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"
16 #include "block/blockjob_int.h"
17 #include "block/block_int.h"
18 #include "sysemu/block-backend.h"
19 #include "qapi/error.h"
20 #include "qapi/qmp/qerror.h"
21 #include "qemu/ratelimit.h"
22 #include "qemu/bitmap.h"
24 #define SLICE_TIME 100000000ULL /* ns */
25 #define MAX_IN_FLIGHT 16
26 #define MAX_IO_SECTORS ((1 << 20) >> BDRV_SECTOR_BITS) /* 1 Mb */
27 #define DEFAULT_MIRROR_BUF_SIZE \
28 (MAX_IN_FLIGHT * MAX_IO_SECTORS * BDRV_SECTOR_SIZE)
30 /* The mirroring buffer is a list of granularity-sized chunks.
31 * Free chunks are organized in a list.
33 typedef struct MirrorBuffer
{
34 QSIMPLEQ_ENTRY(MirrorBuffer
) next
;
37 typedef struct MirrorBlockJob
{
41 BlockDriverState
*base
;
42 /* The name of the graph node to replace */
44 /* The BDS to replace */
45 BlockDriverState
*to_replace
;
46 /* Used to block operations on the drive-mirror-replace target */
47 Error
*replace_blocker
;
49 BlockMirrorBackingMode backing_mode
;
50 BlockdevOnError on_source_error
, on_target_error
;
56 unsigned long *cow_bitmap
;
57 BdrvDirtyBitmap
*dirty_bitmap
;
58 BdrvDirtyBitmapIter
*dbi
;
60 QSIMPLEQ_HEAD(, MirrorBuffer
) buf_free
;
63 uint64_t last_pause_ns
;
64 unsigned long *in_flight_bitmap
;
66 int64_t sectors_in_flight
;
70 int target_cluster_sectors
;
74 typedef struct MirrorOp
{
81 static BlockErrorAction
mirror_error_action(MirrorBlockJob
*s
, bool read
,
86 return block_job_error_action(&s
->common
, s
->on_source_error
,
89 return block_job_error_action(&s
->common
, s
->on_target_error
,
94 static void mirror_iteration_done(MirrorOp
*op
, int ret
)
96 MirrorBlockJob
*s
= op
->s
;
99 int i
, nb_chunks
, sectors_per_chunk
;
101 trace_mirror_iteration_done(s
, op
->sector_num
, op
->nb_sectors
, ret
);
104 s
->sectors_in_flight
-= op
->nb_sectors
;
106 for (i
= 0; i
< op
->qiov
.niov
; i
++) {
107 MirrorBuffer
*buf
= (MirrorBuffer
*) iov
[i
].iov_base
;
108 QSIMPLEQ_INSERT_TAIL(&s
->buf_free
, buf
, next
);
112 sectors_per_chunk
= s
->granularity
>> BDRV_SECTOR_BITS
;
113 chunk_num
= op
->sector_num
/ sectors_per_chunk
;
114 nb_chunks
= DIV_ROUND_UP(op
->nb_sectors
, sectors_per_chunk
);
115 bitmap_clear(s
->in_flight_bitmap
, chunk_num
, nb_chunks
);
118 bitmap_set(s
->cow_bitmap
, chunk_num
, nb_chunks
);
120 s
->common
.offset
+= (uint64_t)op
->nb_sectors
* BDRV_SECTOR_SIZE
;
123 qemu_iovec_destroy(&op
->qiov
);
126 if (s
->waiting_for_io
) {
127 qemu_coroutine_enter(s
->common
.co
);
131 static void mirror_write_complete(void *opaque
, int ret
)
133 MirrorOp
*op
= opaque
;
134 MirrorBlockJob
*s
= op
->s
;
136 aio_context_acquire(blk_get_aio_context(s
->common
.blk
));
138 BlockErrorAction action
;
140 bdrv_set_dirty_bitmap(s
->dirty_bitmap
, op
->sector_num
, op
->nb_sectors
);
141 action
= mirror_error_action(s
, false, -ret
);
142 if (action
== BLOCK_ERROR_ACTION_REPORT
&& s
->ret
>= 0) {
146 mirror_iteration_done(op
, ret
);
147 aio_context_release(blk_get_aio_context(s
->common
.blk
));
150 static void mirror_read_complete(void *opaque
, int ret
)
152 MirrorOp
*op
= opaque
;
153 MirrorBlockJob
*s
= op
->s
;
155 aio_context_acquire(blk_get_aio_context(s
->common
.blk
));
157 BlockErrorAction action
;
159 bdrv_set_dirty_bitmap(s
->dirty_bitmap
, op
->sector_num
, op
->nb_sectors
);
160 action
= mirror_error_action(s
, true, -ret
