block/backup.c

   1 /*
   2  * QEMU backup
   3  *
   4  * Copyright (C) 2013 Proxmox Server Solutions
   5  * Copyright (c) 2019 Virtuozzo International GmbH.
   6  *
   7  * Authors:
   8  *  Dietmar Maurer (dietmar@proxmox.com)
   9  *
  10  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  11  * See the COPYING file in the top-level directory.
  12  *
  13  */
  14
  15 #include "qemu/osdep.h"
  16
  17 #include "trace.h"
  18 #include "block/block.h"
  19 #include "block/block_int.h"
  20 #include "block/blockjob_int.h"
  21 #include "block/block_backup.h"
  22 #include "block/block-copy.h"
  23 #include "qapi/error.h"
  24 #include "qapi/qmp/qerror.h"
  25 #include "qemu/cutils.h"
  26 #include "sysemu/block-backend.h"
  27 #include "qemu/bitmap.h"
  28 #include "qemu/error-report.h"
  29
  30 #include "block/backup-top.h"
  31
  32 #define BACKUP_CLUSTER_SIZE_DEFAULT (1 << 16)
  33
  34 typedef struct BackupBlockJob {
  35     BlockJob common;
  36     BlockDriverState *backup_top;
  37     BlockDriverState *source_bs;
  38     BlockDriverState *target_bs;
  39
  40     BdrvDirtyBitmap *sync_bitmap;
  41
  42     MirrorSyncMode sync_mode;
  43     BitmapSyncMode bitmap_mode;
  44     BlockdevOnError on_source_error;
  45     BlockdevOnError on_target_error;
  46     uint64_t len;
  47     int64_t cluster_size;
  48     BackupPerf perf;
  49
  50     BlockCopyState *bcs;
  51
  52     bool wait;
  53     BlockCopyCallState *bg_bcs_call;
  54 } BackupBlockJob;
  55
  56 static const BlockJobDriver backup_job_driver;
  57
  58 static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
  59 {
  60     BdrvDirtyBitmap *bm;
  61     bool sync = (((ret == 0) || (job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS)) \
  62                  && (job->bitmap_mode != BITMAP_SYNC_MODE_NEVER));
  63
  64     if (sync) {
  65         /*
  66          * We succeeded, or we always intended to sync the bitmap.
  67          * Delete this bitmap and install the child.
  68          */
  69         bm = bdrv_dirty_bitmap_abdicate(job->sync_bitmap, NULL);
  70     } else {
  71         /*
  72          * We failed, or we never intended to sync the bitmap anyway.
  73          * Merge the successor back into the parent, keeping all data.
  74          */
  75         bm = bdrv_reclaim_dirty_bitmap(job->sync_bitmap, NULL);
  76     }
  77
  78     assert(bm);
  79
  80     if (ret < 0 && job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS) {
  81         /* If we failed and synced, merge in the bits we didn't copy: */
  82         bdrv_dirty_bitmap_merge_internal(bm, block_copy_dirty_bitmap(job->bcs),
  83                                          NULL, true);
  84     }
  85 }
  86
  87 static void backup_commit(Job *job)
  88 {
  89     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
  90     if (s->sync_bitmap) {
  91         backup_cleanup_sync_bitmap(s, 0);
  92     }
  93 }
  94
  95 static void backup_abort(Job *job)
  96 {
  97     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
  98     if (s->sync_bitmap) {
  99         backup_cleanup_sync_bitmap(s, -1);
 100     }
 101 }
 102
 103 static void backup_clean(Job *job)
 104 {
 105     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 106     bdrv_backup_top_drop(s->backup_top);
 107 }
 108
 109 void backup_do_checkpoint(BlockJob *job, Error **errp)
 110 {
 111     BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
 112
 113     assert(block_job_driver(job) == &backup_job_driver);
 114
 115     if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
 116         error_setg(errp, "The backup job only supports block checkpoint in"
 117                    " sync=none mode");
 118         return;
 119     }
 120
 121     bdrv_set_dirty_bitmap(block_copy_dirty_bitmap(backup_job->bcs), 0,
 122                           backup_job->len);
 123 }
 124
 125 static BlockErrorAction backup_error_action(BackupBlockJob *job,
 126                                             bool read, int error)
 127 {
 128     if (read) {
 129         return block_job_error_action(&job->common, job->on_source_error,
 130                                       true, error);
 131     } else {
 132         return block_job_error_action(&job->common, job->on_target_error,
 133                                       false, error);
 134     }
 135 }
 136
 137 static void coroutine_fn backup_block_copy_callback(void *opaque)
 138 {
 139     BackupBlockJob *s = opaque;
 140
 141     if (s->wait) {
 142         s->wait = false;
 143         aio_co_wake(s->common.job.co);
 144     } else {
 145         job_enter(&s->common.job);
 146     }
 147 }
 148
 149 static int coroutine_fn backup_loop(BackupBlockJob *job)
 150 {
 151     BlockCopyCallState *s = NULL;
 152     int ret = 0;
 153     bool error_is_read;
 154     BlockErrorAction act;
 155
 156     while (true) { /* retry loop */
 157         job->bg_bcs_call = s = block_copy_async(job->bcs, 0,
 158                 QEMU_ALIGN_UP(job->len, job->cluster_size),
 159                 job->perf.max_workers, job->perf.max_chunk,
 160                 backup_block_copy_callback, job);
 161
 162         while (!block_copy_call_finished(s) &&
 163                !job_is_cancelled(&job->common.job))
 164         {
 165             job_yield(&job->common.job);
 166         }
 167
 168         if (!block_copy_call_finished(s)) {
 169             assert(job_is_cancelled(&job->common.job));
 170             /*
 171              * Note that we can't use job_yield() here, as it doesn't work for
 172              * cancelled job.
 173              */
 174             block_copy_call_cancel(s);
 175             job->wait = true;
 176             qemu_coroutine_yield();
 177             assert(block_copy_call_finished(s));
 178             ret = 0;
 179             goto out;
 180         }
 181
 182         if (job_is_cancelled(&job->common.job) ||
 183             block_copy_call_succeeded(s))
 184         {
 185             ret = 0;
 186             goto out;
 187         }
 188
 189         if (block_copy_call_cancelled(s)) {
 190             /*
 191              * Job is not cancelled but only block-copy call. This is possible
 192              * after job pause. Now the pause is finished, start new block-copy
 193              * iteration.
 194              */
 195             block_copy_call_free(s);
 196             continue;
 197         }
 198
 199         /* The only remaining case is failed block-copy call. */
 200         assert(block_copy_call_failed(s));
 201
 202         ret = block_copy_call_status(s, &error_is_read);
 203         act = backup_error_action(job, error_is_read, -ret);
 204         switch (act) {
 205         case BLOCK_ERROR_ACTION_REPORT:
 206             goto out;
 207         case BLOCK_ERROR_ACTION_STOP:
 208             /*
 209              * Go to pause prior to starting new block-copy call on the next
 210              * iteration.
 211              */
 212             job_pause_point(&job->common.job);
 213             break;
 214         case BLOCK_ERROR_ACTION_IGNORE:
 215             /* Proceed to new block-copy call to retry. */
 216             break;
 217         default:
 218             abort();
 219         }
 220
 221         block_copy_call_free(s);
 222     }
 223
 224 out:
 225     block_copy_call_free(s);
 226     job->bg_bcs_call = NULL;
 227     return ret;
 228 }
 229
 230 static void backup_init_bcs_bitmap(BackupBlockJob *job)
 231 {
 232     bool ret;
 233     uint64_t estimate;
 234     BdrvDirtyBitmap *bcs_bitmap = block_copy_dirty_bitmap(job->bcs);
 235
 236     if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) {
 237         ret = bdrv_dirty_bitmap_merge_internal(bcs_bitmap, job->sync_bitmap,
 238                                                NULL, true);
 239         assert(ret);
 240     } else {
 241         if (job->sync_mode == MIRROR_SYNC_MODE_TOP) {
 242             /*
 243              * We can't hog the coroutine to initialize this thoroughly.
 244              * Set a flag and resume work when we are able to yield safely.
 245              */
 246             block_copy_set_skip_unallocated(job->bcs, true);
 247         }
 248         bdrv_set_dirty_bitmap(bcs_bitmap, 0, job->len);
 249     }
 250
 251     estimate = bdrv_get_dirty_count(bcs_bitmap);
 252     job_progress_set_remaining(&job->common.job, estimate);
 253 }
 254
 255 static int coroutine_fn backup_run(Job *job, Error **errp)
 256 {
 257     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 258     int ret;
 259
 260     backup_init_bcs_bitmap(s);
 261
 262     if (s->sync_mode == MIRROR_SYNC_MODE_TOP) {
 263         int64_t offset = 0;
 264         int64_t count;
 265
 266         for (offset = 0; offset < s->len; ) {
 267             if (job_is_cancelled(job)) {
 268                 return -ECANCELED;
 269             }
 270
 271             job_pause_point(job);
 272
 273             if (job_is_cancelled(job)) {
 274                 return -ECANCELED;
 275             }
 276
 277             ret = block_copy_reset_unallocated(s->bcs, offset, &count);
 278             if (ret < 0) {
 279                 return ret;
 280             }
 281
 282             offset += count;
 283         }
 284         block_copy_set_skip_unallocated(s->bcs, false);
 285     }
 286
 287     if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
 288         /*
 289          * All bits are set in bcs bitmap to allow any cluster to be copied.
 290          * This does not actually require them to be copied.
 291          */
 292         while (!job_is_cancelled(job)) {
 293             /*
 294              * Yield until the job is cancelled.  We just let our before_write
 295              * notify callback service CoW requests.
 296              */
 297             job_yield(job);
 298         }
 299     } else {
 300         return backup_loop(s);
 301     }
 302
 303     return 0;
 304 }
 305
 306 static void coroutine_fn backup_pause(Job *job)
 307 {
 308     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 309
 310     if (s->bg_bcs_call && !block_copy_call_finished(s->bg_bcs_call)) {
 311         block_copy_call_cancel(s->bg_bcs_call);
 312         s->wait = true;
 313         qemu_coroutine_yield();
 314     }
 315 }
 316
 317 static void coroutine_fn backup_set_speed(BlockJob *job, int64_t speed)
 318 {
 319     BackupBlockJob *s = container_of(job, BackupBlockJob, common);
 320
 321     /*
 322      * block_job_set_speed() is called first from block_job_create(), when we
 323      * don't yet have s->bcs.
 324      */
 325     if (s->bcs) {
 326         block_copy_set_speed(s->bcs, speed);
 327         if (s->bg_bcs_call) {
 328             block_copy_kick(s->bg_bcs_call);
 329         }
 330     }
 331 }
 332
 333 static void backup_cancel(Job *job)
 334 {
 335     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 336
 337     bdrv_cancel_in_flight(s->target_bs);
 338 }
 339
 340 static const BlockJobDriver backup_job_driver = {
 341     .job_driver = {
 342         .instance_size          = sizeof(BackupBlockJob),
 343         .job_type               = JOB_TYPE_BACKUP,
 344         .free                   = block_job_free,
 345         .user_resume            = block_job_user_resume,
 346         .run                    = backup_run,
 347         .commit                 = backup_commit,
 348         .abort                  = backup_abort,
 349         .clean                  = backup_clean,
 350         .pause                  = backup_pause,
 351         .cancel                 = backup_cancel,
 352     },
 353     .set_speed = backup_set_speed,
 354 };
 355
 356 static int64_t backup_calculate_cluster_size(BlockDriverState *target,
 357                                              Error **errp)
 358 {
 359     int ret;
 360     BlockDriverInfo bdi;
 361     bool target_does_cow = bdrv_backing_chain_next(target);
 362
 363     /*
 364      * If there is no backing file on the target, we cannot rely on COW if our
 365      * backup cluster size is smaller than the target cluster size. Even for
 366      * targets with a backing file, try to avoid COW if possible.
 367      */
 368     ret = bdrv_get_info(target, &bdi);
 369     if (ret == -ENOTSUP && !target_does_cow) {
 370         /* Cluster size is not defined */
 371         warn_report("The target block device doesn't provide "
 372                     "information about the block size and it doesn't have a "
 373                     "backing file. The default block size of %u bytes is "
 374                     "used. If the actual block size of the target exceeds "
 375                     "this default, the backup may be unusable",
 376                     BACKUP_CLUSTER_SIZE_DEFAULT);
 377         return BACKUP_CLUSTER_SIZE_DEFAULT;
 378     } else if (ret < 0 && !target_does_cow) {
 379         error_setg_errno(errp, -ret,
 380             "Couldn't determine the cluster size of the target image, "
 381             "which has no backing file");
 382         error_append_hint(errp,
 383             "Aborting, since this may create an unusable destination image\n");
 384         return ret;
 385     } else if (ret < 0 && target_does_cow) {
 386         /* Not fatal; just trudge on ahead. */
 387         return BACKUP_CLUSTER_SIZE_DEFAULT;
 388     }
 389
 390     return MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
 391 }
 392
 393 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
 394                   BlockDriverState *target, int64_t speed,
 395                   MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
 396                   BitmapSyncMode bitmap_mode,
 397                   bool compress,
 398                   const char *filter_node_name,
 399                   BackupPerf *perf,
 400                   BlockdevOnError on_source_error,
 401                   BlockdevOnError on_target_error,
 402                   int creation_flags,
 403                   BlockCompletionFunc *cb, void *opaque,
 404                   JobTxn *txn, Error **errp)
 405 {
 406     int64_t len, target_len;
 407     BackupBlockJob *job = NULL;
 408     int64_t cluster_size;
 409     BdrvRequestFlags write_flags;
 410     BlockDriverState *backup_top = NULL;
 411     BlockCopyState *bcs = NULL;
 412
 413     assert(bs);
 414     assert(target);
 415
 416     /* QMP interface protects us from these cases */
 417     assert(sync_mode != MIRROR_SYNC_MODE_INCREMENTAL);
 418     assert(sync_bitmap || sync_mode != MIRROR_SYNC_MODE_BITMAP);
 419
 420     if (bs == target) {
 421         error_setg(errp, "Source and target cannot be the same");
 422         return NULL;
 423     }
 424
 425     if (!bdrv_is_inserted(bs)) {
 426         error_setg(errp, "Device is not inserted: %s",
 427                    bdrv_get_device_name(bs));
 428         return NULL;
 429     }
 430
 431     if (!bdrv_is_inserted(target)) {
 432         error_setg(errp, "Device is not inserted: %s",
 433                    bdrv_get_device_name(target));
 434         return NULL;
 435     }
 436
 437     if (compress && !bdrv_supports_compressed_writes(target)) {
 438         error_setg(errp, "Compression is not supported for this drive %s",
 439                    bdrv_get_device_name(target));
 440         return NULL;
 441     }
 442
 443     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
 444         return NULL;
 445     }
 446
 447     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
 448         return NULL;
 449     }
 450
 451     cluster_size = backup_calculate_cluster_size(target, errp);
 452     if (cluster_size < 0) {
 453         goto error;
 454     }
 455
 456     if (perf->max_workers < 1) {
 457         error_setg(errp, "max-workers must be greater than zero");
 458         return NULL;
 459     }
 460
 461     if (perf->max_chunk < 0) {
 462         error_setg(errp, "max-chunk must be zero (which means no limit) or "
 463                    "positive");
 464         return NULL;
 465     }
 466
 467     if (perf->max_chunk && perf->max_chunk < cluster_size) {
 468         error_setg(errp, "Required max-chunk (%" PRIi64 ") is less than backup "
 469                    "cluster size (%" PRIi64 ")", perf->max_chunk, cluster_size);
 470         return NULL;
 471     }
 472
 473
 474     if (sync_bitmap) {
 475         /* If we need to write to this bitmap, check that we can: */
 476         if (bitmap_mode != BITMAP_SYNC_MODE_NEVER &&
 477             bdrv_dirty_bitmap_check(sync_bitmap, BDRV_BITMAP_DEFAULT, errp)) {
 478             return NULL;
 479         }
 480
 481         /* Create a new bitmap, and freeze/disable this one. */
 482         if (bdrv_dirty_bitmap_create_successor(sync_bitmap, errp) < 0) {
 483             return NULL;
 484         }
 485     }
 486
 487     len = bdrv_getlength(bs);
 488     if (len < 0) {
 489         error_setg_errno(errp, -len, "Unable to get length for '%s'",
 490                          bdrv_get_device_or_node_name(bs));
 491         goto error;
 492     }
 493
 494     target_len = bdrv_getlength(target);
 495     if (target_len < 0) {
 496         error_setg_errno(errp, -target_len, "Unable to get length for '%s'",
 497                          bdrv_get_device_or_node_name(bs));
 498         goto error;
 499     }
 500
 501     if (target_len != len) {
 502         error_setg(errp, "Source and target image have different sizes");
 503         goto error;
 504     }
 505
 506     /*
 507      * If source is in backing chain of target assume that target is going to be
 508      * used for "image fleecing", i.e. it should represent a kind of snapshot of
 509      * source at backup-start point in time. And target is going to be read by
 510      * somebody (for example, used as NBD export) during backup job.
 511      *
 512      * In this case, we need to add BDRV_REQ_SERIALISING write flag to avoid
 513      * intersection of backup writes and third party reads from target,
 514      * otherwise reading from target we may occasionally read already updated by
 515      * guest data.
 516      *
 517      * For more information see commit f8d59dfb40bb and test
 518      * tests/qemu-iotests/222
 519      */
 520     write_flags = (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |
 521                   (compress ? BDRV_REQ_WRITE_COMPRESSED : 0),
 522
 523     backup_top = bdrv_backup_top_append(bs, target, filter_node_name,
 524                                         cluster_size, perf,
 525                                         write_flags, &bcs, errp);
 526     if (!backup_top) {
 527         goto error;
 528     }
 529
 530     /* job->len is fixed, so we can't allow resize */
 531     job = block_job_create(job_id, &backup_job_driver, txn, backup_top,
 532                            0, BLK_PERM_ALL,
 533                            speed, creation_flags, cb, opaque, errp);
 534     if (!job) {
 535         goto error;
 536     }
 537
 538     job->backup_top = backup_top;
 539     job->source_bs = bs;
 540     job->target_bs = target;
 541     job->on_source_error = on_source_error;
 542     job->on_target_error = on_target_error;
 543     job->sync_mode = sync_mode;
 544     job->sync_bitmap = sync_bitmap;
 545     job->bitmap_mode = bitmap_mode;
 546     job->bcs = bcs;
 547     job->cluster_size = cluster_size;
 548     job->len = len;
 549     job->perf = *perf;
 550
 551     block_copy_set_progress_meter(bcs, &job->common.job.progress);
 552     block_copy_set_speed(bcs, speed);
 553
 554     /* Required permissions are already taken by backup-top target */
 555     block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
 556                        &error_abort);
 557
 558     return &job->common;
 559
 560  error:
 561     if (sync_bitmap) {
 562         bdrv_reclaim_dirty_bitmap(sync_bitmap, NULL);
 563     }
 564     if (backup_top) {
 565         bdrv_backup_top_drop(backup_top);
 566     }
 567
 568     return NULL;
 569 }