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 "block/dirty-bitmap.h"
  24 #include "qapi/error.h"
  25 #include "qapi/qmp/qerror.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/copy-before-write.h"
  32
  33 typedef struct BackupBlockJob {
  34     BlockJob common;
  35     BlockDriverState *cbw;
  36     BlockDriverState *source_bs;
  37     BlockDriverState *target_bs;
  38
  39     BdrvDirtyBitmap *sync_bitmap;
  40
  41     MirrorSyncMode sync_mode;
  42     BitmapSyncMode bitmap_mode;
  43     BlockdevOnError on_source_error;
  44     BlockdevOnError on_target_error;
  45     uint64_t len;
  46     int64_t cluster_size;
  47     BackupPerf perf;
  48
  49     BlockCopyState *bcs;
  50
  51     bool wait;
  52     BlockCopyCallState *bg_bcs_call;
  53 } BackupBlockJob;
  54
  55 static const BlockJobDriver backup_job_driver;
  56
  57 static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
  58 {
  59     BdrvDirtyBitmap *bm;
  60     bool sync = (((ret == 0) || (job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS)) \
  61                  && (job->bitmap_mode != BITMAP_SYNC_MODE_NEVER));
  62
  63     if (sync) {
  64         /*
  65          * We succeeded, or we always intended to sync the bitmap.
  66          * Delete this bitmap and install the child.
  67          */
  68         bm = bdrv_dirty_bitmap_abdicate(job->sync_bitmap, NULL);
  69     } else {
  70         /*
  71          * We failed, or we never intended to sync the bitmap anyway.
  72          * Merge the successor back into the parent, keeping all data.
  73          */
  74         bm = bdrv_reclaim_dirty_bitmap(job->sync_bitmap, NULL);
  75     }
  76
  77     assert(bm);
  78
  79     if (ret < 0 && job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS) {
  80         /* If we failed and synced, merge in the bits we didn't copy: */
  81         bdrv_dirty_bitmap_merge_internal(bm, block_copy_dirty_bitmap(job->bcs),
  82                                          NULL, true);
  83     }
  84 }
  85
  86 static void backup_commit(Job *job)
  87 {
  88     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
  89     if (s->sync_bitmap) {
  90         backup_cleanup_sync_bitmap(s, 0);
  91     }
  92 }
  93
  94 static void backup_abort(Job *job)
  95 {
  96     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
  97     if (s->sync_bitmap) {
  98         backup_cleanup_sync_bitmap(s, -1);
  99     }
 100 }
 101
 102 static void backup_clean(Job *job)
 103 {
 104     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 105     block_job_remove_all_bdrv(&s->common);
 106     bdrv_cbw_drop(s->cbw);
 107 }
 108
 109 void backup_do_checkpoint(BlockJob *job, Error **errp)
 110 {
 111     BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
 112
 113     assert(block_job_driver(job) == &backup_job_driver);
 114
 115     if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
 116         error_setg(errp, "The backup job only supports block checkpoint in"
 117                    " sync=none mode");
 118         return;
 119     }
 120
 121     bdrv_set_dirty_bitmap(block_copy_dirty_bitmap(backup_job->bcs), 0,
 122                           backup_job->len);
 123 }
 124
 125 static BlockErrorAction backup_error_action(BackupBlockJob *job,
 126                                             bool read, int error)
 127 {
 128     if (read) {
 129         return block_job_error_action(&job->common, job->on_source_error,
 130                                       true, error);
 131     } else {
 132         return block_job_error_action(&job->common, job->on_target_error,
 133                                       false, error);
 134     }
 135 }
 136
 137 static void coroutine_fn backup_block_copy_callback(void *opaque)
 138 {
 139     BackupBlockJob *s = opaque;
 140
 141     if (s->wait) {
 142         s->wait = false;
 143         aio_co_wake(s->common.job.co);
 144     } else {
 145         job_enter(&s->common.job);
 146     }
 147 }
 148
 149 static int coroutine_fn backup_loop(BackupBlockJob *job)
 150 {
 151     BlockCopyCallState *s = NULL;
 152     int ret = 0;
 153     bool error_is_read;
 154     BlockErrorAction act;
 155
 156     while (true) { /* retry loop */
 157         job->bg_bcs_call = s = block_copy_async(job->bcs, 0,
 158                 QEMU_ALIGN_UP(job->len, job->cluster_size),
 159                 job->perf.max_workers, job->perf.max_chunk,
 160                 backup_block_copy_callback, job);
 161
 162         while (!block_copy_call_finished(s) &&
 163                !job_is_cancelled(&job->common.job))
 164         {
 165             job_yield(&job->common.job);
 166         }
 167
 168         if (!block_copy_call_finished(s)) {
 169             assert(job_is_cancelled(&job->common.job));
 170             /*
 171              * Note that we can't use job_yield() here, as it doesn't work for
 172              * cancelled job.