);
161 if (action
== BLOCK_ERROR_ACTION_REPORT
&& s
->ret
>= 0) {
165 mirror_iteration_done(op
, ret
);
167 blk_aio_pwritev(s
->target
, op
->sector_num
* BDRV_SECTOR_SIZE
, &op
->qiov
,
168 0, mirror_write_complete
, op
);
170 aio_context_release(blk_get_aio_context(s
->common
.blk
));
173 static inline void mirror_clip_sectors(MirrorBlockJob
*s
,
177 *nb_sectors
= MIN(*nb_sectors
,
178 s
->bdev_length
/ BDRV_SECTOR_SIZE
- sector_num
);
181 /* Round sector_num and/or nb_sectors to target cluster if COW is needed, and
182 * return the offset of the adjusted tail sector against original. */
183 static int mirror_cow_align(MirrorBlockJob
*s
,
189 int chunk_sectors
= s
->granularity
>> BDRV_SECTOR_BITS
;
190 int64_t align_sector_num
= *sector_num
;
191 int align_nb_sectors
= *nb_sectors
;
192 int max_sectors
= chunk_sectors
* s
->max_iov
;
194 need_cow
= !test_bit(*sector_num
/ chunk_sectors
, s
->cow_bitmap
);
195 need_cow
|= !test_bit((*sector_num
+ *nb_sectors
- 1) / chunk_sectors
,
198 bdrv_round_sectors_to_clusters(blk_bs(s
->target
), *sector_num
,
199 *nb_sectors
, &align_sector_num
,
203 if (align_nb_sectors
> max_sectors
) {
204 align_nb_sectors
= max_sectors
;
206 align_nb_sectors
= QEMU_ALIGN_DOWN(align_nb_sectors
,
207 s
->target_cluster_sectors
);
210 /* Clipping may result in align_nb_sectors unaligned to chunk boundary, but
211 * that doesn't matter because it's already the end of source image. */
212 mirror_clip_sectors(s
, align_sector_num
, &align_nb_sectors
);
214 ret
= align_sector_num
+ align_nb_sectors
- (*sector_num
+ *nb_sectors
);
215 *sector_num
= align_sector_num
;
216 *nb_sectors
= align_nb_sectors
;
221 static inline void mirror_wait_for_io(MirrorBlockJob
*s
)
223 assert(!s
->waiting_for_io
);
224 s
->waiting_for_io
= true;
225 qemu_coroutine_yield();
226 s
->waiting_for_io
= false;
229 /* Submit async read while handling COW.
230 * Returns: The number of sectors copied after and including sector_num,
231 * excluding any sectors copied prior to sector_num due to alignment.
232 * This will be nb_sectors if no alignment is necessary, or
233 * (new_end - sector_num) if tail is rounded up or down due to
234 * alignment or buffer limit.
236 static int mirror_do_read(MirrorBlockJob
*s
, int64_t sector_num
,
239 BlockBackend
*source
= s
->common
.blk
;
240 int sectors_per_chunk
, nb_chunks
;
245 sectors_per_chunk
= s
->granularity
>> BDRV_SECTOR_BITS
;
246 max_sectors
= sectors_per_chunk
* s
->max_iov
;
248 /* We can only handle as much as buf_size at a time. */
249 nb_sectors
= MIN(s
->buf_size
>> BDRV_SECTOR_BITS
, nb_sectors
);
250 nb_sectors
= MIN(max_sectors
, nb_sectors
);
255 ret
+= mirror_cow_align(s
, §or_num
, &nb_sectors
);
257 assert(nb_sectors
<< BDRV_SECTOR_BITS
<= s
->buf_size
);
258 /* The sector range must meet granularity because:
259 * 1) Caller passes in aligned values;
260 * 2) mirror_cow_align is used only when target cluster is larger. */
261 assert(!(sector_num
% sectors_per_chunk
));
262 nb_chunks
= DIV_ROUND_UP(nb_sectors
, sectors_per_chunk
);
264 while (s
->buf_free_count
< nb_chunks
) {
265 trace_mirror_yield_in_flight(s
, sector_num
, s
->in_flight
);
266 mirror_wait_for_io(s
);
269 /* Allocate a MirrorOp that is used as an AIO callback. */
270 op
= g_new(MirrorOp
, 1);
272 op
->sector_num
= sector_num
;
273 op
->nb_sectors
= nb_sectors
;
