4 * Copyright (C) 2013 Proxmox Server Solutions
7 * Dietmar Maurer (dietmar@proxmox.com)
9 * This work is licensed under the terms of the GNU GPL, version 2 or later.
10 * See the COPYING file in the top-level directory.
14 #include "qemu/osdep.h"
17 #include "block/block.h"
18 #include "block/block_int.h"
19 #include "block/blockjob_int.h"
20 #include "block/block_backup.h"
21 #include "qapi/error.h"
22 #include "qapi/qmp/qerror.h"
23 #include "qemu/ratelimit.h"
24 #include "qemu/cutils.h"
25 #include "sysemu/block-backend.h"
26 #include "qemu/bitmap.h"
27 #include "qemu/error-report.h"
29 #define BACKUP_CLUSTER_SIZE_DEFAULT (1 << 16)
31 typedef struct BackupBlockJob
{
34 /* bitmap for sync=incremental */
35 BdrvDirtyBitmap
*sync_bitmap
;
36 MirrorSyncMode sync_mode
;
37 BlockdevOnError on_source_error
;
38 BlockdevOnError on_target_error
;
39 CoRwlock flush_rwlock
;
44 NotifierWithReturn before_write
;
45 QLIST_HEAD(, CowRequest
) inflight_reqs
;
50 static const BlockJobDriver backup_job_driver
;
52 /* See if in-flight requests overlap and wait for them to complete */
53 static void coroutine_fn
wait_for_overlapping_requests(BackupBlockJob
*job
,
62 QLIST_FOREACH(req
, &job
->inflight_reqs
, list
) {
63 if (end
> req
->start_byte
&& start
< req
->end_byte
) {
64 qemu_co_queue_wait(&req
->wait_queue
, NULL
);
72 /* Keep track of an in-flight request */
73 static void cow_request_begin(CowRequest
*req
, BackupBlockJob
*job
,
74 int64_t start
, int64_t end
)
76 req
->start_byte
= start
;
78 qemu_co_queue_init(&req
->wait_queue
);
79 QLIST_INSERT_HEAD(&job
->inflight_reqs
, req
, list
);
82 /* Forget about a completed request */
83 static void cow_request_end(CowRequest
*req
)
85 QLIST_REMOVE(req
, list
);
86 qemu_co_queue_restart_all(&req
->wait_queue
);
89 static int coroutine_fn
backup_do_cow(BackupBlockJob
*job
,
90 int64_t offset
, uint64_t bytes
,
92 bool is_write_notifier
)
94 BlockBackend
*blk
= job
->common
.blk
;
95 CowRequest cow_request
;
97 QEMUIOVector bounce_qiov
;
98 void *bounce_buffer
= NULL
;
100 int64_t start
, end
; /* bytes */
103 qemu_co_rwlock_rdlock(&job
->flush_rwlock
);
105 start
= QEMU_ALIGN_DOWN(offset
, job
->cluster_size
);
106 end
= QEMU_ALIGN_UP(bytes
+ offset
, job
->cluster_size
);
108 trace_backup_do_cow_enter(job
, start
, offset
, bytes
);
110 wait_for_overlapping_requests(job
, start
, end
);
111 cow_request_begin(&cow_request
, job
, start
, end
);
113 for (; start
< end
; start
+= job
->cluster_size
) {
114 if (!hbitmap_get(job
->copy_bitmap
, start
/ job
->cluster_size
)) {
115 trace_backup_do_cow_skip(job
, start
);
116 continue; /* already copied */
118 hbitmap_reset(job
->copy_bitmap
, start
/ job
->cluster_size
, 1);
120 trace_backup_do_cow_process(job
, start
);
122 n
= MIN(job
->cluster_size
, job
->len
- start
);
124 if (!bounce_buffer
) {
125 bounce_buffer
= blk_blockalign(blk
, job
->cluster_size
);
127 iov
.iov_base
= bounce_buffer
;
129 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
131 ret
= blk_co_preadv(blk
, start
, bounce_qiov
.size
, &bounce_qiov
,
132 is_write_notifier
? BDRV_REQ_NO_SERIALISING
: 0);
134 trace_backup_do_cow_read_fail(job
, start
, ret
);
136 *error_is_read
= true;
138 hbitmap_set(job
->copy_bitmap
, start
/ job
->cluster_size
, 1);
142 if (buffer_is_zero(iov
.iov_base
, iov
.iov_len
)) {
143 ret
= blk_co_pwrite_zeroes(job
->target
, start
,
144 bounce_qiov
.size
, BDRV_REQ_MAY_UNMAP
);
146 ret
= blk_co_pwritev(job
->target
, start
,
147 bounce_qiov
.size
, &bounce_qiov
,
148 job
->compress
? BDRV_REQ_WRITE_COMPRESSED
: 0);
151 trace_backup_do_cow_write_fail(job
, start
, ret
);
153 *error_is_read
= false;
155 hbitmap_set(job
->copy_bitmap
, start
/ job
->cluster_size
, 1);
159 /* Publish progress, guest I/O counts as progress too. Note that the
160 * offset field is an opaque progress value, it is not a disk offset.
