4 * Copyright (C) 2013 Proxmox Server Solutions
5 * Copyright (c) 2019 Virtuozzo International GmbH.
8 * Dietmar Maurer (dietmar@proxmox.com)
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.
15 #include "qemu/osdep.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/cutils.h"
26 #include "sysemu/block-backend.h"
27 #include "qemu/bitmap.h"
28 #include "qemu/error-report.h"
30 #include "block/backup-top.h"
32 #define BACKUP_CLUSTER_SIZE_DEFAULT (1 << 16)
34 typedef struct BackupBlockJob {
36 BlockDriverState *backup_top;
37 BlockDriverState *source_bs;
38 BlockDriverState *target_bs;
40 BdrvDirtyBitmap *sync_bitmap;
42 MirrorSyncMode sync_mode;
43 BitmapSyncMode bitmap_mode;
44 BlockdevOnError on_source_error;
45 BlockdevOnError on_target_error;
53 BlockCopyCallState *bg_bcs_call;
56 static const BlockJobDriver backup_job_driver;
58 static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
61 bool sync = (((ret == 0) || (job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS)) \
62 && (job->bitmap_mode != BITMAP_SYNC_MODE_NEVER));
66 * We succeeded, or we always intended to sync the bitmap.
67 * Delete this bitmap and install the child.
69 bm = bdrv_dirty_bitmap_abdicate(job->sync_bitmap, NULL);
72 * We failed, or we never intended to sync the bitmap anyway.
73 * Merge the successor back into the parent, keeping all data.
75 bm = bdrv_reclaim_dirty_bitmap(job->sync_bitmap, NULL);
80 if (ret < 0 && job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS) {
81 /* If we failed and synced, merge in the bits we didn't copy: */
82 bdrv_dirty_bitmap_merge_internal(bm, block_copy_dirty_bitmap(job->bcs),
87 static void backup_commit(Job *job)
89 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
91 backup_cleanup_sync_bitmap(s, 0);
95 static void backup_abort(Job *job)
97 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
99 backup_cleanup_sync_bitmap(s, -1);
103 static void backup_clean(Job *job)
105 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
106 block_job_remove_all_bdrv(&s->common);
107 bdrv_backup_top_drop(s->backup_top);
110 void backup_do_checkpoint(BlockJob *job, Error **errp)
112 BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
114 assert(block_job_driver(job) == &backup_job_driver);
116 if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
117 error_setg(errp, "The backup job only supports block checkpoint in"
122 bdrv_set_dirty_bitmap(block_copy_dirty_bitmap(backup_job->bcs), 0,
126 static BlockErrorAction backup_error_action(BackupBlockJob *job,
127 bool read, int error)
130 return block_job_error_action(&job->common, job->on_source_error,
133 return block_job_error_action(&job->common, job->on_target_error,
138 static void coroutine_fn backup_block_copy_callback(void *opaque)
140 BackupBlockJob *s = opaque;
144 aio_co_wake(s->common.job.co);
146 job_enter(&s->common.job);
150 static int coroutine_fn backup_loop(BackupBlockJob *job)
152 BlockCopyCallState *s = NULL;
155 BlockErrorAction act;
157 while (true) { /* retry loop */
158 job->bg_bcs_call = s = block_copy_async(job->bcs, 0,
159 QEMU_ALIGN_UP(job->len, job->cluster_size),
160 job->perf.max_workers, job->perf.max_chunk,
161 backup_block_copy_callback, job);
163 while (!block_copy_call_finished(s) &&
164 !job_is_cancelled(&job->common.job))
166 job_yield(&job->common.job);
169 if (!block_copy_call_finished(s)) {
170 assert(job_is_cancelled(&job->common.job));
172 * Note that we can't use job_yield() here, as it doesn't work for
175 block_copy_call_cancel(s);
177 qemu_coroutine_yield();
178 assert(block_copy_call_finished(s));
183 if (job_is_cancelled(&job->common.job) ||
184 block_copy_call_succeeded(s))
190 if (block_copy_call_cancelled(s)) {
192 * Job is not cancelled but only block-copy call. This is possible
193 * after job pause. Now the pause is finished, start new block-copy
196 block_copy_call_free(s);
200 /* The only remaining case is failed block-copy call. */
201 assert(block_copy_call_failed(s));
203 ret = block_copy_call_status(s, &error_is_read);
204 act = backup_error_action(job, error_is_read, -ret);
206 case BLOCK_ERROR_ACTION_REPORT:
208 case BLOCK_ERROR_ACTION_STOP:
210 * Go to pause prior to starting new block-copy call on the next
213 job_pause_point(&job->common.job);
215 case BLOCK_ERROR_ACTION_IGNORE:
216 /* Proceed to new block-copy call to retry. */
222 block_copy_call_free(s);
226 block_copy_call_free(s);
227 job->bg_bcs_call = NULL;
231 static void backup_init_bcs_bitmap(BackupBlockJob *job)
235 BdrvDirtyBitmap *bcs_bitmap = block_copy_dirty_bitmap(job->bcs);
237 if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) {
238 ret = bdrv_dirty_bitmap_merge_internal(bcs_bitmap, job->sync_bitmap,
242 if (job->sync_mode == MIRROR_SYNC_MODE_TOP) {
244 * We can't hog the coroutine to initialize this thoroughly.
