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 "qapi/qmp/qdict.h"
  24 #include "replication.h"
  25
  26 typedef enum {
  27     BLOCK_REPLICATION_NONE,             /* block replication is not started */
  28     BLOCK_REPLICATION_RUNNING,          /* block replication is running */
  29     BLOCK_REPLICATION_FAILOVER,         /* failover is running in background */
  30     BLOCK_REPLICATION_FAILOVER_FAILED,  /* failover failed */
  31     BLOCK_REPLICATION_DONE,             /* block replication is done */
  32 } ReplicationStage;
  33
  34 typedef struct BDRVReplicationState {
  35     ReplicationMode mode;
  36     ReplicationStage stage;
  37     BdrvChild *active_disk;
  38     BdrvChild *hidden_disk;
  39     BdrvChild *secondary_disk;
  40     char *top_id;
  41     ReplicationState *rs;
  42     Error *blocker;
  43     bool orig_hidden_read_only;
  44     bool orig_secondary_read_only;
  45     int error;
  46 } BDRVReplicationState;
  47
  48 static void replication_start(ReplicationState *rs, ReplicationMode mode,
  49                               Error **errp);
  50 static void replication_do_checkpoint(ReplicationState *rs, Error **errp);
  51 static void replication_get_error(ReplicationState *rs, Error **errp);
  52 static void replication_stop(ReplicationState *rs, bool failover,
  53                              Error **errp);
  54
  55 #define REPLICATION_MODE        "mode"
  56 #define REPLICATION_TOP_ID      "top-id"
  57 static QemuOptsList replication_runtime_opts = {
  58     .name = "replication",
  59     .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head),
  60     .desc = {
  61         {
  62             .name = REPLICATION_MODE,
  63             .type = QEMU_OPT_STRING,
  64         },
  65         {
  66             .name = REPLICATION_TOP_ID,
  67             .type = QEMU_OPT_STRING,
  68         },
  69         { /* end of list */ }
  70     },
  71 };
  72
  73 static ReplicationOps replication_ops = {
  74     .start = replication_start,
  75     .checkpoint = replication_do_checkpoint,
  76     .get_error = replication_get_error,
  77     .stop = replication_stop,
  78 };
  79
  80 static int replication_open(BlockDriverState *bs, QDict *options,
  81                             int flags, Error **errp)
  82 {
  83     int ret;
  84     BDRVReplicationState *s = bs->opaque;
  85     Error *local_err = NULL;
  86     QemuOpts *opts = NULL;
  87     const char *mode;
  88     const char *top_id;
  89
  90     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
  91                                false, errp);
  92     if (!bs->file) {
  93         return -EINVAL;
  94     }
  95
  96     ret = -EINVAL;
  97     opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
  98     qemu_opts_absorb_qdict(opts, options, &local_err);
  99     if (local_err) {
 100         goto fail;
 101     }
 102
 103     mode = qemu_opt_get(opts, REPLICATION_MODE);
 104     if (!mode) {
 105         error_setg(&local_err, "Missing the option mode");
 106         goto fail;
 107     }
 108
 109     if (!strcmp(mode, "primary")) {
 110         s->mode = REPLICATION_MODE_PRIMARY;
 111         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 112         if (top_id) {
 113             error_setg(&local_err, "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(&local_err, "Missing the option top-id");
 122             goto fail;
 123         }
 124     } else {
 125         error_setg(&local_err,
 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     error_propagate(errp, local_err);
 137
 138     return ret;
 139 }
 140
 141 static void replication_close(BlockDriverState *bs)
 142 {
 143     BDRVReplicationState *s = bs->opaque;
 144
 145     if (s->stage == BLOCK_REPLICATION_RUNNING) {
 146         replication_stop(s->rs, false, NULL);
 147     }
 148     if (s->stage == BLOCK_REPLICATION_FAILOVER) {
 149         job_cancel_sync(&s->active_disk->bs->job->job);
 150     }
 151
 152     if (s->mode == REPLICATION_MODE_SECONDARY) {
 153         g_free(s->top_id);
 154     }
 155
 156     replication_remove(s->rs);
 157 }
 158
 159 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
 160                                    const BdrvChildRole *role,
 161                                    BlockReopenQueue *reopen_queue,
 162                                    uint64_t perm, uint64_t shared,
 163                                    uint64_t *nperm, uint64_t *nshared)
 164 {
 165     *nperm = BLK_PERM_CONSISTENT_READ;
 166     if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) {
 167         *nperm |= BLK_PERM_WRITE;
 168     }
 169     *nshared = BLK_PERM_CONSISTENT_READ \
 170                | BLK_PERM_WRITE \
 171                | BLK_PERM_WRITE_UNCHANGED;
 172     return;
 173 }
 174
 175 static int64_t replication_getlength(BlockDriverState *bs)
 176 {
 177     return bdrv_getlength(bs->file->bs);
 178 }
 179
 180 static int replication_get_io_status(BDRVReplicationState *s)
 181 {
 182     switch (s->stage) {
 183     case BLOCK_REPLICATION_NONE:
 184         return -EIO;
 185     case BLOCK_REPLICATION_RUNNING:
 186         return 0;
 187     case BLOCK_REPLICATION_FAILOVER:
 188         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 189     case BLOCK_REPLICATION_FAILOVER_FAILED:
 190         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
 191     case BLOCK_REPLICATION_DONE:
 192         /*
 193          * active commit job completes, and active disk and secondary_disk
 194          * is swapped, so we can operate bs->file directly
 195          */
 196         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 197     default:
 198         abort();
 199     }
 200 }
 201
 202 static int replication_return_value(BDRVReplicationState *s, int ret)
 203 {
 204     if (s->mode == REPLICATION_MODE_SECONDARY) {
 205         return ret;
 206     }
 207
 208     if (ret < 0) {
 209         s->error = ret;
 210         ret = 0;
 211     }
 212
 213     return ret;
 214 }
 215
 216 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
 217                                              int64_t sector_num,
 218                                              int remaining_sectors,
 219                                              QEMUIOVector *qiov)
 220 {
 221     BDRVReplicationState *s = bs->opaque;
 222     int ret;
 223
 224     if (s->mode == REPLICATION_MODE_PRIMARY) {
 225         /* We only use it to forward primary write requests */
 226         return -EIO;
 227     }
 228
 229     ret = replication_get_io_status(s);
 230     if (ret < 0) {
 231         return ret;
 232     }
 233
 234     ret = bdrv_co_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
 235                          remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 236
 237     return replication_return_value(s, ret);
 238 }
 239
 240 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
 241                                               int64_t sector_num,
 242                                               int remaining_sectors,
 243                                               QEMUIOVector *qiov,
 244                                               int flags)
 245 {
 246     BDRVReplicationState *s = bs->opaque;
 247     QEMUIOVector hd_qiov;
 248     uint64_t bytes_done = 0;
 249     BdrvChild *top = bs->file;
 250     BdrvChild *base = s->secondary_disk;
 251     BdrvChild *target;
 252     int ret;
 253     int64_t n;
 254
 255     assert(!flags);
 256     ret = replication_get_io_status(s);
 257     if (ret < 0) {
 258         goto out;
 259     }
 260
 261     if (ret == 0) {
 262         ret = bdrv_co_pwritev(top, sector_num * BDRV_SECTOR_SIZE,
 263                               remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 264         return replication_return_value(s, ret);
 265     }
 266
 267     /*
 268      * Failover failed, only write to active disk if the sectors
 269      * have already been allocated in active disk/hidden disk.
 270      */
 271     qemu_iovec_init(&hd_qiov, qiov->niov);
 272     while (remaining_sectors > 0) {
 273         int64_t count;
 274
 275         ret = bdrv_is_allocated_above(top->bs, base->bs,
 276                                       sector_num * BDRV_SECTOR_SIZE,
 277                                       remaining_sectors * BDRV_SECTOR_SIZE,
 278                                       &count);
 279         if (ret < 0) {
 280             goto out1;
 281         }
 282
 283         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
 284         n = count >> BDRV_SECTOR_BITS;
 285         qemu_iovec_reset(&hd_qiov);
 286         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
 287
 288         target = ret ? top : base;
 289         ret = bdrv_co_pwritev(target, sector_num * BDRV_SECTOR_SIZE,
 290                               n * BDRV_SECTOR_SIZE, &hd_qiov, 0);
 291         if (ret < 0) {
 292             goto out1;
 293         }
 294
 295         remaining_sectors -= n;
 296         sector_num += n;
 297         bytes_done += count;
 298     }
 299
 300 out1:
 301     qemu_iovec_destroy(&hd_qiov);
 302 out:
 303     return ret;
 304 }
 305
 306 static bool replication_recurse_is_first_non_filter(BlockDriverState *bs,
 307                                                     BlockDriverState *candidate)
 308 {
 309     return bdrv_recurse_is_first_non_filter(bs->file->bs, candidate);
 310 }
 311
 312 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
 313 {
 314     Error *local_err = NULL;
 315     int ret;
 316
 317     if (!s->secondary_disk->bs->job) {
 318         error_setg(errp, "Backup job was cancelled unexpectedly");
 319         return;
 320     }
 321
 322     backup_do_checkpoint(s->secondary_disk->bs->job, &local_err);
 323     if (local_err) {
 324         error_propagate(errp, local_err);
 325         return;
 326     }
 327
 328     if (!s->active_disk->bs->drv) {
 329         error_setg(errp, "Active disk %s is ejected",
 330                    s->active_disk->bs->node_name);
 331         return;
 332     }
 333
 334     ret = s->active_disk->bs->drv->bdrv_make_empty(s->active_disk->bs);
 335     if (ret < 0) {
 336         error_setg(errp, "Cannot make active disk empty");
 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     ret = s->hidden_disk->bs->drv->bdrv_make_empty(s->hidden_disk->bs);
 347     if (ret < 0) {
 348         error_setg(errp, "Cannot make hidden disk empty");
 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     BlockReopenQueue *reopen_queue = NULL;
 363     Error *local_err = NULL;
 364
 365     if (writable) {
 366         s->orig_hidden_read_only = bdrv_is_read_only(s->hidden_disk->bs);
 367         s->orig_secondary_read_only = bdrv_is_read_only(s->secondary_disk->bs);
 368     }
 369
 370     bdrv_subtree_drained_begin(s->hidden_disk->bs);
 371     bdrv_subtree_drained_begin(s->secondary_disk->bs);
 372
 373     if (s->orig_hidden_read_only) {
 374         int flags = bdrv_get_flags(s->hidden_disk->bs);
 375         QDict *opts = qdict_new();
 376         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 377         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs,
 378                                          opts, flags);
 379     }
 380
 381     if (s->orig_secondary_read_only) {
 382         int flags = bdrv_get_flags(s->secondary_disk->bs);
 383         QDict *opts = qdict_new();
 384         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 385         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
 386                                          opts, flags);
 387     }
 388
 389     if (reopen_queue) {
 390         bdrv_reopen_multiple(bdrv_get_aio_context(bs),
 391                              reopen_queue, &local_err);
 392         error_propagate(errp, local_err);
 393     }
 394
 395     bdrv_subtree_drained_end(s->hidden_disk->bs);
 396     bdrv_subtree_drained_end(s->secondary_disk->bs);
 397 }
 398
 399 static void backup_job_cleanup(BlockDriverState *bs)
 400 {
 401     BDRVReplicationState *s = bs->opaque;
 402     BlockDriverState *top_bs;
 403
 404     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 405     if (!top_bs) {
 406         return;
 407     }
 408     bdrv_op_unblock_all(top_bs, s->blocker);
 409     error_free(s->blocker);
 410     reopen_backing_file(bs, false, NULL);
 411 }
 412
 413 static void backup_job_completed(void *opaque, int ret)
 414 {
 415     BlockDriverState *bs = opaque;
 416     BDRVReplicationState *s = bs->opaque;
 417
 418     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 419         /* The backup job is cancelled unexpectedly */
 420         s->error = -EIO;
 421     }
 422
 423     backup_job_cleanup(bs);
 424 }
 425
 426 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 427 {
 428     BdrvChild *child;
 429
 430     /* The bs itself is the top_bs */
 431     if (top_bs == bs) {
 432         return true;
 433     }
 434
 435     /* Iterate over top_bs's children */
 436     QLIST_FOREACH(child, &top_bs->children, next) {
 437         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 438             return true;
 439         }
 440     }
 441
 442     return false;
 443 }
 444
 445 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 446                               Error **errp)
 447 {
 448     BlockDriverState *bs = rs->opaque;
 449     BDRVReplicationState *s;
 450     BlockDriverState *top_bs;
 451     int64_t active_length, hidden_length, disk_length;
 452     AioContext *aio_context;
 453     Error *local_err = NULL;
 454     BlockJob *job;
 455
 456     aio_context = bdrv_get_aio_context(bs);
 457     aio_context_acquire(aio_context);
 458     s = bs->opaque;
 459
 460     if (s->stage != BLOCK_REPLICATION_NONE) {
 461         error_setg(errp, "Block replication is running or done");
 462         aio_context_release(aio_context);
 463         return;
 464     }
 465
 466     if (s->mode != mode) {
 467         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 468                    " but got %d", s->mode, mode);
 469         aio_context_release(aio_context);
 470         return;
 471     }
 472
 473     switch (s->mode) {
 474     case REPLICATION_MODE_PRIMARY:
 475         break;
 476     case REPLICATION_MODE_SECONDARY:
 477         s->active_disk = bs->file;
 478         if (!s->active_disk || !s->active_disk->bs ||
 479                                     !s->active_disk->bs->backing) {
 480             error_setg(errp, "Active disk doesn't have backing file");
 481             aio_context_release(aio_context);
 482             return;
 483         }
 484
 485         s->hidden_disk = s->active_disk->bs->backing;
 486         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
 487             error_setg(errp, "Hidden disk doesn't have backing file");
 488             aio_context_release(aio_context);
 489             return;
 490         }
 491
 492         s->secondary_disk = s->hidden_disk->bs->backing;
 493         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
 494             error_setg(errp, "The secondary disk doesn't have block backend");
 495             aio_context_release(aio_context);
 496             return;
 497         }
 498
 499         /* verify the length */
 500         active_length = bdrv_getlength(s->active_disk->bs);
 501         hidden_length = bdrv_getlength(s->hidden_disk->bs);
 502         disk_length = bdrv_getlength(s->secondary_disk->bs);
 503         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 504             active_length != hidden_length || hidden_length != disk_length) {
 505             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 506                        " are not the same");
 507             aio_context_release(aio_context);
 508             return;
 509         }
 510
 511         /* Must be true, or the bdrv_getlength() calls would have failed */
 512         assert(s->active_disk->bs->drv && s->hidden_disk->bs->drv);
 513
 514         if (!s->active_disk->bs->drv->bdrv_make_empty ||
 515             !s->hidden_disk->bs->drv->bdrv_make_empty) {
 516             error_setg(errp,
 517                        "Active disk or hidden disk doesn't support make_empty");
 518             aio_context_release(aio_context);
 519             return;
 520         }
 521
 522         /* reopen the backing file in r/w mode */
 523         reopen_backing_file(bs, true, &local_err);
 524         if (local_err) {
 525             error_propagate(errp, local_err);
 526             aio_context_release(aio_context);
 527             return;
 528         }
 529
 530         /* start backup job now */
 531         error_setg(&s->blocker,
 532                    "Block device is in use by internal backup job");
 533
 534         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 535         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 536             !check_top_bs(top_bs, bs)) {
 537             error_setg(errp, "No top_bs or it is invalid");
 538             reopen_backing_file(bs, false, NULL);
 539             aio_context_release(aio_context);
 540             return;
 541         }
 542         bdrv_op_block_all(top_bs, s->blocker);
 543         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 544
 545         job = backup_job_create(NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 546                                 0, MIRROR_SYNC_MODE_NONE, NULL, false,
 547                                 BLOCKDEV_ON_ERROR_REPORT,
 548                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
 549                                 backup_job_completed, bs, NULL, &local_err);
 550         if (local_err) {
 551             error_propagate(errp, local_err);
 552             backup_job_cleanup(bs);