162 job
->bytes_read
+= n
;
163 block_job_progress_update(&job
->common
, n
);
168 qemu_vfree(bounce_buffer
);
171 cow_request_end(&cow_request
);
173 trace_backup_do_cow_return(job
, offset
, bytes
, ret
);
175 qemu_co_rwlock_unlock(&job
->flush_rwlock
);
180 static int coroutine_fn
backup_before_write_notify(
181 NotifierWithReturn
*notifier
,
184 BackupBlockJob
*job
= container_of(notifier
, BackupBlockJob
, before_write
);
185 BdrvTrackedRequest
*req
= opaque
;
187 assert(req
->bs
== blk_bs(job
->common
.blk
));
188 assert(QEMU_IS_ALIGNED(req
->offset
, BDRV_SECTOR_SIZE
));
189 assert(QEMU_IS_ALIGNED(req
->bytes
, BDRV_SECTOR_SIZE
));
191 return backup_do_cow(job
, req
->offset
, req
->bytes
, NULL
, true);
194 static void backup_cleanup_sync_bitmap(BackupBlockJob
*job
, int ret
)
197 BlockDriverState
*bs
= blk_bs(job
->common
.blk
);
200 /* Merge the successor back into the parent, delete nothing. */
201 bm
= bdrv_reclaim_dirty_bitmap(bs
, job
->sync_bitmap
, NULL
);
204 /* Everything is fine, delete this bitmap and install the backup. */
205 bm
= bdrv_dirty_bitmap_abdicate(bs
, job
->sync_bitmap
, NULL
);
210 static void backup_commit(BlockJob
*job
)
212 BackupBlockJob
*s
= container_of(job
, BackupBlockJob
, common
);
213 if (s
->sync_bitmap
) {
214 backup_cleanup_sync_bitmap(s
, 0);
218 static void backup_abort(BlockJob
*job
)
220 BackupBlockJob
*s
= container_of(job
, BackupBlockJob
, common
);
221 if (s
->sync_bitmap
) {
222 backup_cleanup_sync_bitmap(s
, -1);
226 static void backup_clean(BlockJob
*job
)
228 BackupBlockJob
*s
= container_of(job
, BackupBlockJob
, common
);
230 blk_unref(s
->target
);
234 static void backup_attached_aio_context(BlockJob
*job
, AioContext
*aio_context
)
236 BackupBlockJob
*s
= container_of(job
, BackupBlockJob
, common
);
238 blk_set_aio_context(s
->target
, aio_context
);
241 void backup_do_checkpoint(BlockJob
*job
, Error
**errp
)
243 BackupBlockJob
*backup_job
= container_of(job
, BackupBlockJob
, common
);
246 assert(block_job_driver(job
) == &backup_job_driver
);
248 if (backup_job
->sync_mode
!= MIRROR_SYNC_MODE_NONE
) {
249 error_setg(errp
, "The backup job only supports block checkpoint in"
254 len
= DIV_ROUND_UP(backup_job
->len
, backup_job
->cluster_size
);
255 hbitmap_set(backup_job
->copy_bitmap
, 0, len
);
258 void backup_wait_for_overlapping_requests(BlockJob
*job
, int64_t offset
,
261 BackupBlockJob
*backup_job
= container_of(job
, BackupBlockJob
, common
);
264 assert(block_job_driver(job
) == &backup_job_driver
);
266 start
= QEMU_ALIGN_DOWN(offset
, backup_job
->cluster_size
);
267 end
= QEMU_ALIGN_UP(offset
+ bytes
, backup_job
->cluster_size
);
268 wait_for_overlapping_requests(backup_job
, start
, end
);
271 void backup_cow_request_begin(CowRequest
*req
, BlockJob
*job
,
272 int64_t offset
, uint64_t bytes
)
274 BackupBlockJob
*backup_job
= container_of(job
, BackupBlockJob
, common
);
277 assert(block_job_driver(job
) == &backup_job_driver
);
279 start
= QEMU_ALIGN_DOWN(offset
, backup_job
->cluster_size
);
280 end
= QEMU_ALIGN_UP(offset
+ bytes
, backup_job
->cluster_size
);
281 cow_request_begin(req
, backup_job
, start
, end
);
284 void backup_cow_request_end(CowRequest
*req
)
286 cow_request_end(req
);
289 static void backup_drain(BlockJob
*job
)
291 BackupBlockJob
*s
= container_of(job
, BackupBlockJob
, common
);
293 /* Need to keep a reference in case blk_drain triggers execution
294 * of backup_complete...
297 BlockBackend
*target
= s
->target
;
304 static BlockErrorAction
backup_error_action(BackupBlockJob
*job
,
305 bool read
, int error
)
308 return block_job_error_action(&job
->common
, job
->on_source_error
,
311 return block_job_error_action(&job
->common
, job
->on_target_error
,
318 } BackupCompleteData
;
320 static void backup_complete(BlockJob
*job
, void *opaque
)
322 BackupCompleteData
*data
= opaque
;
324 block_job_completed(job
, data
->ret
);
328 static bool coroutine_fn
yield_and_check(BackupBlockJob
*job
)
332 if (block_job_is_cancelled(&job
->common
)) {
336 /* We need to yield even for delay_ns = 0 so that bdrv_drain_all() can
337 * return. Without a yield, the VM would not reboot. */
338 delay_ns
= block_job_ratelimit_get_delay(&job
->common
, job
->bytes_read
);
340 block_job_sleep_ns(&job
->common
, delay_ns
);
342 if (block_job_is_cancelled(&job
->common
)) {
349 static int coroutine_fn
backup_run_incremental(BackupBlockJob
*job
)
356 hbitmap_iter_init(&hbi
, job
->copy_bitmap
, 0);
357 while ((cluster
= hbitmap_iter_next(&hbi
)) != -1) {
359 if (yield_and_check(job
)) {
362 ret
= backup_do_cow(job
, cluster
* job
->cluster_size
,
363 job
->cluster_size
, &error_is_read
, false);
364 if (ret
< 0 && backup_error_action(job
, error_is_read
, -ret
) ==
365 BLOCK_ERROR_ACTION_REPORT
)
375 /* init copy_bitmap from sync_bitmap */
376 static void backup_incremental_init_copy_bitmap(BackupBlockJob
*job
)
378 BdrvDirtyBitmapIter
*dbi
;
380 int64_t end
= DIV_ROUND_UP(bdrv_dirty_bitmap_size(job
->sync_bitmap
),
383 dbi
= bdrv_dirty_iter_new(job
->sync_bitmap
);
384 while ((offset
= bdrv_dirty_iter_next(dbi
)) != -1) {
385 int64_t cluster
= offset
/ job
->cluster_size
;
386 int64_t next_cluster
;
388 offset
+= bdrv_dirty_bitmap_granularity(job
->sync_bitmap
);
389 if (offset
>= bdrv_dirty_bitmap_size(job
->sync_bitmap
)) {
390 hbitmap_set(job
->copy_bitmap
, cluster
, end
- cluster
);
394 offset
= bdrv_dirty_bitmap_next_zero(job
->sync_bitmap
, offset
);
396 hbitmap_set(job
->copy_bitmap
, cluster
, end
- cluster
);
400 next_cluster
= DIV_ROUND_UP(offset
, job
->cluster_size
);
401 hbitmap_set(job
->copy_bitmap
, cluster
, next_cluster
- cluster
);
402 if (next_cluster
>= end
) {
406 bdrv_set_dirty_iter(dbi
, next_cluster
* job
->cluster_size
);
409 /* TODO block_job_progress_set_remaining() would make more sense */
410 block_job_progress_update(&job
->common
,
411 job
->len
- hbitmap_count(job
->copy_bitmap
) * job
->cluster_size
);
413 bdrv_dirty_iter_free(dbi
);
416 static void coroutine_fn
backup_run(void *opaque
)
418 BackupBlockJob
*job
= opaque
;
419 BackupCompleteData
*data
;
420 BlockDriverState
*bs
= blk_bs(job
->common
.blk
);
421 int64_t offset
, nb_clusters
;
424 QLIST_INIT(&job
->inflight_reqs
);
425 qemu_co_rwlock_init(&job
->flush_rwlock
);
427 nb_clusters
= DIV_ROUND_UP(job
->len
, job
->cluster_size
);
428 block_job_progress_set_remaining(&job
->common
, job
->len
);
430 job
->copy_bitmap
= hbitmap_alloc(nb_clusters
, 0);
431 if (job
->sync_mode
== MIRROR_SYNC_MODE_INCREMENTAL
) {
432 backup_incremental_init_copy_bitmap(job
);
434 hbitmap_set(job
->copy_bitmap
, 0, nb_clusters
);
438 job
->before_write
.notify
= backup_before_write_notify
;
439 bdrv_add_before_write_notifier(bs
, &job
->before_write
);
441 if (job
->sync_mode
== MIRROR_SYNC_MODE_NONE
) {
442 /* All bits are set in copy_bitmap to allow any cluster to be copied.