 173              */
 174             block_copy_call_cancel(s);
 175             job->wait = true;
 176             qemu_coroutine_yield();
 177             assert(block_copy_call_finished(s));
 178             ret = 0;
 179             goto out;
 180         }
 181
 182         if (job_is_cancelled(&job->common.job) ||
 183             block_copy_call_succeeded(s))
 184         {
 185             ret = 0;
 186             goto out;
 187         }
 188
 189         if (block_copy_call_cancelled(s)) {
 190             /*
 191              * Job is not cancelled but only block-copy call. This is possible
 192              * after job pause. Now the pause is finished, start new block-copy
 193              * iteration.
 194              */
 195             block_copy_call_free(s);
 196             continue;
 197         }
 198
 199         /* The only remaining case is failed block-copy call. */
 200         assert(block_copy_call_failed(s));
 201
 202         ret = block_copy_call_status(s, &error_is_read);
 203         act = backup_error_action(job, error_is_read, -ret);
 204         switch (act) {
 205         case BLOCK_ERROR_ACTION_REPORT:
 206             goto out;
 207         case BLOCK_ERROR_ACTION_STOP:
 208             /*
 209              * Go to pause prior to starting new block-copy call on the next
 210              * iteration.
 211              */
 212             job_pause_point(&job->common.job);
 213             break;
 214         case BLOCK_ERROR_ACTION_IGNORE:
 215             /* Proceed to new block-copy call to retry. */
 216             break;
 217         default:
 218             abort();
 219         }
 220
 221         block_copy_call_free(s);
 222     }
 223
 224 out:
 225     block_copy_call_free(s);
 226     job->bg_bcs_call = NULL;
 227     return ret;
 228 }
 229
 230 static void backup_init_bcs_bitmap(BackupBlockJob *job)
 231 {
 232     uint64_t estimate;
 233     BdrvDirtyBitmap *bcs_bitmap = block_copy_dirty_bitmap(job->bcs);
 234
 235     if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) {
 236         bdrv_clear_dirty_bitmap(bcs_bitmap, NULL);
 237         bdrv_dirty_bitmap_merge_internal(bcs_bitmap, job->sync_bitmap, NULL,
 238                                          true);
 239     } else if (job->sync_mode == MIRROR_SYNC_MODE_TOP) {
 240         /*
 241          * We can't hog the coroutine to initialize this thoroughly.
 242          * Set a flag and resume work when we are able to yield safely.
 243          */
 244         block_copy_set_skip_unallocated(job->bcs, true);
 245     }
 246
 247     estimate = bdrv_get_dirty_count(bcs_bitmap);
 248     job_progress_set_remaining(&job->common.job, estimate);
 249 }
 250
 251 static int coroutine_fn backup_run(Job *job, Error **errp)
 252 {
 253     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 254     int ret;
 255
 256     backup_init_bcs_bitmap(s);
 257
 258     if (s->sync_mode == MIRROR_SYNC_MODE_TOP) {
 259         int64_t offset = 0;
 260         int64_t count;
 261
 262         for (offset = 0; offset < s->len; ) {
 263             if (job_is_cancelled(job)) {
 264                 return -ECANCELED;
 265             }
 266
 267             job_pause_point(job);
 268
 269             if (job_is_cancelled(job)) {
 270                 return -ECANCELED;
 271             }
 272
 273             ret = block_copy_reset_unallocated(s->bcs, offset, &count);
 274             if (ret < 0) {
 275                 return ret;
 276             }
 277
 278             offset += count;
 279         }
 280         block_copy_set_skip_unallocated(s->bcs, false);
 281     }
 282
 283     if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
 284         /*
 285          * All bits are set in bcs bitmap to allow any cluster to be copied.
