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 typedef struct CommitBlockJob
{
36 BlockDriverState
*commit_top_bs
;
39 BlockdevOnError on_error
;
41 char *backing_file_str
;
44 static int coroutine_fn
commit_populate(BlockBackend
*bs
, BlockBackend
*base
,
45 int64_t offset
, uint64_t bytes
,
55 assert(bytes
< SIZE_MAX
);
56 qemu_iovec_init_external(&qiov
, &iov
, 1);
58 ret
= blk_co_preadv(bs
, offset
, qiov
.size
, &qiov
, 0);
63 ret
= blk_co_pwritev(base
, offset
, qiov
.size
, &qiov
, 0);
71 static void commit_exit(Job
*job
)
73 CommitBlockJob
*s
= container_of(job
, CommitBlockJob
, common
.job
);
74 BlockJob
*bjob
= &s
->common
;
75 BlockDriverState
*top
= blk_bs(s
->top
);
76 BlockDriverState
*base
= blk_bs(s
->base
);
77 BlockDriverState
*commit_top_bs
= s
->commit_top_bs
;
78 bool remove_commit_top_bs
= false;
80 /* Make sure commit_top_bs and top stay around until bdrv_replace_node() */
82 bdrv_ref(commit_top_bs
);
84 /* Remove base node parent that still uses BLK_PERM_WRITE/RESIZE before
85 * the normal backing chain can be restored. */
88 if (!job_is_cancelled(job
) && job
->ret
== 0) {
90 job
->ret
= bdrv_drop_intermediate(s
->commit_top_bs
, base
,
93 /* XXX Can (or should) we somehow keep 'consistent read' blocked even
94 * after the failed/cancelled commit job is gone? If we already wrote
95 * something to base, the intermediate images aren't valid any more. */
96 remove_commit_top_bs
= true;
99 /* restore base open flags here if appropriate (e.g., change the base back
100 * to r/o). These reopens do not need to be atomic, since we won't abort
101 * even on failure here */
102 if (s
->base_flags
!= bdrv_get_flags(base
)) {
103 bdrv_reopen(base
, s
->base_flags
, NULL
);
105 g_free(s
->backing_file_str
);
108 /* If there is more than one reference to the job (e.g. if called from
109 * job_finish_sync()), job_completed() won't free it and therefore the
110 * blockers on the intermediate nodes remain. This would cause
111 * bdrv_set_backing_hd() to fail. */
112 block_job_remove_all_bdrv(bjob
);
114 /* If bdrv_drop_intermediate() didn't already do that, remove the commit
115 * filter driver from the backing chain. Do this as the final step so that
116 * the 'consistent read' permission can be granted. */
117 if (remove_commit_top_bs
) {
118 bdrv_child_try_set_perm(commit_top_bs
->backing
, 0, BLK_PERM_ALL
,
120 bdrv_replace_node(commit_top_bs
, backing_bs(commit_top_bs
),
124 bdrv_unref(commit_top_bs
);
128 static int coroutine_fn
commit_run(Job
*job
, Error
**errp
)
130 CommitBlockJob
*s
= container_of(job
, CommitBlockJob
, common
.job
);
132 uint64_t delay_ns
= 0;
134 int64_t n
= 0; /* bytes */
136 int bytes_written
= 0;
137 int64_t len
, base_len
;
139 ret
= len
= blk_getlength(s
->top
);
143 job_progress_set_remaining(&s
->common
.job
, len
);
145 ret
= base_len
= blk_getlength(s
->base
);
150 if (base_len
< len
) {
151 ret
= blk_truncate(s
->base
, len
, PREALLOC_MODE_OFF
, NULL
);
157 buf
= blk_blockalign(s
->top
, COMMIT_BUFFER_SIZE
);
159 for (offset
= 0; offset
< len
; offset
+= n
) {
162 /* Note that even when no rate limit is applied we need to yield
163 * with no pending I/O here so that bdrv_drain_all() returns.