443 * This does not actually require them to be copied. */
444 while (!block_job_is_cancelled(&job
->common
)) {
445 /* Yield until the job is cancelled. We just let our before_write
446 * notify callback service CoW requests. */
447 block_job_yield(&job
->common
);
449 } else if (job
->sync_mode
== MIRROR_SYNC_MODE_INCREMENTAL
) {
450 ret
= backup_run_incremental(job
);
452 /* Both FULL and TOP SYNC_MODE's require copying.. */
453 for (offset
= 0; offset
< job
->len
;
454 offset
+= job
->cluster_size
) {
458 if (yield_and_check(job
)) {
462 if (job
->sync_mode
== MIRROR_SYNC_MODE_TOP
) {
466 /* Check to see if these blocks are already in the
469 for (i
= 0; i
< job
->cluster_size
;) {
470 /* bdrv_is_allocated() only returns true/false based
471 * on the first set of sectors it comes across that
472 * are are all in the same state.
473 * For that reason we must verify each sector in the
474 * backup cluster length. We end up copying more than
475 * needed but at some point that is always the case. */
477 bdrv_is_allocated(bs
, offset
+ i
,
478 job
->cluster_size
- i
, &n
);
481 if (alloced
|| n
== 0) {
486 /* If the above loop never found any sectors that are in
487 * the topmost image, skip this backup. */
492 /* FULL sync mode we copy the whole drive. */
496 ret
= backup_do_cow(job
, offset
, job
->cluster_size
,
497 &error_is_read
, false);
500 /* Depending on error action, fail now or retry cluster */
501 BlockErrorAction action
=
502 backup_error_action(job
, error_is_read
, -ret
);
503 if (action
== BLOCK_ERROR_ACTION_REPORT
) {
506 offset
-= job
->cluster_size
;
513 notifier_with_return_remove(&job
->before_write
);
515 /* wait until pending backup_do_cow() calls have completed */
516 qemu_co_rwlock_wrlock(&job
->flush_rwlock
);
517 qemu_co_rwlock_unlock(&job
->flush_rwlock
);
518 hbitmap_free(job
->copy_bitmap
);
520 data
= g_malloc(sizeof(*data
));
522 block_job_defer_to_main_loop(&job
->common
, backup_complete
, data
);
525 static const BlockJobDriver backup_job_driver
= {
526 .instance_size
= sizeof(BackupBlockJob
),
527 .job_type
= BLOCK_JOB_TYPE_BACKUP
,
529 .commit
= backup_commit
,
530 .abort
= backup_abort
,
531 .clean
= backup_clean
,
532 .attached_aio_context
= backup_attached_aio_context
,
533 .drain
= backup_drain
,
536 BlockJob
*backup_job_create(const char *job_id
, BlockDriverState
*bs
,
537 BlockDriverState
*target
, int64_t speed
,
538 MirrorSyncMode sync_mode
, BdrvDirtyBitmap
*sync_bitmap
,
540 BlockdevOnError on_source_error
,
541 BlockdevOnError on_target_error
,
543 BlockCompletionFunc
*cb
, void *opaque
,
544 BlockJobTxn
*txn
, Error
**errp
)
548 BackupBlockJob
*job
= NULL
;
555 error_setg(errp
, "Source and target cannot be the same");
559 if (!bdrv_is_inserted(bs
)) {
560 error_setg(errp
, "Device is not inserted: %s",
561 bdrv_get_device_name(bs
));
565 if (!bdrv_is_inserted(target
)) {
566 error_setg(errp
, "Device is not inserted: %s",
567 bdrv_get_device_name(target
));
571 if (compress
&& target
->drv
->bdrv_co_pwritev_compressed
== NULL
) {
572 error_setg(errp
, "Compression is not supported for this drive %s",
573 bdrv_get_device_name(target
