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 while (!flist_empty(&du->slaves)) {
28 struct disk_util *slave;
30 slave = flist_entry(du->slaves.next, struct disk_util, slavelist);
31 flist_del(&slave->slavelist);
35 fio_mutex_remove(du->lock);
40 static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
48 dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
50 f = fopen(du->path, "r");
54 p = fgets(line, sizeof(line), f);
60 dprint(FD_DISKUTIL, "%s: %s", du->path, p);
62 ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0],
63 &dus->merges[0], &dus->sectors[0],
64 &dus->ticks[0], &dus->ios[1],
65 &dus->merges[1], &dus->sectors[1],
66 &dus->ticks[1], &in_flight,
67 &dus->io_ticks, &dus->time_in_queue);
69 dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
73 static void update_io_tick_disk(struct disk_util *du)
75 struct disk_util_stat __dus, *dus, *ldus;
80 if (get_io_ticks(du, &__dus))
86 dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
87 dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
88 dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
89 dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
90 dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
91 dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
92 dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
93 dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
94 dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
95 dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
97 fio_gettime(&t, NULL);
98 du->msec += mtime_since(&du->time, &t);
99 memcpy(&du->time, &t, sizeof(t));
100 memcpy(ldus, &__dus, sizeof(__dus));
103 void update_io_ticks(void)
105 struct flist_head *entry;
106 struct disk_util *du;
108 dprint(FD_DISKUTIL, "update io ticks\n");
110 flist_for_each(entry, &disk_list) {
111 du = flist_entry(entry, struct disk_util, list);
112 update_io_tick_disk(du);
116 static struct disk_util *disk_util_exists(int major, int minor)
118 struct flist_head *entry;
119 struct disk_util *du;
121 flist_for_each(entry, &disk_list) {
122 du = flist_entry(entry, struct disk_util, list);
124 if (major == du->major && minor == du->minor)
131 static int get_device_numbers(char *file_name, int *maj, int *min)
135 char tempname[PATH_MAX], *p;
137 if (!lstat(file_name, &st)) {
138 if (S_ISBLK(st.st_mode)) {
139 majdev = major(st.st_rdev);
140 mindev = minor(st.st_rdev);
141 } else if (S_ISCHR(st.st_mode)) {
142 majdev = major(st.st_rdev);
143 mindev = minor(st.st_rdev);
144 if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
146 } else if (S_ISFIFO(st.st_mode))
149 majdev = major(st.st_dev);
150 mindev = minor(st.st_dev);
154 * must be a file, open "." in that path
156 strncpy(tempname, file_name, PATH_MAX - 1);
157 p = dirname(tempname);
159 perror("disk util stat");
163 majdev = major(st.st_dev);
164 mindev = minor(st.st_dev);
173 static int read_block_dev_entry(char *path, int *maj, int *min)
178 f = fopen(path, "r");
184 p = fgets(line, sizeof(line), f);
190 if (sscanf(p, "%u:%u", maj, min) != 2)
196 static void find_add_disk_slaves(struct thread_data *td, char *path,
197 struct disk_util *masterdu)
199 DIR *dirhandle = NULL;
200 struct dirent *dirent = NULL;
201 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
202 struct disk_util *slavedu = NULL;
206 sprintf(slavesdir, "%s/%s", path, "slaves");
207 dirhandle = opendir(slavesdir);
211 while ((dirent = readdir(dirhandle)) != NULL) {
212 if (!strcmp(dirent->d_name, ".") ||
213 !strcmp(dirent->d_name, ".."))
216 sprintf(temppath, "%s/%s", slavesdir, dirent->d_name);
217 /* Can we always assume that the slaves device entries
218 * are links to the real directories for the slave
221 linklen = readlink(temppath, slavepath, PATH_MAX - 0);
223 perror("readlink() for slave device.");
226 slavepath[linklen] = '\0';
228 sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
229 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
230 perror("Error getting slave device numbers.");
235 * See if this maj,min already exists
237 slavedu = disk_util_exists(majdev, mindev);
241 sprintf(temppath, "%s/%s", slavesdir, slavepath);
242 __init_per_file_disk_util(td, majdev, mindev, temppath);
243 slavedu = disk_util_exists(majdev, mindev);
245 /* Should probably use an assert here. slavedu should
246 * always be present at this point. */
249 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
256 static struct disk_util *disk_util_add(struct thread_data * td, int majdev,
257 int mindev, char *path)
259 struct disk_util *du, *__du;
260 struct flist_head *entry;
262 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
264 du = smalloc(sizeof(*du));
265 memset(du, 0, sizeof(*du));
266 INIT_FLIST_HEAD(&du->list);
267 sprintf(du->path, "%s/stat", path);
268 du->name = smalloc_strdup(basename(path));
269 du->sysfs_root = path;
272 INIT_FLIST_HEAD(&du->slavelist);
273 INIT_FLIST_HEAD(&du->slaves);
274 du->lock = fio_mutex_init(1);
277 flist_for_each(entry, &disk_list) {
278 __du = flist_entry(entry, struct disk_util, list);
280 dprint(FD_DISKUTIL, "found %s in list\n", __du->name);
282 if (!strcmp(du->name, __du->name)) {
288 dprint(FD_DISKUTIL, "add %s to list\n", du->name);
290 fio_gettime(&du->time, NULL);
291 get_io_ticks(du, &du->last_dus);
293 flist_add_tail(&du->list, &disk_list);
294 find_add_disk_slaves(td, path, du);
298 static int check_dev_match(int majdev, int mindev, char *path)
302 if (read_block_dev_entry(path, &major, &minor))
305 if (majdev == major && mindev == minor)
311 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
322 while ((dir = readdir(D)) != NULL) {
325 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
328 sprintf(full_path, "%s/%s", path, dir->d_name);
330 if (!strcmp(dir->d_name, "dev")) {
331 if (!check_dev_match(majdev, mindev, full_path)) {
338 if (stat(full_path, &st) == -1) {
343 if (lstat(full_path, &st) == -1) {
349 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
352 found = find_block_dir(majdev, mindev, full_path, 0);
354 strcpy(path, full_path);
363 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
364 int majdev, int mindev,
372 * If there's a ../queue/ directory there, we are inside a partition.