 553             aio_context_release(aio_context);
 554             return;
 555         }
 556         job_start(&job->job);
 557         break;
 558     default:
 559         aio_context_release(aio_context);
 560         abort();
 561     }
 562
 563     s->stage = BLOCK_REPLICATION_RUNNING;
 564
 565     if (s->mode == REPLICATION_MODE_SECONDARY) {
 566         secondary_do_checkpoint(s, errp);
 567     }
 568
 569     s->error = 0;
 570     aio_context_release(aio_context);
 571 }
 572
 573 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
 574 {
 575     BlockDriverState *bs = rs->opaque;
 576     BDRVReplicationState *s;
 577     AioContext *aio_context;
 578
 579     aio_context = bdrv_get_aio_context(bs);
 580     aio_context_acquire(aio_context);
 581     s = bs->opaque;
 582
 583     if (s->mode == REPLICATION_MODE_SECONDARY) {
 584         secondary_do_checkpoint(s, errp);
 585     }
 586     aio_context_release(aio_context);
 587 }
 588
 589 static void replication_get_error(ReplicationState *rs, Error **errp)
 590 {
 591     BlockDriverState *bs = rs->opaque;
 592     BDRVReplicationState *s;
 593     AioContext *aio_context;
 594
 595     aio_context = bdrv_get_aio_context(bs);
 596     aio_context_acquire(aio_context);
 597     s = bs->opaque;
 598
 599     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 600         error_setg(errp, "Block replication is not running");
 601         aio_context_release(aio_context);
 602         return;
 603     }
 604
 605     if (s->error) {
 606         error_setg(errp, "I/O error occurred");
 607         aio_context_release(aio_context);
 608         return;
 609     }
 610     aio_context_release(aio_context);
 611 }
 612
 613 static void replication_done(void *opaque, int ret)
 614 {
 615     BlockDriverState *bs = opaque;
 616     BDRVReplicationState *s = bs->opaque;
 617
 618     if (ret == 0) {
 619         s->stage = BLOCK_REPLICATION_DONE;
 620
 621         /* refresh top bs's filename */
 622         bdrv_refresh_filename(bs);
 623         s->active_disk = NULL;
 624         s->secondary_disk = NULL;
 625         s->hidden_disk = NULL;
 626         s->error = 0;
 627     } else {
 628         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 629         s->error = -EIO;
 630     }
 631 }
 632
 633 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 634 {
 635     BlockDriverState *bs = rs->opaque;
 636     BDRVReplicationState *s;
 637     AioContext *aio_context;
 638
 639     aio_context = bdrv_get_aio_context(bs);
 640     aio_context_acquire(aio_context);
 641     s = bs->opaque;
 642
 643     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 644         error_setg(errp, "Block replication is not running");
 645         aio_context_release(aio_context);
 646         return;
 647     }
 648
 649     switch (s->mode) {
 650     case REPLICATION_MODE_PRIMARY:
 651         s->stage = BLOCK_REPLICATION_DONE;
 652         s->error = 0;
 653         break;
 654     case REPLICATION_MODE_SECONDARY:
 655         /*
 656          * This BDS will be closed, and the job should be completed
 657          * before the BDS is closed, because we will access hidden
 658          * disk, secondary disk in backup_job_completed().
 659          */
 660         if (s->secondary_disk->bs->job) {
 661             job_cancel_sync(&s->secondary_disk->bs->job->job);
 662         }
 663
 664         if (!failover) {
 665             secondary_do_checkpoint(s, errp);
 666             s->stage = BLOCK_REPLICATION_DONE;
 667             aio_context_release(aio_context);
 668             return;
 669         }
 670
 671         s->stage = BLOCK_REPLICATION_FAILOVER;
 672         commit_active_start(NULL, s->active_disk->bs, s->secondary_disk->bs,
 673                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 674                             NULL, replication_done, bs, true, errp);
 675         break;
 676     default:
 677         aio_context_release(aio_context);
 678         abort();
 679     }
 680     aio_context_release(aio_context);
 681 }
 682
 683 BlockDriver bdrv_replication = {
 684     .format_name                = "replication",
 685     .instance_size              = sizeof(BDRVReplicationState),
 686
 687     .bdrv_open                  = replication_open,
 688     .bdrv_close                 = replication_close,
 689     .bdrv_child_perm            = replication_child_perm,
 690
 691     .bdrv_getlength             = replication_getlength,
 692     .bdrv_co_readv              = replication_co_readv,
 693     .bdrv_co_writev             = replication_co_writev,
 694
 695     .is_filter                  = true,
 696     .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter,
 697
 698     .has_variable_length        = true,
 699 };
 700
 701 static void bdrv_replication_init(void)
 702 {
 703     bdrv_register(&bdrv_replication);
 704 }
 705
 706 block_init(bdrv_replication_init);