block/replication.c

   1 /*
   2  * Replication Block filter
   3  *
   4  * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
   5  * Copyright (c) 2016 Intel Corporation
   6  * Copyright (c) 2016 FUJITSU LIMITED
   7  *
   8  * Author:
   9  *   Wen Congyang <wency@cn.fujitsu.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  12  * See the COPYING file in the top-level directory.
  13  */
  14
  15 #include "qemu/osdep.h"
  16 #include "qemu/option.h"
  17 #include "block/nbd.h"
  18 #include "block/blockjob.h"
  19 #include "block/block_int.h"
  20 #include "block/block_backup.h"
  21 #include "sysemu/block-backend.h"
  22 #include "qapi/error.h"
  23 #include "replication.h"
  24
  25 typedef enum {
  26     BLOCK_REPLICATION_NONE,             /* block replication is not started */
  27     BLOCK_REPLICATION_RUNNING,          /* block replication is running */
  28     BLOCK_REPLICATION_FAILOVER,         /* failover is running in background */
  29     BLOCK_REPLICATION_FAILOVER_FAILED,  /* failover failed */
  30     BLOCK_REPLICATION_DONE,             /* block replication is done */
  31 } ReplicationStage;
  32
  33 typedef struct BDRVReplicationState {
  34     ReplicationMode mode;
  35     ReplicationStage stage;
  36     BdrvChild *active_disk;
  37     BdrvChild *hidden_disk;
  38     BdrvChild *secondary_disk;
  39     char *top_id;
  40     ReplicationState *rs;
  41     Error *blocker;
  42     int orig_hidden_flags;
  43     int orig_secondary_flags;
  44     int error;
  45 } BDRVReplicationState;
  46
  47 static void replication_start(ReplicationState *rs, ReplicationMode mode,
  48                               Error **errp);
  49 static void replication_do_checkpoint(ReplicationState *rs, Error **errp);
  50 static void replication_get_error(ReplicationState *rs, Error **errp);
  51 static void replication_stop(ReplicationState *rs, bool failover,
  52                              Error **errp);
  53
  54 #define REPLICATION_MODE        "mode"
  55 #define REPLICATION_TOP_ID      "top-id"
  56 static QemuOptsList replication_runtime_opts = {
  57     .name = "replication",
  58     .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head),
  59     .desc = {
  60         {
  61             .name = REPLICATION_MODE,
  62             .type = QEMU_OPT_STRING,
  63         },
  64         {
  65             .name = REPLICATION_TOP_ID,
  66             .type = QEMU_OPT_STRING,
  67         },
  68         { /* end of list */ }
  69     },
  70 };
  71
  72 static ReplicationOps replication_ops = {
  73     .start = replication_start,
  74     .checkpoint = replication_do_checkpoint,
  75     .get_error = replication_get_error,
  76     .stop = replication_stop,
  77 };
  78
  79 static int replication_open(BlockDriverState *bs, QDict *options,
  80                             int flags, Error **errp)
  81 {
  82     int ret;
  83     BDRVReplicationState *s = bs->opaque;
  84     Error *local_err = NULL;
  85     QemuOpts *opts = NULL;
  86     const char *mode;
  87     const char *top_id;
  88
  89     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
  90                                false, errp);
  91     if (!bs->file) {
  92         return -EINVAL;
  93     }
  94
  95     ret = -EINVAL;
  96     opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
  97     qemu_opts_absorb_qdict(opts, options, &local_err);
  98     if (local_err) {
  99         goto fail;
 100     }
 101
 102     mode = qemu_opt_get(opts, REPLICATION_MODE);
 103     if (!mode) {
 104         error_setg(&local_err, "Missing the option mode");
 105         goto fail;
 106     }
 107
 108     if (!strcmp(mode, "primary")) {
 109         s->mode = REPLICATION_MODE_PRIMARY;
 110         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 111         if (top_id) {
 112             error_setg(&local_err, "The primary side does not support option top-id");
 113             goto fail;
 114         }
 115     } else if (!strcmp(mode, "secondary")) {
