse)
495 if (!agg->slavecount)
499 log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
500 " aggrin_queue=%u, aggrutil=%3.2f%%",
501 agg->ios[0] / agg->slavecount,
502 agg->ios[1] / agg->slavecount,
503 agg->merges[0] / agg->slavecount,
504 agg->merges[1] / agg->slavecount,
505 agg->ticks[0] / agg->slavecount,
506 agg->ticks[1] / agg->slavecount,
507 agg->time_in_queue / agg->slavecount,
510 log_info(";slaves;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
511 agg->ios[0] / agg->slavecount,
512 agg->ios[1] / agg->slavecount,
513 agg->merges[0] / agg->slavecount,
514 agg->merges[1] / agg->slavecount,
515 agg->ticks[0] / agg->slavecount,
516 agg->ticks[1] / agg->slavecount,
517 agg->time_in_queue / agg->slavecount,
522 static void aggregate_slaves_stats(struct disk_util *masterdu)
524 struct disk_util_agg *agg = &masterdu->agg;
525 struct disk_util_stat *dus;
526 struct flist_head *entry;
527 struct disk_util *slavedu;
530 flist_for_each(entry, &masterdu->slaves) {
531 slavedu = flist_entry(entry, struct disk_util, slavelist);
533 agg->ios[0] += dus->s.ios[0];
534 agg->ios[1] += dus->s.ios[1];
535 agg->merges[0] += dus->s.merges[0];
536 agg->merges[1] += dus->s.merges[1];
537 agg->sectors[0] += dus->s.sectors[0];
538 agg->sectors[1] += dus->s.sectors[1];
539 agg->ticks[0] += dus->s.ticks[0];
540 agg->ticks[1] += dus->s.ticks[1];
541 agg->time_in_queue += dus->s.time_in_queue;
544 util = (double) (100 * dus->s.io_ticks / (double) slavedu->dus.s.msec);
545 /* System utilization is the utilization of the
546 * component with the highest utilization.
548 if (util > agg->max_util.u.f)
549 agg->max_util.u.f = util;
553 if (agg->max_util.u.f > 100.0)
554 agg->max_util.u.f = 100.0;
557 void disk_util_prune_entries(void)
559 fio_mutex_down(disk_util_mutex);
561 while (!flist_empty(&disk_list)) {
562 struct disk_util *du;
564 du = flist_entry(disk_list.next, struct disk_util, list);
565 flist_del(&du->list);
569 last_majdev = last_mindev = -1;
570 fio_mutex_up(disk_util_mutex);
571 fio_mutex_remove(disk_util_mutex);
574 void print_disk_util(struct disk_util_stat *dus, struct disk_util_agg *agg,
580 util = (double) 100 * dus->s.io_ticks / (double) dus->s.msec;
588 log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
589 "in_queue=%u, util=%3.2f%%", dus->name,
590 dus->s.ios[0], dus->s.ios[1],
591 dus->s.merges[0], dus->s.merges[1],
592 dus->s.ticks[0], dus->s.ticks[1],
593 dus->s.time_in_queue, util);
595 log_info(";%s;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
596 dus->name, dus->s.ios[0],
597 dus->s.ios[1], dus->s.merges[0],
598 dus->s.merges[1], dus->s.ticks[0],
600 dus->s.time_in_queue, util);
604 * If the device has slaves, aggregate the stats for
605 * those slave devices also.
607 show_agg_stats(agg, terse);
613 void json_array_add_disk_util(struct disk_util_stat *dus,
614 struct disk_util_agg *agg, struct json_array *array)
616 struct json_object *obj;
620 util = (double) 100 * dus->s.io_ticks / (double) dus->s.msec;
624 obj = json_create_object();
625 json_array_add_value_object(array, obj);
627 json_object_add_value_string(obj, "name", dus->name);
628 json_object_add_value_int(obj, "read_ios", dus->s.ios[0]);
629 json_object_add_value_int(obj, "write_ios", dus->s.ios[1]);
630 json_object_add_value_int(obj, "read_merges", dus->s.merges[0]);
631 json_object_add_value_int(obj, "write_merges", dus->s.merges[1]);
632 json_object_add_value_int(obj, "read_ticks", dus->s.ticks[0]);
633 json_object_add_value_int(obj, "write_ticks", dus->s.ticks[1]);
634 json_object_add_value_int(obj, "in_queue", dus->s.time_in_queue);
635 json_object_add_value_float(obj, "util", util);
638 * If the device has slaves, aggregate the stats for
639 * those slave devices also.
641 if (!agg->slavecount)
643 json_object_add_value_int(obj, "aggr_read_ios",
644 agg->ios[0] / agg->slavecount);
645 json_object_add_value_int(obj, "aggr_write_ios",
646 agg->ios[1] / agg->slavecount);
647 json_object_add_value_int(obj, "aggr_read_merges",
648 agg->merges[0] / agg->slavecount);
649 json_object_add_value_int(obj, "aggr_write_merge",
650 agg->merges[1] / agg->slavecount);
651 json_object_add_value_int(obj, "aggr_read_ticks",
652 agg->ticks[0] / agg->slavecount);
653 json_object_add_value_int(obj, "aggr_write_ticks",
654 agg->ticks[1] / agg->slavecount);
655 json_object_add_value_int(obj, "aggr_in_queue",
656 agg->time_in_queue / agg->slavecount);
657 json_object_add_value_float(obj, "aggr_util", agg->max_util.u.f);
660 static void json_object_add_disk_utils(struct json_object *obj,
661 struct flist_head *head)
663 struct json_array *array = json_create_array();
664 struct flist_head *entry;
665 struct disk_util *du;
667 json_object_add_value_array(obj, "disk_util", array);
669 flist_for_each(entry, head) {
670 du = flist_entry(entry, struct disk_util, list);
672 aggregate_slaves_stats(du);
673 json_array_add_disk_util(&du->dus, &du->agg, array);
677 void show_disk_util(int terse, struct json_object *parent)
679 struct flist_head *entry;
680 struct disk_util *du;
682 fio_mutex_down(disk_util_mutex);
684 if (flist_empty(&disk_list)) {
685 fio_mutex_up(disk_util_mutex);
689 if (!terse && !parent)
690 log_info("\nDisk stats (read/write):\n");
692 if (output_format == FIO_OUTPUT_JSON) {
693 json_object_add_disk_utils(parent, &disk_list);
695 flist_for_each(entry, &disk_list) {
696 du = flist_entry(entry, struct disk_util, list);
698 aggregate_slaves_stats(du);
699 print_disk_util(&du->dus, &du->agg, terse);
702 fio_mutex_up(disk_util_mutex);
705 void setup_disk_util(void)
707 disk_util_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED);