1 /* 2 * Replication Block filter 3 * 4 * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD. 5 * Copyright (c) 2016 Intel Corporation 6 * Copyright (c) 2016 FUJITSU LIMITED 7 * 8 * Author: 9 * Wen Congyang <wency@cn.fujitsu.com> 10 * 11 * This work is licensed under the terms of the GNU GPL, version 2 or later. 12 * See the COPYING file in the top-level directory. 13 */ 14 15 #include "qemu/osdep.h" 16 #include "qemu-common.h" 17 #include "block/nbd.h" 18 #include "block/blockjob.h" 19 #include "block/block_int.h" 20 #include "block/block_backup.h" 21 #include "sysemu/block-backend.h" 22 #include "qapi/error.h" 23 #include "replication.h" 24 25 typedef enum { 26 BLOCK_REPLICATION_NONE, /* block replication is not started */ 27 BLOCK_REPLICATION_RUNNING, /* block replication is running */ 28 BLOCK_REPLICATION_FAILOVER, /* failover is running in background */ 29 BLOCK_REPLICATION_FAILOVER_FAILED, /* failover failed */ 30 BLOCK_REPLICATION_DONE, /* block replication is done */ 31 } ReplicationStage; 32 33 typedef struct BDRVReplicationState { 34 ReplicationMode mode; 35 ReplicationStage stage; 36 BdrvChild *active_disk; 37 BdrvChild *hidden_disk; 38 BdrvChild *secondary_disk; 39 char *top_id; 40 ReplicationState *rs; 41 Error *blocker; 42 int orig_hidden_flags; 43 int orig_secondary_flags; 44 int error; 45 } BDRVReplicationState; 46 47 static void replication_start(ReplicationState *rs, ReplicationMode mode, 48 Error **errp); 49 static void replication_do_checkpoint(ReplicationState *rs, Error **errp); 50 static void replication_get_error(ReplicationState *rs, Error **errp); 51 static void replication_stop(ReplicationState *rs, bool failover, 52 Error **errp); 53 54 #define REPLICATION_MODE "mode" 55 #define REPLICATION_TOP_ID "top-id" 56 static QemuOptsList replication_runtime_opts = { 57 .name = "replication", 58 .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head), 59 .desc = { 60 { 61 .name = REPLICATION_MODE, 62 .type = QEMU_OPT_STRING, 63 }, 64 { 65 .name = REPLICATION_TOP_ID, 66 .type = QEMU_OPT_STRING, 67 }, 68 { /* end of list */ } 69 }, 70 }; 71 72 static ReplicationOps replication_ops = { 73 .start = replication_start, 74 .checkpoint = replication_do_checkpoint, 75 .get_error = replication_get_error, 76 .stop = replication_stop, 77 }; 78 79 static int replication_open(BlockDriverState *bs, QDict *options, 80 int flags, Error **errp) 81 { 82 int ret; 83 BDRVReplicationState *s = bs->opaque; 84 Error *local_err = NULL; 85 QemuOpts *opts = NULL; 86 const char *mode; 87 const char *top_id; 88 89 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 90 false, errp); 91 if (!bs->file) { 92 return -EINVAL; 93 } 94 95 ret = -EINVAL; 96 opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort); 97 qemu_opts_absorb_qdict(opts, options, &local_err); 98 if (local_err) { 99 goto fail; 100 } 101 102 mode = qemu_opt_get(opts, REPLICATION_MODE); 103 if (!mode) { 104 error_setg(&local_err, "Missing the option mode"); 105 goto fail; 106 } 107 108 if (!strcmp(mode, "primary")) { 109 s->mode = REPLICATION_MODE_PRIMARY; 110 top_id = qemu_opt_get(opts, REPLICATION_TOP_ID); 111 if (top_id) { 112 error_setg(&local_err, "The primary side does not support option top-id"); 113 goto fail; 114 } 115 } else if (!strcmp(mode, "secondary")) { 116 s->mode = REPLICATION_MODE_SECONDARY; 117 top_id = qemu_opt_get(opts, REPLICATION_TOP_ID); 118 s->top_id = g_strdup(top_id); 119 if (!s->top_id) { 120 error_setg(&local_err, "Missing the option top-id"); 121 goto fail; 122 } 123 } else { 124 error_setg(&local_err, 125 "The option mode's value should be primary or secondary"); 126 goto fail; 127 } 128 129 s->rs = replication_new(bs, &replication_ops); 130 131 ret = 0; 132 133 fail: 134 qemu_opts_del(opts); 135 error_propagate(errp, local_err); 136 137 return ret; 138 } 139 140 static void replication_close(BlockDriverState *bs) 141 { 142 BDRVReplicationState *s = bs->opaque; 143 144 if (s->stage == BLOCK_REPLICATION_RUNNING) { 145 replication_stop(s->rs, false, NULL); 146 } 147 if (s->stage == BLOCK_REPLICATION_FAILOVER) { 148 block_job_cancel_sync(s->active_disk->bs->job); 149 } 150 151 if (s->mode == REPLICATION_MODE_SECONDARY) { 152 g_free(s->top_id); 153 } 154 155 replication_remove(s->rs); 156 } 157 158 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c, 159 const BdrvChildRole *role, 160 BlockReopenQueue *reopen_queue, 161 uint64_t perm, uint64_t shared, 162 uint64_t *nperm, uint64_t *nshared) 163 { 164 *nperm = BLK_PERM_CONSISTENT_READ; 165 if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) { 166 *nperm |= BLK_PERM_WRITE; 167 } 168 *nshared = BLK_PERM_CONSISTENT_READ \ 169 | BLK_PERM_WRITE \ 170 | BLK_PERM_WRITE_UNCHANGED; 171 return; 172 } 173 174 static int64_t replication_getlength(BlockDriverState *bs) 175 { 176 return bdrv_getlength(bs->file->bs); 177 } 178 179 static int replication_get_io_status(BDRVReplicationState *s) 180 { 181 switch (s->stage) { 182 case BLOCK_REPLICATION_NONE: 183 return -EIO; 184 case BLOCK_REPLICATION_RUNNING: 185 return 0; 186 case BLOCK_REPLICATION_FAILOVER: 187 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0; 188 case BLOCK_REPLICATION_FAILOVER_FAILED: 189 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1; 190 case BLOCK_REPLICATION_DONE: 191 /* 192 * active commit job completes, and active disk and secondary_disk 193 * is swapped, so we can operate bs->file directly 194 */ 195 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0; 196 default: 197 abort(); 198 } 199 } 200 201 static int replication_return_value(BDRVReplicationState *s, int ret) 202 { 203 if (s->mode == REPLICATION_MODE_SECONDARY) { 204 return ret; 205 } 206 207 if (ret < 0) { 208 s->error = ret; 209 ret = 0; 210 } 211 212 return ret; 213 } 214 215 static coroutine_fn int replication_co_readv(BlockDriverState *bs, 216 int64_t sector_num, 217 int remaining_sectors, 218 QEMUIOVector *qiov) 219 { 220 BDRVReplicationState *s = bs->opaque; 221 BdrvChild *child = s->secondary_disk; 222 BlockJob *job = NULL; 223 CowRequest req; 224 int ret; 225 226 if (s->mode == REPLICATION_MODE_PRIMARY) { 227 /* We only use it to forward primary write requests */ 228 return -EIO; 229 } 230 231 ret = replication_get_io_status(s); 232 if (ret < 0) { 233 return ret; 234 } 235 236 if (child && child->bs) { 237 job = child->bs->job; 238 } 239 240 if (job) { 241 uint64_t remaining_bytes = remaining_sectors * BDRV_SECTOR_SIZE; 242 243 backup_wait_for_overlapping_requests(child->bs->job, 244 sector_num * BDRV_SECTOR_SIZE, 245 remaining_bytes); 246 backup_cow_request_begin(&req, child->bs->job, 247 