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"
30 #include "block/block_int.h"
31 #include "block/qdict.h"
32 #include "sysemu/block-backend.h"
33 #include "qemu/module.h"
35 #include "qemu/error-report.h"
36 #include "qapi/error.h"
37 #include "qapi/qapi-events-block-core.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qstring.h"
41 #include "qemu/option_int.h"
42 #include "qemu/cutils.h"
43 #include "qemu/bswap.h"
44 #include "qapi/qobject-input-visitor.h"
45 #include "qapi/qapi-visit-block-core.h"
47 #include "block/thread-pool.h"
50 Differences with QCOW:
52 - Support for multiple incremental snapshots.
53 - Memory management by reference counts.
54 - Clusters which have a reference count of one have the bit
55 QCOW_OFLAG_COPIED to optimize write performance.
56 - Size of compressed clusters is stored in sectors to reduce bit usage
57 in the cluster offsets.
58 - Support for storing additional data (such as the VM state) in the
60 - If a backing store is used, the cluster size is not constrained
61 (could be backported to QCOW).
62 - L2 tables have always a size of one cluster.
69 } QEMU_PACKED QCowExtension
;
71 #define QCOW2_EXT_MAGIC_END 0
72 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
73 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
74 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
75 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
76 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
78 static int coroutine_fn
79 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
80 uint64_t file_cluster_offset
,
85 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
87 const QCowHeader
*cow_header
= (const void *)buf
;
89 if (buf_size
>= sizeof(QCowHeader
) &&
90 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
91 be32_to_cpu(cow_header
->version
) >= 2)
98 static ssize_t
qcow2_crypto_hdr_read_func(QCryptoBlock
*block
, size_t offset
,
99 uint8_t *buf
, size_t buflen
,
100 void *opaque
, Error
**errp
)
102 BlockDriverState
*bs
= opaque
;
103 BDRVQcow2State
*s
= bs
->opaque
;
106 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
107 error_setg(errp
, "Request for data outside of extension header");
111 ret
= bdrv_pread(bs
->file
,
112 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
114 error_setg_errno(errp
, -ret
, "Could not read encryption header");
121 static ssize_t
qcow2_crypto_hdr_init_func(QCryptoBlock
*block
, size_t headerlen
,
122 void *opaque
, Error
**errp
)
124 BlockDriverState
*bs
= opaque
;
125 BDRVQcow2State
*s
= bs
->opaque
;
129 ret
= qcow2_alloc_clusters(bs
, headerlen
);
131 error_setg_errno(errp
, -ret
,
132 "Cannot allocate cluster for LUKS header size %zu",
137 s
->crypto_header
.length
= headerlen
;
138 s
->crypto_header
.offset
= ret
;
140 /* Zero fill remaining space in cluster so it has predictable
141 * content in case of future spec changes */
142 clusterlen
= size_to_clusters(s
, headerlen
) * s
->cluster_size
;
143 assert(qcow2_pre_write_overlap_check(bs
, 0, ret
, clusterlen
) == 0);
144 ret
= bdrv_pwrite_zeroes(bs
->file
,
146 clusterlen
- headerlen
, 0);
148 error_setg_errno(errp
, -ret
, "Could not zero fill encryption header");
156 static ssize_t
qcow2_crypto_hdr_write_func(QCryptoBlock
*block
, size_t offset
,
157 const uint8_t *buf
, size_t buflen
,
158 void *opaque
, Error
**errp
)
160 BlockDriverState
*bs
= opaque
;
161 BDRVQcow2State
*s
= bs
->opaque
;
164 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
165 error_setg(errp
, "Request for data outside of extension header");
169 ret
= bdrv_pwrite(bs
->file
,
170 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
172 error_setg_errno(errp
, -ret
, "Could not read encryption header");
180 * read qcow2 extension and fill bs
181 * start reading from start_offset
182 * finish reading upon magic of value 0 or when end_offset reached
183 * unknown magic is skipped (future extension this version knows nothing about)
184 * return 0 upon success, non-0 otherwise
186 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
187 uint64_t end_offset
, void **p_feature_table
,
188 int flags
, bool *need_update_header
,
191 BDRVQcow2State
*s
= bs
->opaque
;
195 Qcow2BitmapHeaderExt bitmaps_ext
;
197 if (need_update_header
!= NULL
) {
198 *need_update_header
= false;
202 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
204 offset
= start_offset
;
205 while (offset
< end_offset
) {
209 if (offset
> s
->cluster_size
)
210 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
212 printf("attempting to read extended header in offset %lu\n", offset
);
215 ret
= bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
));
217 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
218 "pread fail from offset %" PRIu64
, offset
);
221 ext
.magic
= be32_to_cpu(ext
.magic
);
222 ext
.len
= be32_to_cpu(ext
.len
);
223 offset
+= sizeof(ext
);
225 printf("ext.magic = 0x%x\n", ext
.magic
);
227 if (offset
> end_offset
|| ext
.len
> end_offset
- offset
) {
228 error_setg(errp
, "Header extension too large");
233 case QCOW2_EXT_MAGIC_END
:
236 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
237 if (ext
.len
>= sizeof(bs
->backing_format
)) {
238 error_setg(errp
, "ERROR: ext_backing_format: len=%" PRIu32
239 " too large (>=%zu)", ext
.len
,
240 sizeof(bs
->backing_format
));
243 ret
= bdrv_pread(bs
->file
, offset
, bs
->backing_format
, ext
.len
);
245 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
246 "Could not read format name");
249 bs
->backing_format
[ext
.len
] = '\0';
250 s
->image_backing_format
= g_strdup(bs
->backing_format
);
252 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
256 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
257 if (p_feature_table
!= NULL
) {
258 void* feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
259 ret
= bdrv_pread(bs
->file
, offset
, feature_table
, ext
.len
);
261 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
262 "Could not read table");
266 *p_feature_table
= feature_table
;
270 case QCOW2_EXT_MAGIC_CRYPTO_HEADER
: {
271 unsigned int cflags
= 0;
272 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
273 error_setg(errp
, "CRYPTO header extension only "
274 "expected with LUKS encryption method");
277 if (ext
.len
!= sizeof(Qcow2CryptoHeaderExtension
)) {
278 error_setg(errp
, "CRYPTO header extension size %u, "
279 "but expected size %zu", ext
.len
,
280 sizeof(Qcow2CryptoHeaderExtension
));
284 ret
= bdrv_pread(bs
->file
, offset
, &s
->crypto_header
, ext
.len
);
286 error_setg_errno(errp
, -ret
,
287 "Unable to read CRYPTO header extension");
290 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
291 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
293 if ((s
->crypto_header
.offset
% s
->cluster_size
) != 0) {
294 error_setg(errp
, "Encryption header offset '%" PRIu64
"' is "
295 "not a multiple of cluster size '%u'",
296 s
->crypto_header
.offset
, s
->cluster_size
);
300 if (flags
& BDRV_O_NO_IO
) {
301 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
303 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
304 qcow2_crypto_hdr_read_func
,
305 bs
, cflags
, 1, errp
);
311 case QCOW2_EXT_MAGIC_BITMAPS
:
312 if (ext
.len
!= sizeof(bitmaps_ext
)) {
313 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
314 "Invalid extension length");
318 if (!(s
->autoclear_features
& QCOW2_AUTOCLEAR_BITMAPS
)) {
319 if (s
->qcow_version
< 3) {
320 /* Let's be a bit more specific */
321 warn_report("This qcow2 v2 image contains bitmaps, but "
322 "they may have been modified by a program "
323 "without persistent bitmap support; so now "
324 "they must all be considered inconsistent");
326 warn_report("a program lacking bitmap support "
327 "modified this file, so all bitmaps are now "
328 "considered inconsistent");
330 error_printf("Some clusters may be leaked, "
331 "run 'qemu-img check -r' on the image "
333 if (need_update_header
!= NULL
) {
334 /* Updating is needed to drop invalid bitmap extension. */
335 *need_update_header
= true;
340 ret
= bdrv_pread(bs
->file
, offset
, &bitmaps_ext
, ext
.len
);
342 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
343 "Could not read ext header");
347 if (bitmaps_ext
.reserved32
!= 0) {
348 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
349 "Reserved field is not zero");
353 bitmaps_ext
.nb_bitmaps
= be32_to_cpu(bitmaps_ext
.nb_bitmaps
);
354 bitmaps_ext
.bitmap_directory_size
=
355 be64_to_cpu(bitmaps_ext
.bitmap_directory_size
);
356 bitmaps_ext
.bitmap_directory_offset
=
357 be64_to_cpu(bitmaps_ext
.bitmap_directory_offset
);
359 if (bitmaps_ext
.nb_bitmaps
> QCOW2_MAX_BITMAPS
) {
361 "bitmaps_ext: Image has %" PRIu32
" bitmaps, "
362 "exceeding the QEMU supported maximum of %d",
363 bitmaps_ext
.nb_bitmaps
, QCOW2_MAX_BITMAPS
);
367 if (bitmaps_ext
.nb_bitmaps
== 0) {
368 error_setg(errp
, "found bitmaps extension with zero bitmaps");
372 if (bitmaps_ext
.bitmap_directory_offset
& (s
->cluster_size
- 1)) {
373 error_setg(errp
, "bitmaps_ext: "
374 "invalid bitmap directory offset");
378 if (bitmaps_ext
.bitmap_directory_size
>
379 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
) {
380 error_setg(errp
, "bitmaps_ext: "
381 "bitmap directory size (%" PRIu64
") exceeds "
382 "the maximum supported size (%d)",
383 bitmaps_ext
.bitmap_directory_size
,
384 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
);
388 s
->nb_bitmaps
= bitmaps_ext
.nb_bitmaps
;
389 s
->bitmap_directory_offset
=
390 bitmaps_ext
.bitmap_directory_offset
;
391 s
->bitmap_directory_size
=
392 bitmaps_ext
.bitmap_directory_size
;
395 printf("Qcow2: Got bitmaps extension: "
396 "offset=%" PRIu64
" nb_bitmaps=%" PRIu32
"\n",
397 s
->bitmap_directory_offset
, s
->nb_bitmaps
);
402 /* unknown magic - save it in case we need to rewrite the header */
403 /* If you add a new feature, make sure to also update the fast
404 * path of qcow2_make_empty() to deal with it. */
406 Qcow2UnknownHeaderExtension
*uext
;
408 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
409 uext
->magic
= ext
.magic
;
411 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
413 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
415 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
416 "Could not read data");
423 offset
+= ((ext
.len
+ 7) & ~7);
429 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
431 BDRVQcow2State
*s
= bs
->opaque
;
432 Qcow2UnknownHeaderExtension
*uext
, *next
;
434 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
435 QLIST_REMOVE(uext
, next
);
440 static void report_unsupported_feature(Error
**errp
, Qcow2Feature
*table
,
443 char *features
= g_strdup("");
446 while (table
&& table
->name
[0] != '\0') {
447 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
448 if (mask
& (1ULL << table
->bit
)) {
450 features
= g_strdup_printf("%s%s%.46s", old
, *old
? ", " : "",
453 mask
&= ~(1ULL << table
->bit
);
461 features
= g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64
,
462 old
, *old
? ", " : "", mask
);
466 error_setg(errp
, "Unsupported qcow2 feature(s): %s", features
);
471 * Sets the dirty bit and flushes afterwards if necessary.
