14 static int last_majdev, last_mindev;
15 static struct disk_util *last_du;
17 static struct flist_head disk_list = FLIST_HEAD_INIT(disk_list);
19 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
20 int majdev, int mindev, char *path);
22 static void disk_util_free(struct disk_util *du)
27 fio_mutex_remove(du->lock);
32 static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
40 dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
42 f = fopen(du->path, "r");
46 p = fgets(line, sizeof(line), f);
52 dprint(FD_DISKUTIL, "%s: %s", du->path, p);
54 ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0],
55 &dus->merges[0], &dus->sectors[0],
56 &dus->ticks[0], &dus->ios[1],
57 &dus->merges[1], &dus->sectors[1],
58 &dus->ticks[1], &in_flight,
59 &dus->io_ticks, &dus->time_in_queue);
61 dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
65 static void update_io_tick_disk(struct disk_util *du)
67 struct disk_util_stat __dus, *dus, *ldus;
70 if (get_io_ticks(du, &__dus))
78 dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
79 dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
80 dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
81 dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
82 dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
83 dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
84 dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
85 dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
86 dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
87 dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
89 fio_gettime(&t, NULL);
90 du->msec += mtime_since(&du->time, &t);
91 memcpy(&du->time, &t, sizeof(t));
92 memcpy(ldus, &__dus, sizeof(__dus));
95 void update_io_ticks(void)
97 struct flist_head *entry;
100 dprint(FD_DISKUTIL, "update io ticks\n");
102 flist_for_each(entry, &disk_list) {
103 du = flist_entry(entry, struct disk_util, list);
104 update_io_tick_disk(du);
108 static struct disk_util *disk_util_exists(int major, int minor)
110 struct flist_head *entry;
111 struct disk_util *du;
113 flist_for_each(entry, &disk_list) {
114 du = flist_entry(entry, struct disk_util, list);
116 if (major == du->major && minor == du->minor)
123 static int get_device_numbers(char *file_name, int *maj, int *min)
127 char tempname[PATH_MAX], *p;
129 if (!lstat(file_name, &st)) {
130 if (S_ISBLK(st.st_mode)) {
131 majdev = major(st.st_rdev);
132 mindev = minor(st.st_rdev);
133 } else if (S_ISCHR(st.st_mode)) {
134 majdev = major(st.st_rdev);
135 mindev = minor(st.st_rdev);
136 if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
138 } else if (S_ISFIFO(st.st_mode))
141 majdev = major(st.st_dev);
142 mindev = minor(st.st_dev);
146 * must be a file, open "." in that path
148 strncpy(tempname, file_name, PATH_MAX - 1);
149 p = dirname(tempname);
151 perror("disk util stat");
155 majdev = major(st.st_dev);
156 mindev = minor(st.st_dev);
165 static int read_block_dev_entry(char *path, int *maj, int *min)
170 f = fopen(path, "r");
176 p = fgets(line, sizeof(line), f);
182 if (sscanf(p, "%u:%u", maj, min) != 2)
188 static void find_add_disk_slaves(struct thread_data *td, char *path,
189 struct disk_util *masterdu)
191 DIR *dirhandle = NULL;
192 struct dirent *dirent = NULL;
193 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
194 struct disk_util *slavedu = NULL;
198 sprintf(slavesdir, "%s/%s", path, "slaves");
199 dirhandle = opendir(slavesdir);
203 while ((dirent = readdir(dirhandle)) != NULL) {
204 if (!strcmp(dirent->d_name, ".") ||
205 !strcmp(dirent->d_name, ".."))
208 sprintf(temppath, "%s/%s", slavesdir, dirent->d_name);
209 /* Can we always assume that the slaves device entries
210 * are links to the real directories for the slave
213 linklen = readlink(temppath, slavepath, PATH_MAX - 0);
215 perror("readlink() for slave device.");
218 slavepath[linklen] = '\0';
220 sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
221 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
222 perror("Error getting slave device numbers.");
227 * See if this maj,min already exists
229 slavedu = disk_util_exists(majdev, mindev);
233 sprintf(temppath, "%s/%s", slavesdir, slavepath);
234 __init_per_file_disk_util(td, majdev, mindev, temppath);
235 slavedu = disk_util_exists(majdev, mindev);
237 /* Should probably use an assert here. slavedu should
238 * always be present at this point. */
240 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
246 static struct disk_util *disk_util_add(struct thread_data * td, int majdev,
247 int mindev, char *path)
249 struct disk_util *du, *__du;
250 struct flist_head *entry;
252 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
254 du = smalloc(sizeof(*du));
255 memset(du, 0, sizeof(*du));
256 INIT_FLIST_HEAD(&du->list);
257 sprintf(du->path, "%s/stat", path);
258 du->name = smalloc_strdup(basename(path));
259 du->sysfs_root = path;
262 INIT_FLIST_HEAD(&du->slavelist);
263 INIT_FLIST_HEAD(&du->slaves);
264 du->lock = fio_mutex_init(1);
267 flist_for_each(entry, &disk_list) {
268 __du = flist_entry(entry, struct disk_util, list);
270 dprint(FD_DISKUTIL, "found %s in list\n", __du->name);
272 if (!strcmp(du->name, __du->name)) {
278 dprint(FD_DISKUTIL, "add %s to list\n", du->name);
280 fio_gettime(&du->time, NULL);
281 get_io_ticks(du, &du->last_dus);
283 flist_add_tail(&du->list, &disk_list);
284 find_add_disk_slaves(td, path, du);
288 static int check_dev_match(int majdev, int mindev, char *path)
292 if (read_block_dev_entry(path, &major, &minor))
295 if (majdev == major && mindev == minor)
301 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
312 while ((dir = readdir(D)) != NULL) {
315 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
318 sprintf(full_path, "%s/%s", path, dir->d_name);
320 if (!strcmp(dir->d_name, "dev")) {
321 if (!check_dev_match(majdev, mindev, full_path)) {
328 if (stat(full_path, &st) == -1) {
333 if (lstat(full_path, &st) == -1) {
339 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
342 found = find_block_dir(majdev, mindev, full_path, 0);
344 strcpy(path, full_path);
353 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
354 int majdev, int mindev,
362 * If there's a ../queue/ directory there, we are inside a partition.