245 * Set a flag and resume work when we are able to yield safely.
247 block_copy_set_skip_unallocated(job->bcs, true);
249 bdrv_set_dirty_bitmap(bcs_bitmap, 0, job->len);
252 estimate = bdrv_get_dirty_count(bcs_bitmap);
253 job_progress_set_remaining(&job->common.job, estimate);
256 static int coroutine_fn backup_run(Job *job, Error **errp)
258 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
261 backup_init_bcs_bitmap(s);
263 if (s->sync_mode == MIRROR_SYNC_MODE_TOP) {
267 for (offset = 0; offset < s->len; ) {
268 if (job_is_cancelled(job)) {
272 job_pause_point(job);
274 if (job_is_cancelled(job)) {
278 ret = block_copy_reset_unallocated(s->bcs, offset, &count);
285 block_copy_set_skip_unallocated(s->bcs, false);
288 if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
290 * All bits are set in bcs bitmap to allow any cluster to be copied.
291 * This does not actually require them to be copied.
293 while (!job_is_cancelled(job)) {
295 * Yield until the job is cancelled. We just let our before_write
296 * notify callback service CoW requests.
301 return backup_loop(s);
307 static void coroutine_fn backup_pause(Job *job)
309 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
311 if (s->bg_bcs_call && !block_copy_call_finished(s->bg_bcs_call)) {
312 block_copy_call_cancel(s->bg_bcs_call);
314 qemu_coroutine_yield();
318 static void coroutine_fn backup_set_speed(BlockJob *job, int64_t speed)
320 BackupBlockJob *s = container_of(job, BackupBlockJob, common);
323 * block_job_set_speed() is called first from block_job_create(), when we
324 * don't yet have s->bcs.
327 block_copy_set_speed(s->bcs, speed);
328 if (s->bg_bcs_call) {
329 block_copy_kick(s->bg_bcs_call);
334 static void backup_cancel(Job *job)
336 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
338 bdrv_cancel_in_flight(s->target_bs);
341 static const BlockJobDriver backup_job_driver = {
343 .instance_size = sizeof(BackupBlockJob),
344 .job_type = JOB_TYPE_BACKUP,
345 .free = block_job_free,
346 .user_resume = block_job_user_resume,
348 .commit = backup_commit,
349 .abort = backup_abort,
350 .clean = backup_clean,
351 .pause = backup_pause,
352 .cancel = backup_cancel,
354 .set_speed = backup_set_speed,
357 static int64_t backup_calculate_cluster_size(BlockDriverState *target,
362 bool target_does_cow = bdrv_backing_chain_next(target);
365 * If there is no backing file on the target, we cannot rely on COW if our
366 * backup cluster size is smaller than the target cluster size. Even for
367 * targets with a backing file, try to avoid COW if possible.
