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/module.h"
  17 #include "qemu/option.h"
  18 #include "block/nbd.h"
  19 #include "block/blockjob.h"
  20 #include "block/block_int.h"
  21 #include "block/block_backup.h"
  22 #include "sysemu/block-backend.h"
  23 #include "qapi/error.h"
  24 #include "qapi/qmp/qdict.h"
  25 #include "replication.h"
  26
  27 typedef enum {
  28     BLOCK_REPLICATION_NONE,             /* block replication is not started */
  29     BLOCK_REPLICATION_RUNNING,          /* block replication is running */
  30     BLOCK_REPLICATION_FAILOVER,         /* failover is running in background */
  31     BLOCK_REPLICATION_FAILOVER_FAILED,  /* failover failed */
  32     BLOCK_REPLICATION_DONE,             /* block replication is done */
  33 } ReplicationStage;
  34
  35 typedef struct BDRVReplicationState {
  36     ReplicationMode mode;
  37     ReplicationStage stage;
  38     BdrvChild *active_disk;
  39     BlockJob *commit_job;
  40     BdrvChild *hidden_disk;
  41     BdrvChild *secondary_disk;
  42     BlockJob *backup_job;
  43     char *top_id;
  44     ReplicationState *rs;
  45     Error *blocker;
  46     bool orig_hidden_read_only;
  47     bool orig_secondary_read_only;
  48     int error;
  49 } BDRVReplicationState;
  50
  51 static void replication_start(ReplicationState *rs, ReplicationMode mode,
  52                               Error **errp);
  53 static void replication_do_checkpoint(ReplicationState *rs, Error **errp);
  54 static void replication_get_error(ReplicationState *rs, Error **errp);
  55 static void replication_stop(ReplicationState *rs, bool failover,
  56                              Error **errp);
  57
  58 #define REPLICATION_MODE        "mode"
  59 #define REPLICATION_TOP_ID      "top-id"
  60 static QemuOptsList replication_runtime_opts = {
  61     .name = "replication",
  62     .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head),
  63     .desc = {
  64         {
  65             .name = REPLICATION_MODE,
  66             .type = QEMU_OPT_STRING,
  67         },
  68         {
  69             .name = REPLICATION_TOP_ID,
  70             .type = QEMU_OPT_STRING,
  71         },
  72         { /* end of list */ }
  73     },
  74 };
  75
  76 static ReplicationOps replication_ops = {
  77     .start = replication_start,
  78     .checkpoint = replication_do_checkpoint,
  79     .get_error = replication_get_error,
  80     .stop = replication_stop,
  81 };
  82
  83 static int replication_open(BlockDriverState *bs, QDict *options,
  84                             int flags, Error **errp)
  85 {
  86     int ret;
  87     BDRVReplicationState *s = bs->opaque;
  88     Error *local_err = NULL;
  89     QemuOpts *opts = NULL;
  90     const char *mode;
  91     const char *top_id;
  92
  93     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
  94                                BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
  95                                false, errp);
  96     if (!bs->file) {
  97         return -EINVAL;
  98     }
  99
 100     ret = -EINVAL;
 101     opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
 102     if (!qemu_opts_absorb_qdict(opts, options, &local_err)) {
 103         goto fail;
 104     }
 105
 106     mode = qemu_opt_get(opts, REPLICATION_MODE);
 107     if (!mode) {
 108         error_setg(errp, "Missing the option mode");
 109         goto fail;
 110     }
 111
 112     if (!strcmp(mode, "primary")) {
 113         s->mode = REPLICATION_MODE_PRIMARY;
 114         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 115         if (top_id) {
 116             error_setg(errp,
 117                        "The primary side does not support option top-id");
 118             goto fail;
 119         }
 120     } else if (!strcmp(mode, "secondary")) {
 121         s->mode = REPLICATION_MODE_SECONDARY;
 122         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 123         s->top_id = g_strdup(top_id);
 124         if (!s->top_id) {
 125             error_setg(errp, "Missing the option top-id");
 126             goto fail;
 127         }
 128     } else {
 129         error_setg(errp,
 130                    "The option mode's value should be primary or secondary");
 131         goto fail;
 132     }
 133
 134     s->rs = replication_new(bs, &replication_ops);
 135
 136     ret = 0;
 137
 138 fail:
 139     qemu_opts_del(opts);
 140     return ret;
 141 }
 142
 143 static void replication_close(BlockDriverState *bs)