473 * The incompatible_features bit is only set if the image file header was
474 * updated successfully. Therefore it is not required to check the return
475 * value of this function.
477 int qcow2_mark_dirty(BlockDriverState
*bs
)
479 BDRVQcow2State
*s
= bs
->opaque
;
483 assert(s
->qcow_version
>= 3);
485 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
486 return 0; /* already dirty */
489 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
490 ret
= bdrv_pwrite(bs
->file
, offsetof(QCowHeader
, incompatible_features
),
495 ret
= bdrv_flush(bs
->file
->bs
);
500 /* Only treat image as dirty if the header was updated successfully */
501 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
506 * Clears the dirty bit and flushes before if necessary. Only call this
507 * function when there are no pending requests, it does not guard against
508 * concurrent requests dirtying the image.
510 static int qcow2_mark_clean(BlockDriverState
*bs
)
512 BDRVQcow2State
*s
= bs
->opaque
;
514 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
517 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
519 ret
= qcow2_flush_caches(bs
);
524 return qcow2_update_header(bs
);
530 * Marks the image as corrupt.
532 int qcow2_mark_corrupt(BlockDriverState
*bs
)
534 BDRVQcow2State
*s
= bs
->opaque
;
536 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
537 return qcow2_update_header(bs
);
541 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
542 * before if necessary.
544 int qcow2_mark_consistent(BlockDriverState
*bs
)
546 BDRVQcow2State
*s
= bs
->opaque
;
548 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
549 int ret
= qcow2_flush_caches(bs
);
554 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
555 return qcow2_update_header(bs
);
560 static int coroutine_fn
qcow2_co_check_locked(BlockDriverState
*bs
,
561 BdrvCheckResult
*result
,
564 int ret
= qcow2_check_refcounts(bs
, result
, fix
);
569 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
570 ret
= qcow2_mark_clean(bs
);
574 return qcow2_mark_consistent(bs
);
579 static int coroutine_fn
qcow2_co_check(BlockDriverState
*bs
,
580 BdrvCheckResult
*result
,
583 BDRVQcow2State
*s
= bs
->opaque
;
586 qemu_co_mutex_lock(&s
->lock
);
587 ret
= qcow2_co_check_locked(bs
, result
, fix
);
588 qemu_co_mutex_unlock(&s
->lock
);
592 int qcow2_validate_table(BlockDriverState
*bs
, uint64_t offset
,
593 uint64_t entries
, size_t entry_len
,
594 int64_t max_size_bytes
, const char *table_name
,
597 BDRVQcow2State
*s
= bs
->opaque
;
599 if (entries
> max_size_bytes
/ entry_len
) {
600 error_setg(errp
, "%s too large", table_name
);
604 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
605 * because values will be passed to qemu functions taking int64_t. */
606 if ((INT64_MAX
- entries
* entry_len
< offset
) ||
607 (offset_into_cluster(s
, offset
) != 0)) {
608 error_setg(errp
, "%s offset invalid", table_name
);
615 static QemuOptsList qcow2_runtime_opts
= {
617 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
620 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
621 .type
= QEMU_OPT_BOOL
,
622 .help
= "Postpone refcount updates",
625 .name
= QCOW2_OPT_DISCARD_REQUEST
,
626 .type
= QEMU_OPT_BOOL
,
627 .help
= "Pass guest discard requests to the layer below",
630 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
631 .type
= QEMU_OPT_BOOL
,
632 .help
= "Generate discard requests when snapshot related space "
636 .name
= QCOW2_OPT_DISCARD_OTHER
,
637 .type
= QEMU_OPT_BOOL
,
638 .help
= "Generate discard requests when other clusters are freed",
641 .name
= QCOW2_OPT_OVERLAP
,
642 .type
= QEMU_OPT_STRING
,
643 .help
= "Selects which overlap checks to perform from a range of "
644 "templates (none, constant, cached, all)",
647 .name
= QCOW2_OPT_OVERLAP_TEMPLATE
,
648 .type
= QEMU_OPT_STRING
,
649 .help
= "Selects which overlap checks to perform from a range of "
650 "templates (none, constant, cached, all)",
653 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
654 .type
= QEMU_OPT_BOOL
,
655 .help
= "Check for unintended writes into the main qcow2 header",
658 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
659 .type
= QEMU_OPT_BOOL
,
660 .help
= "Check for unintended writes into the active L1 table",
663 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
664 .type
= QEMU_OPT_BOOL
,
665 .help
= "Check for unintended writes into an active L2 table",
668 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
669 .type
= QEMU_OPT_BOOL
,
670 .help
= "Check for unintended writes into the refcount table",
673 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
674 .type
= QEMU_OPT_BOOL
,
675 .help
= "Check for unintended writes into a refcount block",
678 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
679 .type
= QEMU_OPT_BOOL
,
680 .help
= "Check for unintended writes into the snapshot table",
683 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
684 .type
= QEMU_OPT_BOOL
,
685 .help
= "Check for unintended writes into an inactive L1 table",
688 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
689 .type
= QEMU_OPT_BOOL
,
690 .help
= "Check for unintended writes into an inactive L2 table",
693 .name
= QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
694 .type
= QEMU_OPT_BOOL
,
695 .help
= "Check for unintended writes into the bitmap directory",
698 .name
= QCOW2_OPT_CACHE_SIZE
,
699 .type
= QEMU_OPT_SIZE
,
700 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
704 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
705 .type
= QEMU_OPT_SIZE
,
706 .help
= "Maximum L2 table cache size",
709 .name
= QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
710 .type
= QEMU_OPT_SIZE
,
711 .help
= "Size of each entry in the L2 cache",
714 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
715 .type
= QEMU_OPT_SIZE
,
716 .help
= "Maximum refcount block cache size",
719 .name
= QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
720 .type
= QEMU_OPT_NUMBER
,
721 .help
= "Clean unused cache entries after this time (in seconds)",
723 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
724 "ID of secret providing qcow2 AES key or LUKS passphrase"),
725 { /* end of list */ }
729 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
730 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
731 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
732 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
733 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
734 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
735 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
736 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
737 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
738 [QCOW2_OL_BITMAP_DIRECTORY_BITNR
] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
741 static void cache_clean_timer_cb(void *opaque
)
743 BlockDriverState
*bs
= opaque
;
744 BDRVQcow2State
*s
= bs
->opaque
;
745 qcow2_cache_clean_unused(s
->l2_table_cache
);
746 qcow2_cache_clean_unused(s
->refcount_block_cache
);
747 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
748 (int64_t) s
->cache_clean_interval
* 1000);
751 static void cache_clean_timer_init(BlockDriverState
*bs
, AioContext
*context
)
753 BDRVQcow2State
*s
= bs
->opaque
;
754 if (s
->cache_clean_interval
> 0) {
755 s
->cache_clean_timer
= aio_timer_new(context
, QEMU_CLOCK_VIRTUAL
,
756 SCALE_MS
, cache_clean_timer_cb
,
758 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
759 (int64_t) s
->cache_clean_interval
* 1000);
763 static void cache_clean_timer_del(BlockDriverState
*bs
)
765 BDRVQcow2State
*s
= bs
->opaque
;
766 if (s
->cache_clean_timer
) {
767 timer_del(s
->cache_clean_timer
);
768 timer_free(s
->cache_clean_timer
);
769 s
->cache_clean_timer
= NULL
;
773 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
775 cache_clean_timer_del(bs
);
778 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
779 AioContext
*new_context
)
781 cache_clean_timer_init(bs
, new_context
);
784 static void read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
785 uint64_t *l2_cache_size
,
786 uint64_t *l2_cache_entry_size
,
787 uint64_t *refcount_cache_size
, Error
**errp
)
789 BDRVQcow2State
*s
= bs
->opaque
;
790 uint64_t combined_cache_size
, l2_cache_max_setting
;
791 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
792 bool l2_cache_entry_size_set
;
793 int min_refcount_cache
= MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
794 uint64_t virtual_disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
795 uint64_t max_l2_cache
= virtual_disk_size
/ (s
->cluster_size
/ 8);
797 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
798 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
799 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
800 l2_cache_entry_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
);
802 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
803 l2_cache_max_setting
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
,
804 DEFAULT_L2_CACHE_MAX_SIZE
);
805 *refcount_cache_size
= qemu_opt_get_size(opts
,
806 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
808 *l2_cache_entry_size
= qemu_opt_get_size(
809 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
811 *l2_cache_size
= MIN(max_l2_cache
, l2_cache_max_setting
);
813 if (combined_cache_size_set
) {
814 if (l2_cache_size_set
&& refcount_cache_size_set
) {
815 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
816 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
819 } else if (l2_cache_size_set
&&
820 (l2_cache_max_setting
> combined_cache_size
)) {
821 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
822 QCOW2_OPT_CACHE_SIZE
);
824 } else if (*refcount_cache_size
> combined_cache_size
) {
825 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
826 QCOW2_OPT_CACHE_SIZE
);
830 if (l2_cache_size_set
) {
831 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
832 } else if (refcount_cache_size_set
) {
833 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
835 /* Assign as much memory as possible to the L2 cache, and
836 * use the remainder for the refcount cache */
837 if (combined_cache_size
>= max_l2_cache
+ min_refcount_cache
) {
838 *l2_cache_size
= max_l2_cache
;
839 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
841 *refcount_cache_size
=
842 MIN(combined_cache_size
, min_refcount_cache
);
843 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
849 * If the L2 cache is not enough to cover the whole disk then
850 * default to 4KB entries. Smaller entries reduce the cost of
851 * loads and evictions and increase I/O performance.
