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_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
 250                              remaining_bytes, qiov, 0);
 251         backup_cow_request_end(&req);
 252         goto out;
 253     }
 254
 255     ret = bdrv_co_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
 256                          remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 257 out:
 258     return replication_return_value(s, ret);
 259 }
 260
 261 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
 262                                               int64_t sector_num,
 263                                               int remaining_sectors,
 264                                               QEMUIOVector *qiov,
 265                                               int flags)
 266 {
 267     BDRVReplicationState *s = bs->opaque;
 268     QEMUIOVector hd_qiov;
 269     uint64_t bytes_done = 0;
 270     BdrvChild *top = bs->file;
 271     BdrvChild *base = s->secondary_disk;
 272     BdrvChild *target;
 273     int ret;
 274     int64_t n;
 275
 276     assert(!flags);
 277     ret = replication_get_io_status(s);
 278     if (ret < 0) {
 279         goto out;
 280     }
 281
 282     if (ret == 0) {
 283         ret = bdrv_co_pwritev(top, sector_num * BDRV_SECTOR_SIZE,
 284                               remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 285         return replication_return_value(s, ret);
 286     }
 287
 288     /*
 289      * Failover failed, only write to active disk if the sectors
 290      * have already been allocated in active disk/hidden disk.
 291      */
 292     qemu_iovec_init(&hd_qiov, qiov->niov);
 293     while (remaining_sectors > 0) {
 294         int64_t count;
 295
 296         ret = bdrv_is_allocated_above(top->bs, base->bs,
 297                                       sector_num * BDRV_SECTOR_SIZE,
 298                                       remaining_sectors * BDRV_SECTOR_SIZE,
 299                                       &count);
 300         if (ret < 0) {
 301             goto out1;
 302         }
 303
 304         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
 305         n = count >> BDRV_SECTOR_BITS;
 306         qemu_iovec_reset(&hd_qiov);
 307         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
 308
 309         target = ret ? top : base;
 310         ret = bdrv_co_pwritev(target, sector_num * BDRV_SECTOR_SIZE,
 311                               n * BDRV_SECTOR_SIZE, &hd_qiov, 0);
 312         if (ret < 0) {
 313             goto out1;
 314         }
 315
 316         remaining_sectors -= n;
 317         sector_num += n;
 318         bytes_done += count;
 319     }
 320
 321 out1:
 322     qemu_iovec_destroy(&hd_qiov);
 323 out:
 324     return ret;
 325 }
 326
 327 static bool replication_recurse_is_first_non_filter(BlockDriverState *bs,
 328                                                     BlockDriverState *candidate)
 329 {
 330     return bdrv_recurse_is_first_non_filter(bs->file->bs, candidate);
 331 }
 332
 333 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
 334 {
 335     Error *local_err = NULL;
 336     int ret;
 337
 338     if (!s->secondary_disk->bs->job) {
 339         error_setg(errp, "Backup job was cancelled unexpectedly");
 340         return;
 341     }
 342
 343     backup_do_checkpoint(s->secondary_disk->bs->job, &local_err);
 344     if (local_err) {
 345         error_propagate(errp, local_err);
 346         return;
 347     }
 348
 349     if (!s->active_disk->bs->drv) {
 350         error_setg(errp, "Active disk %s is ejected",
 351                    s->active_disk->bs->node_name);
 352         return;
 353     }
 354
 355     ret = s->active_disk->bs->drv->bdrv_make_empty(s->active_disk->bs);
 356     if (ret < 0) {
 357         error_setg(errp, "Cannot make active disk empty");
 358         return;
 359     }
 360
 361     if (!s->hidden_disk->bs->drv) {
 362         error_setg(errp, "Hidden disk %s is ejected",
 363                    s->hidden_disk->bs->node_name);
 364         return;
 365     }
 366
 367     ret = s->hidden_disk->bs->drv->bdrv_make_empty(s->hidden_disk->bs);
 368     if (ret < 0) {
 369         error_setg(errp, "Cannot make hidden disk empty");
 370         return;
 371     }
 372 }
 373