165 job_sleep_ns(&s
->common
.job
, delay_ns
);
166 if (job_is_cancelled(&s
->common
.job
)) {
169 /* Copy if allocated above the base */
170 ret
= bdrv_is_allocated_above(blk_bs(s
->top
), blk_bs(s
->base
),
171 offset
, COMMIT_BUFFER_SIZE
, &n
);
173 trace_commit_one_iteration(s
, offset
, n
, ret
);
175 ret
= commit_populate(s
->top
, s
->base
, offset
, n
, buf
);
179 BlockErrorAction action
=
180 block_job_error_action(&s
->common
, false, s
->on_error
, -ret
);
181 if (action
== BLOCK_ERROR_ACTION_REPORT
) {
188 /* Publish progress */
189 job_progress_update(&s
->common
.job
, n
);
192 delay_ns
= block_job_ratelimit_get_delay(&s
->common
, n
);
206 static const BlockJobDriver commit_job_driver
= {
208 .instance_size
= sizeof(CommitBlockJob
),
209 .job_type
= JOB_TYPE_COMMIT
,
210 .free
= block_job_free
,
211 .user_resume
= block_job_user_resume
,
212 .drain
= block_job_drain
,
218 static int coroutine_fn
bdrv_commit_top_preadv(BlockDriverState
*bs
,
219 uint64_t offset
, uint64_t bytes
, QEMUIOVector
*qiov
, int flags
)
221 return bdrv_co_preadv(bs
->backing
, offset
, bytes
, qiov
, flags
);
224 static void bdrv_commit_top_refresh_filename(BlockDriverState
*bs
, QDict
*opts
)
226 bdrv_refresh_filename(bs
->backing
->bs
);
227 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
228 bs
->backing
->bs
->filename
);
231 static void bdrv_commit_top_child_perm(BlockDriverState
*bs
, BdrvChild
*c
,
232 const BdrvChildRole
*role
,
233 BlockReopenQueue
*reopen_queue
,
234 uint64_t perm
, uint64_t shared
,
235 uint64_t *nperm
, uint64_t *nshared
)
238 *nshared
= BLK_PERM_ALL
;
241 /* Dummy node that provides consistent read to its users without requiring it
242 * from its backing file and that allows writes on the backing file chain. */
243 static BlockDriver bdrv_commit_top
= {
244 .format_name
= "commit_top",
245 .bdrv_co_preadv
= bdrv_commit_top_preadv
,
246 .bdrv_co_block_status
= bdrv_co_block_status_from_backing
,
247 .bdrv_refresh_filename
= bdrv_commit_top_refresh_filename
,
248 .bdrv_child_perm
= bdrv_commit_top_child_perm
,
251 void commit_start(const char *job_id
, BlockDriverState
*bs
,
252 BlockDriverState
*base
, BlockDriverState
*top
, int64_t speed
,
253 BlockdevOnError on_error
, const char *backing_file_str
,
254 const char *filter_node_name
, Error
**errp
)
258 BlockDriverState
*iter
;
259 BlockDriverState
*commit_top_bs
= NULL
;
260 Error
*local_err
= NULL
;
265 error_setg(errp
, "Invalid files for merge: top and base are the same");
269 s
= block_job_create(job_id
, &commit_job_driver
, NULL
, bs
, 0, BLK_PERM_ALL
,
270 speed
, JOB_DEFAULT
, NULL
, NULL
, errp
);
275 /* convert base to r/w, if necessary */
276 orig_base_flags
= bdrv_get_flags(base
);
277 if (!(orig_base_flags
& BDRV_O_RDWR
)) {
278 bdrv_reopen(base
, orig_base_flags
| BDRV_O_RDWR
, &local_err
);
279 if (local_err
!= NULL
) {
280 error_propagate(errp
, local_err
);
285 /* Insert commit_top block node above top, so we can block consistent read
286 * on the backing chain below it */
287 commit_top_bs
= bdrv_new_open_driver(&bdrv_commit_top
, filter_node_name
, 0,
289 if (commit_top_bs
== NULL
) {
292 if (!filter_node_name
) {
293 commit_top_bs
->implicit
= true;
295 commit_top_bs
->total_sectors
= top
->total_sectors
;
296 bdrv_set_aio_context(commit_top_bs
, bdrv_get_aio_context(top
));
298 bdrv_set_backing_hd(commit_top_bs
, top
, &local_err
);
300 bdrv_unref(commit_top_bs
);
301 commit_top_bs
= NULL
;
302 error_propagate(errp
, local_err
);
305 bdrv_replace_node(top
, commit_top_bs
, &local_err
);
307 bdrv_unref(commit_top_bs
);
308 commit_top_bs
= NULL
;
309 error_propagate(errp
, local_err
);
313 s
->commit_top_bs
= commit_top_bs
;
314 bdrv_unref(commit_top_bs
);
316 /* Block all nodes between top and base, because they will
317 * disappear from the chain after this operation. */
318 assert(bdrv_chain_contains(top
, base
));
319 for (iter
