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     block_job_remove_all_bdrv(&s->common);
 107     bdrv_backup_top_drop(s->backup_top);
 108 }
 109
 110 void backup_do_checkpoint(BlockJob *job, Error **errp)
 111 {
 112     BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
 113
 114     assert(block_job_driver(job) == &backup_job_driver);
 115
 116     if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
 117         error_setg(errp, "The backup job only supports block checkpoint in"
 118                    " sync=none mode");
 119         return;
 120     }
 121
 122     bdrv_set_dirty_bitmap(block_copy_dirty_bitmap(backup_job->bcs), 0,
 123                           backup_job->len);
 124 }
 125
 126 static BlockErrorAction backup_error_action(BackupBlockJob *job,
 127                                             bool read, int error)
 128 {
 129     if (read) {
 130         return block_job_error_action(&job->common, job->on_source_error,
 131                                       true, error);
 132     } else {
 133         return block_job_error_action(&job->common, job->on_target_error,
 134                                       false, error);
 135     }
 136 }
 137
 138 static void coroutine_fn backup_block_copy_callback(void *opaque)
 139 {
 140     BackupBlockJob *s = opaque;
 141
 142     if (s->wait) {
 143         s->wait = false;
 144         aio_co_wake(s->common.job.co);
 145     } else {
 146         job_enter(&s->common.job);
 147     }
 148 }
 149
 150 static int coroutine_fn backup_loop(BackupBlockJob *job)
 151 {
 152     BlockCopyCallState *s = NULL;
 153     int ret = 0;
 154     bool error_is_read;
 155     BlockErrorAction act;
 156
 157     while (true) { /* retry loop */
 158         job->bg_bcs_call = s = block_copy_async(job->bcs, 0,
 159                 QEMU_ALIGN_UP(job->len, job->cluster_size),
 160                 job->perf.max_workers, job->perf.max_chunk,
 161                 backup_block_copy_callback, job);
 162
 163         while (!block_copy_call_finished(s) &&
 164                !job_is_cancelled(&job->common.job))
 165         {
 166             job_yield(&job->common.job);
 167         }
 168
 169         if (!block_copy_call_finished(s)) {
 170             assert(job_is_cancelled(&job->common.job));
 171             /*
 172              * Note that we can't use job_yield() here, as it doesn't work for
 173              * cancelled job.
 174              */
 175             block_copy_call_cancel(s);
 176             job->wait = true;
 177             qemu_coroutine_yield();
 178             assert(block_copy_call_finished(s));
 179             ret = 0;
 180             goto out;
 181         }
 182
 183         if (job_is_cancelled(&job->common.job) ||
 184             block_copy_call_succeeded(s))
 185         {
 186             ret = 0;
 187             goto out;
 188         }
 189
 190         if (block_copy_call_cancelled(s)) {
 191             /*
 192              * Job is not cancelled but only block-copy call. This is possible
 193              * after job pause. Now the pause is finished, start new block-copy
 194              * iteration.
 195              */
 196             block_copy_call_free(s);
 197             continue;
 198         }
 199
 200         /* The only remaining case is failed block-copy call. */
 201         assert(block_copy_call_failed(s));
 202
 203         ret = block_copy_call_status(s, &error_is_read);
 204         act = backup_error_action(job, error_is_read, -ret);
 205         switch (act) {
 206         case BLOCK_ERROR_ACTION_REPORT:
 207             goto out;
 208         case BLOCK_ERROR_ACTION_STOP:
 209             /*
 210              * Go to pause prior to starting new block-copy call on the next
 211              * iteration.
 212              */
 213             job_pause_point(&job->common.job);
 214             break;
 215         case BLOCK_ERROR_ACTION_IGNORE:
 216             /* Proceed to new block-copy call to retry. */
 217             break;
 218         default:
 219             abort();
 220         }
 221
 222         block_copy_call_free(s);
 223     }
 224
 225 out:
 226     block_copy_call_free(s);
 227     job->bg_bcs_call = NULL;
 228     return ret;
 229 }
 230
 231 static void backup_init_bcs_bitmap(BackupBlockJob *job)
 232 {
 233     bool ret;
 234     uint64_t estimate;
 235     BdrvDirtyBitmap *bcs_bitmap = block_copy_dirty_bitmap(job->bcs);
 236
 237     if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) {
 238         ret = bdrv_dirty_bitmap_merge_internal(bcs_bitmap, job->sync_bitmap,
 239                                                NULL, true);
 240         assert(ret);
 241     } else {
 242         if (job->sync_mode == MIRROR_SYNC_MODE_TOP) {
 243             /*
 244              * We can't hog the coroutine to initialize this thoroughly.
 245              * Set a flag and resume work when we are able to yield safely.
 246              */
 247             block_copy_set_skip_unallocated(job->bcs, true);
 248         }
 249         bdrv_set_dirty_bitmap(bcs_bitmap, 0, job->len);
 250     }
 251
 252     estimate = bdrv_get_dirty_count(bcs_bitmap);
 253     job_progress_set_remaining(&job->common.job, estimate);
 254 }
 255
 256 static int coroutine_fn backup_run(Job *job, Error **errp)
 257 {
 258     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 259     int ret;
 260
 261     backup_init_bcs_bitmap(s);
 262
 263     if (s->sync_mode == MIRROR_SYNC_MODE_TOP) {
 264         int64_t offset = 0;
 265         int64_t count;
 266
 267         for (offset = 0; offset < s->len; ) {
 268             if (job_is_cancelled(job)) {
 269                 return -ECANCELED;
 270             }
 271
 272             job_pause_point(job);
 273
 274             if (job_is_cancelled(job)) {
 275                 return -ECANCELED;
 276             }
 277
 278             ret = block_copy_reset_unallocated(s->bcs, offset, &count);
 279             if (ret < 0) {
 280                 return ret;
 281             }
 282
 283             offset += count;
 284         }
 285         block_copy_set_skip_unallocated(s->bcs, false);
 286     }
 287
 288     if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
 289         /*
 290          * All bits are set in bcs bitmap to allow any cluster to be copied.
 291          * This does not actually require them to be copied.
 292          */
 293         while (!job_is_cancelled(job)) {
 294             /*
 295              * Yield until the job is cancelled.  We just let our before_write
 296              * notify callback service CoW requests.
 297              */
 298             job_yield(job);
 299         }
 300     } else {
 301         return backup_loop(s);
 302     }
 303
 304     return 0;
 305 }
 306
 307 static void coroutine_fn backup_pause(Job *job)
 308 {
 309     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 310
 311     if (s->bg_bcs_call && !block_copy_call_finished(s->bg_bcs_call)) {
 312         block_copy_call_cancel(s->bg_bcs_call);
 313         s->wait = true;
 314         qemu_coroutine_yield();
 315     }
 316 }
 317
 318 static void coroutine_fn backup_set_speed(BlockJob *job, int64_t speed)
 319 {
 320     BackupBlockJob *s = container_of(job, BackupBlockJob, common);
 321
 322     /*
 323      * block_job_set_speed() is called first from block_job_create(), when we
 324      * don't yet have s->bcs.
 325      */
 326     if (s->bcs) {
 327         block_copy_set_speed(s->bcs, speed);
 328         if (s->bg_bcs_call) {
 329             block_copy_kick(s->bg_bcs_call);
 330         }
 331     }
 332 }
 333
 334 static void backup_cancel(Job *job)
 335 {
 336     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 337
 338     bdrv_cancel_in_flight(s->target_bs);
 339 }
 340
 341 static const BlockJobDriver backup_job_driver = {
 342     .job_driver = {
 343         .instance_size          = sizeof(BackupBlockJob),
 344         .job_type               = JOB_TYPE_BACKUP,
 345         .free                   = block_job_free,
 346         .user_resume            = block_job_user_resume,
 347         .run                    = backup_run,
 348         .commit                 = backup_commit,
 349         .abort                  = backup_abort,
 350         .clean                  = backup_clean,
 351         .pause                  = backup_pause,
 352         .cancel                 = backup_cancel,
 353     },
 354     .set_speed = backup_set_speed,
 355 };
 356
 357 static int64_t backup_calculate_cluster_size(BlockDriverState *target,
 358                                              Error **errp)
 359 {
 360     int ret;
 361     BlockDriverInfo bdi;
 362     bool target_does_cow = bdrv_backing_chain_next(target);
 363
 364     /*
 365      * If there is no backing file on the target, we cannot rely on COW if our
 366      * backup cluster size is smaller than the target cluster size. Even for
 367      * targets with a backing file, try to avoid COW if possible.
 368      */
 369     ret = bdrv_get_info(target, &bdi);
 370     if (ret == -ENOTSUP && !target_does_cow) {
 371         /* Cluster size is not defined */
 372         warn_report("The target block device doesn't provide "
 373                     "information about the block size and it doesn't have a "
 374                     "backing file. The default block size of %u bytes is "
 375                     "used. If the actual block size of the target exceeds "
 376                     "this default, the backup may be unusable",
 377                     BACKUP_CLUSTER_SIZE_DEFAULT);
 378         return BACKUP_CLUSTER_SIZE_DEFAULT;
 379     } else if (ret < 0 && !target_does_cow) {
 380         error_setg_errno(errp, -ret,
 381             "Couldn't determine the cluster size of the target image, "
 382             "which has no backing file");
 383         error_append_hint(errp,
 384             "Aborting, since this may create an unusable destination image\n");
 385         return ret;
 386     } else if (ret < 0 && target_does_cow) {
 387         /* Not fatal; just trudge on ahead. */
 388         return BACKUP_CLUSTER_SIZE_DEFAULT;
 389     }
 390
 391     return MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
 392 }
 393
 394 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
 395                   BlockDriverState *target, int64_t speed,
 396                   MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
 397                   BitmapSyncMode bitmap_mode,
 398                   bool compress,
 399                   const char *filter_node_name,
 400                   BackupPerf *perf,
 401                   BlockdevOnError on_source_error,
 402                   BlockdevOnError on_target_error,
 403                   int creation_flags,
 404                   BlockCompletionFunc *cb, void *opaque,
 405                   JobTxn *txn, Error **errp)
 406 {
 407     int64_t len, target_len;
 408     BackupBlockJob *job = NULL;
 409     int64_t cluster_size;
 410     BdrvRequestFlags write_flags;
 411     BlockDriverState *backup_top = NULL;
 412     BlockCopyState *bcs = NULL;
 413
 414     assert(bs);
 415     assert(target);
 416
 417     /* QMP interface protects us from these cases */
 418     assert(sync_mode != MIRROR_SYNC_MODE_INCREMENTAL);
 419     assert(sync_bitmap || sync_mode != MIRROR_SYNC_MODE_BITMAP);
 420
 421     if (bs == target) {
 422         error_setg(errp, "Source and target cannot be the same");
 423         return NULL;
 424     }
 425
 426     if (!bdrv_is_inserted(bs)) {
 427         error_setg(errp, "Device is not inserted: %s",
 428                    bdrv_get_device_name(bs));
 429         return NULL;
 430     }
 431
 432     if (!bdrv_is_inserted(target)) {
 433         error_setg(errp, "Device is not inserted: %s",
 434                    bdrv_get_device_name(target));
 435         return NULL;
 436     }
 437
 438     if (compress && !bdrv_supports_compressed_writes(target)) {
 439         error_setg(errp, "Compression is not supported for this drive %s",
 440                    bdrv_get_device_name(target));
 441         return NULL;
 442     }
 443
 444     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
 445         return NULL;
 446     }
 447
 448     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
 449         return NULL;
 450     }
 451
 452     cluster_size = backup_calculate_cluster_size(target, errp);
 453     if (cluster_size < 0) {
 454         goto error;
 455     }
 456
 457     if (perf->max_workers < 1) {
 458         error_setg(errp, "max-workers must be greater than zero");
 459         return NULL;
 460     }
 461
 462     if (perf->max_chunk < 0) {
 463         error_setg(errp, "max-chunk must be zero (which means no limit) or "
 464                    "positive");
 465         return NULL;
 466     }
 467
 468     if (perf->max_chunk && perf->max_chunk < cluster_size) {
 469         error_setg(errp, "Required max-chunk (%" PRIi64 ") is less than backup "
 470                    "cluster size (%" PRIi64 ")", perf->max_chunk, cluster_size);
 471         return NULL;
 472     }
 473
 474
 475     if (sync_bitmap) {
 476         /* If we need to write to this bitmap, check that we can: */
 477         if (bitmap_mode != BITMAP_SYNC_MODE_NEVER &&
 478             bdrv_dirty_bitmap_check(sync_bitmap, BDRV_BITMAP_DEFAULT, errp)) {
 479             return NULL;
 480         }
 481
 482         /* Create a new bitmap, and freeze/disable this one. */
 483         if (bdrv_dirty_bitmap_create_successor(sync_bitmap, errp) < 0) {
 484             return NULL;
 485         }
 486     }
 487
 488     len = bdrv_getlength(bs);
 489     if (len < 0) {
 490         error_setg_errno(errp, -len, "Unable to get length for '%s'",
 491                          bdrv_get_device_or_node_name(bs));
 492         goto error;
 493     }
 494
 495     target_len = bdrv_getlength(target);
 496     if (target_len < 0) {
 497         error_setg_errno(errp, -target_len, "Unable to get length for '%s'",
 498                          bdrv_get_device_or_node_name(bs));
 499         goto error;
 500     }
 501
 502     if (target_len != len) {
 503         error_setg(errp, "Source and target image have different sizes");
 504         goto error;
 505     }
 506
 507     /*
 508      * If source is in backing chain of target assume that target is going to be
 509      * used for "image fleecing", i.e. it should represent a kind of snapshot of
 510      * source at backup-start point in time. And target is going to be read by
 511      * somebody (for example, used as NBD export) during backup job.
 512      *
 513      * In this case, we need to add BDRV_REQ_SERIALISING write flag to avoid
 514      * intersection of backup writes and third party reads from target,
 515      * otherwise reading from target we may occasionally read already updated by
 516      * guest data.
 517      *
 518      * For more information see commit f8d59dfb40bb and test
 519      * tests/qemu-iotests/222
 520      */
 521     write_flags = (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |
 522                   (compress ? BDRV_REQ_WRITE_COMPRESSED : 0),
 523
 524     backup_top = bdrv_backup_top_append(bs, target, filter_node_name,
 525                                         cluster_size, perf,
 526                                         write_flags, &bcs, errp);
 527     if (!backup_top) {
 528         goto error;
 529     }
 530
 531     /* job->len is fixed, so we can't allow resize */
 532     job = block_job_create(job_id, &backup_job_driver, txn, backup_top,
 533                            0, BLK_PERM_ALL,
 534                            speed, creation_flags, cb, opaque, errp);
 535     if (!job) {
 536         goto error;
 537     }
 538
 539     job->backup_top = backup_top;
 540     job->source_bs = bs;
 541     job->target_bs = target;
 542     job->on_source_error = on_source_error;
 543     job->on_target_error = on_target_error;
 544     job->sync_mode = sync_mode;
 545     job->sync_bitmap = sync_bitmap;
 546     job->bitmap_mode = bitmap_mode;
 547     job->bcs = bcs;
 548     job->cluster_size = cluster_size;
 549     job->len = len;
 550     job->perf = *perf;
 551
 552     block_copy_set_progress_meter(bcs, &job->common.job.progress);
 553     block_copy_set_speed(bcs, speed);
 554
 555     /* Required permissions are already taken by backup-top target */
 556     block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
 557                        &error_abort);
 558
 559     return &job->common;
 560
 561  error:
 562     if (sync_bitmap) {
 563         bdrv_reclaim_dirty_bitmap(sync_bitmap, NULL);
 564     }
 565     if (backup_top) {
 566         bdrv_backup_top_drop(backup_top);
 567     }
 568
 569     return NULL;
 570 }