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)
49 unsigned long long sectors[2];
55 dprint(FD_DISKUTIL, "open stat file: %s\n", du->path);
57 f = fopen(du->path, "r");
61 p = fgets(line, sizeof(line), f);
67 dprint(FD_DISKUTIL, "%s: %s", du->path, p);
69 ret = sscanf(p, "%"SCNu64" %"SCNu64" %llu %"SCNu64" "
70 "%"SCNu64" %"SCNu64" %llu %"SCNu64" "
71 "%u %"SCNu64" %"SCNu64"\n",
72 &dus->s.ios[0], &dus->s.merges[0], §ors[0],
74 &dus->s.ios[1], &dus->s.merges[1], §ors[1],
76 &in_flight, &dus->s.io_ticks, &dus->s.time_in_queue);
78 dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1);
79 dus->s.sectors[0] = sectors[0];
80 dus->s.sectors[1] = sectors[1];
84 static void update_io_tick_disk(struct disk_util *du)
86 struct disk_util_stat __dus, *dus, *ldus;
91 if (get_io_ticks(du, &__dus))
97 dus->s.sectors[0] += (__dus.s.sectors[0] - ldus->s.sectors[0]);
98 dus->s.sectors[1] += (__dus.s.sectors[1] - ldus->s.sectors[1]);
99 dus->s.ios[0] += (__dus.s.ios[0] - ldus->s.ios[0]);
100 dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]);
101 dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]);
102 dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]);
103 dus->s.ticks[0] += (__dus.s.ticks[0] - ldus->s.ticks[0]);
104 dus->s.ticks[1] += (__dus.s.ticks[1] - ldus->s.ticks[1]);
105 dus->s.io_ticks += (__dus.s.io_ticks - ldus->s.io_ticks);
106 dus->s.time_in_queue += (__dus.s.time_in_queue - ldus->s.time_in_queue);
108 fio_gettime(&t, NULL);
109 dus->s.msec += mtime_since(&du->time, &t);
110 memcpy(&du->time, &t, sizeof(t));
111 memcpy(&ldus->s, &__dus.s, sizeof(__dus.s));
114 int update_io_ticks(void)
116 struct flist_head *entry;
117 struct disk_util *du;
120 dprint(FD_DISKUTIL, "update io ticks\n");
122 fio_sem_down(disk_util_sem);
124 if (!helper_should_exit()) {
125 flist_for_each(entry, &disk_list) {
126 du = flist_entry(entry, struct disk_util, list);
127 update_io_tick_disk(du);
132 fio_sem_up(disk_util_sem);
136 static struct disk_util *disk_util_exists(int major, int minor)
138 struct flist_head *entry;
139 struct disk_util *du;
141 fio_sem_down(disk_util_sem);
143 flist_for_each(entry, &disk_list) {
144 du = flist_entry(entry, struct disk_util, list);
146 if (major == du->major && minor == du->minor) {
147 fio_sem_up(disk_util_sem);
152 fio_sem_up(disk_util_sem);
156 static int get_device_numbers(char *file_name, int *maj, int *min)
160 char tempname[PATH_MAX], *p;
162 if (!lstat(file_name, &st)) {
163 if (S_ISBLK(st.st_mode)) {
164 majdev = major(st.st_rdev);
165 mindev = minor(st.st_rdev);
166 } else if (S_ISCHR(st.st_mode) ||
167 S_ISFIFO(st.st_mode)) {
170 majdev = major(st.st_dev);
171 mindev = minor(st.st_dev);
175 * must be a file, open "." in that path
177 snprintf(tempname, FIO_ARRAY_SIZE(tempname), "%s", file_name);
178 p = dirname(tempname);
180 perror("disk util stat");
184 majdev = major(st.st_dev);
185 mindev = minor(st.st_dev);
194 static int read_block_dev_entry(char *path, int *maj, int *min)
199 f = fopen(path, "r");
205 p = fgets(line, sizeof(line), f);
211 if (sscanf(p, "%u:%u", maj, min) != 2)
217 static void find_add_disk_slaves(struct thread_data *td, char *path,
218 struct disk_util *masterdu)
220 DIR *dirhandle = NULL;
221 struct dirent *dirent = NULL;
222 char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX];
223 struct disk_util *slavedu = NULL;
227 sprintf(slavesdir, "%s/%s", path, "slaves");
228 dirhandle = opendir(slavesdir);
232 while ((dirent = readdir(dirhandle)) != NULL) {
233 if (!strcmp(dirent->d_name, ".") ||
234 !strcmp(dirent->d_name, ".."))
