| 1 | #include <unistd.h> |
| 2 | #include <fcntl.h> |
| 3 | #include <string.h> |
| 4 | #include <assert.h> |
| 5 | #include <dirent.h> |
| 6 | #include <sys/stat.h> |
| 7 | #include <sys/mman.h> |
| 8 | #include <sys/types.h> |
| 9 | |
| 10 | #include "fio.h" |
| 11 | #include "os.h" |
| 12 | |
| 13 | static int extend_file(struct thread_data *td, struct fio_file *f) |
| 14 | { |
| 15 | int r, new_layout = 0, flags; |
| 16 | unsigned long long left; |
| 17 | unsigned int bs; |
| 18 | char *b; |
| 19 | |
| 20 | /* |
| 21 | * check if we need to lay the file out complete again. fio |
| 22 | * does that for operations involving reads, or for writes |
| 23 | * where overwrite is set |
| 24 | */ |
| 25 | if (td_read(td) || (td_write(td) && td->o.overwrite)) |
| 26 | new_layout = 1; |
| 27 | |
| 28 | if (new_layout && (f->flags & FIO_FILE_EXISTS)) { |
| 29 | if (unlink(f->file_name) < 0) { |
| 30 | td_verror(td, errno, "unlink"); |
| 31 | return 1; |
| 32 | } |
| 33 | } |
| 34 | |
| 35 | flags = O_WRONLY | O_CREAT; |
| 36 | if (new_layout) |
| 37 | flags |= O_TRUNC; |
| 38 | |
| 39 | f->fd = open(f->file_name, flags, 0644); |
| 40 | if (f->fd < 0) { |
| 41 | td_verror(td, errno, "open"); |
| 42 | return 1; |
| 43 | } |
| 44 | |
| 45 | if (ftruncate(f->fd, f->real_file_size) == -1) { |
| 46 | td_verror(td, errno, "ftruncate"); |
| 47 | goto err; |
| 48 | } |
| 49 | |
| 50 | if (posix_fallocate(f->fd, 0, f->real_file_size) < 0) { |
| 51 | td_verror(td, errno, "posix_fallocate"); |
| 52 | goto err; |
| 53 | } |
| 54 | |
| 55 | if (!new_layout) |
| 56 | goto done; |
| 57 | |
| 58 | b = malloc(td->o.max_bs[DDIR_WRITE]); |
| 59 | memset(b, 0, td->o.max_bs[DDIR_WRITE]); |
| 60 | |
| 61 | left = f->real_file_size; |
| 62 | while (left && !td->terminate) { |
| 63 | bs = td->o.max_bs[DDIR_WRITE]; |
| 64 | if (bs > left) |
| 65 | bs = left; |
| 66 | |
| 67 | r = write(f->fd, b, bs); |
| 68 | |
| 69 | if (r == (int) bs) { |
| 70 | left -= bs; |
| 71 | continue; |
| 72 | } else { |
| 73 | if (r < 0) |
| 74 | td_verror(td, errno, "write"); |
| 75 | else |
| 76 | td_verror(td, EIO, "write"); |
| 77 | |
| 78 | break; |
| 79 | } |
| 80 | } |
| 81 | |
| 82 | if (td->terminate) |
| 83 | unlink(f->file_name); |
| 84 | else if (td->o.create_fsync) |
| 85 | fsync(f->fd); |
| 86 | |
| 87 | free(b); |
| 88 | done: |
| 89 | close(f->fd); |
| 90 | f->fd = -1; |
| 91 | return 0; |
| 92 | err: |
| 93 | close(f->fd); |
| 94 | f->fd = -1; |
| 95 | return 1; |
| 96 | } |
| 97 | |
| 98 | static unsigned long long get_rand_file_size(struct thread_data *td) |
| 99 | { |
| 100 | unsigned long long ret; |
| 101 | long r; |
| 102 | |
| 103 | r = os_random_long(&td->file_size_state); |
| 104 | ret = td->o.file_size_low + (unsigned long long) ((double) td->o.file_size_high * (r / (RAND_MAX + 1.0))); |
| 105 | ret -= (ret % td->o.rw_min_bs); |
