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", &dus->ios[0],
65 &dus->merges[0], §ors[0],
66 &dus->ticks[0], &dus->ios[1],
67 &dus->merges[1], §ors[1],
68 &dus->ticks[1], &in_flight,
69 &dus->io_ticks, &dus->time_in_queue);
71 dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
72 dus->sectors[0] = sectors[0];
73 dus->sectors[1] = sectors[1];
77 static void update_io_tick_disk(struct disk_util *du)
79 struct disk_util_stat __dus, *dus, *ldus;
84 if (get_io_ticks(du, &__dus))
90 dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
91 dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
92 dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
93 dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
94 dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
95 dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
96 dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
97 dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
98 dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
99 dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
101 fio_gettime(&t, NULL);
102 dus->msec += mtime_since(&du->time, &t);
103 memcpy(&du->time, &t, sizeof(t));
104 memcpy(ldus, &__dus, sizeof(__dus));
107 int update_io_ticks(void)
109 struct flist_head *entry;
110 struct disk_util *du;
113 dprint(FD_DISKUTIL, "update io ticks\n");
115 fio_mutex_down(disk_util_mutex);
117 if (!disk_util_exit) {
118 flist_for_each(entry, &disk_list) {
119 du = flist_entry(entry, struct disk_util, list);
120 update_io_tick_disk(du);
125 fio_mutex_up(disk_util_mutex);
129 static struct disk_util *disk_util_exists(int major, int minor)
131 struct flist_head *entry;
132 struct disk_util *du;
134 fio_mutex_down(disk_util_mutex);
136 flist_for_each(entry, &disk_list) {
137 du = flist_entry(entry, struct disk_util, list);
139 if (major == du->major && minor == du->minor) {
140 fio_mutex_up(disk_util_mutex);
145 fio_mutex_up(disk_util_mutex);
149 static int get_device_numbers(char *file_name, int *maj, int *min)
153 char tempname[PATH_MAX], *p;
155 if (!lstat(file_name, &st)) {
156 if (S_ISBLK(st.st_mode)) {
157 majdev = major(st.st_rdev);
158 mindev = minor(st.st_rdev);
159 } else if (S_ISCHR(st.st_mode)) {
160 majdev = major(st.st_rdev);
161 mindev = minor(st.st_rdev);
162 if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
164 } else if (S_ISFIFO(st.st_mode))
167 majdev = major(st.st_dev);
168 mindev = minor(st.st_dev);
172 * must be a file, open "." in that path
174 strncpy(tempname, file_name, PATH_MAX - 1);
175 p = dirname(tempname);
177 perror("disk util stat");
181 majdev = major(st.st_dev);
182 mindev = minor(st.st_dev);
191 static int read_block_dev_entry(char *path, int *maj, int *min)
196 f = fopen(path, "r");
202 p = fgets(line, sizeof(line), f);
208 if (sscanf(p, "%u:%u", maj, min) != 2)
214 static void find_add_disk_slaves(struct thread_data *td, char *path,
215 struct disk_util *masterdu)
217 DIR *dirhandle = NULL;
218 struct dirent *dirent = NULL;
219 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
220 struct disk_util *slavedu = NULL;
224 sprintf(slavesdir, "%s/%s", path, "slaves");
225 dirhandle = opendir(slavesdir);
229 while ((dirent = readdir(dirhandle)) != NULL) {
230 if (!strcmp(dirent->d_name, ".") ||
231 !strcmp(dirent->d_name, ".."))
234 sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, dirent->d_name);
235 /* Can we always assume that the slaves device entries
236 * are links to the real directories for the slave
239 linklen = readlink(temppath, slavepath, PATH_MAX - 0);
241 perror("readlink() for slave device.");
244 slavepath[linklen] = '\0';
246 sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
247 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
248 perror("Error getting slave device numbers.");
253 * See if this maj,min already exists
255 slavedu = disk_util_exists(majdev, mindev);
259 sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, slavepath);
260 __init_per_file_disk_util(td, majdev, mindev, temppath);
261 slavedu = disk_util_exists(majdev, mindev);
263 /* Should probably use an assert here. slavedu should
264 * always be present at this point. */
267 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
274 static struct disk_util *disk_util_add(struct thread_data *td, int majdev,
275 int mindev, char *path)
277 struct disk_util *du, *__du;
278 struct flist_head *entry;
281 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
283 du = smalloc(sizeof(*du));
285 log_err("fio: smalloc() pool exhausted\n");
289 memset(du, 0, sizeof(*du));
290 INIT_FLIST_HEAD(&du->list);
291 l = snprintf(du->path, sizeof(du->path), "%s/stat", path);
292 if (l < 0 || l >= sizeof(du->path)) {
293 log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n",
294 path, sizeof(du->path) - 1);
298 strncpy((char *) du->dus.name, basename(path), FIO_DU_NAME_SZ);
299 du->sysfs_root = path;
302 INIT_FLIST_HEAD(&du->slavelist);
303 INIT_FLIST_HEAD(&du->slaves);
304 du->lock = fio_mutex_init(FIO_MUTEX_UNLOCKED);
307 fio_mutex_down(disk_util_mutex);
309 flist_for_each(entry, &disk_list) {
310 __du = flist_entry(entry, struct disk_util, list);
312 dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
314 if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) {
316 fio_mutex_up(disk_util_mutex);
321 dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
323 fio_gettime(&du->time, NULL);
324 get_io_ticks(du, &du->last_dus);
326 flist_add_tail(&du->list, &disk_list);
327 fio_mutex_up(disk_util_mutex);
329 find_add_disk_slaves(td, path, du);
333 static int check_dev_match(int majdev, int mindev, char *path)
337 if (read_block_dev_entry(path, &major, &minor))
340 if (majdev == major && mindev == minor)
346 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
357 while ((dir = readdir(D)) != NULL) {
360 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
363 sprintf(full_path, "%s%s%s", path, FIO_OS_PATH_SEPARATOR, dir->d_name);
365 if (!strcmp(dir->d_name, "dev")) {
366 if (!check_dev_match(majdev, mindev, full_path)) {
373 if (stat(full_path, &st) == -1) {
378 if (lstat(full_path, &st) == -1) {
384 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
387 found = find_block_dir(majdev, mindev, full_path, 0);
389 strcpy(path, full_path);
398 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
399 int majdev, int mindev,
407 * If there's a ../queue/ directory there, we are inside a partition.
