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/ratelimit.h"
  26 #include "qemu/cutils.h"
  27 #include "sysemu/block-backend.h"
  28 #include "qemu/bitmap.h"
  29 #include "qemu/error-report.h"
  30
  31 #include "block/backup-top.h"
  32
  33 #define BACKUP_CLUSTER_SIZE_DEFAULT (1 << 16)
  34
  35 typedef struct BackupBlockJob {
  36     BlockJob common;
  37     BlockDriverState *backup_top;
  38     BlockDriverState *source_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     uint64_t bytes_read;
  48     int64_t cluster_size;
  49
  50     BlockCopyState *bcs;
  51 } BackupBlockJob;
  52
  53 static const BlockJobDriver backup_job_driver;
  54
  55 static void backup_progress_bytes_callback(int64_t bytes, void *opaque)
  56 {
  57     BackupBlockJob *s = opaque;
  58
  59     s->bytes_read += bytes;
  60 }
  61
  62 static int coroutine_fn backup_do_cow(BackupBlockJob *job,
  63                                       int64_t offset, uint64_t bytes,
  64                                       bool *error_is_read)
  65 {
  66     int ret = 0;
  67     int64_t start, end; /* bytes */
  68
  69     start = QEMU_ALIGN_DOWN(offset, job->cluster_size);
  70     end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size);
  71
  72     trace_backup_do_cow_enter(job, start, offset, bytes);
  73
  74     ret = block_copy(job->bcs, start, end - start, error_is_read);
  75
  76     trace_backup_do_cow_return(job, offset, bytes, ret);
  77
  78     return ret;
  79 }
  80
  81 static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
  82 {
  83     BdrvDirtyBitmap *bm;
  84     bool sync = (((ret == 0) || (job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS)) \
  85                  && (job->bitmap_mode != BITMAP_SYNC_MODE_NEVER));
  86
  87     if (sync) {
  88         /*
  89          * We succeeded, or we always intended to sync the bitmap.
  90          * Delete this bitmap and install the child.
  91          */
  92         bm = bdrv_dirty_bitmap_abdicate(job->sync_bitmap, NULL);
  93     } else {
  94         /*
  95          * We failed, or we never intended to sync the bitmap anyway.
  96          * Merge the successor back into the parent, keeping all data.
  97          */
  98         bm = bdrv_reclaim_dirty_bitmap(job->sync_bitmap, NULL);
  99     }
 100
 101     assert(bm);
 102
 103     if (ret < 0 && job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS) {
 104         /* If we failed and synced, merge in the bits we didn't copy: */
 105         bdrv_dirty_bitmap_merge_internal(bm, block_copy_dirty_bitmap(job->bcs),
 106                                          NULL, true);
 107     }
 108 }
 109
 110 static void backup_commit(Job *job)
 111 {
 112     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 113     if (s->sync_bitmap) {
 114         backup_cleanup_sync_bitmap(s, 0);
 115     }
 116 }
 117
 118 static void backup_abort(Job *job)
 119 {
 120     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 121     if (s->sync_bitmap) {
 122         backup_cleanup_sync_bitmap(s, -1);
 123     }
 124 }
 125
 126 static void backup_clean(Job *job)
 127 {
 128     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 129     AioContext *aio_context = bdrv_get_aio_context(s->backup_top);
 130
 131     aio_context_acquire(aio_context);
 132     bdrv_backup_top_drop(s->backup_top);
 133     aio_context_release(aio_context);
 134 }
 135
 136 void backup_do_checkpoint(BlockJob *job, Error **errp)
 137 {
 138     BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
 139
 140     assert(block_job_driver(job) == &backup_job_driver);
 141
 142     if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
 143         error_setg(errp, "The backup job only supports block checkpoint in"
 144                    " sync=none mode");
 145         return;
 146     }
 147
 148     bdrv_set_dirty_bitmap(block_copy_dirty_bitmap(backup_job->bcs), 0,
 149                           backup_job->len);
 150 }
 151
 152 static BlockErrorAction backup_error_action(BackupBlockJob *job,
 153                                             bool read, int error)
 154 {
 155     if (read) {
 156         return block_job_error_action(&job->common, job->on_source_error,
 157                                       true, error);
 158     } else {
 159         return block_job_error_action(&job->common, job->on_target_error,
 160                                       false, error);
 161     }
 162 }
 163
 164 static bool coroutine_fn yield_and_check(BackupBlockJob *job)
 165 {
 166     uint64_t delay_ns;
 167
 168     if (job_is_cancelled(&job->common.job)) {
 169         return true;
 170     }
 171
 172     /*
 173      * We need to yield even for delay_ns = 0 so that bdrv_drain_all() can
 174      * return. Without a yield, the VM would not reboot.