| 106 | return ret; |
| 107 | } |
| 108 | |
| 109 | static int file_size(struct thread_data *td, struct fio_file *f) |
| 110 | { |
| 111 | struct stat st; |
| 112 | |
| 113 | if (fstat(f->fd, &st) == -1) { |
| 114 | td_verror(td, errno, "fstat"); |
| 115 | return 1; |
| 116 | } |
| 117 | |
| 118 | f->real_file_size = st.st_size; |
| 119 | return 0; |
| 120 | } |
| 121 | |
| 122 | static int bdev_size(struct thread_data *td, struct fio_file *f) |
| 123 | { |
| 124 | unsigned long long bytes; |
| 125 | int r; |
| 126 | |
| 127 | r = blockdev_size(f->fd, &bytes); |
| 128 | if (r) { |
| 129 | td_verror(td, r, "blockdev_size"); |
| 130 | return 1; |
| 131 | } |
| 132 | |
| 133 | f->real_file_size = bytes; |
| 134 | return 0; |
| 135 | } |
| 136 | |
| 137 | static int get_file_size(struct thread_data *td, struct fio_file *f) |
| 138 | { |
| 139 | int ret = 0; |
| 140 | |
| 141 | if (f->flags & FIO_SIZE_KNOWN) |
| 142 | return 0; |
| 143 | |
| 144 | if (f->filetype == FIO_TYPE_FILE) |
| 145 | ret = file_size(td, f); |
| 146 | else if (f->filetype == FIO_TYPE_BD) |
| 147 | ret = bdev_size(td, f); |
| 148 | else |
| 149 | f->real_file_size = -1; |
| 150 | |
| 151 | if (ret) |
| 152 | return ret; |
| 153 | |
| 154 | if (f->file_offset > f->real_file_size) { |
| 155 | log_err("%s: offset extends end (%Lu > %Lu)\n", td->o.name, f->file_offset, f->real_file_size); |
| 156 | return 1; |
| 157 | } |
| 158 | |
| 159 | f->flags |= FIO_SIZE_KNOWN; |
| 160 | return 0; |
| 161 | } |
| 162 | |
| 163 | int file_invalidate_cache(struct thread_data *td, struct fio_file *f) |
| 164 | { |
| 165 | int ret = 0; |
| 166 | |
| 167 | if (td->o.odirect) |
| 168 | return 0; |
| 169 | |
| 170 | /* |
| 171 | * FIXME: add blockdev flushing too |
| 172 | */ |
| 173 | if (f->mmap) |
| 174 | ret = madvise(f->mmap, f->io_size, MADV_DONTNEED); |
| 175 | else if (f->filetype == FIO_TYPE_FILE) |
| 176 | ret = fadvise(f->fd, f->file_offset, f->io_size, POSIX_FADV_DONTNEED); |
| 177 | else if (f->filetype == FIO_TYPE_BD) { |
| 178 | ret = blockdev_invalidate_cache(f->fd); |
| 179 | if (ret < 0 && errno == EACCES && geteuid()) { |
| 180 | log_err("fio: only root may flush block devices. Cache flush bypassed!\n"); |
| 181 | ret = 0; |
| 182 | } |
| 183 | } else if (f->filetype == FIO_TYPE_CHAR) |
| 184 | ret = 0; |
| 185 | |
| 186 | if (ret < 0) { |
| 187 | td_verror(td, errno, "invalidate_cache"); |
| 188 | return 1; |
| 189 | } |
| 190 | |
| 191 | return ret; |
| 192 | } |
| 193 | |
| 194 | void generic_close_file(struct thread_data fio_unused *td, struct fio_file *f) |
| 195 | { |
| 196 | close(f->fd); |
| 197 | f->fd = -1; |
| 198 | } |
| 199 | |
| 200 | int generic_open_file(struct thread_data *td, struct fio_file *f) |
| 201 | { |
| 202 | int flags = 0; |
| 203 | |
| 204 | if (td->o.odirect) |
| 205 | flags |= OS_O_DIRECT; |
