4 * Copyright Red Hat, Inc. 2012
7 * Jeff Cody <jcody@redhat.com>
8 * Based on stream.c by Stefan Hajnoczi
10 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
11 * See the COPYING.LIB file in the top-level directory.
15 #include "qemu/osdep.h"
16 #include "qemu/cutils.h"
18 #include "block/block_int.h"
19 #include "block/blockjob_int.h"
20 #include "qapi/error.h"
21 #include "qapi/qmp/qerror.h"
22 #include "qemu/ratelimit.h"
23 #include "sysemu/block-backend.h"
27 * Size of data buffer for populating the image file. This should be large
28 * enough to process multiple clusters in a single call, so that populating
29 * contiguous regions of the image is efficient.
31 COMMIT_BUFFER_SIZE
= 512 * 1024, /* in bytes */
34 #define SLICE_TIME 100000000ULL /* ns */
36 typedef struct CommitBlockJob
{
39 BlockDriverState
*active
;
40 BlockDriverState
*commit_top_bs
;
43 BlockdevOnError on_error
;
45 int orig_overlay_flags
;
46 char *backing_file_str
;
49 static int coroutine_fn
commit_populate(BlockBackend
*bs
, BlockBackend
*base
,
50 int64_t sector_num
, int nb_sectors
,
57 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
60 qemu_iovec_init_external(&qiov
, &iov
, 1);
62 ret
= blk_co_preadv(bs
, sector_num
* BDRV_SECTOR_SIZE
,
68 ret
= blk_co_pwritev(base
, sector_num
* BDRV_SECTOR_SIZE
,
81 static void commit_complete(BlockJob
*job
, void *opaque
)
83 CommitBlockJob
*s
= container_of(job
, CommitBlockJob
, common
);
84 CommitCompleteData
*data
= opaque
;
85 BlockDriverState
*active
= s
->active
;
86 BlockDriverState
*top
= blk_bs(s
->top
);
87 BlockDriverState
*base
= blk_bs(s
->base
);
88 BlockDriverState
*overlay_bs
= bdrv_find_overlay(active
, s
->commit_top_bs
);
90 bool remove_commit_top_bs
= false;
92 /* Make sure overlay_bs and top stay around until bdrv_set_backing_hd() */
96 /* Remove base node parent that still uses BLK_PERM_WRITE/RESIZE before
97 * the normal backing chain can be restored. */
100 if (!block_job_is_cancelled(&s
->common
) && ret
== 0) {
102 ret
= bdrv_drop_intermediate(active
, s
->commit_top_bs
, base
,
103 s
->backing_file_str
);
104 } else if (overlay_bs
) {
105 /* XXX Can (or should) we somehow keep 'consistent read' blocked even
106 * after the failed/cancelled commit job is gone? If we already wrote
107 * something to base, the intermediate images aren't valid any more. */
108 remove_commit_top_bs
= true;
111 /* restore base open flags here if appropriate (e.g., change the base back
112 * to r/o). These reopens do not need to be atomic, since we won't abort
113 * even on failure here */
114 if (s
->base_flags
!= bdrv_get_flags(base
)) {
115 bdrv_reopen(base
, s
->base_flags
, NULL
);
117 if (overlay_bs
&& s
->orig_overlay_flags
!= bdrv_get_flags(overlay_bs
)) {
118 bdrv_reopen(overlay_bs
, s
->orig_overlay_flags
, NULL
);
120 g_free(s
->backing_file_str
);
122 block_job_completed(&s
->common
, ret
);
125 /* If bdrv_drop_intermediate() didn't already do that, remove the commit
126 * filter driver from the backing chain. Do this as the final step so that
127 * the 'consistent read' permission can be granted. */
128 if (remove_commit_top_bs
) {
129 bdrv_set_backing_hd(overlay_bs
, top
, &error_abort
);
132 bdrv_unref(overlay_bs
);
136 static void coroutine_fn
commit_run(void *opaque
)
138 CommitBlockJob
*s
= opaque
;
139 CommitCompleteData
*data
;
140 int64_t sector_num
, end
;
141 uint64_t delay_ns
= 0;
145 int bytes_written
= 0;
148 ret
= s
->common
.len
= blk_getlength(s
->top
);
151 if (s
->common
.len
< 0) {
155 ret
= base_len
= blk_getlength(s
->base
);
160 if (base_len
< s
->common
.len
) {
161 ret
= blk_truncate(s
->base
, s
->common
.len
, NULL
);
167 end
= s
->common
.len
>> BDRV_SECTOR_BITS
;
168 buf
= blk_blockalign(s
->top
, COMMIT_BUFFER_SIZE
);
170 for (sector_num
= 0; sector_num
< end
; sector_num
+= n
) {
173 /* Note that even when no rate limit is applied we need to yield
174 * with no pending I/O here so that bdrv_drain_all() returns.
