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     BackupPerf perf;
  50
  51     BlockCopyState *bcs;
  52 } BackupBlockJob;
  53
  54 static const BlockJobDriver backup_job_driver;
  55
  56 static void backup_progress_bytes_callback(int64_t bytes, void *opaque)
  57 {
  58     BackupBlockJob *s = opaque;
  59
  60     s->bytes_read += bytes;
  61 }
  62
  63 static int coroutine_fn backup_do_cow(BackupBlockJob *job,
  64                                       int64_t offset, uint64_t bytes,
  65                                       bool *error_is_read)
  66 {
  67     int ret = 0;
  68     int64_t start, end; /* bytes */
  69
  70     start = QEMU_ALIGN_DOWN(offset, job->cluster_size);
  71     end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size);
  72
  73     trace_backup_do_cow_enter(job, start, offset, bytes);
  74
  75     ret = block_copy(job->bcs, start, end - start, error_is_read);
  76
  77     trace_backup_do_cow_return(job, offset, bytes, ret);
  78
  79     return ret;
  80 }
  81
  82 static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
  83 {
  84     BdrvDirtyBitmap *bm;
  85     bool sync = (((ret == 0) || (job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS)) \
  86                  && (job->bitmap_mode != BITMAP_SYNC_MODE_NEVER));
  87
  88     if (sync) {
  89         /*
  90          * We succeeded, or we always intended to sync the bitmap.
  91          * Delete this bitmap and install the child.
  92          */
  93         bm = bdrv_dirty_bitmap_abdicate(job->sync_bitmap, NULL);
  94     } else {
  95         /*
  96          * We failed, or we never intended to sync the bitmap anyway.
  97          * Merge the successor back into the parent, keeping all data.
  98          */
  99         bm = bdrv_reclaim_dirty_bitmap(job->sync_bitmap, NULL);
 100     }
 101
 102     assert(bm);
 103
 104     if (ret < 0 && job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS) {
 105         /* If we failed and synced, merge in the bits we didn't copy: */
 106         bdrv_dirty_bitmap_merge_internal(bm, block_copy_dirty_bitmap(job->bcs),
 107                                          NULL, true);
 108     }
 109 }
 110
 111 static void backup_commit(Job *job)
 112 {
 113     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 114     if (s->sync_bitmap) {
 115         backup_cleanup_sync_bitmap(s, 0);
 116     }
 117 }
 118
 119 static void backup_abort(Job *job)
 120 {
 121     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 122     if (s->sync_bitmap) {
 123         backup_cleanup_sync_bitmap(s, -1);
 124     }
 125 }
 126
 127 static void backup_clean(Job *job)
 128 {
 129     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 130     bdrv_backup_top_drop(s->backup_top);
 131 }
 132
 133 void backup_do_checkpoint(BlockJob *job, Error **errp)
 134 {
 135     BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
 136
 137     assert(block_job_driver(job) == &backup_job_driver);
 138
 139     if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
 140         error_setg(errp, "The backup job only supports block checkpoint in"
 141                    " sync=none mode");
 142         return;
 143     }
 144
 145     bdrv_set_dirty_bitmap(block_copy_dirty_bitmap(backup_job->bcs), 0,
 146                           backup_job->len);
 147 }
 148
 149 static BlockErrorAction backup_error_action(BackupBlockJob *job,
 150                                             bool read, int error)
 151 {
 152     if (read) {
 153         return block_job_error_action(&job->common, job->on_source_error,
 154                                       true, error);
 155     } else {
 156         return block_job_error_action(&job->common, job->on_target_error,
 157                                       false, error);
 158     }
 159 }
 160
 161 static bool coroutine_fn yield_and_check(BackupBlockJob *job)
 162 {
 163     uint64_t delay_ns;
 164
 165     if (job_is_cancelled(&job->common.job)) {
 166         return true;
 167     }
 168
 169     /*
 170      * We need to yield even for delay_ns = 0 so that bdrv_drain_all() can
 171      * return. Without a yield, the VM would not reboot.
 172      */
 173     delay_ns = block_job_ratelimit_get_delay(&job->common, job->bytes_read);
 174     job->bytes_read = 0;
 175     job_sleep_ns(&job->common.job, delay_ns);
 176
 177     if (job_is_cancelled(&job->common.job)) {
 178         return true;
 179     }
 180
 181     return false;
 182 }
 183
 184 static int coroutine_fn backup_loop(BackupBlockJob *job)
 185 {
 186     bool error_is_read;
 187     int64_t offset;
 188     BdrvDirtyBitmapIter *bdbi;
 189     int ret = 0;
 190
 191     bdbi = bdrv_dirty_iter_new(block_copy_dirty_bitmap(job->bcs));
 192     while ((offset = bdrv_dirty_iter_next(bdbi)) != -1) {
 193         do {
 194             if (yield_and_check(job)) {
 195                 goto out;
 196             }
 197             ret = backup_do_cow(job, offset, job->cluster_size, &error_is_read);
 198             if (ret < 0 && backup_error_action(job, error_is_read, -ret) ==
 199                            BLOCK_ERROR_ACTION_REPORT)
 200             {
 201                 goto out;
 202             }
 203         } while (ret < 0);
 204     }
 205
 206  out:
 207     bdrv_dirty_iter_free(bdbi);
 208     return ret;
 209 }
 210
 211 static void backup_init_bcs_bitmap(BackupBlockJob *job)
 212 {
 213     bool ret;
 214     uint64_t estimate;
 215     BdrvDirtyBitmap *bcs_bitmap = block_copy_dirty_bitmap(job->bcs);
 216
 217     if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) {
 218         ret = bdrv_dirty_bitmap_merge_internal(bcs_bitmap, job->sync_bitmap,
 219                                                NULL, true);
 220         assert(ret);
 221     } else {
 222         if (job->sync_mode == MIRROR_SYNC_MODE_TOP) {
 223             /*
 224              * We can't hog the coroutine to initialize this thoroughly.
 225              * Set a flag and resume work when we are able to yield safely.
 226              */
 227             block_copy_set_skip_unallocated(job->bcs, true);
 228         }
 229         bdrv_set_dirty_bitmap(bcs_bitmap, 0, job->len);
 230     }
 231
 232     estimate = bdrv_get_dirty_count(bcs_bitmap);
 233     job_progress_set_remaining(&job->common.job, estimate);
 234 }
 235
 236 static int coroutine_fn backup_run(Job *job, Error **errp)
 237 {
 238     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 239     int ret = 0;
 240
 241     backup_init_bcs_bitmap(s);
 242
 243     if (s->sync_mode == MIRROR_SYNC_MODE_TOP) {
 244         int64_t offset = 0;
 245         int64_t count;
 246
 247         for (offset = 0; offset < s->len; ) {
 248             if (yield_and_check(s)) {
 249                 ret = -ECANCELED;
 250                 goto out;
 251             }
 252
 253             ret = block_copy_reset_unallocated(s->bcs, offset, &count);
 254             if (ret < 0) {
 255                 goto out;
 256             }
 257
 258             offset += count;
 259         }
 260         block_copy_set_skip_unallocated(s->bcs, false);
 261     }
 262
 263     if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
 264         /*
 265          * All bits are set in bcs bitmap to allow any cluster to be copied.
 266          * This does not actually require them to be copied.
 267          */
 268         while (!job_is_cancelled(job)) {
 269             /*
 270              * Yield until the job is cancelled.  We just let our before_write
 271              * notify callback service CoW requests.
 272              */
 273             job_yield(job);
 274         }
 275     } else {
 276         ret = backup_loop(s);
 277     }
 278
 279  out:
 280     return ret;
 281 }
 282
 283 static const BlockJobDriver backup_job_driver = {
 284     .job_driver = {
 285         .instance_size          = sizeof(BackupBlockJob),
 286         .job_type               = JOB_TYPE_BACKUP,
 287         .free                   = block_job_free,
 288         .user_resume            = block_job_user_resume,
 289         .run                    = backup_run,
 290         .commit                 = backup_commit,
 291         .abort                  = backup_abort,
 292         .clean                  = backup_clean,
 293     }
 294 };
 295
 296 static int64_t backup_calculate_cluster_size(BlockDriverState *target,
 297                                              Error **errp)
 298 {
 299     int ret;
 300     BlockDriverInfo bdi;
 301     bool target_does_cow = bdrv_backing_chain_next(target);
 302
 303     /*
 304      * If there is no backing file on the target, we cannot rely on COW if our
 305      * backup cluster size is smaller than the target cluster size. Even for
 306      * targets with a backing file, try to avoid COW if possible.
 307      */
 308     ret = bdrv_get_info(target, &bdi);
 309     if (ret == -ENOTSUP && !target_does_cow) {
 310         /* Cluster size is not defined */
 311         warn_report("The target block device doesn't provide "
 312                     "information about the block size and it doesn't have a "
 313                     "backing file. The default block size of %u bytes is "
 314                     "used. If the actual block size of the target exceeds "
 315                     "this default, the backup may be unusable",
 316                     BACKUP_CLUSTER_SIZE_DEFAULT);
 317         return BACKUP_CLUSTER_SIZE_DEFAULT;
 318     } else if (ret < 0 && !target_does_cow) {
 319         error_setg_errno(errp, -ret,
 320             "Couldn't determine the cluster size of the target image, "
 321             "which has no backing file");
 322         error_append_hint(errp,
 323             "Aborting, since this may create an unusable destination image\n");
 324         return ret;
 325     } else if (ret < 0 && target_does_cow) {
 326         /* Not fatal; just trudge on ahead. */
 327         return BACKUP_CLUSTER_SIZE_DEFAULT;
 328     }
 329
 330     return MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
 331 }
 332
 333 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
 334                   BlockDriverState *target, int64_t speed,
 335                   MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
 336                   BitmapSyncMode bitmap_mode,
 337                   bool compress,
 338                   const char *filter_node_name,
 339                   BackupPerf *perf,
 340                   BlockdevOnError on_source_error,
 341                   BlockdevOnError on_target_error,
 342                   int creation_flags,
 343                   BlockCompletionFunc *cb, void *opaque,
 344                   JobTxn *txn, Error **errp)
 345 {
 346     int64_t len, target_len;
 347     BackupBlockJob *job = NULL;
 348     int64_t cluster_size;
 349     BdrvRequestFlags write_flags;
 350     BlockDriverState *backup_top = NULL;
 351     BlockCopyState *bcs = NULL;
 352
 353     assert(bs);
 354     assert(target);
 355
 356     /* QMP interface protects us from these cases */
 357     assert(sync_mode != MIRROR_SYNC_MODE_INCREMENTAL);
 358     assert(sync_bitmap || sync_mode != MIRROR_SYNC_MODE_BITMAP);
 359
 360     if (bs == target) {
 361         error_setg(errp, "Source and target cannot be the same");
 362         return NULL;
 363     }
 364
 365     if (!bdrv_is_inserted(bs)) {
 366         error_setg(errp, "Device is not inserted: %s",
 367                    bdrv_get_device_name(bs));
 368         return NULL;
 369     }
 370
 371     if (!bdrv_is_inserted(target)) {
 372         error_setg(errp, "Device is not inserted: %s",
 373                    bdrv_get_device_name(target));
 374         return NULL;
 375     }
 376
 377     if (compress && !bdrv_supports_compressed_writes(target)) {
 378         error_setg(errp, "Compression is not supported for this drive %s",
 379                    bdrv_get_device_name(target));
 380         return NULL;
 381     }
 382
 383     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
 384         return NULL;
 385     }
 386
 387     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
 388         return NULL;
 389     }
 390
 391     if (sync_bitmap) {
 392         /* If we need to write to this bitmap, check that we can: */
 393         if (bitmap_mode != BITMAP_SYNC_MODE_NEVER &&
 394             bdrv_dirty_bitmap_check(sync_bitmap, BDRV_BITMAP_DEFAULT, errp)) {
 395             return NULL;
 396         }
 397
 398         /* Create a new bitmap, and freeze/disable this one. */
 399         if (bdrv_dirty_bitmap_create_successor(sync_bitmap, errp) < 0) {
 400             return NULL;
 401         }
 402     }
 403
 404     len = bdrv_getlength(bs);
 405     if (len < 0) {
 406         error_setg_errno(errp, -len, "Unable to get length for '%s'",
 407                          bdrv_get_device_or_node_name(bs));
 408         goto error;
 409     }
 410
 411     target_len = bdrv_getlength(target);
 412     if (target_len < 0) {
 413         error_setg_errno(errp, -target_len, "Unable to get length for '%s'",
 414                          bdrv_get_device_or_node_name(bs));
 415         goto error;
 416     }
 417
 418     if (target_len != len) {
 419         error_setg(errp, "Source and target image have different sizes");
 420         goto error;
 421     }
 422
 423     cluster_size = backup_calculate_cluster_size(target, errp);
 424     if (cluster_size < 0) {
 425         goto error;
 426     }
 427
 428     /*
 429      * If source is in backing chain of target assume that target is going to be
 430      * used for "image fleecing", i.e. it should represent a kind of snapshot of
 431      * source at backup-start point in time. And target is going to be read by
 432      * somebody (for example, used as NBD export) during backup job.
 433      *
 434      * In this case, we need to add BDRV_REQ_SERIALISING write flag to avoid
 435      * intersection of backup writes and third party reads from target,
 436      * otherwise reading from target we may occasionally read already updated by
 437      * guest data.
 438      *
 439      * For more information see commit f8d59dfb40bb and test
 440      * tests/qemu-iotests/222
 441      */
 442     write_flags = (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |
 443                   (compress ? BDRV_REQ_WRITE_COMPRESSED : 0),
 444
 445     backup_top = bdrv_backup_top_append(bs, target, filter_node_name,
 446                                         cluster_size, perf,
 447                                         write_flags, &bcs, errp);
 448     if (!backup_top) {
 449         goto error;
 450     }
 451
 452     /* job->len is fixed, so we can't allow resize */
 453     job = block_job_create(job_id, &backup_job_driver, txn, backup_top,
 454                            0, BLK_PERM_ALL,
 455                            speed, creation_flags, cb, opaque, errp);
 456     if (!job) {
 457         goto error;
 458     }
 459
 460     job->backup_top = backup_top;
 461     job->source_bs = bs;
 462     job->on_source_error = on_source_error;
 463     job->on_target_error = on_target_error;
 464     job->sync_mode = sync_mode;
 465     job->sync_bitmap = sync_bitmap;
 466     job->bitmap_mode = bitmap_mode;
 467     job->bcs = bcs;
 468     job->cluster_size = cluster_size;
 469     job->len = len;
 470     job->perf = *perf;
 471
 472     block_copy_set_progress_callback(bcs, backup_progress_bytes_callback, job);
 473     block_copy_set_progress_meter(bcs, &job->common.job.progress);
 474
 475     /* Required permissions are already taken by backup-top target */
 476     block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
 477                        &error_abort);
 478
 479     return &job->common;
 480
 481  error:
 482     if (sync_bitmap) {
 483         bdrv_reclaim_dirty_bitmap(sync_bitmap, NULL);
 484     }
 485     if (backup_top) {
 486         bdrv_backup_top_drop(backup_top);
 487     }
 488
 489     return NULL;
 490 }