1 /* 2 * QEMU Block driver for RADOS (Ceph) 3 * 4 * Copyright (C) 2010-2011 Christian Brunner <chb@muc.de>, 5 * Josh Durgin <josh.durgin@dreamhost.com> 6 * 7 * This work is licensed under the terms of the GNU GPL, version 2. See 8 * the COPYING file in the top-level directory. 9 * 10 * Contributions after 2012-01-13 are licensed under the terms of the 11 * GNU GPL, version 2 or (at your option) any later version. 12 */ 13 14 #include <inttypes.h> 15 16 #include "qemu-common.h" 17 #include "qemu/error-report.h" 18 #include "block/block_int.h" 19 20 #include <rbd/librbd.h> 21 22 /* 23 * When specifying the image filename use: 24 * 25 * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]] 26 * 27 * poolname must be the name of an existing rados pool. 28 * 29 * devicename is the name of the rbd image. 30 * 31 * Each option given is used to configure rados, and may be any valid 32 * Ceph option, "id", or "conf". 33 * 34 * The "id" option indicates what user we should authenticate as to 35 * the Ceph cluster. If it is excluded we will use the Ceph default 36 * (normally 'admin'). 37 * 38 * The "conf" option specifies a Ceph configuration file to read. If 39 * it is not specified, we will read from the default Ceph locations 40 * (e.g., /etc/ceph/ceph.conf). To avoid reading _any_ configuration 41 * file, specify conf=/dev/null. 42 * 43 * Configuration values containing :, @, or = can be escaped with a 44 * leading "\". 45 */ 46 47 /* rbd_aio_discard added in 0.1.2 */ 48 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2) 49 #define LIBRBD_SUPPORTS_DISCARD 50 #else 51 #undef LIBRBD_SUPPORTS_DISCARD 52 #endif 53 54 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER) 55 56 #define RBD_MAX_CONF_NAME_SIZE 128 57 #define RBD_MAX_CONF_VAL_SIZE 512 58 #define RBD_MAX_CONF_SIZE 1024 59 #define RBD_MAX_POOL_NAME_SIZE 128 60 #define RBD_MAX_SNAP_NAME_SIZE 128 61 #define RBD_MAX_SNAPS 100 62 63 typedef enum { 64 RBD_AIO_READ, 65 RBD_AIO_WRITE, 66 RBD_AIO_DISCARD 67 } RBDAIOCmd; 68 69 typedef struct RBDAIOCB { 70 BlockDriverAIOCB common; 71 QEMUBH *bh; 72 int64_t ret; 73 QEMUIOVector *qiov; 74 char *bounce; 75 RBDAIOCmd cmd; 76 int64_t sector_num; 77 int error; 78 struct BDRVRBDState *s; 79 int cancelled; 80 int status; 81 } RBDAIOCB; 82 83 typedef struct RADOSCB { 84 int rcbid; 85 RBDAIOCB *acb; 86 struct BDRVRBDState *s; 87 int done; 88 int64_t size; 89 char *buf; 90 int64_t ret; 91 } RADOSCB; 92 93 #define RBD_FD_READ 0 94 #define RBD_FD_WRITE 1 95 96 typedef struct BDRVRBDState { 97 int fds[2]; 98 rados_t cluster; 99 rados_ioctx_t io_ctx; 100 rbd_image_t image; 101 char name[RBD_MAX_IMAGE_NAME_SIZE]; 102 int qemu_aio_count; 103 char *snap; 104 int event_reader_pos; 105 RADOSCB *event_rcb; 106 } BDRVRBDState; 107 108 static void rbd_aio_bh_cb(void *opaque); 109 110 static int qemu_rbd_next_tok(char *dst, int dst_len, 111 char *src, char delim, 112 const char *name, 113 char **p) 114 { 115 int l; 116 char *end; 117 118 *p = NULL; 119 120 if (delim != '\0') { 121 for (end = src; *end; ++end) { 122 if (*end == delim) { 123 break; 124 } 125 if (*end == '\\' && end[1] != '\0') { 126 end++; 127 } 128 } 129 if (*end == delim) { 130 *p = end + 1; 131 *end = '\0'; 132 } 133 } 134 l = strlen(src); 135 if (l >= dst_len) { 136 error_report("%s too long", name); 137 return -EINVAL; 138 } else if (l == 0) { 139 error_report("%s too short", name); 140 return -EINVAL; 141 } 142 143 pstrcpy(dst, dst_len, src); 144 145 return 0; 146 } 147 148 static void qemu_rbd_unescape(char *src) 149 { 150 char *p; 151 152 for (p = src; *src; ++src, ++p) { 153 if (*src == '\\' && src[1] != '\0') { 154 src++; 155 } 156 *p = *src; 157 } 158 *p = '\0'; 159 } 160 161 static int qemu_rbd_parsename(const char *filename, 162 char *pool, int pool_len, 163 char *snap, int snap_len, 164 char *name, int name_len, 165 char *conf, int conf_len) 166 { 167 const char *start; 168 char *p, *buf; 169 int ret; 170 171 if (!strstart(filename, "rbd:", &start)) { 172 return -EINVAL; 173 } 174 175 buf = g_strdup(start); 176 p = buf; 177 *snap = '\0'; 178 *conf = '\0'; 179 180 ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p); 181 if (ret < 0 || !p) { 182 ret = -EINVAL; 183 goto done; 184 } 185 qemu_rbd_unescape(pool); 186 187 if (strchr(p, '@')) { 188 ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p); 189 if (ret < 0) { 190 goto done; 191 } 192 ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p); 193 qemu_rbd_unescape(snap); 194 } else { 195 ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p); 196 } 197 qemu_rbd_unescape(name); 198 if (ret < 0 || !p) { 199 goto done; 200 } 201 202 ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p); 203 204 done: 205 g_free(buf); 206 return ret; 207 } 208 209 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname) 210 { 211 const char *p = conf; 212 213 while (*p) { 214 int len; 215 const char *end = strchr(p, ':'); 216 217 if (end) { 218 len = end - p; 219 } else { 220 len = strlen(p); 221 } 222 223 if (strncmp(p, "id=", 3) == 0) { 224 len -= 3; 225 strncpy(clientname, p + 3, len); 226 clientname[len] = '\0'; 227 return clientname; 228 } 229 if (end == NULL) { 230 break; 231 } 232 p = end + 1; 233 } 234 return NULL; 235 } 236 237 static int qemu_rbd_set_conf(rados_t cluster, const char *conf) 238 { 239 char *p, *buf; 240 char name[RBD_MAX_CONF_NAME_SIZE]; 241 char value[RBD_MAX_CONF_VAL_SIZE]; 242 int ret = 0; 243 244 buf = g_strdup(conf); 245 p = buf; 246 247 while (p) { 248 ret = qemu_rbd_next_tok(name, sizeof(name), p, 249 '=', "conf option name", &p); 250 if (ret < 0) { 251 break; 252 } 253 qemu_rbd_unescape(name); 254 255 if (!p) { 256 error_report("conf option %s has no value", name); 257 ret = -EINVAL; 258 break; 259 } 260 261 ret = qemu_rbd_next_tok(value, sizeof(value), p, 262 ':', "conf option value", &p); 263 if (ret < 0) { 264 break; 265 } 266 qemu_rbd_unescape(value); 267 268 if (strcmp(name, "conf") == 0) { 269 ret = rados_conf_read_file(cluster, value); 270 if (ret < 0) { 271 error_report("error reading conf file %s", value); 272 break; 273 } 274 } else if (strcmp(name, "id") == 0) { 275 /* ignore, this is parsed by qemu_rbd_parse_clientname() */ 276 } else { 277 ret = rados_conf_set(cluster, name, value); 278 if (ret < 0) { 279 error_report("invalid conf option %s", name); 280 ret = -EINVAL; 281 break; 282 } 283 } 284 } 285 286 g_free(buf); 287 return ret; 288 } 289 290 static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options) 291 { 292 int64_t bytes = 0; 293 int64_t objsize; 294 int obj_order = 0; 295 char pool[RBD_MAX_POOL_NAME_SIZE]; 296 char name[RBD_MAX_IMAGE_NAME_SIZE]; 297 char snap_buf[RBD_MAX_SNAP_NAME_SIZE]; 298 char conf[RBD_MAX_CONF_SIZE]; 299 char clientname_buf[RBD_MAX_CONF_SIZE]; 300 char *clientname; 301 rados_t cluster; 302 rados_ioctx_t io_ctx; 303 int ret; 304 305 if (qemu_rbd_parsename(filename, pool, sizeof(pool), 306 snap_buf, sizeof(snap_buf), 307 name, sizeof(name), 308 conf, sizeof(conf)) < 0) { 309 return -EINVAL; 310 } 311 312 /* Read out options */ 313 while (options && options->name) { 314 if (!strcmp(options->name, BLOCK_OPT_SIZE)) { 315 bytes = options->value.n; 316 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { 317 if (options->value.n) { 318 objsize = options->value.n; 319 if ((objsize - 1) & objsize) { /* not a power of 2? */ 320 error_report("obj size needs to be power of 2"); 321 return -EINVAL; 322 } 323 if (objsize < 4096) { 324 error_report("obj size too small"); 325 return -EINVAL; 326 } 327 obj_order = ffs(objsize) - 1; 328 } 329 } 330 options++; 331 } 332 333 clientname = qemu_rbd_parse_clientname(conf, clientname_buf); 334 if (rados_create(&cluster, clientname) < 0) { 335 error_report("error initializing"); 336 return -EIO; 337 } 338 339 if (strstr(conf, "conf=") == NULL) { 340 /* try default location, but ignore failure */ 341 rados_conf_read_file(cluster, NULL); 342 } 343 344 if (conf[0] != '\0' && 345 qemu_rbd_set_conf(cluster, conf) < 0) { 346 error_report("error setting config options"); 347 rados_shutdown(cluster); 348 return -EIO; 349 } 350 351 if (rados_connect(cluster) < 0) { 352 error_report("error connecting"); 353 rados_shutdown(cluster); 354 return -EIO; 355 } 356 357 if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) { 358 error_report("error opening pool %s", pool); 359 rados_shutdown(cluster); 360 return -EIO; 361 } 362 363 ret = rbd_create(io_ctx, name, bytes, &obj_order); 364 rados_ioctx_destroy(io_ctx); 365 rados_shutdown(cluster); 366 367 return ret; 368 } 369 370 /* 371 * This aio completion is being called from qemu_rbd_aio_event_reader() 372 * and runs in qemu context. It schedules a bh, but just in case the aio 373 * was not cancelled before. 374 */ 375 static void qemu_rbd_complete_aio(RADOSCB *rcb) 376 { 377 RBDAIOCB *acb = rcb->acb; 378 int64_t r; 379 380 r = rcb->ret; 381 382 if (acb->cmd == RBD_AIO_WRITE || 383 acb->cmd == RBD_AIO_DISCARD) { 384 if (r < 0) { 385 acb->ret = r; 386 acb->error = 1; 387 } else if (!acb->error) { 388 acb->ret = rcb->size; 389 } 390 } else { 391 if (r < 0) { 392 memset(rcb->buf, 0, rcb->size); 393 acb->ret = r; 394 acb->error = 1; 395 } else if (r < rcb->size) { 396 memset(rcb->buf + r, 0, rcb->size - r); 397 if (!acb->error) { 398 acb->ret = rcb->size; 399 } 400 } else if (!acb->error) { 401 acb->ret = r; 402 } 403 } 404 /* Note that acb->bh can be NULL in case where the aio was cancelled */ 405 acb->bh = qemu_bh_new(rbd_aio_bh_cb, acb); 406 qemu_bh_schedule(acb->bh); 407 g_free(rcb); 408 } 409 410 /* 411 * aio fd read handler. It runs in the qemu context and calls the 412 * completion handling of completed rados aio operations. 