275 /* Now make a QEMUIOVector taking enough granularity-sized chunks
278 qemu_iovec_init(&op
->qiov
, nb_chunks
);
279 while (nb_chunks
-- > 0) {
280 MirrorBuffer
*buf
= QSIMPLEQ_FIRST(&s
->buf_free
);
281 size_t remaining
= nb_sectors
* BDRV_SECTOR_SIZE
- op
->qiov
.size
;
283 QSIMPLEQ_REMOVE_HEAD(&s
->buf_free
, next
);
285 qemu_iovec_add(&op
->qiov
, buf
, MIN(s
->granularity
, remaining
));
288 /* Copy the dirty cluster. */
290 s
->sectors_in_flight
+= nb_sectors
;
291 trace_mirror_one_iteration(s
, sector_num
, nb_sectors
);
293 blk_aio_preadv(source
, sector_num
* BDRV_SECTOR_SIZE
, &op
->qiov
, 0,
294 mirror_read_complete
, op
);
298 static void mirror_do_zero_or_discard(MirrorBlockJob
*s
,
305 /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed
306 * so the freeing in mirror_iteration_done is nop. */
307 op
= g_new0(MirrorOp
, 1);
309 op
->sector_num
= sector_num
;
310 op
->nb_sectors
= nb_sectors
;
313 s
->sectors_in_flight
+= nb_sectors
;
315 blk_aio_pdiscard(s
->target
, sector_num
<< BDRV_SECTOR_BITS
,
316 op
->nb_sectors
<< BDRV_SECTOR_BITS
,
317 mirror_write_complete
, op
);
319 blk_aio_pwrite_zeroes(s
->target
, sector_num
* BDRV_SECTOR_SIZE
,
320 op
->nb_sectors
* BDRV_SECTOR_SIZE
,
321 s
->unmap
? BDRV_REQ_MAY_UNMAP
: 0,
322 mirror_write_complete
, op
);
326 static uint64_t coroutine_fn
mirror_iteration(MirrorBlockJob
*s
)
328 BlockDriverState
*source
= blk_bs(s
->common
.blk
);
329 int64_t sector_num
, first_chunk
;
330 uint64_t delay_ns
= 0;
331 /* At least the first dirty chunk is mirrored in one iteration. */
333 int64_t end
= s
->bdev_length
/ BDRV_SECTOR_SIZE
;
334 int sectors_per_chunk
= s
->granularity
>> BDRV_SECTOR_BITS
;
335 bool write_zeroes_ok
= bdrv_can_write_zeroes_with_unmap(blk_bs(s
->target
));
336 int max_io_sectors
= MAX((s
->buf_size
>> BDRV_SECTOR_BITS
) / MAX_IN_FLIGHT
,
339 sector_num
= bdrv_dirty_iter_next(s
->dbi
);
340 if (sector_num
< 0) {
341 bdrv_set_dirty_iter(s
->dbi
, 0);
342 sector_num
= bdrv_dirty_iter_next(s
->dbi
);
343 trace_mirror_restart_iter(s
, bdrv_get_dirty_count(s
->dirty_bitmap
));
344 assert(sector_num
>= 0);
347 first_chunk
= sector_num
/ sectors_per_chunk
;
348 while (test_bit(first_chunk
, s
->in_flight_bitmap
)) {
349 trace_mirror_yield_in_flight(s
, sector_num
, s
->in_flight
);
350 mirror_wait_for_io(s
);
353 block_job_pause_point(&s
->common
);
355 /* Find the number of consective dirty chunks following the first dirty
356 * one, and wait for in flight requests in them. */
357 while (nb_chunks
* sectors_per_chunk
< (s
->buf_size
>> BDRV_SECTOR_BITS
)) {
359 int64_t next_sector
= sector_num
+ nb_chunks
* sectors_per_chunk
;
360 int64_t next_chunk
= next_sector
/ sectors_per_chunk
;
361 if (next_sector
>= end
||
362 !bdrv_get_dirty(source
, s
->dirty_bitmap
, next_sector
)) {
365 if (test_bit(next_chunk
, s
->in_flight_bitmap
)) {
369 next_dirty
= bdrv_dirty_iter_next(s
->dbi
);
370 if (next_dirty
> next_sector
|| next_dirty
< 0) {
371 /* The bitmap iterator's cache is stale, refresh it */
372 bdrv_set_dirty_iter(s
->dbi
, next_sector
);
373 next_dirty
= bdrv_dirty_iter_next(s
->dbi
);
375 assert(next_dirty
== next_sector
);
379 /* Clear dirty bits before querying the block status, because
380 * calling bdrv_get_block_status_above could yield - if some blocks are
381 * marked dirty in this window, we need to know.
383 bdrv_reset_dirty_bitmap(s
->dirty_bitmap
, sector_num
,
384 nb_chunks
* sectors_per_chunk
);
385 bitmap_set(s
->in_flight_bitmap
, sector_num
/ sectors_per_chunk
, nb_chunks
);
386 while (nb_chunks
> 0 && sector_num
< end
) {
388 int io_sectors
, io_sectors_acct
;
389 BlockDriverState
*file
;
393 MIRROR_METHOD_DISCARD
394 } mirror_method
= MIRROR_METHOD_COPY
;
396 assert(!(sector_num
% sectors_per_chunk
));
397 ret
= bdrv_get_block_status_above(source
, NULL
, sector_num
,
398 nb_chunks
* sectors_per_chunk
,
401 io_sectors
= MIN(nb_chunks
* sectors_per_chunk
, max_io_sectors
);
402 } else if (ret
& BDRV_BLOCK_DATA
) {
403 io_sectors
= MIN(io_sectors
, max_io_sectors
);
406 io_sectors
-= io_sectors
% sectors_per_chunk
;
407 if (io_sectors
< sectors_per_chunk
) {
408 io_sectors
= sectors_per_chunk
;
409 } else if (ret
>= 0 && !(ret
& BDRV_BLOCK_DATA
)) {
410 int64_t target_sector_num
;
411 int target_nb_sectors
;
412 bdrv_round_sectors_to_clusters(blk_bs(s
->target
), sector_num
,
413 io_sectors
, &target_sector_num
,
415 if (target_sector_num
== sector_num
&&
416 target_nb_sectors
== io_sectors
) {
417 mirror_method
= ret
& BDRV_BLOCK_ZERO
?
419 MIRROR_METHOD_DISCARD
;
423 while (s
->in_flight
>= MAX_IN_FLIGHT
) {
424 trace_mirror_yield_in_flight(s
, sector_num
, s
->in_flight
);
425 mirror_wait_for_io(s
);
432 mirror_clip_sectors(s
, sector_num
, &io_sectors
);
433 switch (mirror_method
) {
434 case MIRROR_METHOD_COPY
:
435 io_sectors
= mirror_do_read(s
, sector_num
, io_sectors
);
436 io_sectors_acct
= io_sectors
;
438 case MIRROR_METHOD_ZERO
:
439 case MIRROR_METHOD_DISCARD
:
440 mirror_do_zero_or_discard(s
, sector_num
, io_sectors
,
441 mirror_method
== MIRROR_METHOD_DISCARD
);
442 if (write_zeroes_ok
) {
445 io_sectors_acct
= io_sectors
;
452 sector_num
+= io_sectors
;
453 nb_chunks
-= DIV_ROUND_UP(io_sectors
, sectors_per_chunk
);
454 if (s
->common
.speed
) {
455 delay_ns
= ratelimit_calculate_delay(&s
->limit
, io_sectors_acct
);
461 static void mirror_free_init(MirrorBlockJob
*s
)
463 int granularity
= s
->granularity
;
464 size_t buf_size
= s
->buf_size
;
465 uint8_t *buf
= s
->buf
;
467 assert(s
->buf_free_count
== 0);
468 QSIMPLEQ_INIT(&s
->buf_free
);
469 while (buf_size
!= 0) {
470 MirrorBuffer
*cur
= (MirrorBuffer
*)buf
;
471 QSIMPLEQ_INSERT_TAIL(&s
->buf_free
, cur
, next
);
473 buf_size
-= granularity
;
478 /* This is also used for the .pause callback. There is no matching
479 * mirror_resume() because mirror_run() will begin iterating again
480 * when the job is resumed.
482 static void mirror_wait_for_all_io(MirrorBlockJob
*s
)
484 while (s
->in_flight
> 0) {
485 mirror_wait_for_io(s
);
493 static void mirror_exit(BlockJob
*job
, void *opaque
)
495 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
);
496 MirrorExitData
*data
= opaque
;
497 AioContext
*replace_aio_context
= NULL
;
498 BlockDriverState
*src
= blk_bs(s
->common
.blk
);
499 BlockDriverState
*target_bs
= blk_bs(s
->target
);
501 /* Make sure that the source BDS doesn't go away before we called
502 * block_job_completed(). */
506 replace_aio_context
= bdrv_get_aio_context(s
->to_replace
);
507 aio_context_acquire(replace_aio_context
);
510 if (s
->should_complete
&& data
->ret
== 0) {
511 BlockDriverState
*to_replace
= src
;
513 to_replace
= s
->to_replace
;
516 if (bdrv_get_flags(target_bs
) != bdrv_get_flags(to_replace
)) {
517 bdrv_reopen(target_bs
, bdrv_get_flags(to_replace
), NULL
);
520 /* The mirror job has no requests in flight any more, but we need to
521 * drain potential other users of the BDS before changing the graph. */
522 bdrv_drained_begin(target_bs
);
