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 flist_for_each(entry, &disk_list) {
136 du = flist_entry(entry, struct disk_util, list);
138 if (major == du->major && minor == du->minor)
145 static int get_device_numbers(char *file_name, int *maj, int *min)
149 char tempname[PATH_MAX], *p;
151 if (!lstat(file_name, &st)) {
152 if (S_ISBLK(st.st_mode)) {
153 majdev = major(st.st_rdev);
154 mindev = minor(st.st_rdev);
155 } else if (S_ISCHR(st.st_mode)) {
156 majdev = major(st.st_rdev);
157 mindev = minor(st.st_rdev);
158 if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
160 } else if (S_ISFIFO(st.st_mode))
163 majdev = major(st.st_dev);
164 mindev = minor(st.st_dev);
168 * must be a file, open "." in that path
170 strncpy(tempname, file_name, PATH_MAX - 1);
171 p = dirname(tempname);
173 perror("disk util stat");
177 majdev = major(st.st_dev);
178 mindev = minor(st.st_dev);
187 static int read_block_dev_entry(char *path, int *maj, int *min)
192 f = fopen(path, "r");
198 p = fgets(line, sizeof(line), f);
204 if (sscanf(p, "%u:%u", maj, min) != 2)
210 static void find_add_disk_slaves(struct thread_data *td, char *path,
211 struct disk_util *masterdu)
213 DIR *dirhandle = NULL;
214 struct dirent *dirent = NULL;
215 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
216 struct disk_util *slavedu = NULL;
220 sprintf(slavesdir, "%s/%s", path, "slaves");
221 dirhandle = opendir(slavesdir);
225 while ((dirent = readdir(dirhandle)) != NULL) {
226 if (!strcmp(dirent->d_name, ".") ||
227 !strcmp(dirent->d_name, ".."))
230 sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, dirent->d_name);
231 /* Can we always assume that the slaves device entries
232 * are links to the real directories for the slave
235 linklen = readlink(temppath, slavepath, PATH_MAX - 0);
237 perror("readlink() for slave device.");
240 slavepath[linklen] = '\0';
242 sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
243 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
244 perror("Error getting slave device numbers.");
249 * See if this maj,min already exists
251 slavedu = disk_util_exists(majdev, mindev);
255 sprintf(temppath, "%s%s%s", slavesdir, FIO_OS_PATH_SEPARATOR, slavepath);
256 __init_per_file_disk_util(td, majdev, mindev, temppath);
257 slavedu = disk_util_exists(majdev, mindev);
259 /* Should probably use an assert here. slavedu should
260 * always be present at this point. */
263 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
270 static struct disk_util *disk_util_add(struct thread_data *td, int majdev,
271 int mindev, char *path)
273 struct disk_util *du, *__du;
274 struct flist_head *entry;
276 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
278 du = smalloc(sizeof(*du));
279 memset(du, 0, sizeof(*du));
280 INIT_FLIST_HEAD(&du->list);
281 sprintf(du->path, "%s/stat", path);
282 strncpy((char *) du->dus.name, basename(path), FIO_DU_NAME_SZ);
283 du->sysfs_root = path;
286 INIT_FLIST_HEAD(&du->slavelist);
287 INIT_FLIST_HEAD(&du->slaves);
288 du->lock = fio_mutex_init(1);
291 fio_mutex_down(disk_util_mutex);
293 flist_for_each(entry, &disk_list) {
294 __du = flist_entry(entry, struct disk_util, list);
296 dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
298 if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) {
300 fio_mutex_up(disk_util_mutex);
305 dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
307 fio_gettime(&du->time, NULL);
308 get_io_ticks(du, &du->last_dus);
310 flist_add_tail(&du->list, &disk_list);
311 find_add_disk_slaves(td, path, du);
312 fio_mutex_up(disk_util_mutex);
316 static int check_dev_match(int majdev, int mindev, char *path)
320 if (read_block_dev_entry(path, &major, &minor))
323 if (majdev == major && mindev == minor)
329 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
340 while ((dir = readdir(D)) != NULL) {
343 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
346 sprintf(full_path, "%s%s%s", path, FIO_OS_PATH_SEPARATOR, dir->d_name);
348 if (!strcmp(dir->d_name, "dev")) {
349 if (!check_dev_match(majdev, mindev, full_path)) {
356 if (stat(full_path, &st) == -1) {
361 if (lstat(full_path, &st) == -1) {
367 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
370 found = find_block_dir(majdev, mindev, full_path, 0);
372 strcpy(path, full_path);
381 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
382 int majdev, int mindev,
390 * If there's a ../queue/ directory there, we are inside a partition.
391 * Check if that is the case and jump back. For loop/md/dm etc we
392 * are already in the right spot.
