block/replication.c

   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/option.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     bdrv_subtree_drained_begin(s->hidden_disk->bs);
 398     bdrv_subtree_drained_begin(s->secondary_disk->bs);
 399
 400     if (orig_hidden_flags != new_hidden_flags) {
 401         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs, NULL,
 402                                          new_hidden_flags);
 403     }
 404
 405     if (!(orig_secondary_flags & BDRV_O_RDWR)) {
 406         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
 407                                          NULL, new_secondary_flags);
 408     }
 409
 410     if (reopen_queue) {
 411         bdrv_reopen_multiple(bdrv_get_aio_context(bs),
 412                              reopen_queue, &local_err);
 413         error_propagate(errp, local_err);
 414     }
 415
 416     bdrv_subtree_drained_end(s->hidden_disk->bs);
 417     bdrv_subtree_drained_end(s->secondary_disk->bs);
 418 }
 419
 420 static void backup_job_cleanup(BlockDriverState *bs)
 421 {
 422     BDRVReplicationState *s = bs->opaque;
 423     BlockDriverState *top_bs;
 424
 425     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 426     if (!top_bs) {
 427         return;
 428     }
 429     bdrv_op_unblock_all(top_bs, s->blocker);
 430     error_free(s->blocker);
 431     reopen_backing_file(bs, false, NULL);
 432 }
 433
 434 static void backup_job_completed(void *opaque, int ret)
 435 {
 436     BlockDriverState *bs = opaque;
 437     BDRVReplicationState *s = bs->opaque;
 438
 439     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 440         /* The backup job is cancelled unexpectedly */
 441         s->error = -EIO;
 442     }
 443
 444     backup_job_cleanup(bs);
 445 }
 446
 447 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 448 {
 449     BdrvChild *child;
 450
 451     /* The bs itself is the top_bs */
 452     if (top_bs == bs) {
 453         return true;
 454     }
 455
 456     /* Iterate over top_bs's children */
 457     QLIST_FOREACH(child, &top_bs->children, next) {
 458         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 459             return true;
 460         }
 461     }
 462
 463     return false;
 464 }
 465
 466 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 467                               Error **errp)
 468 {
 469     BlockDriverState *bs = rs->opaque;
 470     BDRVReplicationState *s;
 471     BlockDriverState *top_bs;
 472     int64_t active_length, hidden_length, disk_length;
 473     AioContext *aio_context;
 474     Error *local_err = NULL;
 475     BlockJob *job;
 476
 477     aio_context = bdrv_get_aio_context(bs);
 478     aio_context_acquire(aio_context);
 479     s = bs->opaque;
 480
 481     if (s->stage != BLOCK_REPLICATION_NONE) {
 482         error_setg(errp, "Block replication is running or done");
 483         aio_context_release(aio_context);
 484         return;
 485     }
 486
 487     if (s->mode != mode) {
 488         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 489                    " but got %d", s->mode, mode);
 490         aio_context_release(aio_context);
 491         return;
 492     }
 493
 494     switch (s->mode) {
 495     case REPLICATION_MODE_PRIMARY:
 496         break;
 497     case REPLICATION_MODE_SECONDARY:
 498         s->active_disk = bs->file;
 499         if (!s->active_disk || !s->active_disk->bs ||
 500                                     !s->active_disk->bs->backing) {
 501             error_setg(errp, "Active disk doesn't have backing file");
 502             aio_context_release(aio_context);
 503             return;
 504         }
 505
 506         s->hidden_disk = s->active_disk->bs->backing;
 507         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
 508             error_setg(errp, "Hidden disk doesn't have backing file");
 509             aio_context_release(aio_context);
 510             return;
 511         }
 512
 513         s->secondary_disk = s->hidden_disk->bs->backing;
 514         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
 515             error_setg(errp, "The secondary disk doesn't have block backend");
 516             aio_context_release(aio_context);
 517             return;
 518         }
 519
 520         /* verify the length */
 521         active_length = bdrv_getlength(s->active_disk->bs);
 522         hidden_length = bdrv_getlength(s->hidden_disk->bs);
 523         disk_length = bdrv_getlength(s->secondary_disk->bs);
 524         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 525             active_length != hidden_length || hidden_length != disk_length) {
 526             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 527                        " are not the same");
 528             aio_context_release(aio_context);
 529             return;
 530         }
 531
 532         /* Must be true, or the bdrv_getlength() calls would have failed */
 533         assert(s->active_disk->bs->drv && s->hidden_disk->bs->drv);
 534
 535         if (!s->active_disk->bs->drv->bdrv_make_empty ||
 536             !s->hidden_disk->bs->drv->bdrv_make_empty) {
 537             error_setg(errp,
 538                        "Active disk or hidden disk doesn't support make_empty");
 539             aio_context_release(aio_context);
 540             return;
 541         }
 542
 543         /* reopen the backing file in r/w mode */
 544         reopen_backing_file(bs, true, &local_err);
 545         if (local_err) {
 546             error_propagate(errp, local_err);
 547             aio_context_release(aio_context);
 548             return;
 549         }
 550
 551         /* start backup job now */
 552         error_setg(&s->blocker,
 553                    "Block device is in use by internal backup job");
 554
 555         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 556         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 557             !check_top_bs(top_bs, bs)) {