 175      */
 176     delay_ns = block_job_ratelimit_get_delay(&job->common, job->bytes_read);
 177     job->bytes_read = 0;
 178     job_sleep_ns(&job->common.job, delay_ns);
 179
 180     if (job_is_cancelled(&job->common.job)) {
 181         return true;
 182     }
 183
 184     return false;
 185 }
 186
 187 static int coroutine_fn backup_loop(BackupBlockJob *job)
 188 {
 189     bool error_is_read;
 190     int64_t offset;
 191     BdrvDirtyBitmapIter *bdbi;
 192     int ret = 0;
 193
 194     bdbi = bdrv_dirty_iter_new(block_copy_dirty_bitmap(job->bcs));
 195     while ((offset = bdrv_dirty_iter_next(bdbi)) != -1) {
 196         do {
 197             if (yield_and_check(job)) {
 198                 goto out;
 199             }
 200             ret = backup_do_cow(job, offset, job->cluster_size, &error_is_read);
 201             if (ret < 0 && backup_error_action(job, error_is_read, -ret) ==
 202                            BLOCK_ERROR_ACTION_REPORT)
 203             {
 204                 goto out;
 205             }
 206         } while (ret < 0);
 207     }
 208
 209  out:
 210     bdrv_dirty_iter_free(bdbi);
 211     return ret;
 212 }
 213
 214 static void backup_init_bcs_bitmap(BackupBlockJob *job)
 215 {
 216     bool ret;
 217     uint64_t estimate;
 218     BdrvDirtyBitmap *bcs_bitmap = block_copy_dirty_bitmap(job->bcs);
 219
 220     if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) {
 221         ret = bdrv_dirty_bitmap_merge_internal(bcs_bitmap, job->sync_bitmap,
 222                                                NULL, true);
 223         assert(ret);
 224     } else {
 225         if (job->sync_mode == MIRROR_SYNC_MODE_TOP) {
 226             /*
 227              * We can't hog the coroutine to initialize this thoroughly.
 228              * Set a flag and resume work when we are able to yield safely.
 229              */
 230             block_copy_set_skip_unallocated(job->bcs, true);
 231         }
 232         bdrv_set_dirty_bitmap(bcs_bitmap, 0, job->len);
 233     }
 234
 235     estimate = bdrv_get_dirty_count(bcs_bitmap);
 236     job_progress_set_remaining(&job->common.job, estimate);
 237 }
 238
 239 static int coroutine_fn backup_run(Job *job, Error **errp)
 240 {
 241     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 242     int ret = 0;
 243
 244     backup_init_bcs_bitmap(s);
 245
 246     if (s->sync_mode == MIRROR_SYNC_MODE_TOP) {
 247         int64_t offset = 0;
 248         int64_t count;
 249
 250         for (offset = 0; offset < s->len; ) {
 251             if (yield_and_check(s)) {
 252                 ret = -ECANCELED;
 253                 goto out;
 254             }
 255
 256             ret = block_copy_reset_unallocated(s->bcs, offset, &count);
 257             if (ret < 0) {
 258                 goto out;
 259             }
 260
 261             offset += count;
 262         }
 263         block_copy_set_skip_unallocated(s->bcs, false);
 264     }
 265
 266     if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
 267         /*
 268          * All bits are set in bcs bitmap to allow any cluster to be copied.
 269          * This does not actually require them to be copied.
 270          */
 271         while (!job_is_cancelled(job)) {
 272             /*
 273              * Yield until the job is cancelled.  We just let our before_write
 274              * notify callback service CoW requests.
 275              */
 276             job_yield(job);
 277         }
 278     } else {
 279         ret = backup_loop(s);
 280     }
 281
 282  out:
 283     return ret;
 284 }
 285
 286 static const BlockJobDriver backup_job_driver = {
 287     .job_driver = {
 288         .instance_size          = sizeof(BackupBlockJob),
 289         .job_type               = JOB_TYPE_BACKUP,
 290         .free                   = block_job_free,
 291         .user_resume            = block_job_user_resume,
 292         .run                    = backup_run,
 293         .commit                 = backup_commit,
 294         .abort                  = backup_abort,
 295         .clean                  = backup_clean,
 296     }
 297 };
 298
 299 static int64_t backup_calculate_cluster_size(BlockDriverState *target,
 300                                              Error **errp)
 301 {
 302     int ret;
 303     BlockDriverInfo bdi;
 304
 305     /*
 306      * If there is no backing file on the target, we cannot rely on COW if our
 307      * backup cluster size is smaller than the target cluster size. Even for
 308      * targets with a backing file, try to avoid COW if possible.
 309      */
 310     ret = bdrv_get_info(target, &bdi);
 311     if (ret == -ENOTSUP && !target->backing) {
 312         /* Cluster size is not defined */
 313         warn_report("The target block device doesn't provide "
 314                     "information about the block size and it doesn't have a "
 315                     "backing file. The default block size of %u bytes is "
 316                     "used. If the actual block size of the target exceeds "
 317                     "this default, the backup may be unusable",
 318                     BACKUP_CLUSTER_SIZE_DEFAULT);
 319         return BACKUP_CLUSTER_SIZE_DEFAULT;
 320     } else if (ret < 0 && !target->backing) {
 321         error_setg_errno(errp, -ret,
 322             "Couldn't determine the cluster size of the target image, "
 323             "which has no backing file");
 324         error_append_hint(errp,
 325             "Aborting, since this may create an unusable destination image\n");
 326         return ret;
 327     } else if (ret < 0 && target->backing) {
 328         /* Not fatal; just trudge on ahead. */
 329         return BACKUP_CLUSTER_SIZE_DEFAULT;
 330     }
 331
 332     return MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
 333 }
 334
 335 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
 336                   BlockDriverState *target, int64_t speed,
 337                   MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
 338                   BitmapSyncMode bitmap_mode,
 339                   bool compress,
 340                   const char *filter_node_name,
 341                   BlockdevOnError on_source_error,
 342                   BlockdevOnError on_target_error,
 343                   int creation_flags,
 344                   BlockCompletionFunc *cb, void *opaque,
 345                   JobTxn *txn, Error **errp)
 346 {
 347     int64_t len;
 348     BackupBlockJob *job = NULL;
 349     int64_t cluster_size;
 350     BdrvRequestFlags write_flags;
 351     BlockDriverState *backup_top = NULL;
 352     BlockCopyState *bcs = NULL;
 353
 354     assert(bs);
 355     assert(target);
 356
 357     /* QMP interface protects us from these cases */
 358     assert(sync_mode != MIRROR_SYNC_MODE_INCREMENTAL);
 359     assert(sync_bitmap || sync_mode != MIRROR_SYNC_MODE_BITMAP);
 360
 361     if (bs == target) {
 362         error_setg(errp, "Source and target cannot be the same");
 363         return NULL;
 364     }
 365
 366     if (!bdrv_is_inserted(bs)) {
 367         error_setg(errp, "Device is not inserted: %s",
 368                    bdrv_get_device_name(bs));
 369         return NULL;
 370     }
 371
 372     if (!bdrv_is_inserted(target)) {
 373         error_setg(errp, "Device is not inserted: %s",
 374                    bdrv_get_device_name(target));
 375         return NULL;
 376     }
 377
 378     if (compress && !block_driver_can_compress(target->drv)) {
 379         error_setg(errp, "Compression is not supported for this drive %s",
 380                    bdrv_get_device_name(target));
 381         return NULL;
 382     }
 383
 384     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
 385         return NULL;
 386     }
 387
 388     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
 389         return NULL;
 390     }
 391
 392     if (sync_bitmap) {
 393         /* If we need to write to this bitmap, check that we can: */
 394         if (bitmap_mode != BITMAP_SYNC_MODE_NEVER &&
 395             bdrv_dirty_bitmap_check(sync_bitmap, BDRV_BITMAP_DEFAULT, errp)) {
 396             return NULL;
 397         }
 398
 399         /* Create a new bitmap, and freeze/disable this one. */
 400         if (bdrv_dirty_bitmap_create_successor(sync_bitmap, errp) < 0) {
 401             return NULL;
 402         }
 403     }
 404
 405     len = bdrv_getlength(bs);
 406     if (len < 0) {
 407         error_setg_errno(errp, -len, "unable to get length for '%s'",
 408                          bdrv_get_device_name(bs));
 409         goto error;
 410     }
 411
 412     cluster_size = backup_calculate_cluster_size(target, errp);
 413     if (cluster_size < 0) {
 414         goto error;
 415     }
 416
 417     /*
 418      * If source is in backing chain of target assume that target is going to be
 419      * used for "image fleecing", i.e. it should represent a kind of snapshot of
 420      * source at backup-start point in time. And target is going to be read by
 421      * somebody (for example, used as NBD export) during backup job.
 422      *
 423      * In this case, we need to add BDRV_REQ_SERIALISING write flag to avoid
 424      * intersection of backup writes and third party reads from target,
 425      * otherwise reading from target we may occasionally read already updated by
 426      * guest data.
 427      *
 428      * For more information see commit f8d59dfb40bb and test
 429      * tests/qemu-iotests/222
 430      */
 431     write_flags = (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |
 432                   (compress ? BDRV_REQ_WRITE_COMPRESSED : 0),
 433
 434     backup_top = bdrv_backup_top_append(bs, target, filter_node_name,
 435                                         cluster_size, write_flags, &bcs, errp);
 436     if (!backup_top) {
 437         goto error;
 438     }
 439
 440     /* job->len is fixed, so we can't allow resize */
 441     job = block_job_create(job_id, &backup_job_driver, txn, backup_top,
 442                            0, BLK_PERM_ALL,
 443                            speed, creation_flags, cb, opaque, errp);
 444     if (!job) {
 445         goto error;
 446     }
 447
 448     job->backup_top = backup_top;
 449     job->source_bs = bs;
 450     job->on_source_error = on_source_error;
 451     job->on_target_error = on_target_error;
 452     job->sync_mode = sync_mode;
 453     job->sync_bitmap = sync_bitmap;
 454     job->bitmap_mode = bitmap_mode;
 455     job->bcs = bcs;
 456     job->cluster_size = cluster_size;
 457     job->len = len;
 458
 459     block_copy_set_progress_callback(bcs, backup_progress_bytes_callback, job);
 460     block_copy_set_progress_meter(bcs, &job->common.job.progress);
 461
 462     /* Required permissions are already taken by backup-top target */
 463     block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
 464                        &error_abort);
 465
 466     return &job->common;
 467
 468  error:
 469     if (sync_bitmap) {
 470         bdrv_reclaim_dirty_bitmap(sync_bitmap, NULL);
 471     }
 472     if (backup_top) {
 473         bdrv_backup_top_drop(backup_top);
 474     }
 475
 476     return NULL;
 477 }