413 */ 414 static void qemu_rbd_aio_event_reader(void *opaque) 415 { 416 BDRVRBDState *s = opaque; 417 418 ssize_t ret; 419 420 do { 421 char *p = (char *)&s->event_rcb; 422 423 /* now read the rcb pointer that was sent from a non qemu thread */ 424 ret = read(s->fds[RBD_FD_READ], p + s->event_reader_pos, 425 sizeof(s->event_rcb) - s->event_reader_pos); 426 if (ret > 0) { 427 s->event_reader_pos += ret; 428 if (s->event_reader_pos == sizeof(s->event_rcb)) { 429 s->event_reader_pos = 0; 430 qemu_rbd_complete_aio(s->event_rcb); 431 s->qemu_aio_count--; 432 } 433 } 434 } while (ret < 0 && errno == EINTR); 435 } 436 437 static int qemu_rbd_aio_flush_cb(void *opaque) 438 { 439 BDRVRBDState *s = opaque; 440 441 return (s->qemu_aio_count > 0); 442 } 443 444 static int qemu_rbd_open(BlockDriverState *bs, const char *filename, int flags) 445 { 446 BDRVRBDState *s = bs->opaque; 447 char pool[RBD_MAX_POOL_NAME_SIZE]; 448 char snap_buf[RBD_MAX_SNAP_NAME_SIZE]; 449 char conf[RBD_MAX_CONF_SIZE]; 450 char clientname_buf[RBD_MAX_CONF_SIZE]; 451 char *clientname; 452 int r; 453 454 if (qemu_rbd_parsename(filename, pool, sizeof(pool), 455 snap_buf, sizeof(snap_buf), 456 s->name, sizeof(s->name), 457 conf, sizeof(conf)) < 0) { 458 return -EINVAL; 459 } 460 461 clientname = qemu_rbd_parse_clientname(conf, clientname_buf); 462 r = rados_create(&s->cluster, clientname); 463 if (r < 0) { 464 error_report("error initializing"); 465 return r; 466 } 467 468 s->snap = NULL; 469 if (snap_buf[0] != '\0') { 470 s->snap = g_strdup(snap_buf); 471 } 472 473 /* 474 * Fallback to more conservative semantics if setting cache 475 * options fails. Ignore errors from setting rbd_cache because the 476 * only possible error is that the option does not exist, and 477 * librbd defaults to no caching. If write through caching cannot 478 * be set up, fall back to no caching. 479 */ 480 if (flags & BDRV_O_NOCACHE) { 481 rados_conf_set(s->cluster, "rbd_cache", "false"); 482 } else { 483 rados_conf_set(s->cluster, "rbd_cache", "true"); 484 } 485 486 if (strstr(conf, "conf=") == NULL) { 487 /* try default location, but ignore failure */ 488 rados_conf_read_file(s->cluster, NULL); 489 } 490 491 if (conf[0] != '\0') { 492 r = qemu_rbd_set_conf(s->cluster, conf); 493 if (r < 0) { 494 error_report("error setting config options"); 495 goto failed_shutdown; 496 } 497 } 498 499 r = rados_connect(s->cluster); 500 if (r < 0) { 501 error_report("error connecting"); 502 goto failed_shutdown; 503 } 504 505 r = rados_ioctx_create(s->cluster, pool, &s->io_ctx); 506 if (r < 0) { 507 error_report("error opening pool %s", pool); 508 goto failed_shutdown; 509 } 510 511 r = rbd_open(s->io_ctx, s->name, &s->image, s->snap); 512 if (r < 0) { 513 error_report("error reading header from %s", s->name); 514 goto failed_open; 515 } 516 517 bs->read_only = (s->snap != NULL); 518 519 s->event_reader_pos = 0; 520 r = qemu_pipe(s->fds); 521 if (r < 0) { 522 error_report("error opening eventfd"); 523 goto failed; 524 } 525 fcntl(s->fds[0], F_SETFL, O_NONBLOCK); 526 fcntl(s->fds[1], F_SETFL, O_NONBLOCK); 527 qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader, 528 NULL, qemu_rbd_aio_flush_cb, s); 529 530 531 return 0; 532 533 failed: 534 rbd_close(s->image); 535 failed_open: 536 rados_ioctx_destroy(s->io_ctx); 537 failed_shutdown: 538 rados_shutdown(s->cluster); 539 g_free(s->snap); 540 return r; 541 } 542 543 static void qemu_rbd_close(BlockDriverState *bs) 544 { 545 BDRVRBDState *s = bs->opaque; 546 547 close(s->fds[0]); 548 close(s->fds[1]); 549 qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL, NULL, NULL, NULL); 550 551 