| 1 | /* |
| 2 | * The io parts of the fio tool, includes workers for sync and mmap'ed |
| 3 | * io, as well as both posix and linux libaio support. |
| 4 | * |
| 5 | * sync io is implemented on top of aio. |
| 6 | * |
| 7 | * This is not really specific to fio, if the get_io_u/put_io_u and |
| 8 | * structures was pulled into this as well it would be a perfectly |
| 9 | * generic io engine that could be used for other projects. |
| 10 | * |
| 11 | */ |
| 12 | #include <stdio.h> |
| 13 | #include <stdlib.h> |
| 14 | #include <unistd.h> |
| 15 | #include <errno.h> |
| 16 | #include <assert.h> |
| 17 | #include <time.h> |
| 18 | #include <string.h> |
| 19 | #include <sys/mman.h> |
| 20 | #include <sys/poll.h> |
| 21 | #include "fio.h" |
| 22 | #include "os.h" |
| 23 | |
| 24 | static int fill_timespec(struct timespec *ts) |
| 25 | { |
| 26 | #ifdef _POSIX_TIMERS |
| 27 | if (!clock_gettime(CLOCK_MONOTONIC, ts)) |
| 28 | return 0; |
| 29 | |
| 30 | perror("clock_gettime"); |
| 31 | #endif |
| 32 | return 1; |
| 33 | } |
| 34 | |
| 35 | static unsigned long long ts_utime_since_now(struct timespec *t) |
| 36 | { |
| 37 | long long sec, nsec; |
| 38 | struct timespec now; |
| 39 | |
| 40 | if (fill_timespec(&now)) |
| 41 | return 0; |
| 42 | |
| 43 | sec = now.tv_sec - t->tv_sec; |
| 44 | nsec = now.tv_nsec - t->tv_nsec; |
| 45 | if (sec > 0 && nsec < 0) { |
| 46 | sec--; |
| 47 | nsec += 1000000000; |
| 48 | } |
| 49 | |
| 50 | sec *= 1000000; |
| 51 | nsec /= 1000; |
| 52 | return sec + nsec; |
| 53 | } |
| 54 | |
| 55 | static int fio_io_sync(struct thread_data *td) |
| 56 | { |
| 57 | return fsync(td->fd); |
| 58 | } |
| 59 | |
| 60 | #ifdef FIO_HAVE_LIBAIO |
| 61 | |
| 62 | #define ev_to_iou(ev) (struct io_u *) ((unsigned long) (ev)->obj) |
| 63 | |
| 64 | struct libaio_data { |
| 65 | io_context_t aio_ctx; |
| 66 | struct io_event *aio_events; |
| 67 | }; |
| 68 | |
| 69 | static int fio_libaio_io_prep(struct thread_data *td, struct io_u *io_u) |
| 70 | { |
| 71 | if (io_u->ddir == DDIR_READ) |
| 72 | io_prep_pread(&io_u->iocb, td->fd, io_u->buf, io_u->buflen, io_u->offset); |
| 73 | else |
| 74 | io_prep_pwrite(&io_u->iocb, td->fd, io_u->buf, io_u->buflen, io_u->offset); |
| 75 | |
| 76 | return 0; |
| 77 | } |
| 78 | |
| 79 | static struct io_u *fio_libaio_event(struct thread_data *td, int event) |
| 80 | { |
| 81 | struct libaio_data *ld = td->io_data; |
| 82 | |
| 83 | return ev_to_iou(ld->aio_events + event); |
| 84 | } |
| 85 | |
| 86 | static int fio_libaio_getevents(struct thread_data *td, int min, int max, |
| 87 | struct timespec *t) |
| 88 | { |
| 89 | struct libaio_data *ld = td->io_data; |
| 90 | long r; |
| 91 | |
| 92 | do { |
| 93 | r = io_getevents(ld->aio_ctx, min, max, ld->aio_events, t); |
| 94 | if (r == -EAGAIN) { |
| 95 | usleep(100); |
| 96 | continue; |
| 97 | } else if (r == -EINTR) |
| 98 | continue; |
| 99 | else |
| 100 | break; |
| 101 | } while (1); |
| 102 | |
| 103 | return (int) r; |
| 104 | } |
| 105 | |
| 106 | static int fio_libaio_queue(struct thread_data *td, struct io_u *io_u) |
| 107 | { |
| 108 | struct libaio_data *ld = td->io_data; |
| 109 | struct iocb *iocb = &io_u->iocb; |
| 110 | long ret; |
| 111 | |
| 112 | do { |
| 113 | ret = io_submit(ld->aio_ctx, 1, &iocb); |