523 bdrv_replace_in_backing_chain(to_replace
, target_bs
);
524 bdrv_drained_end(target_bs
);
526 /* We just changed the BDS the job BB refers to */
527 blk_remove_bs(job
->blk
);
528 blk_insert_bs(job
->blk
, src
);
531 bdrv_op_unblock_all(s
->to_replace
, s
->replace_blocker
);
532 error_free(s
->replace_blocker
);
533 bdrv_unref(s
->to_replace
);
535 if (replace_aio_context
) {
536 aio_context_release(replace_aio_context
);
539 blk_unref(s
->target
);
541 block_job_completed(&s
->common
, data
->ret
);
543 bdrv_drained_end(src
);
547 static void mirror_throttle(MirrorBlockJob
*s
)
549 int64_t now
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
551 if (now
- s
->last_pause_ns
> SLICE_TIME
) {
552 s
->last_pause_ns
= now
;
553 block_job_sleep_ns(&s
->common
, QEMU_CLOCK_REALTIME
, 0);
555 block_job_pause_point(&s
->common
);
559 static int coroutine_fn
mirror_dirty_init(MirrorBlockJob
*s
)
561 int64_t sector_num
, end
;
562 BlockDriverState
*base
= s
->base
;
563 BlockDriverState
*bs
= blk_bs(s
->common
.blk
);
564 BlockDriverState
*target_bs
= blk_bs(s
->target
);
567 end
= s
->bdev_length
/ BDRV_SECTOR_SIZE
;
569 if (base
== NULL
&& !bdrv_has_zero_init(target_bs
)) {
570 if (!bdrv_can_write_zeroes_with_unmap(target_bs
)) {
571 bdrv_set_dirty_bitmap(s
->dirty_bitmap
, 0, end
);
575 for (sector_num
= 0; sector_num
< end
; ) {
576 int nb_sectors
= MIN(end
- sector_num
,
577 QEMU_ALIGN_DOWN(INT_MAX
, s
->granularity
) >> BDRV_SECTOR_BITS
);
581 if (block_job_is_cancelled(&s
->common
)) {
585 if (s
->in_flight
>= MAX_IN_FLIGHT
) {
586 trace_mirror_yield(s
, s
->in_flight
, s
->buf_free_count
, -1);
587 mirror_wait_for_io(s
);
591 mirror_do_zero_or_discard(s
, sector_num
, nb_sectors
, false);
592 sector_num
+= nb_sectors
;
595 mirror_wait_for_all_io(s
);
598 /* First part, loop on the sectors and initialize the dirty bitmap. */
599 for (sector_num
= 0; sector_num
< end
; ) {
600 /* Just to make sure we are not exceeding int limit. */
601 int nb_sectors
= MIN(INT_MAX
>> BDRV_SECTOR_BITS
,
606 if (block_job_is_cancelled(&s
->common
)) {
610 ret
= bdrv_is_allocated_above(bs
, base
, sector_num
, nb_sectors
, &n
);
617 bdrv_set_dirty_bitmap(s
->dirty_bitmap
, sector_num
, n
);
624 /* Called when going out of the streaming phase to flush the bulk of the
625 * data to the medium, or just before completing.
627 static int mirror_flush(MirrorBlockJob
*s
)
629 int ret
= blk_flush(s
->target
);
631 if (mirror_error_action(s
, false, -ret
) == BLOCK_ERROR_ACTION_REPORT
) {
638 static void coroutine_fn
mirror_run(void *opaque
)
640 MirrorBlockJob
*s
= opaque
;
641 MirrorExitData
*data
;
642 BlockDriverState
*bs
= blk_bs(s
->common
.blk
);
643 BlockDriverState
*target_bs
= blk_bs(s
->target
);
644 bool need_drain
= true;
647 char backing_filename
[2]; /* we only need 2 characters because we are only
648 checking for a NULL string */
650 int target_cluster_size
= BDRV_SECTOR_SIZE
;
652 if (block_job_is_cancelled(&s
->common
)) {
656 s
->bdev_length
= bdrv_getlength(bs
);
657 if (s
->bdev_length
< 0) {
658 ret
= s
->bdev_length
;
660 } else if (s
->bdev_length
== 0) {
661 /* Report BLOCK_JOB_READY and wait for complete. */
662 block_job_event_ready(&s
->common
);
664 while (!block_job_is_cancelled(&s
->common
) && !s
->should_complete
) {
665 block_job_yield(&s
->common
);
667 s
->common
.cancelled
= false;
671 length
= DIV_ROUND_UP(s
->bdev_length
, s
->granularity
);
672 s
->in_flight_bitmap
= bitmap_new(length
);
674 /* If we have no backing file yet in the destination, we cannot let
675 * the destination do COW. Instead, we copy sectors around the
676 * dirty data if needed. We need a bitmap to do that.
