| 1 | #include <stdio.h> |
| 2 | #include <string.h> |
| 3 | #include <sys/types.h> |
| 4 | #include <sys/stat.h> |
| 5 | #include <sys/sysmacros.h> |
| 6 | #include <dirent.h> |
| 7 | #include <libgen.h> |
| 8 | #ifdef CONFIG_VALGRIND_DEV |
| 9 | #include <valgrind/drd.h> |
| 10 | #else |
| 11 | #define DRD_IGNORE_VAR(x) do { } while (0) |
| 12 | #endif |
| 13 | |
| 14 | #include "fio.h" |
| 15 | #include "smalloc.h" |
| 16 | #include "diskutil.h" |
| 17 | #include "helper_thread.h" |
| 18 | |
| 19 | static int last_majdev, last_mindev; |
| 20 | static struct disk_util *last_du; |
| 21 | |
| 22 | static struct fio_sem *disk_util_sem; |
| 23 | |
| 24 | static struct disk_util *__init_per_file_disk_util(struct thread_data *td, |
| 25 | int majdev, int mindev, char *path); |
| 26 | |
| 27 | static void disk_util_free(struct disk_util *du) |
| 28 | { |
| 29 | if (du == last_du) |
| 30 | last_du = NULL; |
| 31 | |
| 32 | while (!flist_empty(&du->slaves)) { |
| 33 | struct disk_util *slave; |
| 34 | |
| 35 | slave = flist_first_entry(&du->slaves, struct disk_util, slavelist); |
| 36 | flist_del(&slave->slavelist); |
| 37 | slave->users--; |
| 38 | } |
| 39 | |
| 40 | fio_sem_remove(du->lock); |
| 41 | free(du->sysfs_root); |
| 42 | sfree(du); |
| 43 | } |
| 44 | |
| 45 | static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus) |
| 46 | { |
| 47 | unsigned in_flight; |
| 48 | unsigned long long sectors[2]; |
| 49 | char line[256]; |
| 50 | FILE *f; |
| 51 | char *p; |
| 52 | int ret; |
| 53 | |
| 54 | dprint(FD_DISKUTIL, "open stat file: %s\n", du->path); |
| 55 | |
| 56 | f = fopen(du->path, "r"); |
| 57 | if (!f) |
| 58 | return 1; |
| 59 | |
| 60 | p = fgets(line, sizeof(line), f); |
| 61 | if (!p) { |
| 62 | fclose(f); |
| 63 | return 1; |
| 64 | } |
| 65 | |
| 66 | dprint(FD_DISKUTIL, "%s: %s", du->path, p); |
| 67 | |
| 68 | ret = sscanf(p, "%llu %llu %llu %llu %llu %llu %llu %llu %u %llu %llu\n", |
| 69 | (unsigned long long *) &dus->s.ios[0], |
| 70 | (unsigned long long *) &dus->s.merges[0], |
| 71 | §ors[0], |
| 72 | (unsigned long long *) &dus->s.ticks[0], |
| 73 | (unsigned long long *) &dus->s.ios[1], |
| 74 | (unsigned long long *) &dus->s.merges[1], |
| 75 | §ors[1], |
| 76 | (unsigned long long *) &dus->s.ticks[1], |
| 77 | &in_flight, |
| 78 | (unsigned long long *) &dus->s.io_ticks, |
| 79 | (unsigned long long *) &dus->s.time_in_queue); |
| 80 | fclose(f); |
| 81 | dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1); |
| 82 | dus->s.sectors[0] = sectors[0]; |
| 83 | dus->s.sectors[1] = sectors[1]; |
| 84 | return ret != 11; |
| 85 | } |
| 86 | |
| 87 | static void update_io_tick_disk(struct disk_util *du) |
| 88 | { |
| 89 | struct disk_util_stat __dus, *dus, *ldus; |
| 90 | struct timespec t; |
| 91 | |
| 92 | if (!du->users) |
| 93 | return; |
| 94 | if (get_io_ticks(du, &__dus)) |
| 95 | return; |
| 96 | |
| 97 | dus = &du->dus; |
| 98 | ldus = &du->last_dus; |
| 99 | |
| 100 | dus->s.sectors[0] += (__dus.s.sectors[0] - ldus->s.sectors[0]); |
| 101 | dus->s.sectors[1] += (__dus.s.sectors[1] - ldus->s.sectors[1]); |
| 102 | dus->s.ios[0] += (__dus.s.ios[0] - ldus->s.ios[0]); |
