block/backup.c

   1 /*
   2  * QEMU backup
   3  *
   4  * Copyright (C) 2013 Proxmox Server Solutions
   5  *
   6  * Authors:
   7  *  Dietmar Maurer ([email protected])
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  10  * See the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "qemu/osdep.h"
  15
  16 #include "trace.h"
  17 #include "block/block.h"
  18 #include "block/block_int.h"
  19 #include "block/blockjob_int.h"
  20 #include "block/block_backup.h"
  21 #include "qapi/error.h"
  22 #include "qapi/qmp/qerror.h"
  23 #include "qemu/ratelimit.h"
  24 #include "qemu/cutils.h"
  25 #include "sysemu/block-backend.h"
  26 #include "qemu/bitmap.h"
  27 #include "qemu/error-report.h"
  28
  29 #define BACKUP_CLUSTER_SIZE_DEFAULT (1 << 16)
  30 #define SLICE_TIME 100000000ULL /* ns */
  31
  32 typedef struct BackupBlockJob {
  33     BlockJob common;
  34     BlockBackend *target;
  35     /* bitmap for sync=incremental */
  36     BdrvDirtyBitmap *sync_bitmap;
  37     MirrorSyncMode sync_mode;
  38     RateLimit limit;
  39     BlockdevOnError on_source_error;
  40     BlockdevOnError on_target_error;
  41     CoRwlock flush_rwlock;
  42     uint64_t sectors_read;
  43     unsigned long *done_bitmap;
  44     int64_t cluster_size;
  45     bool compress;
  46     NotifierWithReturn before_write;
  47     QLIST_HEAD(, CowRequest) inflight_reqs;
  48 } BackupBlockJob;
  49
  50 /* Size of a cluster in sectors, instead of bytes. */
  51 static inline int64_t cluster_size_sectors(BackupBlockJob *job)
  52 {
  53   return job->cluster_size / BDRV_SECTOR_SIZE;
  54 }
  55
  56 /* See if in-flight requests overlap and wait for them to complete */
  57 static void coroutine_fn wait_for_overlapping_requests(BackupBlockJob *job,
  58                                                        int64_t start,
  59                                                        int64_t end)
  60 {
  61     CowRequest *req;
  62     bool retry;
  63
  64     do {
  65         retry = false;
  66         QLIST_FOREACH(req, &job->inflight_reqs, list) {
  67             if (end > req->start && start < req->end) {
  68                 qemu_co_queue_wait(&req->wait_queue, NULL);
  69                 retry = true;
  70                 break;
  71             }
  72         }
  73     } while (retry);
  74 }
  75
  76 /* Keep track of an in-flight request */
  77 static void cow_request_begin(CowRequest *req, BackupBlockJob *job,
  78                                      int64_t start, int64_t end)
  79 {
  80     req->start = start;
  81     req->end = end;
  82     qemu_co_queue_init(&req->wait_queue);
  83     QLIST_INSERT_HEAD(&job->inflight_reqs, req, list);
  84 }
  85
  86 /* Forget about a completed request */
  87 static void cow_request_end(CowRequest *req)
  88 {
  89     QLIST_REMOVE(req, list);
  90     qemu_co_queue_restart_all(&req->wait_queue);
  91 }
  92
  93 static int coroutine_fn backup_do_cow(BackupBlockJob *job,
  94                                       int64_t sector_num, int nb_sectors,
  95                                       bool *error_is_read,
  96                                       bool is_write_notifier)
  97 {
  98     BlockBackend *blk = job->common.blk;
  99     CowRequest cow_request;
 100     struct iovec iov;
 101     QEMUIOVector bounce_qiov;
 102     void *bounce_buffer = NULL;
 103     int ret = 0;
 104     int64_t sectors_per_cluster = cluster_size_sectors(job);
 105     int64_t start, end;
 106     int n;
 107
 108     qemu_co_rwlock_rdlock(&job->flush_rwlock);
 109
 110     start = sector_num / sectors_per_cluster;
 111     end = DIV_ROUND_UP(sector_num + nb_sectors, sectors_per_cluster);
 112
 113     trace_backup_do_cow_enter(job, start, sector_num, nb_sectors);
 114
 115     wait_for_overlapping_requests(job, start, end);
 116     cow_request_begin(&cow_request, job, start, end);
 117
 118     for (; start < end; start++) {
 119         if (test_bit(start, job->done_bitmap)) {
 120             trace_backup_do_cow_skip(job, start);
 121             continue; /* already copied */
 122         }
 123
 124         trace_backup_do_cow_process(job, start);
 125
 126         n = MIN(sectors_per_cluster,
 127                 job->common.len / BDRV_SECTOR_SIZE -
 128                 start * sectors_per_cluster);
 129
 130         if (!bounce_buffer) {
 131             bounce_buffer = blk_blockalign(blk, job->cluster_size);
 132         }
 133         iov.iov_base = bounce_buffer;
 134         iov.iov_len = n * BDRV_SECTOR_SIZE;
 135         qemu_iovec_init_external(&bounce_qiov, &iov, 1);
 136
 137         ret = blk_co_preadv(blk, start * job->cluster_size,
 138                             bounce_qiov.size, &bounce_qiov,
 139                             is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0);
 140         if (ret < 0) {
 141             trace_backup_do_cow_read_fail(job, start, ret);
 142             if (error_is_read) {
 143                 *error_is_read = true;
 144             }
 145             goto out;
 146         }
 147
 148         if (buffer_is_zero(iov.iov_base, iov.iov_len)) {
 149             ret = blk_co_pwrite_zeroes(job->target, start * job->cluster_size,
 150                                        bounce_qiov.size, BDRV_REQ_MAY_UNMAP);
 151         } else {
 152             ret = blk_co_pwritev(job->target, start * job->cluster_size,
 153                                  bounce_qiov.size, &bounce_qiov,
 154                                  job->compress ? BDRV_REQ_WRITE_COMPRESSED : 0);
 155         }
 156         if (ret < 0) {
 157             trace_backup_do_cow_write_fail(job, start, ret);
 158             if (error_is_read) {
 159                 *error_is_read = false;
 160             }
 161             goto out;
 162         }
 163
 164         set_bit(start, job->done_bitmap);
 165
 166         /* Publish progress, guest I/O counts as progress too.  Note that the
 167          * offset field is an opaque progress value, it is not a disk offset.
 168          */
 169         job->sectors_read += n;
 170         job->common.offset += n * BDRV_SECTOR_SIZE;
 171     }
 172
 173 out:
 174     if (bounce_buffer) {
 175         qemu_vfree(bounce_buffer);
 176     }
 177
 178     cow_request_end(&cow_request);
 179
 180     trace_backup_do_cow_return(job, sector_num, nb_sectors, ret);
 181
 182     qemu_co_rwlock_unlock(&job->flush_rwlock);
 183
 184     return ret;
 185 }
 186
 187 static int coroutine_fn backup_before_write_notify(
 188         NotifierWithReturn *notifier,
 189         void *opaque)
 190 {
 191     BackupBlockJob *job = container_of(notifier, BackupBlockJob, before_write);
 192     BdrvTrackedRequest *req = opaque;
 193     int64_t sector_num = req->offset >> BDRV_SECTOR_BITS;
 194     int nb_sectors = req->bytes >> BDRV_SECTOR_BITS;
 195
 196     assert(req->bs == blk_bs(job->common.blk));
 197     assert((req->offset & (BDRV_SECTOR_SIZE - 1)) == 0);
 198     assert((req->bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
 199
 200     return backup_do_cow(job, sector_num, nb_sectors, NULL, true);
 201 }
 202
 203 static void backup_set_speed(BlockJob *job, int64_t speed, Error **errp)
 204 {
 205     BackupBlockJob *s = container_of(job, BackupBlockJob, common);
 206
 207     if (speed < 0) {
 208         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 209         return;
 210     }
 211     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 212 }
 213
 214 static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
 215 {
 216     BdrvDirtyBitmap *bm;
 217     BlockDriverState *bs = blk_bs(job->common.blk);
 218
 219     if (ret < 0 || block_job_is_cancelled(&job->common)) {
 220         /* Merge the successor back into the parent, delete nothing. */
 221         bm = bdrv_reclaim_dirty_bitmap(bs, job->sync_bitmap, NULL);
 222         assert(bm);
 223     } else {
 224         /* Everything is fine, delete this bitmap and install the backup. */
 225         bm = bdrv_dirty_bitmap_abdicate(bs, job->sync_bitmap, NULL);
 226         assert(bm);
 227     }
 228 }
 229
 230 static void backup_commit(BlockJob *job)
 231 {
 232     BackupBlockJob *s = container_of(job, BackupBlockJob, common);
 233     if (s->sync_bitmap) {
 234         backup_cleanup_sync_bitmap(s, 0);
 235     }
 236 }
 237
 238 static void backup_abort(BlockJob *job)
 239 {
 240     BackupBlockJob *s = container_of(job, BackupBlockJob, common);
 241     if (s->sync_bitmap) {
 242         backup_cleanup_sync_bitmap(s, -1);
 243     }
 244 }
 245
 246 static void backup_clean(BlockJob *job)
 247 {
 248     BackupBlockJob *s = container_of(job, BackupBlockJob, common);
 249     assert(s->target);
 250     blk_unref(s->target);
 251     s->target = NULL;
 252 }
 253
 254 static void backup_attached_aio_context(BlockJob *job, AioContext *aio_context)
 255 {
 256     BackupBlockJob *s = container_of(job, BackupBlockJob, common);
 257
 258     blk_set_aio_context(s->target, aio_context);
 259 }
 260
 261 void backup_do_checkpoint(BlockJob *job, Error **errp)
 262 {
 263     BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
 264     int64_t len;
 265
 266     assert(job->driver->job_type == BLOCK_JOB_TYPE_BACKUP);
 267
 268     if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
 269         error_setg(errp, "The backup job only supports block checkpoint in"
 270                    " sync=none mode");
 271         return;
 272     }
 273
 274     len = DIV_ROUND_UP(backup_job->common.len, backup_job->cluster_size);
 275     bitmap_zero(backup_job->done_bitmap, len);
 276 }
 277
 278 void backup_wait_for_overlapping_requests(BlockJob *job, int64_t sector_num,
 279                                           int nb_sectors)
 280 {
 281     BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
 282     int64_t sectors_per_cluster = cluster_size_sectors(backup_job);
 283     int64_t start, end;
 284
 285     assert(job->driver->job_type == BLOCK_JOB_TYPE_BACKUP);
 286
 287     start = sector_num / sectors_per_cluster;
 288     end = DIV_ROUND_UP(sector_num + nb_sectors, sectors_per_cluster);
 289     wait_for_overlapping_requests(backup_job, start, end);
 290 }
 291
 292 void backup_cow_request_begin(CowRequest *req, BlockJob *job,
 293                               int64_t sector_num,
 294                               int nb_sectors)
 295 {
 296     BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
 297     int64_t sectors_per_cluster = cluster_size_sectors(backup_job);
 298     int64_t start, end;
 299
 300     assert(job->driver->job_type == BLOCK_JOB_TYPE_BACKUP);
 301
 302     start = sector_num / sectors_per_cluster;
 303     end = DIV_ROUND_UP(sector_num + nb_sectors, sectors_per_cluster);
 304     cow_request_begin(req, backup_job, start, end);
 305 }
 306
 307 void backup_cow_request_end(CowRequest *req)
 308 {
 309     cow_request_end(req);
 310 }
 311
 312 static void backup_drain(BlockJob *job)
 313 {
 314     BackupBlockJob *s = container_of(job, BackupBlockJob, common);
 315
 316     /* Need to keep a reference in case blk_drain triggers execution
 317      * of backup_complete...
 318      */
 319     if (s->target) {
 320         BlockBackend *target = s->target;
 321         blk_ref(target);
 322         blk_drain(target);
 323         blk_unref(target);
 324     }
 325 }
 326
 327 static BlockErrorAction backup_error_action(BackupBlockJob *job,
 328                                             bool read, int error)
 329 {
 330     if (read) {
 331         return block_job_error_action(&job->common, job->on_source_error,
 332                                       true, error);
 333     } else {
 334         return block_job_error_action(&job->common, job->on_target_error,
 335                                       false, error);
 336     }
 337 }
 338
 339 typedef struct {
 340     int ret;
 341 } BackupCompleteData;
 342
 343 static void backup_complete(BlockJob *job, void *opaque)
 344 {
 345     BackupCompleteData *data = opaque;
 346
 347     block_job_completed(job, data->ret);
 348     g_free(data);
 349 }
 350
 351 static bool coroutine_fn yield_and_check(BackupBlockJob *job)
 352 {
 353     if (block_job_is_cancelled(&job->common)) {
 354         return true;
 355     }
 356
 357     /* we need to yield so that bdrv_drain_all() returns.
 358      * (without, VM does not reboot)
 359      */
 360     if (job->common.speed) {
 361         uint64_t delay_ns = ratelimit_calculate_delay(&job->limit,
 362                                                       job->sectors_read);
 363         job->sectors_read = 0;
 364         block_job_sleep_ns(&job->common, QEMU_CLOCK_REALTIME, delay_ns);
 365     } else {
 366         block_job_sleep_ns(&job->common, QEMU_CLOCK_REALTIME, 0);
 367     }
 368
 369     if (block_job_is_cancelled(&job->common)) {
 370         return true;
 371     }
 372
 373     return false;
 374 }
 375
 376 static int coroutine_fn backup_run_incremental(BackupBlockJob *job)
 377 {
 378     bool error_is_read;
 379     int ret = 0;
 380     int clusters_per_iter;
 381     uint32_t granularity;
 382     int64_t sector;
 383     int64_t cluster;
 384     int64_t end;
 385     int64_t last_cluster = -1;
 386     int64_t sectors_per_cluster = cluster_size_sectors(job);
 387     BdrvDirtyBitmapIter *dbi;
 388
 389     granularity = bdrv_dirty_bitmap_granularity(job->sync_bitmap);
 390     clusters_per_iter = MAX((granularity / job->cluster_size), 1);
 391     dbi = bdrv_dirty_iter_new(job->sync_bitmap, 0);
 392
 393     /* Find the next dirty sector(s) */
 394     while ((sector = bdrv_dirty_iter_next(dbi)) != -1) {
 395         cluster = sector / sectors_per_cluster;
 396
 397         /* Fake progress updates for any clusters we skipped */
 398         if (cluster != last_cluster + 1) {
 399             job->common.offset += ((cluster - last_cluster - 1) *
 400                                    job->cluster_size);
 401         }
 402
 403         for (end = cluster + clusters_per_iter; cluster < end; cluster++) {
 404             do {
 405                 if (yield_and_check(job)) {
 406                     goto out;
 407                 }
 408                 ret = backup_do_cow(job, cluster * sectors_per_cluster,
 409                                     sectors_per_cluster, &error_is_read,
 410                                     false);
 411                 if ((ret < 0) &&
 412                     backup_error_action(job, error_is_read, -ret) ==
 413                     BLOCK_ERROR_ACTION_REPORT) {
 414                     goto out;
 415                 }
 416             } while (ret < 0);
 417         }
 418
 419         /* If the bitmap granularity is smaller than the backup granularity,
 420          * we need to advance the iterator pointer to the next cluster. */
 421         if (granularity < job->cluster_size) {
 422             bdrv_set_dirty_iter(dbi, cluster * sectors_per_cluster);
 423         }
 424
 425         last_cluster = cluster - 1;
 426     }
 427
 428     /* Play some final catchup with the progress meter */
 429     end = DIV_ROUND_UP(job->common.len, job->cluster_size);
 430     if (last_cluster + 1 < end) {
 431         job->common.offset += ((end - last_cluster - 1) * job->cluster_size);
 432     }
 433
 434 out:
 435     bdrv_dirty_iter_free(dbi);
 436     return ret;
 437 }
 438
 439 static void coroutine_fn backup_run(void *opaque)
 440 {
 441     BackupBlockJob *job = opaque;
 442     BackupCompleteData *data;
 443     BlockDriverState *bs = blk_bs(job->common.blk);
 444     int64_t start, end;
 445     int64_t sectors_per_cluster = cluster_size_sectors(job);
 446     int ret = 0;
 447
 448     QLIST_INIT(&job->inflight_reqs);
 449     qemu_co_rwlock_init(&job->flush_rwlock);
 450
 451     start = 0;
 452     end = DIV_ROUND_UP(job->common.len, job->cluster_size);
 453
 454     job->done_bitmap = bitmap_new(end);
 455
 456     job->before_write.notify = backup_before_write_notify;
 457     bdrv_add_before_write_notifier(bs, &job->before_write);
 458
 459     if (job->sync_mode == MIRROR_SYNC_MODE_NONE) {
 460         while (!block_job_is_cancelled(&job->common)) {
 461             /* Yield until the job is cancelled.  We just let our before_write
 462              * notify callback service CoW requests. */
 463             block_job_yield(&job->common);
 464         }
 465     } else if (job->sync_mode == MIRROR_SYNC_MODE_INCREMENTAL) {
 466         ret = backup_run_incremental(job);
 467     } else {
 468         /* Both FULL and TOP SYNC_MODE's require copying.. */
 469         for (; start < end; start++) {
 470             bool error_is_read;
 471             int alloced = 0;
 472
 473             if (yield_and_check(job)) {
 474                 break;
 475             }
 476
 477             if (job->sync_mode == MIRROR_SYNC_MODE_TOP) {
 478                 int i, n;
 479
 480                 /* Check to see if these blocks are already in the
 481                  * backing file. */
 482
 483                 for (i = 0; i < sectors_per_cluster;) {
 484                     /* bdrv_is_allocated() only returns true/false based
 485                      * on the first set of sectors it comes across that
 486                      * are are all in the same state.
 487                      * For that reason we must verify each sector in the
 488                      * backup cluster length.  We end up copying more than
 489                      * needed but at some point that is always the case. */
 490                     alloced =
 491                         bdrv_is_allocated(bs,
 492                                 start * sectors_per_cluster + i,
 493                                 sectors_per_cluster - i, &n);
 494                     i += n;
 495
 496                     if (alloced || n == 0) {
 497                         break;
 498                     }
 499                 }
 500
 501                 /* If the above loop never found any sectors that are in
 502                  * the topmost image, skip this backup. */
 503                 if (alloced == 0) {
 504                     continue;
 505                 }
 506             }
 507             /* FULL sync mode we copy the whole drive. */
 508             if (alloced < 0) {
 509                 ret = alloced;
 510             } else {
 511                 ret = backup_do_cow(job, start * sectors_per_cluster,
 512                                     sectors_per_cluster, &error_is_read,
 513                                     false);
 514             }
 515             if (ret < 0) {
 516                 /* Depending on error action, fail now or retry cluster */
 517                 BlockErrorAction action =
 518                     backup_error_action(job, error_is_read, -ret);
 519                 if (action == BLOCK_ERROR_ACTION_REPORT) {
 520                     break;
 521                 } else {
 522                     start--;
 523                     continue;
 524                 }
 525             }
 526         }
 527     }
 528
 529     notifier_with_return_remove(&job->before_write);
 530
 531     /* wait until pending backup_do_cow() calls have completed */
 532     qemu_co_rwlock_wrlock(&job->flush_rwlock);
 533     qemu_co_rwlock_unlock(&job->flush_rwlock);
 534     g_free(job->done_bitmap);
 535
 536     data = g_malloc(sizeof(*data));
 537     data->ret = ret;
 538     block_job_defer_to_main_loop(&job->common, backup_complete, data);
 539 }
 540
 541 static const BlockJobDriver backup_job_driver = {
 542     .instance_size          = sizeof(BackupBlockJob),
 543     .job_type               = BLOCK_JOB_TYPE_BACKUP,
 544     .start                  = backup_run,
 545     .set_speed              = backup_set_speed,
 546     .commit                 = backup_commit,
 547     .abort                  = backup_abort,
 548     .clean                  = backup_clean,
 549     .attached_aio_context   = backup_attached_aio_context,
 550     .drain                  = backup_drain,
 551 };
 552
 553 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
 554                   BlockDriverState *target, int64_t speed,
 555                   MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
 556                   bool compress,
 557                   BlockdevOnError on_source_error,
 558                   BlockdevOnError on_target_error,
 559                   int creation_flags,
 560                   BlockCompletionFunc *cb, void *opaque,
 561                   BlockJobTxn *txn, Error **errp)
 562 {
 563     int64_t len;
 564     BlockDriverInfo bdi;
 565     BackupBlockJob *job = NULL;
 566     int ret;
 567
 568     assert(bs);
 569     assert(target);
 570
 571     if (bs == target) {
 572         error_setg(errp, "Source and target cannot be the same");
 573         return NULL;
 574     }
 575
 576     if (!bdrv_is_inserted(bs)) {
 577         error_setg(errp, "Device is not inserted: %s",
 578                    bdrv_get_device_name(bs));
 579         return NULL;
 580     }
 581
 582     if (!bdrv_is_inserted(target)) {
 583         error_setg(errp, "Device is not inserted: %s",
 584                    bdrv_get_device_name(target));
 585         return NULL;
 586     }
 587
 588     if (compress && target->drv->bdrv_co_pwritev_compressed == NULL) {
 589         error_setg(errp, "Compression is not supported for this drive %s",
 590                    bdrv_get_device_name(target));
 591         return NULL;
 592     }
 593
 594     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
 595         return NULL;
 596     }
 597
 598     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
 599         return NULL;
 600     }
 601
 602     if (sync_mode == MIRROR_SYNC_MODE_INCREMENTAL) {
 603         if (!sync_bitmap) {
 604             error_setg(errp, "must provide a valid bitmap name for "
 605                              "\"incremental\" sync mode");
 606             return NULL;
 607         }
 608
 609         /* Create a new bitmap, and freeze/disable this one. */
 610         if (bdrv_dirty_bitmap_create_successor(bs, sync_bitmap, errp) < 0) {
 611             return NULL;
 612         }
 613     } else if (sync_bitmap) {
 614         error_setg(errp,
 615                    "a sync_bitmap was provided to backup_run, "
 616                    "but received an incompatible sync_mode (%s)",
 617                    MirrorSyncMode_lookup[sync_mode]);
 618         return NULL;
 619     }
 620
 621     len = bdrv_getlength(bs);
 622     if (len < 0) {
 623         error_setg_errno(errp, -len, "unable to get length for '%s'",
 624                          bdrv_get_device_name(bs));
 625         goto error;
 626     }
 627
 628     /* job->common.len is fixed, so we can't allow resize */
 629     job = block_job_create(job_id, &backup_job_driver, bs,
 630                            BLK_PERM_CONSISTENT_READ,
 631                            BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
 632                            BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD,
 633                            speed, creation_flags, cb, opaque, errp);
 634     if (!job) {
 635         goto error;
 636     }
 637
 638     /* The target must match the source in size, so no resize here either */
 639     job->target = blk_new(BLK_PERM_WRITE,
 640                           BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
 641                           BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD);
 642     ret = blk_insert_bs(job->target, target, errp);
 643     if (ret < 0) {
 644         goto error;
 645     }
 646
 647     job->on_source_error = on_source_error;
 648     job->on_target_error = on_target_error;
 649     job->sync_mode = sync_mode;
 650     job->sync_bitmap = sync_mode == MIRROR_SYNC_MODE_INCREMENTAL ?
 651                        sync_bitmap : NULL;
 652     job->compress = compress;
 653
 654     /* If there is no backing file on the target, we cannot rely on COW if our
 655      * backup cluster size is smaller than the target cluster size. Even for
 656      * targets with a backing file, try to avoid COW if possible. */
 657     ret = bdrv_get_info(target, &bdi);
 658     if (ret == -ENOTSUP && !target->backing) {
 659         /* Cluster size is not defined */
 660         error_report("WARNING: The target block device doesn't provide "
 661                      "information about the block size and it doesn't have a "
 662                      "backing file. The default block size of %u bytes is "
 663                      "used. If the actual block size of the target exceeds "
 664                      "this default, the backup may be unusable",
 665                      BACKUP_CLUSTER_SIZE_DEFAULT);
 666         job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
 667     } else if (ret < 0 && !target->backing) {
 668         error_setg_errno(errp, -ret,
 669             "Couldn't determine the cluster size of the target image, "
 670             "which has no backing file");
 671         error_append_hint(errp,
 672             "Aborting, since this may create an unusable destination image\n");
 673         goto error;
 674     } else if (ret < 0 && target->backing) {
 675         /* Not fatal; just trudge on ahead. */
 676         job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
 677     } else {
 678         job->cluster_size = MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
 679     }
 680
 681     /* Required permissions are already taken with target's blk_new() */
 682     block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
 683                        &error_abort);
 684     job->common.len = len;
 685     block_job_txn_add_job(txn, &job->common);
 686
 687     return &job->common;
 688
 689  error:
 690     if (sync_bitmap) {
 691         bdrv_reclaim_dirty_bitmap(bs, sync_bitmap, NULL);
 692     }
 693     if (job) {
 694         backup_clean(&job->common);
 695         block_job_unref(&job->common);
 696     }
 697
 698     return NULL;
 699 }