 374 static void reopen_backing_file(BlockDriverState *bs, bool writable,
 375                                 Error **errp)
 376 {
 377     BDRVReplicationState *s = bs->opaque;
 378     BlockReopenQueue *reopen_queue = NULL;
 379     int orig_hidden_flags, orig_secondary_flags;
 380     int new_hidden_flags, new_secondary_flags;
 381     Error *local_err = NULL;
 382
 383     if (writable) {
 384         orig_hidden_flags = s->orig_hidden_flags =
 385                                 bdrv_get_flags(s->hidden_disk->bs);
 386         new_hidden_flags = (orig_hidden_flags | BDRV_O_RDWR) &
 387                                                     ~BDRV_O_INACTIVE;
 388         orig_secondary_flags = s->orig_secondary_flags =
 389                                 bdrv_get_flags(s->secondary_disk->bs);
 390         new_secondary_flags = (orig_secondary_flags | BDRV_O_RDWR) &
 391                                                      ~BDRV_O_INACTIVE;
 392     } else {
 393         orig_hidden_flags = (s->orig_hidden_flags | BDRV_O_RDWR) &
 394                                                     ~BDRV_O_INACTIVE;
 395         new_hidden_flags = s->orig_hidden_flags;
 396         orig_secondary_flags = (s->orig_secondary_flags | BDRV_O_RDWR) &
 397                                                     ~BDRV_O_INACTIVE;
 398         new_secondary_flags = s->orig_secondary_flags;
 399     }
 400
 401     bdrv_subtree_drained_begin(s->hidden_disk->bs);
 402     bdrv_subtree_drained_begin(s->secondary_disk->bs);
 403
 404     if (orig_hidden_flags != new_hidden_flags) {
 405         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs, NULL,
 406                                          new_hidden_flags);
 407     }
 408
 409     if (!(orig_secondary_flags & BDRV_O_RDWR)) {
 410         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
 411                                          NULL, new_secondary_flags);
 412     }
 413
 414     if (reopen_queue) {
 415         bdrv_reopen_multiple(bdrv_get_aio_context(bs),
 416                              reopen_queue, &local_err);
 417         error_propagate(errp, local_err);
 418     }
 419
 420     bdrv_subtree_drained_end(s->hidden_disk->bs);
 421     bdrv_subtree_drained_end(s->secondary_disk->bs);
 422 }
 423
 424 static void backup_job_cleanup(BlockDriverState *bs)
 425 {
 426     BDRVReplicationState *s = bs->opaque;
 427     BlockDriverState *top_bs;
 428
 429     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 430     if (!top_bs) {
 431         return;
 432     }
 433     bdrv_op_unblock_all(top_bs, s->blocker);
 434     error_free(s->blocker);
 435     reopen_backing_file(bs, false, NULL);
 436 }
 437
 438 static void backup_job_completed(void *opaque, int ret)
 439 {
 440     BlockDriverState *bs = opaque;
 441     BDRVReplicationState *s = bs->opaque;
 442
 443     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 444         /* The backup job is cancelled unexpectedly */
 445         s->error = -EIO;
 446     }
 447
 448     backup_job_cleanup(bs);
 449 }
 450
 451 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 452 {
 453     BdrvChild *child;
 454
 455     /* The bs itself is the top_bs */
 456     if (top_bs == bs) {
 457         return true;
 458     }
 459
 460     /* Iterate over top_bs's children */
 461     QLIST_FOREACH(child, &top_bs->children, next) {
 462         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 463             return true;
 464         }
 465     }
 466
 467     return false;
 468 }
 469
 470 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 471                               Error **errp)
 472 {
 473     BlockDriverState *bs = rs->opaque;
 474     BDRVReplicationState *s;
 475     BlockDriverState *top_bs;
 476     int64_t active_length, hidden_length, disk_length;
 477     AioContext *aio_context;
 478     Error *local_err = NULL;
 479     BlockJob *job;
 480
 481     aio_context = bdrv_get_aio_context(bs);
 482     aio_context_acquire(aio_context);
 483     s = bs->opaque;
 484
 485     if (s->stage != BLOCK_REPLICATION_NONE) {
 486         error_setg(errp, "Block replication is running or done");
 487         aio_context_release(aio_context);
 488         return;
 489     }
 490
 491     if (s->mode != mode) {
 492         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 493                    " but got %d", s->mode, mode);
 494         aio_context_release(aio_context);
 495         return;
 496     }
 497
 498     switch (s->mode) {
 499     case REPLICATION_MODE_PRIMARY:
 500         break;
 501     case REPLICATION_MODE_SECONDARY:
 502         s->active_disk = bs->file;
 503         if (!s->active_disk || !s->active_disk->bs ||
 504                                     !s->active_disk->bs->backing) {
 505             error_setg(errp, "Active disk doesn't have backing file");
 506             aio_context_release(aio_context);
 507             return;
 508         }
 509
 510         s->hidden_disk = s->active_disk->bs->backing;
 511         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
 512             error_setg(errp, "Hidden disk doesn't have backing file");
 513             aio_context_release(aio_context);
 514             return;
 515         }
 516
 517         s->secondary_disk = s->hidden_disk->bs->backing;
 518         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
 519             error_setg(errp, "The secondary disk doesn't have block backend");
 520             aio_context_release(aio_context);
 521             return;
 522         }
 523
 524         /* verify the length */
 525         active_length = bdrv_getlength(s->active_disk->bs);
 526         hidden_length = bdrv_getlength(s->hidden_disk->bs);
 527         disk_length = bdrv_getlength(s->secondary_disk->bs);
 528         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 529             active_length != hidden_length || hidden_length != disk_length) {
 530             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 531                        " are not the same");
 532             aio_context_release(aio_context);
 533             return;
 534         }
 535
 536         /* Must be true, or the bdrv_getlength() calls would have failed */
 537         assert(s->active_disk->bs->drv && s->hidden_disk->bs->drv);
 538
 539         if (!s->active_disk->bs->drv->bdrv_make_empty ||
 540             !s->hidden_disk->bs->drv->bdrv_make_empty) {
 541             error_setg(errp,
 542                        "Active disk or hidden disk doesn't support make_empty");
 543             aio_context_release(aio_context);
 544             return;
 545         }
 546
 547         /* reopen the backing file in r/w mode */
 548         reopen_backing_file(bs, true, &local_err);
 549         if (local_err) {
 550             error_propagate(errp, local_err);
 551             aio_context_release(aio_context);
 552             return;
 553         }
 554
 555         /* start backup job now */
 556         error_setg(&s->blocker,
 557                    "Block device is in use by internal backup job");
 558
 559         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 560         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 561             !check_top_bs(top_bs, bs)) {
 562             error_setg(errp, "No top_bs or it is invalid");
 563             reopen_backing_file(bs, false, NULL);
 564             aio_context_release(aio_context);
 565             return;
 566         }
 567         bdrv_op_block_all(top_bs, s->blocker);
 568         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 569
 570         job = backup_job_create(NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 571                                 0, MIRROR_SYNC_MODE_NONE, NULL, false,
 572                                 BLOCKDEV_ON_ERROR_REPORT,
 573                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
 574                                 backup_job_completed, bs, NULL, &local_err);
 575         if (local_err) {
 576             error_propagate(errp, local_err);
 577             backup_job_cleanup(bs);
 578             aio_context_release(aio_context);
 579             return;
 580         }
 581         job_start(&job->job);
 582         break;
 583     default:
 584         aio_context_release(aio_context);
 585         abort();
 586     }
 587
 588     s->stage = BLOCK_REPLICATION_RUNNING;
 589
 590     if (s->mode == REPLICATION_MODE_SECONDARY) {
 591         secondary_do_checkpoint(s, errp);
 592     }
 593
 594     s->error = 0;
 595     aio_context_release(aio_context);
 596 }
 597
 598 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
 599 {
 600     BlockDriverState *bs = rs->opaque;
