2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "block/block_int.h"
27 #include "sysemu/block-backend.h"
28 #include "qemu/module.h"
30 #include "block/qcow2.h"
31 #include "qemu/error-report.h"
32 #include "qapi/error.h"
33 #include "qapi/qapi-events-block-core.h"
34 #include "qapi/qmp/qdict.h"
35 #include "qapi/qmp/qstring.h"
37 #include "qemu/option_int.h"
38 #include "qemu/cutils.h"
39 #include "qemu/bswap.h"
40 #include "qapi/opts-visitor.h"
41 #include "block/crypto.h"
44 Differences with QCOW:
46 - Support for multiple incremental snapshots.
47 - Memory management by reference counts.
48 - Clusters which have a reference count of one have the bit
49 QCOW_OFLAG_COPIED to optimize write performance.
50 - Size of compressed clusters is stored in sectors to reduce bit usage
51 in the cluster offsets.
52 - Support for storing additional data (such as the VM state) in the
54 - If a backing store is used, the cluster size is not constrained
55 (could be backported to QCOW).
56 - L2 tables have always a size of one cluster.
63 } QEMU_PACKED QCowExtension
;
65 #define QCOW2_EXT_MAGIC_END 0
66 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
67 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
68 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
69 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
71 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
73 const QCowHeader
*cow_header
= (const void *)buf
;
75 if (buf_size
>= sizeof(QCowHeader
) &&
76 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
77 be32_to_cpu(cow_header
->version
) >= 2)
84 static ssize_t
qcow2_crypto_hdr_read_func(QCryptoBlock
*block
, size_t offset
,
85 uint8_t *buf
, size_t buflen
,
86 void *opaque
, Error
**errp
)
88 BlockDriverState
*bs
= opaque
;
89 BDRVQcow2State
*s
= bs
->opaque
;
92 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
93 error_setg(errp
, "Request for data outside of extension header");
97 ret
= bdrv_pread(bs
->file
,
98 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
100 error_setg_errno(errp
, -ret
, "Could not read encryption header");
107 static ssize_t
qcow2_crypto_hdr_init_func(QCryptoBlock
*block
, size_t headerlen
,
108 void *opaque
, Error
**errp
)
110 BlockDriverState
*bs
= opaque
;
111 BDRVQcow2State
*s
= bs
->opaque
;
115 ret
= qcow2_alloc_clusters(bs
, headerlen
);
117 error_setg_errno(errp
, -ret
,
118 "Cannot allocate cluster for LUKS header size %zu",
123 s
->crypto_header
.length
= headerlen
;
124 s
->crypto_header
.offset
= ret
;
126 /* Zero fill remaining space in cluster so it has predictable
127 * content in case of future spec changes */
128 clusterlen
= size_to_clusters(s
, headerlen
) * s
->cluster_size
;
129 assert(qcow2_pre_write_overlap_check(bs
, 0, ret
, clusterlen
) == 0);
130 ret
= bdrv_pwrite_zeroes(bs
->file
,
132 clusterlen
- headerlen
, 0);
134 error_setg_errno(errp
, -ret
, "Could not zero fill encryption header");
142 static ssize_t
qcow2_crypto_hdr_write_func(QCryptoBlock
*block
, size_t offset
,
143 const uint8_t *buf
, size_t buflen
,
144 void *opaque
, Error
**errp
)
146 BlockDriverState
*bs
= opaque
;
147 BDRVQcow2State
*s
= bs
->opaque
;
150 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
151 error_setg(errp
, "Request for data outside of extension header");
155 ret
= bdrv_pwrite(bs
->file
,
156 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
158 error_setg_errno(errp
, -ret
, "Could not read encryption header");
166 * read qcow2 extension and fill bs
167 * start reading from start_offset
168 * finish reading upon magic of value 0 or when end_offset reached
169 * unknown magic is skipped (future extension this version knows nothing about)
170 * return 0 upon success, non-0 otherwise
172 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
173 uint64_t end_offset
, void **p_feature_table
,
174 int flags
, bool *need_update_header
,
177 BDRVQcow2State
*s
= bs
->opaque
;
181 Qcow2BitmapHeaderExt bitmaps_ext
;
183 if (need_update_header
!= NULL
) {
184 *need_update_header
= false;
188 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
190 offset
= start_offset
;
191 while (offset
< end_offset
) {
195 if (offset
> s
->cluster_size
)
196 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
198 printf("attempting to read extended header in offset %lu\n", offset
);
201 ret
= bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
));
203 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
204 "pread fail from offset %" PRIu64
, offset
);
207 be32_to_cpus(&ext
.magic
);
208 be32_to_cpus(&ext
.len
);
209 offset
+= sizeof(ext
);
211 printf("ext.magic = 0x%x\n", ext
.magic
);
213 if (offset
> end_offset
|| ext
.len
> end_offset
- offset
) {
214 error_setg(errp
, "Header extension too large");
219 case QCOW2_EXT_MAGIC_END
:
222 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
223 if (ext
.len
>= sizeof(bs
->backing_format
)) {
224 error_setg(errp
, "ERROR: ext_backing_format: len=%" PRIu32
225 " too large (>=%zu)", ext
.len
,
226 sizeof(bs
->backing_format
));
229 ret
= bdrv_pread(bs
->file
, offset
, bs
->backing_format
, ext
.len
);
231 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
232 "Could not read format name");
235 bs
->backing_format
[ext
.len
] = '\0';
236 s
->image_backing_format
= g_strdup(bs
->backing_format
);
238 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
242 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
243 if (p_feature_table
!= NULL
) {
244 void* feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
245 ret
= bdrv_pread(bs
->file
, offset
, feature_table
, ext
.len
);
247 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
248 "Could not read table");
252 *p_feature_table
= feature_table
;
256 case QCOW2_EXT_MAGIC_CRYPTO_HEADER
: {
257 unsigned int cflags
= 0;
258 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
259 error_setg(errp
, "CRYPTO header extension only "
260 "expected with LUKS encryption method");
263 if (ext
.len
!= sizeof(Qcow2CryptoHeaderExtension
)) {
264 error_setg(errp
, "CRYPTO header extension size %u, "
265 "but expected size %zu", ext
.len
,
266 sizeof(Qcow2CryptoHeaderExtension
));
270 ret
= bdrv_pread(bs
->file
, offset
, &s
->crypto_header
, ext
.len
);
272 error_setg_errno(errp
, -ret
,
273 "Unable to read CRYPTO header extension");
276 be64_to_cpus(&s
->crypto_header
.offset
);
277 be64_to_cpus(&s
->crypto_header
.length
);
279 if ((s
->crypto_header
.offset
% s
->cluster_size
) != 0) {
280 error_setg(errp
, "Encryption header offset '%" PRIu64
"' is "
281 "not a multiple of cluster size '%u'",
282 s
->crypto_header
.offset
, s
->cluster_size
);
286 if (flags
& BDRV_O_NO_IO
) {
287 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
289 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
290 qcow2_crypto_hdr_read_func
,
297 case QCOW2_EXT_MAGIC_BITMAPS
:
298 if (ext
.len
!= sizeof(bitmaps_ext
)) {
299 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
300 "Invalid extension length");
304 if (!(s
->autoclear_features
& QCOW2_AUTOCLEAR_BITMAPS
)) {
305 if (s
->qcow_version
< 3) {
306 /* Let's be a bit more specific */
307 warn_report("This qcow2 v2 image contains bitmaps, but "
308 "they may have been modified by a program "
309 "without persistent bitmap support; so now "
310 "they must all be considered inconsistent");
312 warn_report("a program lacking bitmap support "
313 "modified this file, so all bitmaps are now "
314 "considered inconsistent");
316 error_printf("Some clusters may be leaked, "
317 "run 'qemu-img check -r' on the image "
319 if (need_update_header
!= NULL
) {
320 /* Updating is needed to drop invalid bitmap extension. */
321 *need_update_header
= true;
326 ret
= bdrv_pread(bs
->file
, offset
, &bitmaps_ext
, ext
.len
);
328 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
329 "Could not read ext header");
333 if (bitmaps_ext
.reserved32
!= 0) {
334 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
335 "Reserved field is not zero");
339 be32_to_cpus(&bitmaps_ext
.nb_bitmaps
);
340 be64_to_cpus(&bitmaps_ext
.bitmap_directory_size
);
341 be64_to_cpus(&bitmaps_ext
.bitmap_directory_offset
);
343 if (bitmaps_ext
.nb_bitmaps
> QCOW2_MAX_BITMAPS
) {
345 "bitmaps_ext: Image has %" PRIu32
" bitmaps, "
346 "exceeding the QEMU supported maximum of %d",
347 bitmaps_ext
.nb_bitmaps
, QCOW2_MAX_BITMAPS
);
351 if (bitmaps_ext
.nb_bitmaps
== 0) {
352 error_setg(errp
, "found bitmaps extension with zero bitmaps");
356 if (bitmaps_ext
.bitmap_directory_offset
& (s
->cluster_size
- 1)) {
357 error_setg(errp
, "bitmaps_ext: "
358 "invalid bitmap directory offset");
362 if (bitmaps_ext
.bitmap_directory_size
>
363 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
) {
364 error_setg(errp
, "bitmaps_ext: "
365 "bitmap directory size (%" PRIu64
") exceeds "
366 "the maximum supported size (%d)",
367 bitmaps_ext
.bitmap_directory_size
,
368 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
);
372 s
->nb_bitmaps
= bitmaps_ext
.nb_bitmaps
;
373 s
->bitmap_directory_offset
=
374 bitmaps_ext
.bitmap_directory_offset
;
375 s
->bitmap_directory_size
=
376 bitmaps_ext
.bitmap_directory_size
;
379 printf("Qcow2: Got bitmaps extension: "
380 "offset=%" PRIu64
" nb_bitmaps=%" PRIu32
"\n",
381 s
->bitmap_directory_offset
, s
->nb_bitmaps
);
386 /* unknown magic - save it in case we need to rewrite the header */
387 /* If you add a new feature, make sure to also update the fast
388 * path of qcow2_make_empty() to deal with it. */
390 Qcow2UnknownHeaderExtension
*uext
;
392 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
393 uext
->magic
= ext
.magic
;
395 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
397 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
399 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
400 "Could not read data");
407 offset
+= ((ext
.len
+ 7) & ~7);
413 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
415 BDRVQcow2State
*s
= bs
->opaque
;
416 Qcow2UnknownHeaderExtension
*uext
, *next
;
418 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
419 QLIST_REMOVE(uext
, next
);
424 static void report_unsupported_feature(Error
**errp
, Qcow2Feature
*table
,
427 char *features
= g_strdup("");
430 while (table
&& table
->name
[0] != '\0') {
431 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
432 if (mask
& (1ULL << table
->bit
)) {
434 features
= g_strdup_printf("%s%s%.46s", old
, *old
? ", " : "",
437 mask
&= ~(1ULL << table
->bit
);
445 features
= g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64
,
446 old
, *old
? ", " : "", mask
);
450 error_setg(errp
, "Unsupported qcow2 feature(s): %s", features
);
455 * Sets the dirty bit and flushes afterwards if necessary.
457 * The incompatible_features bit is only set if the image file header was
458 * updated successfully. Therefore it is not required to check the return
459 * value of this function.
461 int qcow2_mark_dirty(BlockDriverState
*bs
)
463 BDRVQcow2State
*s
= bs
->opaque
;
467 assert(s
->qcow_version
>= 3);
469 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
470 return 0; /* already dirty */
473 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
474 ret
= bdrv_pwrite(bs
->file
, offsetof(QCowHeader
, incompatible_features
),
479 ret
= bdrv_flush(bs
->file
->bs
);
484 /* Only treat image as dirty if the header was updated successfully */
485 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
490 * Clears the dirty bit and flushes before if necessary. Only call this
491 * function when there are no pending requests, it does not guard against
492 * concurrent requests dirtying the image.
494 static int qcow2_mark_clean(BlockDriverState
*bs
)
496 BDRVQcow2State
*s
= bs
->opaque
;
498 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
501 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
503 ret
= bdrv_flush(bs
);
508 return qcow2_update_header(bs
);
514 * Marks the image as corrupt.
516 int qcow2_mark_corrupt(BlockDriverState
*bs
)
518 BDRVQcow2State
*s
= bs
->opaque
;
520 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
521 return qcow2_update_header(bs
);
525 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
526 * before if necessary.