sector_num * BDRV_SECTOR_SIZE, 248 remaining_bytes); 249 ret = bdrv_co_readv(bs->file, sector_num, remaining_sectors, 250 qiov); 251 backup_cow_request_end(&req); 252 goto out; 253 } 254 255 ret = bdrv_co_readv(bs->file, sector_num, remaining_sectors, qiov); 256 out: 257 return replication_return_value(s, ret); 258 } 259 260 static coroutine_fn int replication_co_writev(BlockDriverState *bs, 261 int64_t sector_num, 262 int remaining_sectors, 263 QEMUIOVector *qiov) 264 { 265 BDRVReplicationState *s = bs->opaque; 266 QEMUIOVector hd_qiov; 267 uint64_t bytes_done = 0; 268 BdrvChild *top = bs->file; 269 BdrvChild *base = s->secondary_disk; 270 BdrvChild *target; 271 int ret; 272 int64_t n; 273 274 ret = replication_get_io_status(s); 275 if (ret < 0) { 276 goto out; 277 } 278 279 if (ret == 0) { 280 ret = bdrv_co_writev(top, sector_num, 281 remaining_sectors, qiov); 282 return replication_return_value(s, ret); 283 } 284 285 /* 286 * Failover failed, only write to active disk if the sectors 287 * have already been allocated in active disk/hidden disk. 288 */ 289 qemu_iovec_init(&hd_qiov, qiov->niov); 290 while (remaining_sectors > 0) { 291 int64_t count; 292 293 ret = bdrv_is_allocated_above(top->bs, base->bs, 294 sector_num * BDRV_SECTOR_SIZE, 295 remaining_sectors * BDRV_SECTOR_SIZE, 296 &count); 297 if (ret < 0) { 298 goto out1; 299 } 300 301 assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE)); 302 n = count >> BDRV_SECTOR_BITS; 303 qemu_iovec_reset(&hd_qiov); 304 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count); 305 306 target = ret ? top : base; 307 ret = bdrv_co_writev(target, sector_num, n, &hd_qiov); 308 if (ret < 0) { 309 goto out1; 310 } 311 312 remaining_sectors -= n; 313 sector_num += n; 314 bytes_done += count; 315 } 316 317 out1: 318 qemu_iovec_destroy(&hd_qiov); 319 out: 320 return ret; 321 } 322 323 static bool replication_recurse_is_first_non_filter(BlockDriverState *bs, 324 BlockDriverState *candidate) 325 { 326 return bdrv_recurse_is_first_non_filter(bs->file->bs, candidate); 327 } 328 329 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp) 330 { 331 Error *local_err = NULL; 332 int ret; 333 334 if (!s->secondary_disk->bs->job) { 335 error_setg(errp, "Backup job was cancelled unexpectedly"); 336 return; 337 } 338 339 backup_do_checkpoint(s->secondary_disk->bs->job, &local_err); 340 if (local_err) { 341 error_propagate(errp, local_err); 342 return; 343 } 344 345 if (!s->active_disk->bs->drv) { 346 error_setg(errp, "Active disk %s is ejected", 347 s->active_disk->bs->node_name); 348 return; 349 } 350 351 ret = s->active_disk->bs->drv->bdrv_make_empty(s->active_disk->bs); 352 if (ret < 0) { 353 error_setg(errp, "Cannot make active disk empty"); 354 return; 355 } 356 357 if (!s->hidden_disk->bs->drv) { 358 error_setg(errp, "Hidden disk %s is ejected", 359 s->hidden_disk->bs->node_name); 360 return; 361 } 362 363 ret = s->hidden_disk->bs->drv->bdrv_make_empty(s->hidden_disk->bs); 364 if (ret < 0) { 365 error_setg(errp, "Cannot make hidden disk empty"); 366 return; 367 } 368 } 369 370 static void reopen_backing_file(BlockDriverState *bs, bool writable, 371 Error **errp) 372 { 373 BDRVReplicationState *s = bs->opaque; 374 BlockReopenQueue *reopen_queue = NULL; 375 int orig_hidden_flags, orig_secondary_flags; 376 