 116         s->mode = REPLICATION_MODE_SECONDARY;
 117         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 118         s->top_id = g_strdup(top_id);
 119         if (!s->top_id) {
 120             error_setg(&local_err, "Missing the option top-id");
 121             goto fail;
 122         }
 123     } else {
 124         error_setg(&local_err,
 125                    "The option mode's value should be primary or secondary");
 126         goto fail;
 127     }
 128
 129     s->rs = replication_new(bs, &replication_ops);
 130
 131     ret = 0;
 132
 133 fail:
 134     qemu_opts_del(opts);
 135     error_propagate(errp, local_err);
 136
 137     return ret;
 138 }
 139
 140 static void replication_close(BlockDriverState *bs)
 141 {
 142     BDRVReplicationState *s = bs->opaque;
 143
 144     if (s->stage == BLOCK_REPLICATION_RUNNING) {
 145         replication_stop(s->rs, false, NULL);
 146     }
 147     if (s->stage == BLOCK_REPLICATION_FAILOVER) {
 148         job_cancel_sync(&s->active_disk->bs->job->job);
 149     }
 150
 151     if (s->mode == REPLICATION_MODE_SECONDARY) {
 152         g_free(s->top_id);
 153     }
 154
 155     replication_remove(s->rs);
 156 }
 157
 158 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
 159                                    const BdrvChildRole *role,
 160                                    BlockReopenQueue *reopen_queue,
 161                                    uint64_t perm, uint64_t shared,
 162                                    uint64_t *nperm, uint64_t *nshared)
 163 {
 164     *nperm = BLK_PERM_CONSISTENT_READ;
 165     if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) {
 166         *nperm |= BLK_PERM_WRITE;
 167     }
 168     *nshared = BLK_PERM_CONSISTENT_READ \
 169                | BLK_PERM_WRITE \
 170                | BLK_PERM_WRITE_UNCHANGED;
 171     return;
 172 }
 173
 174 static int64_t replication_getlength(BlockDriverState *bs)
 175 {
 176     return bdrv_getlength(bs->file->bs);
 177 }
 178
 179 static int replication_get_io_status(BDRVReplicationState *s)
 180 {
 181     switch (s->stage) {
 182     case BLOCK_REPLICATION_NONE:
 183         return -EIO;
 184     case BLOCK_REPLICATION_RUNNING:
 185         return 0;
 186     case BLOCK_REPLICATION_FAILOVER:
 187         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 188     case BLOCK_REPLICATION_FAILOVER_FAILED:
 189         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
 190     case BLOCK_REPLICATION_DONE:
 191         /*
 192          * active commit job completes, and active disk and secondary_disk
 193          * is swapped, so we can operate bs->file directly
 194          */
 195         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 196     default:
 197         abort();
 198     }
 199 }
 200
 201 static int replication_return_value(BDRVReplicationState *s, int ret)
 202 {
 203     if (s->mode == REPLICATION_MODE_SECONDARY) {
 204         return ret;
 205     }
 206
 207     if (ret < 0) {
 208         s->error = ret;
 209         ret = 0;
 210     }
 211
 212     return ret;
 213 }
 214
 215 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
 216                                              int64_t sector_num,
 217                                              int remaining_sectors,
 218                                              QEMUIOVector *qiov)
 219 {
 220     BDRVReplicationState *s = bs->opaque;
 221     BdrvChild *child = s->secondary_disk;
 222     BlockJob *job = NULL;
 223     CowRequest req;
 224     int ret;
 225
 226     if (s->mode == REPLICATION_MODE_PRIMARY) {
 227         /* We only use it to forward primary write requests */
 228         return -EIO;
 229     }
 230
 231     ret = replication_get_io_status(s);
 232     if (ret < 0) {
 233         return ret;
 234     }
 235
 236     if (child && child->bs) {
 237         job = child->bs->job;
 238     }
 239
 240     if (job) {
 241         uint64_t remaining_bytes = remaining_sectors * BDRV_SECTOR_SIZE;
 242
