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 CowRequest
{
34 QLIST_ENTRY(CowRequest
) list
;
35 CoQueue wait_queue
; /* coroutines blocked on this request */
38 typedef struct BackupBlockJob
{
41 /* bitmap for sync=incremental */
42 BdrvDirtyBitmap
*sync_bitmap
;
43 MirrorSyncMode sync_mode
;
44 BlockdevOnError on_source_error
;
45 BlockdevOnError on_target_error
;
46 CoRwlock flush_rwlock
;
51 NotifierWithReturn before_write
;
52 QLIST_HEAD(, CowRequest
) inflight_reqs
;
56 int64_t copy_range_size
;
58 bool serialize_target_writes
;
61 static const BlockJobDriver backup_job_driver
;
63 /* See if in-flight requests overlap and wait for them to complete */
64 static void coroutine_fn
wait_for_overlapping_requests(BackupBlockJob
*job
,
73 QLIST_FOREACH(req
, &job
->inflight_reqs
, list
) {
74 if (end
> req
->start_byte
&& start
< req
->end_byte
) {
75 qemu_co_queue_wait(&req
->wait_queue
, NULL
);
83 /* Keep track of an in-flight request */
84 static void cow_request_begin(CowRequest
*req
, BackupBlockJob
*job
,
85 int64_t start
, int64_t end
)
87 req
->start_byte
= start
;
89 qemu_co_queue_init(&req
->wait_queue
);
90 QLIST_INSERT_HEAD(&job
->inflight_reqs
, req
, list
);
93 /* Forget about a completed request */
94 static void cow_request_end(CowRequest
*req
)
96 QLIST_REMOVE(req
, list
);
97 qemu_co_queue_restart_all(&req
->wait_queue
);
100 /* Copy range to target with a bounce buffer and return the bytes copied. If
101 * error occurred, return a negative error number */
102 static int coroutine_fn
backup_cow_with_bounce_buffer(BackupBlockJob
*job
,
105 bool is_write_notifier
,
107 void **bounce_buffer
)
110 BlockBackend
*blk
= job
->common
.blk
;
112 int read_flags
= is_write_notifier
? BDRV_REQ_NO_SERIALISING
: 0;
113 int write_flags
= job
->serialize_target_writes
? BDRV_REQ_SERIALISING
: 0;
115 assert(QEMU_IS_ALIGNED(start
, job
->cluster_size
));
116 hbitmap_reset(job
->copy_bitmap
, start
, job
->cluster_size
);
117 nbytes
= MIN(job
->cluster_size
, job
->len
- start
);
118 if (!*bounce_buffer
) {
119 *bounce_buffer
= blk_blockalign(blk
, job
->cluster_size
);
122 ret
= blk_co_pread(blk
, start
, nbytes
, *bounce_buffer
, read_flags
);
124 trace_backup_do_cow_read_fail(job
, start
, ret
);
126 *error_is_read
= true;
131 if (buffer_is_zero(*bounce_buffer
, nbytes
)) {
132 ret
= blk_co_pwrite_zeroes(job
->target
, start
,
133 nbytes
, write_flags
| BDRV_REQ_MAY_UNMAP
);
135 ret
= blk_co_pwrite(job
->target
, start
,
136 nbytes
, *bounce_buffer
, write_flags
|
137 (job
->compress
? BDRV_REQ_WRITE_COMPRESSED
: 0));
140 trace_backup_do_cow_write_fail(job
, start
, ret
);
142 *error_is_read
= false;
149 hbitmap_set(job
->copy_bitmap
, start
, job
->cluster_size
);
154 /* Copy range to target and return the bytes copied. If error occurred, return a
155 * negative error number. */
156 static int coroutine_fn
backup_cow_with_offload(BackupBlockJob
*job
,
159 bool is_write_notifier
)
163 BlockBackend
*blk
= job
->common
.blk
;
165 int read_flags
= is_write_notifier
? BDRV_REQ_NO_SERIALISING
: 0;
166 int write_flags
= job
->serialize_target_writes
