4 * Copyright Red Hat, Inc. 2012
7 * Paolo Bonzini <pbonzini@redhat.com>
9 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
10 * See the COPYING.LIB file in the top-level directory.
14 #include "qemu/osdep.h"
15 #include "qemu/cutils.h"
16 #include "qemu/coroutine.h"
17 #include "qemu/range.h"
19 #include "block/blockjob_int.h"
20 #include "block/block_int.h"
21 #include "sysemu/block-backend.h"
22 #include "qapi/error.h"
23 #include "qapi/qmp/qerror.h"
24 #include "qemu/ratelimit.h"
25 #include "qemu/bitmap.h"
27 #define MAX_IN_FLIGHT 16
28 #define MAX_IO_BYTES (1 << 20) /* 1 Mb */
29 #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES)
31 /* The mirroring buffer is a list of granularity-sized chunks.
32 * Free chunks are organized in a list.
34 typedef struct MirrorBuffer
{
35 QSIMPLEQ_ENTRY(MirrorBuffer
) next
;
38 typedef struct MirrorOp MirrorOp
;
40 typedef struct MirrorBlockJob
{
43 BlockDriverState
*mirror_top_bs
;
44 BlockDriverState
*base
;
46 /* The name of the graph node to replace */
48 /* The BDS to replace */
49 BlockDriverState
*to_replace
;
50 /* Used to block operations on the drive-mirror-replace target */
51 Error
*replace_blocker
;
53 BlockMirrorBackingMode backing_mode
;
54 MirrorCopyMode copy_mode
;
55 BlockdevOnError on_source_error
, on_target_error
;
57 /* Set when the target is synced (dirty bitmap is clean, nothing
58 * in flight) and the job is running in active mode */
64 unsigned long *cow_bitmap
;
65 BdrvDirtyBitmap
*dirty_bitmap
;
66 BdrvDirtyBitmapIter
*dbi
;
68 QSIMPLEQ_HEAD(, MirrorBuffer
) buf_free
;
71 uint64_t last_pause_ns
;
72 unsigned long *in_flight_bitmap
;
74 int64_t bytes_in_flight
;
75 QTAILQ_HEAD(, MirrorOp
) ops_in_flight
;
78 int target_cluster_size
;
80 bool initial_zeroing_ongoing
;
81 int in_active_write_counter
;
85 typedef struct MirrorBDSOpaque
{
95 /* The pointee is set by mirror_co_read(), mirror_co_zero(), and
96 * mirror_co_discard() before yielding for the first time */
97 int64_t *bytes_handled
;
100 bool is_active_write
;
101 CoQueue waiting_requests
;
103 QTAILQ_ENTRY(MirrorOp
) next
;
106 typedef enum MirrorMethod
{
109 MIRROR_METHOD_DISCARD
,
112 static BlockErrorAction
mirror_error_action(MirrorBlockJob
*s
, bool read
,
116 s
->actively_synced
= false;
118 return block_job_error_action(&s
->common
, s
->on_source_error
,
121 return block_job_error_action(&s
->common
, s
->on_target_error
,
126 static void coroutine_fn
mirror_wait_on_conflicts(MirrorOp
*self
,
131 uint64_t self_start_chunk
= offset
/ s
->granularity
;
132 uint64_t self_end_chunk
= DIV_ROUND_UP(offset
+ bytes
, s
->granularity
);
133 uint64_t self_nb_chunks
= self_end_chunk
- self_start_chunk
;
135 while (find_next_bit(s
->in_flight_bitmap
, self_end_chunk
,
136 self_start_chunk
) < self_end_chunk
&&
141 QTAILQ_FOREACH(op
, &s
->ops_in_flight
, next
) {
142 uint64_t op_start_chunk
= op
->offset
/ s
->granularity
;
143 uint64_t op_nb_chunks
= DIV_ROUND_UP(op
->offset
+ op
->bytes
,
151 if (ranges_overlap(self_start_chunk
, self_nb_chunks
,
152 op_start_chunk
, op_nb_chunks
))
154 qemu_co_queue_wait(&op
->waiting_requests
, NULL
);
161 static void coroutine_fn
mirror_iteration_done(MirrorOp
*op
, int ret
)
163 MirrorBlockJob
*s
= op
->s
;
168 trace_mirror_iteration_done(s
, op
->offset
, op
->bytes
, ret
);
171 s
->bytes_in_flight
-= op
->bytes
;
173 for (i
= 0; i
< op
->qiov
.niov
; i
++) {
174 MirrorBuffer
*buf
= (MirrorBuffer
*) iov
[i
].iov_base
;
175 QSIMPLEQ_INSERT_TAIL(&s
->buf_free
, buf
, next
);
179 chunk_num
= op
->offset
/ s
->granularity
;
180 nb_chunks
= DIV_ROUND_UP(op
->bytes
, s
->granularity
);
182 bitmap_clear(s
->in_flight_bitmap
, chunk_num
, nb_chunks
);
183 QTAILQ_REMOVE(&s
->ops_in_flight
, op
, next
);
186 bitmap_set(s
->cow_bitmap
, chunk_num
, nb_chunks
);
188 if (!s
->initial_zeroing_ongoing
) {
189 job_progress_update(&s
->common
.job
, op
->bytes
);
192 qemu_iovec_destroy(&op
->qiov
);
194 qemu_co_queue_restart_all(&op
->waiting_requests
);
198 static void coroutine_fn
mirror_write_complete(MirrorOp
*op
, int ret
)
200 MirrorBlockJob
*s
= op
->s
;
203 BlockErrorAction action
;
205 bdrv_set_dirty_bitmap(s
->dirty_bitmap
, op
->offset
, op
->bytes
);
206 action
= mirror_error_action(s
, false, -ret
);
207 if (action
== BLOCK_ERROR_ACTION_REPORT
&& s
->ret
>= 0) {
212 mirror_iteration_done(op
, ret
);
215 static void coroutine_fn
mirror_read_complete(MirrorOp
*op
, int ret
)
217 MirrorBlockJob
*s
= op
->s
;
220 BlockErrorAction action
;
222 bdrv_set_dirty_bitmap(s
->dirty_bitmap
, op
->offset
, op
->bytes
);
223 action
= mirror_error_action(s
, true, -ret
);
224 if (action
== BLOCK_ERROR_ACTION_REPORT
&& s
->ret
>= 0) {
228 mirror_iteration_done(op
, ret
);
232 ret
= blk_co_pwritev(s
->target
, op
->offset
, op
->qiov
.size
, &op
->qiov
, 0);
233 mirror_write_complete(op
, ret
);
236 /* Clip bytes relative to offset to not exceed end-of-file */
237 static inline int64_t mirror_clip_bytes(MirrorBlockJob
*s
,
241 return MIN(bytes
, s
->bdev_length
- offset
);
244 /* Round offset and/or bytes to target cluster if COW is needed, and
245 * return the offset of the adjusted tail against original. */
246 static int mirror_cow_align(MirrorBlockJob
*s
, int64_t *offset
,
251 int64_t align_offset
= *offset
;
252 int64_t align_bytes
= *bytes
;
253 int max_bytes
= s
->granularity
* s
->max_iov
;
255 need_cow
= !test_bit(*offset
/ s
->granularity
, s
->cow_bitmap
);
256 need_cow
|= !test_bit((*offset
+ *bytes
- 1) / s
->granularity
,
259 bdrv_round_to_clusters(blk_bs(s
->target
), *offset
, *bytes
,
260 &align_offset
, &align_bytes
);
263 if (align_bytes
> max_bytes
) {
264 align_bytes
= max_bytes
;
266 align_bytes
= QEMU_ALIGN_DOWN(align_bytes
, s
->target_cluster_size
);
269 /* Clipping may result in align_bytes unaligned to chunk boundary, but
270 * that doesn't matter because it's already the end of source image. */
271 align_bytes
= mirror_clip_bytes(s
, align_offset
, align_bytes
);
273 ret
= align_offset
+ align_bytes
- (*offset
+ *bytes
);
274 *offset
= align_offset
;
275 *bytes
= align_bytes
;
280 static inline void coroutine_fn
281 mirror_wait_for_any_operation(MirrorBlockJob
*s
, bool active
)
285 QTAILQ_FOREACH(op
, &s
->ops_in_flight
, next
) {
286 /* Do not wait on pseudo ops, because it may in turn wait on
287 * some other operation to start, which may in fact be the
288 * caller of this function. Since there is only one pseudo op
289 * at any given time, we will always find some real operation
291 if (!op
->is_pseudo_op
&& op
->is_active_write
== active
) {
292 qemu_co_queue_wait(&op
->waiting_requests
, NULL
);
299 static inline void coroutine_fn
300 mirror_wait_for_free_in_flight_slot(MirrorBlockJob
*s
)
302 /* Only non-active operations use up in-flight slots */
303 mirror_wait_for_any_operation(s
, false);
306 /* Perform a mirror copy operation.
308 * *op->bytes_handled is set to the number of bytes copied after and
309 * including offset, excluding any bytes copied prior to offset due
310 * to alignment. This will be op->bytes if no alignment is necessary,
311 * or (new_end - op->offset) if the tail is rounded up or down due to
312 * alignment or buffer limit.
314 static void coroutine_fn
mirror_co_read(void *opaque
)
316 MirrorOp
*op
= opaque
;
317 MirrorBlockJob
*s
= op
->s
;
322 max_bytes
= s
->granularity
* s
->max_iov
;
324 /* We can only handle as much as buf_size at a time. */
325 op
->bytes
= MIN(s
->buf_size
, MIN(max_bytes
, op
->bytes
));
327 assert(op
->bytes
< BDRV_REQUEST_MAX_BYTES
);
328 *op
->bytes_handled
= op
->bytes
;
331 *op
->bytes_handled
+= mirror_cow_align(s
, &op
->offset
, &op
->bytes
);
333 /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */
334 assert(*op
->bytes_handled
<= UINT_MAX
);
335 assert(op
->bytes
<= s
->buf_size
);
336 /* The offset is granularity-aligned because:
337 * 1) Caller passes in aligned values;
338 * 2) mirror_cow_align is used only when target cluster is larger. */
339 assert(QEMU_IS_ALIGNED(op
->offset
, s
->granularity
));
340 /* The range is sector-aligned, since bdrv_getlength() rounds up. */
341 assert(QEMU_IS_ALIGNED(op
->bytes
, BDRV_SECTOR_SIZE
));
342 nb_chunks
= DIV_ROUND_UP(op
->bytes
, s
->granularity
);
344 while (s
->buf_free_count
< nb_chunks
) {
345 trace_mirror_yield_in_flight(s
, op
->offset
, s
->in_flight
);
346 mirror_wait_for_free_in_flight_slot(s
);
349 /* Now make a QEMUIOVector taking enough granularity-sized chunks
352 qemu_iovec_init(&op
->qiov
, nb_chunks
);
353 while (nb_chunks
-- > 0) {
354 MirrorBuffer
*buf
= QSIMPLEQ_FIRST(&s
->buf_free
);
355 size_t remaining
= op
->bytes
- op
->qiov
.size
;
357 QSIMPLEQ_REMOVE_HEAD(&s
->buf_free
, next
);
359 qemu_iovec_add(&op
->qiov
, buf
, MIN(s
->granularity
, remaining
));
362 /* Copy the dirty cluster. */
364 s
->bytes_in_flight
+= op
->bytes
;
365 trace_mirror_one_iteration(s
, op
->offset
, op
->bytes
);
367 ret
= bdrv_co_preadv(s
->mirror_top_bs
->backing
, op
->offset
, op
->bytes
,
369 mirror_read_complete(op
, ret
);
372 static void coroutine_fn
mirror_co_zero(void *opaque
)
374 MirrorOp
*op
= opaque
;
378 op
->s
->bytes_in_flight
+= op
->bytes
;
379 *op
->bytes_handled
= op
->bytes
;
381 ret
= blk_co_pwrite_zeroes(op
->s
->target
, op
->offset
, op
->bytes
,
382 op
->s
->unmap
? BDRV_REQ_MAY_UNMAP
: 0);
383 mirror_write_complete(op
, ret
);
386 static void coroutine_fn
mirror_co_discard(void *opaque
)
388 MirrorOp
*op
= opaque
;
392 op
->s
->bytes_in_flight
+= op
->bytes
;
393 *op
->bytes_handled
= op
->bytes
;
395 ret
= blk_co_pdiscard(op
->s
->target
, op
->offset
, op
->bytes
);
396 mirror_write_complete(op
, ret
);
399 static unsigned mirror_perform(MirrorBlockJob
*s
, int64_t offset
,
400 unsigned bytes
, MirrorMethod mirror_method
)
404 int64_t bytes_handled
= -1;
406 op
= g_new(MirrorOp
, 1);
411 .bytes_handled
= &bytes_handled
,
413 qemu_co_queue_init(&op
->waiting_requests
);
415 switch (mirror_method
) {
416 case MIRROR_METHOD_COPY
:
417 co
= qemu_coroutine_create(mirror_co_read
, op
);
419 case MIRROR_METHOD_ZERO
:
420 co
= qemu_coroutine_create(mirror_co_zero
, op
);
422 case MIRROR_METHOD_DISCARD
:
423 co
= qemu_coroutine_create(mirror_co_discard
, op
);
429 QTAILQ_INSERT_TAIL(&s
->ops_in_flight
, op
, next
);
430 qemu_coroutine_enter(co
);
431 /* At this point, ownership of op has been moved to the coroutine
432 * and the object may already be freed */
434 /* Assert that this value has been set */
435 assert(bytes_handled
>= 0);
437 /* Same assertion as in mirror_co_read() (and for mirror_co_read()
438 * and mirror_co_discard(), bytes_handled == op->bytes, which
439 * is the @bytes parameter given to this function) */
440 assert(bytes_handled
<= UINT_MAX
);
441 return bytes_handled
;
444 static uint64_t coroutine_fn
mirror_iteration(MirrorBlockJob
*s
)
446 BlockDriverState
*source
= s
->mirror_top_bs
->backing
->bs
;
449 uint64_t delay_ns
= 0, ret
= 0;
450 /* At least the first dirty chunk is mirrored in one iteration. */
452 bool write_zeroes_ok
= bdrv_can_write_zeroes_with_unmap(blk_bs(s
->target
));
453 int max_io_bytes
= MAX(s
->buf_size
/ MAX_IN_FLIGHT
, MAX_IO_BYTES
);
455 bdrv_dirty_bitmap_lock(s
->dirty_bitmap
);
456 offset
= bdrv_dirty_iter_next(s
->dbi
);
458 bdrv_set_dirty_iter(s
->dbi
, 0);
459 offset
= bdrv_dirty_iter_next(s
->dbi
);
460 trace_mirror_restart_iter(s
, bdrv_get_dirty_count(s
->dirty_bitmap
));
463 bdrv_dirty_bitmap_unlock(s
->dirty_bitmap
);
465 mirror_wait_on_conflicts(NULL
, s
, offset
, 1);
467 job_pause_point(&s
->common
.job
);
469 /* Find the number of consective dirty chunks following the first dirty
470 * one, and wait for in flight requests in them. */
471 bdrv_dirty_bitmap_lock(s
->dirty_bitmap
);
472 while (nb_chunks
* s
->granularity
< s
->buf_size
) {
474 int64_t next_offset
= offset
+ nb_chunks
* s
->granularity
;
475 int64_t next_chunk
= next_offset
/ s
->granularity
;
476 if (next_offset
>= s
->bdev_length
||
477 !bdrv_get_dirty_locked(source
, s
->dirty_bitmap
, next_offset
)) {
480 if (test_bit(next_chunk
, s
->in_flight_bitmap
)) {
484 next_dirty
= bdrv_dirty_iter_next(s
->dbi
);
485 if (next_dirty
> next_offset
|| next_dirty
< 0) {
486 /* The bitmap iterator's cache is stale, refresh it */
487 bdrv_set_dirty_iter(s
->dbi
, next_offset
);
488 next_dirty
= bdrv_dirty_iter_next(s
->dbi
);
490 assert(next_dirty
== next_offset
);
494 /* Clear dirty bits before querying the block status, because
495 * calling bdrv_block_status_above could yield - if some blocks are
496 * marked dirty in this window, we need to know.
