| 1 | #include <inttypes.h> |
| 2 | #include <stdio.h> |
| 3 | #include <string.h> |
| 4 | #include <sys/types.h> |
| 5 | #include <sys/stat.h> |
| 6 | #include <sys/sysmacros.h> |
| 7 | #include <dirent.h> |
| 8 | #include <libgen.h> |
| 9 | #ifdef CONFIG_VALGRIND_DEV |
| 10 | #include <valgrind/drd.h> |
| 11 | #else |
| 12 | #define DRD_IGNORE_VAR(x) do { } while (0) |
| 13 | #endif |
| 14 | |
| 15 | #include "fio.h" |
| 16 | #include "smalloc.h" |
| 17 | #include "diskutil.h" |
| 18 | #include "helper_thread.h" |
| 19 | |
| 20 | static int last_majdev, last_mindev; |
| 21 | static struct disk_util *last_du; |
| 22 | |
| 23 | static struct fio_sem *disk_util_sem; |
| 24 | |
| 25 | static struct disk_util *__init_per_file_disk_util(struct thread_data *td, |
| 26 | int majdev, int mindev, char *path); |
| 27 | |
| 28 | static void disk_util_free(struct disk_util *du) |
| 29 | { |
| 30 | if (du == last_du) |
| 31 | last_du = NULL; |
| 32 | |
| 33 | while (!flist_empty(&du->slaves)) { |
| 34 | struct disk_util *slave; |
| 35 | |
| 36 | slave = flist_first_entry(&du->slaves, struct disk_util, slavelist); |
| 37 | flist_del(&slave->slavelist); |
| 38 | slave->users--; |
| 39 | } |
| 40 | |
| 41 | fio_sem_remove(du->lock); |
| 42 | free(du->sysfs_root); |
| 43 | sfree(du); |
| 44 | } |
| 45 | |
| 46 | static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus) |
| 47 | { |
| 48 | char line[256]; |
| 49 | FILE *f; |
| 50 | char *p; |
| 51 | int ret; |
| 52 | |
| 53 | dprint(FD_DISKUTIL, "open stat file: %s\n", du->path); |
| 54 | |
| 55 | f = fopen(du->path, "r"); |
| 56 | if (!f) |
| 57 | return 1; |
| 58 | |
| 59 | p = fgets(line, sizeof(line), f); |
| 60 | if (!p) { |
| 61 | fclose(f); |
| 62 | return 1; |
| 63 | } |
| 64 | |
| 65 | dprint(FD_DISKUTIL, "%s: %s", du->path, p); |
| 66 | |
| 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], |
| 71 | &dus->s.ticks[0], |
| 72 | &dus->s.ios[1], &dus->s.merges[1], &dus->s.sectors[1], |
| 73 | &dus->s.ticks[1], |
| 74 | &dus->s.io_ticks, &dus->s.time_in_queue); |
| 75 | fclose(f); |
| 76 | dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 10); |
| 77 | return ret != 10; |
| 78 | } |
| 79 | |
| 80 | static uint64_t safe_32bit_diff(uint64_t nval, uint64_t oval) |
| 81 | { |
| 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 |
| 89 | * successive polls. |
| 90 | */ |
| 91 | if (oval <= nval || oval >= (1ull << 32)) |
| 92 | return nval - oval; |
| 93 | else |
| 94 | return (1ull << 32) + nval - oval; |
| 95 | } |
| 96 | |
| 97 | static void update_io_tick_disk(struct disk_util *du) |
| 98 | { |
| 99 | struct disk_util_stat __dus, *dus, *ldus; |
| 100 | struct timespec t; |
| 101 | |
| 102 | if (!du->users) |
| 103 | return; |
| 104 | if (get_io_ticks(du, &__dus)) |
| 105 | return; |
| 106 | |
| 107 | dus = &du->dus; |
| 108 | ldus = &du->last_dus; |
| 109 | |
| 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); |
| 121 | |
| 122 | fio_gettime(&t, NULL); |
| 123 | dus->s.msec += mtime_since(&du->time, &t); |
| 124 | du->time = t; |
| 125 | ldus->s = __dus.s; |
| 126 | } |
| 127 | |
| 128 | int update_io_ticks(void) |
| 129 | { |
| 130 | struct flist_head *entry; |
| 131 | struct disk_util *du; |
| 132 | int ret = 0; |
| 133 | |
| 134 | dprint(FD_DISKUTIL, "update io ticks\n"); |
| 135 | |
| 136 | fio_sem_down(disk_util_sem); |
| 137 | |
| 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); |
| 142 | } |
| 143 | } else |
| 144 | ret = 1; |
| 145 | |