363 * Check if that is the case and jump back. For loop/md/dm etc we
364 * are already in the right spot.
366 sprintf(tmp, "%s/../queue", path);
367 if (!stat(tmp, &st)) {
369 sprintf(tmp, "%s/queue", p);
370 if (stat(tmp, &st)) {
371 log_err("unknown sysfs layout\n");
374 strncpy(tmp, p, PATH_MAX - 1);
375 sprintf(path, "%s", tmp);
378 if (td->o.ioscheduler && !td->sysfs_root)
379 td->sysfs_root = strdup(path);
381 return disk_util_add(td, majdev, mindev, path);
384 static struct disk_util *init_per_file_disk_util(struct thread_data *td,
389 struct disk_util *du;
392 if (get_device_numbers(filename, &majdev, &mindev))
395 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
398 du = disk_util_exists(majdev, mindev);
400 if (td->o.ioscheduler && !td->sysfs_root)
401 td->sysfs_root = strdup(du->sysfs_root);
407 * for an fs without a device, we will repeatedly stat through
408 * sysfs which can take oodles of time for thousands of files. so
409 * cache the last lookup and compare with that before going through
412 if (mindev == last_mindev && majdev == last_majdev)
415 last_mindev = mindev;
416 last_majdev = majdev;
418 sprintf(foo, "/sys/block");
419 if (!find_block_dir(majdev, mindev, foo, 1))
422 return __init_per_file_disk_util(td, majdev, mindev, foo);
425 static struct disk_util *__init_disk_util(struct thread_data *td,
428 return init_per_file_disk_util(td, f->file_name);
431 void init_disk_util(struct thread_data *td)
436 if (!td->o.do_disk_util ||
437 (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL)))
440 for_each_file(td, f, i)
441 f->du = __init_disk_util(td, f);
444 static void aggregate_slaves_stats(struct disk_util *masterdu)
446 struct disk_util_stat *dus;
447 struct flist_head *entry;
448 struct disk_util *slavedu;
449 double util, max_util = 0;
452 unsigned merges[2] = { 0, };
453 unsigned ticks[2] = { 0, };
454 unsigned time_in_queue = { 0, };
455 unsigned long long sectors[2] = { 0, };
456 unsigned ios[2] = { 0, };
458 flist_for_each(entry, &masterdu->slaves) {
459 slavedu = flist_entry(entry, struct disk_util, slavelist);
461 ios[0] += dus->ios[0];
462 ios[1] += dus->ios[1];
463 merges[0] += dus->merges[0];
464 merges[1] += dus->merges[1];
465 sectors[0] += dus->sectors[0];
466 sectors[1] += dus->sectors[1];
467 ticks[0] += dus->ticks[0];
468 ticks[1] += dus->ticks[1];
469 time_in_queue += dus->time_in_queue;
472 util = (double) (100 * dus->io_ticks / (double) slavedu->msec);
473 /* System utilization is the utilization of the
474 * component with the highest utilization.
481 if (max_util > 100.0)
484 log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
485 " aggrin_queue=%u, aggrutil=%3.2f%%",
486 ios[0]/slavecount, ios[1]/slavecount,
487 merges[0]/slavecount, merges[1]/slavecount,
488 ticks[0]/slavecount, ticks[1]/slavecount,
489 time_in_queue/slavecount, max_util);
493 void show_disk_util(void)
495 struct disk_util_stat *dus;
496 struct flist_head *entry, *next;
497 struct disk_util *du;
500 if (flist_empty(&disk_list))
503 log_info("\nDisk stats (read/write):\n");
505 flist_for_each(entry, &disk_list) {
506 du = flist_entry(entry, struct disk_util, list);
509 util = (double) 100 * du->dus.io_ticks / (double) du->msec;
513 /* If this node is the slave of a master device, as
514 * happens in case of software RAIDs, inward-indent
515 * this stats line to reflect a master-slave
516 * relationship. Because the master device gets added
517 * before the slave devices, we can safely assume that
518 * the master's stats line has been displayed in a
519 * previous iteration of this loop.
521 if (!flist_empty(&du->slavelist))
524 log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
525 "in_queue=%u, util=%3.2f%%", du->name,
526 dus->ios[0], dus->ios[1],
527 dus->merges[0], dus->merges[1],
528 dus->ticks[0], dus->ticks[1],
529 dus->time_in_queue, util);
531 /* If the device has slaves, aggregate the stats for
532 * those slave devices also.
534 if (!flist_empty(&du->slaves))
535 aggregate_slaves_stats(du);
543 flist_for_each_safe(entry, next, &disk_list) {
545 du = flist_entry(entry, struct disk_util, list);