 144 {
 145     BDRVReplicationState *s = bs->opaque;
 146     Job *commit_job;
 147
 148     if (s->stage == BLOCK_REPLICATION_RUNNING) {
 149         replication_stop(s->rs, false, NULL);
 150     }
 151     if (s->stage == BLOCK_REPLICATION_FAILOVER) {
 152         commit_job = &s->commit_job->job;
 153         assert(commit_job->aio_context == qemu_get_current_aio_context());
 154         job_cancel_sync(commit_job);
 155     }
 156
 157     if (s->mode == REPLICATION_MODE_SECONDARY) {
 158         g_free(s->top_id);
 159     }
 160
 161     replication_remove(s->rs);
 162 }
 163
 164 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
 165                                    BdrvChildRole role,
 166                                    BlockReopenQueue *reopen_queue,
 167                                    uint64_t perm, uint64_t shared,
 168                                    uint64_t *nperm, uint64_t *nshared)
 169 {
 170     *nperm = BLK_PERM_CONSISTENT_READ;
 171     if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) {
 172         *nperm |= BLK_PERM_WRITE;
 173     }
 174     *nshared = BLK_PERM_CONSISTENT_READ
 175                | BLK_PERM_WRITE
 176                | BLK_PERM_WRITE_UNCHANGED;
 177     return;
 178 }
 179
 180 static int64_t replication_getlength(BlockDriverState *bs)
 181 {
 182     return bdrv_getlength(bs->file->bs);
 183 }
 184
 185 static int replication_get_io_status(BDRVReplicationState *s)
 186 {
 187     switch (s->stage) {
 188     case BLOCK_REPLICATION_NONE:
 189         return -EIO;
 190     case BLOCK_REPLICATION_RUNNING:
 191         return 0;
 192     case BLOCK_REPLICATION_FAILOVER:
 193         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 194     case BLOCK_REPLICATION_FAILOVER_FAILED:
 195         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
 196     case BLOCK_REPLICATION_DONE:
 197         /*
 198          * active commit job completes, and active disk and secondary_disk
 199          * is swapped, so we can operate bs->file directly
 200          */
 201         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 202     default:
 203         abort();
 204     }
 205 }
 206
 207 static int replication_return_value(BDRVReplicationState *s, int ret)
 208 {
 209     if (s->mode == REPLICATION_MODE_SECONDARY) {
 210         return ret;
 211     }
 212
 213     if (ret < 0) {
 214         s->error = ret;
 215         ret = 0;
 216     }
 217
 218     return ret;
 219 }
 220
 221 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
 222                                              int64_t sector_num,
 223                                              int remaining_sectors,
 224                                              QEMUIOVector *qiov)
 225 {
 226     BDRVReplicationState *s = bs->opaque;
 227     int ret;
 228
 229     if (s->mode == REPLICATION_MODE_PRIMARY) {
 230         /* We only use it to forward primary write requests */
 231         return -EIO;
 232     }
 233
 234     ret = replication_get_io_status(s);
 235     if (ret < 0) {
 236         return ret;
 237     }
 238
 239     ret = bdrv_co_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
 240                          remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 241
 242     return replication_return_value(s, ret);
 243 }
 244
 245 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
 246                                               int64_t sector_num,
 247                                               int remaining_sectors,
 248                                               QEMUIOVector *qiov,
 249                                               int flags)
 250 {
 251     BDRVReplicationState *s = bs->opaque;
 252     QEMUIOVector hd_qiov;
 253     uint64_t bytes_done = 0;
 254     BdrvChild *top = bs->file;
 255     BdrvChild *base = s->secondary_disk;
 256     BdrvChild *target;
 257     int ret;
 258     int64_t n;
 259
 260     assert(!flags);
 261     ret = replication_get_io_status(s);
 262     if (ret < 0) {
 263         goto out;
 264     }
 265
 266     if (ret == 0) {
 267         ret = bdrv_co_pwritev(top, sector_num * BDRV_SECTOR_SIZE,
 268                               remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 269         return replication_return_value(s, ret);
 270     }
 271
 272     /*
 273      * Failover failed, only write to active disk if the sectors
 274      * have already been allocated in active disk/hidden disk.