528 int qcow2_mark_consistent(BlockDriverState
*bs
)
530 BDRVQcow2State
*s
= bs
->opaque
;
532 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
533 int ret
= bdrv_flush(bs
);
538 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
539 return qcow2_update_header(bs
);
544 static int qcow2_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
547 int ret
= qcow2_check_refcounts(bs
, result
, fix
);
552 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
553 ret
= qcow2_mark_clean(bs
);
557 return qcow2_mark_consistent(bs
);
562 static int validate_table_offset(BlockDriverState
*bs
, uint64_t offset
,
563 uint64_t entries
, size_t entry_len
)
565 BDRVQcow2State
*s
= bs
->opaque
;
568 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
569 * because values will be passed to qemu functions taking int64_t. */
570 if (entries
> INT64_MAX
/ entry_len
) {
574 size
= entries
* entry_len
;
576 if (INT64_MAX
- size
< offset
) {
580 /* Tables must be cluster aligned */
581 if (offset_into_cluster(s
, offset
) != 0) {
588 static QemuOptsList qcow2_runtime_opts
= {
590 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
593 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
594 .type
= QEMU_OPT_BOOL
,
595 .help
= "Postpone refcount updates",
598 .name
= QCOW2_OPT_DISCARD_REQUEST
,
599 .type
= QEMU_OPT_BOOL
,
600 .help
= "Pass guest discard requests to the layer below",
603 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
604 .type
= QEMU_OPT_BOOL
,
605 .help
= "Generate discard requests when snapshot related space "
609 .name
= QCOW2_OPT_DISCARD_OTHER
,
610 .type
= QEMU_OPT_BOOL
,
611 .help
= "Generate discard requests when other clusters are freed",
614 .name
= QCOW2_OPT_OVERLAP
,
615 .type
= QEMU_OPT_STRING
,
616 .help
= "Selects which overlap checks to perform from a range of "
617 "templates (none, constant, cached, all)",
620 .name
= QCOW2_OPT_OVERLAP_TEMPLATE
,
621 .type
= QEMU_OPT_STRING
,
622 .help
= "Selects which overlap checks to perform from a range of "
623 "templates (none, constant, cached, all)",
626 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
627 .type
= QEMU_OPT_BOOL
,
628 .help
= "Check for unintended writes into the main qcow2 header",
631 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
632 .type
= QEMU_OPT_BOOL
,
633 .help
= "Check for unintended writes into the active L1 table",
636 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
637 .type
= QEMU_OPT_BOOL
,
638 .help
= "Check for unintended writes into an active L2 table",
641 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
642 .type
= QEMU_OPT_BOOL
,
643 .help
= "Check for unintended writes into the refcount table",
646 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
647 .type
= QEMU_OPT_BOOL
,
648 .help
= "Check for unintended writes into a refcount block",
651 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
652 .type
= QEMU_OPT_BOOL
,
653 .help
= "Check for unintended writes into the snapshot table",
656 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
657 .type
= QEMU_OPT_BOOL
,
658 .help
= "Check for unintended writes into an inactive L1 table",
661 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
662 .type
= QEMU_OPT_BOOL
,
663 .help
= "Check for unintended writes into an inactive L2 table",
666 .name
= QCOW2_OPT_CACHE_SIZE
,
667 .type
= QEMU_OPT_SIZE
,
668 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
672 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
673 .type
= QEMU_OPT_SIZE
,
674 .help
= "Maximum L2 table cache size",
677 .name
= QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
678 .type
= QEMU_OPT_SIZE
,
679 .help
= "Size of each entry in the L2 cache",
682 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
683 .type
= QEMU_OPT_SIZE
,
684 .help
= "Maximum refcount block cache size",
687 .name
= QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
688 .type
= QEMU_OPT_NUMBER
,
689 .help
= "Clean unused cache entries after this time (in seconds)",
691 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
692 "ID of secret providing qcow2 AES key or LUKS passphrase"),
693 { /* end of list */ }
697 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
698 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
699 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
700 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
701 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
702 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
703 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
704 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
705 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
708 static void cache_clean_timer_cb(void *opaque
)
710 BlockDriverState
*bs
= opaque
;
711 BDRVQcow2State
*s
= bs
->opaque
;
712 qcow2_cache_clean_unused(s
->l2_table_cache
);
713 qcow2_cache_clean_unused(s
->refcount_block_cache
);
714 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
715 (int64_t) s
->cache_clean_interval
* 1000);
718 static void cache_clean_timer_init(BlockDriverState
*bs
, AioContext
*context
)
720 BDRVQcow2State
*s
= bs
->opaque
;
721 if (s
->cache_clean_interval
> 0) {
722 s
->cache_clean_timer
= aio_timer_new(context
, QEMU_CLOCK_VIRTUAL
,
723 SCALE_MS
, cache_clean_timer_cb
,
725 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
726 (int64_t) s
->cache_clean_interval
* 1000);
730 static void cache_clean_timer_del(BlockDriverState
*bs
)
732 BDRVQcow2State
*s
= bs
->opaque
;
733 if (s
->cache_clean_timer
) {
734 timer_del(s
->cache_clean_timer
);
735 timer_free(s
->cache_clean_timer
);
736 s
->cache_clean_timer
= NULL
;
740 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
742 cache_clean_timer_del(bs
);
745 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
746 AioContext
*new_context
)
748 cache_clean_timer_init(bs
, new_context
);
751 static void read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
752 uint64_t *l2_cache_size
,
753 uint64_t *l2_cache_entry_size
,
754 uint64_t *refcount_cache_size
, Error
**errp
)
756 BDRVQcow2State
*s
= bs
->opaque
;
757 uint64_t combined_cache_size
;
758 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
760 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
761 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
762 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
764 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
765 *l2_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
, 0);
766 *refcount_cache_size
= qemu_opt_get_size(opts
,
767 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
769 *l2_cache_entry_size
= qemu_opt_get_size(
770 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
772 if (combined_cache_size_set
) {
773 if (l2_cache_size_set
&& refcount_cache_size_set
) {
774 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
775 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
778 } else if (*l2_cache_size
> combined_cache_size
) {
779 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
780 QCOW2_OPT_CACHE_SIZE
);
782 } else if (*refcount_cache_size
> combined_cache_size
) {
783 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
784 QCOW2_OPT_CACHE_SIZE
);
788 if (l2_cache_size_set
) {
789 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
790 } else if (refcount_cache_size_set
) {
791 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
793 *refcount_cache_size
= combined_cache_size
794 / (DEFAULT_L2_REFCOUNT_SIZE_RATIO
+ 1);
795 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
798 if (!l2_cache_size_set
&& !refcount_cache_size_set
) {
799 *l2_cache_size
= MAX(DEFAULT_L2_CACHE_BYTE_SIZE
,
800 (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
802 *refcount_cache_size
= *l2_cache_size
803 / DEFAULT_L2_REFCOUNT_SIZE_RATIO
;
804 } else if (!l2_cache_size_set
) {
805 *l2_cache_size
= *refcount_cache_size
806 * DEFAULT_L2_REFCOUNT_SIZE_RATIO
;
807 } else if (!refcount_cache_size_set
) {
808 *refcount_cache_size
= *l2_cache_size
809 / DEFAULT_L2_REFCOUNT_SIZE_RATIO
;
813 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
814 *l2_cache_entry_size
> s
->cluster_size
||
815 !is_power_of_2(*l2_cache_entry_size
)) {
816 error_setg(errp
, "L2 cache entry size must be a power of two "
817 "between %d and the cluster size (%d)",
818 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
823 typedef struct Qcow2ReopenState
{
824 Qcow2Cache
*l2_table_cache
;
825 Qcow2Cache
*refcount_block_cache
;
826 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
827 bool use_lazy_refcounts
;
829 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
830 uint64_t cache_clean_interval
;
831 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
834 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
836 QDict
*options
, int flags
,
839 BDRVQcow2State
*s
= bs
->opaque
;
840 QemuOpts
*opts
= NULL
;
841 const char *opt_overlap_check
, *opt_overlap_check_template
;
842 int overlap_check_template
= 0;
843 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
845 const char *encryptfmt
;
846 QDict
*encryptopts
= NULL
;
847 Error
*local_err
= NULL
;
850 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
851 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
853 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
854 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
856 error_propagate(errp
, local_err
);
861 /* get L2 table/refcount block cache size from command line options */
862 read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
863 &refcount_cache_size
, &local_err
);
865 error_propagate(errp
, local_err
);
870 l2_cache_size
/= l2_cache_entry_size
;
871 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
872 l2_cache_size
= MIN_L2_CACHE_SIZE
;
874 if (l2_cache_size
> INT_MAX
) {
875 error_setg(errp
, "L2 cache size too big");
880 refcount_cache_size
/= s
->cluster_size
;
881 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
882 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
884 if (refcount_cache_size
> INT_MAX
) {
885 error_setg(errp
, "Refcount cache size too big");
890 /* alloc new L2 table/refcount block cache, flush old one */
891 if (s
->l2_table_cache
) {
892 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
894 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
899 if (s
->refcount_block_cache
) {
900 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
902 error_setg_errno(errp
, -ret
,
903 "Failed to flush the refcount block cache");
908 r
->l2_slice_size
= l2_cache_entry_size
/ sizeof(uint64_t);
909 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
910 l2_cache_entry_size
);
911 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
913 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
914 error_setg(errp
, "Could not allocate metadata caches");
919 /* New interval for cache cleanup timer */
920 r
->cache_clean_interval
=
921 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
922 s
->cache_clean_interval
);
924 if (r
->cache_clean_interval
!= 0) {
925 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
926 " not supported on this host");
931 if (r
->cache_clean_interval
> UINT_MAX
) {
932 error_setg(errp
, "Cache clean interval too big");
937 /* lazy-refcounts; flush if going from enabled to disabled */
938 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
939 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
940 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
941 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
942 "qemu 1.1 compatibility level");
947 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
948 ret
= qcow2_mark_clean(bs
);
950 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
955 /* Overlap check options */
956 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
957 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
958 if (opt_overlap_check_template
&& opt_overlap_check
&&
959 strcmp(opt_overlap_check_template
, opt_overlap_check
))
961 error_setg(errp
, "Conflicting values for qcow2 options '"
962 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
963 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
967 if (!opt_overlap_check
) {
968 opt_overlap_check
= opt_overlap_check_template
?: "cached";
971 if (!strcmp(opt_overlap_check
, "none")) {
972 overlap_check_template
= 0;
973 } else if (!strcmp(opt_overlap_check
, "constant")) {
974 overlap_check_template
= QCOW2_OL_CONSTANT
;
975 } else if (!strcmp(opt_overlap_check
, "cached")) {
976 overlap_check_template
= QCOW2_OL_CACHED
;
977 } else if (!strcmp(opt_overlap_check
, "all")) {
978 overlap_check_template
= QCOW2_OL_ALL
;
980 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
981 "'overlap-check'. Allowed are any of the following: "
982 "none, constant, cached, all", opt_overlap_check
);
987 r
->overlap_check
= 0;
988 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
989 /* overlap-check defines a template bitmask, but every flag may be
990 * overwritten through the associated boolean option */
992 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
993 overlap_check_template
& (1 << i
)) << i
;
996 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
997 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
998 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
999 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1000 flags
& BDRV_O_UNMAP
);
1001 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1002 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1003 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1004 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1006 switch (s
->crypt_method_header
) {
1007 case QCOW_CRYPT_NONE
:
1009 error_setg(errp
, "No encryption in image header, but options "
1010 "specified format '%s'", encryptfmt
);
1016 case QCOW_CRYPT_AES
:
1017 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1019 "Header reported 'aes' encryption format but "
1020 "options specify '%s'", encryptfmt
);
1024 qdict_del(encryptopts
, "format");
1025 r
->crypto_opts
= block_crypto_open_opts_init(
1026 Q_CRYPTO_BLOCK_FORMAT_QCOW
, encryptopts
, errp
);
1029 case QCOW_CRYPT_LUKS
:
1030 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1032 "Header reported 'luks' encryption format but "
1033 "options specify '%s'", encryptfmt
);
1037 qdict_del(encryptopts
, "format");
1038 r
->crypto_opts
= block_crypto_open_opts_init(
1039 Q_CRYPTO_BLOCK_FORMAT_LUKS
, encryptopts
, errp
);
1043 error_setg(errp
, "Unsupported encryption method %d",
1044 s
->crypt_method_header
);
1047 if (s
->crypt_method_header
!= QCOW_CRYPT_NONE
&& !r
->crypto_opts
) {
1054 QDECREF(encryptopts
);
1055 qemu_opts_del(opts
);
1060 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1061 Qcow2ReopenState
*r
)
1063 BDRVQcow2State
*s
= bs
->opaque
;
1066 if (s
->l2_table_cache
) {
1067 qcow2_cache_destroy(s
->l2_table_cache
);
1069 if (s
->refcount_block_cache
) {
1070 qcow2_cache_destroy(s
->refcount_block_cache
);
1072 s
->l2_table_cache
= r
->l2_table_cache
;
1073 s
->refcount_block_cache
= r
->refcount_block_cache
;
1074 s
->l2_slice_size
= r
->l2_slice_size
;
1076 s
->overlap_check
= r
->overlap_check
;
1077 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1079 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1080 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1083 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1084 cache_clean_timer_del(bs
);
1085 s
->cache_clean_interval
= r
->cache_clean_interval
;
1086 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1089 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1090 s
->crypto_opts
= r
->crypto_opts
;
1093 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1094 Qcow2ReopenState
*r
)
1096 if (r
->l2_table_cache
) {
1097 qcow2_cache_destroy(r
->l2_table_cache
);
1099 if (r
->refcount_block_cache
) {
1100 qcow2_cache_destroy(r
->refcount_block_cache
);
1102 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1105 static int qcow2_update_options(BlockDriverState
*bs
, QDict
*options
,
1106 int flags
, Error
**errp
)
1108 Qcow2ReopenState r
= {};
1111 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1113 qcow2_update_options_commit(bs
, &r
);
1115 qcow2_update_options_abort(bs
, &r
);
1121 static int qcow2_do_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1124 BDRVQcow2State
*s
= bs
->opaque
;
1125 unsigned int len
, i
;
1128 Error
*local_err
= NULL
;
1130 uint64_t l1_vm_state_index
;
1131 bool update_header
= false;
1133 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
1135 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1138 be32_to_cpus(&header
.magic
);
1139 be32_to_cpus(&header
.version
);
1140 be64_to_cpus(&header
.backing_file_offset
);
1141 be32_to_cpus(&header
.backing_file_size
);
1142 be64_to_cpus(&header
.size
);
1143 be32_to_cpus(&header
.cluster_bits
);
1144 be32_to_cpus(&header
.crypt_method
);
1145 be64_to_cpus(&header
.l1_table_offset
);
1146 be32_to_cpus(&header
.l1_size
);
1147 be64_to_cpus(&header
.refcount_table_offset
);
1148 be32_to_cpus(&header
.refcount_table_clusters
);
1149 be64_to_cpus(&header
.snapshots_offset
);
1150 be32_to_cpus(&header
.nb_snapshots
);
1152 if (header
.magic
!= QCOW_MAGIC
) {
1153 error_setg(errp
, "Image is not in qcow2 format");
1157 if (header
.version
< 2 || header
.version
> 3) {
1158 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1163 s
->qcow_version
= header
.version
;
1165 /* Initialise cluster size */
1166 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1167 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1168 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1169 header
.cluster_bits
);
1174 s
->cluster_bits
= header
.cluster_bits
;
1175 s
->cluster_size
= 1 << s
->cluster_bits
;
1176 s
->cluster_sectors
= 1 << (s
->cluster_bits
- BDRV_SECTOR_BITS
);
1178 /* Initialise version 3 header fields */
1179 if (header
.version
== 2) {
1180 header
.incompatible_features
= 0;
1181 header
.compatible_features
= 0;
1182 header
.autoclear_features
= 0;
1183 header
.refcount_order
= 4;
1184 header
.header_length
= 72;
1186 be64_to_cpus(&header
.incompatible_features
);
1187 be64_to_cpus(&header
.compatible_features
);
1188 be64_to_cpus(&header
.autoclear_features
);
1189 be32_to_cpus(&header
.refcount_order
);
1190 be32_to_cpus(&header
.header_length
);
1192 if (header
.header_length
< 104) {
1193 error_setg(errp
, "qcow2 header too short");
1199 if (header
.header_length
> s
->cluster_size
) {
1200 error_setg(errp
, "qcow2 header exceeds cluster size");
1205 if (header
.header_length
> sizeof(header
)) {
1206 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1207 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1208 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
1209 s
->unknown_header_fields_size
);
1211 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1217 if (header
.backing_file_offset
> s
->cluster_size
) {
1218 error_setg(errp
, "Invalid backing file offset");
1223 if (header
.backing_file_offset
) {
1224 ext_end
= header
.backing_file_offset
;
1226 ext_end
= 1 << header
.cluster_bits
;
1229 /* Handle feature bits */
1230 s
->incompatible_features
= header
.incompatible_features
;
1231 s
->compatible_features
= header
.compatible_features
;
1232 s
->autoclear_features
= header
.autoclear_features
;
1234 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1235 void *feature_table
= NULL
;
1236 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1237 &feature_table
, flags
, NULL
, NULL
);
1238 report_unsupported_feature(errp
, feature_table
,
1239 s
->incompatible_features
&
1240 ~QCOW2_INCOMPAT_MASK
);
1242 g_free(feature_table
);
1246 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1247 /* Corrupt images may not be written to unless they are being repaired
1249 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1250 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1257 /* Check support for various header values */
1258 if (header
.refcount_order
> 6) {
1259 error_setg(errp
, "Reference count entry width too large; may not "
1264 s
->refcount_order
= header
.refcount_order
;
1265 s
->refcount_bits
= 1 << s
->refcount_order
;
1266 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1267 s
->refcount_max
+= s
->refcount_max
- 1;
1269 s
->crypt_method_header
= header
.crypt_method
;
1270 if (s
->crypt_method_header
) {
1271 if (bdrv_uses_whitelist() &&
1272 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1274 "Use of AES-CBC encrypted qcow2 images is no longer "
1275 "supported in system emulators");
1276 error_append_hint(errp
,
1277 "You can use 'qemu-img convert' to convert your "
1278 "image to an alternative supported format, such "
1279 "as unencrypted qcow2, or raw with the LUKS "
1280 "format instead.\n");
1285 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1286 s
->crypt_physical_offset
= false;
1288 /* Assuming LUKS and any future crypt methods we
1289 * add will all use physical offsets, due to the
1290 * fact that the alternative is insecure... */
1291 s
->crypt_physical_offset
= true;
1294 bs
->encrypted
= true;
1297 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
1298 s
->l2_size
= 1 << s
->l2_bits
;
1299 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1300 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1301 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1302 bs
->total_sectors
= header
.size
/ 512;
1303 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1304 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1305 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1307 s
->refcount_table_offset
= header
.refcount_table_offset
;
1308 s
->refcount_table_size
=
1309 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1311 if (header
.refcount_table_clusters
> qcow2_max_refcount_clusters(s
)) {
1312 error_setg(errp
, "Reference count table too large");
1317 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1318 error_setg(errp
, "Image does not contain a reference count table");
1323 ret
= validate_table_offset(bs
, s
->refcount_table_offset
,
1324 s
->refcount_table_size
, sizeof(uint64_t));
1326 error_setg(errp
, "Invalid reference count table offset");
1330 /* Snapshot table offset/length */
1331 if (header
.nb_snapshots
> QCOW_MAX_SNAPSHOTS
) {
1332 error_setg(errp
, "Too many snapshots");
1337 ret
= validate_table_offset(bs
, header
.snapshots_offset
,
1338 header
.nb_snapshots
,
1339 sizeof(QCowSnapshotHeader
));
1341 error_setg(errp
, "Invalid snapshot table offset");
1345 /* read the level 1 table */
1346 if (header
.l1_size
> QCOW_MAX_L1_SIZE
/ sizeof(uint64_t)) {
1347 error_setg(errp
, "Active L1 table too large");
1351 s
->l1_size
= header
.l1_size
;
1353 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1354 if (l1_vm_state_index
> INT_MAX
) {
1355 error_setg(errp
, "Image is too big");
1359 s
->l1_vm_state_index
= l1_vm_state_index
;
1361 /* the L1 table must contain at least enough entries to put
1362 header.size bytes */
1363 if (s
->l1_size
< s
->l1_vm_state_index
) {
1364 error_setg(errp
, "L1 table is too small");
1369 ret
= validate_table_offset(bs
, header
.l1_table_offset
,
1370 header
.l1_size
, sizeof(uint64_t));
1372 error_setg(errp
, "Invalid L1 table offset");
1375 s
->l1_table_offset
= header
.l1_table_offset
;
1378 if (s
->l1_size
> 0) {
1379 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
,
1380 ROUND_UP(s
->l1_size
* sizeof(uint64_t), 512));
1381 if (s
->l1_table
== NULL
) {
1382 error_setg(errp
, "Could not allocate L1 table");
1386 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1387 s
->l1_size
* sizeof(uint64_t));
1389 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1392 for(i
= 0;i
< s
->l1_size
; i
++) {
1393 be64_to_cpus(&s
->l1_table
[i
]);
1397 /* Parse driver-specific options */
1398 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1403 s
->cluster_cache_offset
= -1;
1406 ret
= qcow2_refcount_init(bs
);
1408 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1412 QLIST_INIT(&s
->cluster_allocs
);
1413 QTAILQ_INIT(&s
->discards
);
1415 /* read qcow2 extensions */
1416 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1417 flags
, &update_header
, &local_err
)) {
1418 error_propagate(errp
, local_err
);
1423 /* qcow2_read_extension may have set up the crypto context
1424 * if the crypt method needs a header region, some methods
1425 * don't need header extensions, so must check here
1427 if (s
->crypt_method_header
&& !s
->crypto
) {
1428 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1429 unsigned int cflags
= 0;
1430 if (flags
& BDRV_O_NO_IO
) {
1431 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1433 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1434 NULL
, NULL
, cflags
, errp
);
1439 } else if (!(flags
& BDRV_O_NO_IO
)) {
1440 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1441 s
->crypt_method_header
);
1447 /* read the backing file name */
1448 if (header
.backing_file_offset
!= 0) {
1449 len
= header
.backing_file_size
;
1450 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1451 len
>= sizeof(bs
->backing_file
)) {
1452 error_setg(errp
, "Backing file name too long");
1456 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
1457 bs
->backing_file
, len
);
1459 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1462 bs
->backing_file
[len
] = '\0';
1463 s
->image_backing_file
= g_strdup(bs
->backing_file
);
1466 /* Internal snapshots */
1467 s
->snapshots_offset
= header
.snapshots_offset
;
1468 s
->nb_snapshots
= header
.nb_snapshots
;
1470 ret
= qcow2_read_snapshots(bs
);
1472 error_setg_errno(errp
, -ret
, "Could not read snapshots");
1476 /* Clear unknown autoclear feature bits */
1477 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1479 update_header
&& !bs
->read_only
&& !(flags
& BDRV_O_INACTIVE
);
1480 if (update_header
) {
1481 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1484 if (qcow2_load_dirty_bitmaps(bs
, &local_err
)) {
1485 update_header
= false;
1487 if (local_err
!= NULL
) {
1488 error_propagate(errp
, local_err
);
1493 if (update_header
) {
1494 ret
= qcow2_update_header(bs
);
1496 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1501 /* Initialise locks */
1502 qemu_co_mutex_init(&s
->lock
);
1503 bs
->supported_zero_flags
= header
.version
>= 3 ? BDRV_REQ_MAY_UNMAP
: 0;
1505 /* Repair image if dirty */
1506 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INACTIVE
)) && !bs
->read_only
&&
1507 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1508 BdrvCheckResult result
= {0};
1510 ret
= qcow2_check(bs
, &result
, BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1511 if (ret
< 0 || result
.check_errors
) {
1515 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1522 BdrvCheckResult result
= {0};
1523 qcow2_check_refcounts(bs
, &result
, 0);
1529 g_free(s
->unknown_header_fields
);
1530 cleanup_unknown_header_ext(bs
);
1531 qcow2_free_snapshots(bs
);
1532 qcow2_refcount_close(bs
);
1533 qemu_vfree(s
->l1_table
);
1534 /* else pre-write overlap checks in cache_destroy may crash */
1536 cache_clean_timer_del(bs
);
1537 if (s
->l2_table_cache
) {
1538 qcow2_cache_destroy(s
->l2_table_cache
);
1540 if (s
->refcount_block_cache
) {
1541 qcow2_cache_destroy(s
->refcount_block_cache
);
1543 qcrypto_block_free(s
->crypto
);
1544 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1548 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1551 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_file
,
1557 return qcow2_do_open(bs
, options
, flags
, errp
);
1560 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1562 BDRVQcow2State
*s
= bs
->opaque
;
1564 if (bs
->encrypted
) {
1565 /* Encryption works on a sector granularity */
1566 bs
->bl
.request_alignment
= BDRV_SECTOR_SIZE
;
1568 bs
->bl
.pwrite_zeroes_alignment
= s
->cluster_size
;
1569 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1572 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1573 BlockReopenQueue
*queue
, Error
**errp
)
1575 Qcow2ReopenState
*r
;
1578 r
= g_new0(Qcow2ReopenState
, 1);
1581 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1582 state
->flags
, errp
);
1587 /* We need to write out any unwritten data if we reopen read-only. */
1588 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1589 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1594 ret
= bdrv_flush(state
->bs
);
1599 ret
= qcow2_mark_clean(state
->bs
);
1608 qcow2_update_options_abort(state
->bs
, r
);
1613 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1615 qcow2_update_options_commit(state
->bs
, state
->opaque
);
1616 g_free(state
->opaque
);
1619 static void qcow2_reopen_abort(BDRVReopenState
*state
)
1621 qcow2_update_options_abort(state
->bs
, state
->opaque
);
1622 g_free(state
->opaque
);
1625 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
1627 bool has_new_overlap_template
=
1628 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
1629 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1630 bool has_new_total_cache_size
=
1631 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
1632 bool has_all_cache_options
;
1634 /* New overlap template overrides all old overlap options */
1635 if (has_new_overlap_template
) {
1636 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
1637 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1638 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
1639 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
1640 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
1641 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
1642 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
1643 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
1644 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
1645 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
1648 /* New total cache size overrides all old options */
1649 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
1650 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
1651 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1654 qdict_join(options
, old_options
, false);
1657 * If after merging all cache size options are set, an old total size is
1658 * overwritten. Do keep all options, however, if all three are new. The
1659 * resulting error message is what we want to happen.
