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     qemu_opts_absorb_qdict(opts, options, &local_err);
 103     if (local_err) {
 104         goto fail;
 105     }
 106
 107     mode = qemu_opt_get(opts, REPLICATION_MODE);
 108     if (!mode) {
 109         error_setg(&local_err, "Missing the option mode");
 110         goto fail;
 111     }
 112
 113     if (!strcmp(mode, "primary")) {
 114         s->mode = REPLICATION_MODE_PRIMARY;
 115         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 116         if (top_id) {
 117             error_setg(&local_err, "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(&local_err, "Missing the option top-id");
 126             goto fail;
 127         }
 128     } else {
 129         error_setg(&local_err,
 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     error_propagate(errp, local_err);
 141
 142     return ret;
 143 }
 144
 145 static void replication_close(BlockDriverState *bs)
 146 {
 147     BDRVReplicationState *s = bs->opaque;
 148     Job *commit_job;
 149
 150     if (s->stage == BLOCK_REPLICATION_RUNNING) {
 151         replication_stop(s->rs, false, NULL);
 152     }
 153     if (s->stage == BLOCK_REPLICATION_FAILOVER) {
 154         commit_job = &s->commit_job->job;
 155         assert(commit_job->aio_context == qemu_get_current_aio_context());
 156         job_cancel_sync(commit_job);
 157     }
 158
 159     if (s->mode == REPLICATION_MODE_SECONDARY) {
 160         g_free(s->top_id);
 161     }
 162
 163     replication_remove(s->rs);
 164 }
 165
 166 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
 167                                    BdrvChildRole role,
 168                                    BlockReopenQueue *reopen_queue,
 169                                    uint64_t perm, uint64_t shared,
 170                                    uint64_t *nperm, uint64_t *nshared)
 171 {
 172     *nperm = BLK_PERM_CONSISTENT_READ;
 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     assert(!flags);
 263     ret = replication_get_io_status(s);
 264     if (ret < 0) {
 265         goto out;
 266     }
 267
 268     if (ret == 0) {
 269         ret = bdrv_co_pwritev(top, sector_num * BDRV_SECTOR_SIZE,
 270                               remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 271         return replication_return_value(s, ret);
 272     }
 273
 274     /*
 275      * Failover failed, only write to active disk if the sectors
 276      * have already been allocated in active disk/hidden disk.
 277      */
 278     qemu_iovec_init(&hd_qiov, qiov->niov);
 279     while (remaining_sectors > 0) {
 280         int64_t count;
 281
 282         ret = bdrv_is_allocated_above(top->bs, base->bs, false,
 283                                       sector_num * BDRV_SECTOR_SIZE,
 284                                       remaining_sectors * BDRV_SECTOR_SIZE,
 285                                       &count);
 286         if (ret < 0) {
 287             goto out1;
 288         }
 289
 290         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
 291         n = count >> BDRV_SECTOR_BITS;
 292         qemu_iovec_reset(&hd_qiov);
 293         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
 294
 295         target = ret ? top : base;
 296         ret = bdrv_co_pwritev(target, sector_num * BDRV_SECTOR_SIZE,
 297                               n * BDRV_SECTOR_SIZE, &hd_qiov, 0);
 298         if (ret < 0) {
 299             goto out1;
 300         }
 301
 302         remaining_sectors -= n;
 303         sector_num += n;
 304         bytes_done += count;
 305     }
 306
 307 out1:
 308     qemu_iovec_destroy(&hd_qiov);
 309 out:
 310     return ret;
 311 }
 312
 313 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
 314 {
 315     Error *local_err = NULL;
 316     int ret;
 317
 318     if (!s->backup_job) {
 319         error_setg(errp, "Backup job was cancelled unexpectedly");
 320         return;
 321     }
 322
 323     backup_do_checkpoint(s->backup_job, &local_err);
 324     if (local_err) {
 325         error_propagate(errp, local_err);
 326         return;
 327     }
 328
 329     if (!s->active_disk->bs->drv) {
 330         error_setg(errp, "Active disk %s is ejected",
 331                    s->active_disk->bs->node_name);
 332         return;
 333     }
 334
 335     ret = bdrv_make_empty(s->active_disk, errp);
 336     if (ret < 0) {
 337         return;
 338     }
 339
 340     if (!s->hidden_disk->bs->drv) {
 341         error_setg(errp, "Hidden disk %s is ejected",
 342                    s->hidden_disk->bs->node_name);
 343         return;
 344     }
 345
 346     BlockBackend *blk = blk_new(qemu_get_current_aio_context(),
 347                                 BLK_PERM_WRITE, BLK_PERM_ALL);
 348     blk_insert_bs(blk, s->hidden_disk->bs, &local_err);
 349     if (local_err) {
 350         error_propagate(errp, local_err);
 351         blk_unref(blk);
 352         return;
 353     }
 354
 355     ret = blk_make_empty(blk, errp);
 356     blk_unref(blk);
 357     if (ret < 0) {
 358         return;
 359     }
 360 }
 361
 362 /* This function is supposed to be called twice:
 363  * first with writable = true, then with writable = false.
