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