| 103 | dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]); |
| 104 | dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]); |
| 105 | dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]); |
| 106 | dus->s.ticks[0] += (__dus.s.ticks[0] - ldus->s.ticks[0]); |
| 107 | dus->s.ticks[1] += (__dus.s.ticks[1] - ldus->s.ticks[1]); |
| 108 | dus->s.io_ticks += (__dus.s.io_ticks - ldus->s.io_ticks); |
| 109 | dus->s.time_in_queue += (__dus.s.time_in_queue - ldus->s.time_in_queue); |
| 110 | |
| 111 | fio_gettime(&t, NULL); |
| 112 | dus->s.msec += mtime_since(&du->time, &t); |
| 113 | memcpy(&du->time, &t, sizeof(t)); |
| 114 | memcpy(&ldus->s, &__dus.s, sizeof(__dus.s)); |
| 115 | } |
| 116 | |
| 117 | int update_io_ticks(void) |
| 118 | { |
| 119 | struct flist_head *entry; |
| 120 | struct disk_util *du; |
| 121 | int ret = 0; |
| 122 | |
| 123 | dprint(FD_DISKUTIL, "update io ticks\n"); |
| 124 | |
| 125 | fio_sem_down(disk_util_sem); |
| 126 | |
| 127 | if (!helper_should_exit()) { |
| 128 | flist_for_each(entry, &disk_list) { |
| 129 | du = flist_entry(entry, struct disk_util, list); |
| 130 | update_io_tick_disk(du); |
| 131 | } |
| 132 | } else |
| 133 | ret = 1; |
| 134 | |
| 135 | fio_sem_up(disk_util_sem); |
| 136 | return ret; |
| 137 | } |
| 138 | |
| 139 | static struct disk_util *disk_util_exists(int major, int minor) |
| 140 | { |
| 141 | struct flist_head *entry; |
| 142 | struct disk_util *du; |
| 143 | |
| 144 | fio_sem_down(disk_util_sem); |
| 145 | |
| 146 | flist_for_each(entry, &disk_list) { |
| 147 | du = flist_entry(entry, struct disk_util, list); |
| 148 | |
| 149 | if (major == du->major && minor == du->minor) { |
| 150 | fio_sem_up(disk_util_sem); |
| 151 | return du; |
| 152 | } |
| 153 | } |
| 154 | |
| 155 | fio_sem_up(disk_util_sem); |
| 156 | return NULL; |
| 157 | } |
| 158 | |
| 159 | static int get_device_numbers(char *file_name, int *maj, int *min) |
| 160 | { |
| 161 | struct stat st; |
| 162 | int majdev, mindev; |
| 163 | char tempname[PATH_MAX], *p; |
| 164 | |
| 165 | if (!lstat(file_name, &st)) { |
| 166 | if (S_ISBLK(st.st_mode)) { |
| 167 | majdev = major(st.st_rdev); |
| 168 | mindev = minor(st.st_rdev); |
| 169 | } else if (S_ISCHR(st.st_mode) || |
| 170 | S_ISFIFO(st.st_mode)) { |
| 171 | return -1; |
| 172 | } else { |
| 173 | majdev = major(st.st_dev); |
| 174 | mindev = minor(st.st_dev); |
| 175 | } |
| 176 | } else { |
| 177 | /* |
| 178 | * must be a file, open "." in that path |
| 179 | */ |
| 180 | snprintf(tempname, FIO_ARRAY_SIZE(tempname), "%s", file_name); |
| 181 | p = dirname(tempname); |
| 182 | if (stat(p, &st)) { |
| 183 | perror("disk util stat"); |
| 184 | return -1; |
| 185 | } |
| 186 | |
| 187 | majdev = major(st.st_dev); |
| 188 | mindev = minor(st.st_dev); |
| 189 | } |
| 190 | |
| 191 | *min = mindev; |
| 192 | *maj = majdev; |
| 193 | |
| 194 | return 0; |
| 195 | } |
| 196 | |
| 197 | static int read_block_dev_entry(char *path, int *maj, int *min) |
| 198 | { |
| 199 | char line[256], *p; |
| 200 | FILE *f; |
| 201 | |
| 202 | f = fopen(path, "r"); |
| 203 | if (!f) { |
| 204 | perror("open path"); |
| 205 | return 1; |
| 206 | } |
| 207 | |
| 208 | p = fgets(line, sizeof(line), f); |