176 block_job_sleep_ns(&s
->common
, QEMU_CLOCK_REALTIME
, delay_ns
);
177 if (block_job_is_cancelled(&s
->common
)) {
180 /* Copy if allocated above the base */
181 ret
= bdrv_is_allocated_above(blk_bs(s
->top
), blk_bs(s
->base
),
183 COMMIT_BUFFER_SIZE
/ BDRV_SECTOR_SIZE
,
186 trace_commit_one_iteration(s
, sector_num
, n
, ret
);
188 ret
= commit_populate(s
->top
, s
->base
, sector_num
, n
, buf
);
189 bytes_written
+= n
* BDRV_SECTOR_SIZE
;
192 BlockErrorAction action
=
193 block_job_error_action(&s
->common
, false, s
->on_error
, -ret
);
194 if (action
== BLOCK_ERROR_ACTION_REPORT
) {
201 /* Publish progress */
202 s
->common
.offset
+= n
* BDRV_SECTOR_SIZE
;
204 if (copy
&& s
->common
.speed
) {
205 delay_ns
= ratelimit_calculate_delay(&s
->limit
, n
);
214 data
= g_malloc(sizeof(*data
));
216 block_job_defer_to_main_loop(&s
->common
, commit_complete
, data
);
219 static void commit_set_speed(BlockJob
*job
, int64_t speed
, Error
**errp
)
221 CommitBlockJob
*s
= container_of(job
, CommitBlockJob
, common
);
224 error_setg(errp
, QERR_INVALID_PARAMETER
, "speed");
227 ratelimit_set_speed(&s
->limit
, speed
/ BDRV_SECTOR_SIZE
, SLICE_TIME
);
230 static const BlockJobDriver commit_job_driver
= {
231 .instance_size
= sizeof(CommitBlockJob
),
232 .job_type
= BLOCK_JOB_TYPE_COMMIT
,
233 .set_speed
= commit_set_speed
,
237 static int coroutine_fn
bdrv_commit_top_preadv(BlockDriverState
*bs
,
238 uint64_t offset
, uint64_t bytes
, QEMUIOVector
*qiov
, int flags
)
240 return bdrv_co_preadv(bs
->backing
, offset
, bytes
, qiov
, flags
);
243 static int64_t coroutine_fn
bdrv_commit_top_get_block_status(
244 BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
, int *pnum
,
245 BlockDriverState
**file
)
248 *file
= bs
->backing
->bs
;
249 return BDRV_BLOCK_RAW
| BDRV_BLOCK_OFFSET_VALID
| BDRV_BLOCK_DATA
|
250 (sector_num
<< BDRV_SECTOR_BITS
);
253 static void bdrv_commit_top_refresh_filename(BlockDriverState
*bs
, QDict
*opts
)
255 bdrv_refresh_filename(bs
->backing
->bs
);
256 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
257 bs
->backing
->bs
->filename
);
260 static void bdrv_commit_top_close(BlockDriverState
*bs
)
264 static void bdrv_commit_top_child_perm(BlockDriverState
*bs
, BdrvChild
*c
,
265 const BdrvChildRole
*role
,
266 uint64_t perm
, uint64_t shared
,
267 uint64_t *nperm
, uint64_t *nshared
)
270 *nshared
= BLK_PERM_ALL
;
273 /* Dummy node that provides consistent read to its users without requiring it
274 * from its backing file and that allows writes on the backing file chain. */
275 static BlockDriver bdrv_commit_top
= {
276 .format_name
= "commit_top",
277 .bdrv_co_preadv
= bdrv_commit_top_preadv
,
278 .bdrv_co_get_block_status
= bdrv_commit_top_get_block_status
,
279 .bdrv_refresh_filename
= bdrv_commit_top_refresh_filename
,
280 .bdrv_close
= bdrv_commit_top_close
,
281 .bdrv_child_perm
= bdrv_commit_top_child_perm
,
284 void commit_start(const char *job_id
, BlockDriverState
*bs
,
285 BlockDriverState
*base
, BlockDriverState
*top
, int64_t speed
,
286 BlockdevOnError on_error
, const char *backing_file_str
,
287 const char *filter_node_name
, Error
**errp
)
290 BlockReopenQueue
*reopen_queue
= NULL
;
291 int orig_overlay_flags