1661 has_all_cache_options
=
1662 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
1663 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
1664 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1666 if (has_all_cache_options
&& !has_new_total_cache_size
) {
1667 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
1671 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
1673 int64_t offset
, int64_t count
,
1674 int64_t *pnum
, int64_t *map
,
1675 BlockDriverState
**file
)
1677 BDRVQcow2State
*s
= bs
->opaque
;
1678 uint64_t cluster_offset
;
1679 int index_in_cluster
, ret
;
1683 bytes
= MIN(INT_MAX
, count
);
1684 qemu_co_mutex_lock(&s
->lock
);
1685 ret
= qcow2_get_cluster_offset(bs
, offset
, &bytes
, &cluster_offset
);
1686 qemu_co_mutex_unlock(&s
->lock
);
1693 if (cluster_offset
!= 0 && ret
!= QCOW2_CLUSTER_COMPRESSED
&&
1695 index_in_cluster
= offset
& (s
->cluster_size
- 1);
1696 *map
= cluster_offset
| index_in_cluster
;
1697 *file
= bs
->file
->bs
;
1698 status
|= BDRV_BLOCK_OFFSET_VALID
;
1700 if (ret
== QCOW2_CLUSTER_ZERO_PLAIN
|| ret
== QCOW2_CLUSTER_ZERO_ALLOC
) {
1701 status
|= BDRV_BLOCK_ZERO
;
1702 } else if (ret
!= QCOW2_CLUSTER_UNALLOCATED
) {
1703 status
|= BDRV_BLOCK_DATA
;
1708 static coroutine_fn
int qcow2_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1709 uint64_t bytes
, QEMUIOVector
*qiov
,
1712 BDRVQcow2State
*s
= bs
->opaque
;
1713 int offset_in_cluster
;
1715 unsigned int cur_bytes
; /* number of bytes in current iteration */
1716 uint64_t cluster_offset
= 0;
1717 uint64_t bytes_done
= 0;
1718 QEMUIOVector hd_qiov
;
1719 uint8_t *cluster_data
= NULL
;
1721 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1723 qemu_co_mutex_lock(&s
->lock
);
1725 while (bytes
!= 0) {
1727 /* prepare next request */
1728 cur_bytes
= MIN(bytes
, INT_MAX
);
1730 cur_bytes
= MIN(cur_bytes
,
1731 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1734 ret
= qcow2_get_cluster_offset(bs
, offset
, &cur_bytes
, &cluster_offset
);
1739 offset_in_cluster
= offset_into_cluster(s
, offset
);
1741 qemu_iovec_reset(&hd_qiov
);
1742 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
1745 case QCOW2_CLUSTER_UNALLOCATED
:
1748 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
1749 qemu_co_mutex_unlock(&s
->lock
);
1750 ret
= bdrv_co_preadv(bs
->backing
, offset
, cur_bytes
,
1752 qemu_co_mutex_lock(&s
->lock
);
1757 /* Note: in this case, no need to wait */
1758 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1762 case QCOW2_CLUSTER_ZERO_PLAIN
:
1763 case QCOW2_CLUSTER_ZERO_ALLOC
:
1764 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1767 case QCOW2_CLUSTER_COMPRESSED
:
1768 /* add AIO support for compressed blocks ? */
1769 ret
= qcow2_decompress_cluster(bs
, cluster_offset
);
1774 qemu_iovec_from_buf(&hd_qiov
, 0,
1775 s
->cluster_cache
+ offset_in_cluster
,
1779 case QCOW2_CLUSTER_NORMAL
:
1780 if ((cluster_offset
& 511) != 0) {
1785 if (bs
->encrypted
) {
1789 * For encrypted images, read everything into a temporary
1790 * contiguous buffer on which the AES functions can work.
1792 if (!cluster_data
) {
1794 qemu_try_blockalign(bs
->file
->bs
,
1795 QCOW_MAX_CRYPT_CLUSTERS
1797 if (cluster_data
== NULL
) {
1803 assert(cur_bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1804 qemu_iovec_reset(&hd_qiov
);
1805 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
1808 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1809 qemu_co_mutex_unlock(&s
->lock
);
1810 ret
= bdrv_co_preadv(bs
->file
,
1811 cluster_offset
+ offset_in_cluster
,
1812 cur_bytes
, &hd_qiov
, 0);
1813 qemu_co_mutex_lock(&s
->lock
);
1817 if (bs
->encrypted
) {
1819 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1820 assert((cur_bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1821 if (qcrypto_block_decrypt(s
->crypto
,
1822 (s
->crypt_physical_offset
?
1823 cluster_offset
+ offset_in_cluster
:
1831 qemu_iovec_from_buf(qiov
, bytes_done
, cluster_data
, cur_bytes
);
1836 g_assert_not_reached();
1842 offset
+= cur_bytes
;
1843 bytes_done
+= cur_bytes
;
1848 qemu_co_mutex_unlock(&s
->lock
);
1850 qemu_iovec_destroy(&hd_qiov
);
1851 qemu_vfree(cluster_data
);
1856 /* Check if it's possible to merge a write request with the writing of
1857 * the data from the COW regions */
1858 static bool merge_cow(uint64_t offset
, unsigned bytes
,
1859 QEMUIOVector
*hd_qiov
, QCowL2Meta
*l2meta
)
1863 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
1864 /* If both COW regions are empty then there's nothing to merge */
1865 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
1869 /* The data (middle) region must be immediately after the
1871 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
1875 /* The end region must be immediately after the data (middle)
1877 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
1881 /* Make sure that adding both COW regions to the QEMUIOVector
1882 * does not exceed IOV_MAX */
1883 if (hd_qiov
->niov
> IOV_MAX
- 2) {
1887 m
->data_qiov
= hd_qiov
;
1894 static coroutine_fn
int qcow2_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1895 uint64_t bytes
, QEMUIOVector
*qiov
,
1898 BDRVQcow2State
*s
= bs
->opaque
;
1899 int offset_in_cluster
;
1901 unsigned int cur_bytes
; /* number of sectors in current iteration */
1902 uint64_t cluster_offset
;
1903 QEMUIOVector hd_qiov
;
1904 uint64_t bytes_done
= 0;
1905 uint8_t *cluster_data
= NULL
;
1906 QCowL2Meta
*l2meta
= NULL
;
1908 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
1910 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1912 s
->cluster_cache_offset
= -1; /* disable compressed cache */
1914 qemu_co_mutex_lock(&s
->lock
);
1916 while (bytes
!= 0) {
1920 trace_qcow2_writev_start_part(qemu_coroutine_self());
1921 offset_in_cluster
= offset_into_cluster(s
, offset
);
1922 cur_bytes
= MIN(bytes
, INT_MAX
);
1923 if (bs
->encrypted
) {
1924 cur_bytes
= MIN(cur_bytes
,
1925 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
1926 - offset_in_cluster
);
1929 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
1930 &cluster_offset
, &l2meta
);
1935 assert((cluster_offset
& 511) == 0);
1937 qemu_iovec_reset(&hd_qiov
);
1938 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
1940 if (bs
->encrypted
) {
1942 if (!cluster_data
) {
1943 cluster_data
= qemu_try_blockalign(bs
->file
->bs
,
1944 QCOW_MAX_CRYPT_CLUSTERS
1946 if (cluster_data
== NULL
) {
1952 assert(hd_qiov
.size
<=
1953 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1954 qemu_iovec_to_buf(&hd_qiov
, 0, cluster_data
, hd_qiov
.size
);
1956 if (qcrypto_block_encrypt(s
->crypto
,
1957 (s
->crypt_physical_offset
?
1958 cluster_offset
+ offset_in_cluster
:
1961 cur_bytes
, NULL
) < 0) {
1966 qemu_iovec_reset(&hd_qiov
);
1967 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
1970 ret
= qcow2_pre_write_overlap_check(bs
, 0,
1971 cluster_offset
+ offset_in_cluster
, cur_bytes
);
1976 /* If we need to do COW, check if it's possible to merge the
1977 * writing of the guest data together with that of the COW regions.
1978 * If it's not possible (or not necessary) then write the
1979 * guest data now. */
1980 if (!merge_cow(offset
, cur_bytes
, &hd_qiov
, l2meta
)) {
1981 qemu_co_mutex_unlock(&s
->lock
);
1982 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
1983 trace_qcow2_writev_data(qemu_coroutine_self(),
1984 cluster_offset
+ offset_in_cluster
);
1985 ret
= bdrv_co_pwritev(bs
->file
,
1986 cluster_offset
+ offset_in_cluster
,
1987 cur_bytes
, &hd_qiov
, 0);
1988 qemu_co_mutex_lock(&s
->lock
);
1994 while (l2meta
!= NULL
) {
1997 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
2002 /* Take the request off the list of running requests */
2003 if (l2meta
->nb_clusters
!= 0) {
2004 QLIST_REMOVE(l2meta
, next_in_flight
);
2007 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2009 next
= l2meta
->next
;
2015 offset
+= cur_bytes
;
2016 bytes_done
+= cur_bytes
;
2017 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2022 while (l2meta
!= NULL
) {
2025 if (l2meta
->nb_clusters
!= 0) {
2026 QLIST_REMOVE(l2meta
, next_in_flight
);
2028 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2030 next
= l2meta
->next
;
2035 qemu_co_mutex_unlock(&s
->lock
);
2037 qemu_iovec_destroy(&hd_qiov
);
2038 qemu_vfree(cluster_data
);
2039 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2044 static int qcow2_inactivate(BlockDriverState
*bs
)
2046 BDRVQcow2State
*s
= bs
->opaque
;
2047 int ret
, result
= 0;
2048 Error
*local_err
= NULL
;
2050 qcow2_store_persistent_dirty_bitmaps(bs
, &local_err
);
2051 if (local_err
!= NULL
) {
2053 error_report_err(local_err
);
2054 error_report("Persistent bitmaps are lost for node '%s'",
2055 bdrv_get_device_or_node_name(bs
));
2058 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2061 error_report("Failed to flush the L2 table cache: %s",
2065 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2068 error_report("Failed to flush the refcount block cache: %s",
2073 qcow2_mark_clean(bs
);
2079 static void qcow2_close(BlockDriverState
*bs
)
2081 BDRVQcow2State
*s
= bs
->opaque
;
2082 qemu_vfree(s
->l1_table
);
2083 /* else pre-write overlap checks in cache_destroy may crash */
2086 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2087 qcow2_inactivate(bs
);
2090 cache_clean_timer_del(bs
);
2091 qcow2_cache_destroy(s
->l2_table_cache
);
2092 qcow2_cache_destroy(s
->refcount_block_cache
);
2094 qcrypto_block_free(s
->crypto
);
2097 g_free(s
->unknown_header_fields
);
2098 cleanup_unknown_header_ext(bs
);
2100 g_free(s
->image_backing_file
);
2101 g_free(s
->image_backing_format
);
2103 g_free(s
->cluster_cache
);
2104 qemu_vfree(s
->cluster_data
);
2105 qcow2_refcount_close(bs
);
2106 qcow2_free_snapshots(bs
);
2109 static void qcow2_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
2111 BDRVQcow2State
*s
= bs
->opaque
;
2112 int flags
= s
->flags
;
2113 QCryptoBlock
*crypto
= NULL
;
2115 Error
*local_err
= NULL
;
2119 * Backing files are read-only which makes all of their metadata immutable,
2120 * that means we don't have to worry about reopening them here.
2128 memset(s
, 0, sizeof(BDRVQcow2State
));
2129 options
= qdict_clone_shallow(bs
->options
);
2131 flags
&= ~BDRV_O_INACTIVE
;
2132 ret
= qcow2_do_open(bs
, options
, flags
, &local_err
);
2135 error_propagate(errp
, local_err
);
2136 error_prepend(errp
, "Could not reopen qcow2 layer: ");
2139 } else if (ret
< 0) {
2140 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
2148 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2149 size_t len
, size_t buflen
)
2151 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2152 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2154 if (buflen
< ext_len
) {
2158 *ext_backing_fmt
= (QCowExtension
) {
2159 .magic
= cpu_to_be32(magic
),
2160 .len
= cpu_to_be32(len
),
2164 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2171 * Updates the qcow2 header, including the variable length parts of it, i.e.
2172 * the backing file name and all extensions. qcow2 was not designed to allow
2173 * such changes, so if we run out of space (we can only use the first cluster)
2174 * this function may fail.
2176 * Returns 0 on success, -errno in error cases.
2178 int qcow2_update_header(BlockDriverState
*bs
)
2180 BDRVQcow2State
*s
= bs
->opaque
;
2183 size_t buflen
= s
->cluster_size
;
2185 uint64_t total_size
;
2186 uint32_t refcount_table_clusters
;
2187 size_t header_length
;
2188 Qcow2UnknownHeaderExtension
*uext
;
2190 buf
= qemu_blockalign(bs
, buflen
);
2192 /* Header structure */
2193 header
= (QCowHeader
*) buf
;
2195 if (buflen
< sizeof(*header
)) {
2200 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2201 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2202 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2204 *header
= (QCowHeader
) {
2205 /* Version 2 fields */
2206 .magic
= cpu_to_be32(QCOW_MAGIC
),
2207 .version
= cpu_to_be32(s
->qcow_version
),
2208 .backing_file_offset
= 0,
2209 .backing_file_size
= 0,
2210 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2211 .size
= cpu_to_be64(total_size
),
2212 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2213 .l1_size
= cpu_to_be32(s
->l1_size
),
2214 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2215 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2216 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2217 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2218 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2220 /* Version 3 fields */
2221 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2222 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2223 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2224 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2225 .header_length
= cpu_to_be32(header_length
),
2228 /* For older versions, write a shorter header */
2229 switch (s
->qcow_version
) {
2231 ret
= offsetof(QCowHeader
, incompatible_features
);
2234 ret
= sizeof(*header
);
2243 memset(buf
, 0, buflen
);
2245 /* Preserve any unknown field in the header */
2246 if (s
->unknown_header_fields_size
) {
2247 if (buflen
< s
->unknown_header_fields_size
) {
2252 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2253 buf
+= s
->unknown_header_fields_size
;
2254 buflen
-= s
->unknown_header_fields_size
;
2257 /* Backing file format header extension */
2258 if (s
->image_backing_format
) {
2259 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2260 s
->image_backing_format
,
2261 strlen(s
->image_backing_format
),
2271 /* Full disk encryption header pointer extension */
2272 if (s
->crypto_header
.offset
!= 0) {
2273 cpu_to_be64s(&s
->crypto_header
.offset
);
2274 cpu_to_be64s(&s
->crypto_header
.length
);
2275 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2276 &s
->crypto_header
, sizeof(s
->crypto_header
),
2278 be64_to_cpus(&s
->crypto_header
.offset
);
2279 be64_to_cpus(&s
->crypto_header
.length
);
2288 if (s
->qcow_version
>= 3) {
2289 Qcow2Feature features
[] = {
2291 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2292 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2293 .name
= "dirty bit",
2296 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2297 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
2298 .name
= "corrupt bit",
2301 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
2302 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
2303 .name
= "lazy refcounts",
2307 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
2308 features
, sizeof(features
), buflen
);
2316 /* Bitmap extension */
2317 if (s
->nb_bitmaps
> 0) {
2318 Qcow2BitmapHeaderExt bitmaps_header
= {
2319 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
2320 .bitmap_directory_size
=
2321 cpu_to_be64(s
->bitmap_directory_size
),
2322 .bitmap_directory_offset
=
2323 cpu_to_be64(s
->bitmap_directory_offset
)
2325 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
2326 &bitmaps_header
, sizeof(bitmaps_header
),
2335 /* Keep unknown header extensions */
2336 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
2337 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
2346 /* End of header extensions */
2347 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
2355 /* Backing file name */
2356 if (s
->image_backing_file
) {
2357 size_t backing_file_len
= strlen(s
->image_backing_file
);
2359 if (buflen
< backing_file_len
) {
2364 /* Using strncpy is ok here, since buf is not NUL-terminated. */
2365 strncpy(buf
, s
->image_backing_file
, buflen
);
2367 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
2368 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
2371 /* Write the new header */
2372 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
2383 static int qcow2_change_backing_file(BlockDriverState
*bs
,
2384 const char *backing_file
, const char *backing_fmt
)
2386 BDRVQcow2State
*s
= bs
->opaque
;
2388 if (backing_file
&& strlen(backing_file
) > 1023) {
2392 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
2393 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
2395 g_free(s
->image_backing_file
);
2396 g_free(s
->image_backing_format
);
2398 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
2399 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
2401 return qcow2_update_header(bs
);
2404 static int qcow2_crypt_method_from_format(const char *encryptfmt
)
2406 if (g_str_equal(encryptfmt
, "luks")) {
2407 return QCOW_CRYPT_LUKS
;
2408 } else if (g_str_equal(encryptfmt
, "aes")) {
2409 return QCOW_CRYPT_AES
;
2415 static int qcow2_set_up_encryption(BlockDriverState
*bs
, const char *encryptfmt
,
2416 QemuOpts
*opts
, Error
**errp
)
2418 BDRVQcow2State
*s
= bs
->opaque
;
2419 QCryptoBlockCreateOptions
*cryptoopts
= NULL
;
2420 QCryptoBlock
*crypto
= NULL
;
2422 QDict
*options
, *encryptopts
;
2425 options
= qemu_opts_to_qdict(opts
, NULL
);
2426 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
2429 fmt
= qcow2_crypt_method_from_format(encryptfmt
);
2432 case QCOW_CRYPT_LUKS
:
2433 cryptoopts
= block_crypto_create_opts_init(
2434 Q_CRYPTO_BLOCK_FORMAT_LUKS
, encryptopts
, errp
);
2436 case QCOW_CRYPT_AES
:
2437 cryptoopts
= block_crypto_create_opts_init(
2438 Q_CRYPTO_BLOCK_FORMAT_QCOW
, encryptopts
, errp
);
2441 error_setg(errp
, "Unknown encryption format '%s'", encryptfmt
);
2448 s
->crypt_method_header
= fmt
;
2450 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
2451 qcow2_crypto_hdr_init_func
,
2452 qcow2_crypto_hdr_write_func
,
2459 ret
= qcow2_update_header(bs
);
2461 error_setg_errno(errp
, -ret
, "Could not write encryption header");
2466 QDECREF(encryptopts
);
2467 qcrypto_block_free(crypto
);
2468 qapi_free_QCryptoBlockCreateOptions(cryptoopts
);
2473 typedef struct PreallocCo
{
2474 BlockDriverState
*bs
;
2476 uint64_t new_length
;
2482 * Preallocates metadata structures for data clusters between @offset (in the
2483 * guest disk) and @new_length (which is thus generally the new guest disk
2486 * Returns: 0 on success, -errno on failure.
2488 static void coroutine_fn
preallocate_co(void *opaque
)
2490 PreallocCo
*params
= opaque
;
2491 BlockDriverState
*bs
= params
->bs
;
2492 uint64_t offset
= params
->offset
;
2493 uint64_t new_length
= params
->new_length
;
2494 BDRVQcow2State
*s
= bs
->opaque
;
2496 uint64_t host_offset
= 0;
2497 unsigned int cur_bytes
;
2501 qemu_co_mutex_lock(&s
->lock
);
2503 assert(offset
<= new_length
);
2504 bytes
= new_length
- offset
;
2507 cur_bytes
= MIN(bytes
, INT_MAX
);
2508 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2509 &host_offset
, &meta
);
2515 QCowL2Meta
*next
= meta
->next
;
2517 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
2519 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
2520 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
2524 /* There are no dependent requests, but we need to remove our
2525 * request from the list of in-flight requests */
2526 QLIST_REMOVE(meta
, next_in_flight
);
2532 /* TODO Preallocate data if requested */
2535 offset
+= cur_bytes
;
2539 * It is expected that the image file is large enough to actually contain
2540 * all of the allocated clusters (otherwise we get failing reads after
2541 * EOF). Extend the image to the last allocated sector.
2543 if (host_offset
!= 0) {
2545 ret
= bdrv_pwrite(bs
->file
, (host_offset
+ cur_bytes
) - 1,
2555 qemu_co_mutex_unlock(&s
->lock
);
2559 static int preallocate(BlockDriverState
*bs
,
2560 uint64_t offset
, uint64_t new_length
)
2562 PreallocCo params
= {
2565 .new_length
= new_length
,
2566 .ret
= -EINPROGRESS
,
2569 if (qemu_in_coroutine()) {
2570 preallocate_co(¶ms
);
2572 Coroutine
*co
= qemu_coroutine_create(preallocate_co
, ¶ms
);
2573 bdrv_coroutine_enter(bs
, co
);
2574 BDRV_POLL_WHILE(bs
, params
.ret
== -EINPROGRESS
);
2579 /* qcow2_refcount_metadata_size:
2580 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2581 * @cluster_size: size of a cluster, in bytes
2582 * @refcount_order: refcount bits power-of-2 exponent
2583 * @generous_increase: allow for the refcount table to be 1.5x as large as it
2586 * Returns: Number of bytes required for refcount blocks and table metadata.
2588 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
2589 int refcount_order
, bool generous_increase
,
2590 uint64_t *refblock_count
)
2593 * Every host cluster is reference-counted, including metadata (even
2594 * refcount metadata is recursively included).
2596 * An accurate formula for the size of refcount metadata size is difficult
2597 * to derive. An easier method of calculation is finding the fixed point
2598 * where no further refcount blocks or table clusters are required to
2599 * reference count every cluster.
2601 int64_t blocks_per_table_cluster
= cluster_size
/ sizeof(uint64_t);
2602 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
2603 int64_t table
= 0; /* number of refcount table clusters */
2604 int64_t blocks
= 0; /* number of refcount block clusters */
2610 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
2611 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
2612 n
= clusters
+ blocks
+ table
;
2614 if (n
== last
&& generous_increase
) {
2615 clusters
+= DIV_ROUND_UP(table
, 2);
2616 n
= 0; /* force another loop */
2617 generous_increase
= false;
2619 } while (n
!= last
);
2621 if (refblock_count
) {
2622 *refblock_count
= blocks
;
2625 return (blocks
+ table
) * cluster_size
;
2629 * qcow2_calc_prealloc_size:
2630 * @total_size: virtual disk size in bytes
2631 * @cluster_size: cluster size in bytes
2632 * @refcount_order: refcount bits power-of-2 exponent
2634 * Returns: Total number of bytes required for the fully allocated image
2635 * (including metadata).
2637 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
2638 size_t cluster_size
,
2641 int64_t meta_size
= 0;
2642 uint64_t nl1e
, nl2e
;
2643 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
2645 /* header: 1 cluster */
2646 meta_size
+= cluster_size
;
2648 /* total size of L2 tables */
2649 nl2e
= aligned_total_size
/ cluster_size
;
2650 nl2e
= ROUND_UP(nl2e
, cluster_size
/ sizeof(uint64_t));
2651 meta_size
+= nl2e
* sizeof(uint64_t);
2653 /* total size of L1 tables */
2654 nl1e
= nl2e
* sizeof(uint64_t) / cluster_size
;
2655 nl1e
= ROUND_UP(nl1e
, cluster_size
/ sizeof(uint64_t));
2656 meta_size
+= nl1e
* sizeof(uint64_t);
2658 /* total size of refcount table and blocks */
2659 meta_size
+= qcow2_refcount_metadata_size(
2660 (meta_size
+ aligned_total_size
) / cluster_size
,
2661 cluster_size
, refcount_order
, false, NULL
);
2663 return meta_size
+ aligned_total_size
;
2666 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, Error
**errp
)
2668 size_t cluster_size
;
2671 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
2672 DEFAULT_CLUSTER_SIZE
);
2673 cluster_bits
= ctz32(cluster_size
);
2674 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
2675 (1 << cluster_bits
) != cluster_size
)
2677 error_setg(errp
, "Cluster size must be a power of two between %d and "
2678 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
2681 return cluster_size
;
2684 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
2689 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
2691 ret
= 3; /* default */
2692 } else if (!strcmp(buf
, "0.10")) {
2694 } else if (!strcmp(buf
, "1.1")) {
2697 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
2704 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
2707 uint64_t refcount_bits
;
2709 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
2710 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
2711 error_setg(errp
, "Refcount width must be a power of two and may not "
2716 if (version
< 3 && refcount_bits
!= 16) {
2717 error_setg(errp
, "Different refcount widths than 16 bits require "
2718 "compatibility level 1.1 or above (use compat=1.1 or "
2723 return refcount_bits
;
2726 static int coroutine_fn
2727 qcow2_co_create2(const char *filename
, int64_t total_size
,
2728 const char *backing_file
, const char *backing_format
,
2729 int flags
, size_t cluster_size
, PreallocMode prealloc
,
2730 QemuOpts
*opts
, int version
, int refcount_order
,
2731 const char *encryptfmt
, Error
**errp
)
2736 * Open the image file and write a minimal qcow2 header.
2738 * We keep things simple and start with a zero-sized image. We also
2739 * do without refcount blocks or a L1 table for now. We'll fix the
2740 * inconsistency later.
2742 * We do need a refcount table because growing the refcount table means
2743 * allocating two new refcount blocks - the seconds of which would be at
2744 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2745 * size for any qcow2 image.