 243         backup_wait_for_overlapping_requests(child->bs->job,
 244                                              sector_num * BDRV_SECTOR_SIZE,
 245                                              remaining_bytes);
 246         backup_cow_request_begin(&req, child->bs->job,
 247                                  sector_num * BDRV_SECTOR_SIZE,
 248                                  remaining_bytes);
 249         ret = bdrv_co_readv(bs->file, sector_num, remaining_sectors,
 250                             qiov);
 251         backup_cow_request_end(&req);
 252         goto out;
 253     }
 254
 255     ret = bdrv_co_readv(bs->file, sector_num, remaining_sectors, qiov);
 256 out:
 257     return replication_return_value(s, ret);
 258 }
 259
 260 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
 261                                               int64_t sector_num,
 262                                               int remaining_sectors,
 263                                               QEMUIOVector *qiov,
 264                                               int flags)
 265 {
 266     BDRVReplicationState *s = bs->opaque;
 267     QEMUIOVector hd_qiov;
 268     uint64_t bytes_done = 0;
 269     BdrvChild *top = bs->file;
 270     BdrvChild *base = s->secondary_disk;
 271     BdrvChild *target;
 272     int ret;
 273     int64_t n;
 274
 275     assert(!flags);
 276     ret = replication_get_io_status(s);
 277     if (ret < 0) {
 278         goto out;
 279     }
 280
 281     if (ret == 0) {
 282         ret = bdrv_co_writev(top, sector_num,
 283                              remaining_sectors, qiov);
 284         return replication_return_value(s, ret);
 285     }
 286
 287     /*
 288      * Failover failed, only write to active disk if the sectors
 289      * have already been allocated in active disk/hidden disk.
 290      */
 291     qemu_iovec_init(&hd_qiov, qiov->niov);
 292     while (remaining_sectors > 0) {
 293         int64_t count;
 294
 295         ret = bdrv_is_allocated_above(top->bs, base->bs,
 296                                       sector_num * BDRV_SECTOR_SIZE,
 297                                       remaining_sectors * BDRV_SECTOR_SIZE,
 298                                       &count);
 299         if (ret < 0) {
 300             goto out1;
 301         }
 302
 303         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
 304         n = count >> BDRV_SECTOR_BITS;
 305         qemu_iovec_reset(&hd_qiov);
 306         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
 307
 308         target = ret ? top : base;
 309         ret = bdrv_co_writev(target, sector_num, n, &hd_qiov);
 310         if (ret < 0) {
 311             goto out1;
 312         }
 313
 314         remaining_sectors -= n;
 315         sector_num += n;
 316         bytes_done += count;
 317     }
 318
 319 out1:
 320     qemu_iovec_destroy(&hd_qiov);
 321 out:
 322     return ret;
 323 }
 324
 325 static bool replication_recurse_is_first_non_filter(BlockDriverState *bs,
 326                                                     BlockDriverState *candidate)
 327 {
 328     return bdrv_recurse_is_first_non_filter(bs->file->bs, candidate);
 329 }
 330
 331 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
 332 {
 333     Error *local_err = NULL;
 334     int ret;
 335
 336     if (!s->secondary_disk->bs->job) {
 337         error_setg(errp, "Backup job was cancelled unexpectedly");
 338         return;
 339     }
 340
 341     backup_do_checkpoint(s->secondary_disk->bs->job, &local_err);
 342     if (local_err) {
 343         error_propagate(errp, local_err);
 344         return;
 345     }
 346
 347     if (!s->active_disk->bs->drv) {
 348         error_setg(errp, "Active disk %s is ejected",
 349                    s->active_disk->bs->node_name);
 350         return;
 351     }
 352
 353     ret = s->active_disk->bs->drv->bdrv_make_empty(s->active_disk->bs);
 354     if (ret < 0) {
 355         error_setg(errp, "Cannot make active disk empty");
 356         return;