| 209 | fclose(f); |
| 210 | |
| 211 | if (!p) |
| 212 | return 1; |
| 213 | |
| 214 | if (sscanf(p, "%u:%u", maj, min) != 2) |
| 215 | return 1; |
| 216 | |
| 217 | return 0; |
| 218 | } |
| 219 | |
| 220 | static void find_add_disk_slaves(struct thread_data *td, char *path, |
| 221 | struct disk_util *masterdu) |
| 222 | { |
| 223 | DIR *dirhandle = NULL; |
| 224 | struct dirent *dirent = NULL; |
| 225 | char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX]; |
| 226 | struct disk_util *slavedu = NULL; |
| 227 | int majdev, mindev; |
| 228 | ssize_t linklen; |
| 229 | |
| 230 | sprintf(slavesdir, "%s/%s", path, "slaves"); |
| 231 | dirhandle = opendir(slavesdir); |
| 232 | if (!dirhandle) |
| 233 | return; |
| 234 | |
| 235 | while ((dirent = readdir(dirhandle)) != NULL) { |
| 236 | if (!strcmp(dirent->d_name, ".") || |
| 237 | !strcmp(dirent->d_name, "..")) |
| 238 | continue; |
| 239 | |
| 240 | nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir, |
| 241 | dirent->d_name); |
| 242 | /* Can we always assume that the slaves device entries |
| 243 | * are links to the real directories for the slave |
| 244 | * devices? |
| 245 | */ |
| 246 | linklen = readlink(temppath, slavepath, PATH_MAX - 1); |
| 247 | if (linklen < 0) { |
| 248 | perror("readlink() for slave device."); |
| 249 | closedir(dirhandle); |
| 250 | return; |
| 251 | } |
| 252 | slavepath[linklen] = '\0'; |
| 253 | |
| 254 | nowarn_snprintf(temppath, sizeof(temppath), "%s/%s/dev", |
| 255 | slavesdir, slavepath); |
| 256 | if (access(temppath, F_OK) != 0) |
| 257 | nowarn_snprintf(temppath, sizeof(temppath), |
| 258 | "%s/%s/device/dev", slavesdir, |
| 259 | slavepath); |
| 260 | if (read_block_dev_entry(temppath, &majdev, &mindev)) { |
| 261 | perror("Error getting slave device numbers"); |
| 262 | closedir(dirhandle); |
| 263 | return; |
| 264 | } |
| 265 | |
| 266 | /* |
| 267 | * See if this maj,min already exists |
| 268 | */ |
| 269 | slavedu = disk_util_exists(majdev, mindev); |
| 270 | if (slavedu) |
| 271 | continue; |
| 272 | |
| 273 | nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir, |
| 274 | slavepath); |
| 275 | __init_per_file_disk_util(td, majdev, mindev, temppath); |
| 276 | slavedu = disk_util_exists(majdev, mindev); |
| 277 | |
| 278 | /* Should probably use an assert here. slavedu should |
| 279 | * always be present at this point. */ |
| 280 | if (slavedu) { |
| 281 | slavedu->users++; |
| 282 | flist_add_tail(&slavedu->slavelist, &masterdu->slaves); |
| 283 | } |
| 284 | } |
| 285 | |
| 286 | closedir(dirhandle); |
| 287 | } |
| 288 | |
| 289 | static struct disk_util *disk_util_add(struct thread_data *td, int majdev, |
| 290 | int mindev, char *path) |
| 291 | { |
| 292 | struct disk_util *du, *__du; |
| 293 | struct flist_head *entry; |
| 294 | int l; |
| 295 | |
| 296 | dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path); |
| 297 | |
| 298 | du = smalloc(sizeof(*du)); |
| 299 | if (!du) |
| 300 | return NULL; |
| 301 | |
| 302 | DRD_IGNORE_VAR(du->users); |
| 303 | memset(du, 0, sizeof(*du)); |
| 304 | INIT_FLIST_HEAD(&du->list); |
| 305 | l = snprintf(du->path, sizeof(du->path), "%s/stat", path); |