 286          * This does not actually require them to be copied.
 287          */
 288         while (!job_is_cancelled(job)) {
 289             /*
 290              * Yield until the job is cancelled.  We just let our before_write
 291              * notify callback service CoW requests.
 292              */
 293             job_yield(job);
 294         }
 295     } else {
 296         return backup_loop(s);
 297     }
 298
 299     return 0;
 300 }
 301
 302 static void coroutine_fn backup_pause(Job *job)
 303 {
 304     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 305
 306     if (s->bg_bcs_call && !block_copy_call_finished(s->bg_bcs_call)) {
 307         block_copy_call_cancel(s->bg_bcs_call);
 308         s->wait = true;
 309         qemu_coroutine_yield();
 310     }
 311 }
 312
 313 static void backup_set_speed(BlockJob *job, int64_t speed)
 314 {
 315     BackupBlockJob *s = container_of(job, BackupBlockJob, common);
 316
 317     /*
 318      * block_job_set_speed() is called first from block_job_create(), when we
 319      * don't yet have s->bcs.
 320      */
 321     if (s->bcs) {
 322         block_copy_set_speed(s->bcs, speed);
 323         if (s->bg_bcs_call) {
 324             block_copy_kick(s->bg_bcs_call);
 325         }
 326     }
 327 }
 328
 329 static bool backup_cancel(Job *job, bool force)
 330 {
 331     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
 332
 333     bdrv_cancel_in_flight(s->target_bs);
 334     return true;
 335 }
 336
 337 static const BlockJobDriver backup_job_driver = {
 338     .job_driver = {
 339         .instance_size          = sizeof(BackupBlockJob),
 340         .job_type               = JOB_TYPE_BACKUP,
 341         .free                   = block_job_free,
 342         .user_resume            = block_job_user_resume,
 343         .run                    = backup_run,
 344         .commit                 = backup_commit,
 345         .abort                  = backup_abort,
 346         .clean                  = backup_clean,
 347         .pause                  = backup_pause,
 348         .cancel                 = backup_cancel,
 349     },
 350     .set_speed = backup_set_speed,
 351 };
 352
 353 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
 354                   BlockDriverState *target, int64_t speed,
 355                   MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
 356                   BitmapSyncMode bitmap_mode,
 357                   bool compress,
 358                   const char *filter_node_name,
 359                   BackupPerf *perf,
 360                   BlockdevOnError on_source_error,
 361                   BlockdevOnError on_target_error,
 362                   int creation_flags,
 363                   BlockCompletionFunc *cb, void *opaque,
 364                   JobTxn *txn, Error **errp)
 365 {
 366     int64_t len, target_len;
 367     BackupBlockJob *job = NULL;
 368     int64_t cluster_size;
 369     BlockDriverState *cbw = NULL;
 370     BlockCopyState *bcs = NULL;
 371
 372     assert(bs);
 373     assert(target);
 374     GLOBAL_STATE_CODE();
 375
 376     /* QMP interface protects us from these cases */
 377     assert(sync_mode != MIRROR_SYNC_MODE_INCREMENTAL);
 378     assert(sync_bitmap || sync_mode != MIRROR_SYNC_MODE_BITMAP);
 379
 380     if (bs == target) {
 381         error_setg(errp, "Source and target cannot be the same");
 382         return NULL;
 383     }
 384
 385     if (!bdrv_is_inserted(bs)) {
 386         error_setg(errp, "Device is not inserted: %s",
 387                    bdrv_get_device_name(bs));
 388         return NULL;
 389     }
 390
 391     if (!bdrv_is_inserted(target)) {
 392         error_setg(errp, "Device is not inserted: %s",
 393                    bdrv_get_device_name(target));
 394         return NULL;
 395     }
 396