? BDRV_REQ_SERIALISING
: 0;
168 assert(QEMU_IS_ALIGNED(job
->copy_range_size
, job
->cluster_size
));
169 assert(QEMU_IS_ALIGNED(start
, job
->cluster_size
));
170 nbytes
= MIN(job
->copy_range_size
, end
- start
);
171 nr_clusters
= DIV_ROUND_UP(nbytes
, job
->cluster_size
);
172 hbitmap_reset(job
->copy_bitmap
, start
, job
->cluster_size
* nr_clusters
);
173 ret
= blk_co_copy_range(blk
, start
, job
->target
, start
, nbytes
,
174 read_flags
, write_flags
);
176 trace_backup_do_cow_copy_range_fail(job
, start
, ret
);
177 hbitmap_set(job
->copy_bitmap
, start
, job
->cluster_size
* nr_clusters
);
184 static int coroutine_fn
backup_do_cow(BackupBlockJob
*job
,
185 int64_t offset
, uint64_t bytes
,
187 bool is_write_notifier
)
189 CowRequest cow_request
;
191 int64_t start
, end
; /* bytes */
192 void *bounce_buffer
= NULL
;
194 qemu_co_rwlock_rdlock(&job
->flush_rwlock
);
196 start
= QEMU_ALIGN_DOWN(offset
, job
->cluster_size
);
197 end
= QEMU_ALIGN_UP(bytes
+ offset
, job
->cluster_size
);
199 trace_backup_do_cow_enter(job
, start
, offset
, bytes
);
201 wait_for_overlapping_requests(job
, start
, end
);
202 cow_request_begin(&cow_request
, job
, start
, end
);
204 while (start
< end
) {
205 if (!hbitmap_get(job
->copy_bitmap
, start
)) {
206 trace_backup_do_cow_skip(job
, start
);
207 start
+= job
->cluster_size
;
208 continue; /* already copied */
211 trace_backup_do_cow_process(job
, start
);
213 if (job
->use_copy_range
) {
214 ret
= backup_cow_with_offload(job
, start
, end
, is_write_notifier
);
216 job
->use_copy_range
= false;
219 if (!job
->use_copy_range
) {
220 ret
= backup_cow_with_bounce_buffer(job
, start
, end
, is_write_notifier
,
221 error_is_read
, &bounce_buffer
);
227 /* Publish progress, guest I/O counts as progress too. Note that the
228 * offset field is an opaque progress value, it is not a disk offset.
231 job
->bytes_read
+= ret
;
232 job_progress_update(&job
->common
.job
, ret
);
237 qemu_vfree(bounce_buffer
);
240 cow_request_end(&cow_request
);
242 trace_backup_do_cow_return(job
, offset
, bytes
, ret
);
244 qemu_co_rwlock_unlock(&job
->flush_rwlock
);
249 static int coroutine_fn
backup_before_write_notify(
250 NotifierWithReturn
*notifier
,
253 BackupBlockJob
*job
= container_of(notifier
, BackupBlockJob
, before_write
);
254 BdrvTrackedRequest
*req
= opaque
;
256 assert(req
->bs
== blk_bs(job
->common
.blk
));
257 assert(QEMU_IS_ALIGNED(req
->offset
, BDRV_SECTOR_SIZE
));
258 assert(QEMU_IS_ALIGNED(req
->bytes
, BDRV_SECTOR_SIZE
));
260 return backup_do_cow(job
, req
->offset
, req
->bytes
, NULL
, true);
263 static void backup_cleanup_sync_bitmap(BackupBlockJob
*job
, int ret
)
266 BlockDriverState
*bs
= blk_bs(job
->common
.blk
);
269 /* Merge the successor back into the parent, delete nothing. */
270 bm
= bdrv_reclaim_dirty_bitmap(bs
, job
->sync_bitmap
, NULL
);
273 /* Everything is fine, delete this bitmap and install the backup. */
274 bm
= bdrv_dirty_bitmap_abdicate(bs
, job
->sync_bitmap
, NULL
);
279 static void backup_commit(Job
*job
)
281 BackupBlockJob
*s
= container_of(job