678 bdrv_get_backing_filename(target_bs
, backing_filename
,
679 sizeof(backing_filename
));
680 if (!bdrv_get_info(target_bs
, &bdi
) && bdi
.cluster_size
) {
681 target_cluster_size
= bdi
.cluster_size
;
683 if (backing_filename
[0] && !target_bs
->backing
684 && s
->granularity
< target_cluster_size
) {
685 s
->buf_size
= MAX(s
->buf_size
, target_cluster_size
);
686 s
->cow_bitmap
= bitmap_new(length
);
688 s
->target_cluster_sectors
= target_cluster_size
>> BDRV_SECTOR_BITS
;
689 s
->max_iov
= MIN(bs
->bl
.max_iov
, target_bs
->bl
.max_iov
);
691 s
->buf
= qemu_try_blockalign(bs
, s
->buf_size
);
692 if (s
->buf
== NULL
) {
699 s
->last_pause_ns
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
700 if (!s
->is_none_mode
) {
701 ret
= mirror_dirty_init(s
);
702 if (ret
< 0 || block_job_is_cancelled(&s
->common
)) {
708 s
->dbi
= bdrv_dirty_iter_new(s
->dirty_bitmap
, 0);
710 uint64_t delay_ns
= 0;
712 bool should_complete
;
719 block_job_pause_point(&s
->common
);
721 cnt
= bdrv_get_dirty_count(s
->dirty_bitmap
);
722 /* s->common.offset contains the number of bytes already processed so
723 * far, cnt is the number of dirty sectors remaining and
724 * s->sectors_in_flight is the number of sectors currently being
725 * processed; together those are the current total operation length */
726 s
->common
.len
= s
->common
.offset
+
727 (cnt
+ s
->sectors_in_flight
) * BDRV_SECTOR_SIZE
;
729 /* Note that even when no rate limit is applied we need to yield
730 * periodically with no pending I/O so that bdrv_drain_all() returns.
731 * We do so every SLICE_TIME nanoseconds, or when there is an error,
732 * or when the source is clean, whichever comes first.
734 delta
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - s
->last_pause_ns
;
735 if (delta
< SLICE_TIME
&&
736 s
->common
.iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
737 if (s
->in_flight
>= MAX_IN_FLIGHT
|| s
->buf_free_count
== 0 ||
738 (cnt
== 0 && s
->in_flight
> 0)) {
739 trace_mirror_yield(s
, s
->in_flight
, s
->buf_free_count
, cnt
);
740 mirror_wait_for_io(s
);
742 } else if (cnt
!= 0) {
743 delay_ns
= mirror_iteration(s
);
747 should_complete
= false;
748 if (s
->in_flight
== 0 && cnt
== 0) {
749 trace_mirror_before_flush(s
);
751 if (mirror_flush(s
) < 0) {
752 /* Go check s->ret. */
755 /* We're out of the streaming phase. From now on, if the job
756 * is cancelled we will actually complete all pending I/O and
757 * report completion. This way, block-job-cancel will leave
758 * the target in a consistent state.
760 block_job_event_ready(&s
->common
);
764 should_complete
= s
->should_complete
||
765 block_job_is_cancelled(&s
->common
);
766 cnt
= bdrv_get_dirty_count(s
->dirty_bitmap
);
769 if (cnt
== 0 && should_complete
) {
770 /* The dirty bitmap is not updated while operations are pending.
771 * If we're about to exit, wait for pending operations before
772 * calling bdrv_get_dirty_count(bs), or we may exit while the
773 * source has dirty data to copy!
775 * Note that I/O can be submitted by the guest while
776 * mirror_populate runs, so pause it now. Before deciding
777 * whether to switch to target check one last time if I/O has
778 * come in the meanwhile, and if not flush the data to disk.
780 trace_mirror_before_drain(s
, cnt
);
782 bdrv_drained_begin(bs
);
783 cnt
= bdrv_get_dirty_count(s
->dirty_bitmap
);
784 if (cnt
> 0 || mirror_flush(s
) < 0) {
785 bdrv_drained_end(bs
);
789 /* The two disks are in sync. Exit and report successful
792 assert(QLIST_EMPTY(&bs
->tracked_requests
));
793 s
->common
.cancelled
= false;
799 trace_mirror_before_sleep(s
, cnt
, s
->synced
, delay_ns
);
801 block_job_sleep_ns(&s
->common
, QEMU_CLOCK_REALTIME
, delay_ns
);
802 if (block_job_is_cancelled(&s
->common
)) {
805 } else if (!should_complete
) {
806 delay_ns
= (s
->in_flight
== 0 && cnt
== 0 ? SLICE_TIME
: 0);
807 block_job_sleep_ns(&s
->common
, QEMU_CLOCK_REALTIME
, delay_ns
);
809 s
->last_pause_ns
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
813 if (s
->in_flight
> 0) {
814 /* We get here only if something went wrong. Either the job failed,
815 * or it was cancelled prematurely so that we do not guarantee that
816 * the target is a copy of the source.