498 bdrv_reset_dirty_bitmap_locked(s
->dirty_bitmap
, offset
,
499 nb_chunks
* s
->granularity
);
500 bdrv_dirty_bitmap_unlock(s
->dirty_bitmap
);
502 /* Before claiming an area in the in-flight bitmap, we have to
503 * create a MirrorOp for it so that conflicting requests can wait
504 * for it. mirror_perform() will create the real MirrorOps later,
505 * for now we just create a pseudo operation that will wake up all
506 * conflicting requests once all real operations have been
508 pseudo_op
= g_new(MirrorOp
, 1);
509 *pseudo_op
= (MirrorOp
){
511 .bytes
= nb_chunks
* s
->granularity
,
512 .is_pseudo_op
= true,
514 qemu_co_queue_init(&pseudo_op
->waiting_requests
);
515 QTAILQ_INSERT_TAIL(&s
->ops_in_flight
, pseudo_op
, next
);
517 bitmap_set(s
->in_flight_bitmap
, offset
/ s
->granularity
, nb_chunks
);
518 while (nb_chunks
> 0 && offset
< s
->bdev_length
) {
521 int64_t io_bytes_acct
;
522 MirrorMethod mirror_method
= MIRROR_METHOD_COPY
;
524 assert(!(offset
% s
->granularity
));
525 ret
= bdrv_block_status_above(source
, NULL
, offset
,
526 nb_chunks
* s
->granularity
,
527 &io_bytes
, NULL
, NULL
);
529 io_bytes
= MIN(nb_chunks
* s
->granularity
, max_io_bytes
);
530 } else if (ret
& BDRV_BLOCK_DATA
) {
531 io_bytes
= MIN(io_bytes
, max_io_bytes
);
534 io_bytes
-= io_bytes
% s
->granularity
;
535 if (io_bytes
< s
->granularity
) {
536 io_bytes
= s
->granularity
;
537 } else if (ret
>= 0 && !(ret
& BDRV_BLOCK_DATA
)) {
538 int64_t target_offset
;
539 int64_t target_bytes
;
540 bdrv_round_to_clusters(blk_bs(s
->target
), offset
, io_bytes
,
541 &target_offset
, &target_bytes
);
542 if (target_offset
== offset
&&
543 target_bytes
== io_bytes
) {
544 mirror_method
= ret
& BDRV_BLOCK_ZERO
?
546 MIRROR_METHOD_DISCARD
;
550 while (s
->in_flight
>= MAX_IN_FLIGHT
) {
551 trace_mirror_yield_in_flight(s
, offset
, s
->in_flight
);
552 mirror_wait_for_free_in_flight_slot(s
);
560 io_bytes
= mirror_clip_bytes(s
, offset
, io_bytes
);
561 io_bytes
= mirror_perform(s
, offset
, io_bytes
, mirror_method
);
562 if (mirror_method
!= MIRROR_METHOD_COPY
&& write_zeroes_ok
) {
565 io_bytes_acct
= io_bytes
;
569 nb_chunks
-= DIV_ROUND_UP(io_bytes
, s
->granularity
);
570 delay_ns
= block_job_ratelimit_get_delay(&s
->common
, io_bytes_acct
);
575 QTAILQ_REMOVE(&s
->ops_in_flight
, pseudo_op
, next
);
576 qemu_co_queue_restart_all(&pseudo_op
->waiting_requests
);
582 static void mirror_free_init(MirrorBlockJob
*s
)
584 int granularity
= s
->granularity
;
585 size_t buf_size
= s
->buf_size
;
586 uint8_t *buf
= s
->buf
;
588 assert(s
->buf_free_count
== 0);
589 QSIMPLEQ_INIT(&s
->buf_free
);
590 while (buf_size
!= 0) {
591 MirrorBuffer
*cur
= (MirrorBuffer
*)buf
;
592 QSIMPLEQ_INSERT_TAIL(&s
->buf_free
, cur
, next
);
594 buf_size
-= granularity
;
599 /* This is also used for the .pause callback. There is no matching
600 * mirror_resume() because mirror_run() will begin iterating again
601 * when the job is resumed.
603 static void coroutine_fn
mirror_wait_for_all_io(MirrorBlockJob
*s
)
605 while (s
->in_flight
> 0) {
606 mirror_wait_for_free_in_flight_slot(s
);
611 * mirror_exit_common: handle both abort() and prepare() cases.
612 * for .prepare, returns 0 on success and -errno on failure.
613 * for .abort cases, denoted by abort = true, MUST return 0.
