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