14 static int last_majdev, last_mindev;
15 static struct disk_util *last_du;
17 static struct fio_mutex *disk_util_mutex;
18 static int disk_util_exit;
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_entry(du->slaves.next, 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", &dus->ios[0],
66 &dus->merges[0], §ors[0],
67 &dus->ticks[0], &dus->ios[1],
68 &dus->merges[1], §ors[1],
69 &dus->ticks[1], &in_flight,
70 &dus->io_ticks, &dus->time_in_queue);
72 dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
73 dus->sectors[0] = sectors[0];
74 dus->sectors[1] = sectors[1];
78 static void update_io_tick_disk(struct disk_util *du)
80 struct disk_util_stat __dus, *dus, *ldus;
85 if (get_io_ticks(du, &__dus))
91 dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
92 dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
93 dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
94 dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
95 dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
96 dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
97 dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
98 dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
99 dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
100 dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
102 fio_gettime(&t, NULL);
103 dus->msec += mtime_since(&du->time, &t);
104 memcpy(&du->time, &t, sizeof(t));
105 memcpy(ldus, &__dus, sizeof(__dus));
108 int update_io_ticks(void)
110 struct flist_head *entry;
111 struct disk_util *du;
114 dprint(FD_DISKUTIL, "update io ticks\n");
116 fio_mutex_down(disk_util_mutex);
118 if (!disk_util_exit) {
119 flist_for_each(entry, &disk_list) {
120 du = flist_entry(entry, struct disk_util, list);
121 update_io_tick_disk(du);
126 fio_mutex_up(disk_util_mutex);
130 static struct disk_util *disk_util_exists(int major, int minor)
132 struct flist_head *entry;
133 struct disk_util *du;
135 fio_mutex_down(disk_util_mutex);
137 flist_for_each(entry, &disk_list) {
138 du = flist_entry(entry, struct disk_util, list);
140 if (major == du->major && minor == du->minor) {
141 fio_mutex_up(disk_util_mutex);
146 fio_mutex_up(disk_util_mutex);
150 static int get_device_numbers(char *file_name, int *maj, int *min)
154 char tempname[PATH_MAX], *p;
156 if (!lstat(file_name, &st)) {
157 if (S_ISBLK(st.st_mode)) {
158 majdev = major(st.st_rdev);
159 mindev = minor(st.st_rdev);
160 } else if (S_ISCHR(st.st_mode)) {
161 majdev = major(st.st_rdev);
162 mindev = minor(st.st_rdev);
163 if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
165 } else if (S_ISFIFO(st.st_mode))
168 majdev = major(st.st_dev);
169 mindev = minor(st.st_dev);
173 * must be a file, open "." in that path
175 strncpy(tempname, file_name, PATH_MAX - 1);
176 p = dirname(tempname);
178 perror("disk util stat");
182 majdev = major(st.st_dev);
183 mindev = minor(st.st_dev);
192 static int read_block_dev_entry(char *path, int *maj, int *min)
197 f = fopen(path, "r");
203 p = fgets(line, sizeof(line), f);
209 if (sscanf(p, "%u:%u", maj, min) != 2)
215 static void find_add_disk_slaves(struct thread_data *td, char *path,
216 struct disk_util *masterdu)
218 DIR *dirhandle = NULL;
219 struct dirent *dirent = NULL;
220 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
221 struct disk_util *slavedu = NULL;
225 sprintf(slavesdir, "%s/%s", path, "slaves");
226 dirhandle = opendir(slavesdir);
230 while ((dirent = readdir(dirhandle)) != NULL) {
231 if (!strcmp(dirent->d_name, ".") ||
232 !strcmp(dirent->d_name, ".."))
