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 uint64_t safe_32bit_diff(uint64_t nval, uint64_t oval)
82 /* Linux kernel prints some of the stat fields as 32-bit integers. It is
83 * possible that the value overflows, but since fio uses unsigned 64-bit
84 * arithmetic in update_io_tick_disk(), it instead results in a huge
85 * bogus value being added to the respective accumulating field. Just
86 * in case Linux starts reporting these metrics as 64-bit values in the
87 * future, check that overflow actually happens around the 32-bit
88 * unsigned boundary; assume overflow only happens once between
91 if (oval <= nval || oval >= (1ull << 32))
94 return (1ull << 32) + nval - oval;
97 static void update_io_tick_disk(struct disk_util *du)
99 struct disk_util_stat __dus, *dus, *ldus;
104 if (get_io_ticks(du, &__dus))
108 ldus = &du->last_dus;
110 dus->s.sectors[0] += (__dus.s.sectors[0] - ldus->s.sectors[0]);
111 dus->s.sectors[1] += (__dus.s.sectors[1] - ldus->s.sectors[1]);
112 dus->s.ios[0] += (__dus.s.ios[0] - ldus->s.ios[0]);
113 dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]);
114 dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]);
115 dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]);
116 dus->s.ticks[0] += safe_32bit_diff(__dus.s.ticks[0], ldus->s.ticks[0]);
117 dus->s.ticks[1] += safe_32bit_diff(__dus.s.ticks[1], ldus->s.ticks[1]);
118 dus->s.io_ticks += safe_32bit_diff(__dus.s.io_ticks, ldus->s.io_ticks);
119 dus->s.time_in_queue +=
120 safe_32bit_diff(__dus.s.time_in_queue, ldus->s.time_in_queue);
122 fio_gettime(&t, NULL);
123 dus->s.msec += mtime_since(&du->time, &t);
128 int update_io_ticks(void)
130 struct flist_head *entry;
131 struct disk_util *du;
134 dprint(FD_DISKUTIL, "update io ticks\n");
136 fio_sem_down(disk_util_sem);
138 if (!helper_should_exit()) {
139 flist_for_each(entry, &disk_list) {
140 du = flist_entry(entry, struct disk_util, list);
141 update_io_tick_disk(du);
146 fio_sem_up(disk_util_sem);
150 static struct disk_util *disk_util_exists(int major, int minor)
152 struct flist_head *entry;
153 struct disk_util *du;
155 fio_sem_down(disk_util_sem);
157 flist_for_each(entry, &disk_list) {
158 du = flist_entry(entry, struct disk_util, list);
160 if (major == du->major && minor == du->minor) {
161 fio_sem_up(disk_util_sem);
166 fio_sem_up(disk_util_sem);
170 static int get_device_numbers(char *file_name, int *maj, int *min)
174 char tempname[PATH_MAX], *p;
176 if (!lstat(file_name, &st)) {
177 if (S_ISBLK(st.st_mode)) {
178 majdev = major(st.st_rdev);
179 mindev = minor(st.st_rdev);
180 } else if (S_ISCHR(st.st_mode) ||
181 S_ISFIFO(st.st_mode)) {
184 majdev = major(st.st_dev);
185 mindev = minor(st.st_dev);
189 * must be a file, open "." in that path
191 snprintf(tempname, FIO_ARRAY_SIZE(tempname), "%s", file_name);
192 p = dirname(tempname);
194 perror("disk util stat");
198 majdev = major(st.st_dev);
199 mindev = minor(st.st_dev);
208 static int read_block_dev_entry(char *path, int *maj, int *min)
213 f = fopen(path, "r");
219 p = fgets(line, sizeof(line), f);
225 if (sscanf(p, "%u:%u", maj, min) != 2)
231 static void find_add_disk_slaves(struct thread_data *td, char *path,
232 struct disk_util *masterdu)
234 DIR *dirhandle = NULL;
235 struct dirent *dirent = NULL;
236 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
237 struct disk_util *slavedu = NULL;
241 sprintf(slavesdir, "%s/%s", path, "slaves");
242 dirhandle = opendir(slavesdir);
246 while ((dirent = readdir(dirhandle)) != NULL) {
247 if (!strcmp(dirent->d_name, ".") ||
248 !strcmp(dirent->d_name, ".."))