2749 uint64_t* refcount_table
;
2750 Error
*local_err
= NULL
;
2753 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
2754 int64_t prealloc_size
=
2755 qcow2_calc_prealloc_size(total_size
, cluster_size
, refcount_order
);
2756 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, prealloc_size
, &error_abort
);
2757 qemu_opt_set(opts
, BLOCK_OPT_PREALLOC
, PreallocMode_str(prealloc
),
2761 ret
= bdrv_create_file(filename
, opts
, &local_err
);
2763 error_propagate(errp
, local_err
);
2767 blk
= blk_new_open(filename
, NULL
, NULL
,
2768 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
,
2771 error_propagate(errp
, local_err
);
2775 blk_set_allow_write_beyond_eof(blk
, true);
2777 /* Write the header */
2778 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
2779 header
= g_malloc0(cluster_size
);
2780 *header
= (QCowHeader
) {
2781 .magic
= cpu_to_be32(QCOW_MAGIC
),
2782 .version
= cpu_to_be32(version
),
2783 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
2784 .size
= cpu_to_be64(0),
2785 .l1_table_offset
= cpu_to_be64(0),
2786 .l1_size
= cpu_to_be32(0),
2787 .refcount_table_offset
= cpu_to_be64(cluster_size
),
2788 .refcount_table_clusters
= cpu_to_be32(1),
2789 .refcount_order
= cpu_to_be32(refcount_order
),
2790 .header_length
= cpu_to_be32(sizeof(*header
)),
2793 /* We'll update this to correct value later */
2794 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
2796 if (flags
& BLOCK_FLAG_LAZY_REFCOUNTS
) {
2797 header
->compatible_features
|=
2798 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
2801 ret
= blk_pwrite(blk
, 0, header
, cluster_size
, 0);
2804 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
2808 /* Write a refcount table with one refcount block */
2809 refcount_table
= g_malloc0(2 * cluster_size
);
2810 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
2811 ret
= blk_pwrite(blk
, cluster_size
, refcount_table
, 2 * cluster_size
, 0);
2812 g_free(refcount_table
);
2815 error_setg_errno(errp
, -ret
, "Could not write refcount table");
2823 * And now open the image and make it consistent first (i.e. increase the
2824 * refcount of the cluster that is occupied by the header and the refcount
2827 options
= qdict_new();
2828 qdict_put_str(options
, "driver", "qcow2");
2829 blk
= blk_new_open(filename
, NULL
, options
,
2830 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
2833 error_propagate(errp
, local_err
);
2838 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
2840 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
2841 "header and refcount table");
2844 } else if (ret
!= 0) {
2845 error_report("Huh, first cluster in empty image is already in use?");
2849 /* Create a full header (including things like feature table) */
2850 ret
= qcow2_update_header(blk_bs(blk
));
2852 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
2856 /* Okay, now that we have a valid image, let's give it the right size */
2857 ret
= blk_truncate(blk
, total_size
, PREALLOC_MODE_OFF
, errp
);
2859 error_prepend(errp
, "Could not resize image: ");
2863 /* Want a backing file? There you go.*/
2865 ret
= bdrv_change_backing_file(blk_bs(blk
), backing_file
, backing_format
);
2867 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
2868 "with format '%s'", backing_file
, backing_format
);
2873 /* Want encryption? There you go. */
2875 ret
= qcow2_set_up_encryption(blk_bs(blk
), encryptfmt
, opts
, errp
);
2881 /* And if we're supposed to preallocate metadata, do that now */
2882 if (prealloc
!= PREALLOC_MODE_OFF
) {
2883 ret
= preallocate(blk_bs(blk
), 0, total_size
);
2885 error_setg_errno(errp
, -ret
, "Could not preallocate metadata");
2893 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
2894 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
2895 * have to setup decryption context. We're not doing any I/O on the top
2896 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
2899 options
= qdict_new();
2900 qdict_put_str(options
, "driver", "qcow2");
2901 blk
= blk_new_open(filename
, NULL
, options
,
2902 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
2905 error_propagate(errp
, local_err
);
2918 static int coroutine_fn
qcow2_co_create_opts(const char *filename
, QemuOpts
*opts
,
2921 char *backing_file
= NULL
;
2922 char *backing_fmt
= NULL
;
2926 size_t cluster_size
= DEFAULT_CLUSTER_SIZE
;
2927 PreallocMode prealloc
;
2929 uint64_t refcount_bits
;
2931 char *encryptfmt
= NULL
;
2932 Error
*local_err
= NULL
;
2935 /* Read out options */
2936 size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2938 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
2939 backing_fmt
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FMT
);
2940 encryptfmt
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
2942 if (qemu_opt_get(opts
, BLOCK_OPT_ENCRYPT
)) {
2943 error_setg(errp
, "Options " BLOCK_OPT_ENCRYPT
" and "
2944 BLOCK_OPT_ENCRYPT_FORMAT
" are mutually exclusive");
2948 } else if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_ENCRYPT
, false)) {
2949 encryptfmt
= g_strdup("aes");
2951 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, &local_err
);
2953 error_propagate(errp
, local_err
);
2957 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2958 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2959 PREALLOC_MODE_OFF
, &local_err
);
2961 error_propagate(errp
, local_err
);
2966 version
= qcow2_opt_get_version_del(opts
, &local_err
);
2968 error_propagate(errp
, local_err
);
2973 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_LAZY_REFCOUNTS
, false)) {
2974 flags
|= BLOCK_FLAG_LAZY_REFCOUNTS
;
2977 if (backing_file
&& prealloc
!= PREALLOC_MODE_OFF
) {
2978 error_setg(errp
, "Backing file and preallocation cannot be used at "
2984 if (version
< 3 && (flags
& BLOCK_FLAG_LAZY_REFCOUNTS
)) {
2985 error_setg(errp
, "Lazy refcounts only supported with compatibility "
2986 "level 1.1 and above (use compat=1.1 or greater)");
2991 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
2993 error_propagate(errp
, local_err
);
2998 refcount_order
= ctz32(refcount_bits
);
3000 ret
= qcow2_co_create2(filename
, size
, backing_file
, backing_fmt
, flags
,
3001 cluster_size
, prealloc
, opts
, version
, refcount_order
,
3002 encryptfmt
, &local_err
);
3003 error_propagate(errp
, local_err
);
3006 g_free(backing_file
);
3007 g_free(backing_fmt
);
3014 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3019 /* Clamp to image length, before checking status of underlying sectors */
3020 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3021 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3027 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3028 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
== bytes
;
3031 static coroutine_fn
int qcow2_co_pwrite_zeroes(BlockDriverState
*bs
,
3032 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3035 BDRVQcow2State
*s
= bs
->opaque
;
3037 uint32_t head
= offset
% s
->cluster_size
;
3038 uint32_t tail
= (offset
+ bytes
) % s
->cluster_size
;
3040 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3041 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3049 assert(head
+ bytes
<= s
->cluster_size
);
3051 /* check whether remainder of cluster already reads as zero */
3052 if (!(is_zero(bs
, offset
- head
, head
) &&
3053 is_zero(bs
, offset
+ bytes
,
3054 tail
? s
->cluster_size
- tail
: 0))) {
3058 qemu_co_mutex_lock(&s
->lock
);
3059 /* We can have new write after previous check */
3060 offset
= QEMU_ALIGN_DOWN(offset
, s
->cluster_size
);
3061 bytes
= s
->cluster_size
;
3062 nr
= s
->cluster_size
;
3063 ret
= qcow2_get_cluster_offset(bs
, offset
, &nr
, &off
);
3064 if (ret
!= QCOW2_CLUSTER_UNALLOCATED
&&
3065 ret
!= QCOW2_CLUSTER_ZERO_PLAIN
&&
3066 ret
!= QCOW2_CLUSTER_ZERO_ALLOC
) {
3067 qemu_co_mutex_unlock(&s
->lock
);
3071 qemu_co_mutex_lock(&s
->lock
);
3074 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
3076 /* Whatever is left can use real zero clusters */
3077 ret
= qcow2_cluster_zeroize(bs
, offset
, bytes
, flags
);
3078 qemu_co_mutex_unlock(&s
->lock
);
3083 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
3084 int64_t offset
, int bytes
)
3087 BDRVQcow2State
*s
= bs
->opaque
;
3089 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
3090 assert(bytes
< s
->cluster_size
);
3091 /* Ignore partial clusters, except for the special case of the
3092 * complete partial cluster at the end of an unaligned file */
3093 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
3094 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3099 qemu_co_mutex_lock(&s
->lock
);
3100 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
3102 qemu_co_mutex_unlock(&s
->lock
);
3106 static int qcow2_truncate(BlockDriverState
*bs
, int64_t offset
,
3107 PreallocMode prealloc
, Error
**errp
)
3109 BDRVQcow2State
*s
= bs
->opaque
;
3110 uint64_t old_length
;
3111 int64_t new_l1_size
;
3114 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
3115 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
3117 error_setg(errp
, "Unsupported preallocation mode '%s'",
3118 PreallocMode_str(prealloc
));
3123 error_setg(errp
, "The new size must be a multiple of 512");
3127 /* cannot proceed if image has snapshots */
3128 if (s
->nb_snapshots
) {
3129 error_setg(errp
, "Can't resize an image which has snapshots");
3133 /* cannot proceed if image has bitmaps */
3134 if (s
->nb_bitmaps
) {
3135 /* TODO: resize bitmaps in the image */
3136 error_setg(errp
, "Can't resize an image which has bitmaps");
3140 old_length
= bs
->total_sectors
* 512;
3141 new_l1_size
= size_to_l1(s
, offset
);
3143 if (offset
< old_length
) {
3144 int64_t last_cluster
, old_file_size
;
3145 if (prealloc
!= PREALLOC_MODE_OFF
) {
3147 "Preallocation can't be used for shrinking an image");
3151 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
3152 old_length
- ROUND_UP(offset
,
3154 QCOW2_DISCARD_ALWAYS
, true);
3156 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
3160 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
3162 error_setg_errno(errp
, -ret
,
3163 "Failed to reduce the number of L2 tables");
3167 ret
= qcow2_shrink_reftable(bs
);
3169 error_setg_errno(errp
, -ret
,
3170 "Failed to discard unused refblocks");
3174 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3175 if (old_file_size
< 0) {
3176 error_setg_errno(errp
, -old_file_size
,
3177 "Failed to inquire current file length");
3178 return old_file_size
;
3180 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
3181 if (last_cluster
< 0) {
3182 error_setg_errno(errp
, -last_cluster
,
3183 "Failed to find the last cluster");
3184 return last_cluster
;
3186 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
3187 Error
*local_err
= NULL
;
3189 bdrv_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
3190 PREALLOC_MODE_OFF
, &local_err
);
3192 warn_reportf_err(local_err
,
3193 "Failed to truncate the tail of the image: ");
3197 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
3199 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
3205 case PREALLOC_MODE_OFF
:
3208 case PREALLOC_MODE_METADATA
:
3209 ret
= preallocate(bs
, old_length
, offset
);
3211 error_setg_errno(errp
, -ret
, "Preallocation failed");
3216 case PREALLOC_MODE_FALLOC
:
3217 case PREALLOC_MODE_FULL
:
3219 int64_t allocation_start
, host_offset
, guest_offset
;
3220 int64_t clusters_allocated
;
3221 int64_t old_file_size
, new_file_size
;
3222 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
3224 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3225 if (old_file_size
< 0) {
3226 error_setg_errno(errp
, -old_file_size
,
3227 "Failed to inquire current file length");
3228 return old_file_size
;
3230 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
3232 nb_new_data_clusters
= DIV_ROUND_UP(offset
- old_length
,
3235 /* This is an overestimation; we will not actually allocate space for
3236 * these in the file but just make sure the new refcount structures are
3237 * able to cover them so we will not have to allocate new refblocks
3238 * while entering the data blocks in the potentially new L2 tables.
3239 * (We do not actually care where the L2 tables are placed. Maybe they
3240 * are already allocated or they can be placed somewhere before
3241 * @old_file_size. It does not matter because they will be fully
3242 * allocated automatically, so they do not need to be covered by the
3243 * preallocation. All that matters is that we will not have to allocate
3244 * new refcount structures for them.) */
3245 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
3246 s
->cluster_size
/ sizeof(uint64_t));
3247 /* The cluster range may not be aligned to L2 boundaries, so add one L2
3248 * table for a potential head/tail */
3251 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
3252 nb_new_data_clusters
+
3255 if (allocation_start
< 0) {
3256 error_setg_errno(errp
, -allocation_start
,
3257 "Failed to resize refcount structures");
3258 return allocation_start
;
3261 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
3262 nb_new_data_clusters
);
3263 if (clusters_allocated
< 0) {
3264 error_setg_errno(errp
, -clusters_allocated
,
3265 "Failed to allocate data clusters");
3266 return -clusters_allocated
;
3269 assert(clusters_allocated
== nb_new_data_clusters
);
3271 /* Allocate the data area */
3272 new_file_size
= allocation_start
+
3273 nb_new_data_clusters
* s
->cluster_size
;
3274 ret
= bdrv_truncate(bs
->file
, new_file_size
, prealloc
, errp
);
3276 error_prepend(errp
, "Failed to resize underlying file: ");
3277 qcow2_free_clusters(bs
, allocation_start
,
3278 nb_new_data_clusters
* s
->cluster_size
,
3279 QCOW2_DISCARD_OTHER
);
3283 /* Create the necessary L2 entries */
3284 host_offset
= allocation_start
;
3285 guest_offset
= old_length
;
3286 while (nb_new_data_clusters
) {
3287 int64_t nb_clusters
= MIN(
3288 nb_new_data_clusters
,
3289 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
3290 QCowL2Meta allocation
= {
3291 .offset
= guest_offset
,
3292 .alloc_offset
= host_offset
,
3293 .nb_clusters
= nb_clusters
,
3295 qemu_co_queue_init(&allocation
.dependent_requests
);
3297 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
3299 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
3300 qcow2_free_clusters(bs
, host_offset
,
3301 nb_new_data_clusters
* s
->cluster_size
,
3302 QCOW2_DISCARD_OTHER
);
3306 guest_offset
+= nb_clusters
* s
->cluster_size
;
3307 host_offset
+= nb_clusters
* s
->cluster_size
;
3308 nb_new_data_clusters
-= nb_clusters
;
3314 g_assert_not_reached();
3317 if (prealloc
!= PREALLOC_MODE_OFF
) {
3318 /* Flush metadata before actually changing the image size */
3319 ret
= bdrv_flush(bs
);
3321 error_setg_errno(errp
, -ret
,
3322 "Failed to flush the preallocated area to disk");
3327 /* write updated header.size */
3328 offset
= cpu_to_be64(offset
);
3329 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
3330 &offset
, sizeof(uint64_t));
3332 error_setg_errno(errp
, -ret
, "Failed to update the image size");
3336 s
->l1_vm_state_index
= new_l1_size
;
3340 /* XXX: put compressed sectors first, then all the cluster aligned
3341 tables to avoid losing bytes in alignment */
3342 static coroutine_fn
int
3343 qcow2_co_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
3344 uint64_t bytes
, QEMUIOVector
*qiov
)
3346 BDRVQcow2State
*s
= bs
->opaque
;
3347 QEMUIOVector hd_qiov
;
3351 uint8_t *buf
, *out_buf
;
3352 int64_t cluster_offset
;
3355 /* align end of file to a sector boundary to ease reading with
3356 sector based I/Os */
3357 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
3358 if (cluster_offset
< 0) {
3359 return cluster_offset
;
3361 return bdrv_truncate(bs
->file
, cluster_offset
, PREALLOC_MODE_OFF
, NULL
);
3364 if (offset_into_cluster(s
, offset
)) {
3368 buf
= qemu_blockalign(bs
, s
->cluster_size
);
3369 if (bytes
!= s
->cluster_size
) {
3370 if (bytes
> s
->cluster_size
||
3371 offset
+ bytes
!= bs
->total_sectors
<< BDRV_SECTOR_BITS
)
3376 /* Zero-pad last write if image size is not cluster aligned */
3377 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
3379 qemu_iovec_to_buf(qiov
, 0, buf
, bytes
);
3381 out_buf
= g_malloc(s
->cluster_size
);
3383 /* best compression, small window, no zlib header */
3384 memset(&strm
, 0, sizeof(strm
));
3385 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
3387 9, Z_DEFAULT_STRATEGY
);
3393 strm
.avail_in
= s
->cluster_size
;
3394 strm
.next_in
= (uint8_t *)buf
;
3395 strm
.avail_out
= s
->cluster_size
;
3396 strm
.next_out
= out_buf
;
3398 ret
= deflate(&strm
, Z_FINISH
);
3399 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
3404 out_len
= strm
.next_out
- out_buf
;
3408 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
3409 /* could not compress: write normal cluster */
3410 ret
= qcow2_co_pwritev(bs
, offset
, bytes
, qiov
, 0);
3417 qemu_co_mutex_lock(&s
->lock
);
3419 qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
);
3420 if (!cluster_offset
) {
3421 qemu_co_mutex_unlock(&s
->lock
);
3425 cluster_offset
&= s
->cluster_offset_mask
;
3427 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
);
3428 qemu_co_mutex_unlock(&s
->lock
);
3433 iov
= (struct iovec
) {
3434 .iov_base
= out_buf
,
3437 qemu_iovec_init_external(&hd_qiov
, &iov
, 1);
3439 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
3440 ret
= bdrv_co_pwritev(bs
->file
, cluster_offset
, out_len
, &hd_qiov
, 0);
3452 static int make_completely_empty(BlockDriverState
*bs
)
3454 BDRVQcow2State
*s
= bs
->opaque
;
3455 Error
*local_err
= NULL
;
3456 int ret
, l1_clusters
;
3458 uint64_t *new_reftable
= NULL
;
3459 uint64_t rt_entry
, l1_size2
;
3462 uint64_t reftable_offset
;
3463 uint32_t reftable_clusters
;
3464 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
3466 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
3471 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
3476 /* Refcounts will be broken utterly */
3477 ret
= qcow2_mark_dirty(bs
);
3482 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
3484 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
3485 l1_size2
= (uint64_t)s
->l1_size
* sizeof(uint64_t);
3487 /* After this call, neither the in-memory nor the on-disk refcount
3488 * information accurately describe the actual references */
3490 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
3491 l1_clusters
* s
->cluster_size
, 0);
3493 goto fail_broken_refcounts
;
3495 memset(s
->l1_table
, 0, l1_size2
);
3497 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
3499 /* Overwrite enough clusters at the beginning of the sectors to place
3500 * the refcount table, a refcount block and the L1 table in; this may
3501 * overwrite parts of the existing refcount and L1 table, which is not
3502 * an issue because the dirty flag is set, complete data loss is in fact
3503 * desired and partial data loss is consequently fine as well */
3504 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
3505 (2 + l1_clusters
) * s
->cluster_size
, 0);
3506 /* This call (even if it failed overall) may have overwritten on-disk
3507 * refcount structures; in that case, the in-memory refcount information
3508 * will probably differ from the on-disk information which makes the BDS
3511 goto fail_broken_refcounts
;
3514 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
3515 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
3517 /* "Create" an empty reftable (one cluster) directly after the image
3518 * header and an empty L1 table three clusters after the image header;
3519 * the cluster between those two will be used as the first refblock */
3520 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
3521 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
3522 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
3523 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
3524 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
3526 goto fail_broken_refcounts
;
3529 s
->l1_table_offset
= 3 * s
->cluster_size
;
3531 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ sizeof(uint64_t));
3532 if (!new_reftable
) {
3534 goto fail_broken_refcounts
;
3537 s
->refcount_table_offset
= s
->cluster_size
;
3538 s
->refcount_table_size
= s
->cluster_size
/ sizeof(uint64_t);
3539 s
->max_refcount_table_index
= 0;
3541 g_free(s
->refcount_table
);
3542 s
->refcount_table
= new_reftable
;
3543 new_reftable
= NULL
;
3545 /* Now the in-memory refcount information again corresponds to the on-disk
3546 * information (reftable is empty and no refblocks (the refblock cache is
3547 * empty)); however, this means some clusters (e.g. the image header) are
3548 * referenced, but not refcounted, but the normal qcow2 code assumes that
3549 * the in-memory information is always correct */
3551 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
3553 /* Enter the first refblock into the reftable */
3554 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
3555 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
3556 &rt_entry
, sizeof(rt_entry
));
3558 goto fail_broken_refcounts
;
3560 s
->refcount_table
[0] = 2 * s
->cluster_size
;
3562 s
->free_cluster_index
= 0;
3563 assert(3 + l1_clusters
<= s
->refcount_block_size
);
3564 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
3567 goto fail_broken_refcounts
;
3568 } else if (offset
> 0) {
3569 error_report("First cluster in emptied image is in use");
3573 /* Now finally the in-memory information corresponds to the on-disk
3574 * structures and is correct */
3575 ret
= qcow2_mark_clean(bs
);
3580 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
,
3581 PREALLOC_MODE_OFF
, &local_err
);
3583 error_report_err(local_err
);
3589 fail_broken_refcounts
:
3590 /* The BDS is unusable at this point. If we wanted to make it usable, we
3591 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
3592 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
3593 * again. However, because the functions which could have caused this error
3594 * path to be taken are used by those functions as well, it's very likely
3595 * that that sequence will fail as well. Therefore, just eject the BDS. */
3599 g_free(new_reftable
);
3603 static int qcow2_make_empty(BlockDriverState
*bs
)
3605 BDRVQcow2State
*s
= bs
->opaque
;
3606 uint64_t offset
, end_offset
;
3607 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
3608 int l1_clusters
, ret
= 0;
3610 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
3612 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
3613 3 + l1_clusters
<= s
->refcount_block_size
&&
3614 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
3615 /* The following function only works for qcow2 v3 images (it
3616 * requires the dirty flag) and only as long as there are no
3617 * features that reserve extra clusters (such as snapshots,
3618 * LUKS header, or persistent bitmaps), because it completely
3619 * empties the image. Furthermore, the L1 table and three
3620 * additional clusters (image header, refcount table, one
3621 * refcount block) have to fit inside one refcount block. */
3622 return make_completely_empty(bs
);
3625 /* This fallback code simply discards every active cluster; this is slow,
3626 * but works in all cases */
3627 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3628 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
3629 /* As this function is generally used after committing an external
3630 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
3631 * default action for this kind of discard is to pass the discard,
3632 * which will ideally result in an actually smaller image file, as
3633 * is probably desired. */
3634 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
3635 QCOW2_DISCARD_SNAPSHOT
, true);
3644 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
3646 BDRVQcow2State
*s
= bs
->opaque
;
3649 qemu_co_mutex_lock(&s
->lock
);
3650 ret
= qcow2_cache_write(bs
, s
->l2_table_cache
);
3652 qemu_co_mutex_unlock(&s
->lock
);
3656 if (qcow2_need_accurate_refcounts(s
)) {
3657 ret
= qcow2_cache_write(bs
, s
->refcount_block_cache
);
3659 qemu_co_mutex_unlock(&s
->lock
);
3663 qemu_co_mutex_unlock(&s
->lock
);
3668 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
3671 Error
*local_err
= NULL
;
3672 BlockMeasureInfo
*info
;
3673 uint64_t required
= 0; /* bytes that contribute to required size */
3674 uint64_t virtual_size
; /* disk size as seen by guest */
3675 uint64_t refcount_bits
;
3677 size_t cluster_size
;
3680 PreallocMode prealloc
;
3681 bool has_backing_file
;
3683 /* Parse image creation options */
3684 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, &local_err
);
3689 version
= qcow2_opt_get_version_del(opts
, &local_err
);
3694 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
3699 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
3700 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
3701 PREALLOC_MODE_OFF
, &local_err
);
3707 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
3708 has_backing_file
= !!optstr
;
3711 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
3712 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
3714 /* Check that virtual disk size is valid */
3715 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
3716 cluster_size
/ sizeof(uint64_t));
3717 if (l2_tables
* sizeof(uint64_t) > QCOW_MAX_L1_SIZE
) {
3718 error_setg(&local_err
, "The image size is too large "
3719 "(try using a larger cluster size)");
3723 /* Account for input image */
3725 int64_t ssize
= bdrv_getlength(in_bs
);
3727 error_setg_errno(&local_err
, -ssize
,
3728 "Unable to get image virtual_size");
3732 virtual_size
= ROUND_UP(ssize
, cluster_size
);
3734 if (has_backing_file
) {
3735 /* We don't how much of the backing chain is shared by the input
3736 * image and the new image file. In the worst case the new image's
3737 * backing file has nothing in common with the input image. Be
3738 * conservative and assume all clusters need to be written.
3740 required
= virtual_size
;
3745 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
3748 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
3749 ssize
- offset
, &pnum
, NULL
,
3752 error_setg_errno(&local_err
, -ret
,
3753 "Unable to get block status");
3757 if (ret
& BDRV_BLOCK_ZERO
) {
3758 /* Skip zero regions (safe with no backing file) */
3759 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
3760 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
3761 /* Extend pnum to end of cluster for next iteration */
3762 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
3764 /* Count clusters we've seen */
3765 required
+= offset
% cluster_size
+ pnum
;
3771 /* Take into account preallocation. Nothing special is needed for
3772 * PREALLOC_MODE_METADATA since metadata is always counted.
3774 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
3775 required
= virtual_size
;
3778 info
= g_new(BlockMeasureInfo
, 1);
3779 info
->fully_allocated
=
3780 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
3781 ctz32(refcount_bits
));
3783 /* Remove data clusters that are not required. This overestimates the
3784 * required size because metadata needed for the fully allocated file is
3787 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
3791 error_propagate(errp
, local_err
);
3795 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3797 BDRVQcow2State
*s
= bs
->opaque
;
3798 bdi
->unallocated_blocks_are_zero
= true;
3799 bdi
->cluster_size
= s
->cluster_size
;
3800 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
3804 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
)
3806 BDRVQcow2State
*s
= bs
->opaque
;
3807 ImageInfoSpecific
*spec_info
;
3808 QCryptoBlockInfo
*encrypt_info
= NULL
;
3810 if (s
->crypto
!= NULL
) {
3811 encrypt_info
= qcrypto_block_get_info(s
->crypto
, &error_abort
);
3814 spec_info
= g_new(ImageInfoSpecific
, 1);
3815 *spec_info
= (ImageInfoSpecific
){
3816 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
3817 .u
.qcow2
.data
= g_new(ImageInfoSpecificQCow2
, 1),
3819 if (s
->qcow_version
== 2) {
3820 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
3821 .compat
= g_strdup("0.10"),
3822 .refcount_bits
= s
->refcount_bits
,
3824 } else if (s
->qcow_version
== 3) {
3825 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
3826 .compat
= g_strdup("1.1"),
3827 .lazy_refcounts
= s
->compatible_features
&
3828 QCOW2_COMPAT_LAZY_REFCOUNTS
,
3829 .has_lazy_refcounts
= true,
3830 .corrupt
= s
->incompatible_features
&
3831 QCOW2_INCOMPAT_CORRUPT
,
3832 .has_corrupt
= true,
3833 .refcount_bits
= s
->refcount_bits
,
3836 /* if this assertion fails, this probably means a new version was
3837 * added without having it covered here */
3842 ImageInfoSpecificQCow2Encryption
*qencrypt
=
3843 g_new(ImageInfoSpecificQCow2Encryption
, 1);
3844 switch (encrypt_info
->format
) {
3845 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
3846 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
3847 qencrypt
->u
.aes
= encrypt_info
->u
.qcow
;
3849 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
3850 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
3851 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
3856 /* Since we did shallow copy above, erase any pointers
3857 * in the original info */
3858 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
3859 qapi_free_QCryptoBlockInfo(encrypt_info
);
3861 spec_info
->u
.qcow2
.data
->has_encrypt
= true;
3862 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
3868 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
3871 BDRVQcow2State
*s
= bs
->opaque
;
3873 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
3874 return bs
->drv
->bdrv_co_pwritev(bs
, qcow2_vm_state_offset(s
) + pos
,
3875 qiov
->size
, qiov
, 0);
3878 static int qcow2_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
3881 BDRVQcow2State
*s
= bs
->opaque
;
3883 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
3884 return bs
->drv
->bdrv_co_preadv(bs
, qcow2_vm_state_offset(s
) + pos
,
3885 qiov
->size
, qiov
, 0);
3889 * Downgrades an image's version. To achieve this, any incompatible features
3890 * have to be removed.
3892 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
3893 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
)
3895 BDRVQcow2State
*s
= bs
->opaque
;
3896 int current_version
= s
->qcow_version
;
3899 if (target_version
== current_version
) {
3901 } else if (target_version
> current_version
) {
3903 } else if (target_version
!= 2) {
3907 if (s
->refcount_order
!= 4) {
3908 error_report("compat=0.10 requires refcount_bits=16");
3912 /* clear incompatible features */
3913 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
3914 ret
= qcow2_mark_clean(bs
);
3920 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
3921 * the first place; if that happens nonetheless, returning -ENOTSUP is the
3922 * best thing to do anyway */
3924 if (s
->incompatible_features
) {
3928 /* since we can ignore compatible features, we can set them to 0 as well */
3929 s
->compatible_features
= 0;
3930 /* if lazy refcounts have been used, they have already been fixed through
3931 * clearing the dirty flag */
3933 /* clearing autoclear features is trivial */
3934 s
->autoclear_features
= 0;
3936 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
3941 s
->qcow_version
= target_version
;
3942 ret
= qcow2_update_header(bs
);
3944 s
->qcow_version
= current_version
;
3950 typedef enum Qcow2AmendOperation
{
3951 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
3952 * statically initialized to so that the helper CB can discern the first
3953 * invocation from an operation change */
3954 QCOW2_NO_OPERATION
= 0,
3956 QCOW2_CHANGING_REFCOUNT_ORDER
,
3958 } Qcow2AmendOperation
;
3960 typedef struct Qcow2AmendHelperCBInfo
{
3961 /* The code coordinating the amend operations should only modify
3962 * these four fields; the rest will be managed by the CB */
3963 BlockDriverAmendStatusCB
*original_status_cb
;
3964 void *original_cb_opaque
;
3966 Qcow2AmendOperation current_operation
;
3968 /* Total number of operations to perform (only set once) */
3969 int total_operations
;
3971 /* The following fields are managed by the CB */
3973 /* Number of operations completed */
3974 int operations_completed
;
3976 /* Cumulative offset of all completed operations */
3977 int64_t offset_completed
;
3979 Qcow2AmendOperation last_operation
;
3980 int64_t last_work_size
;
3981 } Qcow2AmendHelperCBInfo
;
3983 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
3984 int64_t operation_offset
,
3985 int64_t operation_work_size
, void *opaque
)
3987 Qcow2AmendHelperCBInfo
*info
= opaque
;
3988 int64_t current_work_size
;
3989 int64_t projected_work_size
;
3991 if (info
->current_operation
!= info
->last_operation
) {
3992 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
3993 info
->offset_completed
+= info
->last_work_size
;
3994 info
->operations_completed
++;
3997 info
->last_operation
= info
->current_operation
;
4000 assert(info
->total_operations
> 0);
4001 assert(info
->operations_completed
< info
->total_operations
);
4003 info
->last_work_size
= operation_work_size
;
4005 current_work_size
= info
->offset_completed
+ operation_work_size
;
4007 /* current_work_size is the total work size for (operations_completed + 1)
4008 * operations (which includes this one), so multiply it by the number of
4009 * operations not covered and divide it by the number of operations
4010 * covered to get a projection for the operations not covered */
4011 projected_work_size
= current_work_size
* (info
->total_operations
-
4012 info
->operations_completed
- 1)
4013 / (info
->operations_completed
+ 1);
4015 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
4016 current_work_size
+ projected_work_size
,
4017 info
->original_cb_opaque
);
4020 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4021 BlockDriverAmendStatusCB
*status_cb
,
4024 BDRVQcow2State
*s
= bs
->opaque
;
4025 int old_version
= s
->qcow_version
, new_version
= old_version
;
4026 uint64_t new_size
= 0;
4027 const char *backing_file
= NULL
, *backing_format
= NULL
;
4028 bool lazy_refcounts
= s
->use_lazy_refcounts
;
4029 const char *compat
= NULL
;
4030 uint64_t cluster_size
= s
->cluster_size
;
4033 int refcount_bits
= s
->refcount_bits
;
4034 Error
*local_err
= NULL
;
4036 QemuOptDesc
*desc
= opts
->list
->desc
;
4037 Qcow2AmendHelperCBInfo helper_cb_info
;
4039 while (desc
&& desc
->name
) {
4040 if (!qemu_opt_find(opts
, desc
->name
)) {
4041 /* only change explicitly defined options */
4046 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
4047 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
4049 /* preserve default */
4050 } else if (!strcmp(compat
, "0.10")) {
4052 } else if (!strcmp(compat
, "1.1")) {
4055 error_report("Unknown compatibility level %s", compat
);
4058 } else if (!strcmp(desc
->name
, BLOCK_OPT_PREALLOC
)) {
4059 error_report("Cannot change preallocation mode");
4061 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
4062 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
4063 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
4064 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4065 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
4066 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4067 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT
)) {
4068 encrypt
= qemu_opt_get_bool(opts
, BLOCK_OPT_ENCRYPT
,
4071 if (encrypt
!= !!s
->crypto
) {
4072 error_report("Changing the encryption flag is not supported");
4075 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT_FORMAT
)) {
4076 encformat
= qcow2_crypt_method_from_format(
4077 qemu_opt_get(opts
, BLOCK_OPT_ENCRYPT_FORMAT
));
4079 if (encformat
!= s
->crypt_method_header
) {
4080 error_report("Changing the encryption format is not supported");
4083 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
4084 error_report("Changing the encryption parameters is not supported");
4086 } else if (!strcmp(desc
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
4087 cluster_size
= qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
,
4089 if (cluster_size
!= s
->cluster_size
) {
4090 error_report("Changing the cluster size is not supported");
4093 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
4094 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
4096 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
4097 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
4100 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
4101 !is_power_of_2(refcount_bits
))
4103 error_report("Refcount width must be a power of two and may "
4104 "not exceed 64 bits");
4108 /* if this point is reached, this probably means a new option was
4109 * added without having it covered here */
4116 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
4117 .original_status_cb
= status_cb
,
4118 .original_cb_opaque
= cb_opaque
,
4119 .total_operations
= (new_version
< old_version
)
4120 + (s
->refcount_bits
!= refcount_bits
)
4123 /* Upgrade first (some features may require compat=1.1) */
4124 if (new_version
> old_version
) {
4125 s
->qcow_version
= new_version
;
4126 ret
= qcow2_update_header(bs
);
4128 s
->qcow_version
= old_version
;
4133 if (s
->refcount_bits
!= refcount_bits
) {
4134 int refcount_order
= ctz32(refcount_bits
);
4136 if (new_version
< 3 && refcount_bits
!= 16) {
4137 error_report("Different refcount widths than 16 bits require "
4138 "compatibility level 1.1 or above (use compat=1.1 or "
4143 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
4144 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
4145 &qcow2_amend_helper_cb
,
4146 &helper_cb_info
, &local_err
);
4148 error_report_err(local_err
);
4153 if (backing_file
|| backing_format
) {
4154 ret
= qcow2_change_backing_file(bs
,
4155 backing_file
?: s
->image_backing_file
,
4156 backing_format
?: s
->image_backing_format
);
4162 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
4163 if (lazy_refcounts
) {
4164 if (new_version
< 3) {
4165 error_report("Lazy refcounts only supported with compatibility "
4166 "level 1.1 and above (use compat=1.1 or greater)");
4169 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
4170 ret
= qcow2_update_header(bs
);
4172 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
4175 s
->use_lazy_refcounts
= true;
4177 /* make image clean first */
4178 ret
= qcow2_mark_clean(bs
);
4182 /* now disallow lazy refcounts */
4183 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
4184 ret
= qcow2_update_header(bs
);
4186 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
4189 s
->use_lazy_refcounts
= false;
4194 BlockBackend
*blk
= blk_new(BLK_PERM_RESIZE
, BLK_PERM_ALL
);
4195 ret
= blk_insert_bs(blk
, bs
, &local_err
);
4197 error_report_err(local_err
);
4202 ret
= blk_truncate(blk
, new_size
, PREALLOC_MODE_OFF
, &local_err
);
4205 error_report_err(local_err
);
4210 /* Downgrade last (so unsupported features can be removed before) */
4211 if (new_version
< old_version
) {
4212 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
4213 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
4224 * If offset or size are negative, respectively, they will not be included in
4225 * the BLOCK_IMAGE_CORRUPTED event emitted.
4226 * fatal will be ignored for read-only BDS; corruptions found there will always
4227 * be considered non-fatal.
4229 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
4230 int64_t size
, const char *message_format
, ...)
4232 BDRVQcow2State
*s
= bs
->opaque
;
4233 const char *node_name
;
4237 fatal
= fatal
&& !bs
->read_only
;
4239 if (s
->signaled_corruption
&&
4240 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
4245 va_start(ap
, message_format
);
4246 message
= g_strdup_vprintf(message_format
, ap
);
4250 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
4251 "corruption events will be suppressed\n", message
);
4253 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
4254 "corruption events will be suppressed\n", message
);
4257 node_name
= bdrv_get_node_name(bs
);
4258 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
4259 *node_name
!= '\0', node_name
,
4260 message
, offset
>= 0, offset
,
4262 fatal
, &error_abort
);
4266 qcow2_mark_corrupt(bs
);
4267 bs
->drv
= NULL
; /* make BDS unusable */
4270 s
->signaled_corruption
= true;
4273 static QemuOptsList qcow2_create_opts
= {
4274 .name
= "qcow2-create-opts",
4275 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
4278 .name
= BLOCK_OPT_SIZE
,
4279 .type
= QEMU_OPT_SIZE
,
4280 .help
= "Virtual disk size"
4283 .name
= BLOCK_OPT_COMPAT_LEVEL
,
4284 .type
= QEMU_OPT_STRING
,
4285 .help
= "Compatibility level (0.10 or 1.1)"
4288 .name
= BLOCK_OPT_BACKING_FILE
,
4289 .type
= QEMU_OPT_STRING
,
4290 .help
= "File name of a base image"
4293 .name
= BLOCK_OPT_BACKING_FMT
,
4294 .type
= QEMU_OPT_STRING
,
4295 .help
= "Image format of the base image"
4298 .name
= BLOCK_OPT_ENCRYPT
,
4299 .type
= QEMU_OPT_BOOL
,
4300 .help
= "Encrypt the image with format 'aes'. (Deprecated "
4301 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)",
4304 .name
= BLOCK_OPT_ENCRYPT_FORMAT
,
4305 .type
= QEMU_OPT_STRING
,
4306 .help
= "Encrypt the image, format choices: 'aes', 'luks'",
4308 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
4309 "ID of secret providing qcow AES key or LUKS passphrase"),
4310 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
4311 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
4312 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
4313 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
4314 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
4315 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
4317 .name
= BLOCK_OPT_CLUSTER_SIZE
,
4318 .type
= QEMU_OPT_SIZE
,
4319 .help
= "qcow2 cluster size",
4320 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
)
4323 .name
= BLOCK_OPT_PREALLOC
,
4324 .type
= QEMU_OPT_STRING
,
4325 .help
= "Preallocation mode (allowed values: off, metadata, "
4329 .name
= BLOCK_OPT_LAZY_REFCOUNTS
,
4330 .type
= QEMU_OPT_BOOL
,
4331 .help
= "Postpone refcount updates",
4332 .def_value_str
= "off"
4335 .name
= BLOCK_OPT_REFCOUNT_BITS
,
4336 .type
= QEMU_OPT_NUMBER
,
4337 .help
= "Width of a reference count entry in bits",
4338 .def_value_str
= "16"
4340 { /* end of list */ }
4344 BlockDriver bdrv_qcow2
= {
4345 .format_name
= "qcow2",
4346 .instance_size
= sizeof(BDRVQcow2State
),
4347 .bdrv_probe
= qcow2_probe
,
4348 .bdrv_open
= qcow2_open
,
4349 .bdrv_close
= qcow2_close
,
4350 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
4351 .bdrv_reopen_commit
= qcow2_reopen_commit
,
4352 .bdrv_reopen_abort
= qcow2_reopen_abort
,
4353 .bdrv_join_options
= qcow2_join_options
,
4354 .bdrv_child_perm
= bdrv_format_default_perms
,
4355 .bdrv_co_create_opts
= qcow2_co_create_opts
,
4356 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
4357 .bdrv_co_block_status
= qcow2_co_block_status
,
4359 .bdrv_co_preadv
= qcow2_co_preadv
,
4360 .bdrv_co_pwritev
= qcow2_co_pwritev
,
4361 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
4363 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
4364 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
4365 .bdrv_truncate
= qcow2_truncate
,
4366 .bdrv_co_pwritev_compressed
= qcow2_co_pwritev_compressed
,
4367 .bdrv_make_empty
= qcow2_make_empty
,
4369 .bdrv_snapshot_create
= qcow2_snapshot_create
,
4370 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
4371 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
4372 .bdrv_snapshot_list
= qcow2_snapshot_list
,
4373 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
4374 .bdrv_measure
= qcow2_measure
,
4375 .bdrv_get_info
= qcow2_get_info
,
4376 .bdrv_get_specific_info
= qcow2_get_specific_info
,
4378 .bdrv_save_vmstate
= qcow2_save_vmstate
,
4379 .bdrv_load_vmstate
= qcow2_load_vmstate
,
4381 .supports_backing
= true,
4382 .bdrv_change_backing_file
= qcow2_change_backing_file
,
4384 .bdrv_refresh_limits
= qcow2_refresh_limits
,
4385 .bdrv_invalidate_cache
= qcow2_invalidate_cache
,
4386 .bdrv_inactivate
= qcow2_inactivate
,
4388 .create_opts
= &qcow2_create_opts
,
4389 .bdrv_check
= qcow2_check
,
4390 .bdrv_amend_options
= qcow2_amend_options
,
4392 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
4393 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
4395 .bdrv_reopen_bitmaps_rw
= qcow2_reopen_bitmaps_rw
,
4396 .bdrv_can_store_new_dirty_bitmap
= qcow2_can_store_new_dirty_bitmap
,
4397 .bdrv_remove_persistent_dirty_bitmap
= qcow2_remove_persistent_dirty_bitmap
,
4400 static void bdrv_qcow2_init(void)
4402 bdrv_register(&bdrv_qcow2
);
4405 block_init(bdrv_qcow2_init
);