853 if (*l2_cache_size
< max_l2_cache
&& !l2_cache_entry_size_set
) {
854 *l2_cache_entry_size
= MIN(s
->cluster_size
, 4096);
857 /* l2_cache_size and refcount_cache_size are ensured to have at least
858 * their minimum values in qcow2_update_options_prepare() */
860 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
861 *l2_cache_entry_size
> s
->cluster_size
||
862 !is_power_of_2(*l2_cache_entry_size
)) {
863 error_setg(errp
, "L2 cache entry size must be a power of two "
864 "between %d and the cluster size (%d)",
865 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
870 typedef struct Qcow2ReopenState
{
871 Qcow2Cache
*l2_table_cache
;
872 Qcow2Cache
*refcount_block_cache
;
873 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
874 bool use_lazy_refcounts
;
876 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
877 uint64_t cache_clean_interval
;
878 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
881 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
883 QDict
*options
, int flags
,
886 BDRVQcow2State
*s
= bs
->opaque
;
887 QemuOpts
*opts
= NULL
;
888 const char *opt_overlap_check
, *opt_overlap_check_template
;
889 int overlap_check_template
= 0;
890 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
892 const char *encryptfmt
;
893 QDict
*encryptopts
= NULL
;
894 Error
*local_err
= NULL
;
897 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
898 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
900 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
901 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
903 error_propagate(errp
, local_err
);
908 /* get L2 table/refcount block cache size from command line options */
909 read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
910 &refcount_cache_size
, &local_err
);
912 error_propagate(errp
, local_err
);
917 l2_cache_size
/= l2_cache_entry_size
;
918 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
919 l2_cache_size
= MIN_L2_CACHE_SIZE
;
921 if (l2_cache_size
> INT_MAX
) {
922 error_setg(errp
, "L2 cache size too big");
927 refcount_cache_size
/= s
->cluster_size
;
928 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
929 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
931 if (refcount_cache_size
> INT_MAX
) {
932 error_setg(errp
, "Refcount cache size too big");
937 /* alloc new L2 table/refcount block cache, flush old one */
938 if (s
->l2_table_cache
) {
939 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
941 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
946 if (s
->refcount_block_cache
) {
947 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
949 error_setg_errno(errp
, -ret
,
950 "Failed to flush the refcount block cache");
955 r
->l2_slice_size
= l2_cache_entry_size
/ sizeof(uint64_t);
956 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
957 l2_cache_entry_size
);
958 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
960 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
961 error_setg(errp
, "Could not allocate metadata caches");
966 /* New interval for cache cleanup timer */
967 r
->cache_clean_interval
=
968 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
969 DEFAULT_CACHE_CLEAN_INTERVAL
);
971 if (r
->cache_clean_interval
!= 0) {
972 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
973 " not supported on this host");
978 if (r
->cache_clean_interval
> UINT_MAX
) {
979 error_setg(errp
, "Cache clean interval too big");
984 /* lazy-refcounts; flush if going from enabled to disabled */
985 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
986 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
987 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
988 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
989 "qemu 1.1 compatibility level");
994 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
995 ret
= qcow2_mark_clean(bs
);
997 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
1002 /* Overlap check options */
1003 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
1004 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1005 if (opt_overlap_check_template
&& opt_overlap_check
&&
1006 strcmp(opt_overlap_check_template
, opt_overlap_check
))
1008 error_setg(errp
, "Conflicting values for qcow2 options '"
1009 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1010 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
1014 if (!opt_overlap_check
) {
1015 opt_overlap_check
= opt_overlap_check_template
?: "cached";
1018 if (!strcmp(opt_overlap_check
, "none")) {
1019 overlap_check_template
= 0;
1020 } else if (!strcmp(opt_overlap_check
, "constant")) {
1021 overlap_check_template
= QCOW2_OL_CONSTANT
;
1022 } else if (!strcmp(opt_overlap_check
, "cached")) {
1023 overlap_check_template
= QCOW2_OL_CACHED
;
1024 } else if (!strcmp(opt_overlap_check
, "all")) {
1025 overlap_check_template
= QCOW2_OL_ALL
;
1027 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1028 "'overlap-check'. Allowed are any of the following: "
1029 "none, constant, cached, all", opt_overlap_check
);
1034 r
->overlap_check
= 0;
1035 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1036 /* overlap-check defines a template bitmask, but every flag may be
1037 * overwritten through the associated boolean option */
1039 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1040 overlap_check_template
& (1 << i
)) << i
;
1043 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1044 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1045 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1046 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1047 flags
& BDRV_O_UNMAP
);
1048 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1049 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1050 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1051 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1053 switch (s
->crypt_method_header
) {
1054 case QCOW_CRYPT_NONE
:
1056 error_setg(errp
, "No encryption in image header, but options "
1057 "specified format '%s'", encryptfmt
);
1063 case QCOW_CRYPT_AES
:
1064 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1066 "Header reported 'aes' encryption format but "
1067 "options specify '%s'", encryptfmt
);
1071 qdict_put_str(encryptopts
, "format", "qcow");
1072 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1075 case QCOW_CRYPT_LUKS
:
1076 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1078 "Header reported 'luks' encryption format but "
1079 "options specify '%s'", encryptfmt
);
1083 qdict_put_str(encryptopts
, "format", "luks");
1084 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1088 error_setg(errp
, "Unsupported encryption method %d",
1089 s
->crypt_method_header
);
1092 if (s
->crypt_method_header
!= QCOW_CRYPT_NONE
&& !r
->crypto_opts
) {
1099 qobject_unref(encryptopts
);
1100 qemu_opts_del(opts
);
1105 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1106 Qcow2ReopenState
*r
)
1108 BDRVQcow2State
*s
= bs
->opaque
;
1111 if (s
->l2_table_cache
) {
1112 qcow2_cache_destroy(s
->l2_table_cache
);
1114 if (s
->refcount_block_cache
) {
1115 qcow2_cache_destroy(s
->refcount_block_cache
);
1117 s
->l2_table_cache
= r
->l2_table_cache
;
1118 s
->refcount_block_cache
= r
->refcount_block_cache
;
1119 s
->l2_slice_size
= r
->l2_slice_size
;
1121 s
->overlap_check
= r
->overlap_check
;
1122 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1124 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1125 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1128 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1129 cache_clean_timer_del(bs
);
1130 s
->cache_clean_interval
= r
->cache_clean_interval
;
1131 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1134 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1135 s
->crypto_opts
= r
->crypto_opts
;
1138 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1139 Qcow2ReopenState
*r
)
1141 if (r
->l2_table_cache
) {
1142 qcow2_cache_destroy(r
->l2_table_cache
);
1144 if (r
->refcount_block_cache
) {
1145 qcow2_cache_destroy(r
->refcount_block_cache
);
1147 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1150 static int qcow2_update_options(BlockDriverState
*bs
, QDict
*options
,
1151 int flags
, Error
**errp
)
1153 Qcow2ReopenState r
= {};
1156 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1158 qcow2_update_options_commit(bs
, &r
);
1160 qcow2_update_options_abort(bs
, &r
);
1166 /* Called with s->lock held. */
1167 static int coroutine_fn
qcow2_do_open(BlockDriverState
*bs
, QDict
*options
,
1168 int flags
, Error
**errp
)
1170 BDRVQcow2State
*s
= bs
->opaque
;
1171 unsigned int len
, i
;
1174 Error
*local_err
= NULL
;
1176 uint64_t l1_vm_state_index
;
1177 bool update_header
= false;
1179 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
1181 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1184 header
.magic
= be32_to_cpu(header
.magic
);
1185 header
.version
= be32_to_cpu(header
.version
);
1186 header
.backing_file_offset
= be64_to_cpu(header
.backing_file_offset
);
1187 header
.backing_file_size
= be32_to_cpu(header
.backing_file_size
);
1188 header
.size
= be64_to_cpu(header
.size
);
1189 header
.cluster_bits
= be32_to_cpu(header
.cluster_bits
);
1190 header
.crypt_method
= be32_to_cpu(header
.crypt_method
);
1191 header
.l1_table_offset
= be64_to_cpu(header
.l1_table_offset
);
1192 header
.l1_size
= be32_to_cpu(header
.l1_size
);
1193 header
.refcount_table_offset
= be64_to_cpu(header
.refcount_table_offset
);
1194 header
.refcount_table_clusters
=