| 206 | if (td->o.sync_io) |
| 207 | flags |= O_SYNC; |
| 208 | |
| 209 | if (td_write(td) || td_rw(td)) { |
| 210 | flags |= O_RDWR; |
| 211 | |
| 212 | if (f->filetype == FIO_TYPE_FILE) |
| 213 | flags |= O_CREAT; |
| 214 | |
| 215 | f->fd = open(f->file_name, flags, 0600); |
| 216 | } else { |
| 217 | if (f->filetype == FIO_TYPE_CHAR) |
| 218 | flags |= O_RDWR; |
| 219 | else |
| 220 | flags |= O_RDONLY; |
| 221 | |
| 222 | f->fd = open(f->file_name, flags); |
| 223 | } |
| 224 | |
| 225 | if (f->fd == -1) { |
| 226 | char buf[FIO_VERROR_SIZE]; |
| 227 | int __e = errno; |
| 228 | |
| 229 | snprintf(buf, sizeof(buf) - 1, "open(%s)", f->file_name); |
| 230 | |
| 231 | td_verror(td, __e, buf); |
| 232 | } |
| 233 | |
| 234 | if (get_file_size(td, f)) |
| 235 | goto err; |
| 236 | |
| 237 | return 0; |
| 238 | err: |
| 239 | close(f->fd); |
| 240 | return 1; |
| 241 | } |
| 242 | |
| 243 | int open_files(struct thread_data *td) |
| 244 | { |
| 245 | struct fio_file *f; |
| 246 | unsigned int i; |
| 247 | int err = 0; |
| 248 | |
| 249 | for_each_file(td, f, i) { |
| 250 | err = td_io_open_file(td, f); |
| 251 | if (err) { |
| 252 | if (td->error == EMFILE) { |
| 253 | log_err("fio: limited open files to: %d\n", td->nr_open_files); |
| 254 | td->o.open_files = td->nr_open_files; |
| 255 | err = 0; |
| 256 | clear_error(td); |
| 257 | } |
| 258 | break; |
| 259 | } |
| 260 | |
| 261 | if (td->o.open_files == td->nr_open_files) |
| 262 | break; |
| 263 | } |
| 264 | |
| 265 | if (!err) |
| 266 | return 0; |
| 267 | |
| 268 | for_each_file(td, f, i) |
| 269 | td_io_close_file(td, f); |
| 270 | |
| 271 | return err; |
| 272 | } |
| 273 | |
| 274 | /* |
| 275 | * open/close all files, so that ->real_file_size gets set |
| 276 | */ |
| 277 | static void get_file_sizes(struct thread_data *td) |
| 278 | { |
| 279 | struct fio_file *f; |
| 280 | unsigned int i; |
| 281 | |
| 282 | for_each_file(td, f, i) { |
| 283 | if (td->io_ops->open_file(td, f)) |
| 284 | clear_error(td); |
| 285 | else |
| 286 | td->io_ops->close_file(td, f); |
| 287 | |
| 288 | if (f->real_file_size == -1ULL && td->o.size) |
| 289 | f->real_file_size = td->o.size / td->o.nr_files; |
| 290 | } |
| 291 | } |
| 292 | |
| 293 | /* |
| 294 | * Open the files and setup files sizes, creating files if necessary. |
| 295 | */ |
| 296 | int setup_files(struct thread_data *td) |
| 297 | { |
| 298 | unsigned long long total_size, extend_size; |
| 299 | struct fio_file *f; |
| 300 | unsigned int i; |
| 301 | int err = 0, need_extend; |
| 302 | |
| 303 | /* |
| 304 | * if ioengine defines a setup() method, it's responsible for |
| 305 | * opening the files and setting f->real_file_size to indicate |
| 306 | * the valid range for that file. |
| 307 | */ |
| 308 | if (td->io_ops->setup) |
| 309 | err = td->io_ops->setup(td); |
| 310 | else |
| 311 | get_file_sizes(td); |