, BackupBlockJob
, common
.job
);
282 if (s
->sync_bitmap
) {
283 backup_cleanup_sync_bitmap(s
, 0);
287 static void backup_abort(Job
*job
)
289 BackupBlockJob
*s
= container_of(job
, BackupBlockJob
, common
.job
);
290 if (s
->sync_bitmap
) {
291 backup_cleanup_sync_bitmap(s
, -1);
295 static void backup_clean(Job
*job
)
297 BackupBlockJob
*s
= container_of(job
, BackupBlockJob
, common
.job
);
299 blk_unref(s
->target
);
302 if (s
->copy_bitmap
) {
303 hbitmap_free(s
->copy_bitmap
);
304 s
->copy_bitmap
= NULL
;
308 void backup_do_checkpoint(BlockJob
*job
, Error
**errp
)
310 BackupBlockJob
*backup_job
= container_of(job
, BackupBlockJob
, common
);
312 assert(block_job_driver(job
) == &backup_job_driver
);
314 if (backup_job
->sync_mode
!= MIRROR_SYNC_MODE_NONE
) {
315 error_setg(errp
, "The backup job only supports block checkpoint in"
320 hbitmap_set(backup_job
->copy_bitmap
, 0, backup_job
->len
);
323 static void backup_drain(BlockJob
*job
)
325 BackupBlockJob
*s
= container_of(job
, BackupBlockJob
, common
);
327 /* Need to keep a reference in case blk_drain triggers execution
328 * of backup_complete...
331 BlockBackend
*target
= s
->target
;
338 static BlockErrorAction
backup_error_action(BackupBlockJob
*job
,
339 bool read
, int error
)
342 return block_job_error_action(&job
->common
, job
->on_source_error
,
345 return block_job_error_action(&job
->common
, job
->on_target_error
,
350 static bool coroutine_fn
yield_and_check(BackupBlockJob
*job
)
354 if (job_is_cancelled(&job
->common
.job
)) {
358 /* We need to yield even for delay_ns = 0 so that bdrv_drain_all() can
359 * return. Without a yield, the VM would not reboot. */
360 delay_ns
= block_job_ratelimit_get_delay(&job
->common
, job
->bytes_read
);
362 job_sleep_ns(&job
->common
.job
, delay_ns
);
364 if (job_is_cancelled(&job
->common
.job
)) {
371 static bool bdrv_is_unallocated_range(BlockDriverState
*bs
,
372 int64_t offset
, int64_t bytes
)
374 int64_t end
= offset
+ bytes
;
376 while (offset
< end
&& !bdrv_is_allocated(bs
, offset
, bytes
, &bytes
)) {
381 bytes
= end
- offset
;
384 return offset
>= end
;
387 static int coroutine_fn
backup_loop(BackupBlockJob
*job
)
393 BlockDriverState
*bs
= blk_bs(job
->common
.blk
);
395 hbitmap_iter_init(&hbi
, job
->copy_bitmap
, 0);
396 while ((offset
= hbitmap_iter_next(&hbi
)) != -1) {
397 if (job
->sync_mode
== MIRROR_SYNC_MODE_TOP
&&
398 bdrv_is_unallocated_range(bs
, offset
, job
->cluster_size
))
400 hbitmap_reset(job
->copy_bitmap
, offset
, job
->cluster_size
);
405 if (yield_and_check(job
)) {
408 ret
= backup_do_cow(job
, offset
,
409 job
->cluster_size
, &error_is_read
, false);
410 if (ret
< 0 && backup_error_action(job
, error_is_read
, -ret
) ==
411 BLOCK_ERROR_ACTION_REPORT
)
421 /* init copy_bitmap from sync_bitmap */
422 static void backup_incremental_init_copy_bitmap(BackupBlockJob
*job
)
425 uint64_t bytes
= job
->len
;
427 while (bdrv_dirty_bitmap_next_dirty_area(job
->sync_bitmap
,
430 hbitmap_set(job
->copy_bitmap
, offset
, bytes
);
433 if (offset
>= job
->len
) {
436 bytes
= job
->len
- offset
;
439 /* TODO job_progress_set_remaining() would make more sense */
440 job_progress_update(&job
->common
.job
,
441 job
->len
- hbitmap_count(job
->copy_bitmap
));
444 static int coroutine_fn
backup_run(Job
*job
, Error
**errp
)
446 BackupBlockJob
*s
= container_of(job
, BackupBlockJob
, common
.job
);
447 BlockDriverState
*bs
= blk_bs(s
->common
.blk
);
450 QLIST_INIT(&s
->inflight_reqs
);
451 qemu_co_rwlock_init(&s
->flush_rwlock
);
453 job_progress_set_remaining(job
, s
->len
);
455 if (s
->sync_mode
== MIRROR_SYNC_MODE_INCREMENTAL
) {
456 backup_incremental_init_copy_bitmap(s
);
458 hbitmap_set(s
->copy_bitmap
, 0, s
->len
);
461 s
->before_write
.notify
= backup_before_write_notify
;
462 bdrv_add_before_write_notifier(bs
, &s
->before_write
);
464 if (s
->sync_mode
== MIRROR_SYNC_MODE_NONE
) {
465 /* All bits are set in copy_bitmap to allow any cluster to be copied.
466 * This does not actually require them to be copied. */
467 while (!job_is_cancelled(job
)) {
468 /* Yield until the job is cancelled. We just let our before_write
469 * notify callback service CoW requests. */
473 ret
= backup_loop(s
);
476 notifier_with_return_remove(&s
->before_write
);
478 /* wait until pending backup_do_cow() calls have completed */
479 qemu_co_rwlock_wrlock(&s
->flush_rwlock
);
480 qemu_co_rwlock_unlock(&s
->flush_rwlock
);
485 static const BlockJobDriver backup_job_driver
= {
487 .instance_size
= sizeof(BackupBlockJob
),
488 .job_type
= JOB_TYPE_BACKUP
,
489 .free
= block_job_free
,
490 .user_resume
= block_job_user_resume
,
491 .drain
= block_job_drain
,
493 .commit
= backup_commit
,
494 .abort
= backup_abort
,
495 .clean
= backup_clean
,
497 .drain
= backup_drain
,
500 static int64_t backup_calculate_cluster_size(BlockDriverState
*target
,
507 * If there is no backing file on the target, we cannot rely on COW if our
508 * backup cluster size is smaller than the target cluster size. Even for
509 * targets with a backing file, try to avoid COW if possible.
511 ret
= bdrv_get_info(target
, &bdi
);
512 if (ret
== -ENOTSUP
&& !target
->backing
) {
513 /* Cluster size is not defined */
514 warn_report("The target block device doesn't provide "
515 "information about the block size and it doesn't have a "
516 "backing file. The default block size of %u bytes is "
517 "used. If the actual block size of the target exceeds "
518 "this default, the backup may be unusable",
519 BACKUP_CLUSTER_SIZE_DEFAULT
);
520 return BACKUP_CLUSTER_SIZE_DEFAULT
;
521 } else if (ret
< 0 && !target
->backing
) {
522 error_setg_errno(errp
, -ret
,
523 "Couldn't determine the cluster size of the target image, "
524 "which has no backing file");
525 error_append_hint(errp
,
526 "Aborting, since this may create an unusable destination image\n");
528 } else if (ret
< 0 && target
->backing
) {
529 /* Not fatal; just trudge on ahead. */
530 return BACKUP_CLUSTER_SIZE_DEFAULT
;
533 return MAX(BACKUP_CLUSTER_SIZE_DEFAULT
, bdi
.cluster_size
);
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 JobTxn
*txn
, Error
**errp
)
547 BackupBlockJob
*job
= NULL
;
549 int64_t cluster_size
;
550 HBitmap