 558             error_setg(errp, "No top_bs or it is invalid");
 559             reopen_backing_file(bs, false, NULL);
 560             aio_context_release(aio_context);
 561             return;
 562         }
 563         bdrv_op_block_all(top_bs, s->blocker);
 564         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 565
 566         job = backup_job_create(NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 567                                 0, MIRROR_SYNC_MODE_NONE, NULL, false,
 568                                 BLOCKDEV_ON_ERROR_REPORT,
 569                                 BLOCKDEV_ON_ERROR_REPORT, BLOCK_JOB_INTERNAL,
 570                                 backup_job_completed, bs, NULL, &local_err);
 571         if (local_err) {
 572             error_propagate(errp, local_err);
 573             backup_job_cleanup(bs);
 574             aio_context_release(aio_context);
 575             return;
 576         }
 577         block_job_start(job);
 578         break;
 579     default:
 580         aio_context_release(aio_context);
 581         abort();
 582     }
 583
 584     s->stage = BLOCK_REPLICATION_RUNNING;
 585
 586     if (s->mode == REPLICATION_MODE_SECONDARY) {
 587         secondary_do_checkpoint(s, errp);
 588     }
 589
 590     s->error = 0;
 591     aio_context_release(aio_context);
 592 }
 593
 594 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
 595 {
 596     BlockDriverState *bs = rs->opaque;
 597     BDRVReplicationState *s;
 598     AioContext *aio_context;
 599
 600     aio_context = bdrv_get_aio_context(bs);
 601     aio_context_acquire(aio_context);
 602     s = bs->opaque;
 603
 604     if (s->mode == REPLICATION_MODE_SECONDARY) {
 605         secondary_do_checkpoint(s, errp);
 606     }
 607     aio_context_release(aio_context);
 608 }
 609
 610 static void replication_get_error(ReplicationState *rs, Error **errp)
 611 {
 612     BlockDriverState *bs = rs->opaque;
 613     BDRVReplicationState *s;
 614     AioContext *aio_context;
 615
 616     aio_context = bdrv_get_aio_context(bs);
 617     aio_context_acquire(aio_context);
 618     s = bs->opaque;
 619
 620     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 621         error_setg(errp, "Block replication is not running");
 622         aio_context_release(aio_context);
 623         return;
 624     }
 625
 626     if (s->error) {
 627         error_setg(errp, "I/O error occurred");
 628         aio_context_release(aio_context);
 629         return;
 630     }
 631     aio_context_release(aio_context);
 632 }
 633
 634 static void replication_done(void *opaque, int ret)
 635 {
 636     BlockDriverState *bs = opaque;
 637     BDRVReplicationState *s = bs->opaque;
 638
 639     if (ret == 0) {
 640         s->stage = BLOCK_REPLICATION_DONE;
 641
 642         /* refresh top bs's filename */
 643         bdrv_refresh_filename(bs);
 644         s->active_disk = NULL;
 645         s->secondary_disk = NULL;
 646         s->hidden_disk = NULL;
 647         s->error = 0;
 648     } else {
 649         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 650         s->error = -EIO;
 651     }
 652 }
 653
 654 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 655 {
 656     BlockDriverState *bs = rs->opaque;
 657     BDRVReplicationState *s;
 658     AioContext *aio_context;
 659
 660     aio_context = bdrv_get_aio_context(bs);
 661     aio_context_acquire(aio_context);
 662     s = bs->opaque;
 663
 664     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 665         error_setg(errp, "Block replication is not running");
 666         aio_context_release(aio_context);
 667         return;
 668     }
 669
 670     switch (s->mode) {
 671     case REPLICATION_MODE_PRIMARY:
 672         s->stage = BLOCK_REPLICATION_DONE;
 673         s->error = 0;
 674         break;
 675     case REPLICATION_MODE_SECONDARY:
 676         /*
 677          * This BDS will be closed, and the job should be completed
 678          * before the BDS is closed, because we will access hidden
 679          * disk, secondary disk in backup_job_completed().
 680          */
 681         if (s->secondary_disk->bs->job) {
 682             block_job_cancel_sync(s->secondary_disk->bs->job);
 683         }
 684
 685         if (!failover) {
 686             secondary_do_checkpoint(s, errp);
 687             s->stage = BLOCK_REPLICATION_DONE;
 688             aio_context_release(aio_context);
 689             return;
 690         }
 691
 692         s->stage = BLOCK_REPLICATION_FAILOVER;
 693         commit_active_start(NULL, s->active_disk->bs, s->secondary_disk->bs,
 694                             BLOCK_JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 695                             NULL, replication_done, bs, true, errp);
 696         break;
 697     default:
 698         aio_context_release(aio_context);
 699         abort();
 700     }
 701     aio_context_release(aio_context);
 702 }
 703
 704 BlockDriver bdrv_replication = {
 705     .format_name                = "replication",
 706     .protocol_name              = "replication",
 707     .instance_size              = sizeof(BDRVReplicationState),
 708
 709     .bdrv_open                  = replication_open,
 710     .bdrv_close                 = replication_close,
 711     .bdrv_child_perm            = replication_child_perm,
 712
 713     .bdrv_getlength             = replication_getlength,
 714     .bdrv_co_readv              = replication_co_readv,
 715     .bdrv_co_writev             = replication_co_writev,
 716
 717     .is_filter                  = true,
 718     .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter,
 719
 720     .has_variable_length        = true,
 721 };
 722
 723 static void bdrv_replication_init(void)
 724 {
 725     bdrv_register(&bdrv_replication);
 726 }
 727
 728 block_init(bdrv_replication_init);