 275      */
 276     qemu_iovec_init(&hd_qiov, qiov->niov);
 277     while (remaining_sectors > 0) {
 278         int64_t count;
 279
 280         ret = bdrv_is_allocated_above(top->bs, base->bs, false,
 281                                       sector_num * BDRV_SECTOR_SIZE,
 282                                       remaining_sectors * BDRV_SECTOR_SIZE,
 283                                       &count);
 284         if (ret < 0) {
 285             goto out1;
 286         }
 287
 288         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
 289         n = count >> BDRV_SECTOR_BITS;
 290         qemu_iovec_reset(&hd_qiov);
 291         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
 292
 293         target = ret ? top : base;
 294         ret = bdrv_co_pwritev(target, sector_num * BDRV_SECTOR_SIZE,
 295                               n * BDRV_SECTOR_SIZE, &hd_qiov, 0);
 296         if (ret < 0) {
 297             goto out1;
 298         }
 299
 300         remaining_sectors -= n;
 301         sector_num += n;
 302         bytes_done += count;
 303     }
 304
 305 out1:
 306     qemu_iovec_destroy(&hd_qiov);
 307 out:
 308     return ret;
 309 }
 310
 311 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
 312 {
 313     Error *local_err = NULL;
 314     int ret;
 315
 316     if (!s->backup_job) {
 317         error_setg(errp, "Backup job was cancelled unexpectedly");
 318         return;
 319     }
 320
 321     backup_do_checkpoint(s->backup_job, &local_err);
 322     if (local_err) {
 323         error_propagate(errp, local_err);
 324         return;
 325     }
 326
 327     if (!s->active_disk->bs->drv) {
 328         error_setg(errp, "Active disk %s is ejected",
 329                    s->active_disk->bs->node_name);
 330         return;
 331     }
 332
 333     ret = bdrv_make_empty(s->active_disk, errp);
 334     if (ret < 0) {
 335         return;
 336     }
 337
 338     if (!s->hidden_disk->bs->drv) {
 339         error_setg(errp, "Hidden disk %s is ejected",
 340                    s->hidden_disk->bs->node_name);
 341         return;
 342     }
 343
 344     BlockBackend *blk = blk_new(qemu_get_current_aio_context(),
 345                                 BLK_PERM_WRITE, BLK_PERM_ALL);
 346     blk_insert_bs(blk, s->hidden_disk->bs, &local_err);
 347     if (local_err) {
 348         error_propagate(errp, local_err);
 349         blk_unref(blk);
 350         return;
 351     }
 352
 353     ret = blk_make_empty(blk, errp);
 354     blk_unref(blk);
 355     if (ret < 0) {
 356         return;
 357     }
 358 }
 359
 360 /* This function is supposed to be called twice:
 361  * first with writable = true, then with writable = false.
 362  * The first call puts s->hidden_disk and s->secondary_disk in
 363  * r/w mode, and the second puts them back in their original state.