 397     if (compress && !bdrv_supports_compressed_writes(target)) {
 398         error_setg(errp, "Compression is not supported for this drive %s",
 399                    bdrv_get_device_name(target));
 400         return NULL;
 401     }
 402
 403     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
 404         return NULL;
 405     }
 406
 407     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
 408         return NULL;
 409     }
 410
 411     if (perf->max_workers < 1 || perf->max_workers > INT_MAX) {
 412         error_setg(errp, "max-workers must be between 1 and %d", INT_MAX);
 413         return NULL;
 414     }
 415
 416     if (perf->max_chunk < 0) {
 417         error_setg(errp, "max-chunk must be zero (which means no limit) or "
 418                    "positive");
 419         return NULL;
 420     }
 421
 422     if (sync_bitmap) {
 423         /* If we need to write to this bitmap, check that we can: */
 424         if (bitmap_mode != BITMAP_SYNC_MODE_NEVER &&
 425             bdrv_dirty_bitmap_check(sync_bitmap, BDRV_BITMAP_DEFAULT, errp)) {
 426             return NULL;
 427         }
 428
 429         /* Create a new bitmap, and freeze/disable this one. */
 430         if (bdrv_dirty_bitmap_create_successor(sync_bitmap, errp) < 0) {
 431             return NULL;
 432         }
 433     }
 434
 435     len = bdrv_getlength(bs);
 436     if (len < 0) {
 437         error_setg_errno(errp, -len, "Unable to get length for '%s'",
 438                          bdrv_get_device_or_node_name(bs));
 439         goto error;
 440     }
 441
 442     target_len = bdrv_getlength(target);
 443     if (target_len < 0) {
 444         error_setg_errno(errp, -target_len, "Unable to get length for '%s'",
 445                          bdrv_get_device_or_node_name(bs));
 446         goto error;
 447     }
 448
 449     if (target_len != len) {
 450         error_setg(errp, "Source and target image have different sizes");
 451         goto error;
 452     }
 453
 454     cbw = bdrv_cbw_append(bs, target, filter_node_name, &bcs, errp);
 455     if (!cbw) {
 456         goto error;
 457     }
 458
 459     cluster_size = block_copy_cluster_size(bcs);
 460
 461     if (perf->max_chunk && perf->max_chunk < cluster_size) {
 462         error_setg(errp, "Required max-chunk (%" PRIi64 ") is less than backup "
 463                    "cluster size (%" PRIi64 ")", perf->max_chunk, cluster_size);
 464         goto error;
 465     }
 466
 467     /* job->len is fixed, so we can't allow resize */
 468     job = block_job_create(job_id, &backup_job_driver, txn, cbw,
 469                            0, BLK_PERM_ALL,
 470                            speed, creation_flags, cb, opaque, errp);
 471     if (!job) {
 472         goto error;
 473     }
 474
 475     job->cbw = cbw;
 476     job->source_bs = bs;
 477     job->target_bs = target;
 478     job->on_source_error = on_source_error;
 479     job->on_target_error = on_target_error;
 480     job->sync_mode = sync_mode;
 481     job->sync_bitmap = sync_bitmap;
 482     job->bitmap_mode = bitmap_mode;
 483     job->bcs = bcs;
 484     job->cluster_size = cluster_size;
 485     job->len = len;
 486     job->perf = *perf;
 487
 488     block_copy_set_copy_opts(bcs, perf->use_copy_range, compress);
 489     block_copy_set_progress_meter(bcs, &job->common.job.progress);
 490     block_copy_set_speed(bcs, speed);
 491
 492     /* Required permissions are taken by copy-before-write filter target */
 493     block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
 494                        &error_abort);
 495
 496     return &job->common;
 497
 498  error:
 499     if (sync_bitmap) {
 500         bdrv_reclaim_dirty_bitmap(sync_bitmap, NULL);
 501     }
 502     if (cbw) {
 503         bdrv_cbw_drop(cbw);
 504     }
 505
 506     return NULL;
 507 }