237 nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir,
239 /* Can we always assume that the slaves device entries
240 * are links to the real directories for the slave
243 linklen = readlink(temppath, slavepath, PATH_MAX - 1);
245 perror("readlink() for slave device.");
249 slavepath[linklen] = '\0';
251 nowarn_snprintf(temppath, sizeof(temppath), "%s/%s/dev",
252 slavesdir, slavepath);
253 if (access(temppath, F_OK) != 0)
254 nowarn_snprintf(temppath, sizeof(temppath),
255 "%s/%s/device/dev", slavesdir,
257 if (read_block_dev_entry(temppath, &majdev, &mindev)) {
258 perror("Error getting slave device numbers");
264 * See if this maj,min already exists
266 slavedu = disk_util_exists(majdev, mindev);
270 nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir,
272 __init_per_file_disk_util(td, majdev, mindev, temppath);
273 slavedu = disk_util_exists(majdev, mindev);
275 /* Should probably use an assert here. slavedu should
276 * always be present at this point. */
279 flist_add_tail(&slavedu->slavelist, &masterdu->slaves);
286 static struct disk_util *disk_util_add(struct thread_data *td, int majdev,
287 int mindev, char *path)
289 struct disk_util *du, *__du;
290 struct flist_head *entry;
293 dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path);
295 du = smalloc(sizeof(*du));
299 DRD_IGNORE_VAR(du->users);
300 memset(du, 0, sizeof(*du));
301 INIT_FLIST_HEAD(&du->list);
302 l = snprintf(du->path, sizeof(du->path), "%s/stat", path);
303 if (l < 0 || l >= sizeof(du->path)) {
304 log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n",
305 path, sizeof(du->path) - 1);
309 snprintf((char *) du->dus.name, FIO_ARRAY_SIZE(du->dus.name), "%s",
311 du->sysfs_root = strdup(path);
314 INIT_FLIST_HEAD(&du->slavelist);
315 INIT_FLIST_HEAD(&du->slaves);
316 du->lock = fio_sem_init(FIO_SEM_UNLOCKED);
319 fio_sem_down(disk_util_sem);
321 flist_for_each(entry, &disk_list) {
322 __du = flist_entry(entry, struct disk_util, list);
324 dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name);
326 if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) {
328 fio_sem_up(disk_util_sem);
333 dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name);
335 fio_gettime(&du->time, NULL);
336 get_io_ticks(du, &du->last_dus);
338 flist_add_tail(&du->list, &disk_list);
339 fio_sem_up(disk_util_sem);
341 find_add_disk_slaves(td, path, du);
345 static int check_dev_match(int majdev, int mindev, char *path)
349 if (read_block_dev_entry(path, &major, &minor))
352 if (majdev == major && mindev == minor)
358 static int find_block_dir(int majdev, int mindev, char *path, int link_ok)
369 while ((dir = readdir(D)) != NULL) {
372 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
375 sprintf(full_path, "%s/%s", path, dir->d_name);
377 if (!strcmp(dir->d_name, "dev")) {
378 if (!check_dev_match(majdev, mindev, full_path)) {
385 if (stat(full_path, &st) == -1) {
390 if (lstat(full_path, &st) == -1) {
396 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
399 found = find_block_dir(majdev, mindev, full_path, 0);
401 strcpy(path, full_path);
410 static struct disk_util *__init_per_file_disk_util(struct thread_data *td,
411 int majdev, int mindev,
419 * If there's a ../queue/ directory there, we are inside a partition.
420 * Check if that is the case and jump back. For loop/md/dm etc we
421 * are already in the right spot.
423 sprintf(tmp, "%s/../queue", path);
424 if (!stat(tmp, &st)) {
426 sprintf(tmp, "%s/queue", p);
427 if (stat(tmp, &st)) {
428 log_err("unknown sysfs layout\n");
431 snprintf(tmp, FIO_ARRAY_SIZE(tmp), "%s", p);
432 sprintf(path, "%s", tmp);
435 return disk_util_add(td, majdev, mindev, path);
438 static struct disk_util *init_per_file_disk_util(struct thread_data *td,
443 struct disk_util *du;
446 if (get_device_numbers(filename, &majdev, &mindev))
449 dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev,
452 du = disk_util_exists(majdev, mindev);
457 * for an fs without a device, we will repeatedly stat through
458 * sysfs which can take oodles of time for thousands of files. so
459 * cache the last lookup and compare with that before going through
462 if (mindev == last_mindev && majdev == last_majdev)
465 last_mindev = mindev;
466 last_majdev = majdev;
468 sprintf(foo, "/sys/block");
469 if (!find_block_dir(majdev, mindev, foo, 1))
472 return __init_per_file_disk_util(td, majdev, mindev, foo);
475 static struct disk_util *__init_disk_util(struct thread_data *td,
478 return init_per_file_disk_util(td, f->file_name);
481 void init_disk_util(struct thread_data *td)
486 if (!td->o.do_disk_util ||
487 td_ioengine_flagged(td, FIO_DISKLESSIO | FIO_NODISKUTIL))
490 for_each_file(td, f, i)
491 f->du = __init_disk_util(td, f);
494 void disk_util_prune_entries(void)
496 fio_sem_down(disk_util_sem);
498 while (!flist_empty(&disk_list)) {
499 struct disk_util *du;
501 du = flist_first_entry(&disk_list, struct disk_util, list);
502 flist_del(&du->list);
506 last_majdev = last_mindev = -1;
507 fio_sem_up(disk_util_sem);
508 fio_sem_remove(disk_util_sem);
511 void setup_disk_util(void)
513 disk_util_sem = fio_sem_init(FIO_SEM_UNLOCKED);