 364  * The first call puts s->hidden_disk and s->secondary_disk in
 365  * r/w mode, and the second puts them back in their original state.
 366  */
 367 static void reopen_backing_file(BlockDriverState *bs, bool writable,
 368                                 Error **errp)
 369 {
 370     BDRVReplicationState *s = bs->opaque;
 371     BlockReopenQueue *reopen_queue = NULL;
 372     Error *local_err = NULL;
 373
 374     if (writable) {
 375         s->orig_hidden_read_only = bdrv_is_read_only(s->hidden_disk->bs);
 376         s->orig_secondary_read_only = bdrv_is_read_only(s->secondary_disk->bs);
 377     }
 378
 379     bdrv_subtree_drained_begin(s->hidden_disk->bs);
 380     bdrv_subtree_drained_begin(s->secondary_disk->bs);
 381
 382     if (s->orig_hidden_read_only) {
 383         QDict *opts = qdict_new();
 384         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 385         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs,
 386                                          opts, true);
 387     }
 388
 389     if (s->orig_secondary_read_only) {
 390         QDict *opts = qdict_new();
 391         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 392         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
 393                                          opts, true);
 394     }
 395
 396     if (reopen_queue) {
 397         bdrv_reopen_multiple(reopen_queue, &local_err);
 398         error_propagate(errp, local_err);
 399     }
 400
 401     bdrv_subtree_drained_end(s->hidden_disk->bs);
 402     bdrv_subtree_drained_end(s->secondary_disk->bs);
 403 }
 404
 405 static void backup_job_cleanup(BlockDriverState *bs)
 406 {
 407     BDRVReplicationState *s = bs->opaque;
 408     BlockDriverState *top_bs;
 409
 410     s->backup_job = NULL;
 411
 412     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 413     if (!top_bs) {
 414         return;
 415     }
 416     bdrv_op_unblock_all(top_bs, s->blocker);
 417     error_free(s->blocker);
 418     reopen_backing_file(bs, false, NULL);
 419 }
 420
 421 static void backup_job_completed(void *opaque, int ret)
 422 {
 423     BlockDriverState *bs = opaque;
 424     BDRVReplicationState *s = bs->opaque;
 425
 426     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 427         /* The backup job is cancelled unexpectedly */
 428         s->error = -EIO;
 429     }
 430
 431     backup_job_cleanup(bs);
 432 }
 433
 434 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 435 {
 436     BdrvChild *child;
 437
 438     /* The bs itself is the top_bs */
 439     if (top_bs == bs) {
 440         return true;
 441     }
 442
 443     /* Iterate over top_bs's children */
 444     QLIST_FOREACH(child, &top_bs->children, next) {
 445         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 446             return true;
 447         }
 448     }
 449
 450     return false;
 451 }
 452
 453 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 454                               Error **errp)
 455 {
 456     BlockDriverState *bs = rs->opaque;
 457     BDRVReplicationState *s;
 458     BlockDriverState *top_bs;
 459     int64_t active_length, hidden_length, disk_length;
 460     AioContext *aio_context;
 461     Error *local_err = NULL;
 462
 463     aio_context = bdrv_get_aio_context(bs);
 464     aio_context_acquire(aio_context);
 465     s = bs->opaque;
 466
 467     if (s->stage == BLOCK_REPLICATION_DONE ||
 468         s->stage == BLOCK_REPLICATION_FAILOVER) {
 469         /*
 470          * This case happens when a secondary is promoted to primary.
 471          * Ignore the request because the secondary side of replication
 472          * doesn't have to do anything anymore.