1195 be32_to_cpu(header
.refcount_table_clusters
);
1196 header
.snapshots_offset
= be64_to_cpu(header
.snapshots_offset
);
1197 header
.nb_snapshots
= be32_to_cpu(header
.nb_snapshots
);
1199 if (header
.magic
!= QCOW_MAGIC
) {
1200 error_setg(errp
, "Image is not in qcow2 format");
1204 if (header
.version
< 2 || header
.version
> 3) {
1205 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1210 s
->qcow_version
= header
.version
;
1212 /* Initialise cluster size */
1213 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1214 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1215 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1216 header
.cluster_bits
);
1221 s
->cluster_bits
= header
.cluster_bits
;
1222 s
->cluster_size
= 1 << s
->cluster_bits
;
1223 s
->cluster_sectors
= 1 << (s
->cluster_bits
- BDRV_SECTOR_BITS
);
1225 /* Initialise version 3 header fields */
1226 if (header
.version
== 2) {
1227 header
.incompatible_features
= 0;
1228 header
.compatible_features
= 0;
1229 header
.autoclear_features
= 0;
1230 header
.refcount_order
= 4;
1231 header
.header_length
= 72;
1233 header
.incompatible_features
=
1234 be64_to_cpu(header
.incompatible_features
);
1235 header
.compatible_features
= be64_to_cpu(header
.compatible_features
);
1236 header
.autoclear_features
= be64_to_cpu(header
.autoclear_features
);
1237 header
.refcount_order
= be32_to_cpu(header
.refcount_order
);
1238 header
.header_length
= be32_to_cpu(header
.header_length
);
1240 if (header
.header_length
< 104) {
1241 error_setg(errp
, "qcow2 header too short");
1247 if (header
.header_length
> s
->cluster_size
) {
1248 error_setg(errp
, "qcow2 header exceeds cluster size");
1253 if (header
.header_length
> sizeof(header
)) {
1254 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1255 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1256 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
1257 s
->unknown_header_fields_size
);
1259 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1265 if (header
.backing_file_offset
> s
->cluster_size
) {
1266 error_setg(errp
, "Invalid backing file offset");
1271 if (header
.backing_file_offset
) {
1272 ext_end
= header
.backing_file_offset
;
1274 ext_end
= 1 << header
.cluster_bits
;
1277 /* Handle feature bits */
1278 s
->incompatible_features
= header
.incompatible_features
;
1279 s
->compatible_features
= header
.compatible_features
;
1280 s
->autoclear_features
= header
.autoclear_features
;
1282 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1283 void *feature_table
= NULL
;
1284 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1285 &feature_table
, flags
, NULL
, NULL
);
1286 report_unsupported_feature(errp
, feature_table
,
1287 s
->incompatible_features
&
1288 ~QCOW2_INCOMPAT_MASK
);
1290 g_free(feature_table
);
1294 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1295 /* Corrupt images may not be written to unless they are being repaired
1297 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1298 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1305 /* Check support for various header values */
1306 if (header
.refcount_order
> 6) {
1307 error_setg(errp
, "Reference count entry width too large; may not "
1312 s
->refcount_order
= header
.refcount_order
;
1313 s
->refcount_bits
= 1 << s
->refcount_order
;
1314 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1315 s
->refcount_max
+= s
->refcount_max
- 1;
1317 s
->crypt_method_header
= header
.crypt_method
;
1318 if (s
->crypt_method_header
) {
1319 if (bdrv_uses_whitelist() &&
1320 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1322 "Use of AES-CBC encrypted qcow2 images is no longer "
1323 "supported in system emulators");
1324 error_append_hint(errp
,
1325 "You can use 'qemu-img convert' to convert your "
1326 "image to an alternative supported format, such "
1327 "as unencrypted qcow2, or raw with the LUKS "
1328 "format instead.\n");
1333 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1334 s
->crypt_physical_offset
= false;
1336 /* Assuming LUKS and any future crypt methods we
1337 * add will all use physical offsets, due to the
1338 * fact that the alternative is insecure... */
1339 s
->crypt_physical_offset
= true;
1342 bs
->encrypted
= true;
1345 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
1346 s
->l2_size
= 1 << s
->l2_bits
;
1347 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1348 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1349 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1350 bs
->total_sectors
= header
.size
/ BDRV_SECTOR_SIZE
;
1351 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1352 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1353 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1355 s
->refcount_table_offset
= header
.refcount_table_offset
;
1356 s
->refcount_table_size
=
1357 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1359 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1360 error_setg(errp
, "Image does not contain a reference count table");
1365 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1366 header
.refcount_table_clusters
,
1367 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1368 "Reference count table", errp
);
1373 /* The total size in bytes of the snapshot table is checked in
1374 * qcow2_read_snapshots() because the size of each snapshot is
1375 * variable and we don't know it yet.
1376 * Here we only check the offset and number of snapshots. */
1377 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1378 header
.nb_snapshots
,
1379 sizeof(QCowSnapshotHeader
),
1380 sizeof(QCowSnapshotHeader
) * QCOW_MAX_SNAPSHOTS
,
1381 "Snapshot table", errp
);
1386 /* read the level 1 table */
1387 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1388 header
.l1_size
, sizeof(uint64_t),
1389 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1393 s
->l1_size
= header
.l1_size
;
1394 s
->l1_table_offset
= header
.l1_table_offset
;
1396 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1397 if (l1_vm_state_index
> INT_MAX
) {
1398 error_setg(errp
, "Image is too big");
1402 s
->l1_vm_state_index
= l1_vm_state_index
;
1404 /* the L1 table must contain at least enough entries to put
1405 header.size bytes */
1406 if (s
->l1_size
< s
->l1_vm_state_index
) {
1407 error_setg(errp
, "L1 table is too small");
1412 if (s
->l1_size
> 0) {
1413 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
,
1414 ROUND_UP(s
->l1_size
* sizeof(uint64_t), 512));
1415 if (s
->l1_table
== NULL
) {
1416 error_setg(errp
, "Could not allocate L1 table");
1420 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1421 s
->l1_size
* sizeof(uint64_t));
1423 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1426 for(i
= 0;i
< s
->l1_size
; i
++) {
1427 s
->l1_table
[i
] = be64_to_cpu(s
->l1_table
[i
]);
1431 /* Parse driver-specific options */
1432 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1439 ret
= qcow2_refcount_init(bs
);
1441 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1445 QLIST_INIT(&s
->cluster_allocs
);
1446 QTAILQ_INIT(&s
->discards
);
1448 /* read qcow2 extensions */
1449 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1450 flags
, &update_header
, &local_err
)) {
1451 error_propagate(errp
, local_err
);
1456 /* TODO Open external data file */
1457 s
->data_file
= bs
->file
;
1459 /* qcow2_read_extension may have set up the crypto context
1460 * if the crypt method needs a header region, some methods
1461 * don't need header extensions, so must check here
1463 if (s
->crypt_method_header
&& !s
->crypto
) {
1464 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1465 unsigned int cflags
= 0;
1466 if (flags
& BDRV_O_NO_IO
) {
1467 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1469 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1470 NULL
, NULL
, cflags
, 1, errp
);
1475 } else if (!(flags
& BDRV_O_NO_IO
)) {
1476 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1477 s
->crypt_method_header
);
1483 /* read the backing file name */
1484 if (header
.backing_file_offset
!= 0) {
1485 len
= header
.backing_file_size
;
1486 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1487 len
>= sizeof(bs
->backing_file
)) {
1488 error_setg(errp
, "Backing file name too long");
1492 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
1493 bs
->auto_backing_file
, len
);
1495 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1498 bs
->auto_backing_file
[len
] = '\0';
1499 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
),
1500 bs
->auto_backing_file
);
1501 s
->image_backing_file
= g_strdup(bs
->auto_backing_file
);
1504 /* Internal snapshots */
1505 s
->snapshots_offset
= header
.snapshots_offset
;
1506 s
->nb_snapshots
= header
.nb_snapshots
;
1508 ret
= qcow2_read_snapshots(bs
);
1510 error_setg_errno(errp
, -ret
, "Could not read snapshots");
1514 /* Clear unknown autoclear feature bits */
1515 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1517 update_header
&& !bs
->read_only
&& !(flags
& BDRV_O_INACTIVE
);
1518 if (update_header
) {
1519 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1522 /* == Handle persistent dirty bitmaps ==
1524 * We want load dirty bitmaps in three cases:
1526 * 1. Normal open of the disk in active mode, not related to invalidation
1529 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1530 * bitmaps are _not_ migrating through migration channel, i.e.
1531 * 'dirty-bitmaps' capability is disabled.
1533 * 3. Invalidation of source vm after failed or canceled migration.