| 306 | if (l < 0 || l >= sizeof(du->path)) { |
| 307 | log_err("constructed path \"%.100s[...]/stat\" larger than buffer (%zu bytes)\n", |
| 308 | path, sizeof(du->path) - 1); |
| 309 | sfree(du); |
| 310 | return NULL; |
| 311 | } |
| 312 | snprintf((char *) du->dus.name, FIO_ARRAY_SIZE(du->dus.name), "%s", |
| 313 | basename(path)); |
| 314 | du->sysfs_root = strdup(path); |
| 315 | du->major = majdev; |
| 316 | du->minor = mindev; |
| 317 | INIT_FLIST_HEAD(&du->slavelist); |
| 318 | INIT_FLIST_HEAD(&du->slaves); |
| 319 | du->lock = fio_sem_init(FIO_SEM_UNLOCKED); |
| 320 | du->users = 0; |
| 321 | |
| 322 | fio_sem_down(disk_util_sem); |
| 323 | |
| 324 | flist_for_each(entry, &disk_list) { |
| 325 | __du = flist_entry(entry, struct disk_util, list); |
| 326 | |
| 327 | dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name); |
| 328 | |
| 329 | if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) { |
| 330 | disk_util_free(du); |
| 331 | fio_sem_up(disk_util_sem); |
| 332 | return __du; |
| 333 | } |
| 334 | } |
| 335 | |
| 336 | dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name); |
| 337 | |
| 338 | fio_gettime(&du->time, NULL); |
| 339 | get_io_ticks(du, &du->last_dus); |
| 340 | |
| 341 | flist_add_tail(&du->list, &disk_list); |
| 342 | fio_sem_up(disk_util_sem); |
| 343 | |
| 344 | find_add_disk_slaves(td, path, du); |
| 345 | return du; |
| 346 | } |
| 347 | |
| 348 | static int check_dev_match(int majdev, int mindev, char *path) |
| 349 | { |
| 350 | int major, minor; |
| 351 | |
| 352 | if (read_block_dev_entry(path, &major, &minor)) |
| 353 | return 1; |
| 354 | |
| 355 | if (majdev == major && mindev == minor) |
| 356 | return 0; |
| 357 | |
| 358 | return 1; |
| 359 | } |
| 360 | |
| 361 | static int find_block_dir(int majdev, int mindev, char *path, int link_ok) |
| 362 | { |
| 363 | struct dirent *dir; |
| 364 | struct stat st; |
| 365 | int found = 0; |
| 366 | DIR *D; |
| 367 | |
| 368 | D = opendir(path); |
| 369 | if (!D) |
| 370 | return 0; |
| 371 | |
| 372 | while ((dir = readdir(D)) != NULL) { |
| 373 | char full_path[257]; |
| 374 | |
| 375 | if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, "..")) |
| 376 | continue; |
| 377 | |
| 378 | sprintf(full_path, "%s/%s", path, dir->d_name); |
| 379 | |
| 380 | if (!strcmp(dir->d_name, "dev")) { |
| 381 | if (!check_dev_match(majdev, mindev, full_path)) { |
| 382 | found = 1; |
| 383 | break; |
| 384 | } |
| 385 | } |
| 386 | |
| 387 | if (link_ok) { |
| 388 | if (stat(full_path, &st) == -1) { |
| 389 | perror("stat"); |
| 390 | break; |
| 391 | } |
| 392 | } else { |
| 393 | if (lstat(full_path, &st) == -1) { |
| 394 | perror("stat"); |
| 395 | break; |
| 396 | } |
| 397 | } |
| 398 | |
| 399 | if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode)) |
| 400 | continue; |
| 401 | |
| 402 | found = find_block_dir(majdev, mindev, full_path, 0); |
| 403 | if (found) { |
| 404 | strcpy(path, full_path); |
| 405 | break; |
| 406 | } |
| 407 | } |
| 408 | |
| 409 | closedir(D); |
| 410 | return found; |
| 411 | } |
| 412 | |
| 413 | static struct disk_util *__init_per_file_disk_util(struct thread_data *td, |
