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     /* The name of the graph node to replace */
  36     char *replaces;
  37     /* The BDS to replace */
  38     BlockDriverState *to_replace;
  39     /* Used to block operations on the drive-mirror-replace target */
  40     Error *replace_blocker;
  41     bool is_none_mode;
  42     BlockdevOnError on_source_error, on_target_error;
  43     bool synced;
  44     bool should_complete;
  45     int64_t sector_num;
  46     int64_t granularity;
  47     size_t buf_size;
  48     int64_t bdev_length;
  49     unsigned long *cow_bitmap;
  50     BdrvDirtyBitmap *dirty_bitmap;
  51     HBitmapIter hbi;
  52     uint8_t *buf;
  53     QSIMPLEQ_HEAD(, MirrorBuffer) buf_free;
  54     int buf_free_count;
  55
  56     unsigned long *in_flight_bitmap;
  57     int in_flight;
  58     int sectors_in_flight;
  59     int ret;
  60 } MirrorBlockJob;
  61
  62 typedef struct MirrorOp {
  63     MirrorBlockJob *s;
  64     QEMUIOVector qiov;
  65     int64_t sector_num;
  66     int nb_sectors;
  67 } MirrorOp;
  68
  69 static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read,
  70                                             int error)
  71 {
  72     s->synced = false;
  73     if (read) {
  74         return block_job_error_action(&s->common, s->common.bs,
  75                                       s->on_source_error, true, error);
  76     } else {
  77         return block_job_error_action(&s->common, s->target,
  78                                       s->on_target_error, false, error);
  79     }
  80 }
  81
  82 static void mirror_iteration_done(MirrorOp *op, int ret)
  83 {
  84     MirrorBlockJob *s = op->s;
  85     struct iovec *iov;
  86     int64_t chunk_num;
  87     int i, nb_chunks, sectors_per_chunk;
  88
  89     trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret);
  90
  91     s->in_flight--;
  92     s->sectors_in_flight -= op->nb_sectors;
  93     iov = op->qiov.iov;
  94     for (i = 0; i < op->qiov.niov; i++) {
  95         MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base;
  96         QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next);
  97         s->buf_free_count++;
  98     }
  99
 100     sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
 101     chunk_num = op->sector_num / sectors_per_chunk;
 102     nb_chunks = op->nb_sectors / sectors_per_chunk;
 103     bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks);
 104     if (ret >= 0) {
 105         if (s->cow_bitmap) {
 106             bitmap_set(s->cow_bitmap, chunk_num, nb_chunks);
 107         }
 108         s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE;
 109     }
 110
 111     qemu_iovec_destroy(&op->qiov);
 112     g_slice_free(MirrorOp, op);
 113
 114     /* Enter coroutine when it is not sleeping.  The coroutine sleeps to
 115      * rate-limit itself.  The coroutine will eventually resume since there is
 116      * a sleep timeout so don't wake it early.