 601     BDRVReplicationState *s;
 602     AioContext *aio_context;
 603
 604     aio_context = bdrv_get_aio_context(bs);
 605     aio_context_acquire(aio_context);
 606     s = bs->opaque;
 607
 608     if (s->mode == REPLICATION_MODE_SECONDARY) {
 609         secondary_do_checkpoint(s, errp);
 610     }
 611     aio_context_release(aio_context);
 612 }
 613
 614 static void replication_get_error(ReplicationState *rs, Error **errp)
 615 {
 616     BlockDriverState *bs = rs->opaque;
 617     BDRVReplicationState *s;
 618     AioContext *aio_context;
 619
 620     aio_context = bdrv_get_aio_context(bs);
 621     aio_context_acquire(aio_context);
 622     s = bs->opaque;
 623
 624     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 625         error_setg(errp, "Block replication is not running");
 626         aio_context_release(aio_context);
 627         return;
 628     }
 629
 630     if (s->error) {
 631         error_setg(errp, "I/O error occurred");
 632         aio_context_release(aio_context);
 633         return;
 634     }
 635     aio_context_release(aio_context);
 636 }
 637
 638 static void replication_done(void *opaque, int ret)
 639 {
 640     BlockDriverState *bs = opaque;
 641     BDRVReplicationState *s = bs->opaque;
 642
 643     if (ret == 0) {
 644         s->stage = BLOCK_REPLICATION_DONE;
 645
 646         /* refresh top bs's filename */
 647         bdrv_refresh_filename(bs);
 648         s->active_disk = NULL;
 649         s->secondary_disk = NULL;
 650         s->hidden_disk = NULL;
 651         s->error = 0;
 652     } else {
 653         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 654         s->error = -EIO;
 655     }
 656 }
 657
 658 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 659 {
 660     BlockDriverState *bs = rs->opaque;
 661     BDRVReplicationState *s;
 662     AioContext *aio_context;
 663
 664     aio_context = bdrv_get_aio_context(bs);
 665     aio_context_acquire(aio_context);
 666     s = bs->opaque;
 667
 668     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 669         error_setg(errp, "Block replication is not running");
 670         aio_context_release(aio_context);
 671         return;
 672     }
 673
 674     switch (s->mode) {
 675     case REPLICATION_MODE_PRIMARY:
 676         s->stage = BLOCK_REPLICATION_DONE;
 677         s->error = 0;
 678         break;
 679     case REPLICATION_MODE_SECONDARY:
 680         /*
 681          * This BDS will be closed, and the job should be completed
 682          * before the BDS is closed, because we will access hidden
 683          * disk, secondary disk in backup_job_completed().
 684          */
 685         if (s->secondary_disk->bs->job) {
 686             job_cancel_sync(&s->secondary_disk->bs->job->job);
 687         }
 688
 689         if (!failover) {
 690             secondary_do_checkpoint(s, errp);
 691             s->stage = BLOCK_REPLICATION_DONE;
 692             aio_context_release(aio_context);
 693             return;
 694         }
 695
 696         s->stage = BLOCK_REPLICATION_FAILOVER;
 697         commit_active_start(NULL, s->active_disk->bs, s->secondary_disk->bs,
 698                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 699                             NULL, replication_done, bs, true, errp);
 700         break;
 701     default:
 702         aio_context_release(aio_context);
 703         abort();
 704     }
 705     aio_context_release(aio_context);
 706 }
 707
 708 BlockDriver bdrv_replication = {
 709     .format_name                = "replication",
 710     .instance_size              = sizeof(BDRVReplicationState),
 711
 712     .bdrv_open                  = replication_open,
 713     .bdrv_close                 = replication_close,
 714     .bdrv_child_perm            = replication_child_perm,
 715
 716     .bdrv_getlength             = replication_getlength,
 717     .bdrv_co_readv              = replication_co_readv,
 718     .bdrv_co_writev             = replication_co_writev,
 719
 720     .is_filter                  = true,
 721     .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter,
 722
 723     .has_variable_length        = true,
 724 };
 725
 726 static void bdrv_replication_init(void)
 727 {
 728     bdrv_register(&bdrv_replication);
 729 }
 730
 731 block_init(bdrv_replication_init);