 357     }
 358
 359     if (!s->hidden_disk->bs->drv) {
 360         error_setg(errp, "Hidden disk %s is ejected",
 361                    s->hidden_disk->bs->node_name);
 362         return;
 363     }
 364
 365     ret = s->hidden_disk->bs->drv->bdrv_make_empty(s->hidden_disk->bs);
 366     if (ret < 0) {
 367         error_setg(errp, "Cannot make hidden disk empty");
 368         return;
 369     }
 370 }
 371
 372 static void reopen_backing_file(BlockDriverState *bs, bool writable,
 373                                 Error **errp)
 374 {
 375     BDRVReplicationState *s = bs->opaque;
 376     BlockReopenQueue *reopen_queue = NULL;
 377     int orig_hidden_flags, orig_secondary_flags;
 378     int new_hidden_flags, new_secondary_flags;
 379     Error *local_err = NULL;
 380
 381     if (writable) {
 382         orig_hidden_flags = s->orig_hidden_flags =
 383                                 bdrv_get_flags(s->hidden_disk->bs);
 384         new_hidden_flags = (orig_hidden_flags | BDRV_O_RDWR) &
 385                                                     ~BDRV_O_INACTIVE;
 386         orig_secondary_flags = s->orig_secondary_flags =
 387                                 bdrv_get_flags(s->secondary_disk->bs);
 388         new_secondary_flags = (orig_secondary_flags | BDRV_O_RDWR) &
 389                                                      ~BDRV_O_INACTIVE;
 390     } else {
 391         orig_hidden_flags = (s->orig_hidden_flags | BDRV_O_RDWR) &
 392                                                     ~BDRV_O_INACTIVE;
 393         new_hidden_flags = s->orig_hidden_flags;
 394         orig_secondary_flags = (s->orig_secondary_flags | BDRV_O_RDWR) &
 395                                                     ~BDRV_O_INACTIVE;
 396         new_secondary_flags = s->orig_secondary_flags;
 397     }
 398
 399     bdrv_subtree_drained_begin(s->hidden_disk->bs);
 400     bdrv_subtree_drained_begin(s->secondary_disk->bs);
 401
 402     if (orig_hidden_flags != new_hidden_flags) {
 403         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs, NULL,
 404                                          new_hidden_flags);
 405     }
 406
 407     if (!(orig_secondary_flags & BDRV_O_RDWR)) {
 408         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
 409                                          NULL, new_secondary_flags);
 410     }
 411
 412     if (reopen_queue) {
 413         bdrv_reopen_multiple(bdrv_get_aio_context(bs),
 414                              reopen_queue, &local_err);
 415         error_propagate(errp, local_err);
 416     }
 417
 418     bdrv_subtree_drained_end(s->hidden_disk->bs);
 419     bdrv_subtree_drained_end(s->secondary_disk->bs);
 420 }
 421
 422 static void backup_job_cleanup(BlockDriverState *bs)
 423 {
 424     BDRVReplicationState *s = bs->opaque;
 425     BlockDriverState *top_bs;
 426
 427     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 428     if (!top_bs) {
 429         return;
 430     }
 431     bdrv_op_unblock_all(top_bs, s->blocker);
 432     error_free(s->blocker);
 433     reopen_backing_file(bs, false, NULL);
 434 }
 435
 436 static void backup_job_completed(void *opaque, int ret)
 437 {
 438     BlockDriverState *bs = opaque;
 439     BDRVReplicationState *s = bs->opaque;
 440
 441     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 442         /* The backup job is cancelled unexpectedly */
 443         s->error = -EIO;
 444     }
 445
 446     backup_job_cleanup(bs);
 447 }
 448
 449 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 450 {
 451     BdrvChild *child;
 452
 453     /* The bs itself is the top_bs */
 454     if (top_bs == bs) {
 455         return true;
 456     }
 457
 458     /* Iterate over top_bs's children */
 459     QLIST_FOREACH(child, &top_bs->children, next) {
 460         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 461             return true;
 462         }
 463     }
 464
 465     return false;
 466 }
 467