 117      */
 118     if (s->common.busy) {
 119         qemu_coroutine_enter(s->common.co, NULL);
 120     }
 121 }
 122
 123 static void mirror_write_complete(void *opaque, int ret)
 124 {
 125     MirrorOp *op = opaque;
 126     MirrorBlockJob *s = op->s;
 127     if (ret < 0) {
 128         BlockErrorAction action;
 129
 130         bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors);
 131         action = mirror_error_action(s, false, -ret);
 132         if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
 133             s->ret = ret;
 134         }
 135     }
 136     mirror_iteration_done(op, ret);
 137 }
 138
 139 static void mirror_read_complete(void *opaque, int ret)
 140 {
 141     MirrorOp *op = opaque;
 142     MirrorBlockJob *s = op->s;
 143     if (ret < 0) {
 144         BlockErrorAction action;
 145
 146         bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors);
 147         action = mirror_error_action(s, true, -ret);
 148         if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
 149             s->ret = ret;
 150         }
 151
 152         mirror_iteration_done(op, ret);
 153         return;
 154     }
 155     bdrv_aio_writev(s->target, op->sector_num, &op->qiov, op->nb_sectors,
 156                     mirror_write_complete, op);
 157 }
 158
 159 static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)
 160 {
 161     BlockDriverState *source = s->common.bs;
 162     int nb_sectors, sectors_per_chunk, nb_chunks;
 163     int64_t end, sector_num, next_chunk, next_sector, hbitmap_next_sector;
 164     uint64_t delay_ns = 0;
 165     MirrorOp *op;
 166
 167     s->sector_num = hbitmap_iter_next(&s->hbi);
 168     if (s->sector_num < 0) {
 169         bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi);
 170         s->sector_num = hbitmap_iter_next(&s->hbi);
 171         trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap));
 172         assert(s->sector_num >= 0);
 173     }
 174
 175     hbitmap_next_sector = s->sector_num;
 176     sector_num = s->sector_num;
 177     sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
 178     end = s->bdev_length / BDRV_SECTOR_SIZE;
 179
 180     /* Extend the QEMUIOVector to include all adjacent blocks that will
 181      * be copied in this operation.
 182      *
 183      * We have to do this if we have no backing file yet in the destination,
 184      * and the cluster size is very large.  Then we need to do COW ourselves.
 185      * The first time a cluster is copied, copy it entirely.  Note that,
 186      * because both the granularity and the cluster size are powers of two,
 187      * the number of sectors to copy cannot exceed one cluster.
 188      *
 189      * We also want to extend the QEMUIOVector to include more adjacent
 190      * dirty blocks if possible, to limit the number of I/O operations and
 191      * run efficiently even with a small granularity.
 192      */
 193     nb_chunks = 0;
 194     nb_sectors = 0;
 195     next_sector = sector_num;
 196     next_chunk = sector_num / sectors_per_chunk;
 197
 198     /* Wait for I/O to this cluster (from a previous iteration) to be done.  */
 199     while (test_bit(next_chunk, s->in_flight_bitmap)) {
 200         trace_mirror_yield_in_flight(s, sector_num, s->in_flight);
 201         qemu_coroutine_yield();
 202     }
 203
 204     do {
 205         int added_sectors, added_chunks;
 206
 207         if (!bdrv_get_dirty(source, s->dirty_bitmap, next_sector) ||
 208             test_bit(next_chunk, s->in_flight_bitmap)) {
 209             assert(nb_sectors > 0);
 210             break;
 211         }
 212
 213         added_sectors = sectors_per_chunk;
 214         if (s->cow_bitmap && !test_bit(next_chunk, s->cow_bitmap)) {
 215             bdrv_round_to_clusters(s->target,
 216                                    next_sector, added_sectors,
 217                                    &next_sector, &added_sectors);
 218
 219             /* On the first iteration, the rounding may make us copy
 220              * sectors before the first dirty one.
 221              */
 222             if (next_sector < sector_num) {
 223                 assert(nb_sectors == 0);
 224                 sector_num = next_sector;
 225                 next_chunk = next_sector / sectors_per_chunk;
 226             }
 227         }
 228
 229         added_sectors = MIN(added_sectors, end - (sector_num + nb_sectors));
 230         added_chunks = (added_sectors + sectors_per_chunk - 1) / sectors_per_chunk;
 231
 232         /* When doing COW, it may happen that there is not enough space for
 233          * a full cluster.  Wait if that is the case.