int new_hidden_flags, new_secondary_flags; 377 Error *local_err = NULL; 378 379 if (writable) { 380 orig_hidden_flags = s->orig_hidden_flags = 381 bdrv_get_flags(s->hidden_disk->bs); 382 new_hidden_flags = (orig_hidden_flags | BDRV_O_RDWR) & 383 ~BDRV_O_INACTIVE; 384 orig_secondary_flags = s->orig_secondary_flags = 385 bdrv_get_flags(s->secondary_disk->bs); 386 new_secondary_flags = (orig_secondary_flags | BDRV_O_RDWR) & 387 ~BDRV_O_INACTIVE; 388 } else { 389 orig_hidden_flags = (s->orig_hidden_flags | BDRV_O_RDWR) & 390 ~BDRV_O_INACTIVE; 391 new_hidden_flags = s->orig_hidden_flags; 392 orig_secondary_flags = (s->orig_secondary_flags | BDRV_O_RDWR) & 393 ~BDRV_O_INACTIVE; 394 new_secondary_flags = s->orig_secondary_flags; 395 } 396 397 if (orig_hidden_flags != new_hidden_flags) { 398 reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs, NULL, 399 new_hidden_flags); 400 } 401 402 if (!(orig_secondary_flags & BDRV_O_RDWR)) { 403 reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs, 404 NULL, new_secondary_flags); 405 } 406 407 if (reopen_queue) { 408 bdrv_reopen_multiple(bdrv_get_aio_context(bs), 409 reopen_queue, &local_err); 410 error_propagate(errp, local_err); 411 } 412 } 413 414 static void backup_job_cleanup(BlockDriverState *bs) 415 { 416 BDRVReplicationState *s = bs->opaque; 417 BlockDriverState *top_bs; 418 419 top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL); 420 if (!top_bs) { 421 return; 422 } 423 bdrv_op_unblock_all(top_bs, s->blocker); 424 error_free(s->blocker); 425 reopen_backing_file(bs, false, NULL); 426 } 427 428 static void backup_job_completed(void *opaque, int ret) 429 { 430 BlockDriverState *bs = opaque; 431 BDRVReplicationState *s = bs->opaque; 432 433 if (s->stage != BLOCK_REPLICATION_FAILOVER) { 434 /* The backup job is cancelled unexpectedly */ 435 s->error = -EIO; 436 } 437 438 backup_job_cleanup(bs); 439 } 440 441 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs) 442 { 443 BdrvChild *child; 444 445 /* The bs itself is the top_bs */ 446 if (top_bs == bs) { 447 return true; 448 } 449 450 /* Iterate over top_bs's children */ 451 QLIST_FOREACH(child, &top_bs->children, next) { 452 if (child->bs == bs || check_top_bs(child->bs, bs)) { 453 return true; 454 } 455 } 456 457 return false; 458 } 459 460 static void replication_start(ReplicationState *rs, ReplicationMode mode, 461 Error **errp) 462 { 463 BlockDriverState *bs = rs->opaque; 464 BDRVReplicationState *s; 465 BlockDriverState *top_bs; 466 int64_t active_length, hidden_length, disk_length; 467 AioContext *aio_context; 468 Error *local_err = NULL; 469 BlockJob *job; 470 471 aio_context = bdrv_get_aio_context(bs); 472 aio_context_acquire(aio_context); 473 s = bs->opaque; 474 475 if (s->stage != BLOCK_REPLICATION_NONE) { 476 error_setg(errp, "Block replication is running or done"); 477 aio_context_release(aio_context); 478 return; 479 } 480 481 if (s->mode != mode) { 482 error_setg(errp, "The parameter mode's value is invalid, needs %d," 483 " but got %d", s->mode, mode); 484 aio_context_release(aio_context); 485 return; 486 } 487 488 switch (s->mode) { 489 case REPLICATION_MODE_PRIMARY: 490 break; 491 case REPLICATION_MODE_SECONDARY: 492 s->active_disk = bs->file; 493 if (!s->active_disk || !s->active_disk->bs || 494 !s->active_disk->bs->backing) { 