| 146 | fio_sem_up(disk_util_sem); |
| 147 | return ret; |
| 148 | } |
| 149 | |
| 150 | static struct disk_util *disk_util_exists(int major, int minor) |
| 151 | { |
| 152 | struct flist_head *entry; |
| 153 | struct disk_util *du; |
| 154 | |
| 155 | fio_sem_down(disk_util_sem); |
| 156 | |
| 157 | flist_for_each(entry, &disk_list) { |
| 158 | du = flist_entry(entry, struct disk_util, list); |
| 159 | |
| 160 | if (major == du->major && minor == du->minor) { |
| 161 | fio_sem_up(disk_util_sem); |
| 162 | return du; |
| 163 | } |
| 164 | } |
| 165 | |
| 166 | fio_sem_up(disk_util_sem); |
| 167 | return NULL; |
| 168 | } |
| 169 | |
| 170 | static int get_device_numbers(char *file_name, int *maj, int *min) |
| 171 | { |
| 172 | struct stat st; |
| 173 | int majdev, mindev; |
| 174 | char tempname[PATH_MAX], *p; |
| 175 | |
| 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)) { |
| 182 | return -1; |
| 183 | } else { |
| 184 | majdev = major(st.st_dev); |
| 185 | mindev = minor(st.st_dev); |
| 186 | } |
| 187 | } else { |
| 188 | /* |
| 189 | * must be a file, open "." in that path |
| 190 | */ |
| 191 | snprintf(tempname, FIO_ARRAY_SIZE(tempname), "%s", file_name); |
| 192 | p = dirname(tempname); |
| 193 | if (stat(p, &st)) { |
| 194 | perror("disk util stat"); |
| 195 | return -1; |
| 196 | } |
| 197 | |
| 198 | majdev = major(st.st_dev); |
| 199 | mindev = minor(st.st_dev); |
| 200 | } |
| 201 | |
| 202 | *min = mindev; |
| 203 | *maj = majdev; |
| 204 | |
| 205 | return 0; |
| 206 | } |
| 207 | |
| 208 | static int read_block_dev_entry(char *path, int *maj, int *min) |
| 209 | { |
| 210 | char line[256], *p; |
| 211 | FILE *f; |
| 212 | |
| 213 | f = fopen(path, "r"); |
| 214 | if (!f) { |
| 215 | perror("open path"); |
| 216 | return 1; |
| 217 | } |
| 218 | |
| 219 | p = fgets(line, sizeof(line), f); |
| 220 | fclose(f); |
| 221 | |
| 222 | if (!p) |
| 223 | return 1; |
| 224 | |
| 225 | if (sscanf(p, "%u:%u", maj, min) != 2) |
| 226 | return 1; |
| 227 | |
| 228 | return 0; |
| 229 | } |
| 230 | |
| 231 | static void find_add_disk_slaves(struct thread_data *td, char *path, |
| 232 | struct disk_util *masterdu) |
| 233 | { |
| 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; |
| 238 | int majdev, mindev; |
| 239 | ssize_t linklen; |
| 240 | |
| 241 | sprintf(slavesdir, "%s/%s", path, "slaves"); |
| 242 | dirhandle = opendir(slavesdir); |
| 243 | if (!dirhandle) |
| 244 | return; |
| 245 | |
| 246 | while ((dirent = readdir(dirhandle)) != NULL) { |
| 247 | if (!strcmp(dirent->d_name, ".") || |
| 248 | !strcmp(dirent->d_name, "..")) |
| 249 | continue; |
| 250 | |
| 251 | nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir, |
| 252 | dirent->d_name); |
| 253 | /* Can we always assume that the slaves device entries |
| 254 | * are links to the real directories for the slave |
| 255 | * devices? |
| 256 | */ |
| 257 | linklen = readlink(temppath, slavepath, PATH_MAX - 1); |
| 258 | if (linklen < 0) { |
| 259 | perror("readlink() for slave device."); |
| 260 | closedir(dirhandle); |
| 261 | return; |
| 262 | } |
| 263 | slavepath[linklen] = '\0'; |
| 264 | |
| 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, |
| 270 | slavepath); |
| 271 | if (read_block_dev_entry(temppath, &majdev, &mindev)) { |
| 272 | perror("Error getting slave device numbers"); |
| 273 | closedir(dirhandle); |
| 274 | return; |
| 275 | } |
| 276 | |
| 277 | /* |
| 278 | * See if this maj,min already exists |
