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         QDict *opts = qdict_new();
 375         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 376         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs,
 377                                          opts, true);
 378     }
 379
 380     if (s->orig_secondary_read_only) {
 381         QDict *opts = qdict_new();
 382         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 383         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
 384                                          opts, true);
 385     }
 386
 387     if (reopen_queue) {
 388         bdrv_reopen_multiple(reopen_queue, &local_err);
 389         error_propagate(errp, local_err);
 390     }
 391
 392     bdrv_subtree_drained_end(s->hidden_disk->bs);
 393     bdrv_subtree_drained_end(s->secondary_disk->bs);
 394 }
 395
 396 static void backup_job_cleanup(BlockDriverState *bs)
 397 {
 398     BDRVReplicationState *s = bs->opaque;
 399     BlockDriverState *top_bs;
 400
 401     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 402     if (!top_bs) {
 403         return;
 404     }
 405     bdrv_op_unblock_all(top_bs, s->blocker);
 406     error_free(s->blocker);
 407     reopen_backing_file(bs, false, NULL);
 408 }
 409
 410 static void backup_job_completed(void *opaque, int ret)
 411 {
 412     BlockDriverState *bs = opaque;
 413     BDRVReplicationState *s = bs->opaque;
 414
 415     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 416         /* The backup job is cancelled unexpectedly */
 417         s->error = -EIO;
 418     }
 419
 420     backup_job_cleanup(bs);
 421 }
 422
 423 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 424 {
 425     BdrvChild *child;
 426
 427     /* The bs itself is the top_bs */
 428     if (top_bs == bs) {
 429         return true;
 430     }
 431
 432     /* Iterate over top_bs's children */
 433     QLIST_FOREACH(child, &top_bs->children, next) {
 434         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 435             return true;
 436         }
 437     }
 438
 439     return false;
 440 }
 441
 442 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 443                               Error **errp)
 444 {
 445     BlockDriverState *bs = rs->opaque;
 446     BDRVReplicationState *s;
 447     BlockDriverState *top_bs;
 448     int64_t active_length, hidden_length, disk_length;
 449     AioContext *aio_context;
 450     Error *local_err = NULL;
 451     BlockJob *job;
 452
 453     aio_context = bdrv_get_aio_context(bs);
 454     aio_context_acquire(aio_context);
 455     s = bs->opaque;
 456
 457     if (s->stage != BLOCK_REPLICATION_NONE) {
 458         error_setg(errp, "Block replication is running or done");
 459         aio_context_release(aio_context);
 460         return;
 461     }
 462
 463     if (s->mode != mode) {
 464         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 465                    " but got %d", s->mode, mode);
 466         aio_context_release(aio_context);
 467         return;
 468     }
 469
 470     switch (s->mode) {
 471     case REPLICATION_MODE_PRIMARY:
 472         break;
 473     case REPLICATION_MODE_SECONDARY:
 474         s->active_disk = bs->file;
 475         if (!s->active_disk || !s->active_disk->bs ||
 476                                     !s->active_disk->bs->backing) {
 477             error_setg(errp, "Active disk doesn't have backing file");
 478             aio_context_release(aio_context);
 479             return;
 480         }
 481
 482         s->hidden_disk = s->active_disk->bs->backing;
 483         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
 484             error_setg(errp, "Hidden disk doesn't have backing file");
 485             aio_context_release(aio_context);
 486             return;
 487         }
 488
 489         s->secondary_disk = s->hidden_disk->bs->backing;
 490         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
 491             error_setg(errp, "The secondary disk doesn't have block backend");
 492             aio_context_release(aio_context);
 493             return;
 494         }
 495
 496         /* verify the length */
 497         active_length = bdrv_getlength(s->active_disk->bs);
 498         hidden_length = bdrv_getlength(s->hidden_disk->bs);
 499         disk_length = bdrv_getlength(s->secondary_disk->bs);
 500         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 501             active_length != hidden_length || hidden_length != disk_length) {
 502             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 503                        " are not the same");
 504             aio_context_release(aio_context);
 505             return;
 506         }
 507
 508         /* Must be true, or the bdrv_getlength() calls would have failed */
 509         assert(s->active_disk->bs->drv && s->hidden_disk->bs->drv);
 510
 511         if (!s->active_disk->bs->drv->bdrv_make_empty ||
 512             !s->hidden_disk->bs->drv->bdrv_make_empty) {
 513             error_setg(errp,
 514                        "Active disk or hidden disk doesn't support make_empty");
 515             aio_context_release(aio_context);
 516             return;
 517         }
 518
 519         /* reopen the backing file in r/w mode */
 520         reopen_backing_file(bs, true, &local_err);
 521         if (local_err) {
 522             error_propagate(errp, local_err);
 523             aio_context_release(aio_context);
 524             return;
 525         }
 526
 527         /* start backup job now */
 528         error_setg(&s->blocker,
 529                    "Block device is in use by internal backup job");
 530
 531         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 532         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 533             !check_top_bs(top_bs, bs)) {