615 static int mirror_exit_common(Job
*job
)
617 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
.job
);
618 BlockJob
*bjob
= &s
->common
;
619 MirrorBDSOpaque
*bs_opaque
= s
->mirror_top_bs
->opaque
;
620 AioContext
*replace_aio_context
= NULL
;
621 BlockDriverState
*src
= s
->mirror_top_bs
->backing
->bs
;
622 BlockDriverState
*target_bs
= blk_bs(s
->target
);
623 BlockDriverState
*mirror_top_bs
= s
->mirror_top_bs
;
624 Error
*local_err
= NULL
;
625 bool abort
= job
->ret
< 0;
633 if (bdrv_chain_contains(src
, target_bs
)) {
634 bdrv_unfreeze_backing_chain(mirror_top_bs
, target_bs
);
637 bdrv_release_dirty_bitmap(src
, s
->dirty_bitmap
);
639 /* Make sure that the source BDS doesn't go away during bdrv_replace_node,
640 * before we can call bdrv_drained_end */
642 bdrv_ref(mirror_top_bs
);
645 /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before
646 * inserting target_bs at s->to_replace, where we might not be able to get
649 * Note that blk_unref() alone doesn't necessarily drop permissions because
650 * we might be running nested inside mirror_drain(), which takes an extra
651 * reference, so use an explicit blk_set_perm() first. */
652 blk_set_perm(s
->target
, 0, BLK_PERM_ALL
, &error_abort
);
653 blk_unref(s
->target
);
656 /* We don't access the source any more. Dropping any WRITE/RESIZE is
657 * required before it could become a backing file of target_bs. */
658 bdrv_child_try_set_perm(mirror_top_bs
->backing
, 0, BLK_PERM_ALL
,
660 if (!abort
&& s
->backing_mode
== MIRROR_SOURCE_BACKING_CHAIN
) {
661 BlockDriverState
*backing
= s
->is_none_mode
? src
: s
->base
;
662 if (backing_bs(target_bs
) != backing
) {
663 bdrv_set_backing_hd(target_bs
, backing
, &local_err
);
665 error_report_err(local_err
);
672 replace_aio_context
= bdrv_get_aio_context(s
->to_replace
);
673 aio_context_acquire(replace_aio_context
);
676 if (s
->should_complete
&& !abort
) {
677 BlockDriverState
*to_replace
= s
->to_replace
?: src
;
678 bool ro
= bdrv_is_read_only(to_replace
);
680 if (ro
!= bdrv_is_read_only(target_bs
)) {
681 bdrv_reopen_set_read_only(target_bs
, ro
, NULL
);
684 /* The mirror job has no requests in flight any more, but we need to
685 * drain potential other users of the BDS before changing the graph. */
686 bdrv_drained_begin(target_bs
);
687 bdrv_replace_node(to_replace
, target_bs
, &local_err
);
688 bdrv_drained_end(target_bs
);
690 error_report_err(local_err
);
695 bdrv_op_unblock_all(s
->to_replace
, s
->replace_blocker
);
696 error_free(s
->replace_blocker
);
697 bdrv_unref(s
->to_replace
);
699 if (replace_aio_context
) {
700 aio_context_release(replace_aio_context
);
703 bdrv_unref(target_bs
);
705 /* Remove the mirror filter driver from the graph. Before this, get rid of
706 * the blockers on the intermediate nodes so that the resulting state is
707 * valid. Also give up permissions on mirror_top_bs->backing, which might
708 * block the removal. */
709 block_job_remove_all_bdrv(bjob
);
710 bdrv_child_try_set_perm(mirror_top_bs
->backing
, 0, BLK_PERM_ALL
,
712 bdrv_replace_node(mirror_top_bs
, backing_bs(mirror_top_bs
), &error_abort
);
714 /* We just changed the BDS the job BB refers to (with either or both of the
715 * bdrv_replace_node() calls), so switch the BB back so the cleanup does
716 * the right thing. We don't need any permissions any more now. */
717 blk_remove_bs(bjob
->blk
);
718 blk_set_perm(bjob
->blk
, 0, BLK_PERM_ALL
, &error_abort
);
719 blk_insert_bs(bjob
->blk
, mirror_top_bs
, &error_abort
);
721 bs_opaque
->job
= NULL
;
723 bdrv_drained_end(src
);
724 bdrv_unref(mirror_top_bs
);
730 static int mirror_prepare(Job
*job
)
732 return mirror_exit_common(job
);
735 static void mirror_abort(Job
*job
)
737 int ret
= mirror_exit_common(job
);
741 static void coroutine_fn
mirror_throttle(MirrorBlockJob
*s
)
743 int64_t now
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
745 if (now
- s
->last_pause_ns
> BLOCK_JOB_SLICE_TIME
) {
746 s
->last_pause_ns
= now
;
747 job_sleep_ns(&s
->common
.job
, 0);
749 job_pause_point(&s
->common
.job
);
753 static int coroutine_fn
mirror_dirty_init(MirrorBlockJob
*s
)
756 BlockDriverState
*base
= s
->base
;
757 BlockDriverState
*bs
= s
->mirror_top_bs
->backing
->bs
;
758 BlockDriverState
*target_bs
= blk_bs(s
->target
);
762 if (base
== NULL
&& !bdrv_has_zero_init(target_bs
)) {
763 if (!bdrv_can_write_zeroes_with_unmap(target_bs
)) {
764 bdrv_set_dirty_bitmap(s
->dirty_bitmap
, 0, s
->bdev_length
);
768 s
->initial_zeroing_ongoing
= true;
769 for (offset
= 0; offset
< s
->bdev_length
; ) {
770 int bytes
= MIN(s
->bdev_length
- offset
,
771 QEMU_ALIGN_DOWN(INT_MAX
, s
->granularity
));
775 if (job_is_cancelled(&s
->common
.job
)) {
776 s
->initial_zeroing_ongoing
= false;
780 if (s
->in_flight
>= MAX_IN_FLIGHT
) {
781 trace_mirror_yield(s
, UINT64_MAX
, s
->buf_free_count
,
783 mirror_wait_for_free_in_flight_slot(s
);
787 mirror_perform(s
, offset
, bytes
, MIRROR_METHOD_ZERO
);
791 mirror_wait_for_all_io(s
);
792 s
->initial_zeroing_ongoing
= false;
795 /* First part, loop on the sectors and initialize the dirty bitmap. */
796 for (offset
= 0; offset
< s
->bdev_length
; ) {
797 /* Just to make sure we are not exceeding int limit. */
798 int bytes
= MIN(s
->bdev_length
- offset
,
799 QEMU_ALIGN_DOWN(INT_MAX
, s
->granularity
));
803 if (job_is_cancelled(&s
->common
.job
)) {
807 ret
= bdrv_is_allocated_above(bs
, base
, offset
, bytes
, &count
);
814 bdrv_set_dirty_bitmap(s
->dirty_bitmap
, offset
, count
);
821 /* Called when going out of the streaming phase to flush the bulk of the
822 * data to the medium, or just before completing.
824 static int mirror_flush(MirrorBlockJob
*s
)
826 int ret
= blk_flush(s
->target
);
828 if (mirror_error_action(s
, false, -ret
) == BLOCK_ERROR_ACTION_REPORT
) {
835 static int coroutine_fn
mirror_run(Job
*job
, Error
**errp
)
837 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
.job
);
838 BlockDriverState
*bs
= s
->mirror_top_bs
->backing
->bs
;
839 BlockDriverState
*target_bs
= blk_bs(s
->target
);
840 bool need_drain
= true;
843 char backing_filename
[2]; /* we only need 2 characters because we are only
844 checking for a NULL string */
847 if (job_is_cancelled(&s
->common
.job
)) {
851 s
->bdev_length
= bdrv_getlength(bs
);
852 if (s
->bdev_length
< 0) {
853 ret
= s
->bdev_length
;
857 /* Active commit must resize the base image if its size differs from the
859 if (s
->base
== blk_bs(s
->target
)) {
862 base_length
= blk_getlength(s
->target
);
863 if (base_length
< 0) {
868 if (s
->bdev_length
> base_length
) {
869 ret
= blk_truncate(s
->target
, s
->bdev_length
, PREALLOC_MODE_OFF
,
877 if (s
->bdev_length
== 0) {
878 /* Transition to the READY state and wait for complete. */
879 job_transition_to_ready(&s
->common
.job
);
881 s
->actively_synced
= true;
882 while (!job_is_cancelled(&s
->common
.job
) && !s
->should_complete
) {
883 job_yield(&s
->common
.job
);
885 s
->common
.job
.cancelled
= false;
889 length
= DIV_ROUND_UP(s
->bdev_length
, s
->granularity
);
890 s
->in_flight_bitmap
= bitmap_new(length
);
892 /* If we have no backing file yet in the destination, we cannot let
893 * the destination do COW. Instead, we copy sectors around the
894 * dirty data if needed. We need a bitmap to do that.