369 ret = bdrv_get_info(target, &bdi);
370 if (ret == -ENOTSUP && !target_does_cow) {
371 /* Cluster size is not defined */
372 warn_report("The target block device doesn't provide "
373 "information about the block size and it doesn't have a "
374 "backing file. The default block size of %u bytes is "
375 "used. If the actual block size of the target exceeds "
376 "this default, the backup may be unusable",
377 BACKUP_CLUSTER_SIZE_DEFAULT);
378 return BACKUP_CLUSTER_SIZE_DEFAULT;
379 } else if (ret < 0 && !target_does_cow) {
380 error_setg_errno(errp, -ret,
381 "Couldn't determine the cluster size of the target image, "
382 "which has no backing file");
383 error_append_hint(errp,
384 "Aborting, since this may create an unusable destination image\n");
386 } else if (ret < 0 && target_does_cow) {
387 /* Not fatal; just trudge on ahead. */
388 return BACKUP_CLUSTER_SIZE_DEFAULT;
391 return MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
394 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
395 BlockDriverState *target, int64_t speed,
396 MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
397 BitmapSyncMode bitmap_mode,
399 const char *filter_node_name,
401 BlockdevOnError on_source_error,
402 BlockdevOnError on_target_error,
404 BlockCompletionFunc *cb, void *opaque,
405 JobTxn *txn, Error **errp)
407 int64_t len, target_len;
408 BackupBlockJob *job = NULL;
409 int64_t cluster_size;
410 BdrvRequestFlags write_flags;
411 BlockDriverState *backup_top = NULL;
412 BlockCopyState *bcs = NULL;
417 /* QMP interface protects us from these cases */
418 assert(sync_mode != MIRROR_SYNC_MODE_INCREMENTAL);
419 assert(sync_bitmap || sync_mode != MIRROR_SYNC_MODE_BITMAP);
422 error_setg(errp, "Source and target cannot be the same");
426 if (!bdrv_is_inserted(bs)) {
427 error_setg(errp, "Device is not inserted: %s",
428 bdrv_get_device_name(bs));
432 if (!bdrv_is_inserted(target)) {
433 error_setg(errp, "Device is not inserted: %s",
434 bdrv_get_device_name(target));
438 if (compress && !bdrv_supports_compressed_writes(target)) {
439 error_setg(errp, "Compression is not supported for this drive %s",
440 bdrv_get_device_name(target));
444 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
448 if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
452 cluster_size = backup_calculate_cluster_size(target, errp);
453 if (cluster_size < 0) {
457 if (perf->max_workers < 1) {
458 error_setg(errp, "max-workers must be greater than zero");
462 if (perf->max_chunk < 0) {
463 error_setg(errp, "max-chunk must be zero (which means no limit) or "
468 if (perf->max_chunk && perf->max_chunk < cluster_size) {
469 error_setg(errp, "Required max-chunk (%" PRIi64 ") is less than backup "
470 "cluster size (%" PRIi64 ")", perf->max_chunk, cluster_size);
476 /* If we need to write to this bitmap, check that we can: */
477 if (bitmap_mode != BITMAP_SYNC_MODE_NEVER &&
478 bdrv_dirty_bitmap_check(sync_bitmap, BDRV_BITMAP_DEFAULT, errp)) {
482 /* Create a new bitmap, and freeze/disable this one. */
483 if (bdrv_dirty_bitmap_create_successor(sync_bitmap, errp) < 0) {
488 len = bdrv_getlength(bs);
490 error_setg_errno(errp, -len, "Unable to get length for '%s'",
491 bdrv_get_device_or_node_name(bs));
495 target_len = bdrv_getlength(target);
496 if (target_len < 0) {
497 error_setg_errno(errp, -target_len, "Unable to get length for '%s'",
498 bdrv_get_device_or_node_name(bs));
502 if (target_len != len) {
503 error_setg(errp, "Source and target image have different sizes");
508 * If source is in backing chain of target assume that target is going to be
509 * used for "image fleecing", i.e. it should represent a kind of snapshot of
510 * source at backup-start point in time. And target is going to be read by
511 * somebody (for example, used as NBD export) during backup job.
513 * In this case, we need to add BDRV_REQ_SERIALISING write flag to avoid
514 * intersection of backup writes and third party reads from target,
515 * otherwise reading from target we may occasionally read already updated by
518 * For more information see commit f8d59dfb40bb and test
519 * tests/qemu-iotests/222
521 write_flags = (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |
522 (compress ? BDRV_REQ_WRITE_COMPRESSED : 0),
524 backup_top = bdrv_backup_top_append(bs, target, filter_node_name,
526 write_flags, &bcs, errp);
531 /* job->len is fixed, so we can't allow resize */
532 job = block_job_create(job_id, &backup_job_driver, txn, backup_top,
534 speed, creation_flags, cb, opaque, errp);
539 job->backup_top = backup_top;
541 job->target_bs = target;
542 job->on_source_error = on_source_error;
543 job->on_target_error = on_target_error;
544 job->sync_mode = sync_mode;
545 job->sync_bitmap = sync_bitmap;
546 job->bitmap_mode = bitmap_mode;
548 job->cluster_size = cluster_size;
552 block_copy_set_progress_meter(bcs, &job->common.job.progress);
553 block_copy_set_speed(bcs, speed);
555 /* Required permissions are already taken by backup-top target */
556 block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
563 bdrv_reclaim_dirty_bitmap(sync_bitmap, NULL);
566 bdrv_backup_top_drop(backup_top);