12 static int last_majdev, last_mindev;
14 static struct flist_head disk_list = FLIST_HEAD_INIT(disk_list);
16 static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
24 dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
26 f = fopen(du->path, "r");
30 p = fgets(line, sizeof(line), f);
36 dprint(FD_DISKUTIL, "%s: %s", du->path, p);
38 ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0],
39 &dus->merges[0], &dus->sectors[0],
40 &dus->ticks[0], &dus->ios[1],
41 &dus->merges[1], &dus->sectors[1],
42 &dus->ticks[1], &in_flight,
43 &dus->io_ticks, &dus->time_in_queue);
45 dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
49 static void update_io_tick_disk(struct disk_util *du)
51 struct disk_util_stat __dus, *dus, *ldus;
54 if (get_io_ticks(du, &__dus))
60 dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
61 dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
62 dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
63 dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
64 dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
65 dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
66 dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
67 dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
68 dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
69 dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
71 fio_gettime(&t, NULL);
72 du->msec += mtime_since(&du->time, &t);
73 memcpy(&du->time, &t, sizeof(t));
74 memcpy(ldus, &__dus, sizeof(__dus));
77 void update_io_ticks(void)
79 struct flist_head *entry;
82 dprint(FD_DISKUTIL, "update io ticks\n");
84 flist_for_each(entry, &disk_list) {
85 du = flist_entry(entry, struct disk_util, list);
86 update_io_tick_disk(du);
90 static struct disk_util *disk_util_exists(int major, int minor)
92 struct flist_head *entry;
95 flist_for_each(entry, &disk_list) {
96 du = flist_entry(entry, struct disk_util, list);
98 if (major == du->major && minor == du->minor)
105 static int get_device_numbers(char *file_name, int *maj, int *min)
109 char tempname[PATH_MAX], *p;
111 if (!lstat(file_name, &st)) {
112 if (S_ISBLK(st.st_mode)) {
113 majdev = major(st.st_rdev);
114 mindev = minor(st.st_rdev);
115 } else if (S_ISCHR(st.st_mode)) {
116 majdev = major(st.st_rdev);
117 mindev = minor(st.st_rdev);
118 if (fio_lookup_raw(st.st_rdev, &majdev, &mindev))
120 } else if (S_ISFIFO(st.st_mode))
123 majdev = major(st.st_dev);
124 mindev = minor(st.st_dev);
128 * must be a file, open "." in that path
130 strncpy(tempname, file_name, PATH_MAX - 1);
131 p = dirname(tempname);
133 perror("disk util stat");
137 majdev = major(st.st_dev);
138 mindev = minor(st.st_dev);
147 static int read_block_dev_entry(char *path, int *maj, int *min)
152 f = fopen(path, "r");
158 p = fgets(line, sizeof(line), f);
164 if (sscanf(p, "%u:%u", maj, min) != 2)
171 static void __init_per_file_disk_util(struct thread_data *td, int majdev,
172 int mindev, char * path);
174 static void find_add_disk_slaves(struct thread_data *td, char *path,
175 struct disk_util *masterdu)
177 DIR *dirhandle = NULL;
178 struct dirent *dirent = NULL;
179 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
180 struct disk_util *slavedu = NULL;
184 sprintf(slavesdir, "%s/%s",path, "slaves");
185 dirhandle = opendir(slavesdir);
189 while ((dirent = readdir(dirhandle)) != NULL) {
190 if (!strcmp(dirent->d_name, ".") || !strcmp(dirent->d_name, ".."))
193 sprintf(temppath, "%s/%s", slavesdir, dirent->d_name);
194 /* Can we always assume that the slaves device entries
195 * are links to the real directories for the slave
198 if ((linklen = readlink(temppath, slavepath, PATH_MAX-1)) < 0) {
199 perror("readlink() for slave device.");
202 slavepath[linklen]='\0';
204 sprintf(temppath, "%s/%s/dev", slavesdir, slavepath);
205 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
206 perror("Error getting slave device numbers.");
210 sprintf(temppath, "%s/%s", slavesdir, slavepath);
211 __init_per_file_disk_util(td, majdev, mindev, temppath);
212 slavedu = disk_util_exists(majdev, mindev);
214 /* Should probably use an assert here. slavedu should
215 * always be present at this point. */
217 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
226 static void disk_util_add(struct thread_data * td, int majdev, int mindev,
229 struct disk_util *du, *__du;
230 struct flist_head *entry;
232 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
234 du = malloc(sizeof(*du));
235 memset(du, 0, sizeof(*du));
236 INIT_FLIST_HEAD(&du->list);
237 sprintf(du->path, "%s/stat", path);
238 du->name = strdup(basename(path));
239 du->sysfs_root = path;
242 INIT_FLIST_HEAD(&du->slavelist);
243 INIT_FLIST_HEAD(&du->slaves);
245 flist_for_each(entry, &disk_list) {
246 __du = flist_entry(entry, struct disk_util, list);
248 dprint(FD_DISKUTIL, "found %s in list\n", __du->name);
250 if (!strcmp(du->name, __du->name)) {
257 dprint(FD_DISKUTIL, "add %s to list\n", du->name);
259 fio_gettime(&du->time, NULL);
260 get_io_ticks(du, &du->last_dus);
262 flist_add_tail(&du->list, &disk_list);
263 find_add_disk_slaves(td, path, du);
267 static int check_dev_match(int majdev, int mindev, char *path)
271 if (read_block_dev_entry(path, &major, &minor))
274 if (majdev == major && mindev == minor) {
281 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
292 while ((dir = readdir(D)) != NULL) {
295 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
298 sprintf(full_path, "%s/%s", path, dir->d_name);
300 if (!strcmp(dir->d_name, "dev")) {
301 if (!check_dev_match(majdev, mindev, full_path)) {
308 if (stat(full_path, &st) == -1) {
313 if (lstat(full_path, &st) == -1) {
319 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
322 found = find_block_dir(majdev, mindev, full_path, 0);
324 strcpy(path, full_path);
334 static void __init_per_file_disk_util(struct thread_data *td, int majdev,
335 int mindev, char * path)
342 * If there's a ../queue/ directory there, we are inside a partition.
