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/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 "block/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     BlockJob *commit_job;
  39     BdrvChild *hidden_disk;
  40     BdrvChild *secondary_disk;
  41     BlockJob *backup_job;
  42     char *top_id;
  43     ReplicationState *rs;
  44     Error *blocker;
  45     bool orig_hidden_read_only;
  46     bool orig_secondary_read_only;
  47     int error;
  48 } BDRVReplicationState;
  49
  50 static void replication_start(ReplicationState *rs, ReplicationMode mode,
  51                               Error **errp);
  52 static void replication_do_checkpoint(ReplicationState *rs, Error **errp);
  53 static void replication_get_error(ReplicationState *rs, Error **errp);
  54 static void replication_stop(ReplicationState *rs, bool failover,
  55                              Error **errp);
  56
  57 #define REPLICATION_MODE        "mode"
  58 #define REPLICATION_TOP_ID      "top-id"
  59 static QemuOptsList replication_runtime_opts = {
  60     .name = "replication",
  61     .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head),
  62     .desc = {
  63         {
  64             .name = REPLICATION_MODE,
  65             .type = QEMU_OPT_STRING,
  66         },
  67         {
  68             .name = REPLICATION_TOP_ID,
  69             .type = QEMU_OPT_STRING,
  70         },
  71         { /* end of list */ }
  72     },
  73 };
  74
  75 static ReplicationOps replication_ops = {
  76     .start = replication_start,
  77     .checkpoint = replication_do_checkpoint,
  78     .get_error = replication_get_error,
  79     .stop = replication_stop,
  80 };
  81
  82 static int replication_open(BlockDriverState *bs, QDict *options,
  83                             int flags, Error **errp)
  84 {
  85     int ret;
  86     BDRVReplicationState *s = bs->opaque;
  87     QemuOpts *opts = NULL;
  88     const char *mode;
  89     const char *top_id;
  90
  91     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
  92                                BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
  93                                false, errp);
  94     if (!bs->file) {
  95         return -EINVAL;
  96     }
  97
  98     ret = -EINVAL;
  99     opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
 100     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
 101         goto fail;
 102     }
 103
 104     mode = qemu_opt_get(opts, REPLICATION_MODE);
 105     if (!mode) {
 106         error_setg(errp, "Missing the option mode");
 107         goto fail;
 108     }
 109
 110     if (!strcmp(mode, "primary")) {
 111         s->mode = REPLICATION_MODE_PRIMARY;
 112         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 113         if (top_id) {
 114             error_setg(errp,
 115                        "The primary side does not support option top-id");
 116             goto fail;
 117         }
 118     } else if (!strcmp(mode, "secondary")) {
 119         s->mode = REPLICATION_MODE_SECONDARY;
 120         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 121         s->top_id = g_strdup(top_id);
 122         if (!s->top_id) {
 123             error_setg(errp, "Missing the option top-id");
 124             goto fail;
 125         }
 126     } else {
 127         error_setg(errp,
 128                    "The option mode's value should be primary or secondary");
 129         goto fail;
 130     }
 131
 132     s->rs = replication_new(bs, &replication_ops);
 133
 134     ret = 0;
 135
 136 fail:
 137     qemu_opts_del(opts);
 138     return ret;
 139 }
 140
 141 static void replication_close(BlockDriverState *bs)
 142 {
 143     BDRVReplicationState *s = bs->opaque;
 144     Job *commit_job;
 145
 146     if (s->stage == BLOCK_REPLICATION_RUNNING) {
 147         replication_stop(s->rs, false, NULL);
 148     }
 149     if (s->stage == BLOCK_REPLICATION_FAILOVER) {
 150         commit_job = &s->commit_job->job;
 151         assert(commit_job->aio_context == qemu_get_current_aio_context());
 152         job_cancel_sync(commit_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                                    BdrvChildRole role,
 164                                    BlockReopenQueue *reopen_queue,
 165                                    uint64_t perm, uint64_t shared,
 166                                    uint64_t *nperm, uint64_t *nshared)
 167 {
 168     if (role & BDRV_CHILD_PRIMARY) {
 169         *nperm = BLK_PERM_CONSISTENT_READ;
 170     } else {
 171         *nperm = 0;
 172     }
 173
 174     if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) {
 175         *nperm |= BLK_PERM_WRITE;
 176     }
 177     *nshared = BLK_PERM_CONSISTENT_READ
 178                | BLK_PERM_WRITE
 179                | BLK_PERM_WRITE_UNCHANGED;
 180     return;
 181 }
 182
 183 static int64_t replication_getlength(BlockDriverState *bs)
 184 {
 185     return bdrv_getlength(bs->file->bs);
 186 }
 187
 188 static int replication_get_io_status(BDRVReplicationState *s)
 189 {
 190     switch (s->stage) {
 191     case BLOCK_REPLICATION_NONE:
 192         return -EIO;
 193     case BLOCK_REPLICATION_RUNNING:
 194         return 0;
 195     case BLOCK_REPLICATION_FAILOVER:
 196         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 197     case BLOCK_REPLICATION_FAILOVER_FAILED:
 198         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
 199     case BLOCK_REPLICATION_DONE:
 200         /*
 201          * active commit job completes, and active disk and secondary_disk
 202          * is swapped, so we can operate bs->file directly
 203          */
 204         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 205     default:
 206         abort();
 207     }
 208 }
 209
 210 static int replication_return_value(BDRVReplicationState *s, int ret)
 211 {
 212     if (s->mode == REPLICATION_MODE_SECONDARY) {
 213         return ret;
 214     }
 215
 216     if (ret < 0) {
 217         s->error = ret;
 218         ret = 0;
 219     }
 220
 221     return ret;
 222 }
 223
 224 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
 225                                              int64_t sector_num,
 226                                              int remaining_sectors,
 227                                              QEMUIOVector *qiov)
 228 {
 229     BDRVReplicationState *s = bs->opaque;
 230     int ret;
 231
 232     if (s->mode == REPLICATION_MODE_PRIMARY) {
 233         /* We only use it to forward primary write requests */
 234         return -EIO;
 235     }
 236
 237     ret = replication_get_io_status(s);
 238     if (ret < 0) {
 239         return ret;
 240     }
 241
 242     ret = bdrv_co_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
 243                          remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 244
 245     return replication_return_value(s, ret);
 246 }
 247
 248 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
 249                                               int64_t sector_num,
 250                                               int remaining_sectors,
 251                                               QEMUIOVector *qiov,
 252                                               int flags)
 253 {
 254     BDRVReplicationState *s = bs->opaque;
 255     QEMUIOVector hd_qiov;
 256     uint64_t bytes_done = 0;
 257     BdrvChild *top = bs->file;
 258     BdrvChild *base = s->secondary_disk;
 259     BdrvChild *target;
 260     int ret;
 261     int64_t n;
 262
 263     assert(!flags);
 264     ret = replication_get_io_status(s);
 265     if (ret < 0) {
 266         goto out;
 267     }
 268
 269     if (ret == 0) {
 270         ret = bdrv_co_pwritev(top, sector_num * BDRV_SECTOR_SIZE,
 271                               remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 272         return replication_return_value(s, ret);
 273     }
 274
 275     /*
 276      * Failover failed, only write to active disk if the sectors
 277      * have already been allocated in active disk/hidden disk.
 278      */
 279     qemu_iovec_init(&hd_qiov, qiov->niov);
 280     while (remaining_sectors > 0) {
 281         int64_t count;
 282
 283         ret = bdrv_is_allocated_above(top->bs, base->bs, false,
 284                                       sector_num * BDRV_SECTOR_SIZE,
 285                                       remaining_sectors * BDRV_SECTOR_SIZE,
 286                                       &count);
 287         if (ret < 0) {
 288             goto out1;
 289         }
 290
 291         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
 292         n = count >> BDRV_SECTOR_BITS;
 293         qemu_iovec_reset(&hd_qiov);
 294         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
 295
 296         target = ret ? top : base;
 297         ret = bdrv_co_pwritev(target, sector_num * BDRV_SECTOR_SIZE,
 298                               n * BDRV_SECTOR_SIZE, &hd_qiov, 0);
 299         if (ret < 0) {
 300             goto out1;
 301         }
 302
 303         remaining_sectors -= n;
 304         sector_num += n;
 305         bytes_done += count;
 306     }
 307
 308 out1:
 309     qemu_iovec_destroy(&hd_qiov);
 310 out:
 311     return ret;
 312 }
 313
 314 static void secondary_do_checkpoint(BlockDriverState *bs, Error **errp)
 315 {
 316     BDRVReplicationState *s = bs->opaque;
 317     BdrvChild *active_disk = bs->file;
 318     Error *local_err = NULL;
 319     int ret;
 320
 321     if (!s->backup_job) {
 322         error_setg(errp, "Backup job was cancelled unexpectedly");
 323         return;
 324     }
 325
 326     backup_do_checkpoint(s->backup_job, &local_err);
 327     if (local_err) {
 328         error_propagate(errp, local_err);
 329         return;
 330     }
 331
 332     if (!active_disk->bs->drv) {
 333         error_setg(errp, "Active disk %s is ejected",
 334                    active_disk->bs->node_name);
 335         return;
 336     }
 337
 338     ret = bdrv_make_empty(active_disk, errp);
 339     if (ret < 0) {
 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     BlockBackend *blk = blk_new(qemu_get_current_aio_context(),
 350                                 BLK_PERM_WRITE, BLK_PERM_ALL);
 351     blk_insert_bs(blk, s->hidden_disk->bs, &local_err);
 352     if (local_err) {
 353         error_propagate(errp, local_err);
 354         blk_unref(blk);
 355         return;
 356     }
 357
 358     ret = blk_make_empty(blk, errp);
 359     blk_unref(blk);
 360     if (ret < 0) {
 361         return;
 362     }
 363 }
 364
 365 /* This function is supposed to be called twice:
 366  * first with writable = true, then with writable = false.
 367  * The first call puts s->hidden_disk and s->secondary_disk in
 368  * r/w mode, and the second puts them back in their original state.
 369  */
 370 static void reopen_backing_file(BlockDriverState *bs, bool writable,
 371                                 Error **errp)
 372 {
 373     BDRVReplicationState *s = bs->opaque;
 374     BdrvChild *hidden_disk, *secondary_disk;
 375     BlockReopenQueue *reopen_queue = NULL;
 376
 377     /*
 378      * s->hidden_disk and s->secondary_disk may not be set yet, as they will
 379      * only be set after the children are writable.
 380      */
 381     hidden_disk = bs->file->bs->backing;
 382     secondary_disk = hidden_disk->bs->backing;
 383
 384     if (writable) {
 385         s->orig_hidden_read_only = bdrv_is_read_only(hidden_disk->bs);
 386         s->orig_secondary_read_only = bdrv_is_read_only(secondary_disk->bs);
 387     }
 388
 389     bdrv_subtree_drained_begin(hidden_disk->bs);
 390     bdrv_subtree_drained_begin(secondary_disk->bs);
 391
 392     if (s->orig_hidden_read_only) {
 393         QDict *opts = qdict_new();
 394         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 395         reopen_queue = bdrv_reopen_queue(reopen_queue, hidden_disk->bs,
 396                                          opts, true);
 397     }
 398
 399     if (s->orig_secondary_read_only) {
 400         QDict *opts = qdict_new();
 401         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 402         reopen_queue = bdrv_reopen_queue(reopen_queue, secondary_disk->bs,
 403                                          opts, true);
 404     }
 405
 406     if (reopen_queue) {
 407         AioContext *ctx = bdrv_get_aio_context(bs);
 408         if (ctx != qemu_get_aio_context()) {
 409             aio_context_release(ctx);
 410         }
 411         bdrv_reopen_multiple(reopen_queue, errp);
 412         if (ctx != qemu_get_aio_context()) {
 413             aio_context_acquire(ctx);
 414         }
 415     }
 416
 417     bdrv_subtree_drained_end(hidden_disk->bs);
 418     bdrv_subtree_drained_end(secondary_disk->bs);
 419 }
 420
 421 static void backup_job_cleanup(BlockDriverState *bs)
 422 {
 423     BDRVReplicationState *s = bs->opaque;
 424     BlockDriverState *top_bs;
 425
 426     s->backup_job = NULL;
 427
 428     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 429     if (!top_bs) {
 430         return;
 431     }
 432     bdrv_op_unblock_all(top_bs, s->blocker);
 433     error_free(s->blocker);
 434     reopen_backing_file(bs, false, NULL);
 435 }
 436
 437 static void backup_job_completed(void *opaque, int ret)
 438 {
 439     BlockDriverState *bs = opaque;
 440     BDRVReplicationState *s = bs->opaque;
 441
 442     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 443         /* The backup job is cancelled unexpectedly */
 444         s->error = -EIO;
 445     }
 446
 447     backup_job_cleanup(bs);
 448 }
 449
 450 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 451 {
 452     BdrvChild *child;
 453
 454     /* The bs itself is the top_bs */
 455     if (top_bs == bs) {
 456         return true;
 457     }
 458
 459     /* Iterate over top_bs's children */
 460     QLIST_FOREACH(child, &top_bs->children, next) {
 461         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 462             return true;
 463         }
 464     }
 465
 466     return false;
 467 }
 468
 469 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 470                               Error **errp)
 471 {
 472     BlockDriverState *bs = rs->opaque;
 473     BDRVReplicationState *s;
 474     BlockDriverState *top_bs;
 475     BdrvChild *active_disk, *hidden_disk, *secondary_disk;
 476     int64_t active_length, hidden_length, disk_length;
 477     AioContext *aio_context;
 478     Error *local_err = NULL;
 479     BackupPerf perf = { .use_copy_range = true, .max_workers = 1 };
 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_DONE ||
 486         s->stage == BLOCK_REPLICATION_FAILOVER) {
 487         /*
 488          * This case happens when a secondary is promoted to primary.
 489          * Ignore the request because the secondary side of replication
 490          * doesn't have to do anything anymore.
 491          */
 492         aio_context_release(aio_context);
 493         return;
 494     }
 495
 496     if (s->stage != BLOCK_REPLICATION_NONE) {
 497         error_setg(errp, "Block replication is running or done");
 498         aio_context_release(aio_context);
 499         return;
 500     }
 501
 502     if (s->mode != mode) {
 503         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 504                    " but got %d", s->mode, mode);
 505         aio_context_release(aio_context);
 506         return;
 507     }
 508
 509     switch (s->mode) {
 510     case REPLICATION_MODE_PRIMARY:
 511         break;
 512     case REPLICATION_MODE_SECONDARY:
 513         active_disk = bs->file;
 514         if (!active_disk || !active_disk->bs || !active_disk->bs->backing) {
 515             error_setg(errp, "Active disk doesn't have backing file");
 516             aio_context_release(aio_context);
 517             return;
 518         }
 519
 520         hidden_disk = active_disk->bs->backing;
 521         if (!hidden_disk->bs || !hidden_disk->bs->backing) {
 522             error_setg(errp, "Hidden disk doesn't have backing file");
 523             aio_context_release(aio_context);
 524             return;
 525         }
 526
 527         secondary_disk = hidden_disk->bs->backing;
 528         if (!secondary_disk->bs || !bdrv_has_blk(secondary_disk->bs)) {
 529             error_setg(errp, "The secondary disk doesn't have block backend");
 530             aio_context_release(aio_context);
 531             return;
 532         }
 533
 534         /* verify the length */
 535         active_length = bdrv_getlength(active_disk->bs);
 536         hidden_length = bdrv_getlength(hidden_disk->bs);
 537         disk_length = bdrv_getlength(secondary_disk->bs);
 538         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 539             active_length != hidden_length || hidden_length != disk_length) {
 540             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 541                        " are not the same");
 542             aio_context_release(aio_context);
 543             return;
 544         }
 545
 546         /* Must be true, or the bdrv_getlength() calls would have failed */
 547         assert(active_disk->bs->drv && hidden_disk->bs->drv);
 548
 549         if (!active_disk->bs->drv->bdrv_make_empty ||
 550             !hidden_disk->bs->drv->bdrv_make_empty) {
 551             error_setg(errp,
 552                        "Active disk or hidden disk doesn't support make_empty");
 553             aio_context_release(aio_context);
 554             return;
 555         }
 556
 557         /* reopen the backing file in r/w mode */
 558         reopen_backing_file(bs, true, &local_err);
 559         if (local_err) {
 560             error_propagate(errp, local_err);
 561             aio_context_release(aio_context);
 562             return;
 563         }
 564
 565         bdrv_ref(hidden_disk->bs);
 566         s->hidden_disk = bdrv_attach_child(bs, hidden_disk->bs, "hidden disk",
 567                                            &child_of_bds, BDRV_CHILD_DATA,
 568                                            &local_err);
 569         if (local_err) {
 570             error_propagate(errp, local_err);
 571             aio_context_release(aio_context);
 572             return;
 573         }
 574
 575         bdrv_ref(secondary_disk->bs);
 576         s->secondary_disk = bdrv_attach_child(bs, secondary_disk->bs,
 577                                               "secondary disk", &child_of_bds,
 578                                               BDRV_CHILD_DATA, &local_err);
 579         if (local_err) {
 580             error_propagate(errp, local_err);
 581             aio_context_release(aio_context);
 582             return;
 583         }
 584
 585         /* start backup job now */
 586         error_setg(&s->blocker,
 587                    "Block device is in use by internal backup job");
 588
 589         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 590         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 591             !check_top_bs(top_bs, bs)) {
 592             error_setg(errp, "No top_bs or it is invalid");
 593             reopen_backing_file(bs, false, NULL);
 594             aio_context_release(aio_context);
 595             return;
 596         }
 597         bdrv_op_block_all(top_bs, s->blocker);
 598         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 599
 600         s->backup_job = backup_job_create(
 601                                 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 602                                 0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
 603                                 &perf,
 604                                 BLOCKDEV_ON_ERROR_REPORT,
 605                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
 606                                 backup_job_completed, bs, NULL, &local_err);
 607         if (local_err) {
 608             error_propagate(errp, local_err);
 609             backup_job_cleanup(bs);
 610             aio_context_release(aio_context);
 611             return;
 612         }
 613         job_start(&s->backup_job->job);
 614         break;
 615     default:
 616         aio_context_release(aio_context);
 617         abort();
 618     }
 619
 620     s->stage = BLOCK_REPLICATION_RUNNING;
 621
 622     if (s->mode == REPLICATION_MODE_SECONDARY) {
 623         secondary_do_checkpoint(bs, errp);
 624     }
 625
 626     s->error = 0;
 627     aio_context_release(aio_context);
 628 }
 629
 630 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
 631 {
 632     BlockDriverState *bs = rs->opaque;
 633     BDRVReplicationState *s;
 634     AioContext *aio_context;
 635
 636     aio_context = bdrv_get_aio_context(bs);
 637     aio_context_acquire(aio_context);
 638     s = bs->opaque;
 639
 640     if (s->stage == BLOCK_REPLICATION_DONE ||
 641         s->stage == BLOCK_REPLICATION_FAILOVER) {
 642         /*
 643          * This case happens when a secondary was promoted to primary.
 644          * Ignore the request because the secondary side of replication
 645          * doesn't have to do anything anymore.
 646          */
 647         aio_context_release(aio_context);
 648         return;
 649     }
 650
 651     if (s->mode == REPLICATION_MODE_SECONDARY) {
 652         secondary_do_checkpoint(bs, errp);
 653     }
 654     aio_context_release(aio_context);
 655 }
 656
 657 static void replication_get_error(ReplicationState *rs, Error **errp)
 658 {
 659     BlockDriverState *bs = rs->opaque;
 660     BDRVReplicationState *s;
 661     AioContext *aio_context;
 662
 663     aio_context = bdrv_get_aio_context(bs);
 664     aio_context_acquire(aio_context);
 665     s = bs->opaque;
 666
 667     if (s->stage == BLOCK_REPLICATION_NONE) {
 668         error_setg(errp, "Block replication is not running");
 669         aio_context_release(aio_context);
 670         return;
 671     }
 672
 673     if (s->error) {
 674         error_setg(errp, "I/O error occurred");
 675         aio_context_release(aio_context);
 676         return;
 677     }
 678     aio_context_release(aio_context);
 679 }
 680
 681 static void replication_done(void *opaque, int ret)
 682 {
 683     BlockDriverState *bs = opaque;
 684     BDRVReplicationState *s = bs->opaque;
 685
 686     if (ret == 0) {
 687         s->stage = BLOCK_REPLICATION_DONE;
 688
 689         bdrv_unref_child(bs, s->secondary_disk);
 690         s->secondary_disk = NULL;
 691         bdrv_unref_child(bs, s->hidden_disk);
 692         s->hidden_disk = NULL;
 693         s->error = 0;
 694     } else {
 695         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 696         s->error = -EIO;
 697     }
 698 }
 699
 700 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 701 {
 702     BlockDriverState *bs = rs->opaque;
 703     BDRVReplicationState *s;
 704     AioContext *aio_context;
 705
 706     aio_context = bdrv_get_aio_context(bs);
 707     aio_context_acquire(aio_context);
 708     s = bs->opaque;
 709
 710     if (s->stage == BLOCK_REPLICATION_DONE ||
 711         s->stage == BLOCK_REPLICATION_FAILOVER) {
 712         /*
 713          * This case happens when a secondary was promoted to primary.
 714          * Ignore the request because the secondary side of replication
 715          * doesn't have to do anything anymore.
 716          */
 717         aio_context_release(aio_context);
 718         return;
 719     }
 720
 721     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 722         error_setg(errp, "Block replication is not running");
 723         aio_context_release(aio_context);
 724         return;
 725     }
 726
 727     switch (s->mode) {
 728     case REPLICATION_MODE_PRIMARY:
 729         s->stage = BLOCK_REPLICATION_DONE;
 730         s->error = 0;
 731         break;
 732     case REPLICATION_MODE_SECONDARY:
 733         /*
 734          * This BDS will be closed, and the job should be completed
 735          * before the BDS is closed, because we will access hidden
 736          * disk, secondary disk in backup_job_completed().
 737          */
 738         if (s->backup_job) {
 739             job_cancel_sync(&s->backup_job->job);
 740         }
 741
 742         if (!failover) {
 743             secondary_do_checkpoint(bs, errp);
 744             s->stage = BLOCK_REPLICATION_DONE;
 745             aio_context_release(aio_context);
 746             return;
 747         }
 748
 749         s->stage = BLOCK_REPLICATION_FAILOVER;
 750         s->commit_job = commit_active_start(
 751                             NULL, bs->file->bs, s->secondary_disk->bs,
 752                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 753                             NULL, replication_done, bs, true, errp);
 754         break;
 755     default:
 756         aio_context_release(aio_context);
 757         abort();
 758     }
 759     aio_context_release(aio_context);
 760 }
 761
 762 static const char *const replication_strong_runtime_opts[] = {
 763     REPLICATION_MODE,
 764     REPLICATION_TOP_ID,
 765
 766     NULL
 767 };
 768
 769 static BlockDriver bdrv_replication = {
 770     .format_name                = "replication",
 771     .instance_size              = sizeof(BDRVReplicationState),
 772
 773     .bdrv_open                  = replication_open,
 774     .bdrv_close                 = replication_close,
 775     .bdrv_child_perm            = replication_child_perm,
 776
 777     .bdrv_getlength             = replication_getlength,
 778     .bdrv_co_readv              = replication_co_readv,
 779     .bdrv_co_writev             = replication_co_writev,
 780
 781     .is_filter                  = true,
 782
 783     .has_variable_length        = true,
 784     .strong_runtime_opts        = replication_strong_runtime_opts,
 785 };
 786
 787 static void bdrv_replication_init(void)
 788 {
 789     bdrv_register(&bdrv_replication);
 790 }
 791
 792 block_init(bdrv_replication_init);