));
577 if (bdrv_op_is_blocked(bs
, BLOCK_OP_TYPE_BACKUP_SOURCE
, errp
)) {
581 if (bdrv_op_is_blocked(target
, BLOCK_OP_TYPE_BACKUP_TARGET
, errp
)) {
585 if (sync_mode
== MIRROR_SYNC_MODE_INCREMENTAL
) {
587 error_setg(errp
, "must provide a valid bitmap name for "
588 "\"incremental\" sync mode");
592 /* Create a new bitmap, and freeze/disable this one. */
593 if (bdrv_dirty_bitmap_create_successor(bs
, sync_bitmap
, errp
) < 0) {
596 } else if (sync_bitmap
) {
598 "a sync_bitmap was provided to backup_run, "
599 "but received an incompatible sync_mode (%s)",
600 MirrorSyncMode_str(sync_mode
));
604 len
= bdrv_getlength(bs
);
606 error_setg_errno(errp
, -len
, "unable to get length for '%s'",
607 bdrv_get_device_name(bs
));
611 /* job->len is fixed, so we can't allow resize */
612 job
= block_job_create(job_id
, &backup_job_driver
, txn
, bs
,
613 BLK_PERM_CONSISTENT_READ
,
614 BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE
|
615 BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_GRAPH_MOD
,
616 speed
, creation_flags
, cb
, opaque
, errp
);
621 /* The target must match the source in size, so no resize here either */
622 job
->target
= blk_new(BLK_PERM_WRITE
,
623 BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE
|
624 BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_GRAPH_MOD
);
625 ret
= blk_insert_bs(job
->target
, target
, errp
);
630 job
->on_source_error
= on_source_error
;
631 job
->on_target_error
= on_target_error
;
632 job
->sync_mode
= sync_mode
;
633 job
->sync_bitmap
= sync_mode
== MIRROR_SYNC_MODE_INCREMENTAL
?
635 job
->compress
= compress
;
637 /* If there is no backing file on the target, we cannot rely on COW if our
638 * backup cluster size is smaller than the target cluster size. Even for
639 * targets with a backing file, try to avoid COW if possible. */
640 ret
= bdrv_get_info(target
, &bdi
);
641 if (ret
== -ENOTSUP
&& !target
->backing
) {
642 /* Cluster size is not defined */
643 warn_report("The target block device doesn't provide "
644 "information about the block size and it doesn't have a "
645 "backing file. The default block size of %u bytes is "
646 "used. If the actual block size of the target exceeds "
647 "this default, the backup may be unusable",
648 BACKUP_CLUSTER_SIZE_DEFAULT
);
649 job
->cluster_size
= BACKUP_CLUSTER_SIZE_DEFAULT
;
650 } else if (ret
< 0 && !target
->backing
) {
651 error_setg_errno(errp
, -ret
,
652 "Couldn't determine the cluster size of the target image, "
653 "which has no backing file");
654 error_append_hint(errp
,
655 "Aborting, since this may create an unusable destination image\n");
657 } else if (ret
< 0 && target
->backing
) {
658 /* Not fatal; just trudge on ahead. */
659 job
->cluster_size
= BACKUP_CLUSTER_SIZE_DEFAULT
;
661 job
->cluster_size
= MAX(BACKUP_CLUSTER_SIZE_DEFAULT
, bdi
.cluster_size
);
664 /* Required permissions are already taken with target's blk_new() */
665 block_job_add_bdrv(&job
->common
, "target", target
, 0, BLK_PERM_ALL
,
673 bdrv_reclaim_dirty_bitmap(bs
, sync_bitmap
, NULL
);
676 backup_clean(&job
->common
);
677 block_job_early_fail(&job
->common
);