495 error_setg(errp, "Active disk doesn't have backing file"); 496 aio_context_release(aio_context); 497 return; 498 } 499 500 s->hidden_disk = s->active_disk->bs->backing; 501 if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) { 502 error_setg(errp, "Hidden disk doesn't have backing file"); 503 aio_context_release(aio_context); 504 return; 505 } 506 507 s->secondary_disk = s->hidden_disk->bs->backing; 508 if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) { 509 error_setg(errp, "The secondary disk doesn't have block backend"); 510 aio_context_release(aio_context); 511 return; 512 } 513 514 /* verify the length */ 515 active_length = bdrv_getlength(s->active_disk->bs); 516 hidden_length = bdrv_getlength(s->hidden_disk->bs); 517 disk_length = bdrv_getlength(s->secondary_disk->bs); 518 if (active_length < 0 || hidden_length < 0 || disk_length < 0 || 519 active_length != hidden_length || hidden_length != disk_length) { 520 error_setg(errp, "Active disk, hidden disk, secondary disk's length" 521 " are not the same"); 522 aio_context_release(aio_context); 523 return; 524 } 525 526 /* Must be true, or the bdrv_getlength() calls would have failed */ 527 assert(s->active_disk->bs->drv && s->hidden_disk->bs->drv); 528 529 if (!s->active_disk->bs->drv->bdrv_make_empty || 530 !s->hidden_disk->bs->drv->bdrv_make_empty) { 531 error_setg(errp, 532 "Active disk or hidden disk doesn't support make_empty"); 533 aio_context_release(aio_context); 534 return; 535 } 536 537 /* reopen the backing file in r/w mode */ 538 reopen_backing_file(bs, true, &local_err); 539 if (local_err) { 540 error_propagate(errp, local_err); 541 aio_context_release(aio_context); 542 return; 543 } 544 545 /* start backup job now */ 546 error_setg(&s->blocker, 547 "Block device is in use by internal backup job"); 548 549 top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL); 550 if (!top_bs || !bdrv_is_root_node(top_bs) || 551 !check_top_bs(top_bs, bs)) { 552 error_setg(errp, "No top_bs or it is invalid"); 553 reopen_backing_file(bs, false, NULL); 554 aio_context_release(aio_context); 555 return; 556 } 557 bdrv_op_block_all(top_bs, s->blocker); 558 bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker); 559 560 job = backup_job_create(NULL, s->secondary_disk->bs, s->hidden_disk->bs, 561 0, MIRROR_SYNC_MODE_NONE, NULL, false, 562 BLOCKDEV_ON_ERROR_REPORT, 563 BLOCKDEV_ON_ERROR_REPORT, BLOCK_JOB_INTERNAL, 564 backup_job_completed, bs, NULL, &local_err); 565 if (local_err) { 566 error_propagate(errp, local_err); 567 backup_job_cleanup(bs); 568 aio_context_release(aio_context); 569 return; 570 } 571 block_job_start(job); 572 break; 573 default: 574 aio_context_release(aio_context); 575 abort(); 576 } 577 578 s->stage = BLOCK_REPLICATION_RUNNING; 579 580 if (s->mode == REPLICATION_MODE_SECONDARY) { 581 secondary_do_checkpoint(s, errp); 582 } 583 584 s->error = 0; 585 aio_context_release(aio_context); 586 } 587 588 static void replication_do_checkpoint(ReplicationState *rs, Error **errp) 589 { 590 BlockDriverState *bs = rs->opaque; 591 BDRVReplicationState *s; 592 AioContext *aio_context; 593 594 aio_context = bdrv_get_aio_context(bs); 595 aio_context_acquire(aio_context); 596 s = bs->opaque; 597 598 if (s->mode == REPLICATION_MODE_SECONDARY) { 599 secondary_do_checkpoint(s, errp); 