818 assert(ret
< 0 || (!s
->synced
&& block_job_is_cancelled(&s
->common
)));
820 mirror_wait_for_all_io(s
);
823 assert(s
->in_flight
== 0);
825 g_free(s
->cow_bitmap
);
826 g_free(s
->in_flight_bitmap
);
827 bdrv_dirty_iter_free(s
->dbi
);
828 bdrv_release_dirty_bitmap(bs
, s
->dirty_bitmap
);
830 data
= g_malloc(sizeof(*data
));
834 bdrv_drained_begin(bs
);
836 block_job_defer_to_main_loop(&s
->common
, mirror_exit
, data
);
839 static void mirror_set_speed(BlockJob
*job
, int64_t speed
, Error
**errp
)
841 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
);
844 error_setg(errp
, QERR_INVALID_PARAMETER
, "speed");
847 ratelimit_set_speed(&s
->limit
, speed
/ BDRV_SECTOR_SIZE
, SLICE_TIME
);
850 static void mirror_complete(BlockJob
*job
, Error
**errp
)
852 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
);
853 BlockDriverState
*src
, *target
;
855 src
= blk_bs(job
->blk
);
856 target
= blk_bs(s
->target
);
859 error_setg(errp
, "The active block job '%s' cannot be completed",
864 if (s
->backing_mode
== MIRROR_OPEN_BACKING_CHAIN
) {
867 assert(!target
->backing
);
868 ret
= bdrv_open_backing_file(target
, NULL
, "backing", errp
);
874 /* block all operations on to_replace bs */
876 AioContext
*replace_aio_context
;
878 s
->to_replace
= bdrv_find_node(s
->replaces
);
879 if (!s
->to_replace
) {
880 error_setg(errp
, "Node name '%s' not found", s
->replaces
);
884 replace_aio_context
= bdrv_get_aio_context(s
->to_replace
);
885 aio_context_acquire(replace_aio_context
);
887 error_setg(&s
->replace_blocker
,
888 "block device is in use by block-job-complete");
889 bdrv_op_block_all(s
->to_replace
, s
->replace_blocker
);
890 bdrv_ref(s
->to_replace
);
892 aio_context_release(replace_aio_context
);
895 if (s
->backing_mode
== MIRROR_SOURCE_BACKING_CHAIN
) {
896 BlockDriverState
*backing
= s
->is_none_mode
? src
: s
->base
;
897 if (backing_bs(target
) != backing
) {
898 bdrv_set_backing_hd(target
, backing
);
902 s
->should_complete
= true;
903 block_job_enter(&s
->common
);
906 static void mirror_pause(BlockJob
*job
)
908 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
);
910 mirror_wait_for_all_io(s
);
913 static void mirror_attached_aio_context(BlockJob
*job
, AioContext
*new_context
)
915 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
);
917 blk_set_aio_context(s
->target
, new_context
);
920 static void mirror_drain(BlockJob
*job
)
922 MirrorBlockJob
*s
= container_of(job
, MirrorBlockJob
, common
);
924 /* Need to keep a reference in case blk_drain triggers execution
925 * of mirror_complete...
928 BlockBackend
*target
= s
->target
;
935 static const BlockJobDriver mirror_job_driver
= {
936 .instance_size
= sizeof(MirrorBlockJob
),
937 .job_type
= BLOCK_JOB_TYPE_MIRROR
,
938 .set_speed
= mirror_set_speed
,
940 .complete
= mirror_complete
,
941 .pause
= mirror_pause
,
942 .attached_aio_context
= mirror_attached_aio_context
,
943 .drain
= mirror_drain
,
946 static const BlockJobDriver commit_active_job_driver
= {
947 .instance_size
= sizeof(MirrorBlockJob
),
948 .job_type
= BLOCK_JOB_TYPE_COMMIT
,
949 .set_speed
= mirror_set_speed
,
951 .complete
= mirror_complete
,
952 .pause
= mirror_pause
,
953 .attached_aio_context
= mirror_attached_aio_context
,
954 .drain
= mirror_drain
,
957 static void mirror_start_job(const char *job_id
, BlockDriverState
*bs
,
958 int creation_flags
, BlockDriverState
*target
,
959 const char *replaces
, int64_t speed
,
960 uint32_t granularity
, int64_t buf_size
,
961 BlockMirrorBackingMode backing_mode
,
962 BlockdevOnError on_source_error
,
963 BlockdevOnError on_target_error
,
965 BlockCompletionFunc
*cb
,
966 void *opaque
, Error
**errp
,
967 const BlockJobDriver
*driver
,
968 bool is_none_mode
, BlockDriverState
*base
,
973 if (granularity
== 0) {
974 granularity