| 279 | */ |
| 280 | slavedu = disk_util_exists(majdev, mindev); |
| 281 | if (slavedu) |
| 282 | continue; |
| 283 | |
| 284 | nowarn_snprintf(temppath, sizeof(temppath), "%s/%s", slavesdir, |
| 285 | slavepath); |
| 286 | __init_per_file_disk_util(td, majdev, mindev, temppath); |
| 287 | slavedu = disk_util_exists(majdev, mindev); |
| 288 | |
| 289 | /* Should probably use an assert here. slavedu should |
| 290 | * always be present at this point. */ |
| 291 | if (slavedu) { |
| 292 | slavedu->users++; |
| 293 | flist_add_tail(&slavedu->slavelist, &masterdu->slaves); |
| 294 | } |
| 295 | } |
| 296 | |
| 297 | closedir(dirhandle); |
| 298 | } |
| 299 | |
| 300 | static struct disk_util *disk_util_add(struct thread_data *td, int majdev, |
| 301 | int mindev, char *path) |
| 302 | { |
| 303 | struct disk_util *du, *__du; |
| 304 | struct flist_head *entry; |
| 305 | int l; |
| 306 | |
| 307 | dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path); |
| 308 | |
| 309 | du = smalloc(sizeof(*du)); |
| 310 | if (!du) |
| 311 | return NULL; |
| 312 | |
| 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); |
| 320 | sfree(du); |
| 321 | return NULL; |
| 322 | } |
| 323 | snprintf((char *) du->dus.name, FIO_ARRAY_SIZE(du->dus.name), "%s", |
| 324 | basename(path)); |
| 325 | du->sysfs_root = strdup(path); |
| 326 | du->major = majdev; |
| 327 | du->minor = mindev; |
| 328 | INIT_FLIST_HEAD(&du->slavelist); |
| 329 | INIT_FLIST_HEAD(&du->slaves); |
| 330 | du->lock = fio_sem_init(FIO_SEM_UNLOCKED); |
| 331 | du->users = 0; |
| 332 | |
| 333 | fio_sem_down(disk_util_sem); |
| 334 | |
| 335 | flist_for_each(entry, &disk_list) { |
| 336 | __du = flist_entry(entry, struct disk_util, list); |
| 337 | |
| 338 | dprint(FD_DISKUTIL, "found %s in list\n", __du->dus.name); |
| 339 | |
| 340 | if (!strcmp((char *) du->dus.name, (char *) __du->dus.name)) { |
| 341 | disk_util_free(du); |
| 342 | fio_sem_up(disk_util_sem); |
| 343 | return __du; |
| 344 | } |
| 345 | } |
| 346 | |
| 347 | dprint(FD_DISKUTIL, "add %s to list\n", du->dus.name); |
| 348 | |
| 349 | fio_gettime(&du->time, NULL); |
| 350 | get_io_ticks(du, &du->last_dus); |
| 351 | |
| 352 | flist_add_tail(&du->list, &disk_list); |
| 353 | fio_sem_up(disk_util_sem); |
| 354 | |
| 355 | find_add_disk_slaves(td, path, du); |
| 356 | return du; |
| 357 | } |
| 358 | |
| 359 | static int check_dev_match(int majdev, int mindev, char *path) |
| 360 | { |
| 361 | int major, minor; |
| 362 | |
| 363 | if (read_block_dev_entry(path, &major, &minor)) |
| 364 | return 1; |
| 365 | |
| 366 | if (majdev == major && mindev == minor) |
| 367 | return 0; |
| 368 | |
| 369 | return 1; |
| 370 | } |
| 371 | |
| 372 | static int find_block_dir(int majdev, int mindev, char *path, int link_ok) |
| 373 | { |
| 374 | struct dirent *dir; |
| 375 | struct stat st; |
| 376 | int found = 0; |
| 377 | DIR *D; |
| 378 | |
| 379 | D = opendir(path); |
| 380 | if (!D) |
| 381 | return 0; |
| 382 | |
| 383 | while ((dir = readdir(D)) != NULL) { |
| 384 | char full_path[257]; |
| 385 | |
| 386 | if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, "..")) |
| 387 | continue; |
| 388 | |
| 389 | sprintf(full_path, "%s/%s", path, dir->d_name); |
| 390 | |
| 391 | if (!strcmp(dir->d_name, "dev")) { |
| 392 | if (!check_dev_match(majdev, mindev, full_path)) { |
| 393 | found = 1; |
| 394 | break; |
| 395 | } |
| 396 | } |
| 397 | |
| 398 | if (link_ok) { |
| 399 | if (stat(full_path, &st) == -1) { |