896 bdrv_get_backing_filename(target_bs
, backing_filename
,
897 sizeof(backing_filename
));
898 if (!bdrv_get_info(target_bs
, &bdi
) && bdi
.cluster_size
) {
899 s
->target_cluster_size
= bdi
.cluster_size
;
901 s
->target_cluster_size
= BDRV_SECTOR_SIZE
;
903 if (backing_filename
[0] && !target_bs
->backing
&&
904 s
->granularity
< s
->target_cluster_size
) {
905 s
->buf_size
= MAX(s
->buf_size
, s
->target_cluster_size
);
906 s
->cow_bitmap
= bitmap_new(length
);
908 s
->max_iov
= MIN(bs
->bl
.max_iov
, target_bs
->bl
.max_iov
);
910 s
->buf
= qemu_try_blockalign(bs
, s
->buf_size
);
911 if (s
->buf
== NULL
) {
918 s
->last_pause_ns
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
919 if (!s
->is_none_mode
) {
920 ret
= mirror_dirty_init(s
);
921 if (ret
< 0 || job_is_cancelled(&s
->common
.job
)) {
927 s
->dbi
= bdrv_dirty_iter_new(s
->dirty_bitmap
);
929 uint64_t delay_ns
= 0;
931 bool should_complete
;
933 /* Do not start passive operations while there are active
934 * writes in progress */
935 while (s
->in_active_write_counter
) {
936 mirror_wait_for_any_operation(s
, true);
944 job_pause_point(&s
->common
.job
);
946 cnt
= bdrv_get_dirty_count(s
->dirty_bitmap
);
947 /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is
948 * the number of bytes currently being processed; together those are
949 * the current remaining operation length */
950 job_progress_set_remaining(&s
->common
.job
, s
->bytes_in_flight
+ cnt
);
952 /* Note that even when no rate limit is applied we need to yield
953 * periodically with no pending I/O so that bdrv_drain_all() returns.
954 * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is
955 * an error, or when the source is clean, whichever comes first. */
956 delta
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - s
->last_pause_ns
;
957 if (delta
< BLOCK_JOB_SLICE_TIME
&&
958 s
->common
.iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
959 if (s
->in_flight
>= MAX_IN_FLIGHT
|| s
->buf_free_count
== 0 ||
960 (cnt
== 0 && s
->in_flight
> 0)) {
961 trace_mirror_yield(s
, cnt
, s
->buf_free_count
, s
->in_flight
);
962 mirror_wait_for_free_in_flight_slot(s
);
964 } else if (cnt
!= 0) {
965 delay_ns
= mirror_iteration(s
);
969 should_complete
= false;
970 if (s
->in_flight
== 0 && cnt
== 0) {
971 trace_mirror_before_flush(s
);
973 if (mirror_flush(s
) < 0) {
974 /* Go check s->ret. */
977 /* We're out of the streaming phase. From now on, if the job
978 * is cancelled we will actually complete all pending I/O and
979 * report completion. This way, block-job-cancel will leave
980 * the target in a consistent state.
982 job_transition_to_ready(&s
->common
.job
);
984 if (s
->copy_mode
!= MIRROR_COPY_MODE_BACKGROUND
) {
985 s
->actively_synced
= true;
989 should_complete
= s
->should_complete
||
990 job_is_cancelled(&s
->common
.job
);
991 cnt
= bdrv_get_dirty_count(s
->dirty_bitmap
);
994 if (cnt
== 0 && should_complete
) {
995 /* The dirty bitmap is not updated while operations are pending.
996 * If we're about to exit, wait for pending operations before
997 * calling bdrv_get_dirty_count(bs), or we may exit while the
998 * source has dirty data to copy!
1000 * Note that I/O can be submitted by the guest while
1001 * mirror_populate runs, so pause it now. Before deciding
1002 * whether to switch to target check one last time if I/O has
1003 * come in the meanwhile, and if not flush the data to disk.
1005 trace_mirror_before_drain(s
, cnt
);
1007 bdrv_drained_begin(bs
);
1008 cnt
= bdrv_get_dirty_count(s
->dirty_bitmap
);
1009 if (cnt
> 0 || mirror_flush(s
) < 0) {
1010 bdrv_drained_end(bs
);
1014 /* The two disks are in sync. Exit and report successful
1017 assert(QLIST_EMPTY(&bs
->tracked_requests
));
1018 s
->common
.job
.cancelled
= false;
1025 if (s
->synced
&& !should_complete
) {
1026 delay_ns
= (s
->in_flight
== 0 &&
1027 cnt
== 0 ? BLOCK_JOB_SLICE_TIME
: 0);
1029 trace_mirror_before_sleep(s
, cnt
, s
->synced
, delay_ns
);
1030 job_sleep_ns(&s
->common
.job
, delay_ns
);
1031 if (job_is_cancelled(&s
->common
.job
) &&
1032 (!s
->synced
|| s
->common
.job
.force_cancel
))
1036 s
->last_pause_ns
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
1040 if (s
->in_flight
> 0) {
1041 /* We get here only if something went wrong. Either the job failed,
1042 * or it was cancelled prematurely so that we do not guarantee that
1043 * the target is a copy of the source.
1045 assert(ret
< 0 || ((s
->common
.job
.force_cancel
|| !s
->synced
) &&
1046 job_is_cancelled(&s
->common
.job
)));
1048 mirror_wait_for_all_io(s
);
1051 assert(s
->in_flight
== 0);
1053 g_free(s
->cow_bitmap
);
1054 g_free(s
->in_flight_bitmap
);
1055 bdrv_dirty_iter_free(s
->dbi
);
1058 bdrv_drained_begin(bs
);
1064 static void mirror_complete(Job
*job
, Error
**errp
)
1066 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
.job
);
1067 BlockDriverState
*target
;
1069 target
= blk_bs(s
->target
);
1072 error_setg(errp
, "The active block job '%s' cannot be completed",
1077 if (s
->backing_mode
== MIRROR_OPEN_BACKING_CHAIN
) {
1080 assert(!target
->backing
);
1081 ret
= bdrv_open_backing_file(target
, NULL
, "backing", errp
);
1087 /* block all operations on to_replace bs */
1089 AioContext
*replace_aio_context
;
1091 s
->to_replace
= bdrv_find_node(s
->replaces
);
1092 if (!s
->to_replace
) {
1093 error_setg(errp
, "Node name '%s' not found", s
->replaces
);
1097 replace_aio_context
= bdrv_get_aio_context(s
->to_replace
);
1098 aio_context_acquire(replace_aio_context
);
1100 /* TODO Translate this into permission system. Current definition of
1101 * GRAPH_MOD would require to request it for the parents; they might
1102 * not even be BlockDriverStates, however, so a BdrvChild can't address
1103 * them. May need redefinition of GRAPH_MOD. */
1104 error_setg(&s
->replace_blocker
,
1105 "block device is in use by block-job-complete");
1106 bdrv_op_block_all(s
->to_replace
, s
->replace_blocker
);
1107 bdrv_ref(s
->to_replace
);
1109 aio_context_release(replace_aio_context
);
1112 s
->should_complete
= true;
1116 static void coroutine_fn
mirror_pause(Job
*job
)
1118 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
.job
);
1120 mirror_wait_for_all_io(s
);
1123 static bool mirror_drained_poll(BlockJob
*job
)
1125 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
);
1126 return !!s
->in_flight
;
1129 static void mirror_attached_aio_context(BlockJob
*job
, AioContext
*new_context
)
1131 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
);
1133 blk_set_aio_context(s
->target
, new_context
);
1136 static void mirror_drain(BlockJob
*job
)
1138 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
);
1140 /* Need to keep a reference in case blk_drain triggers execution
1141 * of mirror_complete...
1144 BlockBackend
*target
= s
->target
;
1151 static const BlockJobDriver mirror_job_driver
= {
1153 .instance_size
= sizeof(MirrorBlockJob
),
1154 .job_type
= JOB_TYPE_MIRROR
,
1155 .free
= block_job_free
,
1156 .user_resume
= block_job_user_resume
,
1157 .drain
= block_job_drain
,
1159 .prepare
= mirror_prepare
,
1160 .abort
= mirror_abort
,
1161 .pause
= mirror_pause
,
1162 .complete
= mirror_complete
,
1164 .drained_poll
= mirror_drained_poll
,
1165 .attached_aio_context
= mirror_attached_aio_context
,
1166 .drain
= mirror_drain
,
1169 static const BlockJobDriver commit_active_job_driver
= {
1171 .instance_size
= sizeof(MirrorBlockJob
),
1172 .job_type
= JOB_TYPE_COMMIT
,
1173 .free
= block_job_free
,
1174 .user_resume
= block_job_user_resume
,
1175 .drain
= block_job_drain
,
1177 .prepare
= mirror_prepare
,
1178 .abort
= mirror_abort
,
1179 .pause
= mirror_pause
,
1180 .complete
= mirror_complete
,
1182 .drained_poll
= mirror_drained_poll
,
1183 .attached_aio_context
= mirror_attached_aio_context
,
1184 .drain
= mirror_drain
,
1187 static void coroutine_fn
1188 do_sync_target_write(MirrorBlockJob
*job
, MirrorMethod method
,
1189 uint64_t offset
, uint64_t bytes
,
1190 QEMUIOVector
*qiov
, int flags
)
1192 QEMUIOVector target_qiov
;
1193 uint64_t dirty_offset
= offset
;
1194 uint64_t dirty_bytes
;
1197 qemu_iovec_init(&target_qiov
, qiov
->niov
);
1204 bdrv_dirty_bitmap_lock(job
->dirty_bitmap
);
1205 dirty_bytes
= MIN(offset
+ bytes
- dirty_offset
, INT_MAX
);
1206 valid_area
= bdrv_dirty_bitmap_next_dirty_area(job
->dirty_bitmap
,
1210 bdrv_dirty_bitmap_unlock(job
->dirty_bitmap
);
1214 bdrv_reset_dirty_bitmap_locked(job
->dirty_bitmap
,
1215 dirty_offset
, dirty_bytes
);
1216 bdrv_dirty_bitmap_unlock(job
->dirty_bitmap
);
1218 job_progress_increase_remaining(&job
->common
.job
, dirty_bytes
);
1220 assert(dirty_offset
- offset
<= SIZE_MAX
);
1222 qemu_iovec_reset(&target_qiov
);
1223 qemu_iovec_concat(&target_qiov
, qiov
,
1224 dirty_offset
- offset
, dirty_bytes
);
1228 case MIRROR_METHOD_COPY
:
1229 ret
= blk_co_pwritev(job
->target
, dirty_offset
, dirty_bytes
,
1230 qiov
? &target_qiov
: NULL
, flags
);
1233 case MIRROR_METHOD_ZERO
:
1235 ret
= blk_co_pwrite_zeroes(job
->target
, dirty_offset
, dirty_bytes
,
1239 case MIRROR_METHOD_DISCARD
:
1241 ret
= blk_co_pdiscard(job
->target
, dirty_offset
, dirty_bytes
);
1249 job_progress_update(&job
->common
.job
, dirty_bytes
);
1251 BlockErrorAction action
;
1253 bdrv_set_dirty_bitmap(job
->dirty_bitmap
, dirty_offset
, dirty_bytes
);
1254 job
->actively_synced
= false;
1256 action
= mirror_error_action(job
, false, -ret
);
1257 if (action
== BLOCK_ERROR_ACTION_REPORT
) {
1265 dirty_offset
+= dirty_bytes
;
1269 qemu_iovec_destroy(&target_qiov
);
1273 static MirrorOp
*coroutine_fn
active_write_prepare(MirrorBlockJob
*s
,
1278 uint64_t start_chunk
= offset
/ s
->granularity
;
1279 uint64_t end_chunk
= DIV_ROUND_UP(offset
+ bytes
, s
->granularity
);
1281 op
= g_new(MirrorOp
, 1);
1286 .is_active_write
= true,
1288 qemu_co_queue_init(&op
->waiting_requests
);
1289 QTAILQ_INSERT_TAIL(&s
->ops_in_flight
, op
, next
);
1291 s
->in_active_write_counter
++;
1293 mirror_wait_on_conflicts(op
, s
, offset
, bytes
);
1295 bitmap_set(s
->in_flight_bitmap
, start_chunk
, end_chunk
- start_chunk
);
1300 static void coroutine_fn
active_write_settle(MirrorOp
*op
)
1302 uint64_t start_chunk
= op
->offset
/ op
->s
->granularity
;
1303 uint64_t end_chunk
= DIV_ROUND_UP(op
->offset
+ op
->bytes
,
1304 op
->s
->granularity
);
1306 if (!--op
->s
->in_active_write_counter
&& op
->s
->actively_synced
) {
1307 BdrvChild
*source
= op
->s
->mirror_top_bs
->backing
;
1309 if (QLIST_FIRST(&source
->bs
->parents
) == source
&&
1310 QLIST_NEXT(source
, next_parent
) == NULL
)
1312 /* Assert that we are back in sync once all active write
1313 * operations are settled.