*copy_bitmap
= NULL
;
556 error_setg(errp
, "Source and target cannot be the same");
560 if (!bdrv_is_inserted(bs
)) {
561 error_setg(errp
, "Device is not inserted: %s",
562 bdrv_get_device_name(bs
));
566 if (!bdrv_is_inserted(target
)) {
567 error_setg(errp
, "Device is not inserted: %s",
568 bdrv_get_device_name(target
));
572 if (compress
&& target
->drv
->bdrv_co_pwritev_compressed
== NULL
) {
573 error_setg(errp
, "Compression is not supported for this drive %s",
574 bdrv_get_device_name(target
));
578 if (bdrv_op_is_blocked(bs
, BLOCK_OP_TYPE_BACKUP_SOURCE
, errp
)) {
582 if (bdrv_op_is_blocked(target
, BLOCK_OP_TYPE_BACKUP_TARGET
, errp
)) {
586 if (sync_mode
== MIRROR_SYNC_MODE_INCREMENTAL
) {
588 error_setg(errp
, "must provide a valid bitmap name for "
589 "\"incremental\" sync mode");
593 /* Create a new bitmap, and freeze/disable this one. */
594 if (bdrv_dirty_bitmap_create_successor(bs
, sync_bitmap
, errp
) < 0) {
597 } else if (sync_bitmap
) {
599 "a sync_bitmap was provided to backup_run, "
600 "but received an incompatible sync_mode (%s)",
601 MirrorSyncMode_str(sync_mode
));
605 len
= bdrv_getlength(bs
);
607 error_setg_errno(errp
, -len
, "unable to get length for '%s'",
608 bdrv_get_device_name(bs
));
612 cluster_size
= backup_calculate_cluster_size(target
, errp
);
613 if (cluster_size
< 0) {
617 copy_bitmap
= hbitmap_alloc(len
, ctz32(cluster_size
));
619 /* job->len is fixed, so we can't allow resize */
620 job
= block_job_create(job_id
, &backup_job_driver
, txn
, bs
,
621 BLK_PERM_CONSISTENT_READ
,
622 BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE
|
623 BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_GRAPH_MOD
,
624 speed
, creation_flags
, cb
, opaque
, errp
);
629 /* The target must match the source in size, so no resize here either */
630 job
->target
= blk_new(job
->common
.job
.aio_context
,
632 BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE
|
633 BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_GRAPH_MOD
);
634 ret
= blk_insert_bs(job
->target
, target
, errp
);
639 job
->on_source_error
= on_source_error
;
640 job
->on_target_error
= on_target_error
;
641 job
->sync_mode
= sync_mode
;
642 job
->sync_bitmap
= sync_mode
== MIRROR_SYNC_MODE_INCREMENTAL
?
644 job
->compress
= compress
;
646 /* Detect image-fleecing (and similar) schemes */
647 job
->serialize_target_writes
= bdrv_chain_contains(target
, bs
);
648 job
->cluster_size
= cluster_size
;
649 job
->copy_bitmap
= copy_bitmap
;
651 job
->use_copy_range
= true;
652 job
->copy_range_size
= MIN_NON_ZERO(blk_get_max_transfer(job
->common
.blk
),
653 blk_get_max_transfer(job
->target
));
654 job
->copy_range_size
= MAX(job
->cluster_size
,
655 QEMU_ALIGN_UP(job
->copy_range_size
,
658 /* Required permissions are already taken with target's blk_new() */
659 block_job_add_bdrv(&job
->common
, "target", target
, 0, BLK_PERM_ALL
,
667 assert(!job
|| !job
->copy_bitmap
);
668 hbitmap_free(copy_bitmap
);
671 bdrv_reclaim_dirty_bitmap(bs
, sync_bitmap
, NULL
);
674 backup_clean(&job
->common
.job
);
675 job_early_fail(&job
->common
.job
);