251 nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir,
253 /* Can we always assume that the slaves device entries
254 * are links to the real directories for the slave
257 linklen = readlink(temppath, slavepath, PATH_MAX - 1);
259 perror("readlink() for slave device.");
263 slavepath[linklen] = '\0';
265 nowarn_snprintf(temppath, sizeof(temppath), "%s/%s/dev",
266 slavesdir, slavepath);
267 if (access(temppath, F_OK) != 0)
268 nowarn_snprintf(temppath, sizeof(temppath),
269 "%s/%s/device/dev", slavesdir,
271 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
272 perror("Error getting slave device numbers");
278 * See if this maj,min already exists
280 slavedu = disk_util_exists(majdev, mindev);
284 nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir,
286 __init_per_file_disk_util(td, majdev, mindev, temppath);
287 slavedu = disk_util_exists(majdev, mindev);
289 /* Should probably use an assert here. slavedu should
290 * always be present at this point. */
293 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
300 static struct disk_util *disk_util_add(struct thread_data *td, int majdev,
301 int mindev, char *path)
303 struct disk_util *du, *__du;
304 struct flist_head *entry;
307 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
309 du = smalloc(sizeof(*du));
313 DRD_IGNORE_VAR(du->users);
314 memset(du, 0, sizeof(*du));
315 INIT_FLIST_HEAD(&du->list);
316 l = snprintf(du->path, sizeof(du->path), "%s/stat", path);
317 if (l < 0 || l >= sizeof(du->path)) {
318 log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n",
319 path, sizeof(du->path) - 1);
323 snprintf((char *) du->dus.name, FIO_ARRAY_SIZE(du->dus.name), "%s",
325 du->sysfs_root = strdup(path);
328 INIT_FLIST_HEAD(&du->slavelist);
329 INIT_FLIST_HEAD(&du->slaves);
330 du->lock = fio_sem_init(FIO_SEM_UNLOCKED);
333 fio_sem_down(disk_util_sem);
335 flist_for_each(entry, &disk_list) {
336 __du = flist_entry(entry, struct disk_util, list);
338 dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
340 if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) {
342 fio_sem_up(disk_util_sem);
347 dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
349 fio_gettime(&du->time, NULL);
350 get_io_ticks(du, &du->last_dus);
352 flist_add_tail(&du->list, &disk_list);
353 fio_sem_up(disk_util_sem);
355 find_add_disk_slaves(td, path, du);
359 static int check_dev_match(int majdev, int mindev, char *path)
363 if (read_block_dev_entry(path, &major, &minor))
366 if (majdev == major && mindev == minor)
372 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
383 while ((dir = readdir(D)) != NULL) {
386 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
389 sprintf(full_path, "%s/%s", path, dir->d_name);
391 if (!strcmp(dir->d_name, "dev")) {
392 if (!check_dev_match(majdev, mindev, full_path)) {
399 if (stat(full_path, &st) == -1) {
404 if (lstat(full_path, &st) == -1) {
410 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
413 found = find_block_dir(majdev, mindev, full_path, 0);
415 strcpy(path, full_path);
424 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
425 int majdev, int mindev,
433 * If there's a ../queue/ directory there, we are inside a partition.
434 * Check if that is the case and jump back. For loop/md/dm etc we
435 * are already in the right spot.
437 sprintf(tmp, "%s/../queue", path);
438 if (!stat(tmp, &st)) {
440 sprintf(tmp, "%s/queue", p);
441 if (stat(tmp, &st)) {
442 log_err("unknown sysfs layout\n");
445 snprintf(tmp, FIO_ARRAY_SIZE(tmp), "%s", p);
446 sprintf(path, "%s", tmp);
449 return disk_util_add(td, majdev, mindev, path);
452 static struct disk_util *init_per_file_disk_util(struct thread_data *td,
457 struct disk_util *du;
460 if (get_device_numbers(filename, &majdev, &mindev))
463 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
466 du = disk_util_exists(majdev, mindev);
471 * for an fs without a device, we will repeatedly stat through
472 * sysfs which can take oodles of time for thousands of files. so
473 * cache the last lookup and compare with that before going through
476 if (mindev == last_mindev && majdev == last_majdev)
479 last_mindev = mindev;
480 last_majdev = majdev;
482 sprintf(foo, "/sys/block");
483 if (!find_block_dir(majdev, mindev, foo, 1))
486 return __init_per_file_disk_util(td, majdev, mindev, foo);
489 static struct disk_util *__init_disk_util(struct thread_data *td,
492 return init_per_file_disk_util(td, f->file_name);
495 void init_disk_util(struct thread_data *td)
500 if (!td->o.do_disk_util ||
501 td_ioengine_flagged(td, FIO_DISKLESSIO | FIO_NODISKUTIL))
504 for_each_file(td, f, i)
505 f->du = __init_disk_util(td, f);
508 void disk_util_prune_entries(void)
510 fio_sem_down(disk_util_sem);
512 while (!flist_empty(&disk_list)) {
513 struct disk_util *du;
515 du = flist_first_entry(&disk_list, struct disk_util, list);
516 flist_del(&du->list);
520 last_majdev = last_mindev = -1;
521 fio_sem_up(disk_util_sem);
522 fio_sem_remove(disk_util_sem);
525 void setup_disk_util(void)
527 disk_util_sem = fio_sem_init(FIO_SEM_UNLOCKED);