| 114 | if (ret == 1) |
| 115 | return 0; |
| 116 | else if (ret == -EAGAIN) |
| 117 | usleep(100); |
| 118 | else if (ret == -EINTR) |
| 119 | continue; |
| 120 | else |
| 121 | break; |
| 122 | } while (1); |
| 123 | |
| 124 | return (int) ret; |
| 125 | |
| 126 | } |
| 127 | |
| 128 | static int fio_libaio_cancel(struct thread_data *td, struct io_u *io_u) |
| 129 | { |
| 130 | struct libaio_data *ld = td->io_data; |
| 131 | |
| 132 | return io_cancel(ld->aio_ctx, &io_u->iocb, ld->aio_events); |
| 133 | } |
| 134 | |
| 135 | static void fio_libaio_cleanup(struct thread_data *td) |
| 136 | { |
| 137 | struct libaio_data *ld = td->io_data; |
| 138 | |
| 139 | if (ld) { |
| 140 | io_destroy(ld->aio_ctx); |
| 141 | if (ld->aio_events) |
| 142 | free(ld->aio_events); |
| 143 | |
| 144 | free(ld); |
| 145 | td->io_data = NULL; |
| 146 | } |
| 147 | } |
| 148 | |
| 149 | int fio_libaio_init(struct thread_data *td) |
| 150 | { |
| 151 | struct libaio_data *ld = malloc(sizeof(*ld)); |
| 152 | |
| 153 | memset(ld, 0, sizeof(*ld)); |
| 154 | if (io_queue_init(td->iodepth, &ld->aio_ctx)) { |
| 155 | td_verror(td, errno); |
| 156 | return 1; |
| 157 | } |
| 158 | |
| 159 | td->io_prep = fio_libaio_io_prep; |
| 160 | td->io_queue = fio_libaio_queue; |
| 161 | td->io_getevents = fio_libaio_getevents; |
| 162 | td->io_event = fio_libaio_event; |
| 163 | td->io_cancel = fio_libaio_cancel; |
| 164 | td->io_cleanup = fio_libaio_cleanup; |
| 165 | td->io_sync = fio_io_sync; |
| 166 | |
| 167 | ld->aio_events = malloc(td->iodepth * sizeof(struct io_event)); |
| 168 | td->io_data = ld; |
| 169 | return 0; |
| 170 | } |
| 171 | |
| 172 | #else /* FIO_HAVE_LIBAIO */ |
| 173 | |
| 174 | int fio_libaio_init(struct thread_data *td) |
| 175 | { |
| 176 | return EINVAL; |
| 177 | } |
| 178 | |
| 179 | #endif /* FIO_HAVE_LIBAIO */ |
| 180 | |
| 181 | #ifdef FIO_HAVE_POSIXAIO |
| 182 | |
| 183 | struct posixaio_data { |
| 184 | struct io_u **aio_events; |
| 185 | }; |
| 186 | |
| 187 | static int fio_posixaio_cancel(struct thread_data *td, struct io_u *io_u) |
| 188 | { |
| 189 | int r = aio_cancel(td->fd, &io_u->aiocb); |
| 190 | |
| 191 | if (r == 1 || r == AIO_CANCELED) |
| 192 | return 0; |
| 193 | |
| 194 | return 1; |
| 195 | } |
| 196 | |
| 197 | static int fio_posixaio_prep(struct thread_data *td, struct io_u *io_u) |
| 198 | { |
| 199 | struct aiocb *aiocb = &io_u->aiocb; |
| 200 | |
| 201 | aiocb->aio_fildes = td->fd; |
| 202 | aiocb->aio_buf = io_u->buf; |
| 203 | aiocb->aio_nbytes = io_u->buflen; |
| 204 | aiocb->aio_offset = io_u->offset; |
| 205 | |
| 206 | io_u->seen = 0; |
| 207 | return 0; |
| 208 | } |
| 209 | |
| 210 | static int fio_posixaio_getevents(struct thread_data *td, int min, int max, |
| 211 | struct timespec *t) |
| 212 | { |
| 213 | struct posixaio_data *pd = td->io_data; |
| 214 | struct list_head *entry; |
| 215 | struct timespec start; |
| 216 | int r, have_timeout = 0; |
| 217 | |
| 218 | if (t && !fill_timespec(&start)) |
| 219 | have_timeout = 1; |
| 220 | |
| 221 | r = 0; |
| 222 | restart: |
| 223 | list_for_each(entry, &td->io_u_busylist) { |
| 224 | struct io_u *io_u = list_entry(entry, struct io_u, list); |
| 225 | int err; |
| 226 | |
| 227 | if (io_u->seen) |
| 228 | continue; |
| 229 | |
| 230 | err = aio_error(&io_u->aiocb); |
| 231 | switch (err) { |
| 232 | default: |
| 233 | io_u->error = err; |
| 234 | case ECANCELED: |
| 235 | case 0: |
| 236 | pd->aio_events[r++] = io_u; |
| 237 | io_u->seen = 1; |
| 238 | break; |
| 239 | case EINPROGRESS: |
| 240 | break; |
| 241 | } |
| 242 | |
| 243 | if (r >= max) |
| 244 | break; |
| 245 | } |
| 246 | |
| 247 | if (r >= min) |
| 248 | return r; |
| 249 | |
| 250 | if (have_timeout) { |
| 251 | unsigned long long usec; |
| 252 | |
| 253 | usec = (t->tv_sec * 1000000) + (t->tv_nsec / 1000); |
| 254 | if (ts_utime_since_now(&start) > usec) |
| 255 | return r; |
| 256 | } |
| 257 | |
| 258 | /* |
| 259 | * hrmpf, we need to wait for more. we should use aio_suspend, for |
| 260 | * now just sleep a little and recheck status of busy-and-not-seen |
| 261 | */ |
| 262 | usleep(1000); |
| 263 | goto restart; |
| 264 | } |
| 265 | |
| 266 | static struct io_u *fio_posixaio_event(struct thread_data *td, int event) |
| 267 | { |
| 268 | struct posixaio_data *pd = td->io_data; |
| 269 | |
| 270 | return pd->aio_events[event]; |
| 271 | } |
| 272 | |
| 273 | static int fio_posixaio_queue(struct thread_data *td, struct io_u *io_u) |
| 274 | { |
| 275 | struct aiocb *aiocb = &io_u->aiocb; |
| 276 | int ret; |
| 277 | |
| 278 | if (io_u->ddir == DDIR_READ) |
| 279 | ret = aio_read(aiocb); |
| 280 | else |
| 281 | ret = aio_write(aiocb); |
| 282 | |
| 283 | if (ret) |
| 284 | io_u->error = errno; |
| 285 | |
| 286 | return io_u->error; |
| 287 | } |
| 288 | |
| 289 | static void fio_posixaio_cleanup(struct thread_data *td) |
| 290 | { |
| 291 | struct posixaio_data *pd = td->io_data; |
| 292 | |
| 293 | if (pd) { |
| 294 | free(pd->aio_events); |
| 295 | free(pd); |
| 296 | td->io_data = NULL; |
| 297 | } |
| 298 | } |
| 299 | |
| 300 | int fio_posixaio_init(struct thread_data *td) |
| 301 | { |
| 302 | struct posixaio_data *pd = malloc(sizeof(*pd)); |
| 303 | |
| 304 | pd->aio_events = malloc(td->iodepth * sizeof(struct io_u *)); |
| 305 | |
| 306 | td->io_prep = fio_posixaio_prep; |
| 307 | td->io_queue = fio_posixaio_queue; |
| 308 | td->io_getevents = fio_posixaio_getevents; |
| 309 | td->io_event = fio_posixaio_event; |
| 310 | td->io_cancel = fio_posixaio_cancel; |
| 311 | td->io_cleanup = fio_posixaio_cleanup; |
| 312 | td->io_sync = fio_io_sync; |
| 313 | |
| 314 | td->io_data = pd; |
| 315 | return 0; |
| 316 | } |
| 317 | |
| 318 | #else /* FIO_HAVE_POSIXAIO */ |
| 319 | |
| 320 | int fio_posixaio_init(struct thread_data *td) |
| 321 | { |
| 322 | return EINVAL; |
| 323 | } |
| 324 | |
| 325 | #endif /* FIO_HAVE_POSIXAIO */ |
| 326 | |
| 327 | struct syncio_data { |
| 328 | struct io_u *last_io_u; |
| 329 | }; |
| 330 | |
| 331 | static int fio_syncio_getevents(struct thread_data *td, int min, int max, |
| 332 | struct timespec *t) |
| 333 | { |
| 334 | assert(max <= 1); |
| 335 | |
| 336 | /* |
| 337 | * we can only have one finished io_u for sync io, since the depth |
| 338 | * is always 1 |
| 339 | */ |
| 340 | if (list_empty(&td->io_u_busylist)) |
| 341 | return 0; |
| 342 | |
| 343 | return 1; |
| 344 | } |
| 345 | |
| 346 | static struct io_u *fio_syncio_event(struct thread_data *td, int event) |
| 347 | { |
| 348 | struct syncio_data *sd = td->io_data; |
| 349 | |
| 350 | assert(event == 0); |
| 351 | |
| 352 | return sd->last_io_u; |
| 353 | } |
| 354 | |
| 355 | static int fio_syncio_prep(struct thread_data *td, struct io_u *io_u) |
| 356 | { |
| 357 | if (lseek(td->fd, io_u->offset, SEEK_SET) == -1) { |
| 358 | td_verror(td, errno); |
| 359 | return 1; |
| 360 | } |
| 361 | |
| 362 | return 0; |
| 363 | } |
| 364 | |
| 365 | static int fio_syncio_queue(struct thread_data *td, struct io_u *io_u) |
| 366 | { |
| 367 | struct syncio_data *sd = td->io_data; |
| 368 | int ret; |
| 369 | |
| 370 | if (io_u->ddir == DDIR_READ) |
| 371 | ret = read(td->fd, io_u->buf, io_u->buflen); |
| 372 | else |
| 373 | ret = write(td->fd, io_u->buf, io_u->buflen); |
| 374 | |
| 375 | if ((unsigned int) ret != io_u->buflen) { |
| 376 | if (ret > 0) { |
| 377 | io_u->resid = io_u->buflen - ret; |
| 378 | io_u->error = EIO; |
| 379 | } else |
| 380 | io_u->error = errno; |
| 381 | } |
| 382 | |
| 383 | if (!io_u->error) |
| 384 | sd->last_io_u = io_u; |
| 385 | |
| 386 | return io_u->error; |
| 387 | } |
| 388 | |
| 389 | static void fio_syncio_cleanup(struct thread_data *td) |
| 390 | { |
| 391 | if (td->io_data) { |
| 392 | free(td->io_data); |
| 393 | td->io_data = NULL; |
| 394 | } |
| 395 | } |
| 396 | |
| 397 | int fio_syncio_init(struct thread_data *td) |
| 398 | { |
| 399 | struct syncio_data *sd = malloc(sizeof(*sd)); |
| 400 | |
| 401 | td->io_prep = fio_syncio_prep; |
| 402 | td->io_queue = fio_syncio_queue; |
| 403 | td->io_getevents = fio_syncio_getevents; |
| 404 | td->io_event = fio_syncio_event; |
| 405 | td->io_cancel = NULL; |
| 406 | td->io_cleanup = fio_syncio_cleanup; |
| 407 | td->io_sync = fio_io_sync; |
| 408 | |
| 409 | sd->last_io_u = NULL; |
| 410 | td->io_data = sd; |
| 411 | return 0; |
| 412 | } |
| 413 | |
| 414 | static int fio_mmapio_queue(struct thread_data *td, struct io_u *io_u) |
| 415 | { |
| 416 | unsigned long long real_off = io_u->offset - td->file_offset; |
| 417 | struct syncio_data *sd = td->io_data; |
| 418 | |
| 419 | if (io_u->ddir == DDIR_READ) |
| 420 | memcpy(io_u->buf, td->mmap + real_off, io_u->buflen); |
| 421 | else |
| 422 | memcpy(td->mmap + real_off, io_u->buf, io_u->buflen); |
| 423 | |
| 424 | /* |
| 425 | * not really direct, but should drop the pages from the cache |
| 426 | */ |
| 427 | if (td->odirect) { |
| 428 | if (msync(td->mmap + real_off, io_u->buflen, MS_SYNC) < 0) |
| 429 | io_u->error = errno; |
| 430 | if (madvise(td->mmap + real_off, io_u->buflen, MADV_DONTNEED) < 0) |
| 431 | io_u->error = errno; |
| 432 | } |
| 433 | |
| 434 | if (!io_u->error) |
| 435 | sd->last_io_u = io_u; |
| 436 | |
| 437 | return io_u->error; |
| 438 | } |
| 439 | |
| 440 | static int fio_mmapio_sync(struct thread_data *td) |
| 441 | { |
| 442 | return msync(td->mmap, td->file_size, MS_SYNC); |
| 443 | } |
| 444 | |
| 445 | int fio_mmapio_init(struct thread_data *td) |
| 446 | { |
| 447 | struct syncio_data *sd = malloc(sizeof(*sd)); |
| 448 | |
| 449 | td->io_prep = NULL; |
| 450 | td->io_queue = fio_mmapio_queue; |
| 451 | td->io_getevents = fio_syncio_getevents; |
| 452 | td->io_event = fio_syncio_event; |
| 453 | td->io_cancel = NULL; |
| 454 | td->io_cleanup = fio_syncio_cleanup; |
| 455 | td->io_sync = fio_mmapio_sync; |
| 456 | |
| 457 | sd->last_io_u = NULL; |
| 458 | td->io_data = sd; |
| 459 | return 0; |
| 460 | } |
| 461 | |
| 462 | #ifdef FIO_HAVE_SGIO |
| 463 | |
| 464 | struct sgio_cmd { |
| 465 | unsigned char cdb[10]; |
| 466 | int nr; |
| 467 | }; |
| 468 | |
| 469 | struct sgio_data { |
| 470 | struct sgio_cmd *cmds; |
| 471 | struct io_u **events; |
| 472 | unsigned int bs; |
| 473 | }; |
| 474 | |
| 475 | static void sgio_hdr_init(struct sgio_data *sd, struct sg_io_hdr *hdr, |
| 476 | struct io_u *io_u, int fs) |
| 477 | { |
| 478 | struct sgio_cmd *sc = &sd->cmds[io_u->index]; |
| 479 | |
| 480 | memset(hdr, 0, sizeof(*hdr)); |
| 481 | memset(sc->cdb, 0, sizeof(sc->cdb)); |
| 482 | |
| 483 | hdr->interface_id = 'S'; |
| 484 | hdr->cmdp = sc->cdb; |
| 485 | hdr->cmd_len = sizeof(sc->cdb); |
| 486 | hdr->pack_id = io_u->index; |
| 487 | hdr->usr_ptr = io_u; |
| 488 | |
| 489 | if (fs) { |
| 490 | hdr->dxferp = io_u->buf; |
| 491 | hdr->dxfer_len = io_u->buflen; |
| 492 | } |
| 493 | } |
| 494 | |
| 495 | static int fio_sgio_getevents(struct thread_data *td, int min, int max, |
| 496 | struct timespec *t) |
| 497 | { |
| 498 | struct sgio_data *sd = td->io_data; |
| 499 | struct pollfd pfd = { .fd = td->fd, .events = POLLIN }; |
| 500 | void *buf = malloc(max * sizeof(struct sg_io_hdr)); |
| 501 | int left = max, ret, events, i, r = 0, fl = 0; |
| 502 | |
| 503 | /* |
| 504 | * don't block for !events |
| 505 | */ |
| 506 | if (!min) { |
| 507 | fl = fcntl(td->fd, F_GETFL); |
| 508 | fcntl(td->fd, F_SETFL, fl | O_NONBLOCK); |
| 509 | } |
| 510 | |
| 511 | while (left) { |
| 512 | do { |
| 513 | if (!min) |
| 514 | break; |
| 515 | poll(&pfd, 1, -1); |
| 516 | if (pfd.revents & POLLIN) |
| 517 | break; |
| 518 | } while (1); |
| 519 | |
| 520 | ret = read(td->fd, buf, left * sizeof(struct sg_io_hdr)); |
| 521 | if (ret < 0) { |
| 522 | if (errno == EAGAIN) |
| 523 | break; |
| 524 | td_verror(td, errno); |
| 525 | r = -1; |
| 526 | break; |
| 527 | } else if (!ret) |
| 528 | break; |
| 529 | |
| 530 | events = ret / sizeof(struct sg_io_hdr); |
| 531 | left -= events; |
| 532 | r += events; |
| 533 | |
| 534 | for (i = 0; i < events; i++) { |
| 535 | struct sg_io_hdr *hdr = (struct sg_io_hdr *) buf + i; |
| 536 | |
| 537 | sd->events[i] = hdr->usr_ptr; |
| 538 | } |
| 539 | } |
| 540 | |
| 541 | if (!min) |
| 542 | fcntl(td->fd, F_SETFL, fl); |
| 543 | |
| 544 | free(buf); |
| 545 | return r; |
| 546 | } |
| 547 | |
| 548 | static int fio_sgio_ioctl_doio(struct thread_data *td, struct io_u *io_u) |
| 549 | { |
| 550 | struct sgio_data *sd = td->io_data; |
| 551 | struct sg_io_hdr *hdr = &io_u->hdr; |
| 552 | |
| 553 | sd->events[0] = io_u; |
| 554 | |
| 555 | return ioctl(td->fd, SG_IO, hdr); |
| 556 | } |
| 557 | |
| 558 | static int fio_sgio_rw_doio(struct thread_data *td, struct io_u *io_u, int sync) |
| 559 | { |
| 560 | struct sg_io_hdr *hdr = &io_u->hdr; |
| 561 | int ret; |
| 562 | |
| 563 | ret = write(td->fd, hdr, sizeof(*hdr)); |
| 564 | if (ret < 0) |
| 565 | return errno; |
| 566 | |
| 567 | if (sync) { |
| 568 | ret = read(td->fd, hdr, sizeof(*hdr)); |
| 569 | if (ret < 0) |
| 570 | return errno; |
| 571 | } |
| 572 | |
| 573 | return 0; |
| 574 | } |
| 575 | |
| 576 | static int fio_sgio_doio(struct thread_data *td, struct io_u *io_u, int sync) |
| 577 | { |
| 578 | if (td->filetype == FIO_TYPE_BD) |
| 579 | return fio_sgio_ioctl_doio(td, io_u); |
| 580 | |
| 581 | return fio_sgio_rw_doio(td, io_u, sync); |
| 582 | } |
| 583 | |
| 584 | static int fio_sgio_sync(struct thread_data *td) |
| 585 | { |
| 586 | struct sgio_data *sd = td->io_data; |
| 587 | struct sg_io_hdr *hdr; |
| 588 | struct io_u *io_u; |
| 589 | int ret; |
| 590 | |
| 591 | io_u = __get_io_u(td); |
| 592 | if (!io_u) |
| 593 | return ENOMEM; |
| 594 | |
| 595 | hdr = &io_u->hdr; |
| 596 | sgio_hdr_init(sd, hdr, io_u, 0); |
| 597 | hdr->dxfer_direction = SG_DXFER_NONE; |
| 598 | |
| 599 | hdr->cmdp[0] = 0x35; |
| 600 | |
| 601 | ret = fio_sgio_doio(td, io_u, 1); |
| 602 | put_io_u(td, io_u); |
| 603 | return ret; |
| 604 | } |
| 605 | |
| 606 | static int fio_sgio_prep(struct thread_data *td, struct io_u *io_u) |
| 607 | { |
| 608 | struct sg_io_hdr *hdr = &io_u->hdr; |
| 609 | struct sgio_data *sd = td->io_data; |
| 610 | int nr_blocks, lba; |
| 611 | |
| 612 | if (io_u->buflen & (sd->bs - 1)) { |
| 613 | log_err("read/write not sector aligned\n"); |
| 614 | return EINVAL; |
| 615 | } |
| 616 | |
| 617 | sgio_hdr_init(sd, hdr, io_u, 1); |
| 618 | |
| 619 | if (io_u->ddir == DDIR_READ) { |
| 620 | hdr->dxfer_direction = SG_DXFER_FROM_DEV; |
| 621 | hdr->cmdp[0] = 0x28; |
| 622 | } else { |
| 623 | hdr->dxfer_direction = SG_DXFER_TO_DEV; |
| 624 | hdr->cmdp[0] = 0x2a; |
| 625 | } |
| 626 | |
| 627 | nr_blocks = io_u->buflen / sd->bs; |
| 628 | lba = io_u->offset / sd->bs; |
| 629 | hdr->cmdp[2] = (lba >> 24) & 0xff; |
| 630 | hdr->cmdp[3] = (lba >> 16) & 0xff; |
| 631 | hdr->cmdp[4] = (lba >> 8) & 0xff; |
| 632 | hdr->cmdp[5] = lba & 0xff; |
| 633 | hdr->cmdp[7] = (nr_blocks >> 8) & 0xff; |
| 634 | hdr->cmdp[8] = nr_blocks & 0xff; |
| 635 | return 0; |
| 636 | } |
| 637 | |
| 638 | static int fio_sgio_queue(struct thread_data *td, struct io_u *io_u) |
| 639 | { |
| 640 | struct sg_io_hdr *hdr = &io_u->hdr; |
| 641 | int ret; |
| 642 | |
| 643 | ret = fio_sgio_doio(td, io_u, 0); |
| 644 | |
| 645 | if (ret < 0) |
| 646 | io_u->error = errno; |
| 647 | else if (hdr->status) { |
| 648 | io_u->resid = hdr->resid; |
| 649 | io_u->error = EIO; |
| 650 | } |
| 651 | |
| 652 | return io_u->error; |
| 653 | } |
| 654 | |
| 655 | static struct io_u *fio_sgio_event(struct thread_data *td, int event) |
| 656 | { |
| 657 | struct sgio_data *sd = td->io_data; |
| 658 | |
| 659 | return sd->events[event]; |
| 660 | } |
| 661 | |
| 662 | static int fio_sgio_get_bs(struct thread_data *td, unsigned int *bs) |
| 663 | { |
| 664 | struct sgio_data *sd = td->io_data; |
| 665 | struct io_u *io_u; |
| 666 | struct sg_io_hdr *hdr; |
| 667 | unsigned char buf[8]; |
| 668 | int ret; |
| 669 | |
| 670 | io_u = __get_io_u(td); |
| 671 | assert(io_u); |
| 672 | |
| 673 | hdr = &io_u->hdr; |
| 674 | sgio_hdr_init(sd, hdr, io_u, 0); |
| 675 | memset(buf, 0, sizeof(buf)); |
| 676 | |
| 677 | hdr->cmdp[0] = 0x25; |
| 678 | hdr->dxfer_direction = SG_DXFER_FROM_DEV; |
| 679 | hdr->dxferp = buf; |
| 680 | hdr->dxfer_len = sizeof(buf); |
| 681 | |
| 682 | ret = fio_sgio_doio(td, io_u, 1); |
| 683 | if (ret) { |
| 684 | put_io_u(td, io_u); |
| 685 | return ret; |
| 686 | } |
| 687 | |
| 688 | *bs = (buf[4] << 24) | (buf[5] << 16) | (buf[6] << 8) | buf[7]; |
| 689 | put_io_u(td, io_u); |
| 690 | return 0; |
| 691 | } |
| 692 | |
| 693 | int fio_sgio_init(struct thread_data *td) |
| 694 | { |
| 695 | struct sgio_data *sd; |
| 696 | unsigned int bs; |
| 697 | int ret; |
| 698 | |
| 699 | sd = malloc(sizeof(*sd)); |
| 700 | sd->cmds = malloc(td->iodepth * sizeof(struct sgio_cmd)); |
| 701 | sd->events = malloc(td->iodepth * sizeof(struct io_u *)); |
| 702 | td->io_data = sd; |
| 703 | |
| 704 | if (td->filetype == FIO_TYPE_BD) { |
| 705 | if (ioctl(td->fd, BLKSSZGET, &bs) < 0) { |
| 706 | td_verror(td, errno); |
| 707 | return 1; |
| 708 | } |
| 709 | } else if (td->filetype == FIO_TYPE_CHAR) { |
| 710 | int version; |
| 711 | |
| 712 | if (ioctl(td->fd, SG_GET_VERSION_NUM, &version) < 0) { |
| 713 | td_verror(td, errno); |
| 714 | return 1; |
| 715 | } |
| 716 | |
| 717 | ret = fio_sgio_get_bs(td, &bs); |
| 718 | if (ret) |
| 719 | return ret; |
| 720 | } else { |
| 721 | log_err("ioengine sgio only works on block devices\n"); |
| 722 | return 1; |
| 723 | } |
| 724 | |
| 725 | sd->bs = bs; |
| 726 | |
| 727 | td->io_prep = fio_sgio_prep; |
| 728 | td->io_queue = fio_sgio_queue; |
| 729 | |
| 730 | if (td->filetype == FIO_TYPE_BD) |
| 731 | td->io_getevents = fio_syncio_getevents; |
| 732 | else |
| 733 | td->io_getevents = fio_sgio_getevents; |
| 734 | |
| 735 | td->io_event = fio_sgio_event; |
| 736 | td->io_cancel = NULL; |
| 737 | td->io_cleanup = fio_syncio_cleanup; |
| 738 | td->io_sync = fio_sgio_sync; |
| 739 | |
| 740 | /* |
| 741 | * we want to do it, regardless of whether odirect is set or not |
| 742 | */ |
| 743 | td->override_sync = 1; |
| 744 | return 0; |
| 745 | } |
| 746 | |
| 747 | #else /* FIO_HAVE_SGIO */ |
| 748 | |
| 749 | int fio_sgio_init(struct thread_data *td) |
| 750 | { |
| 751 | return EINVAL; |
| 752 | } |
| 753 | |
| 754 | #endif /* FIO_HAVE_SGIO */ |
| 755 | |
| 756 | #ifdef FIO_HAVE_SPLICE |
| 757 | struct spliceio_data { |
| 758 | struct io_u *last_io_u; |
| 759 | int pipe[2]; |
| 760 | }; |
| 761 | |
| 762 | static struct io_u *fio_spliceio_event(struct thread_data *td, int event) |
| 763 | { |
| 764 | struct spliceio_data *sd = td->io_data; |
| 765 | |
| 766 | assert(event == 0); |
| 767 | |
| 768 | return sd->last_io_u; |
| 769 | } |
| 770 | |
| 771 | /* |
| 772 | * For splice reading, we unfortunately cannot (yet) vmsplice the other way. |
| 773 | * So just splice the data from the file into the pipe, and use regular |
| 774 | * read to fill the buffer. Doesn't make a lot of sense, but... |
| 775 | */ |
| 776 | static int fio_splice_read(struct thread_data *td, struct io_u *io_u) |
| 777 | { |
| 778 | struct spliceio_data *sd = td->io_data; |
| 779 | int ret, ret2, buflen; |
| 780 | off_t offset; |
| 781 | void *p; |
| 782 | |
| 783 | offset = io_u->offset; |
| 784 | buflen = io_u->buflen; |
| 785 | p = io_u->buf; |
| 786 | while (buflen) { |
| 787 | int this_len = buflen; |
| 788 | |
| 789 | if (this_len > SPLICE_DEF_SIZE) |
| 790 | this_len = SPLICE_DEF_SIZE; |
| 791 | |
| 792 | ret = splice(td->fd, &offset, sd->pipe[1], NULL, this_len, SPLICE_F_MORE); |
| 793 | if (ret < 0) { |
| 794 | if (errno == ENODATA || errno == EAGAIN) |
| 795 | continue; |
| 796 | |
| 797 | return errno; |
| 798 | } |
| 799 | |
| 800 | buflen -= ret; |
| 801 | |
| 802 | while (ret) { |
| 803 | ret2 = read(sd->pipe[0], p, ret); |
| 804 | if (ret2 < 0) |
| 805 | return errno; |
| 806 | |
| 807 | ret -= ret2; |
| 808 | p += ret2; |
| 809 | } |
| 810 | } |
| 811 | |
| 812 | return io_u->buflen; |
| 813 | } |
| 814 | |
| 815 | /* |
| 816 | * For splice writing, we can vmsplice our data buffer directly into a |
| 817 | * pipe and then splice that to a file. |
| 818 | */ |
| 819 | static int fio_splice_write(struct thread_data *td, struct io_u *io_u) |
| 820 | { |
| 821 | struct spliceio_data *sd = td->io_data; |
| 822 | struct iovec iov[1] = { |
| 823 | { |
| 824 | .iov_base = io_u->buf, |
| 825 | .iov_len = io_u->buflen, |
| 826 | } |
| 827 | }; |
| 828 | struct pollfd pfd = { .fd = sd->pipe[1], .events = POLLOUT, }; |
| 829 | off_t off = io_u->offset; |
| 830 | int ret, ret2; |
| 831 | |
| 832 | while (iov[0].iov_len) { |
| 833 | if (poll(&pfd, 1, -1) < 0) |
| 834 | return errno; |
| 835 | |
| 836 | ret = vmsplice(sd->pipe[1], iov, 1, SPLICE_F_NONBLOCK); |
| 837 | if (ret < 0) |
| 838 | return errno; |
| 839 | |
| 840 | iov[0].iov_len -= ret; |
| 841 | iov[0].iov_base += ret; |
| 842 | |
| 843 | while (ret) { |
| 844 | ret2 = splice(sd->pipe[0], NULL, td->fd, &off, ret, 0); |
| 845 | if (ret2 < 0) |
| 846 | return errno; |
| 847 | |
| 848 | ret -= ret2; |
| 849 | } |
| 850 | } |
| 851 | |
| 852 | return io_u->buflen; |
| 853 | } |
| 854 | |
| 855 | static int fio_spliceio_queue(struct thread_data *td, struct io_u *io_u) |
| 856 | { |
| 857 | struct spliceio_data *sd = td->io_data; |
| 858 | int ret; |
| 859 | |
| 860 | if (io_u->ddir == DDIR_READ) |
| 861 | ret = fio_splice_read(td, io_u); |
| 862 | else |
| 863 | ret = fio_splice_write(td, io_u); |
| 864 | |
| 865 | if ((unsigned int) ret != io_u->buflen) { |
| 866 | if (ret > 0) { |
| 867 | io_u->resid = io_u->buflen - ret; |
| 868 | io_u->error = ENODATA; |
| 869 | } else |
| 870 | io_u->error = errno; |
| 871 | } |
| 872 | |
| 873 | if (!io_u->error) |
| 874 | sd->last_io_u = io_u; |
| 875 | |
| 876 | return io_u->error; |
| 877 | } |
| 878 | |
| 879 | static void fio_spliceio_cleanup(struct thread_data *td) |
| 880 | { |
| 881 | struct spliceio_data *sd = td->io_data; |
| 882 | |
| 883 | if (sd) { |
| 884 | close(sd->pipe[0]); |
| 885 | close(sd->pipe[1]); |
| 886 | free(sd); |
| 887 | td->io_data = NULL; |
| 888 | } |
| 889 | } |
| 890 | |
| 891 | int fio_spliceio_init(struct thread_data *td) |
| 892 | { |
| 893 | struct spliceio_data *sd = malloc(sizeof(*sd)); |
| 894 | |
| 895 | td->io_queue = fio_spliceio_queue; |
| 896 | td->io_getevents = fio_syncio_getevents; |
| 897 | td->io_event = fio_spliceio_event; |
| 898 | td->io_cancel = NULL; |
| 899 | td->io_cleanup = fio_spliceio_cleanup; |
| 900 | td->io_sync = fio_io_sync; |
| 901 | |
| 902 | sd->last_io_u = NULL; |
| 903 | if (pipe(sd->pipe) < 0) { |
| 904 | td_verror(td, errno); |
| 905 | free(sd); |
| 906 | return 1; |
| 907 | } |
| 908 | |
| 909 | td->io_data = sd; |
| 910 | return 0; |
| 911 | } |
| 912 | |
| 913 | #else /* FIO_HAVE_SPLICE */ |
| 914 | |
| 915 | int fio_spliceio_init(struct thread_data *td) |
| 916 | { |
| 917 | return EINVAL; |
| 918 | } |
| 919 | |
| 920 | #endif /* FIO_HAVE_SPLICE */ |