 534             error_setg(errp, "No top_bs or it is invalid");
 535             reopen_backing_file(bs, false, NULL);
 536             aio_context_release(aio_context);
 537             return;
 538         }
 539         bdrv_op_block_all(top_bs, s->blocker);
 540         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 541
 542         job = backup_job_create(NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 543                                 0, MIRROR_SYNC_MODE_NONE, NULL, false,
 544                                 BLOCKDEV_ON_ERROR_REPORT,
 545                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
 546                                 backup_job_completed, bs, NULL, &local_err);
 547         if (local_err) {
 548             error_propagate(errp, local_err);
 549             backup_job_cleanup(bs);
 550             aio_context_release(aio_context);
 551             return;
 552         }
 553         job_start(&job->job);
 554         break;
 555     default:
 556         aio_context_release(aio_context);
 557         abort();
 558     }
 559
 560     s->stage = BLOCK_REPLICATION_RUNNING;
 561
 562     if (s->mode == REPLICATION_MODE_SECONDARY) {
 563         secondary_do_checkpoint(s, errp);
 564     }
 565
 566     s->error = 0;
 567     aio_context_release(aio_context);
 568 }
 569
 570 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
 571 {
 572     BlockDriverState *bs = rs->opaque;
 573     BDRVReplicationState *s;
 574     AioContext *aio_context;
 575
 576     aio_context = bdrv_get_aio_context(bs);
 577     aio_context_acquire(aio_context);
 578     s = bs->opaque;
 579
 580     if (s->mode == REPLICATION_MODE_SECONDARY) {
 581         secondary_do_checkpoint(s, errp);
 582     }
 583     aio_context_release(aio_context);
 584 }
 585
 586 static void replication_get_error(ReplicationState *rs, Error **errp)
 587 {
 588     BlockDriverState *bs = rs->opaque;
 589     BDRVReplicationState *s;
 590     AioContext *aio_context;
 591
 592     aio_context = bdrv_get_aio_context(bs);
 593     aio_context_acquire(aio_context);
 594     s = bs->opaque;
 595
 596     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 597         error_setg(errp, "Block replication is not running");
 598         aio_context_release(aio_context);
 599         return;
 600     }
 601
 602     if (s->error) {
 603         error_setg(errp, "I/O error occurred");
 604         aio_context_release(aio_context);
 605         return;
 606     }
 607     aio_context_release(aio_context);
 608 }
 609
 610 static void replication_done(void *opaque, int ret)
 611 {
 612     BlockDriverState *bs = opaque;
 613     BDRVReplicationState *s = bs->opaque;
 614
 615     if (ret == 0) {
 616         s->stage = BLOCK_REPLICATION_DONE;
 617
 618         s->active_disk = NULL;
 619         s->secondary_disk = NULL;
 620         s->hidden_disk = NULL;
 621         s->error = 0;
 622     } else {
 623         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 624         s->error = -EIO;
 625     }
 626 }
 627
 628 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 629 {
 630     BlockDriverState *bs = rs->opaque;
 631     BDRVReplicationState *s;
 632     AioContext *aio_context;
 633
 634     aio_context = bdrv_get_aio_context(bs);
 635     aio_context_acquire(aio_context);
 636     s = bs->opaque;
 637
 638     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 639         error_setg(errp, "Block replication is not running");
 640         aio_context_release(aio_context);
 641         return;
 642     }
 643
 644     switch (s->mode) {
 645     case REPLICATION_MODE_PRIMARY:
 646         s->stage = BLOCK_REPLICATION_DONE;
 647         s->error = 0;
 648         break;
 649     case REPLICATION_MODE_SECONDARY:
 650         /*
 651          * This BDS will be closed, and the job should be completed
 652          * before the BDS is closed, because we will access hidden
 653          * disk, secondary disk in backup_job_completed().
 654          */
 655         if (s->secondary_disk->bs->job) {
 656             job_cancel_sync(&s->secondary_disk->bs->job->job);
 657         }
 658
 659         if (!failover) {
 660             secondary_do_checkpoint(s, errp);
 661             s->stage = BLOCK_REPLICATION_DONE;
 662             aio_context_release(aio_context);
 663             return;
 664         }
 665
 666         s->stage = BLOCK_REPLICATION_FAILOVER;
 667         commit_active_start(NULL, s->active_disk->bs, s->secondary_disk->bs,
 668                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 669                             NULL, replication_done, bs, true, errp);
 670         break;
 671     default:
 672         aio_context_release(aio_context);
 673         abort();
 674     }
 675     aio_context_release(aio_context);
 676 }
 677
 678 static const char *const replication_strong_runtime_opts[] = {
 679     REPLICATION_MODE,
 680     REPLICATION_TOP_ID,
 681
 682     NULL
 683 };
 684
 685 static BlockDriver bdrv_replication = {
 686     .format_name                = "replication",
 687     .instance_size              = sizeof(BDRVReplicationState),
 688
 689     .bdrv_open                  = replication_open,
 690     .bdrv_close                 = replication_close,
 691     .bdrv_child_perm            = replication_child_perm,
 692
 693     .bdrv_getlength             = replication_getlength,
 694     .bdrv_co_readv              = replication_co_readv,
 695     .bdrv_co_writev             = replication_co_writev,
 696
 697     .is_filter                  = true,
 698     .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter,
 699
 700     .has_variable_length        = true,
 701     .strong_runtime_opts        = replication_strong_runtime_opts,
 702 };
 703
 704 static void bdrv_replication_init(void)
 705 {
 706     bdrv_register(&bdrv_replication);
 707 }
 708
 709 block_init(bdrv_replication_init);