 364  */
 365 static void reopen_backing_file(BlockDriverState *bs, bool writable,
 366                                 Error **errp)
 367 {
 368     BDRVReplicationState *s = bs->opaque;
 369     BlockReopenQueue *reopen_queue = NULL;
 370     Error *local_err = NULL;
 371
 372     if (writable) {
 373         s->orig_hidden_read_only = bdrv_is_read_only(s->hidden_disk->bs);
 374         s->orig_secondary_read_only = bdrv_is_read_only(s->secondary_disk->bs);
 375     }
 376
 377     bdrv_subtree_drained_begin(s->hidden_disk->bs);
 378     bdrv_subtree_drained_begin(s->secondary_disk->bs);
 379
 380     if (s->orig_hidden_read_only) {
 381         QDict *opts = qdict_new();
 382         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 383         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs,
 384                                          opts, true);
 385     }
 386
 387     if (s->orig_secondary_read_only) {
 388         QDict *opts = qdict_new();
 389         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 390         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
 391                                          opts, true);
 392     }
 393
 394     if (reopen_queue) {
 395         bdrv_reopen_multiple(reopen_queue, &local_err);
 396         error_propagate(errp, local_err);
 397     }
 398
 399     bdrv_subtree_drained_end(s->hidden_disk->bs);
 400     bdrv_subtree_drained_end(s->secondary_disk->bs);
 401 }
 402
 403 static void backup_job_cleanup(BlockDriverState *bs)
 404 {
 405     BDRVReplicationState *s = bs->opaque;
 406     BlockDriverState *top_bs;
 407
 408     s->backup_job = NULL;
 409
 410     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 411     if (!top_bs) {
 412         return;
 413     }
 414     bdrv_op_unblock_all(top_bs, s->blocker);
 415     error_free(s->blocker);
 416     reopen_backing_file(bs, false, NULL);
 417 }
 418
 419 static void backup_job_completed(void *opaque, int ret)
 420 {
 421     BlockDriverState *bs = opaque;
 422     BDRVReplicationState *s = bs->opaque;
 423
 424     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 425         /* The backup job is cancelled unexpectedly */
 426         s->error = -EIO;
 427     }
 428
 429     backup_job_cleanup(bs);
 430 }
 431
 432 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 433 {
 434     BdrvChild *child;
 435
 436     /* The bs itself is the top_bs */
 437     if (top_bs == bs) {
 438         return true;
 439     }
 440
 441     /* Iterate over top_bs's children */
 442     QLIST_FOREACH(child, &top_bs->children, next) {
 443         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 444             return true;
 445         }
 446     }
 447
 448     return false;
 449 }
 450
 451 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 452                               Error **errp)
 453 {
 454     BlockDriverState *bs = rs->opaque;
 455     BDRVReplicationState *s;
 456     BlockDriverState *top_bs;
 457     int64_t active_length, hidden_length, disk_length;
 458     AioContext *aio_context;
 459     Error *local_err = NULL;
 460
 461     aio_context = bdrv_get_aio_context(bs);
 462     aio_context_acquire(aio_context);
 463     s = bs->opaque;
 464
 465     if (s->stage == BLOCK_REPLICATION_DONE ||
 466         s->stage == BLOCK_REPLICATION_FAILOVER) {
 467         /*
 468          * This case happens when a secondary is promoted to primary.
 469          * Ignore the request because the secondary side of replication
 470          * doesn't have to do anything anymore.