1534 * This is a very interesting case. There are two possible types of
1537 * A. Stored on inactivation and removed. They should be loaded from the
1540 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1541 * the migration channel (with dirty-bitmaps capability).
1543 * On the other hand, there are two possible sub-cases:
1545 * 3.1 disk was changed by somebody else while were inactive. In this
1546 * case all in-RAM dirty bitmaps (both persistent and not) are
1547 * definitely invalid. And we don't have any method to determine
1550 * Simple and safe thing is to just drop all the bitmaps of type B on
1551 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1553 * On the other hand, resuming source vm, if disk was already changed
1554 * is a bad thing anyway: not only bitmaps, the whole vm state is
1555 * out of sync with disk.
1557 * This means, that user or management tool, who for some reason
1558 * decided to resume source vm, after disk was already changed by
1559 * target vm, should at least drop all dirty bitmaps by hand.
1561 * So, we can ignore this case for now, but TODO: "generation"
1562 * extension for qcow2, to determine, that image was changed after
1563 * last inactivation. And if it is changed, we will drop (or at least
1564 * mark as 'invalid' all the bitmaps of type B, both persistent
1567 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1568 * to disk ('dirty-bitmaps' capability disabled), or not saved
1569 * ('dirty-bitmaps' capability enabled), but we don't need to care
1570 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1571 * and not stored has flag IN_USE=1 in the image and will be skipped
1574 * One remaining possible case when we don't want load bitmaps:
1576 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1577 * will be loaded on invalidation, no needs try loading them before)
1580 if (!(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
)) {
1581 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1582 bool header_updated
= qcow2_load_dirty_bitmaps(bs
, &local_err
);
1584 update_header
= update_header
&& !header_updated
;
1586 if (local_err
!= NULL
) {
1587 error_propagate(errp
, local_err
);
1592 if (update_header
) {
1593 ret
= qcow2_update_header(bs
);
1595 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1600 bs
->supported_zero_flags
= header
.version
>= 3 ? BDRV_REQ_MAY_UNMAP
: 0;
1602 /* Repair image if dirty */
1603 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INACTIVE
)) && !bs
->read_only
&&
1604 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1605 BdrvCheckResult result
= {0};
1607 ret
= qcow2_co_check_locked(bs
, &result
,
1608 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1609 if (ret
< 0 || result
.check_errors
) {
1613 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1620 BdrvCheckResult result
= {0};
1621 qcow2_check_refcounts(bs
, &result
, 0);
1625 qemu_co_queue_init(&s
->compress_wait_queue
);
1630 g_free(s
->unknown_header_fields
);
1631 cleanup_unknown_header_ext(bs
);
1632 qcow2_free_snapshots(bs
);
1633 qcow2_refcount_close(bs
);
1634 qemu_vfree(s
->l1_table
);
1635 /* else pre-write overlap checks in cache_destroy may crash */
1637 cache_clean_timer_del(bs
);
1638 if (s
->l2_table_cache
) {
1639 qcow2_cache_destroy(s
->l2_table_cache
);
1641 if (s
->refcount_block_cache
) {
1642 qcow2_cache_destroy(s
->refcount_block_cache
);
1644 qcrypto_block_free(s
->crypto
);
1645 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1649 typedef struct QCow2OpenCo
{
1650 BlockDriverState
*bs
;
1657 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1659 QCow2OpenCo
*qoc
= opaque
;
1660 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1662 qemu_co_mutex_lock(&s
->lock
);
1663 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, qoc
->errp
);
1664 qemu_co_mutex_unlock(&s
->lock
);
1667 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1670 BDRVQcow2State
*s
= bs
->opaque
;
1679 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_file
,
1685 /* Initialise locks */
1686 qemu_co_mutex_init(&s
->lock
);
1688 if (qemu_in_coroutine()) {
1689 /* From bdrv_co_create. */
1690 qcow2_open_entry(&qoc
);
1692 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1693 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1694 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1699 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1701 BDRVQcow2State
*s
= bs
->opaque
;
1703 if (bs
->encrypted
) {
1704 /* Encryption works on a sector granularity */
1705 bs
->bl
.request_alignment
= qcrypto_block_get_sector_size(s
->crypto
);
1707 bs
->bl
.pwrite_zeroes_alignment
= s
->cluster_size
;
1708 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1711 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1712 BlockReopenQueue
*queue
, Error
**errp
)
1714 Qcow2ReopenState
*r
;
1717 r
= g_new0(Qcow2ReopenState
, 1);
1720 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1721 state
->flags
, errp
);
1726 /* We need to write out any unwritten data if we reopen read-only. */
1727 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1728 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1733 ret
= bdrv_flush(state
->bs
);
1738 ret
= qcow2_mark_clean(state
->bs
);
1747 qcow2_update_options_abort(state
->bs
, r
);
1752 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1754 qcow2_update_options_commit(state
->bs
, state
->opaque
);
1755 g_free(state
->opaque
);
1758 static void qcow2_reopen_abort(BDRVReopenState
*state
)
1760 qcow2_update_options_abort(state
->bs
, state
->opaque
);
1761 g_free(state
->opaque
);
1764 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
1766 bool has_new_overlap_template
=
1767 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
1768 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1769 bool has_new_total_cache_size
=
1770 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
1771 bool has_all_cache_options
;
1773 /* New overlap template overrides all old overlap options */
1774 if (has_new_overlap_template
) {
1775 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
1776 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1777 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
1778 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
1779 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
1780 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
1781 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
1782 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
1783 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
1784 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
1787 /* New total cache size overrides all old options */
1788 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
1789 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
1790 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1793 qdict_join(options
, old_options
, false);
1796 * If after merging all cache size options are set, an old total size is
1797 * overwritten. Do keep all options, however, if all three are new. The
1798 * resulting error message is what we want to happen.
1800 has_all_cache_options
=
1801 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
1802 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
1803 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1805 if (has_all_cache_options
&& !has_new_total_cache_size
) {
1806 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
1810 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
1812 int64_t offset
, int64_t count
,
1813 int64_t *pnum
, int64_t *map
,
1814 BlockDriverState
**file
)
1816 BDRVQcow2State
*s
= bs
->opaque
;
1817 uint64_t cluster_offset
;
1818 int index_in_cluster
, ret
;
1822 bytes
= MIN(INT_MAX
, count
);
1823 qemu_co_mutex_lock(&s
->lock
);
1824 ret
= qcow2_get_cluster_offset(bs
, offset
, &bytes
, &cluster_offset
);
1825 qemu_co_mutex_unlock(&s
->lock
);
1832 if (cluster_offset
!= 0 && ret
!= QCOW2_CLUSTER_COMPRESSED
&&
1834 index_in_cluster
= offset
& (s
->cluster_size
- 1);
1835 *map
= cluster_offset
| index_in_cluster
;
1836 *file
= bs
->file
->bs
;
1837 status
|= BDRV_BLOCK_OFFSET_VALID
;
1839 if (ret
== QCOW2_CLUSTER_ZERO_PLAIN
|| ret
== QCOW2_CLUSTER_ZERO_ALLOC
) {
1840 status
|= BDRV_BLOCK_ZERO
;
1841 } else if (ret
!= QCOW2_CLUSTER_UNALLOCATED
) {
1842 status
|= BDRV_BLOCK_DATA
;
1847 static coroutine_fn
int qcow2_handle_l2meta(BlockDriverState
*bs
,
1848 QCowL2Meta
**pl2meta
,
1852 QCowL2Meta
*l2meta
= *pl2meta
;
1854 while (l2meta
!= NULL
) {
1858 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
1863 qcow2_alloc_cluster_abort(bs
, l2meta
);
1866 /* Take the request off the list of running requests */
1867 if (l2meta
->nb_clusters
!= 0) {
1868 QLIST_REMOVE(l2meta
, next_in_flight
);
1871 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
1873 next
= l2meta
->next
;
1882 static coroutine_fn
int qcow2_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1883 uint64_t bytes
, QEMUIOVector
*qiov
,
1886 BDRVQcow2State
*s
= bs
->opaque
;
1887 int offset_in_cluster
;
1889 unsigned int cur_bytes
; /* number of bytes in current iteration */
1890 uint64_t cluster_offset
= 0;
1891 uint64_t bytes_done
= 0;
1892 QEMUIOVector hd_qiov
;
1893 uint8_t *cluster_data
= NULL
;
1895 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1897 qemu_co_mutex_lock(&s
->lock
);
1899 while (bytes
!= 0) {
1901 /* prepare next request */
1902 cur_bytes
= MIN(bytes
, INT_MAX
);
1904 cur_bytes
= MIN(cur_bytes
,
1905 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1908 ret
= qcow2_get_cluster_offset(bs
, offset
, &cur_bytes
, &cluster_offset
);
1913 offset_in_cluster
= offset_into_cluster(s
, offset
);
1915 qemu_iovec_reset(&hd_qiov
);
1916 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
1919 case QCOW2_CLUSTER_UNALLOCATED
:
1922 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
1923 qemu_co_mutex_unlock(&s
->lock
);
1924 ret
= bdrv_co_preadv(bs
->backing
, offset
, cur_bytes
,
1926 qemu_co_mutex_lock(&s
->lock
);
1931 /* Note: in this case, no need to wait */
1932 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1936 case QCOW2_CLUSTER_ZERO_PLAIN
:
1937 case QCOW2_CLUSTER_ZERO_ALLOC
:
1938 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1941 case QCOW2_CLUSTER_COMPRESSED
:
1942 qemu_co_mutex_unlock(&s
->lock
);
1943 ret
= qcow2_co_preadv_compressed(bs
, cluster_offset
,
1946 qemu_co_mutex_lock(&s
->lock
);
1953 case QCOW2_CLUSTER_NORMAL
:
1954 if ((cluster_offset
& 511) != 0) {
1959 if (bs
->encrypted
) {
1963 * For encrypted images, read everything into a temporary
1964 * contiguous buffer on which the AES functions can work.
1966 if (!cluster_data
) {
1968 qemu_try_blockalign(bs
->file
->bs
,
1969 QCOW_MAX_CRYPT_CLUSTERS
1971 if (cluster_data
== NULL
) {
1977 assert(cur_bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1978 qemu_iovec_reset(&hd_qiov
);
1979 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
1982 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1983 qemu_co_mutex_unlock(&s
->lock
);
1984 ret
= bdrv_co_preadv(bs
->file
,
1985 cluster_offset
+ offset_in_cluster
,
1986 cur_bytes
, &hd_qiov
, 0);
1987 qemu_co_mutex_lock(&s
->lock
);
1991 if (bs
->encrypted
) {
1993 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1994 assert((cur_bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1995 if (qcrypto_block_decrypt(s
->crypto
,
1996 (s
->crypt_physical_offset
?
1997 cluster_offset
+ offset_in_cluster
:
2005 qemu_iovec_from_buf(qiov
, bytes_done
, cluster_data
, cur_bytes
);
2010 g_assert_not_reached();
2016 offset
+= cur_bytes
;
2017 bytes_done
+= cur_bytes
;
2022 qemu_co_mutex_unlock(&s
->lock
);
2024 qemu_iovec_destroy(&hd_qiov
);
2025 qemu_vfree(cluster_data
);
2030 /* Check if it's possible to merge a write request with the writing of
2031 * the data from the COW regions */
2032 static bool merge_cow(uint64_t offset
, unsigned bytes
,
2033 QEMUIOVector
*hd_qiov
, QCowL2Meta
*l2meta
)
2037 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2038 /* If both COW regions are empty then there's nothing to merge */
2039 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
2043 /* The data (middle) region must be immediately after the
2045 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
2049 /* The end region must be immediately after the data (middle)
2051 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
2055 /* Make sure that adding both COW regions to the QEMUIOVector
2056 * does not exceed IOV_MAX */
2057 if (hd_qiov
->niov
> IOV_MAX
- 2) {
2061 m
->data_qiov
= hd_qiov
;
2068 static coroutine_fn
int qcow2_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
2069 uint64_t bytes
, QEMUIOVector
*qiov
,
2072 BDRVQcow2State
*s
= bs
->opaque
;
2073 int offset_in_cluster
;
2075 unsigned int cur_bytes
; /* number of sectors in current iteration */
2076 uint64_t cluster_offset
;
2077 QEMUIOVector hd_qiov
;
2078 uint64_t bytes_done
= 0;
2079 uint8_t *cluster_data
= NULL
;
2080 QCowL2Meta
*l2meta
= NULL
;
2082 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
2084 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
2086 qemu_co_mutex_lock(&s
->lock
);
2088 while (bytes
!= 0) {
2092 trace_qcow2_writev_start_part(qemu_coroutine_self());
2093 offset_in_cluster
= offset_into_cluster(s
, offset
);
2094 cur_bytes
= MIN(bytes
, INT_MAX
);
2095 if (bs
->encrypted
) {
2096 cur_bytes
= MIN(cur_bytes
,
2097 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
2098 - offset_in_cluster
);
2101 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2102 &cluster_offset
, &l2meta
);
2107 assert((cluster_offset
& 511) == 0);
2109 qemu_iovec_reset(&hd_qiov
);
2110 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
2112 if (bs
->encrypted
) {
2114 if (!cluster_data
) {
2115 cluster_data
= qemu_try_blockalign(bs
->file
->bs
,
2116 QCOW_MAX_CRYPT_CLUSTERS
2118 if (cluster_data
== NULL
) {
2124 assert(hd_qiov
.size
<=
2125 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2126 qemu_iovec_to_buf(&hd_qiov
, 0, cluster_data
, hd_qiov
.size
);
2128 if (qcrypto_block_encrypt(s
->crypto
,
2129 (s
->crypt_physical_offset
?
2130 cluster_offset
+ offset_in_cluster
:
2133 cur_bytes
, NULL
) < 0) {
2138 qemu_iovec_reset(&hd_qiov
);
2139 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
2142 ret
= qcow2_pre_write_overlap_check(bs
, 0,
2143 cluster_offset
+ offset_in_cluster
, cur_bytes
);
2148 /* If we need to do COW, check if it's possible to merge the
2149 * writing of the guest data together with that of the COW regions.
2150 * If it's not possible (or not necessary) then write the
2151 * guest data now. */
2152 if (!merge_cow(offset
, cur_bytes
, &hd_qiov
, l2meta
)) {
2153 qemu_co_mutex_unlock(&s
->lock
);
2154 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2155 trace_qcow2_writev_data(qemu_coroutine_self(),
2156 cluster_offset
+ offset_in_cluster
);
2157 ret
= bdrv_co_pwritev(bs
->file
,
2158 cluster_offset
+ offset_in_cluster
,
2159 cur_bytes
, &hd_qiov
, 0);
2160 qemu_co_mutex_lock(&s
->lock
);
2166 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
2172 offset
+= cur_bytes
;
2173 bytes_done
+= cur_bytes
;
2174 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2179 qcow2_handle_l2meta(bs
, &l2meta
, false);
2181 qemu_co_mutex_unlock(&s
->lock
);
2183 qemu_iovec_destroy(&hd_qiov
);
2184 qemu_vfree(cluster_data
);
2185 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2190 static int qcow2_inactivate(BlockDriverState
*bs
)
2192 BDRVQcow2State
*s
= bs
->opaque
;
2193 int ret
, result
= 0;
2194 Error
*local_err
= NULL
;
2196 qcow2_store_persistent_dirty_bitmaps(bs
, &local_err
);
2197 if (local_err
!= NULL
) {
2199 error_reportf_err(local_err
, "Lost persistent bitmaps during "
2200 "inactivation of node '%s': ",
2201 bdrv_get_device_or_node_name(bs
));
2204 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2207 error_report("Failed to flush the L2 table cache: %s",
2211 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2214 error_report("Failed to flush the refcount block cache: %s",
2219 qcow2_mark_clean(bs
);
2225 static void qcow2_close(BlockDriverState
*bs
)
2227 BDRVQcow2State
*s
= bs
->opaque
;
2228 qemu_vfree(s
->l1_table
);
2229 /* else pre-write overlap checks in cache_destroy may crash */
2232 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2233 qcow2_inactivate(bs
);
2236 cache_clean_timer_del(bs
);
2237 qcow2_cache_destroy(s
->l2_table_cache
);
2238 qcow2_cache_destroy(s
->refcount_block_cache
);
2240 qcrypto_block_free(s
->crypto
);
2243 g_free(s
->unknown_header_fields
);
2244 cleanup_unknown_header_ext(bs
);
2246 g_free(s
->image_backing_file
);
2247 g_free(s
->image_backing_format
);
2249 qcow2_refcount_close(bs
);
2250 qcow2_free_snapshots(bs
);
2253 static void coroutine_fn
qcow2_co_invalidate_cache(BlockDriverState
*bs
,
2256 BDRVQcow2State
*s
= bs
->opaque
;
2257 int flags
= s
->flags
;
2258 QCryptoBlock
*crypto
= NULL
;
2260 Error
*local_err
= NULL
;
2264 * Backing files are read-only which makes all of their metadata immutable,
2265 * that means we don't have to worry about reopening them here.
2273 memset(s
, 0, sizeof(BDRVQcow2State
));
2274 options
= qdict_clone_shallow(bs
->options
);
2276 flags
&= ~BDRV_O_INACTIVE
;
2277 qemu_co_mutex_lock(&s
->lock
);
2278 ret
= qcow2_do_open(bs
, options
, flags
, &local_err
);
2279 qemu_co_mutex_unlock(&s
->lock
);
2280 qobject_unref(options
);
2282 error_propagate_prepend(errp
, local_err
,
2283 "Could not reopen qcow2 layer: ");
2286 } else if (ret
< 0) {
2287 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
2295 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2296 size_t len
, size_t buflen
)
2298 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2299 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2301 if (buflen
< ext_len
) {
2305 *ext_backing_fmt
= (QCowExtension
) {
2306 .magic
= cpu_to_be32(magic
),
2307 .len
= cpu_to_be32(len
),
2311 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2318 * Updates the qcow2 header, including the variable length parts of it, i.e.
2319 * the backing file name and all extensions. qcow2 was not designed to allow
2320 * such changes, so if we run out of space (we can only use the first cluster)
2321 * this function may fail.
2323 * Returns 0 on success, -errno in error cases.