rbd_close(s->image); 552 rados_ioctx_destroy(s->io_ctx); 553 g_free(s->snap); 554 rados_shutdown(s->cluster); 555 } 556 557 /* 558 * Cancel aio. Since we don't reference acb in a non qemu threads, 559 * it is safe to access it here. 560 */ 561 static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb) 562 { 563 RBDAIOCB *acb = (RBDAIOCB *) blockacb; 564 acb->cancelled = 1; 565 566 while (acb->status == -EINPROGRESS) { 567 qemu_aio_wait(); 568 } 569 570 qemu_aio_release(acb); 571 } 572 573 static const AIOCBInfo rbd_aiocb_info = { 574 .aiocb_size = sizeof(RBDAIOCB), 575 .cancel = qemu_rbd_aio_cancel, 576 }; 577 578 static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb) 579 { 580 int ret = 0; 581 while (1) { 582 fd_set wfd; 583 int fd = s->fds[RBD_FD_WRITE]; 584 585 /* send the op pointer to the qemu thread that is responsible 586 for the aio/op completion. Must do it in a qemu thread context */ 587 ret = write(fd, (void *)&rcb, sizeof(rcb)); 588 if (ret >= 0) { 589 break; 590 } 591 if (errno == EINTR) { 592 continue; 593 } 594 if (errno != EAGAIN) { 595 break; 596 } 597 598 FD_ZERO(&wfd); 599 FD_SET(fd, &wfd); 600 do { 601 ret = select(fd + 1, NULL, &wfd, NULL, NULL); 602 } while (ret < 0 && errno == EINTR); 603 } 604 605 return ret; 606 } 607 608 /* 609 * This is the callback function for rbd_aio_read and _write 610 * 611 * Note: this function is being called from a non qemu thread so 612 * we need to be careful about what we do here. Generally we only 613 * write to the block notification pipe, and do the rest of the 614 * io completion handling from qemu_rbd_aio_event_reader() which 615 * runs in a qemu context. 616 */ 617 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb) 618 { 619 int ret; 620 rcb->ret = rbd_aio_get_return_value(c); 621 rbd_aio_release(c); 622 ret = qemu_rbd_send_pipe(rcb->s, rcb); 623 if (ret < 0) { 624 error_report("failed writing to acb->s->fds"); 625 g_free(rcb); 626 } 627 } 628 629 /* Callback when all queued rbd_aio requests are complete */ 630 631 static void rbd_aio_bh_cb(void *opaque) 632 { 633 RBDAIOCB *acb = opaque; 634 635 if (acb->cmd == RBD_AIO_READ) { 636 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size); 637 } 638 qemu_vfree(acb->bounce); 639 acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret)); 640 qemu_bh_delete(acb->bh); 641 acb->bh = NULL; 642 acb->status = 0; 643 644 if (!acb->cancelled) { 645 qemu_aio_release(acb); 646 } 647 } 648 649 static int rbd_aio_discard_wrapper(rbd_image_t image, 650 uint64_t off, 651 uint64_t len, 652 rbd_completion_t comp) 653 { 654 #ifdef LIBRBD_SUPPORTS_DISCARD 655 return rbd_aio_discard(image, off, len, comp); 656 #else 657 return -ENOTSUP; 658 #endif 659 } 660 661 static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs, 662 int64_t sector_num, 663 QEMUIOVector *qiov, 664 int nb_sectors, 665 BlockDriverCompletionFunc *cb, 666 void *opaque, 667 RBDAIOCmd cmd) 668 { 669 RBDAIOCB *acb; 670 RADOSCB *rcb; 671 rbd_completion_t c; 672 int64_t off, size; 673 char *buf; 674 int r; 675 676 BDRVRBDState *s = bs->opaque; 677 678 acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque); 679 acb->cmd = cmd; 680 acb->qiov = qiov; 681 if (cmd == RBD_AIO_DISCARD) { 682 acb->bounce = NULL; 683 } else { 684 acb->bounce = qemu_blockalign(bs, qiov->size); 685 } 686 