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     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     QemuOpts *opts = NULL;
  89     const char *mode;
  90     const char *top_id;
  91
  92     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
  93                                BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
  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     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
 102         goto fail;
 103     }
 104
 105     mode = qemu_opt_get(opts, REPLICATION_MODE);
 106     if (!mode) {
 107         error_setg(errp, "Missing the option mode");
 108         goto fail;
 109     }
 110
 111     if (!strcmp(mode, "primary")) {
 112         s->mode = REPLICATION_MODE_PRIMARY;
 113         top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
 114         if (top_id) {
 115             error_setg(errp,
 116                        "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(errp, "Missing the option top-id");
 125             goto fail;
 126         }
 127     } else {
 128         error_setg(errp,
 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     return ret;
 140 }
 141
 142 static void replication_close(BlockDriverState *bs)
 143 {
 144     BDRVReplicationState *s = bs->opaque;
 145     Job *commit_job;
 146
 147     if (s->stage == BLOCK_REPLICATION_RUNNING) {
 148         replication_stop(s->rs, false, NULL);
 149     }
 150     if (s->stage == BLOCK_REPLICATION_FAILOVER) {
 151         commit_job = &s->commit_job->job;
 152         assert(commit_job->aio_context == qemu_get_current_aio_context());
 153         job_cancel_sync(commit_job);
 154     }
 155
 156     if (s->mode == REPLICATION_MODE_SECONDARY) {
 157         g_free(s->top_id);
 158     }
 159
 160     replication_remove(s->rs);
 161 }
 162
 163 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
 164                                    BdrvChildRole role,
 165                                    BlockReopenQueue *reopen_queue,
 166                                    uint64_t perm, uint64_t shared,
 167                                    uint64_t *nperm, uint64_t *nshared)
 168 {
 169     *nperm = BLK_PERM_CONSISTENT_READ;
 170     if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) {
 171         *nperm |= BLK_PERM_WRITE;
 172     }
 173     *nshared = BLK_PERM_CONSISTENT_READ
 174                | BLK_PERM_WRITE
 175                | BLK_PERM_WRITE_UNCHANGED;
 176     return;
 177 }
 178
 179 static int64_t replication_getlength(BlockDriverState *bs)
 180 {
 181     return bdrv_getlength(bs->file->bs);
 182 }
 183
 184 static int replication_get_io_status(BDRVReplicationState *s)
 185 {
 186     switch (s->stage) {
 187     case BLOCK_REPLICATION_NONE:
 188         return -EIO;
 189     case BLOCK_REPLICATION_RUNNING:
 190         return 0;
 191     case BLOCK_REPLICATION_FAILOVER:
 192         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 193     case BLOCK_REPLICATION_FAILOVER_FAILED:
 194         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
 195     case BLOCK_REPLICATION_DONE:
 196         /*
 197          * active commit job completes, and active disk and secondary_disk
 198          * is swapped, so we can operate bs->file directly
 199          */
 200         return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
 201     default:
 202         abort();
 203     }
 204 }
 205
 206 static int replication_return_value(BDRVReplicationState *s, int ret)
 207 {
 208     if (s->mode == REPLICATION_MODE_SECONDARY) {
 209         return ret;
 210     }
 211
 212     if (ret < 0) {
 213         s->error = ret;
 214         ret = 0;
 215     }
 216
 217     return ret;
 218 }
 219
 220 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
 221                                              int64_t sector_num,
 222                                              int remaining_sectors,
 223                                              QEMUIOVector *qiov)
 224 {
 225     BDRVReplicationState *s = bs->opaque;
 226     int ret;
 227
 228     if (s->mode == REPLICATION_MODE_PRIMARY) {
 229         /* We only use it to forward primary write requests */
 230         return -EIO;
 231     }
 232
 233     ret = replication_get_io_status(s);
 234     if (ret < 0) {
 235         return ret;
 236     }
 237
 238     ret = bdrv_co_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
 239                          remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 240
 241     return replication_return_value(s, ret);
 242 }
 243
 244 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
 245                                               int64_t sector_num,
 246                                               int remaining_sectors,
 247                                               QEMUIOVector *qiov,
 248                                               int flags)
 249 {
 250     BDRVReplicationState *s = bs->opaque;
 251     QEMUIOVector hd_qiov;
 252     uint64_t bytes_done = 0;
 253     BdrvChild *top = bs->file;
 254     BdrvChild *base = s->secondary_disk;
 255     BdrvChild *target;
 256     int ret;
 257     int64_t n;
 258
 259     assert(!flags);
 260     ret = replication_get_io_status(s);
 261     if (ret < 0) {
 262         goto out;
 263     }
 264
 265     if (ret == 0) {
 266         ret = bdrv_co_pwritev(top, sector_num * BDRV_SECTOR_SIZE,
 267                               remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
 268         return replication_return_value(s, ret);
 269     }
 270
 271     /*
 272      * Failover failed, only write to active disk if the sectors
 273      * have already been allocated in active disk/hidden disk.
 274      */
 275     qemu_iovec_init(&hd_qiov, qiov->niov);
 276     while (remaining_sectors > 0) {
 277         int64_t count;
 278
 279         ret = bdrv_is_allocated_above(top->bs, base->bs, false,
 280                                       sector_num * BDRV_SECTOR_SIZE,
 281                                       remaining_sectors * BDRV_SECTOR_SIZE,
 282                                       &count);
 283         if (ret < 0) {
 284             goto out1;
 285         }
 286
 287         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
 288         n = count >> BDRV_SECTOR_BITS;
 289         qemu_iovec_reset(&hd_qiov);
 290         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
 291
 292         target = ret ? top : base;
 293         ret = bdrv_co_pwritev(target, sector_num * BDRV_SECTOR_SIZE,
 294                               n * BDRV_SECTOR_SIZE, &hd_qiov, 0);
 295         if (ret < 0) {
 296             goto out1;
 297         }
 298
 299         remaining_sectors -= n;
 300         sector_num += n;
 301         bytes_done += count;
 302     }
 303
 304 out1:
 305     qemu_iovec_destroy(&hd_qiov);
 306 out:
 307     return ret;
 308 }
 309
 310 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
 311 {
 312     Error *local_err = NULL;
 313     int ret;
 314
 315     if (!s->backup_job) {
 316         error_setg(errp, "Backup job was cancelled unexpectedly");
 317         return;
 318     }
 319
 320     backup_do_checkpoint(s->backup_job, &local_err);
 321     if (local_err) {
 322         error_propagate(errp, local_err);
 323         return;
 324     }
 325
 326     if (!s->active_disk->bs->drv) {
 327         error_setg(errp, "Active disk %s is ejected",
 328                    s->active_disk->bs->node_name);
 329         return;
 330     }
 331
 332     ret = bdrv_make_empty(s->active_disk, errp);
 333     if (ret < 0) {
 334         return;
 335     }
 336
 337     if (!s->hidden_disk->bs->drv) {
 338         error_setg(errp, "Hidden disk %s is ejected",
 339                    s->hidden_disk->bs->node_name);
 340         return;
 341     }
 342
 343     BlockBackend *blk = blk_new(qemu_get_current_aio_context(),
 344                                 BLK_PERM_WRITE, BLK_PERM_ALL);
 345     blk_insert_bs(blk, s->hidden_disk->bs, &local_err);
 346     if (local_err) {
 347         error_propagate(errp, local_err);
 348         blk_unref(blk);
 349         return;
 350     }
 351
 352     ret = blk_make_empty(blk, errp);
 353     blk_unref(blk);
 354     if (ret < 0) {
 355         return;
 356     }
 357 }
 358
 359 /* This function is supposed to be called twice:
 360  * first with writable = true, then with writable = false.
 361  * The first call puts s->hidden_disk and s->secondary_disk in
 362  * r/w mode, and the second puts them back in their original state.
 363  */
 364 static void reopen_backing_file(BlockDriverState *bs, bool writable,
 365                                 Error **errp)
 366 {
 367     BDRVReplicationState *s = bs->opaque;
 368     BlockReopenQueue *reopen_queue = NULL;
 369
 370     if (writable) {
 371         s->orig_hidden_read_only = bdrv_is_read_only(s->hidden_disk->bs);
 372         s->orig_secondary_read_only = bdrv_is_read_only(s->secondary_disk->bs);
 373     }
 374
 375     bdrv_subtree_drained_begin(s->hidden_disk->bs);
 376     bdrv_subtree_drained_begin(s->secondary_disk->bs);
 377
 378     if (s->orig_hidden_read_only) {
 379         QDict *opts = qdict_new();
 380         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 381         reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs,
 382                                          opts, true);
 383     }
 384
 385     if (s->orig_secondary_read_only) {
 386         QDict *opts = qdict_new();
 387         qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
 388         reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
 389                                          opts, true);
 390     }
 391
 392     if (reopen_queue) {
 393         AioContext *ctx = bdrv_get_aio_context(bs);
 394         if (ctx != qemu_get_aio_context()) {
 395             aio_context_release(ctx);
 396         }
 397         bdrv_reopen_multiple(reopen_queue, errp);
 398         if (ctx != qemu_get_aio_context()) {
 399             aio_context_acquire(ctx);
 400         }
 401     }
 402
 403     bdrv_subtree_drained_end(s->hidden_disk->bs);
 404     bdrv_subtree_drained_end(s->secondary_disk->bs);
 405 }
 406
 407 static void backup_job_cleanup(BlockDriverState *bs)
 408 {
 409     BDRVReplicationState *s = bs->opaque;
 410     BlockDriverState *top_bs;
 411
 412     s->backup_job = NULL;
 413
 414     top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 415     if (!top_bs) {
 416         return;
 417     }
 418     bdrv_op_unblock_all(top_bs, s->blocker);
 419     error_free(s->blocker);
 420     reopen_backing_file(bs, false, NULL);
 421 }
 422
 423 static void backup_job_completed(void *opaque, int ret)
 424 {
 425     BlockDriverState *bs = opaque;
 426     BDRVReplicationState *s = bs->opaque;
 427
 428     if (s->stage != BLOCK_REPLICATION_FAILOVER) {
 429         /* The backup job is cancelled unexpectedly */
 430         s->error = -EIO;
 431     }
 432
 433     backup_job_cleanup(bs);
 434 }
 435
 436 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
 437 {
 438     BdrvChild *child;
 439
 440     /* The bs itself is the top_bs */
 441     if (top_bs == bs) {
 442         return true;
 443     }
 444
 445     /* Iterate over top_bs's children */
 446     QLIST_FOREACH(child, &top_bs->children, next) {
 447         if (child->bs == bs || check_top_bs(child->bs, bs)) {
 448             return true;
 449         }
 450     }
 451
 452     return false;
 453 }
 454
 455 static void replication_start(ReplicationState *rs, ReplicationMode mode,
 456                               Error **errp)
 457 {
 458     BlockDriverState *bs = rs->opaque;
 459     BDRVReplicationState *s;
 460     BlockDriverState *top_bs;
 461     int64_t active_length, hidden_length, disk_length;
 462     AioContext *aio_context;
 463     Error *local_err = NULL;
 464     BackupPerf perf = { .use_copy_range = true, .max_workers = 1 };
 465
 466     aio_context = bdrv_get_aio_context(bs);
 467     aio_context_acquire(aio_context);
 468     s = bs->opaque;
 469
 470     if (s->stage == BLOCK_REPLICATION_DONE ||
 471         s->stage == BLOCK_REPLICATION_FAILOVER) {
 472         /*
 473          * This case happens when a secondary is promoted to primary.
 474          * Ignore the request because the secondary side of replication
 475          * doesn't have to do anything anymore.
 476          */
 477         aio_context_release(aio_context);
 478         return;
 479     }
 480
 481     if (s->stage != BLOCK_REPLICATION_NONE) {
 482         error_setg(errp, "Block replication is running or done");
 483         aio_context_release(aio_context);
 484         return;
 485     }
 486
 487     if (s->mode != mode) {
 488         error_setg(errp, "The parameter mode's value is invalid, needs %d,"
 489                    " but got %d", s->mode, mode);
 490         aio_context_release(aio_context);
 491         return;
 492     }
 493
 494     switch (s->mode) {
 495     case REPLICATION_MODE_PRIMARY:
 496         break;
 497     case REPLICATION_MODE_SECONDARY:
 498         s->active_disk = bs->file;
 499         if (!s->active_disk || !s->active_disk->bs ||
 500                                     !s->active_disk->bs->backing) {
 501             error_setg(errp, "Active disk doesn't have backing file");
 502             aio_context_release(aio_context);
 503             return;
 504         }
 505
 506         s->hidden_disk = s->active_disk->bs->backing;
 507         if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
 508             error_setg(errp, "Hidden disk doesn't have backing file");
 509             aio_context_release(aio_context);
 510             return;
 511         }
 512
 513         s->secondary_disk = s->hidden_disk->bs->backing;
 514         if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
 515             error_setg(errp, "The secondary disk doesn't have block backend");
 516             aio_context_release(aio_context);
 517             return;
 518         }
 519
 520         /* verify the length */
 521         active_length = bdrv_getlength(s->active_disk->bs);
 522         hidden_length = bdrv_getlength(s->hidden_disk->bs);
 523         disk_length = bdrv_getlength(s->secondary_disk->bs);
 524         if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
 525             active_length != hidden_length || hidden_length != disk_length) {
 526             error_setg(errp, "Active disk, hidden disk, secondary disk's length"
 527                        " are not the same");
 528             aio_context_release(aio_context);
 529             return;
 530         }
 531
 532         /* Must be true, or the bdrv_getlength() calls would have failed */
 533         assert(s->active_disk->bs->drv && s->hidden_disk->bs->drv);
 534
 535         if (!s->active_disk->bs->drv->bdrv_make_empty ||
 536             !s->hidden_disk->bs->drv->bdrv_make_empty) {
 537             error_setg(errp,
 538                        "Active disk or hidden disk doesn't support make_empty");
 539             aio_context_release(aio_context);
 540             return;
 541         }
 542
 543         /* reopen the backing file in r/w mode */
 544         reopen_backing_file(bs, true, &local_err);
 545         if (local_err) {
 546             error_propagate(errp, local_err);
 547             aio_context_release(aio_context);
 548             return;
 549         }
 550
 551         /* start backup job now */
 552         error_setg(&s->blocker,
 553                    "Block device is in use by internal backup job");
 554
 555         top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
 556         if (!top_bs || !bdrv_is_root_node(top_bs) ||
 557             !check_top_bs(top_bs, bs)) {
 558             error_setg(errp, "No top_bs or it is invalid");
 559             reopen_backing_file(bs, false, NULL);
 560             aio_context_release(aio_context);
 561             return;
 562         }
 563         bdrv_op_block_all(top_bs, s->blocker);
 564         bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
 565
 566         s->backup_job = backup_job_create(
 567                                 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
 568                                 0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
 569                                 &perf,
 570                                 BLOCKDEV_ON_ERROR_REPORT,
 571                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
 572                                 backup_job_completed, bs, NULL, &local_err);
 573         if (local_err) {
 574             error_propagate(errp, local_err);
 575             backup_job_cleanup(bs);
 576             aio_context_release(aio_context);
 577             return;
 578         }
 579         job_start(&s->backup_job->job);
 580         break;
 581     default:
 582         aio_context_release(aio_context);
 583         abort();
 584     }
 585
 586     s->stage = BLOCK_REPLICATION_RUNNING;
 587
 588     if (s->mode == REPLICATION_MODE_SECONDARY) {
 589         secondary_do_checkpoint(s, errp);
 590     }
 591
 592     s->error = 0;
 593     aio_context_release(aio_context);
 594 }
 595
 596 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
 597 {
 598     BlockDriverState *bs = rs->opaque;
 599     BDRVReplicationState *s;
 600     AioContext *aio_context;
 601
 602     aio_context = bdrv_get_aio_context(bs);
 603     aio_context_acquire(aio_context);
 604     s = bs->opaque;
 605
 606     if (s->stage == BLOCK_REPLICATION_DONE ||
 607         s->stage == BLOCK_REPLICATION_FAILOVER) {
 608         /*
 609          * This case happens when a secondary was promoted to primary.
 610          * Ignore the request because the secondary side of replication
 611          * doesn't have to do anything anymore.
 612          */
 613         aio_context_release(aio_context);
 614         return;
 615     }
 616
 617     if (s->mode == REPLICATION_MODE_SECONDARY) {
 618         secondary_do_checkpoint(s, errp);
 619     }
 620     aio_context_release(aio_context);
 621 }
 622
 623 static void replication_get_error(ReplicationState *rs, Error **errp)
 624 {
 625     BlockDriverState *bs = rs->opaque;
 626     BDRVReplicationState *s;
 627     AioContext *aio_context;
 628
 629     aio_context = bdrv_get_aio_context(bs);
 630     aio_context_acquire(aio_context);
 631     s = bs->opaque;
 632
 633     if (s->stage == BLOCK_REPLICATION_NONE) {
 634         error_setg(errp, "Block replication is not running");
 635         aio_context_release(aio_context);
 636         return;
 637     }
 638
 639     if (s->error) {
 640         error_setg(errp, "I/O error occurred");
 641         aio_context_release(aio_context);
 642         return;
 643     }
 644     aio_context_release(aio_context);
 645 }
 646
 647 static void replication_done(void *opaque, int ret)
 648 {
 649     BlockDriverState *bs = opaque;
 650     BDRVReplicationState *s = bs->opaque;
 651
 652     if (ret == 0) {
 653         s->stage = BLOCK_REPLICATION_DONE;
 654
 655         s->active_disk = NULL;
 656         s->secondary_disk = NULL;
 657         s->hidden_disk = NULL;
 658         s->error = 0;
 659     } else {
 660         s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
 661         s->error = -EIO;
 662     }
 663 }
 664
 665 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
 666 {
 667     BlockDriverState *bs = rs->opaque;
 668     BDRVReplicationState *s;
 669     AioContext *aio_context;
 670
 671     aio_context = bdrv_get_aio_context(bs);
 672     aio_context_acquire(aio_context);
 673     s = bs->opaque;
 674
 675     if (s->stage == BLOCK_REPLICATION_DONE ||
 676         s->stage == BLOCK_REPLICATION_FAILOVER) {
 677         /*
 678          * This case happens when a secondary was promoted to primary.
 679          * Ignore the request because the secondary side of replication
 680          * doesn't have to do anything anymore.
 681          */
 682         aio_context_release(aio_context);
 683         return;
 684     }
 685
 686     if (s->stage != BLOCK_REPLICATION_RUNNING) {
 687         error_setg(errp, "Block replication is not running");
 688         aio_context_release(aio_context);
 689         return;
 690     }
 691
 692     switch (s->mode) {
 693     case REPLICATION_MODE_PRIMARY:
 694         s->stage = BLOCK_REPLICATION_DONE;
 695         s->error = 0;
 696         break;
 697     case REPLICATION_MODE_SECONDARY:
 698         /*
 699          * This BDS will be closed, and the job should be completed
 700          * before the BDS is closed, because we will access hidden
 701          * disk, secondary disk in backup_job_completed().
 702          */
 703         if (s->backup_job) {
 704             job_cancel_sync(&s->backup_job->job);
 705         }
 706
 707         if (!failover) {
 708             secondary_do_checkpoint(s, errp);
 709             s->stage = BLOCK_REPLICATION_DONE;
 710             aio_context_release(aio_context);
 711             return;
 712         }
 713
 714         s->stage = BLOCK_REPLICATION_FAILOVER;
 715         s->commit_job = commit_active_start(
 716                             NULL, s->active_disk->bs, s->secondary_disk->bs,
 717                             JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
 718                             NULL, replication_done, bs, true, errp);
 719         break;
 720     default:
 721         aio_context_release(aio_context);
 722         abort();
 723     }
 724     aio_context_release(aio_context);
 725 }
 726
 727 static const char *const replication_strong_runtime_opts[] = {
 728     REPLICATION_MODE,
 729     REPLICATION_TOP_ID,
 730
 731     NULL
 732 };
 733
 734 static BlockDriver bdrv_replication = {
 735     .format_name                = "replication",
 736     .instance_size              = sizeof(BDRVReplicationState),
 737
 738     .bdrv_open                  = replication_open,
 739     .bdrv_close                 = replication_close,
 740     .bdrv_child_perm            = replication_child_perm,
 741
 742     .bdrv_getlength             = replication_getlength,
 743     .bdrv_co_readv              = replication_co_readv,
 744     .bdrv_co_writev             = replication_co_writev,
 745
 746     .is_filter                  = true,
 747
 748     .has_variable_length        = true,
 749     .strong_runtime_opts        = replication_strong_runtime_opts,
 750 };
 751
 752 static void bdrv_replication_init(void)
 753 {
 754     bdrv_register(&bdrv_replication);
 755 }
 756
 757 block_init(bdrv_replication_init);