1314 * Note that we can only assert this if the mirror node
1315 * is the source node's only parent. */
1316 assert(!bdrv_get_dirty_count(op
->s
->dirty_bitmap
));
1319 bitmap_clear(op
->s
->in_flight_bitmap
, start_chunk
, end_chunk
- start_chunk
);
1320 QTAILQ_REMOVE(&op
->s
->ops_in_flight
, op
, next
);
1321 qemu_co_queue_restart_all(&op
->waiting_requests
);
1325 static int coroutine_fn
bdrv_mirror_top_preadv(BlockDriverState
*bs
,
1326 uint64_t offset
, uint64_t bytes
, QEMUIOVector
*qiov
, int flags
)
1328 return bdrv_co_preadv(bs
->backing
, offset
, bytes
, qiov
, flags
);
1331 static int coroutine_fn
bdrv_mirror_top_do_write(BlockDriverState
*bs
,
1332 MirrorMethod method
, uint64_t offset
, uint64_t bytes
, QEMUIOVector
*qiov
,
1335 MirrorOp
*op
= NULL
;
1336 MirrorBDSOpaque
*s
= bs
->opaque
;
1338 bool copy_to_target
;
1340 copy_to_target
= s
->job
->ret
>= 0 &&
1341 s
->job
->copy_mode
== MIRROR_COPY_MODE_WRITE_BLOCKING
;
1343 if (copy_to_target
) {
1344 op
= active_write_prepare(s
->job
, offset
, bytes
);
1348 case MIRROR_METHOD_COPY
:
1349 ret
= bdrv_co_pwritev(bs
->backing
, offset
, bytes
, qiov
, flags
);
1352 case MIRROR_METHOD_ZERO
:
1353 ret
= bdrv_co_pwrite_zeroes(bs
->backing
, offset
, bytes
, flags
);
1356 case MIRROR_METHOD_DISCARD
:
1357 ret
= bdrv_co_pdiscard(bs
->backing
, offset
, bytes
);
1368 if (copy_to_target
) {
1369 do_sync_target_write(s
->job
, method
, offset
, bytes
, qiov
, flags
);
1373 if (copy_to_target
) {
1374 active_write_settle(op
);
1379 static int coroutine_fn
bdrv_mirror_top_pwritev(BlockDriverState
*bs
,
1380 uint64_t offset
, uint64_t bytes
, QEMUIOVector
*qiov
, int flags
)
1382 MirrorBDSOpaque
*s
= bs
->opaque
;
1383 QEMUIOVector bounce_qiov
;
1386 bool copy_to_target
;
1388 copy_to_target
= s
->job
->ret
>= 0 &&
1389 s
->job
->copy_mode
== MIRROR_COPY_MODE_WRITE_BLOCKING
;
1391 if (copy_to_target
) {
1392 /* The guest might concurrently modify the data to write; but
1393 * the data on source and destination must match, so we have
1394 * to use a bounce buffer if we are going to write to the
1396 bounce_buf
= qemu_blockalign(bs
, bytes
);
1397 iov_to_buf_full(qiov
->iov
, qiov
->niov
, 0, bounce_buf
, bytes
);
1399 qemu_iovec_init(&bounce_qiov
, 1);
1400 qemu_iovec_add(&bounce_qiov
, bounce_buf
, bytes
);
1401 qiov
= &bounce_qiov
;
1404 ret
= bdrv_mirror_top_do_write(bs
, MIRROR_METHOD_COPY
, offset
, bytes
, qiov
,
1407 if (copy_to_target
) {
1408 qemu_iovec_destroy(&bounce_qiov
);
1409 qemu_vfree(bounce_buf
);
1415 static int coroutine_fn
bdrv_mirror_top_flush(BlockDriverState
*bs
)
1417 if (bs
->backing
== NULL
) {
1418 /* we can be here after failed bdrv_append in mirror_start_job */
1421 return bdrv_co_flush(bs
->backing
->bs
);
1424 static int coroutine_fn
bdrv_mirror_top_pwrite_zeroes(BlockDriverState
*bs
,
1425 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
1427 return bdrv_mirror_top_do_write(bs
, MIRROR_METHOD_ZERO
, offset
, bytes
, NULL
,
1431 static int coroutine_fn
bdrv_mirror_top_pdiscard(BlockDriverState
*bs
,
1432 int64_t offset
, int bytes
)
1434 return bdrv_mirror_top_do_write(bs
, MIRROR_METHOD_DISCARD
, offset
, bytes
,
1438 static void bdrv_mirror_top_refresh_filename(BlockDriverState
*bs
)
1440 if (bs
->backing
== NULL
) {
1441 /* we can be here after failed bdrv_attach_child in
1442 * bdrv_set_backing_hd */
1445 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
1446 bs
->backing
->bs
->filename
);
1449 static void bdrv_mirror_top_child_perm(BlockDriverState
*bs
, BdrvChild
*c
,
1450 const BdrvChildRole
*role
,
1451 BlockReopenQueue
*reopen_queue
,
1452 uint64_t perm
, uint64_t shared
,
1453 uint64_t *nperm
, uint64_t *nshared
)
1455 /* Must be able to forward guest writes to the real image */
1457 if (perm
& BLK_PERM_WRITE
) {
1458 *nperm
|= BLK_PERM_WRITE
;
1461 *nshared
= BLK_PERM_ALL
;
1464 /* Dummy node that provides consistent read to its users without requiring it
1465 * from its backing file and that allows writes on the backing file chain. */
1466 static BlockDriver bdrv_mirror_top
= {
1467 .format_name
= "mirror_top",
1468 .bdrv_co_preadv
= bdrv_mirror_top_preadv
,
1469 .bdrv_co_pwritev
= bdrv_mirror_top_pwritev
,
1470 .bdrv_co_pwrite_zeroes
= bdrv_mirror_top_pwrite_zeroes
,
1471 .bdrv_co_pdiscard
= bdrv_mirror_top_pdiscard
,
1472 .bdrv_co_flush
= bdrv_mirror_top_flush
,
1473 .bdrv_co_block_status
= bdrv_co_block_status_from_backing
,
1474 .bdrv_refresh_filename
= bdrv_mirror_top_refresh_filename
,
1475 .bdrv_child_perm
= bdrv_mirror_top_child_perm
,
1478 static void mirror_start_job(const char *job_id
, BlockDriverState
*bs
,
1479 int creation_flags
, BlockDriverState
*target
,
1480 const char *replaces
, int64_t speed
,
1481 uint32_t granularity
, int64_t buf_size
,
1482 BlockMirrorBackingMode backing_mode
,
1483 BlockdevOnError on_source_error
,
1484 BlockdevOnError on_target_error
,
1486 BlockCompletionFunc
*cb
,
1488 const BlockJobDriver
*driver
,
1489 bool is_none_mode
, BlockDriverState
*base
,
1490 bool auto_complete
, const char *filter_node_name
,
1491 bool is_mirror
, MirrorCopyMode copy_mode
,
1495 MirrorBDSOpaque
*bs_opaque
;
1496 BlockDriverState
*mirror_top_bs
;
1497 bool target_graph_mod
;
1498 bool target_is_backing
;
1499 Error
*local_err
= NULL
;
1502 if (granularity
== 0) {
1503 granularity
= bdrv_get_default_bitmap_granularity(target
);
1506 assert(is_power_of_2(granularity
));
1509 error_setg(errp
, "Invalid parameter 'buf-size'");
1513 if (buf_size
== 0) {
1514 buf_size
= DEFAULT_MIRROR_BUF_SIZE
;
1518 error_setg(errp
, "Can't mirror node into itself");
1522 /* In the case of active commit, add dummy driver to provide consistent
1523 * reads on the top, while disabling it in the intermediate nodes, and make
1524 * the backing chain writable. */
1525 mirror_top_bs
= bdrv_new_open_driver(&bdrv_mirror_top
, filter_node_name
,
1527 if (mirror_top_bs
== NULL
) {
1530 if (!filter_node_name
) {
1531 mirror_top_bs
->implicit
= true;
1533 mirror_top_bs
->total_sectors
= bs
->total_sectors
;
1534 mirror_top_bs
->supported_write_flags
= BDRV_REQ_WRITE_UNCHANGED
;
1535 mirror_top_bs
->supported_zero_flags
= BDRV_REQ_WRITE_UNCHANGED
;
1536 bs_opaque
= g_new0(MirrorBDSOpaque
, 1);
1537 mirror_top_bs
->opaque
= bs_opaque
;
1538 bdrv_set_aio_context(mirror_top_bs
, bdrv_get_aio_context(bs
));
1540 /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep
1541 * it alive until block_job_create() succeeds even if bs has no parent. */
1542 bdrv_ref(mirror_top_bs
);
1543 bdrv_drained_begin(bs
);
1544 bdrv_append(mirror_top_bs
, bs
, &local_err
);
1545 bdrv_drained_end(bs
);
1548 bdrv_unref(mirror_top_bs
);
1549 error_propagate(errp
, local_err
);
1553 /* Make sure that the source is not resized while the job is running */
1554 s
= block_job_create(job_id
, driver
, NULL
, mirror_top_bs
,
1555 BLK_PERM_CONSISTENT_READ
,
1556 BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
|
1557 BLK_PERM_WRITE
| BLK_PERM_GRAPH_MOD
, speed
,
1558 creation_flags
, cb
, opaque
, errp
);
1564 /* The block job now has a reference to this node */
1565 bdrv_unref(mirror_top_bs
);
1567 s
->mirror_top_bs
= mirror_top_bs
;
1569 /* No resize for the target either; while the mirror is still running, a
1570 * consistent read isn't necessarily possible. We could possibly allow
1571 * writes and graph modifications, though it would likely defeat the
1572 * purpose of a mirror, so leave them blocked for now.