acb->ret = 0; 687 acb->error = 0; 688 acb->s = s; 689 acb->cancelled = 0; 690 acb->bh = NULL; 691 acb->status = -EINPROGRESS; 692 693 if (cmd == RBD_AIO_WRITE) { 694 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size); 695 } 696 697 buf = acb->bounce; 698 699 off = sector_num * BDRV_SECTOR_SIZE; 700 size = nb_sectors * BDRV_SECTOR_SIZE; 701 702 s->qemu_aio_count++; /* All the RADOSCB */ 703 704 rcb = g_malloc(sizeof(RADOSCB)); 705 rcb->done = 0; 706 rcb->acb = acb; 707 rcb->buf = buf; 708 rcb->s = acb->s; 709 rcb->size = size; 710 r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c); 711 if (r < 0) { 712 goto failed; 713 } 714 715 switch (cmd) { 716 case RBD_AIO_WRITE: 717 r = rbd_aio_write(s->image, off, size, buf, c); 718 break; 719 case RBD_AIO_READ: 720 r = rbd_aio_read(s->image, off, size, buf, c); 721 break; 722 case RBD_AIO_DISCARD: 723 r = rbd_aio_discard_wrapper(s->image, off, size, c); 724 break; 725 default: 726 r = -EINVAL; 727 } 728 729 if (r < 0) { 730 goto failed; 731 } 732 733 return &acb->common; 734 735 failed: 736 g_free(rcb); 737 s->qemu_aio_count--; 738 qemu_aio_release(acb); 739 return NULL; 740 } 741 742 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs, 743 int64_t sector_num, 744 QEMUIOVector *qiov, 745 int nb_sectors, 746 BlockDriverCompletionFunc *cb, 747 void *opaque) 748 { 749 return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque, 750 RBD_AIO_READ); 751 } 752 753 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs, 754 int64_t sector_num, 755 QEMUIOVector *qiov, 756 int nb_sectors, 757 BlockDriverCompletionFunc *cb, 758 void *opaque) 759 { 760 return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque, 761 RBD_AIO_WRITE); 762 } 763 764 static int qemu_rbd_co_flush(BlockDriverState *bs) 765 { 766 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1) 767 /* rbd_flush added in 0.1.1 */ 768 BDRVRBDState *s = bs->opaque; 769 return rbd_flush(s->image); 770 #else 771 return 0; 772 #endif 773 } 774 775 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi) 776 { 777 BDRVRBDState *s = bs->opaque; 778 rbd_image_info_t info; 779 int r; 780 781 r = rbd_stat(s->image, &info, sizeof(info)); 782 if (r < 0) { 783 return r; 784 } 785 786 bdi->cluster_size = info.obj_size; 787 return 0; 788 } 789 790 static int64_t qemu_rbd_getlength(BlockDriverState *bs) 791 { 792 BDRVRBDState *s = bs->opaque; 793 rbd_image_info_t info; 794 int r; 795 796 r = rbd_stat(s->image, &info, sizeof(info)); 797 if (r < 0) { 798 return r; 799 } 800 801 return info.size; 802 } 803 804 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset) 805 { 806 BDRVRBDState *s = bs->opaque; 807 int r; 808 809 r = rbd_resize(s->image, offset); 810 if (r < 0) { 811 return r; 812 } 813 814 return 0; 815 } 816 817 static int qemu_rbd_snap_create(BlockDriverState *bs, 818 QEMUSnapshotInfo *sn_info) 819 { 820 BDRVRBDState *s = bs->opaque; 821 int r; 822 823 if (sn_info->name[0] == '\0') { 824 return -EINVAL; /* we need a name for rbd snapshots */ 825 } 826 827 /* 828 * rbd snapshots are using the name as the user controlled unique identifier 829 * we can't use the rbd snapid for that purpose, as it can't be set 830 */ 831 if (sn_info->id_str[0] != '\0' && 832 strcmp(sn_info->id_str, sn_info->name) != 0) { 833 return -EINVAL; 834 } 835 836 