 473          */
 474         aio_context_release(aio_context);
 475         return;
 476     }
 477
 478     if (s->stage != BLOCK_REPLICATION_NONE) {
 479         error_setg(errp, "Block replication is running or done");
 480         aio_context_release(aio_context);
 481         return;
 482     }
 483
 484     if (s->mode != mode) {
 485         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 486                    " but got %d", s->mode, mode);
 487         aio_context_release(aio_context);
 488         return;
 489     }
 490
 491     switch (s->mode) {
 492     case REPLICATION_MODE_PRIMARY:
 493         break;
 494     case REPLICATION_MODE_SECONDARY:
 495         s->active_disk = bs->file;
 496         if (!s->active_disk || !s->active_disk->bs ||
 497                                     !s->active_disk->bs->backing) {
 498             error_setg(errp, "Active disk doesn't have backing file");
 499             aio_context_release(aio_context);
 500             return;
 501         }
 502
 503         s->hidden_disk = s->active_disk->bs->backing;
 504         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
 505             error_setg(errp, "Hidden disk doesn't have backing file");
 506             aio_context_release(aio_context);
 507             return;
 508         }
 509
 510         s->secondary_disk = s->hidden_disk->bs->backing;
 511         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
 512             error_setg(errp, "The secondary disk doesn't have block backend");
 513             aio_context_release(aio_context);
 514             return;
 515         }
 516
 517         /* verify the length */
 518         active_length = bdrv_getlength(s->active_disk->bs);
 519         hidden_length = bdrv_getlength(s->hidden_disk->bs);
 520         disk_length = bdrv_getlength(s->secondary_disk->bs);
 521         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 522             active_length != hidden_length || hidden_length != disk_length) {
 523             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 524                        " are not the same");
 525             aio_context_release(aio_context);
 526             return;
 527         }
 528
 529         /* Must be true, or the bdrv_getlength() calls would have failed */
 530         assert(s->active_disk->bs->drv && s->hidden_disk->bs->drv);
 531
 532         if (!s->active_disk->bs->drv->bdrv_make_empty ||
 533             !s->hidden_disk->bs->drv->bdrv_make_empty) {
 534             error_setg(errp,
 535                        "Active disk or hidden disk doesn't support make_empty");
 536             aio_context_release(aio_context);
 537             return;
 538         }
 539
 540         /* reopen the backing file in r/w mode */
 541         reopen_backing_file(bs, true, &local_err);
 542         if (local_err) {
 543             error_propagate(errp, local_err);
 544             aio_context_release(aio_context);
 545             return;
 546         }
 547
 548         /* start backup job now */
 549         error_setg(&s->blocker,
 550                    "Block device is in use by internal backup job");
 551
 552         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 553         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 554             !check_top_bs(top_bs, bs)) {
 555             error_setg(errp, "No top_bs or it is invalid");
 556             reopen_backing_file(bs, false, NULL);
 557             aio_context_release(aio_context);
 558             return;
 559         }
 560         bdrv_op_block_all(top_bs, s->blocker);
 561         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 562
 563         s->backup_job = backup_job_create(
 564                                 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 565                                 0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
 566                                 BLOCKDEV_ON_ERROR_REPORT,
 567                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
 568                                 backup_job_completed, bs, NULL, &local_err);
 569         if (local_err) {
 570             error_propagate(errp, local_err);
 571             backup_job_cleanup(bs);
 572             aio_context_release(aio_context);
 573             return;
 574         }
 575         job_start(&s->backup_job->job);
 576         break;
 577     default:
 578         aio_context_release(aio_context);
 579         abort();
 580     }
 581
 582     s->stage = BLOCK_REPLICATION_RUNNING;
 583
 584     if (s->mode == REPLICATION_MODE_SECONDARY) {
 585         secondary_do_checkpoint(s, errp);
 586     }
 587
 588     s->error = 0;
 589     aio_context_release(aio_context);
 590 }
 591
 592 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
 593 {
 594     BlockDriverState *bs = rs->opaque;
 595     BDRVReplicationState *s;
 596     AioContext *aio_context;
 597
 598     aio_context = bdrv_get_aio_context(bs);
 599     aio_context_acquire(aio_context);
 600     s = bs->opaque;
 601
 602     if (s->stage == BLOCK_REPLICATION_DONE ||
 603         s->stage == BLOCK_REPLICATION_FAILOVER) {
 604         /*
 605          * This case happens when a secondary was promoted to primary.
 606          * Ignore the request because the secondary side of replication
 607          * doesn't have to do anything anymore.
 608          */
 609         aio_context_release(aio_context);
 610         return;
 611     }
 612
 613     if (s->mode == REPLICATION_MODE_SECONDARY) {
 614         secondary_do_checkpoint(s, errp);
 615     }
 616     aio_context_release(aio_context);
 617 }
 618
 619 static void replication_get_error(ReplicationState *rs, Error **errp)
 620 {
 621     BlockDriverState *bs = rs->opaque;
 622     BDRVReplicationState *s;
 623     AioContext *aio_context;
 624
 625     aio_context = bdrv_get_aio_context(bs);
 626     aio_context_acquire(aio_context);
 627     s = bs->opaque;
 628
 629     if (s->stage == BLOCK_REPLICATION_NONE) {
 630         error_setg(errp, "Block replication is not running");
 631         aio_context_release(aio_context);
 632         return;
 633     }
 634
 635     if (s->error) {
 636         error_setg(errp, "I/O error occurred");
 637         aio_context_release(aio_context);
 638         return;
 639     }
 640     aio_context_release(aio_context);
 641 }
 642
 643 static void replication_done(void *opaque, int ret)
 644 {
 645     BlockDriverState *bs = opaque;
 646     BDRVReplicationState *s = bs->opaque;
 647
 648     if (ret == 0) {
 649         s->stage = BLOCK_REPLICATION_DONE;
 650
 651         s->active_disk = NULL;
 652         s->secondary_disk = NULL;
 653         s->hidden_disk = NULL;
 654         s->error = 0;
 655     } else {
 656         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 657         s->error = -EIO;
 658     }
 659 }
 660
 661 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 662 {
 663     BlockDriverState *bs = rs->opaque;
 664     BDRVReplicationState *s;
 665     AioContext *aio_context;
 666
 667     aio_context = bdrv_get_aio_context(bs);
 668     aio_context_acquire(aio_context);
 669     s = bs->opaque;
 670
 671     if (s->stage == BLOCK_REPLICATION_DONE ||
 672         s->stage == BLOCK_REPLICATION_FAILOVER) {
 673         /*
 674          * This case happens when a secondary was promoted to primary.
 675          * Ignore the request because the secondary side of replication
 676          * doesn't have to do anything anymore.
 677          */
 678         aio_context_release(aio_context);
 679         return;
 680     }
 681
 682     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 683         error_setg(errp, "Block replication is not running");
 684         aio_context_release(aio_context);
 685         return;
 686     }
 687
 688     switch (s->mode) {
 689     case REPLICATION_MODE_PRIMARY:
 690         s->stage = BLOCK_REPLICATION_DONE;
 691         s->error = 0;
 692         break;
 693     case REPLICATION_MODE_SECONDARY:
 694         /*
 695          * This BDS will be closed, and the job should be completed
 696          * before the BDS is closed, because we will access hidden
 697          * disk, secondary disk in backup_job_completed().
 698          */
 699         if (s->backup_job) {
 700             job_cancel_sync(&s->backup_job->job);
 701         }
 702
 703         if (!failover) {
 704             secondary_do_checkpoint(s, errp);
 705             s->stage = BLOCK_REPLICATION_DONE;
 706             aio_context_release(aio_context);
 707             return;
 708         }
 709
 710         s->stage = BLOCK_REPLICATION_FAILOVER;
 711         s->commit_job = commit_active_start(
 712                             NULL, s->active_disk->bs, s->secondary_disk->bs,
 713                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 714                             NULL, replication_done, bs, true, errp);
 715         break;
 716     default:
 717         aio_context_release(aio_context);
 718         abort();
 719     }
 720     aio_context_release(aio_context);
 721 }
 722
 723 static const char *const replication_strong_runtime_opts[] = {
 724     REPLICATION_MODE,
 725     REPLICATION_TOP_ID,
 726
 727     NULL
 728 };
 729
 730 static BlockDriver bdrv_replication = {
 731     .format_name                = "replication",
 732     .instance_size              = sizeof(BDRVReplicationState),
 733
 734     .bdrv_open                  = replication_open,
 735     .bdrv_close                 = replication_close,
 736     .bdrv_child_perm            = replication_child_perm,
 737
 738     .bdrv_getlength             = replication_getlength,
 739     .bdrv_co_readv              = replication_co_readv,
 740     .bdrv_co_writev             = replication_co_writev,
 741
 742     .is_filter                  = true,
 743
 744     .has_variable_length        = true,
 745     .strong_runtime_opts        = replication_strong_runtime_opts,
 746 };
 747
 748 static void bdrv_replication_init(void)
 749 {
 750     bdrv_register(&bdrv_replication);
 751 }
 752
 753 block_init(bdrv_replication_init);