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 "block/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     BlockJob *commit_job;
  39     BdrvChild *hidden_disk;
  40     BdrvChild *secondary_disk;
  41     BlockJob *backup_job;
  42     char *top_id;
  43     ReplicationState *rs;
  44     Error *blocker;
  45     bool orig_hidden_read_only;
  46     bool orig_secondary_read_only;
  47     int error;
  48 } BDRVReplicationState;
  49
  50 static void replication_start(ReplicationState *rs, ReplicationMode mode,
  51                               Error **errp);
  52 static void replication_do_checkpoint(ReplicationState *rs, Error **errp);
  53 static void replication_get_error(ReplicationState *rs, Error **errp);
  54 static void replication_stop(ReplicationState *rs, bool failover,
  55                              Error **errp);
  56
  57 #define REPLICATION_MODE        "mode"
  58 #define REPLICATION_TOP_ID      "top-id"
  59 static QemuOptsList replication_runtime_opts = {
  60     .name = "replication",
  61     .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head),
  62     .desc = {
  63         {
  64             .name = REPLICATION_MODE,
  65             .type = QEMU_OPT_STRING,
  66         },
  67         {
  68             .name = REPLICATION_TOP_ID,
  69             .type = QEMU_OPT_STRING,
  70         },
  71         { /* end of list */ }
  72     },
  73 };
  74
  75 static ReplicationOps replication_ops = {
  76     .start = replication_start,
  77     .checkpoint = replication_do_checkpoint,
  78     .get_error = replication_get_error,
  79     .stop = replication_stop,
  80 };
  81
  82 static int replication_open(BlockDriverState *bs, QDict *options,
  83                             int flags, Error **errp)
  84 {
  85     int ret;
  86     BDRVReplicationState *s = bs->opaque;
  87     QemuOpts *opts = NULL;
  88     const char *mode;
  89     const char *top_id;
  90
  91     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
  92     if (ret < 0) {
  93         return ret;
  94     }
  95
  96     ret = -EINVAL;
  97     opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
  98     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
  99         goto fail;
 100     }
 101
 102     mode = qemu_opt_get(opts, REPLICATION_MODE);
 103     if (!mode) {
 104         error_setg(errp, "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(errp,
 113                        "The primary side does not support option top-id");
 114             goto fail;
 115         }
 116     } else if (!strcmp(mode, "secondary")) {
 117         s->mode = REPLICATION_MODE_SECONDARY;
 118         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 119         s->top_id = g_strdup(top_id);
 120         if (!s->top_id) {
 121             error_setg(errp, "Missing the option top-id");
 122             goto fail;
 123         }
 124     } else {
 125         error_setg(errp,
 126                    "The option mode's value should be primary or secondary");
 127         goto fail;
 128     }
 129
 130     s->rs = replication_new(bs, &replication_ops);
 131
 132     ret = 0;
 133
 134 fail:
 135     qemu_opts_del(opts);
 136     return ret;
 137 }
 138
 139 static void replication_close(BlockDriverState *bs)
 140 {
 141     BDRVReplicationState *s = bs->opaque;
 142     Job *commit_job;
 143     GLOBAL_STATE_CODE();
 144
 145     if (s->stage == BLOCK_REPLICATION_RUNNING) {
 146         replication_stop(s->rs, false, NULL);
 147     }
 148     if (s->stage == BLOCK_REPLICATION_FAILOVER) {
 149         commit_job = &s->commit_job->job;
 150         assert(commit_job->aio_context == qemu_get_current_aio_context());
 151         job_cancel_sync(commit_job, false);
 152     }
 153
 154     if (s->mode == REPLICATION_MODE_SECONDARY) {
 155         g_free(s->top_id);
 156     }
 157
 158     replication_remove(s->rs);
 159 }
 160
 161 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
 162                                    BdrvChildRole role,
 163                                    BlockReopenQueue *reopen_queue,
 164                                    uint64_t perm, uint64_t shared,
 165                                    uint64_t *nperm, uint64_t *nshared)
 166 {
 167     if (role & BDRV_CHILD_PRIMARY) {
 168         *nperm = BLK_PERM_CONSISTENT_READ;
 169     } else {
 170         *nperm = 0;
 171     }
 172
 173     if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) {
 174         *nperm |= BLK_PERM_WRITE;
 175     }
 176     *nshared = BLK_PERM_CONSISTENT_READ
 177                | BLK_PERM_WRITE
