2 * copy-before-write filter driver
4 * The driver performs Copy-Before-Write (CBW) operation: it is injected above
5 * some node, and before each write it copies _old_ data to the target node.
7 * Copyright (c) 2018-2021 Virtuozzo International GmbH.
10 * Sementsov-Ogievskiy Vladimir <vsementsov@virtuozzo.com>
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License as published by
14 * the Free Software Foundation; either version 2 of the License, or
15 * (at your option) any later version.
17 * This program is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU General Public License for more details.
22 * You should have received a copy of the GNU General Public License
23 * along with this program. If not, see <http://www.gnu.org/licenses/>.
26 #include "qemu/osdep.h"
27 #include "qapi/qmp/qjson.h"
29 #include "sysemu/block-backend.h"
30 #include "qemu/cutils.h"
31 #include "qapi/error.h"
32 #include "block/block_int.h"
33 #include "block/qdict.h"
34 #include "block/block-copy.h"
36 #include "block/copy-before-write.h"
37 #include "block/reqlist.h"
39 #include "qapi/qapi-visit-block-core.h"
41 typedef struct BDRVCopyBeforeWriteState
{
44 OnCbwError on_cbw_error
;
47 * @lock: protects access to @access_bitmap, @done_bitmap and
53 * @access_bitmap: represents areas allowed for reading by fleecing user.
54 * Reading from non-dirty areas leads to -EACCES.
56 BdrvDirtyBitmap
*access_bitmap
;
59 * @done_bitmap: represents areas that was successfully copied to @target by
60 * copy-before-write operations.
62 BdrvDirtyBitmap
*done_bitmap
;
65 * @frozen_read_reqs: current read requests for fleecing user in bs->file
66 * node. These areas must not be rewritten by guest.
68 BlockReqList frozen_read_reqs
;
71 * @snapshot_error is normally zero. But on first copy-before-write failure
72 * when @on_cbw_error == ON_CBW_ERROR_BREAK_SNAPSHOT, @snapshot_error takes
73 * value of this error (<0). After that all in-flight and further
74 * snapshot-API requests will fail with that error.
77 } BDRVCopyBeforeWriteState
;
79 static coroutine_fn
int cbw_co_preadv(
80 BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
81 QEMUIOVector
*qiov
, BdrvRequestFlags flags
)
83 return bdrv_co_preadv(bs
->file
, offset
, bytes
, qiov
, flags
);
87 * Do copy-before-write operation.
89 * On failure guest request must be failed too.
91 * On success, we also wait for all in-flight fleecing read requests in source
92 * node, and it's guaranteed that after cbw_do_copy_before_write() successful
93 * return there are no such requests and they will never appear.
95 static coroutine_fn
int cbw_do_copy_before_write(BlockDriverState
*bs
,
96 uint64_t offset
, uint64_t bytes
, BdrvRequestFlags flags
)
98 BDRVCopyBeforeWriteState
*s
= bs
->opaque
;
101 int64_t cluster_size
= block_copy_cluster_size(s
->bcs
);
103 if (flags
& BDRV_REQ_WRITE_UNCHANGED
) {
107 if (s
->snapshot_error
) {
111 off
= QEMU_ALIGN_DOWN(offset
, cluster_size
);
112 end
= QEMU_ALIGN_UP(offset
+ bytes
, cluster_size
);
114 ret
= block_copy(s
->bcs
, off
, end
- off
, true, 0, NULL
, NULL
);
115 if (ret
< 0 && s
->on_cbw_error
== ON_CBW_ERROR_BREAK_GUEST_WRITE
) {
119 WITH_QEMU_LOCK_GUARD(&s
->lock
) {
121 assert(s
->on_cbw_error
== ON_CBW_ERROR_BREAK_SNAPSHOT
);
122 if (!s
->snapshot_error
) {
123 s