 471          */
 472         aio_context_release(aio_context);
 473         return;
 474     }
 475
 476     if (s->stage != BLOCK_REPLICATION_NONE) {
 477         error_setg(errp, "Block replication is running or done");
 478         aio_context_release(aio_context);
 479         return;
 480     }
 481
 482     if (s->mode != mode) {
 483         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 484                    " but got %d", s->mode, mode);
 485         aio_context_release(aio_context);
 486         return;
 487     }
 488
 489     switch (s->mode) {
 490     case REPLICATION_MODE_PRIMARY:
 491         break;
 492     case REPLICATION_MODE_SECONDARY:
 493         s->active_disk = bs->file;
 494         if (!s->active_disk || !s->active_disk->bs ||
 495                                     !s->active_disk->bs->backing) {
 496             error_setg(errp, "Active disk doesn't have backing file");
 497             aio_context_release(aio_context);
 498             return;
 499         }
 500
 501         s->hidden_disk = s->active_disk->bs->backing;
 502         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
 503             error_setg(errp, "Hidden disk doesn't have backing file");
 504             aio_context_release(aio_context);
 505             return;
 506         }
 507
 508         s->secondary_disk = s->hidden_disk->bs->backing;
 509         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
 510             error_setg(errp, "The secondary disk doesn't have block backend");
 511             aio_context_release(aio_context);
 512             return;
 513         }
 514
 515         /* verify the length */
 516         active_length = bdrv_getlength(s->active_disk->bs);
 517         hidden_length = bdrv_getlength(s->hidden_disk->bs);
 518         disk_length = bdrv_getlength(s->secondary_disk->bs);
 519         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 520             active_length != hidden_length || hidden_length != disk_length) {
 521             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 522                        " are not the same");
 523             aio_context_release(aio_context);
 524             return;
 525         }
 526
 527         /* Must be true, or the bdrv_getlength() calls would have failed */
 528         assert(s->active_disk->bs->drv && s->hidden_disk->bs->drv);
 529
 530         if (!s->active_disk->bs->drv->bdrv_make_empty ||
 531             !s->hidden_disk->bs->drv->bdrv_make_empty) {
 532             error_setg(errp,
 533                        "Active disk or hidden disk doesn't support make_empty");
 534             aio_context_release(aio_context);
 535             return;
 536         }
 537
 538         /* reopen the backing file in r/w mode */
 539         reopen_backing_file(bs, true, &local_err);
 540         if (local_err) {
 541             error_propagate(errp, local_err);
 542             aio_context_release(aio_context);
 543             return;
 544         }
 545
 546         /* start backup job now */
 547         error_setg(&s->blocker,
 548                    "Block device is in use by internal backup job");
 549
 550         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 551         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 552             !check_top_bs(top_bs, bs)) {
 553             error_setg(errp, "No top_bs or it is invalid");
 554             reopen_backing_file(bs, false, NULL);
 555             aio_context_release(aio_context);
 556             return;
 557         }
 558         bdrv_op_block_all(top_bs, s->blocker);
 559         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 560
 561         s->backup_job = backup_job_create(
 562                                 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 563                                 0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
 564                                 BLOCKDEV_ON_ERROR_REPORT,
 565                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
 566                                 backup_job_completed, bs, NULL, &local_err);
 567         if (local_err) {
 568             error_propagate(errp, local_err);
 569             backup_job_cleanup(bs);
 570             aio_context_release(aio_context);
 571             return;
 572         }
 573         job_start(&s->backup_job->job);
 574         break;
 575     default:
 576         aio_context_release(aio_context);
 577         abort();
 578     }
 579
 580     s->stage = BLOCK_REPLICATION_RUNNING;
 581
 582     if (s->mode == REPLICATION_MODE_SECONDARY) {
 583         secondary_do_checkpoint(s, errp);
 584     }
 585
 586     s->error = 0;
 587     aio_context_release(aio_context);
 588 }
 589
 590 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
 591 {
 592     BlockDriverState *bs = rs->opaque;
 593     BDRVReplicationState *s;
 594     AioContext *aio_context;
 595
 596     aio_context = bdrv_get_aio_context(bs);
 597     aio_context_acquire(aio_context);
 598     s = bs->opaque;
 599
 600     if (s->stage == BLOCK_REPLICATION_DONE ||
 601         s->stage == BLOCK_REPLICATION_FAILOVER) {
 602         /*
 603          * This case happens when a secondary was promoted to primary.
 604          * Ignore the request because the secondary side of replication
 605          * doesn't have to do anything anymore.