2325 int qcow2_update_header(BlockDriverState
*bs
)
2327 BDRVQcow2State
*s
= bs
->opaque
;
2330 size_t buflen
= s
->cluster_size
;
2332 uint64_t total_size
;
2333 uint32_t refcount_table_clusters
;
2334 size_t header_length
;
2335 Qcow2UnknownHeaderExtension
*uext
;
2337 buf
= qemu_blockalign(bs
, buflen
);
2339 /* Header structure */
2340 header
= (QCowHeader
*) buf
;
2342 if (buflen
< sizeof(*header
)) {
2347 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2348 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2349 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2351 *header
= (QCowHeader
) {
2352 /* Version 2 fields */
2353 .magic
= cpu_to_be32(QCOW_MAGIC
),
2354 .version
= cpu_to_be32(s
->qcow_version
),
2355 .backing_file_offset
= 0,
2356 .backing_file_size
= 0,
2357 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2358 .size
= cpu_to_be64(total_size
),
2359 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2360 .l1_size
= cpu_to_be32(s
->l1_size
),
2361 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2362 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2363 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2364 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2365 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2367 /* Version 3 fields */
2368 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2369 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2370 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2371 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2372 .header_length
= cpu_to_be32(header_length
),
2375 /* For older versions, write a shorter header */
2376 switch (s
->qcow_version
) {
2378 ret
= offsetof(QCowHeader
, incompatible_features
);
2381 ret
= sizeof(*header
);
2390 memset(buf
, 0, buflen
);
2392 /* Preserve any unknown field in the header */
2393 if (s
->unknown_header_fields_size
) {
2394 if (buflen
< s
->unknown_header_fields_size
) {
2399 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2400 buf
+= s
->unknown_header_fields_size
;
2401 buflen
-= s
->unknown_header_fields_size
;
2404 /* Backing file format header extension */
2405 if (s
->image_backing_format
) {
2406 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2407 s
->image_backing_format
,
2408 strlen(s
->image_backing_format
),
2418 /* Full disk encryption header pointer extension */
2419 if (s
->crypto_header
.offset
!= 0) {
2420 s
->crypto_header
.offset
= cpu_to_be64(s
->crypto_header
.offset
);
2421 s
->crypto_header
.length
= cpu_to_be64(s
->crypto_header
.length
);
2422 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2423 &s
->crypto_header
, sizeof(s
->crypto_header
),
2425 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
2426 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
2435 if (s
->qcow_version
>= 3) {
2436 Qcow2Feature features
[] = {
2438 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2439 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2440 .name
= "dirty bit",
2443 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2444 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
2445 .name
= "corrupt bit",
2448 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2449 .bit
= QCOW2_INCOMPAT_DATA_FILE_BITNR
,
2450 .name
= "external data file",
2453 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
2454 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
2455 .name
= "lazy refcounts",
2459 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
2460 features
, sizeof(features
), buflen
);
2468 /* Bitmap extension */
2469 if (s
->nb_bitmaps
> 0) {
2470 Qcow2BitmapHeaderExt bitmaps_header
= {
2471 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
2472 .bitmap_directory_size
=
2473 cpu_to_be64(s
->bitmap_directory_size
),
2474 .bitmap_directory_offset
=
2475 cpu_to_be64(s
->bitmap_directory_offset
)
2477 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
2478 &bitmaps_header
, sizeof(bitmaps_header
),
2487 /* Keep unknown header extensions */
2488 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
2489 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
2498 /* End of header extensions */
2499 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
2507 /* Backing file name */
2508 if (s
->image_backing_file
) {
2509 size_t backing_file_len
= strlen(s
->image_backing_file
);
2511 if (buflen
< backing_file_len
) {
2516 /* Using strncpy is ok here, since buf is not NUL-terminated. */
2517 strncpy(buf
, s
->image_backing_file
, buflen
);
2519 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
2520 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
2523 /* Write the new header */
2524 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
2535 static int qcow2_change_backing_file(BlockDriverState
*bs
,
2536 const char *backing_file
, const char *backing_fmt
)
2538 BDRVQcow2State
*s
= bs
->opaque
;
2540 if (backing_file
&& strlen(backing_file
) > 1023) {
2544 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2545 backing_file
?: "");
2546 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
2547 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
2549 g_free(s
->image_backing_file
);
2550 g_free(s
->image_backing_format
);
2552 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
2553 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
2555 return qcow2_update_header(bs
);
2558 static int qcow2_crypt_method_from_format(const char *encryptfmt
)
2560 if (g_str_equal(encryptfmt
, "luks")) {
2561 return QCOW_CRYPT_LUKS
;
2562 } else if (g_str_equal(encryptfmt
, "aes")) {
2563 return QCOW_CRYPT_AES
;
2569 static int qcow2_set_up_encryption(BlockDriverState
*bs
,
2570 QCryptoBlockCreateOptions
*cryptoopts
,
2573 BDRVQcow2State
*s
= bs
->opaque
;
2574 QCryptoBlock
*crypto
= NULL
;
2577 switch (cryptoopts
->format
) {
2578 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
2579 fmt
= QCOW_CRYPT_LUKS
;
2581 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
2582 fmt
= QCOW_CRYPT_AES
;
2585 error_setg(errp
, "Crypto format not supported in qcow2");
2589 s
->crypt_method_header
= fmt
;
2591 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
2592 qcow2_crypto_hdr_init_func
,
2593 qcow2_crypto_hdr_write_func
,
2599 ret
= qcow2_update_header(bs
);
2601 error_setg_errno(errp
, -ret
, "Could not write encryption header");
2607 qcrypto_block_free(crypto
);
2612 * Preallocates metadata structures for data clusters between @offset (in the
2613 * guest disk) and @new_length (which is thus generally the new guest disk
2616 * Returns: 0 on success, -errno on failure.
2618 static int coroutine_fn
preallocate_co(BlockDriverState
*bs
, uint64_t offset
,
2619 uint64_t new_length
)
2622 uint64_t host_offset
= 0;
2623 unsigned int cur_bytes
;
2627 assert(offset
<= new_length
);
2628 bytes
= new_length
- offset
;
2631 cur_bytes
= MIN(bytes
, INT_MAX
);
2632 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2633 &host_offset
, &meta
);
2639 QCowL2Meta
*next
= meta
->next
;
2641 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
2643 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
2644 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
2648 /* There are no dependent requests, but we need to remove our
2649 * request from the list of in-flight requests */
2650 QLIST_REMOVE(meta
, next_in_flight
);
2656 /* TODO Preallocate data if requested */
2659 offset
+= cur_bytes
;
2663 * It is expected that the image file is large enough to actually contain
2664 * all of the allocated clusters (otherwise we get failing reads after
2665 * EOF). Extend the image to the last allocated sector.
2667 if (host_offset
!= 0) {
2669 ret
= bdrv_pwrite(bs
->file
, (host_offset
+ cur_bytes
) - 1,
2679 /* qcow2_refcount_metadata_size:
2680 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2681 * @cluster_size: size of a cluster, in bytes
2682 * @refcount_order: refcount bits power-of-2 exponent
2683 * @generous_increase: allow for the refcount table to be 1.5x as large as it
2686 * Returns: Number of bytes required for refcount blocks and table metadata.
2688 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
2689 int refcount_order
, bool generous_increase
,
2690 uint64_t *refblock_count
)
2693 * Every host cluster is reference-counted, including metadata (even
2694 * refcount metadata is recursively included).
2696 * An accurate formula for the size of refcount metadata size is difficult
2697 * to derive. An easier method of calculation is finding the fixed point
2698 * where no further refcount blocks or table clusters are required to
2699 * reference count every cluster.
2701 int64_t blocks_per_table_cluster
= cluster_size
/ sizeof(uint64_t);
2702 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
2703 int64_t table
= 0; /* number of refcount table clusters */
2704 int64_t blocks
= 0; /* number of refcount block clusters */
2710 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
2711 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
2712 n
= clusters
+ blocks
+ table
;
2714 if (n
== last
&& generous_increase
) {
2715 clusters
+= DIV_ROUND_UP(table
, 2);
2716 n
= 0; /* force another loop */
2717 generous_increase
= false;
2719 } while (n
!= last
);
2721 if (refblock_count
) {
2722 *refblock_count
= blocks
;
2725 return (blocks
+ table
) * cluster_size
;
2729 * qcow2_calc_prealloc_size:
2730 * @total_size: virtual disk size in bytes
2731 * @cluster_size: cluster size in bytes
2732 * @refcount_order: refcount bits power-of-2 exponent
2734 * Returns: Total number of bytes required for the fully allocated image
2735 * (including metadata).
2737 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
2738 size_t cluster_size
,
2741 int64_t meta_size
= 0;
2742 uint64_t nl1e
, nl2e
;
2743 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
2745 /* header: 1 cluster */
2746 meta_size
+= cluster_size
;
2748 /* total size of L2 tables */
2749 nl2e
= aligned_total_size
/ cluster_size
;
2750 nl2e
= ROUND_UP(nl2e
, cluster_size
/ sizeof(uint64_t));
2751 meta_size
+= nl2e
* sizeof(uint64_t);
2753 /* total size of L1 tables */
2754 nl1e
= nl2e
* sizeof(uint64_t) / cluster_size
;
2755 nl1e
= ROUND_UP(nl1e
, cluster_size
/ sizeof(uint64_t));
2756 meta_size
+= nl1e
* sizeof(uint64_t);
2758 /* total size of refcount table and blocks */
2759 meta_size
+= qcow2_refcount_metadata_size(
2760 (meta_size
+ aligned_total_size
) / cluster_size
,
2761 cluster_size
, refcount_order
, false, NULL
);
2763 return meta_size
+ aligned_total_size
;
2766 static bool validate_cluster_size(size_t cluster_size
, Error
**errp
)
2768 int cluster_bits
= ctz32(cluster_size
);
2769 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
2770 (1 << cluster_bits
) != cluster_size
)
2772 error_setg(errp
, "Cluster size must be a power of two between %d and "
2773 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
2779 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, Error
**errp
)
2781 size_t cluster_size
;
2783 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
2784 DEFAULT_CLUSTER_SIZE
);
2785 if (!validate_cluster_size(cluster_size
, errp
)) {
2788 return cluster_size
;
2791 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
2796 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
2798 ret
= 3; /* default */
2799 } else if (!strcmp(buf
, "0.10")) {
2801 } else if (!strcmp(buf
, "1.1")) {
2804 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
2811 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
2814 uint64_t refcount_bits
;
2816 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
2817 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
2818 error_setg(errp
, "Refcount width must be a power of two and may not "
2823 if (version
< 3 && refcount_bits
!= 16) {
2824 error_setg(errp
, "Different refcount widths than 16 bits require "
2825 "compatibility level 1.1 or above (use compat=1.1 or "
2830 return refcount_bits
;
2833 static int coroutine_fn
2834 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
2836 BlockdevCreateOptionsQcow2
*qcow2_opts
;
2840 * Open the image file and write a minimal qcow2 header.
