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 <[email protected]>
  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_file,
  94                                false, errp);
  95     if (!bs->file) {
  96         return -EINVAL;
  97     }
  98
  99     ret = -EINVAL;
 100     opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
 101     qemu_opts_absorb_qdict(opts, options, &local_err);
 102     if (local_err) {
 103         goto fail;
 104     }
 105
 106     mode = qemu_opt_get(opts, REPLICATION_MODE);
 107     if (!mode) {
 108         error_setg(&local_err, "Missing the option mode");
 109         goto fail;
 110     }
 111
 112     if (!strcmp(mode, "primary")) {
 113         s->mode = REPLICATION_MODE_PRIMARY;
 114         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 115         if (top_id) {
 116             error_setg(&local_err, "The primary side does not support option top-id");
 117             goto fail;
 118         }
 119     } else if (!strcmp(mode, "secondary")) {
 120         s->mode = REPLICATION_MODE_SECONDARY;
 121         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 122         s->top_id = g_strdup(top_id);
 123         if (!s->top_id) {
 124             error_setg(&local_err, "Missing the option top-id");
 125             goto fail;
 126         }
 127     } else {
 128         error_setg(&local_err,
 129                    "The option mode's value should be primary or secondary");
 130         goto fail;
 131     }
 132
 133     s->rs = replication_new(bs, &replication_ops);
 134
 135     ret = 0;
 136
 137 fail:
 138     qemu_opts_del(opts);
 139     error_propagate(errp, local_err);
 140
 141     return ret;
 142 }
 143
 144 static void replication_close(BlockDriverState *bs)
 145 {
 146     BDRVReplicationState *s = bs->opaque;
 147
 148     if (s->stage == BLOCK_REPLICATION_RUNNING) {
 149         replication_stop(s->rs, false, NULL);
 150     }
 151     if (s->stage == BLOCK_REPLICATION_FAILOVER) {
 152         job_cancel_sync(&s->commit_job->job);
 153     }
 154
 155     if (s->mode == REPLICATION_MODE_SECONDARY) {
 156         g_free(s->top_id);
 157     }
 158
 159     replication_remove(s->rs);
 160 }
 161
 162 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
 163                                    const BdrvChildRole *role,
 164                                    BlockReopenQueue *reopen_queue,
 165                                    uint64_t perm, uint64_t shared,
 166                                    uint64_t *nperm, uint64_t *nshared)
 167 {
 168     *nperm = BLK_PERM_CONSISTENT_READ;
 169     if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) {
 170         *nperm |= BLK_PERM_WRITE;
 171     }
 172     *nshared = BLK_PERM_CONSISTENT_READ \
 173                | BLK_PERM_WRITE \
 174                | BLK_PERM_WRITE_UNCHANGED;
 175     return;
 176 }
 177
 178 static int64_t replication_getlength(BlockDriverState *bs)
 179 {
 180     return bdrv_getlength(bs->file->bs);
 181 }
 182
 183 static int replication_get_io_status(BDRVReplicationState *s)
 184 {
 185     switch (s->stage) {
 186     case BLOCK_REPLICATION_NONE:
 187         return -EIO;
 188     case BLOCK_REPLICATION_RUNNING:
 189         return 0;
 190     case BLOCK_REPLICATION_FAILOVER:
 191         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 192     case BLOCK_REPLICATION_FAILOVER_FAILED:
 193         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
 194     case BLOCK_REPLICATION_DONE:
 195         /*
 196          * active commit job completes, and active disk and secondary_disk
 197          * is swapped, so we can operate bs->file directly
 198          */
 199         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 200     default:
 201         abort();
 202     }
 203 }
 204
 205 static int replication_return_value(BDRVReplicationState *s, int ret)
 206 {
 207     if (s->mode == REPLICATION_MODE_SECONDARY) {
 208         return ret;
 209     }
 210
 211     if (ret < 0) {
 212         s->error = ret;
 213         ret = 0;
 214     }
 215
 216     return ret;
 217 }
 218
 219 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
 220                                              int64_t sector_num,
 221                                              int remaining_sectors,
 222                                              QEMUIOVector *qiov)
 223 {
 224     BDRVReplicationState *s = bs->opaque;
