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