408 * Check if that is the case and jump back. For loop/md/dm etc we
409 * are already in the right spot.
411 sprintf(tmp, "%s/../queue", path);
412 if (!stat(tmp, &st)) {
414 sprintf(tmp, "%s/queue", p);
415 if (stat(tmp, &st)) {
416 log_err("unknown sysfs layout\n");
419 strncpy(tmp, p, PATH_MAX - 1);
420 sprintf(path, "%s", tmp);
423 if (td->o.ioscheduler && !td->sysfs_root)
424 td->sysfs_root = strdup(path);
426 return disk_util_add(td, majdev, mindev, path);
429 static struct disk_util *init_per_file_disk_util(struct thread_data *td,
434 struct disk_util *du;
437 if (get_device_numbers(filename, &majdev, &mindev))
440 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
443 du = disk_util_exists(majdev, mindev);
445 if (td->o.ioscheduler && !td->sysfs_root)
446 td->sysfs_root = strdup(du->sysfs_root);
452 * for an fs without a device, we will repeatedly stat through
453 * sysfs which can take oodles of time for thousands of files. so
454 * cache the last lookup and compare with that before going through
457 if (mindev == last_mindev && majdev == last_majdev)
460 last_mindev = mindev;
461 last_majdev = majdev;
463 sprintf(foo, "/sys/block");
464 if (!find_block_dir(majdev, mindev, foo, 1))
467 return __init_per_file_disk_util(td, majdev, mindev, foo);
470 static struct disk_util *__init_disk_util(struct thread_data *td,
473 return init_per_file_disk_util(td, f->file_name);
476 void init_disk_util(struct thread_data *td)
481 if (!td->o.do_disk_util ||
482 (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL)))
485 for_each_file(td, f, i)
486 f->du = __init_disk_util(td, f);
489 static void show_agg_stats(struct disk_util_agg *agg, int terse)
491 if (!agg->slavecount)
495 log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
496 " aggrin_queue=%u, aggrutil=%3.2f%%",
497 agg->ios[0] / agg->slavecount,
498 agg->ios[1] / agg->slavecount,
499 agg->merges[0] / agg->slavecount,
500 agg->merges[1] / agg->slavecount,
501 agg->ticks[0] / agg->slavecount,
502 agg->ticks[1] / agg->slavecount,
503 agg->time_in_queue / agg->slavecount,
506 log_info(";slaves;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
507 agg->ios[0] / agg->slavecount,
508 agg->ios[1] / agg->slavecount,
509 agg->merges[0] / agg->slavecount,
510 agg->merges[1] / agg->slavecount,
511 agg->ticks[0] / agg->slavecount,
512 agg->ticks[1] / agg->slavecount,
513 agg->time_in_queue / agg->slavecount,
518 static void aggregate_slaves_stats(struct disk_util *masterdu)
520 struct disk_util_agg *agg = &masterdu->agg;
521 struct disk_util_stat *dus;
522 struct flist_head *entry;
523 struct disk_util *slavedu;
526 flist_for_each(entry, &masterdu->slaves) {
527 slavedu = flist_entry(entry, struct disk_util, slavelist);
529 agg->ios[0] += dus->ios[0];
530 agg->ios[1] += dus->ios[1];
531 agg->merges[0] += dus->merges[0];
532 agg->merges[1] += dus->merges[1];
533 agg->sectors[0] += dus->sectors[0];
534 agg->sectors[1] += dus->sectors[1];
535 agg->ticks[0] += dus->ticks[0];
536 agg->ticks[1] += dus->ticks[1];
537 agg->time_in_queue += dus->time_in_queue;
540 util = (double) (100 * dus->io_ticks / (double) slavedu->dus.msec);
541 /* System utilization is the utilization of the
542 * component with the highest utilization.