->snapshot_error
= ret
;
126 bdrv_set_dirty_bitmap(s
->done_bitmap
, off
, end
- off
);
128 reqlist_wait_all(&s
->frozen_read_reqs
, off
, end
- off
, &s
->lock
);
134 static int coroutine_fn
cbw_co_pdiscard(BlockDriverState
*bs
,
135 int64_t offset
, int64_t bytes
)
137 int ret
= cbw_do_copy_before_write(bs
, offset
, bytes
, 0);
142 return bdrv_co_pdiscard(bs
->file
, offset
, bytes
);
145 static int coroutine_fn
cbw_co_pwrite_zeroes(BlockDriverState
*bs
,
146 int64_t offset
, int64_t bytes
, BdrvRequestFlags flags
)
148 int ret
= cbw_do_copy_before_write(bs
, offset
, bytes
, flags
);
153 return bdrv_co_pwrite_zeroes(bs
->file
, offset
, bytes
, flags
);
156 static coroutine_fn
int cbw_co_pwritev(BlockDriverState
*bs
,
160 BdrvRequestFlags flags
)
162 int ret
= cbw_do_copy_before_write(bs
, offset
, bytes
, flags
);
167 return bdrv_co_pwritev(bs
->file
, offset
, bytes
, qiov
, flags
);
170 static int coroutine_fn
cbw_co_flush(BlockDriverState
*bs
)
176 return bdrv_co_flush(bs
->file
->bs
);
180 * If @offset not accessible - return NULL.
182 * Otherwise, set @pnum to some bytes that accessible from @file (@file is set
183 * to bs->file or to s->target). Return newly allocated BlockReq object that
184 * should be than passed to cbw_snapshot_read_unlock().
186 * It's guaranteed that guest writes will not interact in the region until
187 * cbw_snapshot_read_unlock() called.
189 static BlockReq
*cbw_snapshot_read_lock(BlockDriverState
*bs
,
190 int64_t offset
, int64_t bytes
,
191 int64_t *pnum
, BdrvChild
**file
)
193 BDRVCopyBeforeWriteState
*s
= bs
->opaque
;
194 BlockReq
*req
= g_new(BlockReq
, 1);
197 QEMU_LOCK_GUARD(&s
->lock
);
199 if (s
->snapshot_error
) {
204 if (bdrv_dirty_bitmap_next_zero(s
->access_bitmap
, offset
, bytes
) != -1) {
209 done
= bdrv_dirty_bitmap_status(s
->done_bitmap
, offset
, bytes
, pnum
);
212 * Special invalid BlockReq, that is handled in
213 * cbw_snapshot_read_unlock(). We don't need to lock something to read
216 *req
= (BlockReq
) {.offset
= -1, .bytes
= -1};
219 reqlist_init_req(&s
->frozen_read_reqs
, req
, offset
, bytes
);
226 static void cbw_snapshot_read_unlock(BlockDriverState
*bs
, BlockReq
*req
)
228 BDRVCopyBeforeWriteState
*s
= bs
->opaque
;
230 if (req
->offset
== -1 && req
->bytes
== -1) {
235 QEMU_LOCK_GUARD(&s
->lock
);
237 reqlist_remove_req(req
);
241 static coroutine_fn
int
242 cbw_co_preadv_snapshot(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
243 QEMUIOVector
*qiov
, size_t qiov_offset
)
249 /* TODO: upgrade to async loop using AioTask */
253 req
= cbw_snapshot_read_lock(bs
, offset
, bytes
, &cur_bytes
, &file
);
258 ret
= bdrv_co_preadv_part(file
, offset
, cur_bytes
,
259 qiov
, qiov_offset
, 0);
260 cbw_snapshot_read_unlock(bs
, req
);
267 qiov_offset
+= cur_bytes
;
273 static int coroutine_fn
274 cbw_co_snapshot_block_status(BlockDriverState
*bs
,
275 bool want_zero
, int64_t offset
, int64_t bytes
,
276 int64_t *pnum
, int64_t *map
,
277 BlockDriverState
**file
)
279 BDRVCopyBeforeWriteState
*s
= bs
->opaque
;
285 req
= cbw_snapshot_read_lock(bs
, offset
, bytes
, &cur_bytes
, &child
);
290 ret
= bdrv_block_status(child
->bs
, offset
, cur_bytes
, pnum
, map
, file
);
291 if (child
== s
->target
) {
293 * We refer to s->target only for areas that we've written to it.