| 312 | |
| 313 | if (err) |
| 314 | return err; |
| 315 | |
| 316 | /* |
| 317 | * check sizes. if the files/devices do not exist and the size |
| 318 | * isn't passed to fio, abort. |
| 319 | */ |
| 320 | total_size = 0; |
| 321 | for_each_file(td, f, i) { |
| 322 | if (f->real_file_size == -1ULL) |
| 323 | total_size = -1ULL; |
| 324 | else |
| 325 | total_size += f->real_file_size; |
| 326 | } |
| 327 | |
| 328 | /* |
| 329 | * device/file sizes are zero and no size given, punt |
| 330 | */ |
| 331 | if ((!total_size || total_size == -1ULL) && !td->o.size) { |
| 332 | log_err("%s: you need to specify size=\n", td->o.name); |
| 333 | td_verror(td, EINVAL, "total_file_size"); |
| 334 | return 1; |
| 335 | } |
| 336 | |
| 337 | /* |
| 338 | * now file sizes are known, so we can set ->io_size. if size= is |
| 339 | * not given, ->io_size is just equal to ->real_file_size. if size |
| 340 | * is given, ->io_size is size / nr_files. |
| 341 | */ |
| 342 | extend_size = total_size = 0; |
| 343 | need_extend = 0; |
| 344 | for_each_file(td, f, i) { |
| 345 | if (!td->o.file_size_low) { |
| 346 | /* |
| 347 | * no file size range given, file size is equal to |
| 348 | * total size divided by number of files. if that is |
| 349 | * zero, set it to the real file size. |
| 350 | */ |
| 351 | f->io_size = td->o.size / td->o.nr_files; |
| 352 | if (!f->io_size) |
| 353 | f->io_size = f->real_file_size; |
| 354 | } else if (f->real_file_size < td->o.file_size_low || |
| 355 | f->real_file_size > td->o.file_size_high) { |
| 356 | /* |
| 357 | * file size given. if it's fixed, use that. if it's a |
| 358 | * range, generate a random size in-between. |
| 359 | */ |
| 360 | if (td->o.file_size_low == td->o.file_size_high) |
| 361 | f->io_size = td->o.file_size_low; |
| 362 | else |
| 363 | f->io_size = get_rand_file_size(td); |
| 364 | } else |
| 365 | f->io_size = f->real_file_size; |
| 366 | |
| 367 | if (f->io_size == -1ULL) |
| 368 | total_size = -1ULL; |
| 369 | else |
| 370 | total_size += f->io_size; |
| 371 | |
| 372 | if (f->filetype == FIO_TYPE_FILE && |
| 373 | f->io_size > f->real_file_size && |
| 374 | !(td->io_ops->flags & FIO_DISKLESSIO)) { |
| 375 | need_extend++; |
| 376 | extend_size += f->io_size; |
| 377 | f->flags |= FIO_FILE_EXTEND; |
| 378 | } |
| 379 | } |
| 380 | |
| 381 | if (!td->o.size) |
| 382 | td->o.size = total_size; |
| 383 | |
| 384 | /* |
| 385 | * See if we need to extend some files |
| 386 | */ |
| 387 | if (need_extend) { |
| 388 | temp_stall_ts = 1; |
| 389 | log_info("%s: Laying out IO file(s) (%u files / %LuMiB)\n", |
| 390 | td->o.name, need_extend, extend_size >> 20); |
| 391 | |
| 392 | for_each_file(td, f, i) { |
| 393 | if (!(f->flags & FIO_FILE_EXTEND)) |
| 394 | continue; |
| 395 | |
| 396 | assert(f->filetype == FIO_TYPE_FILE); |
| 397 | f->flags &= ~FIO_FILE_EXTEND; |