544 if (util > agg->max_util.u.f)
545 agg->max_util.u.f = util;
549 if (agg->max_util.u.f > 100.0)
550 agg->max_util.u.f = 100.0;
553 void disk_util_prune_entries(void)
555 fio_mutex_down(disk_util_mutex);
557 while (!flist_empty(&disk_list)) {
558 struct disk_util *du;
560 du = flist_entry(disk_list.next, struct disk_util, list);
561 flist_del(&du->list);
565 last_majdev = last_mindev = -1;
566 fio_mutex_up(disk_util_mutex);
567 fio_mutex_remove(disk_util_mutex);
570 void print_disk_util(struct disk_util_stat *dus, struct disk_util_agg *agg,
576 util = (double) 100 * dus->io_ticks / (double) dus->msec;
584 log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
585 "in_queue=%u, util=%3.2f%%", dus->name,
586 dus->ios[0], dus->ios[1],
587 dus->merges[0], dus->merges[1],
588 dus->ticks[0], dus->ticks[1],
589 dus->time_in_queue, util);
591 log_info(";%s;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
592 dus->name, dus->ios[0], dus->ios[1],
593 dus->merges[0], dus->merges[1],
594 dus->ticks[0], dus->ticks[1],
595 dus->time_in_queue, util);
599 * If the device has slaves, aggregate the stats for
600 * those slave devices also.
602 show_agg_stats(agg, terse);
608 void json_array_add_disk_util(struct disk_util_stat *dus,
609 struct disk_util_agg *agg, struct json_array *array)
611 struct json_object *obj;
615 util = (double) 100 * dus->io_ticks / (double) dus->msec;
619 obj = json_create_object();
620 json_array_add_value_object(array, obj);
622 json_object_add_value_string(obj, "name", dus->name);
623 json_object_add_value_int(obj, "read_ios", dus->ios[0]);
624 json_object_add_value_int(obj, "write_ios", dus->ios[1]);
625 json_object_add_value_int(obj, "read_merges", dus->merges[0]);
626 json_object_add_value_int(obj, "write_merges", dus->merges[1]);
627 json_object_add_value_int(obj, "read_ticks", dus->ticks[0]);
628 json_object_add_value_int(obj, "write_ticks", dus->ticks[1]);
629 json_object_add_value_int(obj, "in_queue", dus->time_in_queue);
630 json_object_add_value_float(obj, "util", util);
633 * If the device has slaves, aggregate the stats for
634 * those slave devices also.
636 if (!agg->slavecount)
638 json_object_add_value_int(obj, "aggr_read_ios",
639 agg->ios[0] / agg->slavecount);
640 json_object_add_value_int(obj, "aggr_write_ios",
641 agg->ios[1] / agg->slavecount);
642 json_object_add_value_int(obj, "aggr_read_merges",
643 agg->merges[0] / agg->slavecount);
644 json_object_add_value_int(obj, "aggr_write_merge",
645 agg->merges[1] / agg->slavecount);
646 json_object_add_value_int(obj, "aggr_read_ticks",
647 agg->ticks[0] / agg->slavecount);
648 json_object_add_value_int(obj, "aggr_write_ticks",
649 agg->ticks[1] / agg->slavecount);
650 json_object_add_value_int(obj, "aggr_in_queue",
651 agg->time_in_queue / agg->slavecount);
652 json_object_add_value_float(obj, "aggr_util", agg->max_util.u.f);
655 void json_object_add_disk_utils(struct json_object *obj,
656 struct flist_head *head)
658 struct json_array *array = json_create_array();
659 struct flist_head *entry;
660 struct disk_util *du;
662 json_object_add_value_array(obj, "disk_util", array);
664 flist_for_each(entry, head) {
665 du = flist_entry(entry, struct disk_util, list);
667 aggregate_slaves_stats(du);
668 json_array_add_disk_util(&du->dus, &du->agg, array);
672 void show_disk_util(int terse, struct json_object *parent)
674 struct flist_head *entry;
675 struct disk_util *du;
677 fio_mutex_down(disk_util_mutex);
679 if (flist_empty(&disk_list)) {
680 fio_mutex_up(disk_util_mutex);
684 if (!terse && !parent)
685 log_info("\nDisk stats (read/write):\n");
687 if (output_format == FIO_OUTPUT_JSON) {
688 json_object_add_disk_utils(parent, &disk_list);
690 flist_for_each(entry, &disk_list) {
691 du = flist_entry(entry, struct disk_util, list);
693 aggregate_slaves_stats(du);
694 print_disk_util(&du->dus, &du->agg, terse);
697 fio_mutex_up(disk_util_mutex);
700 void setup_disk_util(void)
702 disk_util_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED);