6 #include <sys/sysmacros.h>
9 #ifdef CONFIG_VALGRIND_DEV
10 #include <valgrind/drd.h>
12 #define DRD_IGNORE_VAR(x) do { } while (0)
18 #include "helper_thread.h"
20 static int last_majdev, last_mindev;
21 static struct disk_util *last_du;
23 static struct fio_sem *disk_util_sem;
25 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
26 int majdev, int mindev, char *path);
28 static void disk_util_free(struct disk_util *du)
33 while (!flist_empty(&du->slaves)) {
34 struct disk_util *slave;
36 slave = flist_first_entry(&du->slaves, struct disk_util, slavelist);
37 flist_del(&slave->slavelist);
41 fio_sem_remove(du->lock);
46 static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
53 dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
55 f = fopen(du->path, "r");
59 p = fgets(line, sizeof(line), f);
65 dprint(FD_DISKUTIL, "%s: %s", du->path, p);
67 ret = sscanf(p, "%"SCNu64" %"SCNu64" %"SCNu64" %"SCNu64" "
68 "%"SCNu64" %"SCNu64" %"SCNu64" %"SCNu64" "
69 "%*u %"SCNu64" %"SCNu64"\n",
70 &dus->s.ios[0], &dus->s.merges[0], &dus->s.sectors[0],
72 &dus->s.ios[1], &dus->s.merges[1], &dus->s.sectors[1],
74 &dus->s.io_ticks, &dus->s.time_in_queue);
76 dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 10);
80 static void update_io_tick_disk(struct disk_util *du)
82 struct disk_util_stat __dus, *dus, *ldus;
87 if (get_io_ticks(du, &__dus))
93 dus->s.sectors[0] += (__dus.s.sectors[0] - ldus->s.sectors[0]);
94 dus->s.sectors[1] += (__dus.s.sectors[1] - ldus->s.sectors[1]);
95 dus->s.ios[0] += (__dus.s.ios[0] - ldus->s.ios[0]);
96 dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]);
97 dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]);
98 dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]);
99 dus->s.ticks[0] += (__dus.s.ticks[0] - ldus->s.ticks[0]);
100 dus->s.ticks[1] += (__dus.s.ticks[1] - ldus->s.ticks[1]);
101 dus->s.io_ticks += (__dus.s.io_ticks - ldus->s.io_ticks);
102 dus->s.time_in_queue += (__dus.s.time_in_queue - ldus->s.time_in_queue);
104 fio_gettime(&t, NULL);
105 dus->s.msec += mtime_since(&du->time, &t);
106 memcpy(&du->time, &t, sizeof(t));
107 memcpy(&ldus->s, &__dus.s, sizeof(__dus.s));
110 int update_io_ticks(void)
112 struct flist_head *entry;
113 struct disk_util *du;
116 dprint(FD_DISKUTIL, "update io ticks\n");
118 fio_sem_down(disk_util_sem);
120 if (!helper_should_exit()) {
121 flist_for_each(entry, &disk_list) {
122 du = flist_entry(entry, struct disk_util, list);
123 update_io_tick_disk(du);
128 fio_sem_up(disk_util_sem);
132 static struct disk_util *disk_util_exists(int major, int minor)
134 struct flist_head *entry;
135 struct disk_util *du;
137 fio_sem_down(disk_util_sem);
139 flist_for_each(entry, &disk_list) {
140 du = flist_entry(entry, struct disk_util, list);
142 if (major == du->major && minor == du->minor) {
143 fio_sem_up(disk_util_sem);
148 fio_sem_up(disk_util_sem);
152 static int get_device_numbers(char *file_name, int *maj, int *min)
156 char tempname[PATH_MAX], *p;
158 if (!lstat(file_name, &st)) {
159 if (S_ISBLK(st.st_mode)) {
160 majdev = major(st.st_rdev);
161 mindev = minor(st.st_rdev);
162 } else if (S_ISCHR(st.st_mode) ||
163 S_ISFIFO(st.st_mode)) {
166 majdev = major(st.st_dev);
167 mindev = minor(st.st_dev);
171 * must be a file, open "." in that path
173 snprintf(tempname, FIO_ARRAY_SIZE(tempname), "%s", file_name);
174 p = dirname(tempname);
176 perror("disk util stat");
180 majdev = major(st.st_dev);
181 mindev = minor(st.st_dev);
190 static int read_block_dev_entry(char *path, int *maj, int *min)
195 f = fopen(path, "r");
201 p = fgets(line, sizeof(line), f);
207 if (sscanf(p, "%u:%u", maj, min) != 2)
213 static void find_add_disk_slaves(struct thread_data *td, char *path,
214 struct disk_util *masterdu)
216 DIR *dirhandle = NULL;
217 struct dirent *dirent = NULL;
218 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
219 struct disk_util *slavedu = NULL;
223 sprintf(slavesdir, "%s/%s", path, "slaves");
224 dirhandle = opendir(slavesdir);
228 while ((dirent = readdir(dirhandle)) != NULL) {
229 if (!strcmp(dirent->d_name, ".") ||
230 !strcmp(dirent->d_name, ".."))
233 nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir,
235 /* Can we always assume that the slaves device entries
236 * are links to the real directories for the slave
239 linklen = readlink(temppath, slavepath, PATH_MAX - 1);
241 perror("readlink() for slave device.");
245 slavepath[linklen] = '\0';
247 nowarn_snprintf(temppath, sizeof(temppath), "%s/%s/dev",
248 slavesdir, slavepath);
249 if (access(temppath, F_OK) != 0)
250 nowarn_snprintf(temppath, sizeof(temppath),
251 "%s/%s/device/dev", slavesdir,
253 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
254 perror("Error getting slave device numbers");
260 * See if this maj,min already exists
262 slavedu = disk_util_exists(majdev, mindev);
266 nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir,
268 __init_per_file_disk_util(td, majdev, mindev, temppath);
269 slavedu = disk_util_exists(majdev, mindev);
271 /* Should probably use an assert here. slavedu should
272 * always be present at this point. */
275 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
282 static struct disk_util *disk_util_add(struct thread_data *td, int majdev,
283 int mindev, char *path)
285 struct disk_util *du, *__du;
286 struct flist_head *entry;
289 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
291 du = smalloc(sizeof(*du));
295 DRD_IGNORE_VAR(du->users);
296 memset(du, 0, sizeof(*du));
297 INIT_FLIST_HEAD(&du->list);
298 l = snprintf(du->path, sizeof(du->path), "%s/stat", path);
299 if (l < 0 || l >= sizeof(du->path)) {
300 log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n",
301 path, sizeof(du->path) - 1);
305 snprintf((char *) du->dus.name, FIO_ARRAY_SIZE(du->dus.name), "%s",
307 du->sysfs_root = strdup(path);
310 INIT_FLIST_HEAD(&du->slavelist);
311 INIT_FLIST_HEAD(&du->slaves);
312 du->lock = fio_sem_init(FIO_SEM_UNLOCKED);
315 fio_sem_down(disk_util_sem);
317 flist_for_each(entry, &disk_list) {
318 __du = flist_entry(entry, struct disk_util, list);
320 dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
322 if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) {
324 fio_sem_up(disk_util_sem);
329 dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
331 fio_gettime(&du->time, NULL);
332 get_io_ticks(du, &du->last_dus);
334 flist_add_tail(&du->list, &disk_list);
335 fio_sem_up(disk_util_sem);
337 find_add_disk_slaves(td, path, du);
341 static int check_dev_match(int majdev, int mindev, char *path)
345 if (read_block_dev_entry(path, &major, &minor))
348 if (majdev == major && mindev == minor)
354 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
365 while ((dir = readdir(D)) != NULL) {
368 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
371 sprintf(full_path, "%s/%s", path, dir->d_name);
373 if (!strcmp(dir->d_name, "dev")) {
374 if (!check_dev_match(majdev, mindev, full_path)) {
381 if (stat(full_path, &st) == -1) {
386 if (lstat(full_path, &st) == -1) {
392 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
395 found = find_block_dir(majdev, mindev, full_path, 0);
397 strcpy(path, full_path);
406 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
407 int majdev, int mindev,
415 * If there's a ../queue/ directory there, we are inside a partition.
416 * Check if that is the case and jump back. For loop/md/dm etc we
417 * are already in the right spot.
419 sprintf(tmp, "%s/../queue", path);
420 if (!stat(tmp, &st)) {
422 sprintf(tmp, "%s/queue", p);
423 if (stat(tmp, &st)) {
424 log_err("unknown sysfs layout\n");
427 snprintf(tmp, FIO_ARRAY_SIZE(tmp), "%s", p);
428 sprintf(path, "%s", tmp);
431 return disk_util_add(td, majdev, mindev, path);
434 static struct disk_util *init_per_file_disk_util(struct thread_data *td,
439 struct disk_util *du;
442 if (get_device_numbers(filename, &majdev, &mindev))
445 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
448 du = disk_util_exists(majdev, mindev);
453 * for an fs without a device, we will repeatedly stat through
454 * sysfs which can take oodles of time for thousands of files. so
455 * cache the last lookup and compare with that before going through
458 if (mindev == last_mindev && majdev == last_majdev)
461 last_mindev = mindev;
462 last_majdev = majdev;
464 sprintf(foo, "/sys/block");
465 if (!find_block_dir(majdev, mindev, foo, 1))
468 return __init_per_file_disk_util(td, majdev, mindev, foo);
471 static struct disk_util *__init_disk_util(struct thread_data *td,
474 return init_per_file_disk_util(td, f->file_name);
477 void init_disk_util(struct thread_data *td)
482 if (!td->o.do_disk_util ||
483 td_ioengine_flagged(td, FIO_DISKLESSIO | FIO_NODISKUTIL))
486 for_each_file(td, f, i)
487 f->du = __init_disk_util(td, f);
490 void disk_util_prune_entries(void)
492 fio_sem_down(disk_util_sem);
494 while (!flist_empty(&disk_list)) {
495 struct disk_util *du;
497 du = flist_first_entry(&disk_list, struct disk_util, list);
498 flist_del(&du->list);
502 last_majdev = last_mindev = -1;
503 fio_sem_up(disk_util_sem);
504 fio_sem_remove(disk_util_sem);
507 void setup_disk_util(void)
509 disk_util_sem = fio_sem_init(FIO_SEM_UNLOCKED);