 468 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 469                               Error **errp)
 470 {
 471     BlockDriverState *bs = rs->opaque;
 472     BDRVReplicationState *s;
 473     BlockDriverState *top_bs;
 474     int64_t active_length, hidden_length, disk_length;
 475     AioContext *aio_context;
 476     Error *local_err = NULL;
 477     BlockJob *job;
 478
 479     aio_context = bdrv_get_aio_context(bs);
 480     aio_context_acquire(aio_context);
 481     s = bs->opaque;
 482
 483     if (s->stage != BLOCK_REPLICATION_NONE) {
 484         error_setg(errp, "Block replication is running or done");
 485         aio_context_release(aio_context);
 486         return;
 487     }
 488
 489     if (s->mode != mode) {
 490         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 491                    " but got %d", s->mode, mode);
 492         aio_context_release(aio_context);
 493         return;
 494     }
 495
 496     switch (s->mode) {
 497     case REPLICATION_MODE_PRIMARY:
 498         break;
 499     case REPLICATION_MODE_SECONDARY:
 500         s->active_disk = bs->file;
 501         if (!s->active_disk || !s->active_disk->bs ||
 502                                     !s->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         s->hidden_disk = s->active_disk->bs->backing;
 509         if (!s->hidden_disk->bs || !s->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         s->secondary_disk = s->hidden_disk->bs->backing;
 516         if (!s->secondary_disk->bs || !bdrv_has_blk(s->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(s->active_disk->bs);
 524         hidden_length = bdrv_getlength(s->hidden_disk->bs);
 525         disk_length = bdrv_getlength(s->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(s->active_disk->bs->drv && s->hidden_disk->bs->drv);
 536
 537         if (!s->active_disk->bs->drv->bdrv_make_empty ||
 538             !s->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         /* start backup job now */
 554         error_setg(&s->blocker,
 555                    "Block device is in use by internal backup job");
 556
 557         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 558         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 559             !check_top_bs(top_bs, bs)) {
 560             error_setg(errp, "No top_bs or it is invalid");
 561             reopen_backing_file(bs, false, NULL);
 562             aio_context_release(aio_context);
 563             return;
 564         }
 565         bdrv_op_block_all(top_bs, s->blocker);
 566         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 567
 568         job = backup_job_create(NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 569                                 0, MIRROR_SYNC_MODE_NONE, NULL, false,
 570                                 BLOCKDEV_ON_ERROR_REPORT,
 571                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
 572                                 backup_job_completed, bs, NULL, &local_err);
 573         if (local_err) {
 574             error_propagate(errp, local_err);
 575             backup_job_cleanup(bs);
 576             aio_context_release(aio_context);
 577             return;
 578         }
 579         job_start(&job->job);
 580         break;
 581     default:
 582         aio_context_release(aio_context);
 583         abort();
 584     }
 585
 586     s->stage = BLOCK_REPLICATION_RUNNING;
 587
 588     if (s->mode == REPLICATION_MODE_SECONDARY) {
 589         secondary_do_checkpoint(s, errp);
 590     }
 591
 592     s->error = 0;
 593     aio_context_release(aio_context);
 594 }
 595
 596 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
 597 {
 598     BlockDriverState *bs = rs->opaque;
 599     BDRVReplicationState *s;
 600     AioContext *aio_context;
 601
 602     aio_context = bdrv_get_aio_context(bs);
 603     aio_context_acquire(aio_context);
 604     s = bs->opaque;
 605
 606     if (s->mode == REPLICATION_MODE_SECONDARY) {
 607         secondary_do_checkpoint(s, errp);
 608     }