;
293 BlockDriverState
*iter
;
294 BlockDriverState
*overlay_bs
;
295 BlockDriverState
*commit_top_bs
= NULL
;
296 Error
*local_err
= NULL
;
301 error_setg(errp
, "Invalid files for merge: top and base are the same");
305 overlay_bs
= bdrv_find_overlay(bs
, top
);
307 if (overlay_bs
== NULL
) {
308 error_setg(errp
, "Could not find overlay image for %s:", top
->filename
);
312 s
= block_job_create(job_id
, &commit_job_driver
, bs
, 0, BLK_PERM_ALL
,
313 speed
, BLOCK_JOB_DEFAULT
, NULL
, NULL
, errp
);
318 orig_base_flags
= bdrv_get_flags(base
);
319 orig_overlay_flags
= bdrv_get_flags(overlay_bs
);
321 /* convert base & overlay_bs to r/w, if necessary */
322 if (!(orig_base_flags
& BDRV_O_RDWR
)) {
323 reopen_queue
= bdrv_reopen_queue(reopen_queue
, base
, NULL
,
324 orig_base_flags
| BDRV_O_RDWR
);
326 if (!(orig_overlay_flags
& BDRV_O_RDWR
)) {
327 reopen_queue
= bdrv_reopen_queue(reopen_queue
, overlay_bs
, NULL
,
328 orig_overlay_flags
| BDRV_O_RDWR
);
331 bdrv_reopen_multiple(bdrv_get_aio_context(bs
), reopen_queue
, &local_err
);
332 if (local_err
!= NULL
) {
333 error_propagate(errp
, local_err
);
338 /* Insert commit_top block node above top, so we can block consistent read
339 * on the backing chain below it */
340 commit_top_bs
= bdrv_new_open_driver(&bdrv_commit_top
, filter_node_name
, 0,
342 if (commit_top_bs
== NULL
) {
345 commit_top_bs
->total_sectors
= top
->total_sectors
;
346 bdrv_set_aio_context(commit_top_bs
, bdrv_get_aio_context(top
));
348 bdrv_set_backing_hd(commit_top_bs
, top
, &local_err
);
350 bdrv_unref(commit_top_bs
);
351 commit_top_bs
= NULL
;
352 error_propagate(errp
, local_err
);
355 bdrv_set_backing_hd(overlay_bs
, commit_top_bs
, &local_err
);
357 bdrv_unref(commit_top_bs
);
358 commit_top_bs
= NULL
;
359 error_propagate(errp
, local_err
);
363 s
->commit_top_bs
= commit_top_bs
;
364 bdrv_unref(commit_top_bs
);
366 /* Block all nodes between top and base, because they will
367 * disappear from the chain after this operation. */
368 assert(bdrv_chain_contains(top
, base
));
369 for (iter
= top
; iter
!= base
; iter
= backing_bs(iter
)) {
370 /* XXX BLK_PERM_WRITE needs to be allowed so we don't block ourselves
371 * at s->base (if writes are blocked for a node, they are also blocked
372 * for its backing file). The other options would be a second filter
373 * driver above s->base. */
374 ret
= block_job_add_bdrv(&s
->common
, "intermediate node", iter
, 0,
375 BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_WRITE
,
382 ret
= block_job_add_bdrv(&s
->common
, "base", base
, 0, BLK_PERM_ALL
, errp
);
387 /* overlay_bs must be blocked because it needs to be modified to
388 * update the backing image string. */
389 ret
= block_job_add_bdrv(&s
->common
, "overlay of top", overlay_bs
,
390 BLK_PERM_GRAPH_MOD
, BLK_PERM_ALL
, errp
);
395 s
->base
= blk_new(BLK_PERM_CONSISTENT_READ
398 BLK_PERM_CONSISTENT_READ
400 | BLK_PERM_WRITE_UNCHANGED
);
401 ret
= blk_insert_bs(s
->base
, base
, errp
);
406 /* Required permissions are already taken with block_job_add_bdrv() */
407 s
->top
= blk_new(0, BLK_PERM_ALL
);
408 ret
= blk_insert_bs(s
->top
, top
, errp
);
415 s
->base_flags
= orig_base_flags
;
416 s
->orig_overlay_flags
= orig_overlay_flags
;
418 s
->backing_file_str
= g_strdup(backing_file_str
);
420 s
->on_error
= on_error
;
422 trace_commit_start(bs
, base
, top
, s
);
423 block_job_start(&s
->common
);
434 bdrv_set_backing_hd(overlay_bs
, top
, &error_abort
);
436 block_job_early_fail(&s
->common
);
440 #define COMMIT_BUF_SECTORS 2048
442 /* commit COW file into the raw image */
443 int bdrv_commit(BlockDriverState
*bs
)
445 BlockBackend
*src
, *backing
;
446 BlockDriverState
*backing_file_bs
= NULL
;
447 BlockDriverState
*commit_top_bs
= NULL
;
448 BlockDriver
*drv
= bs
->drv
;
449 int64_t sector
, total_sectors
, length
, backing_length
;
450 int n
, ro
, open_flags
;
453 Error
*local_err
= NULL
;
462 if (bdrv_op_is_blocked(bs
, BLOCK_OP_TYPE_COMMIT_SOURCE
, NULL
) ||
463 bdrv_op_is_blocked(bs
->backing
->bs
, BLOCK_OP_TYPE_COMMIT_TARGET
, NULL
)) {
467 ro
= bs
->backing
->bs
->read_only
;
468 open_flags
= bs
->backing
->bs
->open_flags
;
471 if (bdrv_reopen(bs
->backing
->bs
, open_flags
| BDRV_O_RDWR
, NULL
)) {
476 src
= blk_new(BLK_PERM_CONSISTENT_READ
, BLK_PERM_ALL
);
477 backing
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
);
479 ret
= blk_insert_bs(src
, bs
, &local_err
);
481 error_report_err(local_err
);
485 /* Insert commit_top block node above backing, so we can write to it */
486 backing_file_bs
= backing_bs(bs
);
488 commit_top_bs
= bdrv_new_open_driver(&bdrv_commit_top
, NULL
, BDRV_O_RDWR
,
490 if (commit_top_bs
== NULL
) {
491 error_report_err(local_err
);
494 bdrv_set_aio_context(commit_top_bs
, bdrv_get_aio_context(backing_file_bs
));
496 bdrv_set_backing_hd(commit_top_bs
, backing_file_bs
, &error_abort
);
497 bdrv_set_backing_hd(bs
, commit_top_bs
, &error_abort
);
499 ret
= blk_insert_bs(backing
, backing_file_bs
, &local_err
);
501 error_report_err(local_err
);
505 length
= blk_getlength(src
);
511 backing_length
= blk_getlength(backing
);
512 if (backing_length
< 0) {
513 ret
= backing_length
;
517 /* If our top snapshot is larger than the backing file image,
518 * grow the backing file image if possible. If not possible,
519 * we must return an error */
520 if (length
> backing_length
) {
521 ret
= blk_truncate(backing
, length
, &local_err
);
523 error_report_err(local_err
);
528 total_sectors
= length
>> BDRV_SECTOR_BITS
;
530 /* blk_try_blockalign() for src will choose an alignment that works for
531 * backing as well, so no need to compare the alignment manually. */
532 buf
= blk_try_blockalign(src
, COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
538 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
539 ret
= bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
);
544 ret
= blk_pread(src
, sector
* BDRV_SECTOR_SIZE
, buf
,
545 n
* BDRV_SECTOR_SIZE
);
550 ret
= blk_pwrite(backing
, sector
* BDRV_SECTOR_SIZE
, buf
,
551 n
* BDRV_SECTOR_SIZE
, 0);
558 if (drv
->bdrv_make_empty
) {
559 ret
= drv
->bdrv_make_empty(bs
);
567 * Make sure all data we wrote to the backing device is actually
577 if (backing_file_bs
) {
578 bdrv_set_backing_hd(bs
, backing_file_bs
, &error_abort
);
580 bdrv_unref(commit_top_bs
);
584 /* ignoring error return here */
585 bdrv_reopen(bs
->backing
->bs
, open_flags
& ~BDRV_O_RDWR
, NULL
);