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, false);
 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     ret = bdrv_make_empty(s->hidden_disk, errp);
 350     if (ret < 0) {
 351         return;
 352     }
 353 }
 354
 355 /* This function is supposed to be called twice:
 356  * first with writable = true, then with writable = false.
 357  * The first call puts s->hidden_disk and s->secondary_disk in
 358  * r/w mode, and the second puts them back in their original state.
 359  */
 360 static void reopen_backing_file(BlockDriverState *bs, bool writable,
 361                                 Error **errp)
 362 {
 363     BDRVReplicationState *s = bs->opaque;
 364     BdrvChild *hidden_disk, *secondary_disk;
 365     BlockReopenQueue *reopen_queue = NULL;
 366
 367     /*
 368      * s->hidden_disk and s->secondary_disk may not be set yet, as they will
 369      * only be set after the children are writable.
 370      */
 371     hidden_disk = bs->file->bs->backing;
 372     secondary_disk = hidden_disk->bs->backing;
 373
 374     if (writable) {
 375         s->orig_hidden_read_only = bdrv_is_read_only(hidden_disk->bs);
 376         s->orig_secondary_read_only = bdrv_is_read_only(secondary_disk->bs);
 377     }
 378
 379     bdrv_subtree_drained_begin(hidden_disk->bs);
 380     bdrv_subtree_drained_begin(secondary_disk->bs);
 381
 382     if (s->orig_hidden_read_only) {
 383         QDict *opts = qdict_new();
 384         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 385         reopen_queue = bdrv_reopen_queue(reopen_queue, hidden_disk->bs,
 386                                          opts, true);
 387     }
 388
 389     if (s->orig_secondary_read_only) {
 390         QDict *opts = qdict_new();
 391         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 392         reopen_queue = bdrv_reopen_queue(reopen_queue, secondary_disk->bs,
 393                                          opts, true);
 394     }
 395
 396     if (reopen_queue) {
 397         AioContext *ctx = bdrv_get_aio_context(bs);
 398         if (ctx != qemu_get_aio_context()) {
 399             aio_context_release(ctx);
 400         }
 401         bdrv_reopen_multiple(reopen_queue, errp);
 402         if (ctx != qemu_get_aio_context()) {
 403             aio_context_acquire(ctx);
 404         }
 405     }
 406
 407     bdrv_subtree_drained_end(hidden_disk->bs);
 408     bdrv_subtree_drained_end(secondary_disk->bs);
 409 }
 410
 411 static void backup_job_cleanup(BlockDriverState *bs)
 412 {
 413     BDRVReplicationState *s = bs->opaque;
 414     BlockDriverState *top_bs;
 415
 416     s->backup_job = NULL;
 417
 418     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 419     if (!top_bs) {
 420         return;
 421     }
 422     bdrv_op_unblock_all(top_bs, s->blocker);
 423     error_free(s->blocker);
 424     reopen_backing_file(bs, false, NULL);
 425 }
 426
 427 static void backup_job_completed(void *opaque, int ret)
 428 {
 429     BlockDriverState *bs = opaque;
 430     BDRVReplicationState *s = bs->opaque;
 431
 432     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 433         /* The backup job is cancelled unexpectedly */
 434         s->error = -EIO;
 435     }
 436
 437     backup_job_cleanup(bs);
 438 }
 439
 440 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 441 {
 442     BdrvChild *child;
 443
 444     /* The bs itself is the top_bs */
 445     if (top_bs == bs) {
 446         return true;
 447     }
 448
 449     /* Iterate over top_bs's children */
 450     QLIST_FOREACH(child, &top_bs->children, next) {
 451         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 452             return true;
 453         }
 454     }
 455
 456     return false;
 457 }
 458
 459 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 460                               Error **errp)
 461 {
 462     BlockDriverState *bs = rs->opaque;
 463     BDRVReplicationState *s;
 464     BlockDriverState *top_bs;
 465     BdrvChild *active_disk, *hidden_disk, *secondary_disk;
 466     int64_t active_length, hidden_length, disk_length;
 467     AioContext *aio_context;
 468     Error *local_err = NULL;
 469     BackupPerf perf = { .use_copy_range = true, .max_workers = 1 };
 470
 471     aio_context = bdrv_get_aio_context(bs);
 472     aio_context_acquire(aio_context);
 473     s = bs->opaque;
 474
 475     if (s->stage == BLOCK_REPLICATION_DONE ||
 476         s->stage == BLOCK_REPLICATION_FAILOVER) {
 477         /*
 478          * This case happens when a secondary is promoted to primary.
 479          * Ignore the request because the secondary side of replication
 480          * doesn't have to do anything anymore.