 609     aio_context_release(aio_context);
 610 }
 611
 612 static void replication_get_error(ReplicationState *rs, Error **errp)
 613 {
 614     BlockDriverState *bs = rs->opaque;
 615     BDRVReplicationState *s;
 616     AioContext *aio_context;
 617
 618     aio_context = bdrv_get_aio_context(bs);
 619     aio_context_acquire(aio_context);
 620     s = bs->opaque;
 621
 622     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 623         error_setg(errp, "Block replication is not running");
 624         aio_context_release(aio_context);
 625         return;
 626     }
 627
 628     if (s->error) {
 629         error_setg(errp, "I/O error occurred");
 630         aio_context_release(aio_context);
 631         return;
 632     }
 633     aio_context_release(aio_context);
 634 }
 635
 636 static void replication_done(void *opaque, int ret)
 637 {
 638     BlockDriverState *bs = opaque;
 639     BDRVReplicationState *s = bs->opaque;
 640
 641     if (ret == 0) {
 642         s->stage = BLOCK_REPLICATION_DONE;
 643
 644         /* refresh top bs's filename */
 645         bdrv_refresh_filename(bs);
 646         s->active_disk = NULL;
 647         s->secondary_disk = NULL;
 648         s->hidden_disk = NULL;
 649         s->error = 0;
 650     } else {
 651         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 652         s->error = -EIO;
 653     }
 654 }
 655
 656 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 657 {
 658     BlockDriverState *bs = rs->opaque;
 659     BDRVReplicationState *s;
 660     AioContext *aio_context;
 661
 662     aio_context = bdrv_get_aio_context(bs);
 663     aio_context_acquire(aio_context);
 664     s = bs->opaque;
 665
 666     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 667         error_setg(errp, "Block replication is not running");
 668         aio_context_release(aio_context);
 669         return;
 670     }
 671
 672     switch (s->mode) {
 673     case REPLICATION_MODE_PRIMARY:
 674         s->stage = BLOCK_REPLICATION_DONE;
 675         s->error = 0;
 676         break;
 677     case REPLICATION_MODE_SECONDARY:
 678         /*
 679          * This BDS will be closed, and the job should be completed
 680          * before the BDS is closed, because we will access hidden
 681          * disk, secondary disk in backup_job_completed().
 682          */
 683         if (s->secondary_disk->bs->job) {
 684             job_cancel_sync(&s->secondary_disk->bs->job->job);
 685         }
 686
 687         if (!failover) {
 688             secondary_do_checkpoint(s, errp);
 689             s->stage = BLOCK_REPLICATION_DONE;
 690             aio_context_release(aio_context);
 691             return;
 692         }
 693
 694         s->stage = BLOCK_REPLICATION_FAILOVER;
 695         commit_active_start(NULL, s->active_disk->bs, s->secondary_disk->bs,
 696                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 697                             NULL, replication_done, bs, true, errp);
 698         break;
 699     default:
 700         aio_context_release(aio_context);
 701         abort();
 702     }
 703     aio_context_release(aio_context);
 704 }
 705
 706 BlockDriver bdrv_replication = {
 707     .format_name                = "replication",
 708     .instance_size              = sizeof(BDRVReplicationState),
 709
 710     .bdrv_open                  = replication_open,
 711     .bdrv_close                 = replication_close,
 712     .bdrv_child_perm            = replication_child_perm,
 713
 714     .bdrv_getlength             = replication_getlength,
 715     .bdrv_co_readv              = replication_co_readv,
 716     .bdrv_co_writev             = replication_co_writev,
 717
 718     .is_filter                  = true,
 719     .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter,
 720
 721     .has_variable_length        = true,
 722 };
 723
 724 static void bdrv_replication_init(void)
 725 {
 726     bdrv_register(&bdrv_replication);
 727 }
 728
 729 block_init(bdrv_replication_init);