 234          */
 235         while (nb_chunks == 0 && s->buf_free_count < added_chunks) {
 236             trace_mirror_yield_buf_busy(s, nb_chunks, s->in_flight);
 237             qemu_coroutine_yield();
 238         }
 239         if (s->buf_free_count < nb_chunks + added_chunks) {
 240             trace_mirror_break_buf_busy(s, nb_chunks, s->in_flight);
 241             break;
 242         }
 243
 244         /* We have enough free space to copy these sectors.  */
 245         bitmap_set(s->in_flight_bitmap, next_chunk, added_chunks);
 246
 247         nb_sectors += added_sectors;
 248         nb_chunks += added_chunks;
 249         next_sector += added_sectors;
 250         next_chunk += added_chunks;
 251         if (!s->synced && s->common.speed) {
 252             delay_ns = ratelimit_calculate_delay(&s->limit, added_sectors);
 253         }
 254     } while (delay_ns == 0 && next_sector < end);
 255
 256     /* Allocate a MirrorOp that is used as an AIO callback.  */
 257     op = g_slice_new(MirrorOp);
 258     op->s = s;
 259     op->sector_num = sector_num;
 260     op->nb_sectors = nb_sectors;
 261
 262     /* Now make a QEMUIOVector taking enough granularity-sized chunks
 263      * from s->buf_free.
 264      */
 265     qemu_iovec_init(&op->qiov, nb_chunks);
 266     next_sector = sector_num;
 267     while (nb_chunks-- > 0) {
 268         MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free);
 269         size_t remaining = (nb_sectors * BDRV_SECTOR_SIZE) - op->qiov.size;
 270
 271         QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next);
 272         s->buf_free_count--;
 273         qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining));
 274
 275         /* Advance the HBitmapIter in parallel, so that we do not examine
 276          * the same sector twice.
 277          */
 278         if (next_sector > hbitmap_next_sector
 279             && bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) {
 280             hbitmap_next_sector = hbitmap_iter_next(&s->hbi);
 281         }
 282
 283         next_sector += sectors_per_chunk;
 284     }
 285
 286     bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, nb_sectors);
 287
 288     /* Copy the dirty cluster.  */
 289     s->in_flight++;
 290     s->sectors_in_flight += nb_sectors;
 291     trace_mirror_one_iteration(s, sector_num, nb_sectors);
 292     bdrv_aio_readv(source, sector_num, &op->qiov, nb_sectors,
 293                    mirror_read_complete, op);
 294     return delay_ns;
 295 }
 296
 297 static void mirror_free_init(MirrorBlockJob *s)
 298 {
 299     int granularity = s->granularity;
 300     size_t buf_size = s->buf_size;
 301     uint8_t *buf = s->buf;
 302
 303     assert(s->buf_free_count == 0);
 304     QSIMPLEQ_INIT(&s->buf_free);
 305     while (buf_size != 0) {
 306         MirrorBuffer *cur = (MirrorBuffer *)buf;
 307         QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next);
 308         s->buf_free_count++;
 309         buf_size -= granularity;
 310         buf += granularity;
 311     }
 312 }
 313
 314 static void mirror_drain(MirrorBlockJob *s)
 315 {
 316     while (s->in_flight > 0) {
 317         qemu_coroutine_yield();
 318     }
 319 }
 320
 321 typedef struct {
 322     int ret;
 323 } MirrorExitData;
 324
 325 static void mirror_exit(BlockJob *job, void *opaque)
 326 {
 327     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
 328     MirrorExitData *data = opaque;
 329     AioContext *replace_aio_context = NULL;
 330
 331     if (s->to_replace) {
 332         replace_aio_context = bdrv_get_aio_context(s->to_replace);
 333         aio_context_acquire(replace_aio_context);
 334     }
 335
 336     if (s->should_complete && data->ret == 0) {
 337         BlockDriverState *to_replace = s->common.bs;
 338         if (s->to_replace) {
 339             to_replace = s->to_replace;
 340         }
 341         if (bdrv_get_flags(s->target) != bdrv_get_flags(to_replace)) {
 342             bdrv_reopen(s->target, bdrv_get_flags(to_replace), NULL);
 343         }
 344         bdrv_swap(s->target, to_replace);
 345         if (s->common.driver->job_type == BLOCK_JOB_TYPE_COMMIT) {
 346             /* drop the bs loop chain formed by the swap: break the loop then
 347              * trigger the unref from the top one */
 348             BlockDriverState *p = s->base->backing_hd;
 349             bdrv_set_backing_hd(s->base, NULL);
 350             bdrv_unref(p);
 351         }
 352     }
 353     if (s->to_replace) {
 354         bdrv_op_unblock_all(s->to_replace, s->replace_blocker);
 355         error_free(s->replace_blocker);
 356         bdrv_unref(s->to_replace);
 357     }
 358     if (replace_aio_context) {
 359         aio_context_release(replace_aio_context);
 360     }
 361     g_free(s->replaces);
 362     bdrv_unref(s->target);
 363     block_job_completed(&s->common, data->ret);
 364     g_free(data);
 365 }
 366
 367 static void coroutine_fn mirror_run(void *opaque)
 368 {
 369     MirrorBlockJob *s = opaque;
 370     MirrorExitData *data;
 371     BlockDriverState *bs = s->common.bs;
 372     int64_t sector_num, end, sectors_per_chunk, length;
 373     uint64_t last_pause_ns;
 374     BlockDriverInfo bdi;
 375     char backing_filename[2]; /* we only need 2 characters because we are only
 376                                  checking for a NULL string */
 377     int ret = 0;
 378     int n;
 379
 380     if (block_job_is_cancelled(&s->common)) {
 381         goto immediate_exit;
 382     }
 383
 384     s->bdev_length = bdrv_getlength(bs);
 385     if (s->bdev_length < 0) {
 386         ret = s->bdev_length;
 387         goto immediate_exit;
 388     } else if (s->bdev_length == 0) {
 389         /* Report BLOCK_JOB_READY and wait for complete. */
 390         block_job_event_ready(&s->common);
 391         s->synced = true;
 392         while (!block_job_is_cancelled(&s->common) && !s->should_complete) {
 393             block_job_yield(&s->common);
 394         }
 395         s->common.cancelled = false;
 396         goto immediate_exit;
 397     }
 398
 399     length = DIV_ROUND_UP(s->bdev_length, s->granularity);
 400     s->in_flight_bitmap = bitmap_new(length);
 401
 402     /* If we have no backing file yet in the destination, we cannot let
 403      * the destination do COW.  Instead, we copy sectors around the
 404      * dirty data if needed.  We need a bitmap to do that.
 405      */
 406     bdrv_get_backing_filename(s->target, backing_filename,
 407                               sizeof(backing_filename));
 408     if (backing_filename[0] && !s->target->backing_hd) {
 409         ret = bdrv_get_info(s->target, &bdi);
 410         if (ret < 0) {
 411             goto immediate_exit;
 412         }
 413         if (s->granularity < bdi.cluster_size) {
 414             s->buf_size = MAX(s->buf_size, bdi.cluster_size);
 415             s->cow_bitmap = bitmap_new(length);
 416         }
 417     }
 418
 419     end = s->bdev_length / BDRV_SECTOR_SIZE;
 420     s->buf = qemu_try_blockalign(bs, s->buf_size);
 421     if (s->buf == NULL) {
 422         ret = -ENOMEM;
 423         goto immediate_exit;
 424     }
 425
 426     sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
 427     mirror_free_init(s);
 428
 429     if (!s->is_none_mode) {
 430         /* First part, loop on the sectors and initialize the dirty bitmap.  */
 431         BlockDriverState *base = s->base;
 432         for (sector_num = 0; sector_num < end; ) {
 433             int64_t next = (sector_num | (sectors_per_chunk - 1)) + 1;
 434             ret = bdrv_is_allocated_above(bs, base,
 435                                           sector_num, next - sector_num, &n);
 436
 437             if (ret < 0) {
 438                 goto immediate_exit;
 439             }
 440
 441             assert(n > 0);
 442             if (ret == 1) {
 443                 bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n);
 444                 sector_num = next;
 445             } else {
 446                 sector_num += n;
 447             }
 448         }
 449     }
 450
 451     bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi);
 452     last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
 453     for (;;) {
 454         uint64_t delay_ns = 0;
 455         int64_t cnt;
 456         bool should_complete;
 457
 458         if (s->ret < 0) {
 459             ret = s->ret;
 460             goto immediate_exit;
 461         }
 462
 463         cnt = bdrv_get_dirty_count(s->dirty_bitmap);
 464         /* s->common.offset contains the number of bytes already processed so
 465          * far, cnt is the number of dirty sectors remaining and
 466          * s->sectors_in_flight is the number of sectors currently being
 467          * processed; together those are the current total operation length */
 468         s->common.len = s->common.offset +
 469                         (cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE;
 470
 471         /* Note that even when no rate limit is applied we need to yield
 472          * periodically with no pending I/O so that bdrv_drain_all() returns.
 473          * We do so every SLICE_TIME nanoseconds, or when there is an error,
 474          * or when the source is clean, whichever comes first.
 475          */
 476         if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - last_pause_ns < SLICE_TIME &&
 477             s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 478             if (s->in_flight == MAX_IN_FLIGHT || s->buf_free_count == 0 ||
 479                 (cnt == 0 && s->in_flight > 0)) {
 480                 trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt);
 481                 qemu_coroutine_yield();
 482                 continue;
 483             } else if (cnt != 0) {
 484                 delay_ns = mirror_iteration(s);
 485             }
 486         }
 487
 488         should_complete = false;
 489         if (s->in_flight == 0 && cnt == 0) {
 490             trace_mirror_before_flush(s);
 491             ret = bdrv_flush(s->target);
 492             if (ret < 0) {
 493                 if (mirror_error_action(s, false, -ret) ==
 494                     BLOCK_ERROR_ACTION_REPORT) {
 495                     goto immediate_exit;
 496                 }
 497             } else {
 498                 /* We're out of the streaming phase.  From now on, if the job
 499                  * is cancelled we will actually complete all pending I/O and
 500                  * report completion.  This way, block-job-cancel will leave
 501                  * the target in a consistent state.
 502                  */
 503                 if (!s->synced) {
 504                     block_job_event_ready(&s->common);
 505                     s->synced = true;
 506                 }
 507
 508                 should_complete = s->should_complete ||
 509                     block_job_is_cancelled(&s->common);
 510                 cnt = bdrv_get_dirty_count(s->dirty_bitmap);
 511             }
 512         }
 513
 514         if (cnt == 0 && should_complete) {
 515             /* The dirty bitmap is not updated while operations are pending.
 516              * If we're about to exit, wait for pending operations before
 517              * calling bdrv_get_dirty_count(bs), or we may exit while the
 518              * source has dirty data to copy!
 519              *
 520              * Note that I/O can be submitted by the guest while
 521              * mirror_populate runs.
 522              */
 523             trace_mirror_before_drain(s, cnt);
 524             bdrv_drain(bs);
 525             cnt = bdrv_get_dirty_count(s->dirty_bitmap);
 526         }
 527
 528         ret = 0;
 529         trace_mirror_before_sleep(s, cnt, s->synced, delay_ns);
 530         if (!s->synced) {
 531             block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 532             if (block_job_is_cancelled(&s->common)) {
 533                 break;
 534             }
 535         } else if (!should_complete) {
 536             delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0);
 537             block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 538         } else if (cnt == 0) {
 539             /* The two disks are in sync.  Exit and report successful
 540              * completion.
 541              */
 542             assert(QLIST_EMPTY(&bs->tracked_requests));
 543             s->common.cancelled = false;
 544             break;
 545         }
 546         last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
 547     }
 548
 549 immediate_exit:
 550     if (s->in_flight > 0) {
 551         /* We get here only if something went wrong.  Either the job failed,
 552          * or it was cancelled prematurely so that we do not guarantee that
 553          * the target is a copy of the source.
 554          */
 555         assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common)));
 556         mirror_drain(s);
 557     }
 558
 559     assert(s->in_flight == 0);
 560     qemu_vfree(s->buf);
 561     g_free(s->cow_bitmap);
 562     g_free(s->in_flight_bitmap);
 563     bdrv_release_dirty_bitmap(bs, s->dirty_bitmap);
 564     bdrv_iostatus_disable(s->target);
 565
 566     data = g_malloc(sizeof(*data));
 567     data->ret = ret;
 568     block_job_defer_to_main_loop(&s->common, mirror_exit, data);
 569 }
 570
 571 static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp)
 572 {
 573     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
 574
 575     if (speed < 0) {
 576         error_set(errp, QERR_INVALID_PARAMETER, "speed");
 577         return;
 578     }
 579     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 580 }
 581
 582 static void mirror_iostatus_reset(BlockJob *job)
 583 {
 584     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
 585
 586     bdrv_iostatus_reset(s->target);
 587 }
 588
 589 static void mirror_complete(BlockJob *job, Error **errp)
 590 {
 591     MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
 592     Error *local_err = NULL;
 593     int ret;
 594
 595     ret = bdrv_open_backing_file(s->target, NULL, &local_err);
 596     if (ret < 0) {
 597         error_propagate(errp, local_err);
 598         return;
 599     }
 600     if (!s->synced) {
 601         error_set(errp, QERR_BLOCK_JOB_NOT_READY,
 602                   bdrv_get_device_name(job->bs));
 603         return;
 604     }
 605
 606     /* check the target bs is not blocked and block all operations on it */
 607     if (s->replaces) {
 608         AioContext *replace_aio_context;
 609
 610         s->to_replace = check_to_replace_node(s->replaces, &local_err);
 611         if (!s->to_replace) {
 612             error_propagate(errp, local_err);
 613             return;
 614         }
 615
 616         replace_aio_context = bdrv_get_aio_context(s->to_replace);
 617         aio_context_acquire(replace_aio_context);
 618
 619         error_setg(&s->replace_blocker,
 620                    "block device is in use by block-job-complete");
 621         bdrv_op_block_all(s->to_replace, s->replace_blocker);
 622         bdrv_ref(s->to_replace);
 623
 624         aio_context_release(replace_aio_context);
 625     }
 626
 627     s->should_complete = true;
 628     block_job_enter(&s->common);
 629 }
 630
 631 static const BlockJobDriver mirror_job_driver = {
 632     .instance_size = sizeof(MirrorBlockJob),
 633     .job_type      = BLOCK_JOB_TYPE_MIRROR,
 634     .set_speed     = mirror_set_speed,
 635     .iostatus_reset= mirror_iostatus_reset,
 636     .complete      = mirror_complete,
 637 };
 638
 639 static const BlockJobDriver commit_active_job_driver = {
 640     .instance_size = sizeof(MirrorBlockJob),
 641     .job_type      = BLOCK_JOB_TYPE_COMMIT,
 642     .set_speed     = mirror_set_speed,
 643     .iostatus_reset
 644                    = mirror_iostatus_reset,
 645     .complete      = mirror_complete,
 646 };
 647
 648 static void mirror_start_job(BlockDriverState *bs, BlockDriverState *target,
 649                              const char *replaces,
 650                              int64_t speed, uint32_t granularity,
 651                              int64_t buf_size,
 652                              BlockdevOnError on_source_error,
 653                              BlockdevOnError on_target_error,
 654                              BlockCompletionFunc *cb,
 655                              void *opaque, Error **errp,
 656                              const BlockJobDriver *driver,
 657                              bool is_none_mode, BlockDriverState *base)
 658 {
 659     MirrorBlockJob *s;
 660
 661     if (granularity == 0) {
 662         granularity = bdrv_get_default_bitmap_granularity(target);
 663     }
 664
 665     assert ((granularity & (granularity - 1)) == 0);
 666
 667     if ((on_source_error == BLOCKDEV_ON_ERROR_STOP ||
 668          on_source_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
 669         !bdrv_iostatus_is_enabled(bs)) {
 670         error_set(errp, QERR_INVALID_PARAMETER, "on-source-error");
 671         return;
 672     }
 673
 674
 675     s = block_job_create(driver, bs, speed, cb, opaque, errp);
 676     if (!s) {
 677         return;
 678     }
 679
 680     s->replaces = g_strdup(replaces);
 681     s->on_source_error = on_source_error;
 682     s->on_target_error = on_target_error;
 683     s->target = target;
 684     s->is_none_mode = is_none_mode;
 685     s->base = base;
 686     s->granularity = granularity;
 687     s->buf_size = MAX(buf_size, granularity);
 688
 689     s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp);
 690     if (!s->dirty_bitmap) {
 691         return;
 692     }
 693     bdrv_set_enable_write_cache(s->target, true);
 694     bdrv_set_on_error(s->target, on_target_error, on_target_error);
 695     bdrv_iostatus_enable(s->target);
 696     s->common.co = qemu_coroutine_create(mirror_run);
 697     trace_mirror_start(bs, s, s->common.co, opaque);
 698     qemu_coroutine_enter(s->common.co, s);
 699 }
 700
 701 void mirror_start(BlockDriverState *bs, BlockDriverState *target,
 702                   const char *replaces,
 703                   int64_t speed, uint32_t granularity, int64_t buf_size,
 704                   MirrorSyncMode mode, BlockdevOnError on_source_error,
 705                   BlockdevOnError on_target_error,
 706                   BlockCompletionFunc *cb,
 707                   void *opaque, Error **errp)
 708 {
 709     bool is_none_mode;
 710     BlockDriverState *base;
 711
 712     if (mode == MIRROR_SYNC_MODE_DIRTY_BITMAP) {
 713         error_setg(errp, "Sync mode 'dirty-bitmap' not supported");
 714         return;
 715     }
 716     is_none_mode = mode == MIRROR_SYNC_MODE_NONE;
 717     base = mode == MIRROR_SYNC_MODE_TOP ? bs->backing_hd : NULL;
 718     mirror_start_job(bs, target, replaces,
 719                      speed, granularity, buf_size,
 720                      on_source_error, on_target_error, cb, opaque, errp,
 721                      &mirror_job_driver, is_none_mode, base);
 722 }
 723
 724 void commit_active_start(BlockDriverState *bs, BlockDriverState *base,
 725                          int64_t speed,
 726                          BlockdevOnError on_error,
 727                          BlockCompletionFunc *cb,
 728                          void *opaque, Error **errp)
 729 {
 730     int64_t length, base_length;
 731     int orig_base_flags;
 732     int ret;
 733     Error *local_err = NULL;
 734
 735     orig_base_flags = bdrv_get_flags(base);
 736
 737     if (bdrv_reopen(base, bs->open_flags, errp)) {
 738         return;
 739     }
 740
 741     length = bdrv_getlength(bs);
 742     if (length < 0) {
 743         error_setg_errno(errp, -length,
 744                          "Unable to determine length of %s", bs->filename);
 745         goto error_restore_flags;
 746     }
 747
 748     base_length = bdrv_getlength(base);
 749     if (base_length < 0) {
 750         error_setg_errno(errp, -base_length,
 751                          "Unable to determine length of %s", base->filename);
 752         goto error_restore_flags;
 753     }
 754
 755     if (length > base_length) {
 756         ret = bdrv_truncate(base, length);
 757         if (ret < 0) {
 758             error_setg_errno(errp, -ret,
 759                             "Top image %s is larger than base image %s, and "
 760                              "resize of base image failed",
 761                              bs->filename, base->filename);
 762             goto error_restore_flags;
 763         }
 764     }
 765
 766     bdrv_ref(base);
 767     mirror_start_job(bs, base, NULL, speed, 0, 0,
 768                      on_error, on_error, cb, opaque, &local_err,
 769                      &commit_active_job_driver, false, base);
 770     if (local_err) {
 771         error_propagate(errp, local_err);
 772         goto error_restore_flags;
 773     }
 774
 775     return;
 776
 777 error_restore_flags:
 778     /* ignore error and errp for bdrv_reopen, because we want to propagate
 779      * the original error */
 780     bdrv_reopen(base, orig_base_flags, NULL);
 781     return;
 782 }