235 sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, dirent->d_name);
236 /* Can we always assume that the slaves device entries
237 * are links to the real directories for the slave
240 linklen = readlink(temppath, slavepath, PATH_MAX - 0);
242 perror("readlink() for slave device.");
245 slavepath[linklen] = '\0';
247 sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
248 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
249 perror("Error getting slave device numbers.");
254 * See if this maj,min already exists
256 slavedu = disk_util_exists(majdev, mindev);
260 sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, slavepath);
261 __init_per_file_disk_util(td, majdev, mindev, temppath);
262 slavedu = disk_util_exists(majdev, mindev);
264 /* Should probably use an assert here. slavedu should
265 * always be present at this point. */
268 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
275 static struct disk_util *disk_util_add(struct thread_data *td, int majdev,
276 int mindev, char *path)
278 struct disk_util *du, *__du;
279 struct flist_head *entry;
281 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
283 du = smalloc(sizeof(*du));
284 memset(du, 0, sizeof(*du));
285 INIT_FLIST_HEAD(&du->list);
286 sprintf(du->path, "%s/stat", path);
287 strncpy((char *) du->dus.name, basename(path), FIO_DU_NAME_SZ);
288 du->sysfs_root = path;
291 INIT_FLIST_HEAD(&du->slavelist);
292 INIT_FLIST_HEAD(&du->slaves);
293 du->lock = fio_mutex_init(FIO_MUTEX_UNLOCKED);
296 fio_mutex_down(disk_util_mutex);
298 flist_for_each(entry, &disk_list) {
299 __du = flist_entry(entry, struct disk_util, list);
301 dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
303 if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) {
305 fio_mutex_up(disk_util_mutex);
310 dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
312 fio_gettime(&du->time, NULL);
313 get_io_ticks(du, &du->last_dus);
315 flist_add_tail(&du->list, &disk_list);
316 fio_mutex_up(disk_util_mutex);
318 find_add_disk_slaves(td, path, du);
322 static int check_dev_match(int majdev, int mindev, char *path)
326 if (read_block_dev_entry(path, &major, &minor))
329 if (majdev == major && mindev == minor)
335 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
346 while ((dir = readdir(D)) != NULL) {
349 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
352 sprintf(full_path, "%s%s%s", path, FIO_OS_PATH_SEPARATOR, dir->d_name);
354 if (!strcmp(dir->d_name, "dev")) {
355 if (!check_dev_match(majdev, mindev, full_path)) {
362 if (stat(full_path, &st) == -1) {
367 if (lstat(full_path, &st) == -1) {
373 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
376 found = find_block_dir(majdev, mindev, full_path, 0);
378 strcpy(path, full_path);
387 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
388 int majdev, int mindev,
396 * If there's a ../queue/ directory there, we are inside a partition.
397 * Check if that is the case and jump back. For loop/md/dm etc we
398 * are already in the right spot.
400 sprintf(tmp, "%s/../queue", path);
401 if (!stat(tmp, &st)) {
403 sprintf(tmp, "%s/queue", p);
404 if (stat(tmp, &st)) {
405 log_err("unknown sysfs layout\n");
408 strncpy(tmp, p, PATH_MAX - 1);
409 sprintf(path, "%s", tmp);
412 if (td->o.ioscheduler && !td->sysfs_root)
413 td->sysfs_root = strdup(path);
415 return disk_util_add(td, majdev, mindev, path);
418 static struct disk_util *init_per_file_disk_util(struct thread_data *td,
423 struct disk_util *du;
426 if (get_device_numbers(filename, &majdev, &mindev))
429 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
432 du = disk_util_exists(majdev, mindev);
434 if (td->o.ioscheduler && !td->sysfs_root)
435 td->sysfs_root = strdup(du->sysfs_root);
441 * for an fs without a device, we will repeatedly stat through
442 * sysfs which can take oodles of time for thousands of files. so
443 * cache the last lookup and compare with that before going through