 225     int ret;
 226
 227     if (s->mode == REPLICATION_MODE_PRIMARY) {
 228         /* We only use it to forward primary write requests */
 229         return -EIO;
 230     }
 231
 232     ret = replication_get_io_status(s);
 233     if (ret < 0) {
 234         return ret;
 235     }
 236
 237     ret = bdrv_co_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
 238                          remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 239
 240     return replication_return_value(s, ret);
 241 }
 242
 243 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
 244                                               int64_t sector_num,
 245                                               int remaining_sectors,
 246                                               QEMUIOVector *qiov,
 247                                               int flags)
 248 {
 249     BDRVReplicationState *s = bs->opaque;
 250     QEMUIOVector hd_qiov;
 251     uint64_t bytes_done = 0;
 252     BdrvChild *top = bs->file;
 253     BdrvChild *base = s->secondary_disk;
 254     BdrvChild *target;
 255     int ret;
 256     int64_t n;
 257
 258     assert(!flags);
 259     ret = replication_get_io_status(s);
 260     if (ret < 0) {
 261         goto out;
 262     }
 263
 264     if (ret == 0) {
 265         ret = bdrv_co_pwritev(top, sector_num * BDRV_SECTOR_SIZE,
 266                               remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 267         return replication_return_value(s, ret);
 268     }
 269
 270     /*
 271      * Failover failed, only write to active disk if the sectors
 272      * have already been allocated in active disk/hidden disk.
 273      */
 274     qemu_iovec_init(&hd_qiov, qiov->niov);
 275     while (remaining_sectors > 0) {
 276         int64_t count;
 277
 278         ret = bdrv_is_allocated_above(top->bs, base->bs, false,
 279                                       sector_num * BDRV_SECTOR_SIZE,
 280                                       remaining_sectors * BDRV_SECTOR_SIZE,
 281                                       &count);
 282         if (ret < 0) {
 283             goto out1;
 284         }
 285
 286         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
 287         n = count >> BDRV_SECTOR_BITS;
 288         qemu_iovec_reset(&hd_qiov);
 289         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
 290
 291         target = ret ? top : base;
 292         ret = bdrv_co_pwritev(target, sector_num * BDRV_SECTOR_SIZE,
 293                               n * BDRV_SECTOR_SIZE, &hd_qiov, 0);
 294         if (ret < 0) {
 295             goto out1;
 296         }
 297
 298         remaining_sectors -= n;
 299         sector_num += n;
 300         bytes_done += count;
 301     }
 302
 303 out1:
 304     qemu_iovec_destroy(&hd_qiov);
 305 out:
 306     return ret;
 307 }
 308
 309 static bool replication_recurse_is_first_non_filter(BlockDriverState *bs,
 310                                                     BlockDriverState *candidate)
 311 {
 312     return bdrv_recurse_is_first_non_filter(bs->file->bs, candidate);
 313 }
 314
 315 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
 316 {
 317     Error *local_err = NULL;
 318     int ret;
 319
 320     if (!s->backup_job) {
 321         error_setg(errp, "Backup job was cancelled unexpectedly");
 322         return;
 323     }
 324
 325     backup_do_checkpoint(s->backup_job, &local_err);
 326     if (local_err) {
 327         error_propagate(errp, local_err);
 328         return;
 329     }
 330
 331     if (!s->active_disk->bs->drv) {
 332         error_setg(errp, "Active disk %s is ejected",
 333                    s->active_disk->bs->node_name);
 334         return;
 335     }
 336
 337     ret = s->active_disk->bs->drv->bdrv_make_empty(s->active_disk->bs);
 338     if (ret < 0) {
 339         error_setg(errp, "Cannot make active disk empty");
 340         return;
 341     }
 342
 343     if (!s->hidden_disk->bs->drv) {
 344         error_setg(errp, "Hidden disk %s is ejected",
 345                    s->hidden_disk->bs->node_name);
 346         return;
 347     }
 348
 349     ret = s->hidden_disk->bs->drv->bdrv_make_empty(s->hidden_disk->bs);
 350     if (ret < 0) {
 351         error_setg(errp, "Cannot make hidden disk empty");
 352         return;
 353     }
 354 }
 355
 356 /* This function is supposed to be called twice:
 357  * first with writable = true, then with writable = false.