| 400 | perror("stat"); |
| 401 | break; |
| 402 | } |
| 403 | } else { |
| 404 | if (lstat(full_path, &st) == -1) { |
| 405 | perror("stat"); |
| 406 | break; |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode)) |
| 411 | continue; |
| 412 | |
| 413 | found = find_block_dir(majdev, mindev, full_path, 0); |
| 414 | if (found) { |
| 415 | strcpy(path, full_path); |
| 416 | break; |
| 417 | } |
| 418 | } |
| 419 | |
| 420 | closedir(D); |
| 421 | return found; |
| 422 | } |
| 423 | |
| 424 | static struct disk_util *__init_per_file_disk_util(struct thread_data *td, |
| 425 | int majdev, int mindev, |
| 426 | char *path) |
| 427 | { |
| 428 | struct stat st; |
| 429 | char tmp[PATH_MAX]; |
| 430 | char *p; |
| 431 | |
| 432 | /* |
| 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. |
| 436 | */ |
| 437 | sprintf(tmp, "%s/../queue", path); |
| 438 | if (!stat(tmp, &st)) { |
| 439 | p = dirname(path); |
| 440 | sprintf(tmp, "%s/queue", p); |
| 441 | if (stat(tmp, &st)) { |
| 442 | log_err("unknown sysfs layout\n"); |
| 443 | return NULL; |
| 444 | } |
| 445 | snprintf(tmp, FIO_ARRAY_SIZE(tmp), "%s", p); |
| 446 | sprintf(path, "%s", tmp); |
| 447 | } |
| 448 | |
| 449 | return disk_util_add(td, majdev, mindev, path); |
| 450 | } |
| 451 | |
| 452 | static struct disk_util *init_per_file_disk_util(struct thread_data *td, |
| 453 | char *filename) |
| 454 | { |
| 455 | |
| 456 | char foo[PATH_MAX]; |
| 457 | struct disk_util *du; |
| 458 | int mindev, majdev; |
| 459 | |
| 460 | if (get_device_numbers(filename, &majdev, &mindev)) |
| 461 | return NULL; |
| 462 | |
| 463 | dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev, |
| 464 | mindev); |
| 465 | |
| 466 | du = disk_util_exists(majdev, mindev); |
| 467 | if (du) |
| 468 | return du; |
| 469 | |
| 470 | /* |
| 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 |
| 474 | * everything again. |
| 475 | */ |
| 476 | if (mindev == last_mindev && majdev == last_majdev) |
| 477 | return last_du; |
| 478 | |
| 479 | last_mindev = mindev; |
| 480 | last_majdev = majdev; |
| 481 | |
| 482 | sprintf(foo, "/sys/block"); |
| 483 | if (!find_block_dir(majdev, mindev, foo, 1)) |
| 484 | return NULL; |
| 485 | |
| 486 | return __init_per_file_disk_util(td, majdev, mindev, foo); |
| 487 | } |
| 488 | |
| 489 | static struct disk_util *__init_disk_util(struct thread_data *td, |
| 490 | struct fio_file *f) |
| 491 | { |
| 492 | return init_per_file_disk_util(td, f->file_name); |
| 493 | } |
| 494 | |
| 495 | void init_disk_util(struct thread_data *td) |
| 496 | { |
| 497 | struct fio_file *f; |
| 498 | unsigned int i; |
| 499 | |
| 500 | if (!td->o.do_disk_util || |
| 501 | td_ioengine_flagged(td, FIO_DISKLESSIO | FIO_NODISKUTIL)) |
| 502 | return; |
| 503 | |
| 504 | for_each_file(td, f, i) |
| 505 | f->du = __init_disk_util(td, f); |
| 506 | } |
| 507 | |
| 508 | void disk_util_prune_entries(void) |
| 509 | { |
| 510 | fio_sem_down(disk_util_sem); |
| 511 | |
| 512 | while (!flist_empty(&disk_list)) { |
| 513 | struct disk_util *du; |
| 514 | |
| 515 | du = flist_first_entry(&disk_list, struct disk_util, list); |
| 516 | flist_del(&du->list); |
| 517 | disk_util_free(du); |
| 518 | } |
| 519 | |
| 520 | last_majdev = last_mindev = -1; |
| 521 | fio_sem_up(disk_util_sem); |
| 522 | fio_sem_remove(disk_util_sem); |
| 523 | } |
| 524 | |
| 525 | void setup_disk_util(void) |
| 526 | { |
| 527 | disk_util_sem = fio_sem_init(FIO_SEM_UNLOCKED); |
| 528 | } |