394 sprintf(tmp, "%s/../queue", path);
395 if (!stat(tmp, &st)) {
397 sprintf(tmp, "%s/queue", p);
398 if (stat(tmp, &st)) {
399 log_err("unknown sysfs layout\n");
402 strncpy(tmp, p, PATH_MAX - 1);
403 sprintf(path, "%s", tmp);
406 if (td->o.ioscheduler && !td->sysfs_root)
407 td->sysfs_root = strdup(path);
409 return disk_util_add(td, majdev, mindev, path);
412 static struct disk_util *init_per_file_disk_util(struct thread_data *td,
417 struct disk_util *du;
420 if (get_device_numbers(filename, &majdev, &mindev))
423 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
426 du = disk_util_exists(majdev, mindev);
428 if (td->o.ioscheduler && !td->sysfs_root)
429 td->sysfs_root = strdup(du->sysfs_root);
435 * for an fs without a device, we will repeatedly stat through
436 * sysfs which can take oodles of time for thousands of files. so
437 * cache the last lookup and compare with that before going through
440 if (mindev == last_mindev && majdev == last_majdev)
443 last_mindev = mindev;
444 last_majdev = majdev;
446 sprintf(foo, "/sys/block");
447 if (!find_block_dir(majdev, mindev, foo, 1))
450 return __init_per_file_disk_util(td, majdev, mindev, foo);
453 static struct disk_util *__init_disk_util(struct thread_data *td,
456 return init_per_file_disk_util(td, f->file_name);
459 void init_disk_util(struct thread_data *td)
464 if (!td->o.do_disk_util ||
465 (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL)))
468 for_each_file(td, f, i)
469 f->du = __init_disk_util(td, f);
472 static void show_agg_stats(struct disk_util_agg *agg, int terse)
474 if (!agg->slavecount)
478 log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
479 " aggrin_queue=%u, aggrutil=%3.2f%%",
480 agg->ios[0] / agg->slavecount,
481 agg->ios[1] / agg->slavecount,
482 agg->merges[0] / agg->slavecount,
483 agg->merges[1] / agg->slavecount,
484 agg->ticks[0] / agg->slavecount,
485 agg->ticks[1] / agg->slavecount,
486 agg->time_in_queue / agg->slavecount,
489 log_info(";slaves;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
490 agg->ios[0] / agg->slavecount,
491 agg->ios[1] / agg->slavecount,
492 agg->merges[0] / agg->slavecount,
493 agg->merges[1] / agg->slavecount,
494 agg->ticks[0] / agg->slavecount,
495 agg->ticks[1] / agg->slavecount,
496 agg->time_in_queue / agg->slavecount,
501 static void aggregate_slaves_stats(struct disk_util *masterdu)
503 struct disk_util_agg *agg = &masterdu->agg;
504 struct disk_util_stat *dus;
505 struct flist_head *entry;
506 struct disk_util *slavedu;
509 flist_for_each(entry, &masterdu->slaves) {
510 slavedu = flist_entry(entry, struct disk_util, slavelist);
512 agg->ios[0] += dus->ios[0];
513 agg->ios[1] += dus->ios[1];
514 agg->merges[0] += dus->merges[0];
515 agg->merges[1] += dus->merges[1];
516 agg->sectors[0] += dus->sectors[0];
517 agg->sectors[1] += dus->sectors[1];
518 agg->ticks[0] += dus->ticks[0];
519 agg->ticks[1] += dus->ticks[1];
520 agg->time_in_queue += dus->time_in_queue;
523 util = (double) (100 * dus->io_ticks / (double) slavedu->dus.msec);
524 /* System utilization is the utilization of the
525 * component with the highest utilization.
527 if (util > agg->max_util.u.f)
528 agg->max_util.u.f = util;
532 if (agg->max_util.u.f > 100.0)
533 agg->max_util.u.f = 100.0;
536 void free_disk_util(void)
538 struct disk_util *du;
541 wait_for_disk_thread_exit();
543 fio_mutex_down(disk_util_mutex);
545 while (!flist_empty(&disk_list)) {
546 du = flist_entry(disk_list.next, struct disk_util, list);
547 flist_del(&du->list);
551 last_majdev = last_mindev = -1;
552 fio_mutex_up(disk_util_mutex);
553 fio_mutex_remove(disk_util_mutex);
556 void print_disk_util(struct disk_util_stat *dus, struct disk_util_agg *agg,
562 util = (double) 100 * dus->io_ticks / (double) dus->msec;
570 log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
571 "in_queue=%u, util=%3.2f%%", dus->name,
572 dus->ios[0], dus->ios[1],
573 dus->merges[0], dus->merges[1],
574 dus->ticks[0], dus->ticks[1],
575 dus->time_in_queue, util);
577 log_info(";%s;%u;%u;%u;%u;%u;%u;%u;%3.2f%%",
578 dus->name, 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);
585 * If the device has slaves, aggregate the stats for
586 * those slave devices also.
588 show_agg_stats(agg, terse);
594 void show_disk_util(int terse)
596 struct flist_head *entry;
597 struct disk_util *du;
599 fio_mutex_down(disk_util_mutex);
601 if (flist_empty(&disk_list)) {
602 fio_mutex_up(disk_util_mutex);
607 log_info("\nDisk stats (read/write):\n");
609 flist_for_each(entry, &disk_list) {
610 du = flist_entry(entry, struct disk_util, list);
612 aggregate_slaves_stats(du);
613 print_disk_util(&du->dus, &du->agg, terse);
616 fio_mutex_up(disk_util_mutex);
619 void setup_disk_util(void)
621 disk_util_mutex = fio_mutex_init(1);