 358  * The first call puts s->hidden_disk and s->secondary_disk in
 359  * r/w mode, and the second puts them back in their original state.
 360  */
 361 static void reopen_backing_file(BlockDriverState *bs, bool writable,
 362                                 Error **errp)
 363 {
 364     BDRVReplicationState *s = bs->opaque;
 365     BlockReopenQueue *reopen_queue = NULL;
 366     Error *local_err = NULL;
 367
 368     if (writable) {
 369         s->orig_hidden_read_only = bdrv_is_read_only(s->hidden_disk->bs);
 370         s->orig_secondary_read_only = bdrv_is_read_only(s->secondary_disk->bs);
 371     }
 372
 373     bdrv_subtree_drained_begin(s->hidden_disk->bs);
 374     bdrv_subtree_drained_begin(s->secondary_disk->bs);
 375
 376     if (s->orig_hidden_read_only) {
 377         QDict *opts = qdict_new();
 378         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 379         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs,
 380                                          opts, true);
 381     }
 382
 383     if (s->orig_secondary_read_only) {
 384         QDict *opts = qdict_new();
 385         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 386         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
 387                                          opts, true);
 388     }
 389
 390     if (reopen_queue) {
 391         bdrv_reopen_multiple(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
 455     aio_context = bdrv_get_aio_context(bs);
 456     aio_context_acquire(aio_context);
 457     s = bs->opaque;
 458
 459     if (s->stage != BLOCK_REPLICATION_NONE) {
 460         error_setg(errp, "Block replication is running or done");
 461         aio_context_release(aio_context);
 462         return;
 463     }
 464
 465     if (s->mode != mode) {
 466         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 467                    " but got %d", s->mode, mode);
 468         aio_context_release(aio_context);
 469         return;
 470     }
 471
 472     switch (s->mode) {
 473     case REPLICATION_MODE_PRIMARY:
 474         break;
 475     case REPLICATION_MODE_SECONDARY:
 476         s->active_disk = bs->file;
 477         if (!s->active_disk || !s->active_disk->bs ||
 478                                     !s->active_disk->bs->backing) {
 479             error_setg(errp, "Active disk doesn't have backing file");
 480             aio_context_release(aio_context);
 481             return;
 482         }
 483
 484         s->hidden_disk = s->active_disk->bs->backing;
 485         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
 486             error_setg(errp, "Hidden disk doesn't have backing file");
 487             aio_context_release(aio_context);
 488             return;
 489         }
 490
 491         s->secondary_disk = s->hidden_disk->bs->backing;
 492         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
 493             error_setg(errp, "The secondary disk doesn't have block backend");
 494             aio_context_release(aio_context);
 495             return;
 496         }
 497
 498         /* verify the length */
 499         active_length = bdrv_getlength(s->active_disk->bs);
 500         hidden_length = bdrv_getlength(s->hidden_disk->bs);
 501         disk_length = bdrv_getlength(s->secondary_disk->bs);
 502         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 503             active_length != hidden_length || hidden_length != disk_length) {
 504             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 505                        " are not the same");
 506             aio_context_release(aio_context);
 507             return;
 508         }
 509
 510         /* Must be true, or the bdrv_getlength() calls would have failed */
 511         assert(s->active_disk->bs->drv && s->hidden_disk->bs->drv);
 512
 513         if (!s->active_disk->bs->drv->bdrv_make_empty ||
 514             !s->hidden_disk->bs->drv->bdrv_make_empty) {
 515             error_setg(errp,
 516                        "Active disk or hidden disk doesn't support make_empty");
 517             aio_context_release(aio_context);
 518             return;
 519         }
 520
 521         /* reopen the backing file in r/w mode */
 522         reopen_backing_file(bs, true, &local_err);
 523         if (local_err) {
 524             error_propagate(errp, local_err);
 525             aio_context_release(aio_context);
 526             return;
 527         }
 528
 529         /* start backup job now */
 530         error_setg(&s->blocker,
 531                    "Block device is in use by internal backup job");
 532
 533         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 534         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 535             !check_top_bs(top_bs, bs)) {
 536             error_setg(errp, "No top_bs or it is invalid");
 537             reopen_backing_file(bs, false, NULL);
 538             aio_context_release(aio_context);
 539             return;
 540         }
 541         bdrv_op_block_all(top_bs, s->blocker);
 542         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 543
 544         s->backup_job = backup_job_create(
 545                                 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 546                                 0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
 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(&s->backup_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         s->active_disk = NULL;
 622         s->secondary_disk = NULL;
 623         s->hidden_disk = NULL;
 624         s->error = 0;
 625     } else {
 626         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 627         s->error = -EIO;
 628     }
 629 }
 630
 631 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 632 {
 633     BlockDriverState *bs = rs->opaque;
 634     BDRVReplicationState *s;
 635     AioContext *aio_context;
 636
 637     aio_context = bdrv_get_aio_context(bs);
 638     aio_context_acquire(aio_context);
 639     s = bs->opaque;
 640
 641     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 642         error_setg(errp, "Block replication is not running");
 643         aio_context_release(aio_context);
 644         return;
 645     }
 646
 647     switch (s->mode) {
 648     case REPLICATION_MODE_PRIMARY:
 649         s->stage = BLOCK_REPLICATION_DONE;
 650         s->error = 0;
 651         break;
 652     case REPLICATION_MODE_SECONDARY:
 653         /*
 654          * This BDS will be closed, and the job should be completed
 655          * before the BDS is closed, because we will access hidden
 656          * disk, secondary disk in backup_job_completed().
 657          */
 658         if (s->backup_job) {
 659             job_cancel_sync(&s->backup_job->job);
 660         }
 661
 662         if (!failover) {
 663             secondary_do_checkpoint(s, errp);
 664             s->stage = BLOCK_REPLICATION_DONE;
 665             aio_context_release(aio_context);
 666             return;
 667         }
 668
 669         s->stage = BLOCK_REPLICATION_FAILOVER;
 670         s->commit_job = commit_active_start(
 671                             NULL, s->active_disk->bs, s->secondary_disk->bs,
 672                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 673                             NULL, replication_done, bs, true, errp);
 674         break;
 675     default:
 676         aio_context_release(aio_context);
 677         abort();
 678     }
 679     aio_context_release(aio_context);
 680 }
 681
 682 static const char *const replication_strong_runtime_opts[] = {
 683     REPLICATION_MODE,
 684     REPLICATION_TOP_ID,
 685
 686     NULL
 687 };
 688
 689 static BlockDriver bdrv_replication = {
 690     .format_name                = "replication",
 691     .instance_size              = sizeof(BDRVReplicationState),
 692
 693     .bdrv_open                  = replication_open,
 694     .bdrv_close                 = replication_close,
 695     .bdrv_child_perm            = replication_child_perm,
 696
 697     .bdrv_getlength             = replication_getlength,
 698     .bdrv_co_readv              = replication_co_readv,
 699     .bdrv_co_writev             = replication_co_writev,
 700
 701     .is_filter                  = true,
 702     .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter,
 703
 704     .has_variable_length        = true,
 705     .strong_runtime_opts        = replication_strong_runtime_opts,
 706 };
 707
 708 static void bdrv_replication_init(void)
 709 {
 710     bdrv_register(&bdrv_replication);
 711 }
 712
 713 block_init(bdrv_replication_init);