 481          */
 482         aio_context_release(aio_context);
 483         return;
 484     }
 485
 486     if (s->stage != BLOCK_REPLICATION_NONE) {
 487         error_setg(errp, "Block replication is running or done");
 488         aio_context_release(aio_context);
 489         return;
 490     }
 491
 492     if (s->mode != mode) {
 493         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 494                    " but got %d", s->mode, mode);
 495         aio_context_release(aio_context);
 496         return;
 497     }
 498
 499     switch (s->mode) {
 500     case REPLICATION_MODE_PRIMARY:
 501         break;
 502     case REPLICATION_MODE_SECONDARY:
 503         active_disk = bs->file;
 504         if (!active_disk || !active_disk->bs || !active_disk->bs->backing) {
 505             error_setg(errp, "Active disk doesn't have backing file");
 506             aio_context_release(aio_context);
 507             return;
 508         }
 509
 510         hidden_disk = active_disk->bs->backing;
 511         if (!hidden_disk->bs || !hidden_disk->bs->backing) {
 512             error_setg(errp, "Hidden disk doesn't have backing file");
 513             aio_context_release(aio_context);
 514             return;
 515         }
 516
 517         secondary_disk = hidden_disk->bs->backing;
 518         if (!secondary_disk->bs || !bdrv_has_blk(secondary_disk->bs)) {
 519             error_setg(errp, "The secondary disk doesn't have block backend");
 520             aio_context_release(aio_context);
 521             return;
 522         }
 523
 524         /* verify the length */
 525         active_length = bdrv_getlength(active_disk->bs);
 526         hidden_length = bdrv_getlength(hidden_disk->bs);
 527         disk_length = bdrv_getlength(secondary_disk->bs);
 528         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 529             active_length != hidden_length || hidden_length != disk_length) {
 530             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 531                        " are not the same");
 532             aio_context_release(aio_context);
 533             return;
 534         }
 535
 536         /* Must be true, or the bdrv_getlength() calls would have failed */
 537         assert(active_disk->bs->drv && hidden_disk->bs->drv);
 538
 539         if (!active_disk->bs->drv->bdrv_make_empty ||
 540             !hidden_disk->bs->drv->bdrv_make_empty) {
 541             error_setg(errp,
 542                        "Active disk or hidden disk doesn't support make_empty");
 543             aio_context_release(aio_context);
 544             return;
 545         }
 546
 547         /* reopen the backing file in r/w mode */
 548         reopen_backing_file(bs, true, &local_err);
 549         if (local_err) {
 550             error_propagate(errp, local_err);
 551             aio_context_release(aio_context);
 552             return;
 553         }
 554
 555         bdrv_ref(hidden_disk->bs);
 556         s->hidden_disk = bdrv_attach_child(bs, hidden_disk->bs, "hidden disk",
 557                                            &child_of_bds, BDRV_CHILD_DATA,
 558                                            &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(secondary_disk->bs);
 566         s->secondary_disk = bdrv_attach_child(bs, secondary_disk->bs,
 567                                               "secondary disk", &child_of_bds,
 568                                               BDRV_CHILD_DATA, &local_err);
 569         if (local_err) {
 570             error_propagate(errp, local_err);
 571             aio_context_release(aio_context);
 572             return;
 573         }
 574
 575         /* start backup job now */
 576         error_setg(&s->blocker,
 577                    "Block device is in use by internal backup job");
 578
 579         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 580         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 581             !check_top_bs(top_bs, bs)) {
 582             error_setg(errp, "No top_bs or it is invalid");
 583             reopen_backing_file(bs, false, NULL);
 584             aio_context_release(aio_context);
 585             return;
 586         }
 587         bdrv_op_block_all(top_bs, s->blocker);
 588         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 589
 590         s->backup_job = backup_job_create(
 591                                 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 592                                 0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
 593                                 &perf,
 594                                 BLOCKDEV_ON_ERROR_REPORT,
 595                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
 596                                 backup_job_completed, bs, NULL, &local_err);
 597         if (local_err) {
 598             error_propagate(errp, local_err);
 599             backup_job_cleanup(bs);
 600             aio_context_release(aio_context);
 601             return;
 602         }
 603         job_start(&s->backup_job->job);
 604         break;
 605     default:
 606         aio_context_release(aio_context);
 607         abort();
 608     }
 609
 610     s->stage = BLOCK_REPLICATION_RUNNING;
 611
 612     if (s->mode == REPLICATION_MODE_SECONDARY) {
 613         secondary_do_checkpoint(bs, errp);
 614     }
 615
 616     s->error = 0;
 617     aio_context_release(aio_context);
 618 }
 619
 620 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
 621 {
 622     BlockDriverState *bs = rs->opaque;
 623     BDRVReplicationState *s;
 624     AioContext *aio_context;
 625
 626     aio_context = bdrv_get_aio_context(bs);
 627     aio_context_acquire(aio_context);
 628     s = bs->opaque;
 629
 630     if (s->stage == BLOCK_REPLICATION_DONE ||
 631         s->stage == BLOCK_REPLICATION_FAILOVER) {
 632         /*
 633          * This case happens when a secondary was promoted to primary.
 634          * Ignore the request because the secondary side of replication
 635          * doesn't have to do anything anymore.
 636          */
 637         aio_context_release(aio_context);
 638         return;
 639     }
 640
 641     if (s->mode == REPLICATION_MODE_SECONDARY) {
 642         secondary_do_checkpoint(bs, errp);
 643     }
 644     aio_context_release(aio_context);
 645 }
 646
 647 static void replication_get_error(ReplicationState *rs, Error **errp)
 648 {
 649     BlockDriverState *bs = rs->opaque;
 650     BDRVReplicationState *s;
 651     AioContext *aio_context;
 652
 653     aio_context = bdrv_get_aio_context(bs);
 654     aio_context_acquire(aio_context);
 655     s = bs->opaque;
 656
 657     if (s->stage == BLOCK_REPLICATION_NONE) {
 658         error_setg(errp, "Block replication is not running");
 659         aio_context_release(aio_context);
 660         return;
 661     }
 662
 663     if (s->error) {
 664         error_setg(errp, "I/O error occurred");
 665         aio_context_release(aio_context);
 666         return;
 667     }
 668     aio_context_release(aio_context);
 669 }
 670
 671 static void replication_done(void *opaque, int ret)
 672 {
 673     BlockDriverState *bs = opaque;
 674     BDRVReplicationState *s = bs->opaque;
 675
 676     if (ret == 0) {
 677         s->stage = BLOCK_REPLICATION_DONE;
 678
 679         bdrv_unref_child(bs, s->secondary_disk);
 680         s->secondary_disk = NULL;
 681         bdrv_unref_child(bs, s->hidden_disk);
 682         s->hidden_disk = NULL;
 683         s->error = 0;
 684     } else {
 685         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 686         s->error = -EIO;
 687     }
 688 }
 689
 690 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 691 {
 692     BlockDriverState *bs = rs->opaque;
 693     BDRVReplicationState *s;
 694     AioContext *aio_context;
 695
 696     aio_context = bdrv_get_aio_context(bs);
 697     aio_context_acquire(aio_context);
 698     s = bs->opaque;
 699
 700     if (s->stage == BLOCK_REPLICATION_DONE ||
 701         s->stage == BLOCK_REPLICATION_FAILOVER) {
 702         /*
 703          * This case happens when a secondary was promoted to primary.
 704          * Ignore the request because the secondary side of replication
 705          * doesn't have to do anything anymore.
 706          */
 707         aio_context_release(aio_context);
 708         return;
 709     }
 710
 711     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 712         error_setg(errp, "Block replication is not running");
 713         aio_context_release(aio_context);
 714         return;
 715     }
 716
 717     switch (s->mode) {
 718     case REPLICATION_MODE_PRIMARY:
 719         s->stage = BLOCK_REPLICATION_DONE;
 720         s->error = 0;
 721         break;
 722     case REPLICATION_MODE_SECONDARY:
 723         /*
 724          * This BDS will be closed, and the job should be completed
 725          * before the BDS is closed, because we will access hidden
 726          * disk, secondary disk in backup_job_completed().
 727          */
 728         if (s->backup_job) {
 729             job_cancel_sync(&s->backup_job->job, true);
 730         }
 731
 732         if (!failover) {
 733             secondary_do_checkpoint(bs, errp);
 734             s->stage = BLOCK_REPLICATION_DONE;
 735             aio_context_release(aio_context);
 736             return;
 737         }
 738
 739         s->stage = BLOCK_REPLICATION_FAILOVER;
 740         s->commit_job = commit_active_start(
 741                             NULL, bs->file->bs, s->secondary_disk->bs,
 742                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 743                             NULL, replication_done, bs, true, errp);
 744         break;
 745     default:
 746         aio_context_release(aio_context);
 747         abort();
 748     }
 749     aio_context_release(aio_context);
 750 }
 751
 752 static const char *const replication_strong_runtime_opts[] = {
 753     REPLICATION_MODE,
 754     REPLICATION_TOP_ID,
 755
 756     NULL
 757 };
 758
 759 static BlockDriver bdrv_replication = {
 760     .format_name                = "replication",
 761     .instance_size              = sizeof(BDRVReplicationState),
 762
 763     .bdrv_open                  = replication_open,
 764     .bdrv_close                 = replication_close,
 765     .bdrv_child_perm            = replication_child_perm,
 766
 767     .bdrv_getlength             = replication_getlength,
 768     .bdrv_co_readv              = replication_co_readv,
 769     .bdrv_co_writev             = replication_co_writev,
 770
 771     .is_filter                  = true,
 772
 773     .has_variable_length        = true,
 774     .strong_runtime_opts        = replication_strong_runtime_opts,
 775 };
 776
 777 static void bdrv_replication_init(void)
 778 {
 779     bdrv_register(&bdrv_replication);
 780 }
 781
 782 block_init(bdrv_replication_init);