294 * And we can not report unallocated blocks in s->target: this will
295 * break generic block-status-above logic, that will go to
296 * copy-before-write filtered child in this case.
298 assert(ret
& BDRV_BLOCK_ALLOCATED
);
301 cbw_snapshot_read_unlock(bs
, req
);
306 static int coroutine_fn
cbw_co_pdiscard_snapshot(BlockDriverState
*bs
,
307 int64_t offset
, int64_t bytes
)
309 BDRVCopyBeforeWriteState
*s
= bs
->opaque
;
311 WITH_QEMU_LOCK_GUARD(&s
->lock
) {
312 bdrv_reset_dirty_bitmap(s
->access_bitmap
, offset
, bytes
);
315 block_copy_reset(s
->bcs
, offset
, bytes
);
317 return bdrv_co_pdiscard(s
->target
, offset
, bytes
);
320 static void cbw_refresh_filename(BlockDriverState
*bs
)
322 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
323 bs
->file
->bs
->filename
);
326 static void cbw_child_perm(BlockDriverState
*bs
, BdrvChild
*c
,
328 BlockReopenQueue
*reopen_queue
,
329 uint64_t perm
, uint64_t shared
,
330 uint64_t *nperm
, uint64_t *nshared
)
332 if (!(role
& BDRV_CHILD_FILTERED
)) {
336 * Share write to target (child_file), to not interfere
337 * with guest writes to its disk which may be in target backing chain.
338 * Can't resize during a backup block job because we check the size
341 *nshared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
342 *nperm
= BLK_PERM_WRITE
;
345 bdrv_default_perms(bs
, c
, role
, reopen_queue
,
346 perm
, shared
, nperm
, nshared
);
348 if (!QLIST_EMPTY(&bs
->parents
)) {
349 if (perm
& BLK_PERM_WRITE
) {
350 *nperm
= *nperm
| BLK_PERM_CONSISTENT_READ
;
352 *nshared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
357 static BlockdevOptions
*cbw_parse_options(QDict
*options
, Error
**errp
)
359 BlockdevOptions
*opts
= NULL
;
362 qdict_put_str(options
, "driver", "copy-before-write");
364 v
= qobject_input_visitor_new_flat_confused(options
, errp
);
369 visit_type_BlockdevOptions(v
, NULL
, &opts
, errp
);
375 * Delete options which we are going to parse through BlockdevOptions
376 * object for original options.
378 qdict_extract_subqdict(options
, NULL
, "bitmap");
379 qdict_del(options
, "on-cbw-error");
383 qdict_del(options
, "driver");
388 static int cbw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
391 BDRVCopyBeforeWriteState
*s
= bs
->opaque
;
392 BdrvDirtyBitmap
*bitmap
= NULL
;
393 int64_t cluster_size
;
394 g_autoptr(BlockdevOptions
) full_opts
= NULL
;
395 BlockdevOptionsCbw
*opts
;
397 full_opts
= cbw_parse_options(options
, errp
);
401 assert(full_opts
->driver
== BLOCKDEV_DRIVER_COPY_BEFORE_WRITE
);
402 opts
= &full_opts
->u
.copy_before_write
;
404 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_of_bds
,
405 BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
,
411 s
->target
= bdrv_open_child(NULL
, options
, "target", bs
, &child_of_bds
,
412 BDRV_CHILD_DATA
, false, errp
);
417 if (opts
->has_bitmap
) {
418 bitmap
= block_dirty_bitmap_lookup(opts
->bitmap
->node
,
419 opts
->bitmap
->name
, NULL
, errp
);
424 s
->on_cbw_error
= opts
->has_on_cbw_error
? opts
->on_cbw_error
:
425 ON_CBW_ERROR_BREAK_GUEST_WRITE
;
427 bs
->total_sectors
= bs
->file
->bs
->total_sectors
;
428 bs
->supported_write_flags
= BDRV_REQ_WRITE_UNCHANGED
|
429 (BDRV_REQ_FUA
& bs
->file
->bs
->supported_write_flags
);
430 bs
->supported_zero_flags
= BDRV_REQ_WRITE_UNCHANGED
|
431 ((BDRV_REQ_FUA
| BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
) &
432 bs
->file
->bs
->supported_zero_flags
);
434 s
->bcs
= block_copy_state_new(bs
->file
, s
->target
, bitmap
, errp
);
436 error_prepend(errp
, "Cannot create block-copy-state: ");
440 cluster_size
= block_copy_cluster_size(s
->bcs
);
442 s
->done_bitmap
= bdrv_create_dirty_bitmap(bs
, cluster_size
, NULL
, errp
);
443 if (!s
->done_bitmap
) {
446 bdrv_disable_dirty_bitmap(s
->done_bitmap
);
448 /* s->access_bitmap starts equal to bcs bitmap */
449 s
->access_bitmap
= bdrv_create_dirty_bitmap(bs
, cluster_size
, NULL
, errp
);
450 if (!s
->access_bitmap
) {
453 bdrv_disable_dirty_bitmap(s
->access_bitmap
);
454 bdrv_dirty_bitmap_merge_internal(s
->access_bitmap
,
455 block_copy_dirty_bitmap(s
->bcs
), NULL
,
458 qemu_co_mutex_init(&s
->lock
);
459 QLIST_INIT(&s
->frozen_read_reqs
);
464 static void cbw_close(BlockDriverState
*bs
)
466 BDRVCopyBeforeWriteState
*s
= bs
->opaque
;
468 bdrv_release_dirty_bitmap(s
->access_bitmap
);
469 bdrv_release_dirty_bitmap(s
->done_bitmap
);
471 block_copy_state_free(s
->bcs
);
475 BlockDriver bdrv_cbw_filter
= {
476 .format_name
= "copy-before-write",
477 .instance_size
= sizeof(BDRVCopyBeforeWriteState
),
479 .bdrv_open
= cbw_open
,
480 .bdrv_close
= cbw_close
,
482 .bdrv_co_preadv
= cbw_co_preadv
,
483 .bdrv_co_pwritev
= cbw_co_pwritev
,
484 .bdrv_co_pwrite_zeroes
= cbw_co_pwrite_zeroes
,
485 .bdrv_co_pdiscard
= cbw_co_pdiscard
,
486 .bdrv_co_flush
= cbw_co_flush
,
488 .bdrv_co_preadv_snapshot
= cbw_co_preadv_snapshot
,
489 .bdrv_co_pdiscard_snapshot
= cbw_co_pdiscard_snapshot
,
490 .bdrv_co_snapshot_block_status
= cbw_co_snapshot_block_status
,
492 .bdrv_refresh_filename
= cbw_refresh_filename
,
494 .bdrv_child_perm
= cbw_child_perm
,
499 BlockDriverState
*bdrv_cbw_append(BlockDriverState
*source
,
500 BlockDriverState
*target
,
501 const char *filter_node_name
,
502 BlockCopyState
**bcs
,
506 BDRVCopyBeforeWriteState
*state
;
507 BlockDriverState
*top
;
510 assert(source
->total_sectors
== target
->total_sectors
);
514 qdict_put_str(opts
, "driver", "copy-before-write");
515 if (filter_node_name
) {
516 qdict_put_str(opts
, "node-name", filter_node_name
);
518 qdict_put_str(opts
, "file", bdrv_get_node_name(source
));
519 qdict_put_str(opts
, "target", bdrv_get_node_name(target
));
521 top
= bdrv_insert_node(source
, opts
, BDRV_O_RDWR
, errp
);
532 void bdrv_cbw_drop(BlockDriverState
*bs
)
535 bdrv_drop_filter(bs
, &error_abort
);
539 static void cbw_init(void)
541 bdrv_register(&bdrv_cbw_filter
);
544 block_init(cbw_init
);