if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) { 837 return -ERANGE; 838 } 839 840 r = rbd_snap_create(s->image, sn_info->name); 841 if (r < 0) { 842 error_report("failed to create snap: %s", strerror(-r)); 843 return r; 844 } 845 846 return 0; 847 } 848 849 static int qemu_rbd_snap_remove(BlockDriverState *bs, 850 const char *snapshot_name) 851 { 852 BDRVRBDState *s = bs->opaque; 853 int r; 854 855 r = rbd_snap_remove(s->image, snapshot_name); 856 return r; 857 } 858 859 static int qemu_rbd_snap_rollback(BlockDriverState *bs, 860 const char *snapshot_name) 861 { 862 BDRVRBDState *s = bs->opaque; 863 int r; 864 865 r = rbd_snap_rollback(s->image, snapshot_name); 866 return r; 867 } 868 869 static int qemu_rbd_snap_list(BlockDriverState *bs, 870 QEMUSnapshotInfo **psn_tab) 871 { 872 BDRVRBDState *s = bs->opaque; 873 QEMUSnapshotInfo *sn_info, *sn_tab = NULL; 874 int i, snap_count; 875 rbd_snap_info_t *snaps; 876 int max_snaps = RBD_MAX_SNAPS; 877 878 do { 879 snaps = g_malloc(sizeof(*snaps) * max_snaps); 880 snap_count = rbd_snap_list(s->image, snaps, &max_snaps); 881 if (snap_count < 0) { 882 g_free(snaps); 883 } 884 } while (snap_count == -ERANGE); 885 886 if (snap_count <= 0) { 887 goto done; 888 } 889 890 sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo)); 891 892 for (i = 0; i < snap_count; i++) { 893 const char *snap_name = snaps[i].name; 894 895 sn_info = sn_tab + i; 896 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name); 897 pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name); 898 899 sn_info->vm_state_size = snaps[i].size; 900 sn_info->date_sec = 0; 901 sn_info->date_nsec = 0; 902 sn_info->vm_clock_nsec = 0; 903 } 904 rbd_snap_list_end(snaps); 905 906 done: 907 *psn_tab = sn_tab; 908 return snap_count; 909 } 910 911 #ifdef LIBRBD_SUPPORTS_DISCARD 912 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs, 913 int64_t sector_num, 914 int nb_sectors, 915 BlockDriverCompletionFunc *cb, 916 void *opaque) 917 { 918 return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque, 919 RBD_AIO_DISCARD); 920 } 921 #endif 922 923 static QEMUOptionParameter qemu_rbd_create_options[] = { 924 { 925 .name = BLOCK_OPT_SIZE, 926 .type = OPT_SIZE, 927 .help = "Virtual disk size" 928 }, 929 { 930 .name = BLOCK_OPT_CLUSTER_SIZE, 931 .type = OPT_SIZE, 932 .help = "RBD object size" 933 }, 934 {NULL} 935 }; 936 937 static BlockDriver bdrv_rbd = { 938 .format_name = "rbd", 939 .instance_size = sizeof(BDRVRBDState), 940 .bdrv_file_open = qemu_rbd_open, 941 .bdrv_close = qemu_rbd_close, 942 .bdrv_create = qemu_rbd_create, 943 .bdrv_get_info = qemu_rbd_getinfo, 944 .create_options = qemu_rbd_create_options, 945 .bdrv_getlength = qemu_rbd_getlength, 946 .bdrv_truncate = qemu_rbd_truncate, 947 .protocol_name = "rbd", 948 949 .bdrv_aio_readv = qemu_rbd_aio_readv, 950 .bdrv_aio_writev = qemu_rbd_aio_writev, 951 .bdrv_co_flush_to_disk = qemu_rbd_co_flush, 952 953 #ifdef LIBRBD_SUPPORTS_DISCARD 954 .bdrv_aio_discard = qemu_rbd_aio_discard, 955 #endif 956 957 .bdrv_snapshot_create = qemu_rbd_snap_create, 958 .bdrv_snapshot_delete = qemu_rbd_snap_remove, 959 .bdrv_snapshot_list = qemu_rbd_snap_list, 960 .bdrv_snapshot_goto = qemu_rbd_snap_rollback, 961 }; 962 963 static void bdrv_rbd_init(void) 964 { 965 bdrv_register(&bdrv_rbd); 966 } 967 968 block_init(bdrv_rbd_init); 969