1574 * In the case of active commit, things look a bit different, though,
1575 * because the target is an already populated backing file in active use.
1576 * We can allow anything except resize there.*/
1577 target_is_backing
= bdrv_chain_contains(bs
, target
);
1578 target_graph_mod
= (backing_mode
!= MIRROR_LEAVE_BACKING_CHAIN
);
1579 s
->target
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
|
1580 (target_graph_mod
? BLK_PERM_GRAPH_MOD
: 0),
1581 BLK_PERM_WRITE_UNCHANGED
|
1582 (target_is_backing
? BLK_PERM_CONSISTENT_READ
|
1584 BLK_PERM_GRAPH_MOD
: 0));
1585 ret
= blk_insert_bs(s
->target
, target
, errp
);
1590 /* XXX: Mirror target could be a NBD server of target QEMU in the case
1591 * of non-shared block migration. To allow migration completion, we
1592 * have to allow "inactivate" of the target BB. When that happens, we
1593 * know the job is drained, and the vcpus are stopped, so no write
1594 * operation will be performed. Block layer already has assertions to
1596 blk_set_force_allow_inactivate(s
->target
);
1599 s
->replaces
= g_strdup(replaces
);
1600 s
->on_source_error
= on_source_error
;
1601 s
->on_target_error
= on_target_error
;
1602 s
->is_none_mode
= is_none_mode
;
1603 s
->backing_mode
= backing_mode
;
1604 s
->copy_mode
= copy_mode
;
1606 s
->granularity
= granularity
;
1607 s
->buf_size
= ROUND_UP(buf_size
, granularity
);
1609 if (auto_complete
) {
1610 s
->should_complete
= true;
1613 s
->dirty_bitmap
= bdrv_create_dirty_bitmap(bs
, granularity
, NULL
, errp
);
1614 if (!s
->dirty_bitmap
) {
1618 ret
= block_job_add_bdrv(&s
->common
, "source", bs
, 0,
1619 BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_WRITE
|
1620 BLK_PERM_CONSISTENT_READ
,
1626 /* Required permissions are already taken with blk_new() */
1627 block_job_add_bdrv(&s
->common
, "target", target
, 0, BLK_PERM_ALL
,
1630 /* In commit_active_start() all intermediate nodes disappear, so
1631 * any jobs in them must be blocked */
1632 if (target_is_backing
) {
1633 BlockDriverState
*iter
;
1634 for (iter
= backing_bs(bs
); iter
!= target
; iter
= backing_bs(iter
)) {
1635 /* XXX BLK_PERM_WRITE needs to be allowed so we don't block
1636 * ourselves at s->base (if writes are blocked for a node, they are
1637 * also blocked for its backing file). The other options would be a
1638 * second filter driver above s->base (== target). */
1639 ret
= block_job_add_bdrv(&s
->common
, "intermediate node", iter
, 0,
1640 BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_WRITE
,
1647 if (bdrv_freeze_backing_chain(mirror_top_bs
, target
, errp
) < 0) {
1652 QTAILQ_INIT(&s
->ops_in_flight
);
1654 trace_mirror_start(bs
, s
, opaque
);
1655 job_start(&s
->common
.job
);
1660 /* Make sure this BDS does not go away until we have completed the graph
1662 bdrv_ref(mirror_top_bs
);
1664 g_free(s
->replaces
);
1665 blk_unref(s
->target
);
1666 bs_opaque
->job
= NULL
;
1667 if (s
->dirty_bitmap
) {
1668 bdrv_release_dirty_bitmap(bs
, s
->dirty_bitmap
);
1670 job_early_fail(&s
->common
.job
);
1673 bdrv_child_try_set_perm(mirror_top_bs
->backing
, 0, BLK_PERM_ALL
,
1675 bdrv_replace_node(mirror_top_bs
, backing_bs(mirror_top_bs
), &error_abort
);
1677 bdrv_unref(mirror_top_bs
);
1680 void mirror_start(const char *job_id
, BlockDriverState
*bs
,
1681 BlockDriverState
*target
, const char *replaces
,
1682 int creation_flags
, int64_t speed
,
1683 uint32_t granularity
, int64_t buf_size
,
1684 MirrorSyncMode mode
, BlockMirrorBackingMode backing_mode
,
1685 BlockdevOnError on_source_error
,
1686 BlockdevOnError on_target_error
,
1687 bool unmap
, const char *filter_node_name
,
1688 MirrorCopyMode copy_mode
, Error
**errp
)
1691 BlockDriverState
*base
;
1693 if (mode
== MIRROR_SYNC_MODE_INCREMENTAL
) {
1694 error_setg(errp
, "Sync mode 'incremental' not supported");
1697 is_none_mode
= mode
== MIRROR_SYNC_MODE_NONE
;
1698 base
= mode
== MIRROR_SYNC_MODE_TOP
? backing_bs(bs
) : NULL
;
1699 mirror_start_job(job_id
, bs
, creation_flags
, target
, replaces
,
1700 speed
, granularity
, buf_size
, backing_mode
,
1701 on_source_error
, on_target_error
, unmap
, NULL
, NULL
,
1702 &mirror_job_driver
, is_none_mode
, base
, false,
1703 filter_node_name
, true, copy_mode
, errp
);
1706 void commit_active_start(const char *job_id
, BlockDriverState
*bs
,
1707 BlockDriverState
*base
, int creation_flags
,
1708 int64_t speed
, BlockdevOnError on_error
,
1709 const char *filter_node_name
,
1710 BlockCompletionFunc
*cb
, void *opaque
,
1711 bool auto_complete
, Error
**errp
)
1713 bool base_read_only
;
1714 Error
*local_err
= NULL
;
1716 base_read_only
= bdrv_is_read_only(base
);
1718 if (base_read_only
) {
1719 if (bdrv_reopen_set_read_only(base
, false, errp
) < 0) {
1724 mirror_start_job(job_id
, bs
, creation_flags
, base
, NULL
, speed
, 0, 0,
1725 MIRROR_LEAVE_BACKING_CHAIN
,
1726 on_error
, on_error
, true, cb
, opaque
,
1727 &commit_active_job_driver
, false, base
, auto_complete
,
1728 filter_node_name
, false, MIRROR_COPY_MODE_BACKGROUND
,
1731 error_propagate(errp
, local_err
);
1732 goto error_restore_flags
;
1737 error_restore_flags
:
1738 /* ignore error and errp for bdrv_reopen, because we want to propagate
1739 * the original error */
1740 if (base_read_only
) {
1741 bdrv_reopen_set_read_only(base
, true, NULL
);