446 if (mindev == last_mindev && majdev == last_majdev)
449 last_mindev = mindev;
450 last_majdev = majdev;
452 sprintf(foo, "/sys/block");
453 if (!find_block_dir(majdev, mindev, foo, 1))
456 return __init_per_file_disk_util(td, majdev, mindev, foo);
459 static struct disk_util *__init_disk_util(struct thread_data *td,
462 return init_per_file_disk_util(td, f->file_name);
465 void init_disk_util(struct thread_data *td)
470 if (!td->o.do_disk_util ||
471 (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL)))
474 for_each_file(td, f, i)
475 f->du = __init_disk_util(td, f);
478 static void show_agg_stats(struct disk_util_agg *agg, int terse)
480 if (!agg->slavecount)
484 log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
485 " aggrin_queue=%u, aggrutil=%3.2f%%",
486 agg->ios[0] / agg->slavecount,
487 agg->ios[1] / agg->slavecount,
488 agg->merges[0] / agg->slavecount,
489 agg->merges[1] / agg->slavecount,
490 agg->ticks[0] / agg->slavecount,
491 agg->ticks[1] / agg->slavecount,
492 agg->time_in_queue / agg->slavecount,
495 log_info(";slaves;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
496 agg->ios[0] / agg->slavecount,
497 agg->ios[1] / agg->slavecount,
498 agg->merges[0] / agg->slavecount,
499 agg->merges[1] / agg->slavecount,
500 agg->ticks[0] / agg->slavecount,
501 agg->ticks[1] / agg->slavecount,
502 agg->time_in_queue / agg->slavecount,
507 static void aggregate_slaves_stats(struct disk_util *masterdu)
509 struct disk_util_agg *agg = &masterdu->agg;
510 struct disk_util_stat *dus;
511 struct flist_head *entry;
512 struct disk_util *slavedu;
515 flist_for_each(entry, &masterdu->slaves) {
516 slavedu = flist_entry(entry, struct disk_util, slavelist);
518 agg->ios[0] += dus->ios[0];
519 agg->ios[1] += dus->ios[1];
520 agg->merges[0] += dus->merges[0];
521 agg->merges[1] += dus->merges[1];
522 agg->sectors[0] += dus->sectors[0];
523 agg->sectors[1] += dus->sectors[1];
524 agg->ticks[0] += dus->ticks[0];
525 agg->ticks[1] += dus->ticks[1];
526 agg->time_in_queue += dus->time_in_queue;
529 util = (double) (100 * dus->io_ticks / (double) slavedu->dus.msec);
530 /* System utilization is the utilization of the
531 * component with the highest utilization.
533 if (util > agg->max_util.u.f)
534 agg->max_util.u.f = util;
538 if (agg->max_util.u.f > 100.0)
539 agg->max_util.u.f = 100.0;
542 void free_disk_util(void)
544 struct disk_util *du;
547 wait_for_disk_thread_exit();
549 fio_mutex_down(disk_util_mutex);
551 while (!flist_empty(&disk_list)) {
552 du = flist_entry(disk_list.next, struct disk_util, list);
553 flist_del(&du->list);
557 last_majdev = last_mindev = -1;
558 fio_mutex_up(disk_util_mutex);
559 fio_mutex_remove(disk_util_mutex);
562 void print_disk_util(struct disk_util_stat *dus, struct disk_util_agg *agg,
568 util = (double) 100 * dus->io_ticks / (double) dus->msec;
576 log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
577 "in_queue=%u, util=%3.2f%%", dus->name,
578 dus->ios[0], dus->ios[1],
579 dus->merges[0], dus->merges[1],
580 dus->ticks[0], dus->ticks[1],
581 dus->time_in_queue, util);
583 log_info(";%s;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
584 dus->name, dus->ios[0], dus->ios[1],
585 dus->merges[0], dus->merges[1],
586 dus->ticks[0], dus->ticks[1],
587 dus->time_in_queue, util);
591 * If the device has slaves, aggregate the stats for
592 * those slave devices also.
594 show_agg_stats(agg, terse);
600 void show_disk_util(int terse)
602 struct flist_head *entry;
603 struct disk_util *du;
605 fio_mutex_down(disk_util_mutex);
607 if (flist_empty(&disk_list)) {
608 fio_mutex_up(disk_util_mutex);
613 log_info("\nDisk stats (read/write):\n");
615 flist_for_each(entry, &disk_list) {
616 du = flist_entry(entry, struct disk_util, list);
618 aggregate_slaves_stats(du);
619 print_disk_util(&du->dus, &du->agg, terse);
622 fio_mutex_up(disk_util_mutex);
625 void setup_disk_util(void)
627 disk_util_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED);