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