= bdrv_get_default_bitmap_granularity(target
);
977 assert ((granularity
& (granularity
- 1)) == 0);
980 error_setg(errp
, "Invalid parameter 'buf-size'");
985 buf_size
= DEFAULT_MIRROR_BUF_SIZE
;
988 s
= block_job_create(job_id
, driver
, bs
, speed
, creation_flags
,
994 s
->target
= blk_new();
995 blk_insert_bs(s
->target
, target
);
997 s
->replaces
= g_strdup(replaces
);
998 s
->on_source_error
= on_source_error
;
999 s
->on_target_error
= on_target_error
;
1000 s
->is_none_mode
= is_none_mode
;
1001 s
->backing_mode
= backing_mode
;
1003 s
->granularity
= granularity
;
1004 s
->buf_size
= ROUND_UP(buf_size
, granularity
);
1006 if (auto_complete
) {
1007 s
->should_complete
= true;
1010 s
->dirty_bitmap
= bdrv_create_dirty_bitmap(bs
, granularity
, NULL
, errp
);
1011 if (!s
->dirty_bitmap
) {
1012 g_free(s
->replaces
);
1013 blk_unref(s
->target
);
1014 block_job_unref(&s
->common
);
1018 block_job_add_bdrv(&s
->common
, target
);
1019 /* In commit_active_start() all intermediate nodes disappear, so
1020 * any jobs in them must be blocked */
1021 if (bdrv_chain_contains(bs
, target
)) {
1022 BlockDriverState
*iter
;
1023 for (iter
= backing_bs(bs
); iter
!= target
; iter
= backing_bs(iter
)) {
1024 block_job_add_bdrv(&s
->common
, iter
);
1028 trace_mirror_start(bs
, s
, opaque
);
1029 block_job_start(&s
->common
);
1032 void mirror_start(const char *job_id
, BlockDriverState
*bs
,
1033 BlockDriverState
*target
, const char *replaces
,
1034 int64_t speed
, uint32_t granularity
, int64_t buf_size
,
1035 MirrorSyncMode mode
, BlockMirrorBackingMode backing_mode
,
1036 BlockdevOnError on_source_error
,
1037 BlockdevOnError on_target_error
,
1038 bool unmap
, Error
**errp
)
1041 BlockDriverState
*base
;
1043 if (mode
== MIRROR_SYNC_MODE_INCREMENTAL
) {
1044 error_setg(errp
, "Sync mode 'incremental' not supported");
1047 is_none_mode
= mode
== MIRROR_SYNC_MODE_NONE
;
1048 base
= mode
== MIRROR_SYNC_MODE_TOP
? backing_bs(bs
) : NULL
;
1049 mirror_start_job(job_id
, bs
, BLOCK_JOB_DEFAULT
, target
, replaces
,
1050 speed
, granularity
, buf_size
, backing_mode
,
1051 on_source_error
, on_target_error
, unmap
, NULL
, NULL
, errp
,
1052 &mirror_job_driver
, is_none_mode
, base
, false);
1055 void commit_active_start(const char *job_id
, BlockDriverState
*bs
,
1056 BlockDriverState
*base
, int creation_flags
,
1057 int64_t speed
, BlockdevOnError on_error
,
1058 BlockCompletionFunc
*cb
, void *opaque
, Error
**errp
,
1061 int64_t length
, base_length
;
1062 int orig_base_flags
;
1064 Error
*local_err
= NULL
;
1066 orig_base_flags
= bdrv_get_flags(base
);
1068 if (bdrv_reopen(base
, bs
->open_flags
, errp
)) {
1072 length
= bdrv_getlength(bs
);
1074 error_setg_errno(errp
, -length
,
1075 "Unable to determine length of %s", bs
->filename
);
1076 goto error_restore_flags
;
1079 base_length
= bdrv_getlength(base
);
1080 if (base_length
< 0) {
1081 error_setg_errno(errp
, -base_length
,
1082 "Unable to determine length of %s", base
->filename
);
1083 goto error_restore_flags
;
1086 if (length
> base_length
) {
1087 ret
= bdrv_truncate(base
, length
);
1089 error_setg_errno(errp
, -ret
,
1090 "Top image %s is larger than base image %s, and "
1091 "resize of base image failed",
1092 bs
->filename
, base
->filename
);
1093 goto error_restore_flags
;
1097 mirror_start_job(job_id
, bs
, creation_flags
, base
, NULL
, speed
, 0, 0,
1098 MIRROR_LEAVE_BACKING_CHAIN
,
1099 on_error
, on_error
, true, cb
, opaque
, &local_err
,
1100 &commit_active_job_driver
, false, base
, auto_complete
);
1102 error_propagate(errp
, local_err
);
1103 goto error_restore_flags
;
1108 error_restore_flags
:
1109 /* ignore error and errp for bdrv_reopen, because we want to propagate
1110 * the original error */
1111 bdrv_reopen(base
, orig_base_flags
, NULL
);