373 * Check if that is the case and jump back. For loop/md/dm etc we
374 * are already in the right spot.
376 sprintf(tmp, "%s/../queue", path);
377 if (!stat(tmp, &st)) {
379 sprintf(tmp, "%s/queue", p);
380 if (stat(tmp, &st)) {
381 log_err("unknown sysfs layout\n");
384 strncpy(tmp, p, PATH_MAX - 1);
385 sprintf(path, "%s", tmp);
388 if (td->o.ioscheduler && !td->sysfs_root)
389 td->sysfs_root = strdup(path);
391 return disk_util_add(td, majdev, mindev, path);
394 static struct disk_util *init_per_file_disk_util(struct thread_data *td,
399 struct disk_util *du;
402 if (get_device_numbers(filename, &majdev, &mindev))
405 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
408 du = disk_util_exists(majdev, mindev);
410 if (td->o.ioscheduler && !td->sysfs_root)
411 td->sysfs_root = strdup(du->sysfs_root);
417 * for an fs without a device, we will repeatedly stat through
418 * sysfs which can take oodles of time for thousands of files. so
419 * cache the last lookup and compare with that before going through
422 if (mindev == last_mindev && majdev == last_majdev)
425 last_mindev = mindev;
426 last_majdev = majdev;
428 sprintf(foo, "/sys/block");
429 if (!find_block_dir(majdev, mindev, foo, 1))
432 return __init_per_file_disk_util(td, majdev, mindev, foo);
435 static struct disk_util *__init_disk_util(struct thread_data *td,
438 return init_per_file_disk_util(td, f->file_name);
441 void init_disk_util(struct thread_data *td)
446 if (!td->o.do_disk_util ||
447 (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL)))
450 for_each_file(td, f, i)
451 f->du = __init_disk_util(td, f);
454 static void aggregate_slaves_stats(struct disk_util *masterdu)
456 struct disk_util_stat *dus;
457 struct flist_head *entry;
458 struct disk_util *slavedu;
459 double util, max_util = 0;
462 unsigned merges[2] = { 0, };
463 unsigned ticks[2] = { 0, };
464 unsigned time_in_queue = { 0, };
465 unsigned long long sectors[2] = { 0, };
466 unsigned ios[2] = { 0, };
468 flist_for_each(entry, &masterdu->slaves) {
469 slavedu = flist_entry(entry, struct disk_util, slavelist);
471 ios[0] += dus->ios[0];
472 ios[1] += dus->ios[1];
473 merges[0] += dus->merges[0];
474 merges[1] += dus->merges[1];
475 sectors[0] += dus->sectors[0];
476 sectors[1] += dus->sectors[1];
477 ticks[0] += dus->ticks[0];
478 ticks[1] += dus->ticks[1];
479 time_in_queue += dus->time_in_queue;
482 util = (double) (100 * dus->io_ticks / (double) slavedu->msec);
483 /* System utilization is the utilization of the
484 * component with the highest utilization.
491 if (max_util > 100.0)
494 log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
495 " aggrin_queue=%u, aggrutil=%3.2f%%",
496 ios[0]/slavecount, ios[1]/slavecount,
497 merges[0]/slavecount, merges[1]/slavecount,
498 ticks[0]/slavecount, ticks[1]/slavecount,
499 time_in_queue/slavecount, max_util);
503 void show_disk_util(void)
505 struct disk_util_stat *dus;
506 struct flist_head *entry, *next;
507 struct disk_util *du;
510 if (flist_empty(&disk_list))
513 log_info("\nDisk stats (read/write):\n");
515 flist_for_each(entry, &disk_list) {
516 du = flist_entry(entry, struct disk_util, list);
519 util = (double) 100 * du->dus.io_ticks / (double) du->msec;
523 /* If this node is the slave of a master device, as
524 * happens in case of software RAIDs, inward-indent
525 * this stats line to reflect a master-slave
526 * relationship. Because the master device gets added
527 * before the slave devices, we can safely assume that
528 * the master's stats line has been displayed in a
529 * previous iteration of this loop.
531 if (!flist_empty(&du->slavelist))
534 log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
535 "in_queue=%u, util=%3.2f%%", du->name,
536 dus->ios[0], dus->ios[1],
537 dus->merges[0], dus->merges[1],
538 dus->ticks[0], dus->ticks[1],
539 dus->time_in_queue, util);
541 /* If the device has slaves, aggregate the stats for
542 * those slave devices also.
544 if (!flist_empty(&du->slaves))
545 aggregate_slaves_stats(du);
553 flist_for_each_safe(entry, next, &disk_list) {
555 du = flist_entry(entry, struct disk_util, list);