 606          */
 607         aio_context_release(aio_context);
 608         return;
 609     }
 610
 611     if (s->mode == REPLICATION_MODE_SECONDARY) {
 612         secondary_do_checkpoint(s, errp);
 613     }
 614     aio_context_release(aio_context);
 615 }
 616
 617 static void replication_get_error(ReplicationState *rs, Error **errp)
 618 {
 619     BlockDriverState *bs = rs->opaque;
 620     BDRVReplicationState *s;
 621     AioContext *aio_context;
 622
 623     aio_context = bdrv_get_aio_context(bs);
 624     aio_context_acquire(aio_context);
 625     s = bs->opaque;
 626
 627     if (s->stage == BLOCK_REPLICATION_NONE) {
 628         error_setg(errp, "Block replication is not running");
 629         aio_context_release(aio_context);
 630         return;
 631     }
 632
 633     if (s->error) {
 634         error_setg(errp, "I/O error occurred");
 635         aio_context_release(aio_context);
 636         return;
 637     }
 638     aio_context_release(aio_context);
 639 }
 640
 641 static void replication_done(void *opaque, int ret)
 642 {
 643     BlockDriverState *bs = opaque;
 644     BDRVReplicationState *s = bs->opaque;
 645
 646     if (ret == 0) {
 647         s->stage = BLOCK_REPLICATION_DONE;
 648
 649         s->active_disk = NULL;
 650         s->secondary_disk = NULL;
 651         s->hidden_disk = NULL;
 652         s->error = 0;
 653     } else {
 654         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 655         s->error = -EIO;
 656     }
 657 }
 658
 659 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 660 {
 661     BlockDriverState *bs = rs->opaque;
 662     BDRVReplicationState *s;
 663     AioContext *aio_context;
 664
 665     aio_context = bdrv_get_aio_context(bs);
 666     aio_context_acquire(aio_context);
 667     s = bs->opaque;
 668
 669     if (s->stage == BLOCK_REPLICATION_DONE ||
 670         s->stage == BLOCK_REPLICATION_FAILOVER) {
 671         /*
 672          * This case happens when a secondary was promoted to primary.
 673          * Ignore the request because the secondary side of replication
 674          * doesn't have to do anything anymore.
 675          */
 676         aio_context_release(aio_context);
 677         return;
 678     }
 679
 680     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 681         error_setg(errp, "Block replication is not running");
 682         aio_context_release(aio_context);
 683         return;
 684     }
 685
 686     switch (s->mode) {
 687     case REPLICATION_MODE_PRIMARY:
 688         s->stage = BLOCK_REPLICATION_DONE;
 689         s->error = 0;
 690         break;
 691     case REPLICATION_MODE_SECONDARY:
 692         /*
 693          * This BDS will be closed, and the job should be completed
 694          * before the BDS is closed, because we will access hidden
 695          * disk, secondary disk in backup_job_completed().
 696          */
 697         if (s->backup_job) {
 698             job_cancel_sync(&s->backup_job->job);
 699         }
 700
 701         if (!failover) {
 702             secondary_do_checkpoint(s, errp);
 703             s->stage = BLOCK_REPLICATION_DONE;
 704             aio_context_release(aio_context);
 705             return;
 706         }
 707
 708         s->stage = BLOCK_REPLICATION_FAILOVER;
 709         s->commit_job = commit_active_start(
 710                             NULL, s->active_disk->bs, s->secondary_disk->bs,
 711                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 712                             NULL, replication_done, bs, true, errp);
 713         break;
 714     default:
 715         aio_context_release(aio_context);
 716         abort();
 717     }
 718     aio_context_release(aio_context);
 719 }
 720
 721 static const char *const replication_strong_runtime_opts[] = {
 722     REPLICATION_MODE,
 723     REPLICATION_TOP_ID,
 724
 725     NULL
 726 };
 727
 728 static BlockDriver bdrv_replication = {
 729     .format_name                = "replication",
 730     .instance_size              = sizeof(BDRVReplicationState),
 731
 732     .bdrv_open                  = replication_open,
 733     .bdrv_close                 = replication_close,
 734     .bdrv_child_perm            = replication_child_perm,
 735
 736     .bdrv_getlength             = replication_getlength,
 737     .bdrv_co_readv              = replication_co_readv,
 738     .bdrv_co_writev             = replication_co_writev,
 739
 740     .is_filter                  = true,
 741
 742     .has_variable_length        = true,
 743     .strong_runtime_opts        = replication_strong_runtime_opts,
 744 };
 745
 746 static void bdrv_replication_init(void)
 747 {
 748     bdrv_register(&bdrv_replication);
 749 }
 750
 751 block_init(bdrv_replication_init);