343 * Check if that is the case and jump back. For loop/md/dm etc we
344 * are already in the right spot.
346 sprintf(tmp, "%s/../queue", path);
347 if (!stat(tmp, &st)) {
349 sprintf(tmp, "%s/queue", p);
350 if (stat(tmp, &st)) {
351 log_err("unknown sysfs layout\n");
354 strncpy(tmp, p, PATH_MAX - 1);
355 sprintf(path, "%s", tmp);
358 if (td->o.ioscheduler && !td->sysfs_root)
359 td->sysfs_root = strdup(path);
361 disk_util_add(td, majdev, mindev, path);
366 static void init_per_file_disk_util(struct thread_data *td, char * filename)
370 struct disk_util *du;
373 if(get_device_numbers(filename, &majdev, &mindev))
376 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
379 du = disk_util_exists(majdev, mindev);
381 if (td->o.ioscheduler && !td->sysfs_root)
382 td->sysfs_root = strdup(du->sysfs_root);
388 * for an fs without a device, we will repeatedly stat through
389 * sysfs which can take oodles of time for thousands of files. so
390 * cache the last lookup and compare with that before going through
393 if (mindev == last_mindev && majdev == last_majdev)
396 last_mindev = mindev;
397 last_majdev = majdev;
399 sprintf(foo, "/sys/block");
400 if (!find_block_dir(majdev, mindev, foo, 1))
403 __init_per_file_disk_util(td, majdev, mindev, foo);
407 static void __init_disk_util(struct thread_data *td, struct fio_file *f)
409 return init_per_file_disk_util(td, f->file_name);
412 void init_disk_util(struct thread_data *td)
417 if (!td->o.do_disk_util ||
418 (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL)))
421 for_each_file(td, f, i)
422 __init_disk_util(td, f);
425 static void aggregate_slaves_stats(struct disk_util *masterdu)
427 struct disk_util_stat *dus;
428 struct flist_head *entry;
429 struct disk_util *slavedu;
430 double util, max_util = 0;
433 unsigned merges[2] = { 0, };
434 unsigned ticks[2] = { 0, };
435 unsigned time_in_queue = { 0, };
436 unsigned long long sectors[2] = { 0, };
437 unsigned ios[2] = { 0, };
439 flist_for_each(entry, &masterdu->slaves) {
440 slavedu = flist_entry(entry, struct disk_util, slavelist);
442 ios[0] += dus->ios[0];
443 ios[1] += dus->ios[1];
444 merges[0] += dus->merges[0];
445 merges[1] += dus->merges[1];
446 sectors[0] += dus->sectors[0];
447 sectors[1] += dus->sectors[1];
448 ticks[0] += dus->ticks[0];
449 ticks[1] += dus->ticks[1];
450 time_in_queue += dus->time_in_queue;
453 util = (double) (100 * dus->io_ticks / (double) slavedu->msec);
454 /* System utilization is the utilization of the
455 * component with the highest utilization.
462 if (max_util > 100.0)
465 log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u,"
466 " aggrin_queue=%u, aggrutil=%3.2f%%",
467 ios[0]/slavecount, ios[1]/slavecount,
468 merges[0]/slavecount, merges[1]/slavecount,
469 ticks[0]/slavecount, ticks[1]/slavecount,
470 time_in_queue/slavecount, max_util);
476 void show_disk_util(void)
478 struct disk_util_stat *dus;
479 struct flist_head *entry, *next;
480 struct disk_util *du;
483 if (flist_empty(&disk_list))
486 log_info("\nDisk stats (read/write):\n");
488 flist_for_each(entry, &disk_list) {
489 du = flist_entry(entry, struct disk_util, list);
492 util = (double) 100 * du->dus.io_ticks / (double) du->msec;
496 /* If this node is the slave of a master device, as
497 * happens in case of software RAIDs, inward-indent
498 * this stats line to reflect a master-slave
499 * relationship. Because the master device gets added
500 * before the slave devices, we can safely assume that
501 * the master's stats line has been displayed in a
502 * previous iteration of this loop.
504 if(!flist_empty(&du->slavelist))
507 log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, "
508 "in_queue=%u, util=%3.2f%%", du->name,
509 dus->ios[0], dus->ios[1],
510 dus->merges[0], dus->merges[1],
511 dus->ticks[0], dus->ticks[1],
512 dus->time_in_queue, util);
514 /* If the device has slaves, aggregate the stats for
515 * those slave devices also.
517 if(!flist_empty(&du->slaves))
518 aggregate_slaves_stats(du);
526 flist_for_each_safe(entry, next, &disk_list) {
528 du = flist_entry(entry, struct disk_util, list);