 178                | BLK_PERM_WRITE_UNCHANGED;
 179     return;
 180 }
 181
 182 static int64_t replication_getlength(BlockDriverState *bs)
 183 {
 184     return bdrv_getlength(bs->file->bs);
 185 }
 186
 187 static int replication_get_io_status(BDRVReplicationState *s)
 188 {
 189     switch (s->stage) {
 190     case BLOCK_REPLICATION_NONE:
 191         return -EIO;
 192     case BLOCK_REPLICATION_RUNNING:
 193         return 0;
 194     case BLOCK_REPLICATION_FAILOVER:
 195         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 196     case BLOCK_REPLICATION_FAILOVER_FAILED:
 197         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
 198     case BLOCK_REPLICATION_DONE:
 199         /*
 200          * active commit job completes, and active disk and secondary_disk
 201          * is swapped, so we can operate bs->file directly
 202          */
 203         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 204     default:
 205         abort();
 206     }
 207 }
 208
 209 static int replication_return_value(BDRVReplicationState *s, int ret)
 210 {
 211     if (s->mode == REPLICATION_MODE_SECONDARY) {
 212         return ret;
 213     }
 214
 215     if (ret < 0) {
 216         s->error = ret;
 217         ret = 0;
 218     }
 219
 220     return ret;
 221 }
 222
 223 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
 224                                              int64_t sector_num,
 225                                              int remaining_sectors,
 226                                              QEMUIOVector *qiov)
 227 {
 228     BDRVReplicationState *s = bs->opaque;
 229     int ret;
 230
 231     if (s->mode == REPLICATION_MODE_PRIMARY) {
 232         /* We only use it to forward primary write requests */
 233         return -EIO;
 234     }
 235
 236     ret = replication_get_io_status(s);
 237     if (ret < 0) {
 238         return ret;
 239     }
 240
 241     ret = bdrv_co_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
 242                          remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 243
 244     return replication_return_value(s, ret);
 245 }
 246
 247 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
 248                                               int64_t sector_num,
 249                                               int remaining_sectors,
 250                                               QEMUIOVector *qiov,
 251                                               int flags)
 252 {
 253     BDRVReplicationState *s = bs->opaque;
 254     QEMUIOVector hd_qiov;
 255     uint64_t bytes_done = 0;
 256     BdrvChild *top = bs->file;
 257     BdrvChild *base = s->secondary_disk;
 258     BdrvChild *target;
 259     int ret;
 260     int64_t n;
 261
 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(BlockDriverState *bs, Error **errp)
 313 {
 314     BDRVReplicationState *s = bs->opaque;
 315     BdrvChild *active_disk = bs->file;
 316     Error *local_err = NULL;
 317     int ret;
 318
 319     if (!s->backup_job) {
 320         error_setg(errp, "Backup job was cancelled unexpectedly");
 321         return;
 322     }
 323
 324     backup_do_checkpoint(s->backup_job, &local_err);
 325     if (local_err) {
 326         error_propagate(errp, local_err);
 327         return;
 328     }
 329
 330     if (!active_disk->bs->drv) {
 331         error_setg(errp, "Active disk %s is ejected",
 332                    active_disk->bs->node_name);
 333         return;
 334     }
 335
 336     ret = bdrv_make_empty(active_disk, errp);
 337     if (ret < 0) {
 338         return;
 339     }
 340
 341     if (!s->hidden_disk->bs->drv) {
 342         error_setg(errp, "Hidden disk %s is ejected",
 343                    s->hidden_disk->bs->node_name);
 344         return;
 345     }
 346
 347     ret = bdrv_make_empty(s->hidden_disk, errp);
 348     if (ret < 0) {
 349         return;
 350     }
 351 }
 352
 353 /* This function is supposed to be called twice:
 354  * first with writable = true, then with writable = false.
 355  * The first call puts s->hidden_disk and s->secondary_disk in
 356  * r/w mode, and the second puts them back in their original state.
 357  */
 358 static void reopen_backing_file(BlockDriverState *bs, bool writable,
 359                                 Error **errp)
 360 {
 361     BDRVReplicationState *s = bs->opaque;
 362     BdrvChild *hidden_disk, *secondary_disk;
 363     BlockReopenQueue *reopen_queue = NULL;
 364
 365     /*
 366      * s->hidden_disk and s->secondary_disk may not be set yet, as they will
 367      * only be set after the children are writable.
 368      */
 369     hidden_disk = bs->file->bs->backing;
 370     secondary_disk = hidden_disk->bs->backing;
 371
 372     if (writable) {
 373         s->orig_hidden_read_only = bdrv_is_read_only(hidden_disk->bs);
 374         s->orig_secondary_read_only = bdrv_is_read_only(secondary_disk->bs);
 375     }
 376
 377     bdrv_subtree_drained_begin(hidden_disk->bs);
 378     bdrv_subtree_drained_begin(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, 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, secondary_disk->bs,
 391                                          opts, true);
 392     }
 393
 394     if (reopen_queue) {
 395         AioContext *ctx = bdrv_get_aio_context(bs);
 396         if (ctx != qemu_get_aio_context()) {
 397             aio_context_release(ctx);
 398         }
 399         bdrv_reopen_multiple(reopen_queue, errp);
 400         if (ctx != qemu_get_aio_context()) {
 401             aio_context_acquire(ctx);
 402         }
 403     }
 404
 405     bdrv_subtree_drained_end(hidden_disk->bs);
 406     bdrv_subtree_drained_end(secondary_disk->bs);
 407 }
 408
 409 static void backup_job_cleanup(BlockDriverState *bs)
 410 {
 411     BDRVReplicationState *s = bs->opaque;
 412     BlockDriverState *top_bs;
 413
 414     s->backup_job = NULL;
 415
 416     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 417     if (!top_bs) {
 418         return;
 419     }
 420     bdrv_op_unblock_all(top_bs, s->blocker);
 421     error_free(s->blocker);
 422     reopen_backing_file(bs, false, NULL);
 423 }
 424
 425 static void backup_job_completed(void *opaque, int ret)
 426 {
 427     BlockDriverState *bs = opaque;
 428     BDRVReplicationState *s = bs->opaque;
 429
 430     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 431         /* The backup job is cancelled unexpectedly */
 432         s->error = -EIO;
 433     }
 434
 435     backup_job_cleanup(bs);
 436 }
 437
 438 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 439 {
 440     BdrvChild *child;
 441
 442     /* The bs itself is the top_bs */
 443     if (top_bs == bs) {
 444         return true;
 445     }
 446
 447     /* Iterate over top_bs's children */
 448     QLIST_FOREACH(child, &top_bs->children, next) {
 449         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 450             return true;
 451         }
 452     }
 453
 454     return false;
 455 }
 456
 457 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 458                               Error **errp)
 459 {
 460     BlockDriverState *bs = rs->opaque;
 461     BDRVReplicationState *s;
 462     BlockDriverState *top_bs;
 463     BdrvChild *active_disk, *hidden_disk, *secondary_disk;
 464     int64_t active_length, hidden_length, disk_length;
 465     AioContext *aio_context;
 466     Error *local_err = NULL;
 467     BackupPerf perf = { .use_copy_range = true, .max_workers = 1 };
 468
 469     aio_context = bdrv_get_aio_context(bs);
 470     aio_context_acquire(aio_context);
 471     s = bs->opaque;
 472
 473     if (s->stage == BLOCK_REPLICATION_DONE ||
 474         s->stage == BLOCK_REPLICATION_FAILOVER) {
 475         /*
 476          * This case happens when a secondary is promoted to primary.
 477          * Ignore the request because the secondary side of replication
 478          * doesn't have to do anything anymore.
 479          */
 480         aio_context_release(aio_context);
 481         return;
 482     }
 483
 484     if (s->stage != BLOCK_REPLICATION_NONE) {
 485         error_setg(errp, "Block replication is running or done");
 486         aio_context_release(aio_context);
 487         return;
 488     }
 489
 490     if (s->mode != mode) {
 491         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 492                    " but got %d", s->mode, mode);
 493         aio_context_release(aio_context);
 494         return;
 495     }
 496
 497     switch (s->mode) {
 498     case REPLICATION_MODE_PRIMARY:
 499         break;
 500     case REPLICATION_MODE_SECONDARY:
 501         active_disk = bs->file;
 502         if (!active_disk || !active_disk->bs || !active_disk->bs->backing) {
 503             error_setg(errp, "Active disk doesn't have backing file");
 504             aio_context_release(aio_context);
 505             return;
 506         }
 507
 508         hidden_disk = active_disk->bs->backing;
 509         if (!hidden_disk->bs || !hidden_disk->bs->backing) {
 510             error_setg(errp, "Hidden disk doesn't have backing file");
 511             aio_context_release(aio_context);
 512             return;
 513         }
 514
 515         secondary_disk = hidden_disk->bs->backing;
 516         if (!secondary_disk->bs || !bdrv_has_blk(secondary_disk->bs)) {
 517             error_setg(errp, "The secondary disk doesn't have block backend");
 518             aio_context_release(aio_context);
 519             return;
 520         }
 521
 522         /* verify the length */
 523         active_length = bdrv_getlength(active_disk->bs);
 524         hidden_length = bdrv_getlength(hidden_disk->bs);
 525         disk_length = bdrv_getlength(secondary_disk->bs);
 526         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 527             active_length != hidden_length || hidden_length != disk_length) {
 528             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 529                        " are not the same");
 530             aio_context_release(aio_context);
 531             return;
 532         }
 533
 534         /* Must be true, or the bdrv_getlength() calls would have failed */
 535         assert(active_disk->bs->drv && hidden_disk->bs->drv);
 536
 537         if (!active_disk->bs->drv->bdrv_make_empty ||
 538             !hidden_disk->bs->drv->bdrv_make_empty) {
 539             error_setg(errp,
 540                        "Active disk or hidden disk doesn't support make_empty");
 541             aio_context_release(aio_context);
 542             return;
 543         }
 544
 545         /* reopen the backing file in r/w mode */
 546         reopen_backing_file(bs, true, &local_err);
 547         if (local_err) {
 548             error_propagate(errp, local_err);
 549             aio_context_release(aio_context);
 550             return;
 551         }
 552
 553         bdrv_ref(hidden_disk->bs);
 554         s->hidden_disk = bdrv_attach_child(bs, hidden_disk->bs, "hidden disk",
 555                                            &child_of_bds, BDRV_CHILD_DATA,
 556                                            &local_err);
 557         if (local_err) {
 558             error_propagate(errp, local_err);
 559             aio_context_release(aio_context);
 560             return;
 561         }
 562
 563         bdrv_ref(secondary_disk->bs);
 564         s->secondary_disk = bdrv_attach_child(bs, secondary_disk->bs,
 565                                               "secondary disk", &child_of_bds,
 566                                               BDRV_CHILD_DATA, &local_err);
 567         if (local_err) {
 568             error_propagate(errp, local_err);
 569             aio_context_release(aio_context);
 570             return;
 571         }
 572
 573         /* start backup job now */
 574         error_setg(&s->blocker,
 575                    "Block device is in use by internal backup job");
 576
 577         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 578         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 579             !check_top_bs(top_bs, bs)) {
 580             error_setg(errp, "No top_bs or it is invalid");
 581             reopen_backing_file(bs, false, NULL);
 582             aio_context_release(aio_context);
 583             return;
 584         }
 585         bdrv_op_block_all(top_bs, s->blocker);
 586         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 587
 588         s->backup_job = backup_job_create(
 589                                 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 590                                 0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
 591                                 &perf,
 592                                 BLOCKDEV_ON_ERROR_REPORT,
 593                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
 594                                 backup_job_completed, bs, NULL, &local_err);
 595         if (local_err) {
 596             error_propagate(errp, local_err);
 597             backup_job_cleanup(bs);
 598             aio_context_release(aio_context);
 599             return;
 600         }
 601         job_start(&s->backup_job->job);
 602         break;
 603     default:
 604         aio_context_release(aio_context);
 605         abort();
 606     }
 607
 608     s->stage = BLOCK_REPLICATION_RUNNING;
 609
 610     if (s->mode == REPLICATION_MODE_SECONDARY) {
 611         secondary_do_checkpoint(bs, errp);
 612     }
 613
 614     s->error = 0;
 615     aio_context_release(aio_context);
 616 }
 617
 618 static void replication_do_checkpoint(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_DONE ||
 629         s->stage == BLOCK_REPLICATION_FAILOVER) {
 630         /*
 631          * This case happens when a secondary was promoted to primary.
 632          * Ignore the request because the secondary side of replication
 633          * doesn't have to do anything anymore.
 634          */
 635         aio_context_release(aio_context);
 636         return;
 637     }
 638
 639     if (s->mode == REPLICATION_MODE_SECONDARY) {
 640         secondary_do_checkpoint(bs, errp);
 641     }
 642     aio_context_release(aio_context);
 643 }
 644
 645 static void replication_get_error(ReplicationState *rs, Error **errp)
 646 {
 647     BlockDriverState *bs = rs->opaque;
 648     BDRVReplicationState *s;
 649     AioContext *aio_context;
 650
 651     aio_context = bdrv_get_aio_context(bs);
 652     aio_context_acquire(aio_context);
 653     s = bs->opaque;
 654
 655     if (s->stage == BLOCK_REPLICATION_NONE) {
 656         error_setg(errp, "Block replication is not running");
 657         aio_context_release(aio_context);
 658         return;
 659     }
 660
 661     if (s->error) {
 662         error_setg(errp, "I/O error occurred");
 663         aio_context_release(aio_context);
 664         return;
 665     }
 666     aio_context_release(aio_context);
 667 }
 668
 669 static void replication_done(void *opaque, int ret)
 670 {
 671     BlockDriverState *bs = opaque;
 672     BDRVReplicationState *s = bs->opaque;
 673
 674     if (ret == 0) {
 675         s->stage = BLOCK_REPLICATION_DONE;
 676
 677         bdrv_unref_child(bs, s->secondary_disk);
 678         s->secondary_disk = NULL;
 679         bdrv_unref_child(bs, s->hidden_disk);
 680         s->hidden_disk = NULL;
 681         s->error = 0;
 682     } else {
 683         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 684         s->error = -EIO;
 685     }
 686 }
 687
 688 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 689 {
 690     BlockDriverState *bs = rs->opaque;
 691     BDRVReplicationState *s;
 692     AioContext *aio_context;
 693
 694     aio_context = bdrv_get_aio_context(bs);
 695     aio_context_acquire(aio_context);
 696     s = bs->opaque;
 697
 698     if (s->stage == BLOCK_REPLICATION_DONE ||
 699         s->stage == BLOCK_REPLICATION_FAILOVER) {
 700         /*
 701          * This case happens when a secondary was promoted to primary.
 702          * Ignore the request because the secondary side of replication
 703          * doesn't have to do anything anymore.
 704          */
 705         aio_context_release(aio_context);
 706         return;
 707     }
 708
 709     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 710         error_setg(errp, "Block replication is not running");
 711         aio_context_release(aio_context);
 712         return;
 713     }
 714
 715     switch (s->mode) {
 716     case REPLICATION_MODE_PRIMARY:
 717         s->stage = BLOCK_REPLICATION_DONE;
 718         s->error = 0;
 719         break;
 720     case REPLICATION_MODE_SECONDARY:
 721         /*
 722          * This BDS will be closed, and the job should be completed
 723          * before the BDS is closed, because we will access hidden
 724          * disk, secondary disk in backup_job_completed().
 725          */
 726         if (s->backup_job) {
 727             aio_context_release(aio_context);
 728             job_cancel_sync(&s->backup_job->job, true);
 729             aio_context_acquire(aio_context);
 730         }
 731
 732         if (!failover) {
 733             secondary_do_checkpoint(bs, errp);
 734             s->stage = BLOCK_REPLICATION_DONE;
 735             aio_context_release(aio_context);
 736             return;
 737         }
 738
 739         s->stage = BLOCK_REPLICATION_FAILOVER;
 740         s->commit_job = commit_active_start(
 741                             NULL, bs->file->bs, s->secondary_disk->bs,
 742                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 743                             NULL, replication_done, bs, true, errp);
 744         break;
 745     default:
 746         aio_context_release(aio_context);
 747         abort();
 748     }
 749     aio_context_release(aio_context);
 750 }
 751
 752 static const char *const replication_strong_runtime_opts[] = {
 753     REPLICATION_MODE,
 754     REPLICATION_TOP_ID,
 755
 756     NULL
 757 };
 758
 759 static BlockDriver bdrv_replication = {
 760     .format_name                = "replication",
 761     .instance_size              = sizeof(BDRVReplicationState),
 762
 763     .bdrv_open                  = replication_open,
 764     .bdrv_close                 = replication_close,
 765     .bdrv_child_perm            = replication_child_perm,
 766
 767     .bdrv_getlength             = replication_getlength,
 768     .bdrv_co_readv              = replication_co_readv,
 769     .bdrv_co_writev             = replication_co_writev,
 770
 771     .is_filter                  = true,
 772
 773     .has_variable_length        = true,
 774     .strong_runtime_opts        = replication_strong_runtime_opts,
 775 };
 776
 777 static void bdrv_replication_init(void)
 778 {
 779     bdrv_register(&bdrv_replication);
 780 }
 781
 782 block_init(bdrv_replication_init);