block/mirror.c

   1 /*
   2  * Image mirroring
   3  *
   4  * Copyright Red Hat, Inc. 2012
   5  *
   6  * Authors:
   7  *  Paolo Bonzini  <pbonzini@redhat.com>
   8  *
   9  * This work is licensed under the terms of the GNU LGPL, version 2 or later.
  10  * See the COPYING.LIB file in the top-level directory.
  11  *
  12  */
  13
  14 #include "trace.h"
  15 #include "block/blockjob.h"
  16 #include "block/block_int.h"
  17 #include "qemu/ratelimit.h"
  18 #include "qemu/bitmap.h"
  19
  20 #define SLICE_TIME    100000000ULL /* ns */
  21 #define MAX_IN_FLIGHT 16
  22
  23 /* The mirroring buffer is a list of granularity-sized chunks.
  24  * Free chunks are organized in a list.
  25  */
  26 typedef struct MirrorBuffer {
  27     QSIMPLEQ_ENTRY(MirrorBuffer) next;
  28 } MirrorBuffer;
  29
  30 typedef struct MirrorBlockJob {
  31     BlockJob common;
  32     RateLimit limit;
  33     BlockDriverState *target;
  34     BlockDriverState *base;
  35     bool is_none_mode;
  36     BlockdevOnError on_source_error, on_target_error;
  37     bool synced;
  38     bool should_complete;
  39     int64_t sector_num;
  40     int64_t granularity;
  41     size_t buf_size;
  42     unsigned long *cow_bitmap;
  43     BdrvDirtyBitmap *dirty_bitmap;
  44     HBitmapIter hbi;
  45     uint8_t *buf;
  46     QSIMPLEQ_HEAD(, MirrorBuffer) buf_free;
  47     int buf_free_count;
  48
  49     unsigned long *in_flight_bitmap;
  50     int in_flight;
  51     int ret;
  52 } MirrorBlockJob;
  53
  54 typedef struct MirrorOp {
  55     MirrorBlockJob *s;
  56     QEMUIOVector qiov;
  57     int64_t sector_num;
  58     int nb_sectors;
  59 } MirrorOp;
  60
  61 static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read,
  62                                             int error)
  63 {
  64     s->synced = false;
  65     if (read) {
  66         return block_job_error_action(&s->common, s->common.bs,
  67                                       s->on_source_error, true, error);
  68     } else {
  69         return block_job_error_action(&s->common, s->target,
  70                                       s->on_target_error, false, error);
  71     }
  72 }
  73
  74 static void mirror_iteration_done(MirrorOp *op, int ret)
  75 {
  76     MirrorBlockJob *s = op->s;
  77     struct iovec *iov;
  78     int64_t chunk_num;
  79     int i, nb_chunks, sectors_per_chunk;
  80
  81     trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret);
  82
  83     s->in_flight--;
  84     iov = op->qiov.iov;
  85     for (i = 0; i < op->qiov.niov; i++) {
  86         MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base;
  87         QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next);
  88         s->buf_free_count++;
  89     }
  90
  91     sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
  92     chunk_num = op->sector_num / sectors_per_chunk;
  93     nb_chunks = op->nb_sectors / sectors_per_chunk;
  94     bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks);
  95     if (s->cow_bitmap && ret >= 0) {
  96         bitmap_set(s->cow_bitmap, chunk_num, nb_chunks);
  97     }
  98
  99     qemu_iovec_destroy(&op->qiov);
 100     g_slice_free(MirrorOp, op);
 101
 102     /* Enter coroutine when it is not sleeping.  The coroutine sleeps to
 103      * rate-limit itself.  The coroutine will eventually resume since there is
 104      * a sleep timeout so don't wake it early.
 105      */
 106     if (s->common.busy) {
 107         qemu_coroutine_enter(s->common.co, NULL);
 108     }
 109 }
 110
 111 static void mirror_write_complete(void *opaque, int ret)
 112 {
 113     MirrorOp *op = opaque;
 114     MirrorBlockJob *s = op->s;
 115     if (ret < 0) {
 116         BlockDriverState *source = s->common.bs;
 117         BlockErrorAction action;
 118
 119         bdrv_set_dirty(source, op->sector_num, op->nb_sectors);
 120         action = mirror_error_action(s, false, -ret);
 121         if (action == BDRV_ACTION_REPORT && s->ret >= 0) {
 122             s->ret = ret;
 123         }
 124     }
 125     mirror_iteration_done(op, ret);
 126 }
 127
 128 static void mirror_read_complete(void *opaque, int ret)
 129 {
 130     MirrorOp *op = opaque;
 131     MirrorBlockJob *s = op->s;
 132     if (ret < 0) {
 133         BlockDriverState *source = s->common.bs;
 134         BlockErrorAction action;
 135
 136         bdrv_set_dirty(source, op->sector_num, op->nb_sectors);
 137         action = mirror_error_action(s, true, -ret);
 138         if (action == BDRV_ACTION_REPORT && s->ret >= 0) {
 139             s->ret = ret;
 140         }
 141
 142         mirror_iteration_done(op, ret);
 143         return;
 144     }
 145     bdrv_aio_writev(s->target, op->sector_num, &op->qiov, op->nb_sectors,
 146                     mirror_write_complete, op);
 147 }
 148
 149 static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)
 150 {
 151     BlockDriverState *source = s->common.bs;
 152     int nb_sectors, sectors_per_chunk, nb_chunks;
 153     int64_t end, sector_num, next_chunk, next_sector, hbitmap_next_sector;
 154     uint64_t delay_ns;
 155     MirrorOp *op;
 156
 157     s->sector_num = hbitmap_iter_next(&s->hbi);
 158     if (s->sector_num < 0) {
 159         bdrv_dirty_iter_init(source, s->dirty_bitmap, &s->hbi);
 160         s->sector_num = hbitmap_iter_next(&s->hbi);
 161         trace_mirror_restart_iter(s,
 162                                   bdrv_get_dirty_count(source, s->dirty_bitmap));
 163         assert(s->sector_num >= 0);
 164     }
 165
 166     hbitmap_next_sector = s->sector_num;
 167     sector_num = s->sector_num;
 168     sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
 169     end = s->common.len >> BDRV_SECTOR_BITS;
 170
 171     /* Extend the QEMUIOVector to include all adjacent blocks that will
 172      * be copied in this operation.
 173      *
 174      * We have to do this if we have no backing file yet in the destination,
 175      * and the cluster size is very large.  Then we need to do COW ourselves.
 176      * The first time a cluster is copied, copy it entirely.  Note that,
 177      * because both the granularity and the cluster size are powers of two,
 178      * the number of sectors to copy cannot exceed one cluster.
 179      *
 180      * We also want to extend the QEMUIOVector to include more adjacent
 181      * dirty blocks if possible, to limit the number of I/O operations and
 182      * run efficiently even with a small granularity.
 183      */
 184     nb_chunks = 0;
 185     nb_sectors = 0;
 186     next_sector = sector_num;
 187     next_chunk = sector_num / sectors_per_chunk;
 188
 189     /* Wait for I/O to this cluster (from a previous iteration) to be done.  */
 190     while (test_bit(next_chunk, s->in_flight_bitmap)) {
 191         trace_mirror_yield_in_flight(s, sector_num, s->in_flight);
 192         qemu_coroutine_yield();
 193     }
 194
 195     do {
 196         int added_sectors, added_chunks;
 197
 198         if (!bdrv_get_dirty(source, s->dirty_bitmap, next_sector) ||
 199             test_bit(next_chunk, s->in_flight_bitmap)) {
 200             assert(nb_sectors > 0);
 201             break;
 202         }
 203
 204         added_sectors = sectors_per_chunk;
 205         if (s->cow_bitmap && !test_bit(next_chunk, s->cow_bitmap)) {
 206             bdrv_round_to_clusters(s->target,
 207                                    next_sector, added_sectors,
 208                                    &next_sector, &added_sectors);
 209
 210             /* On the first iteration, the rounding may make us copy
 211              * sectors before the first dirty one.
 212              */
 213             if (next_sector < sector_num) {
 214                 assert(nb_sectors == 0);
 215                 sector_num = next_sector;
 216                 next_chunk = next_sector / sectors_per_chunk;
 217             }
 218         }
 219
 220         added_sectors = MIN(added_sectors, end - (sector_num + nb_sectors));
 221         added_chunks = (added_sectors + sectors_per_chunk - 1) / sectors_per_chunk;
 222
 223         /* When doing COW, it may happen that there is not enough space for
 224          * a full cluster.  Wait if that is the case.
 225          */
 226         while (nb_chunks == 0 && s->buf_free_count < added_chunks) {
 227             trace_mirror_yield_buf_busy(s, nb_chunks, s->in_flight);
 228             qemu_coroutine_yield();
 229         }
 230         if (s->buf_free_count < nb_chunks + added_chunks) {
 231             trace_mirror_break_buf_busy(s, nb_chunks, s->in_flight);
 232             break;
 233         }
 234
 235         /* We have enough free space to copy these sectors.  */
 236         bitmap_set(s->in_flight_bitmap, next_chunk, added_chunks);
 237
 238         nb_sectors += added_sectors;
 239         nb_chunks += added_chunks;
 240         next_sector += added_sectors;
 241         next_chunk += added_chunks;
 242         if (!s->synced && s->common.speed) {
 243             delay_ns = ratelimit_calculate_delay(&s->limit, added_sectors);
 244         } else {
 245             delay_ns = 0;
 246         }
 247     } while (delay_ns == 0 && next_sector < end);
 248
 249     /* Allocate a MirrorOp that is used as an AIO callback.  */
 250     op = g_slice_new(MirrorOp);
 251     op->s = s;
 252     op->sector_num = sector_num;
 253     op->nb_sectors = nb_sectors;
 254
 255     /* Now make a QEMUIOVector taking enough granularity-sized chunks
 256      * from s->buf_free.
 257      */
 258     qemu_iovec_init(&op->qiov, nb_chunks);
 259     next_sector = sector_num;
 260     while (nb_chunks-- > 0) {
 261         MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free);
 262         QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next);
 263         s->buf_free_count--;
 264         qemu_iovec_add(&op->qiov, buf, s->granularity);
 265
 266         /* Advance the HBitmapIter in parallel, so that we do not examine
 267          * the same sector twice.
 268          */
 269         if (next_sector > hbitmap_next_sector
 270             && bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) {
 271             hbitmap_next_sector = hbitmap_iter_next(&s->hbi);
 272         }
 273
 274         next_sector += sectors_per_chunk;
 275     }
 276
 277     bdrv_reset_dirty(source, sector_num, nb_sectors);
 278
 279     /* Copy the dirty cluster.  */
 280     s->in_flight++;
 281     trace_mirror_one_iteration(s, sector_num, nb_sectors);
 282     bdrv_aio_readv(source, sector_num, &op->qiov, nb_sectors,
 283                    mirror_read_complete, op);
 284     return delay_ns;
 285 }
 286
 287 static void mirror_free_init(MirrorBlockJob *s)
 288 {
 289     int granularity = s->granularity;
 290     size_t buf_size = s->buf_size;
 291     uint8_t *buf = s->buf;
 292
 293     assert(s->buf_free_count == 0);
 294     QSIMPLEQ_INIT(&s->buf_free);
 295     while (buf_size != 0) {
 296         MirrorBuffer *cur = (MirrorBuffer *)buf;
 297         QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next);
 298         s->buf_free_count++;
 299         buf_size -= granularity;
 300         buf += granularity;
 301     }
 302 }
 303
 304 static void mirror_drain(MirrorBlockJob *s)
 305 {
 306     while (s->in_flight > 0) {
 307         qemu_coroutine_yield();
 308     }
 309 }
 310
 311 static void coroutine_fn mirror_run(void *opaque)
 312 {
 313     MirrorBlockJob *s = opaque;
 314     BlockDriverState *bs = s->common.bs;
 315     int64_t sector_num, end, sectors_per_chunk, length;
 316     uint64_t last_pause_ns;
 317     BlockDriverInfo bdi;
 318     char backing_filename[1024];
 319     int ret = 0;
 320     int n;
 321
 322     if (block_job_is_cancelled(&s->common)) {
 323         goto immediate_exit;
 324     }
 325
 326     s->common.len = bdrv_getlength(bs);
 327     if (s->common.len <= 0) {
 328         ret = s->common.len;
 329         goto immediate_exit;
 330     }
 331
 332     length = DIV_ROUND_UP(s->common.len, s->granularity);
 333     s->in_flight_bitmap = bitmap_new(length);
 334
 335     /* If we have no backing file yet in the destination, we cannot let
 336      * the destination do COW.  Instead, we copy sectors around the
 337      * dirty data if needed.  We need a bitmap to do that.
 338      */
 339     bdrv_get_backing_filename(s->target, backing_filename,
 340                               sizeof(backing_filename));
 341     if (backing_filename[0] && !s->target->backing_hd) {
 342         ret = bdrv_get_info(s->target, &bdi);
 343         if (ret < 0) {
 344             goto immediate_exit;
 345         }
 346         if (s->granularity < bdi.cluster_size) {
 347             s->buf_size = MAX(s->buf_size, bdi.cluster_size);
 348             s->cow_bitmap = bitmap_new(length);
 349         }
 350     }
 351
 352     end = s->common.len >> BDRV_SECTOR_BITS;
 353     s->buf = qemu_blockalign(bs, s->buf_size);
 354     sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
 355     mirror_free_init(s);
 356
 357     if (!s->is_none_mode) {
 358         /* First part, loop on the sectors and initialize the dirty bitmap.  */
 359         BlockDriverState *base = s->base;
 360         for (sector_num = 0; sector_num < end; ) {
 361             int64_t next = (sector_num | (sectors_per_chunk - 1)) + 1;
 362             ret = bdrv_is_allocated_above(bs, base,
 363                                           sector_num, next - sector_num, &n);
 364
 365             if (ret < 0) {
 366                 goto immediate_exit;
 367             }
 368
 369             assert(n > 0);
 370             if (ret == 1) {
 371                 bdrv_set_dirty(bs, sector_num, n);
 372                 sector_num = next;
 373             } else {
 374                 sector_num += n;
 375             }
 376         }
 377     }
 378
 379     bdrv_dirty_iter_init(bs, s->dirty_bitmap, &s->hbi);
 380     last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
 381     for (;;) {
 382         uint64_t delay_ns = 0;
 383         int64_t cnt;
 384         bool should_complete;
 385
 386         if (s->ret < 0) {
 387             ret = s->ret;
 388             goto immediate_exit;
 389         }
 390
 391         cnt = bdrv_get_dirty_count(bs, s->dirty_bitmap);
 392
 393         /* Note that even when no rate limit is applied we need to yield
 394          * periodically with no pending I/O so that qemu_aio_flush() returns.
 395          * We do so every SLICE_TIME nanoseconds, or when there is an error,
 396          * or when the source is clean, whichever comes first.
 397          */
 398         if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - last_pause_ns < SLICE_TIME &&
 399             s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 400             if (s->in_flight == MAX_IN_FLIGHT || s->buf_free_count == 0 ||
 401                 (cnt == 0 && s->in_flight > 0)) {
 402                 trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt);
 403                 qemu_coroutine_yield();
 404                 continue;
 405             } else if (cnt != 0) {
 406                 delay_ns = mirror_iteration(s);
 407                 if (delay_ns == 0) {
 408                     continue;
 409                 }
 410             }
 411         }
 412
 413         should_complete = false;
 414         if (s->in_flight == 0 && cnt == 0) {
 415             trace_mirror_before_flush(s);
 416             ret = bdrv_flush(s->target);
 417             if (ret < 0) {
 418                 if (mirror_error_action(s, false, -ret) == BDRV_ACTION_REPORT) {
 419                     goto immediate_exit;
 420                 }
 421             } else {
 422                 /* We're out of the streaming phase.  From now on, if the job
 423                  * is cancelled we will actually complete all pending I/O and
 424                  * report completion.  This way, block-job-cancel will leave
 425                  * the target in a consistent state.
 426                  */
 427                 s->common.offset = end * BDRV_SECTOR_SIZE;
 428                 if (!s->synced) {
 429                     block_job_ready(&s->common);
 430                     s->synced = true;
 431                 }
 432
 433                 should_complete = s->should_complete ||
 434                     block_job_is_cancelled(&s->common);
 435                 cnt = bdrv_get_dirty_count(bs, s->dirty_bitmap);
 436             }
 437         }
 438
 439         if (cnt == 0 && should_complete) {
 440             /* The dirty bitmap is not updated while operations are pending.
 441              * If we're about to exit, wait for pending operations before
 442              * calling bdrv_get_dirty_count(bs), or we may exit while the
 443              * source has dirty data to copy!
 444              *
 445              * Note that I/O can be submitted by the guest while
 446              * mirror_populate runs.
 447              */
 448             trace_mirror_before_drain(s, cnt);
 449             bdrv_drain_all();
 450             cnt = bdrv_get_dirty_count(bs, s->dirty_bitmap);
 451         }
 452
 453         ret = 0;
 454         trace_mirror_before_sleep(s, cnt, s->synced, delay_ns);
 455         if (!s->synced) {
 456             /* Publish progress */
 457             s->common.offset = (end - cnt) * BDRV_SECTOR_SIZE;
 458             block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 459             if (block_job_is_cancelled(&s->common)) {
 460                 break;
 461             }
 462         } else if (!should_complete) {
 463             delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0);
 464             block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 465         } else if (cnt == 0) {
 466             /* The two disks are in sync.  Exit and report successful
 467              * completion.
 468              */
 469             assert(QLIST_EMPTY(&bs->tracked_requests));
 470             s->common.cancelled = false;
 471             break;
 472         }
 473         last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
 474     }
 475
 476 immediate_exit:
 477     if (s->in_flight > 0) {
 478         /* We get here only if something went wrong.  Either the job failed,
 479          * or it was cancelled prematurely so that we do not guarantee that
 480          * the target is a copy of the source.
 481          */
 482         assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common)));
 483         mirror_drain(s);
 484     }
 485
 486     assert(s->in_flight == 0);
 487     qemu_vfree(s->buf);
 488     g_free(s->cow_bitmap);
 489     g_free(s->in_flight_bitmap);
 490     bdrv_release_dirty_bitmap(bs, s->dirty_bitmap);
 491     bdrv_iostatus_disable(s->target);
 492     if (s->should_complete && ret == 0) {
 493         if (bdrv_get_flags(s->target) != bdrv_get_flags(s->common.bs)) {
 494             bdrv_reopen(s->target, bdrv_get_flags(s->common.bs), NULL);
 495         }
 496         bdrv_swap(s->target, s->common.bs);
 497         if (s->common.driver->job_type == BLOCK_JOB_TYPE_COMMIT) {
 498             /* drop the bs loop chain formed by the swap: break the loop then
 499              * trigger the unref from the top one */
 500             BlockDriverState *p = s->base->backing_hd;
 501             bdrv_set_backing_hd(s->base, NULL);
 502             bdrv_unref(p);
 503         }
 504     }
 505     bdrv_unref(s->target);
 506     block_job_completed(&s->common, ret);
 507 }
 508
 509 static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp)
 510 {
 511     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
 512
 513     if (speed < 0) {
 514         error_set(errp, QERR_INVALID_PARAMETER, "speed");
 515         return;
 516     }
 517     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 518 }
 519
 520 static void mirror_iostatus_reset(BlockJob *job)
 521 {
 522     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
 523
 524     bdrv_iostatus_reset(s->target);
 525 }
 526
 527 static void mirror_complete(BlockJob *job, Error **errp)
 528 {
 529     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
 530     Error *local_err = NULL;
 531     int ret;
 532
 533     ret = bdrv_open_backing_file(s->target, NULL, &local_err);
 534     if (ret < 0) {
 535         error_propagate(errp, local_err);
 536         return;
 537     }
 538     if (!s->synced) {
 539         error_set(errp, QERR_BLOCK_JOB_NOT_READY, job->bs->device_name);
 540         return;
 541     }
 542
 543     s->should_complete = true;
 544     block_job_resume(job);
 545 }
 546
 547 static const BlockJobDriver mirror_job_driver = {
 548     .instance_size = sizeof(MirrorBlockJob),
 549     .job_type      = BLOCK_JOB_TYPE_MIRROR,
 550     .set_speed     = mirror_set_speed,
 551     .iostatus_reset= mirror_iostatus_reset,
 552     .complete      = mirror_complete,
 553 };
 554
 555 static const BlockJobDriver commit_active_job_driver = {
 556     .instance_size = sizeof(MirrorBlockJob),
 557     .job_type      = BLOCK_JOB_TYPE_COMMIT,
 558     .set_speed     = mirror_set_speed,
 559     .iostatus_reset
 560                    = mirror_iostatus_reset,
 561     .complete      = mirror_complete,
 562 };
 563
 564 static void mirror_start_job(BlockDriverState *bs, BlockDriverState *target,
 565                             int64_t speed, int64_t granularity,
 566                             int64_t buf_size,
 567                             BlockdevOnError on_source_error,
 568                             BlockdevOnError on_target_error,
 569                             BlockDriverCompletionFunc *cb,
 570                             void *opaque, Error **errp,
 571                             const BlockJobDriver *driver,
 572                             bool is_none_mode, BlockDriverState *base)
 573 {
 574     MirrorBlockJob *s;
 575
 576     if (granularity == 0) {
 577         /* Choose the default granularity based on the target file's cluster
 578          * size, clamped between 4k and 64k.  */
 579         BlockDriverInfo bdi;
 580         if (bdrv_get_info(target, &bdi) >= 0 && bdi.cluster_size != 0) {
 581             granularity = MAX(4096, bdi.cluster_size);
 582             granularity = MIN(65536, granularity);
 583         } else {
 584             granularity = 65536;
 585         }
 586     }
 587
 588     assert ((granularity & (granularity - 1)) == 0);
 589
 590     if ((on_source_error == BLOCKDEV_ON_ERROR_STOP ||
 591          on_source_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
 592         !bdrv_iostatus_is_enabled(bs)) {
 593         error_set(errp, QERR_INVALID_PARAMETER, "on-source-error");
 594         return;
 595     }
 596
 597
 598     s = block_job_create(driver, bs, speed, cb, opaque, errp);
 599     if (!s) {
 600         return;
 601     }
 602
 603     s->on_source_error = on_source_error;
 604     s->on_target_error = on_target_error;
 605     s->target = target;
 606     s->is_none_mode = is_none_mode;
 607     s->base = base;
 608     s->granularity = granularity;
 609     s->buf_size = MAX(buf_size, granularity);
 610
 611     s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, errp);
 612     if (!s->dirty_bitmap) {
 613         return;
 614     }
 615     bdrv_set_enable_write_cache(s->target, true);
 616     bdrv_set_on_error(s->target, on_target_error, on_target_error);
 617     bdrv_iostatus_enable(s->target);
 618     s->common.co = qemu_coroutine_create(mirror_run);
 619     trace_mirror_start(bs, s, s->common.co, opaque);
 620     qemu_coroutine_enter(s->common.co, s);
 621 }
 622
 623 void mirror_start(BlockDriverState *bs, BlockDriverState *target,
 624                   int64_t speed, int64_t granularity, int64_t buf_size,
 625                   MirrorSyncMode mode, BlockdevOnError on_source_error,
 626                   BlockdevOnError on_target_error,
 627                   BlockDriverCompletionFunc *cb,
 628                   void *opaque, Error **errp)
 629 {
 630     bool is_none_mode;
 631     BlockDriverState *base;
 632
 633     is_none_mode = mode == MIRROR_SYNC_MODE_NONE;
 634     base = mode == MIRROR_SYNC_MODE_TOP ? bs->backing_hd : NULL;
 635     mirror_start_job(bs, target, speed, granularity, buf_size,
 636                      on_source_error, on_target_error, cb, opaque, errp,
 637                      &mirror_job_driver, is_none_mode, base);
 638 }
 639
 640 void commit_active_start(BlockDriverState *bs, BlockDriverState *base,
 641                          int64_t speed,
 642                          BlockdevOnError on_error,
 643                          BlockDriverCompletionFunc *cb,
 644                          void *opaque, Error **errp)
 645 {
 646     int64_t length, base_length;
 647     int orig_base_flags;
 648     int ret;
 649     Error *local_err = NULL;
 650
 651     orig_base_flags = bdrv_get_flags(base);
 652
 653     if (bdrv_reopen(base, bs->open_flags, errp)) {
 654         return;
 655     }
 656
 657     length = bdrv_getlength(bs);
 658     if (length < 0) {
 659         error_setg_errno(errp, -length,
 660                          "Unable to determine length of %s", bs->filename);
 661         goto error_restore_flags;
 662     }
 663
 664     base_length = bdrv_getlength(base);
 665     if (base_length < 0) {
 666         error_setg_errno(errp, -base_length,
 667                          "Unable to determine length of %s", base->filename);
 668         goto error_restore_flags;
 669     }
 670
 671     if (length > base_length) {
 672         ret = bdrv_truncate(base, length);
 673         if (ret < 0) {
 674             error_setg_errno(errp, -ret,
 675                             "Top image %s is larger than base image %s, and "
 676                              "resize of base image failed",
 677                              bs->filename, base->filename);
 678             goto error_restore_flags;
 679         }
 680     }
 681
 682     bdrv_ref(base);
 683     mirror_start_job(bs, base, speed, 0, 0,
 684                      on_error, on_error, cb, opaque, &local_err,
 685                      &commit_active_job_driver, false, base);
 686     if (local_err) {
 687         error_propagate(errp, local_err);
 688         goto error_restore_flags;
 689     }
 690
 691     return;
 692
 693 error_restore_flags:
 694     /* ignore error and errp for bdrv_reopen, because we want to propagate
 695      * the original error */
 696     bdrv_reopen(base, orig_base_flags, NULL);
 697     return;
 698 }