600 } 601 aio_context_release(aio_context); 602 } 603 604 static void replication_get_error(ReplicationState *rs, Error **errp) 605 { 606 BlockDriverState *bs = rs->opaque; 607 BDRVReplicationState *s; 608 AioContext *aio_context; 609 610 aio_context = bdrv_get_aio_context(bs); 611 aio_context_acquire(aio_context); 612 s = bs->opaque; 613 614 if (s->stage != BLOCK_REPLICATION_RUNNING) { 615 error_setg(errp, "Block replication is not running"); 616 aio_context_release(aio_context); 617 return; 618 } 619 620 if (s->error) { 621 error_setg(errp, "I/O error occurred"); 622 aio_context_release(aio_context); 623 return; 624 } 625 aio_context_release(aio_context); 626 } 627 628 static void replication_done(void *opaque, int ret) 629 { 630 BlockDriverState *bs = opaque; 631 BDRVReplicationState *s = bs->opaque; 632 633 if (ret == 0) { 634 s->stage = BLOCK_REPLICATION_DONE; 635 636 /* refresh top bs's filename */ 637 bdrv_refresh_filename(bs); 638 s->active_disk = NULL; 639 s->secondary_disk = NULL; 640 s->hidden_disk = NULL; 641 s->error = 0; 642 } else { 643 s->stage = BLOCK_REPLICATION_FAILOVER_FAILED; 644 s->error = -EIO; 645 } 646 } 647 648 static void replication_stop(ReplicationState *rs, bool failover, Error **errp) 649 { 650 BlockDriverState *bs = rs->opaque; 651 BDRVReplicationState *s; 652 AioContext *aio_context; 653 654 aio_context = bdrv_get_aio_context(bs); 655 aio_context_acquire(aio_context); 656 s = bs->opaque; 657 658 if (s->stage != BLOCK_REPLICATION_RUNNING) { 659 error_setg(errp, "Block replication is not running"); 660 aio_context_release(aio_context); 661 return; 662 } 663 664 switch (s->mode) { 665 case REPLICATION_MODE_PRIMARY: 666 s->stage = BLOCK_REPLICATION_DONE; 667 s->error = 0; 668 break; 669 case REPLICATION_MODE_SECONDARY: 670 /* 671 * This BDS will be closed, and the job should be completed 672 * before the BDS is closed, because we will access hidden 673 * disk, secondary disk in backup_job_completed(). 674 */ 675 if (s->secondary_disk->bs->job) { 676 block_job_cancel_sync(s->secondary_disk->bs->job); 677 } 678 679 if (!failover) { 680 secondary_do_checkpoint(s, errp); 681 s->stage = BLOCK_REPLICATION_DONE; 682 aio_context_release(aio_context); 683 return; 684 } 685 686 s->stage = BLOCK_REPLICATION_FAILOVER; 687 commit_active_start(NULL, s->active_disk->bs, s->secondary_disk->bs, 688 BLOCK_JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT, 689 NULL, replication_done, bs, true, errp); 690 break; 691 default: 692 aio_context_release(aio_context); 693 abort(); 694 } 695 aio_context_release(aio_context); 696 } 697 698 BlockDriver bdrv_replication = { 699 .format_name = "replication", 700 .protocol_name = "replication", 701 .instance_size = sizeof(BDRVReplicationState), 702 703 .bdrv_open = replication_open, 704 .bdrv_close = replication_close, 705 .bdrv_child_perm = replication_child_perm, 706 707 .bdrv_getlength = replication_getlength, 708 .bdrv_co_readv = replication_co_readv, 709 .bdrv_co_writev = replication_co_writev, 710 711 .is_filter = true, 712 .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter, 713 714 .has_variable_length = true, 715 }; 716 717 static void bdrv_replication_init(void) 718 { 719 bdrv_register(&bdrv_replication); 720 } 721 722 block_init(bdrv_replication_init); 723