= top
; iter
!= base
; iter
= backing_bs(iter
)) {
320 /* XXX BLK_PERM_WRITE needs to be allowed so we don't block ourselves
321 * at s->base (if writes are blocked for a node, they are also blocked
322 * for its backing file). The other options would be a second filter
323 * driver above s->base. */
324 ret
= block_job_add_bdrv(&s
->common
, "intermediate node", iter
, 0,
325 BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_WRITE
,
332 ret
= block_job_add_bdrv(&s
->common
, "base", base
, 0, BLK_PERM_ALL
, errp
);
337 s
->base
= blk_new(BLK_PERM_CONSISTENT_READ
340 BLK_PERM_CONSISTENT_READ
342 | BLK_PERM_WRITE_UNCHANGED
);
343 ret
= blk_insert_bs(s
->base
, base
, errp
);
348 /* Required permissions are already taken with block_job_add_bdrv() */
349 s
->top
= blk_new(0, BLK_PERM_ALL
);
350 ret
= blk_insert_bs(s
->top
, top
, errp
);
355 s
->base_flags
= orig_base_flags
;
356 s
->backing_file_str
= g_strdup(backing_file_str
);
357 s
->on_error
= on_error
;
359 trace_commit_start(bs
, base
, top
, s
);
360 job_start(&s
->common
.job
);
371 bdrv_replace_node(commit_top_bs
, top
, &error_abort
);
373 job_early_fail(&s
->common
.job
);
377 #define COMMIT_BUF_SIZE (2048 * BDRV_SECTOR_SIZE)
379 /* commit COW file into the raw image */
380 int bdrv_commit(BlockDriverState
*bs
)
382 BlockBackend
*src
, *backing
;
383 BlockDriverState
*backing_file_bs
= NULL
;
384 BlockDriverState
*commit_top_bs
= NULL
;
385 BlockDriver
*drv
= bs
->drv
;
386 int64_t offset
, length
, backing_length
;
391 Error
*local_err
= NULL
;
400 if (bdrv_op_is_blocked(bs
, BLOCK_OP_TYPE_COMMIT_SOURCE
, NULL
) ||
401 bdrv_op_is_blocked(bs
->backing
->bs
, BLOCK_OP_TYPE_COMMIT_TARGET
, NULL
)) {
405 ro
= bs
->backing
->bs
->read_only
;
406 open_flags
= bs
->backing
->bs
->open_flags
;
409 if (bdrv_reopen(bs
->backing
->bs
, open_flags
| BDRV_O_RDWR
, NULL
)) {
414 src
= blk_new(BLK_PERM_CONSISTENT_READ
, BLK_PERM_ALL
);
415 backing
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
);
417 ret
= blk_insert_bs(src
, bs
, &local_err
);
419 error_report_err(local_err
);
423 /* Insert commit_top block node above backing, so we can write to it */
424 backing_file_bs
= backing_bs(bs
);
426 commit_top_bs
= bdrv_new_open_driver(&bdrv_commit_top
, NULL
, BDRV_O_RDWR
,
428 if (commit_top_bs
== NULL
) {
429 error_report_err(local_err
);
432 bdrv_set_aio_context(commit_top_bs
, bdrv_get_aio_context(backing_file_bs
));
434 bdrv_set_backing_hd(commit_top_bs
, backing_file_bs
, &error_abort
);
435 bdrv_set_backing_hd(bs
, commit_top_bs
, &error_abort
);
437 ret
= blk_insert_bs(backing
, backing_file_bs
, &local_err
);
439 error_report_err(local_err
);
443 length
= blk_getlength(src
);
449 backing_length
= blk_getlength(backing
);
450 if (backing_length
< 0) {
451 ret
= backing_length
;
455 /* If our top snapshot is larger than the backing file image,
456 * grow the backing file image if possible. If not possible,
457 * we must return an error */
458 if (length
> backing_length
) {
459 ret
= blk_truncate(backing
, length
, PREALLOC_MODE_OFF
, &local_err
);
461 error_report_err(local_err
);
466 /* blk_try_blockalign() for src will choose an alignment that works for
467 * backing as well, so no need to compare the alignment manually. */
468 buf
= blk_try_blockalign(src
, COMMIT_BUF_SIZE
);
474 for (offset
= 0; offset
< length
; offset
+= n
) {
475 ret
= bdrv_is_allocated(bs
, offset
, COMMIT_BUF_SIZE
, &n
);
480 ret
= blk_pread(src
, offset
, buf
, n
);
485 ret
= blk_pwrite(backing
, offset
, buf
, n
, 0);
492 if (drv
->bdrv_make_empty
) {
493 ret
= drv
->bdrv_make_empty(bs
);
501 * Make sure all data we wrote to the backing device is actually
511 if (backing_file_bs
) {
512 bdrv_set_backing_hd(bs
, backing_file_bs
, &error_abort
);
514 bdrv_unref(commit_top_bs
);
518 /* ignoring error return here */
519 bdrv_reopen(bs
->backing
->bs
, open_flags
& ~BDRV_O_RDWR
, NULL
);