| 398 | f->real_file_size = f->io_size; |
| 399 | err = extend_file(td, f); |
| 400 | if (err) |
| 401 | break; |
| 402 | } |
| 403 | temp_stall_ts = 0; |
| 404 | } |
| 405 | |
| 406 | if (err) |
| 407 | return err; |
| 408 | |
| 409 | if (!td->o.zone_size) |
| 410 | td->o.zone_size = td->o.size; |
| 411 | |
| 412 | td->total_io_size = td->o.size * td->o.loops; |
| 413 | return 0; |
| 414 | } |
| 415 | |
| 416 | int init_random_map(struct thread_data *td) |
| 417 | { |
| 418 | int num_maps, blocks; |
| 419 | struct fio_file *f; |
| 420 | unsigned int i; |
| 421 | |
| 422 | if (td->o.norandommap) |
| 423 | return 0; |
| 424 | |
| 425 | for_each_file(td, f, i) { |
| 426 | blocks = (f->real_file_size + td->o.rw_min_bs - 1) / td->o.rw_min_bs; |
| 427 | num_maps = (blocks + BLOCKS_PER_MAP-1)/ BLOCKS_PER_MAP; |
| 428 | f->file_map = malloc(num_maps * sizeof(long)); |
| 429 | if (!f->file_map) { |
| 430 | log_err("fio: failed allocating random map. If running a large number of jobs, try the 'norandommap' option\n"); |
| 431 | return 1; |
| 432 | } |
| 433 | f->num_maps = num_maps; |
| 434 | memset(f->file_map, 0, num_maps * sizeof(long)); |
| 435 | } |
| 436 | |
| 437 | return 0; |
| 438 | } |
| 439 | |
| 440 | void close_files(struct thread_data *td) |
| 441 | { |
| 442 | struct fio_file *f; |
| 443 | unsigned int i; |
| 444 | |
| 445 | for_each_file(td, f, i) { |
| 446 | if ((f->flags & FIO_FILE_UNLINK) && |
| 447 | f->filetype == FIO_TYPE_FILE) |
| 448 | unlink(f->file_name); |
| 449 | |
| 450 | td_io_close_file(td, f); |
| 451 | |
| 452 | free(f->file_name); |
| 453 | f->file_name = NULL; |
| 454 | |
| 455 | if (f->file_map) { |
| 456 | free(f->file_map); |
| 457 | f->file_map = NULL; |
| 458 | } |
| 459 | } |
| 460 | |
| 461 | td->o.filename = NULL; |
| 462 | free(td->files); |
| 463 | td->files = NULL; |
| 464 | td->o.nr_files = 0; |
| 465 | } |
| 466 | |
| 467 | static void get_file_type(struct fio_file *f) |
| 468 | { |
| 469 | struct stat sb; |
| 470 | |
| 471 | f->filetype = FIO_TYPE_FILE; |
| 472 | |
| 473 | if (!lstat(f->file_name, &sb)) { |
| 474 | if (S_ISBLK(sb.st_mode)) |
| 475 | f->filetype = FIO_TYPE_BD; |
| 476 | else if (S_ISCHR(sb.st_mode)) |
| 477 | f->filetype = FIO_TYPE_CHAR; |
| 478 | } |
| 479 | } |
| 480 | |
| 481 | void add_file(struct thread_data *td, const char *fname) |
| 482 | { |
| 483 | int cur_files = td->files_index; |
| 484 | char file_name[PATH_MAX]; |
| 485 | struct fio_file *f; |
| 486 | int len = 0; |
| 487 | |
| 488 | td->files = realloc(td->files, (cur_files + 1) * sizeof(*f)); |
| 489 | |
| 490 | f = &td->files[cur_files]; |
| 491 | memset(f, 0, sizeof(*f)); |
| 492 | f->fd = -1; |
| 493 | |
| 494 | if (td->o.directory) |
| 495 | len = sprintf(file_name, "%s/", td->o.directory); |
| 496 | |
| 497 | sprintf(file_name + len, "%s", fname); |