| 414 | int majdev, int mindev, |
| 415 | char *path) |
| 416 | { |
| 417 | struct stat st; |
| 418 | char tmp[PATH_MAX]; |
| 419 | char *p; |
| 420 | |
| 421 | /* |
| 422 | * If there's a ../queue/ directory there, we are inside a partition. |
| 423 | * Check if that is the case and jump back. For loop/md/dm etc we |
| 424 | * are already in the right spot. |
| 425 | */ |
| 426 | sprintf(tmp, "%s/../queue", path); |
| 427 | if (!stat(tmp, &st)) { |
| 428 | p = dirname(path); |
| 429 | sprintf(tmp, "%s/queue", p); |
| 430 | if (stat(tmp, &st)) { |
| 431 | log_err("unknown sysfs layout\n"); |
| 432 | return NULL; |
| 433 | } |
| 434 | snprintf(tmp, FIO_ARRAY_SIZE(tmp), "%s", p); |
| 435 | sprintf(path, "%s", tmp); |
| 436 | } |
| 437 | |
| 438 | return disk_util_add(td, majdev, mindev, path); |
| 439 | } |
| 440 | |
| 441 | static struct disk_util *init_per_file_disk_util(struct thread_data *td, |
| 442 | char *filename) |
| 443 | { |
| 444 | |
| 445 | char foo[PATH_MAX]; |
| 446 | struct disk_util *du; |
| 447 | int mindev, majdev; |
| 448 | |
| 449 | if (get_device_numbers(filename, &majdev, &mindev)) |
| 450 | return NULL; |
| 451 | |
| 452 | dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev, |
| 453 | mindev); |
| 454 | |
| 455 | du = disk_util_exists(majdev, mindev); |
| 456 | if (du) |
| 457 | return du; |
| 458 | |
| 459 | /* |
| 460 | * for an fs without a device, we will repeatedly stat through |
| 461 | * sysfs which can take oodles of time for thousands of files. so |
| 462 | * cache the last lookup and compare with that before going through |
| 463 | * everything again. |
| 464 | */ |
| 465 | if (mindev == last_mindev && majdev == last_majdev) |
| 466 | return last_du; |
| 467 | |
| 468 | last_mindev = mindev; |
| 469 | last_majdev = majdev; |
| 470 | |
| 471 | sprintf(foo, "/sys/block"); |
| 472 | if (!find_block_dir(majdev, mindev, foo, 1)) |
| 473 | return NULL; |
| 474 | |
| 475 | return __init_per_file_disk_util(td, majdev, mindev, foo); |
| 476 | } |
| 477 | |
| 478 | static struct disk_util *__init_disk_util(struct thread_data *td, |
| 479 | struct fio_file *f) |
| 480 | { |
| 481 | return init_per_file_disk_util(td, f->file_name); |
| 482 | } |
| 483 | |
| 484 | void init_disk_util(struct thread_data *td) |
| 485 | { |
| 486 | struct fio_file *f; |
| 487 | unsigned int i; |
| 488 | |
| 489 | if (!td->o.do_disk_util || |
| 490 | td_ioengine_flagged(td, FIO_DISKLESSIO | FIO_NODISKUTIL)) |
| 491 | return; |
| 492 | |
| 493 | for_each_file(td, f, i) |
| 494 | f->du = __init_disk_util(td, f); |
| 495 | } |
| 496 | |
| 497 | void disk_util_prune_entries(void) |
| 498 | { |
| 499 | fio_sem_down(disk_util_sem); |
| 500 | |
| 501 | while (!flist_empty(&disk_list)) { |
| 502 | struct disk_util *du; |
| 503 | |
| 504 | du = flist_first_entry(&disk_list, struct disk_util, list); |
| 505 | flist_del(&du->list); |
| 506 | disk_util_free(du); |
| 507 | } |
| 508 | |
| 509 | last_majdev = last_mindev = -1; |
| 510 | fio_sem_up(disk_util_sem); |
| 511 | fio_sem_remove(disk_util_sem); |
| 512 | } |
| 513 | |
| 514 | void setup_disk_util(void) |
| 515 | { |
| 516 | disk_util_sem = fio_sem_init(FIO_SEM_UNLOCKED); |
| 517 | } |