| 498 | f->file_name = strdup(file_name); |
| 499 | |
| 500 | get_file_type(f); |
| 501 | |
| 502 | td->files_index++; |
| 503 | if (f->filetype == FIO_TYPE_FILE) |
| 504 | td->nr_normal_files++; |
| 505 | } |
| 506 | |
| 507 | void get_file(struct fio_file *f) |
| 508 | { |
| 509 | f->references++; |
| 510 | } |
| 511 | |
| 512 | void put_file(struct thread_data *td, struct fio_file *f) |
| 513 | { |
| 514 | if (!(f->flags & FIO_FILE_OPEN)) |
| 515 | return; |
| 516 | |
| 517 | assert(f->references); |
| 518 | if (--f->references) |
| 519 | return; |
| 520 | |
| 521 | if (should_fsync(td) && td->o.fsync_on_close) |
| 522 | fsync(f->fd); |
| 523 | |
| 524 | if (td->io_ops->close_file) |
| 525 | td->io_ops->close_file(td, f); |
| 526 | td->nr_open_files--; |
| 527 | f->flags &= ~FIO_FILE_OPEN; |
| 528 | } |
| 529 | |
| 530 | static int recurse_dir(struct thread_data *td, const char *dirname) |
| 531 | { |
| 532 | struct dirent *dir; |
| 533 | int ret = 0; |
| 534 | DIR *D; |
| 535 | |
| 536 | D = opendir(dirname); |
| 537 | if (!D) { |
| 538 | char buf[FIO_VERROR_SIZE]; |
| 539 | |
| 540 | snprintf(buf, FIO_VERROR_SIZE - 1, "opendir(%s)", dirname); |
| 541 | td_verror(td, errno, buf); |
| 542 | return 1; |
| 543 | } |
| 544 | |
| 545 | while ((dir = readdir(D)) != NULL) { |
| 546 | char full_path[PATH_MAX]; |
| 547 | struct stat sb; |
| 548 | |
| 549 | if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, "..")) |
| 550 | continue; |
| 551 | |
| 552 | sprintf(full_path, "%s/%s", dirname, dir->d_name); |
| 553 | |
| 554 | if (lstat(full_path, &sb) == -1) { |
| 555 | if (errno != ENOENT) { |
| 556 | td_verror(td, errno, "stat"); |
| 557 | return 1; |
| 558 | } |
| 559 | } |
| 560 | |
| 561 | if (S_ISREG(sb.st_mode)) { |
| 562 | add_file(td, full_path); |
| 563 | td->o.nr_files++; |
| 564 | continue; |
| 565 | } |
| 566 | if (!S_ISDIR(sb.st_mode)) |
| 567 | continue; |
| 568 | |
| 569 | if ((ret = recurse_dir(td, full_path)) != 0) |
| 570 | break; |
| 571 | } |
| 572 | |
| 573 | closedir(D); |
| 574 | return ret; |
| 575 | } |
| 576 | |
| 577 | int add_dir_files(struct thread_data *td, const char *path) |
| 578 | { |
| 579 | int ret = recurse_dir(td, path); |
| 580 | |
| 581 | if (!ret) |
| 582 | log_info("fio: opendir added %d files\n", td->o.nr_files); |
| 583 | |
| 584 | return ret; |
| 585 | } |
| 586 | |
| 587 | void dup_files(struct thread_data *td, struct thread_data *org) |
| 588 | { |
| 589 | struct fio_file *f; |
| 590 | unsigned int i; |
| 591 | size_t bytes; |
| 592 | |
| 593 | if (!org->files) |
| 594 | return; |
| 595 | |
| 596 | bytes = org->files_index * sizeof(*f); |
| 597 | td->files = malloc(bytes); |
| 598 | memcpy(td->files, org->files, bytes); |
| 599 | |
| 600 | for_each_file(td, f, i) { |
| 601 | if (f->file_name) |
| 602 | f->file_name = strdup(f->file_name); |
| 603 | } |
| 604 | } |