2842 * We keep things simple and start with a zero-sized image. We also
2843 * do without refcount blocks or a L1 table for now. We'll fix the
2844 * inconsistency later.
2846 * We do need a refcount table because growing the refcount table means
2847 * allocating two new refcount blocks - the seconds of which would be at
2848 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2849 * size for any qcow2 image.
2851 BlockBackend
*blk
= NULL
;
2852 BlockDriverState
*bs
= NULL
;
2854 size_t cluster_size
;
2857 uint64_t* refcount_table
;
2858 Error
*local_err
= NULL
;
2861 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
2862 qcow2_opts
= &create_options
->u
.qcow2
;
2864 bs
= bdrv_open_blockdev_ref(qcow2_opts
->file
, errp
);
2869 /* Validate options and set default values */
2870 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
2871 error_setg(errp
, "Image size must be a multiple of 512 bytes");
2876 if (qcow2_opts
->has_version
) {
2877 switch (qcow2_opts
->version
) {
2878 case BLOCKDEV_QCOW2_VERSION_V2
:
2881 case BLOCKDEV_QCOW2_VERSION_V3
:
2885 g_assert_not_reached();
2891 if (qcow2_opts
->has_cluster_size
) {
2892 cluster_size
= qcow2_opts
->cluster_size
;
2894 cluster_size
= DEFAULT_CLUSTER_SIZE
;
2897 if (!validate_cluster_size(cluster_size
, errp
)) {
2902 if (!qcow2_opts
->has_preallocation
) {
2903 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
2905 if (qcow2_opts
->has_backing_file
&&
2906 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
)
2908 error_setg(errp
, "Backing file and preallocation cannot be used at "
2913 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->has_backing_file
) {
2914 error_setg(errp
, "Backing format cannot be used without backing file");
2919 if (!qcow2_opts
->has_lazy_refcounts
) {
2920 qcow2_opts
->lazy_refcounts
= false;
2922 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
2923 error_setg(errp
, "Lazy refcounts only supported with compatibility "
2924 "level 1.1 and above (use version=v3 or greater)");
2929 if (!qcow2_opts
->has_refcount_bits
) {
2930 qcow2_opts
->refcount_bits
= 16;
2932 if (qcow2_opts
->refcount_bits
> 64 ||
2933 !is_power_of_2(qcow2_opts
->refcount_bits
))
2935 error_setg(errp
, "Refcount width must be a power of two and may not "
2940 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
2941 error_setg(errp
, "Different refcount widths than 16 bits require "
2942 "compatibility level 1.1 or above (use version=v3 or "
2947 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
2950 /* Create BlockBackend to write to the image */
2951 blk
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
);
2952 ret
= blk_insert_bs(blk
, bs
, errp
);
2956 blk_set_allow_write_beyond_eof(blk
, true);
2958 /* Clear the protocol layer and preallocate it if necessary */
2959 ret
= blk_truncate(blk
, 0, PREALLOC_MODE_OFF
, errp
);
2964 /* Write the header */
2965 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
2966 header
= g_malloc0(cluster_size
);
2967 *header
= (QCowHeader
) {
2968 .magic
= cpu_to_be32(QCOW_MAGIC
),
2969 .version
= cpu_to_be32(version
),
2970 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
2971 .size
= cpu_to_be64(0),
2972 .l1_table_offset
= cpu_to_be64(0),
2973 .l1_size
= cpu_to_be32(0),
2974 .refcount_table_offset
= cpu_to_be64(cluster_size
),
2975 .refcount_table_clusters
= cpu_to_be32(1),
2976 .refcount_order
= cpu_to_be32(refcount_order
),
2977 .header_length
= cpu_to_be32(sizeof(*header
)),
2980 /* We'll update this to correct value later */
2981 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
2983 if (qcow2_opts
->lazy_refcounts
) {
2984 header
->compatible_features
|=
2985 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
2988 ret
= blk_pwrite(blk
, 0, header
, cluster_size
, 0);
2991 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
2995 /* Write a refcount table with one refcount block */
2996 refcount_table
= g_malloc0(2 * cluster_size
);
2997 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
2998 ret
= blk_pwrite(blk
, cluster_size
, refcount_table
, 2 * cluster_size
, 0);
2999 g_free(refcount_table
);
3002 error_setg_errno(errp
, -ret
, "Could not write refcount table");
3010 * And now open the image and make it consistent first (i.e. increase the
3011 * refcount of the cluster that is occupied by the header and the refcount
3014 options
= qdict_new();
3015 qdict_put_str(options
, "driver", "qcow2");
3016 qdict_put_str(options
, "file", bs
->node_name
);
3017 blk
= blk_new_open(NULL
, NULL
, options
,
3018 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
3021 error_propagate(errp
, local_err
);
3026 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
3028 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
3029 "header and refcount table");
3032 } else if (ret
!= 0) {
3033 error_report("Huh, first cluster in empty image is already in use?");
3037 /* Create a full header (including things like feature table) */
3038 ret
= qcow2_update_header(blk_bs(blk
));
3040 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
3044 /* Okay, now that we have a valid image, let's give it the right size */
3045 ret
= blk_truncate(blk
, qcow2_opts
->size
, qcow2_opts
->preallocation
, errp
);
3047 error_prepend(errp
, "Could not resize image: ");
3051 /* Want a backing file? There you go.*/
3052 if (qcow2_opts
->has_backing_file
) {
3053 const char *backing_format
= NULL
;
3055 if (qcow2_opts
->has_backing_fmt
) {
3056 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
3059 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
3062 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
3063 "with format '%s'", qcow2_opts
->backing_file
,
3069 /* Want encryption? There you go. */
3070 if (qcow2_opts
->has_encrypt
) {
3071 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3080 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3081 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3082 * have to setup decryption context. We're not doing any I/O on the top
3083 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3086 options
= qdict_new();
3087 qdict_put_str(options
, "driver", "qcow2");
3088 qdict_put_str(options
, "file", bs
->node_name
);
3089 blk
= blk_new_open(NULL
, NULL
, options
,
3090 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3093 error_propagate(errp
, local_err
);
3105 static int coroutine_fn
qcow2_co_create_opts(const char *filename
, QemuOpts
*opts
,
3108 BlockdevCreateOptions
*create_options
= NULL
;
3111 BlockDriverState
*bs
= NULL
;
3112 Error
*local_err
= NULL
;
3116 /* Only the keyval visitor supports the dotted syntax needed for
3117 * encryption, so go through a QDict before getting a QAPI type. Ignore
3118 * options meant for the protocol layer so that the visitor doesn't
3120 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3123 /* Handle encryption options */
3124 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3125 if (val
&& !strcmp(val
, "on")) {
3126 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3127 } else if (val
&& !strcmp(val
, "off")) {
3128 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3131 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3132 if (val
&& !strcmp(val
, "aes")) {
3133 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3136 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3137 * version=v2/v3 below. */
3138 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3139 if (val
&& !strcmp(val
, "0.10")) {
3140 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3141 } else if (val
&& !strcmp(val
, "1.1")) {
3142 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3145 /* Change legacy command line options into QMP ones */
3146 static const QDictRenames opt_renames
[] = {
3147 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3148 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3149 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3150 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3151 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3152 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3153 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3157 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3162 /* Create and open the file (protocol layer) */
3163 ret
= bdrv_create_file(filename
, opts
, errp
);
3168 bs
= bdrv_open(filename
, NULL
, NULL
,
3169 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3175 /* Set 'driver' and 'node' options */
3176 qdict_put_str(qdict
, "driver", "qcow2");
3177 qdict_put_str(qdict
, "file", bs
->node_name
);
3179 /* Now get the QAPI type BlockdevCreateOptions */
3180 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
3186 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, &local_err
);
3190 error_propagate(errp
, local_err
);
3195 /* Silently round up size */
3196 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3199 /* Create the qcow2 image (format layer) */
3200 ret
= qcow2_co_create(create_options
, errp
);
3207 qobject_unref(qdict
);
3209 qapi_free_BlockdevCreateOptions(create_options
);
3214 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3219 /* Clamp to image length, before checking status of underlying sectors */
3220 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3221 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3227 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3228 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
== bytes
;
3231 static coroutine_fn
int qcow2_co_pwrite_zeroes(BlockDriverState
*bs
,
3232 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3235 BDRVQcow2State
*s
= bs
->opaque
;
3237 uint32_t head
= offset
% s
->cluster_size
;
3238 uint32_t tail
= (offset
+ bytes
) % s
->cluster_size
;
3240 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3241 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3249 assert(head
+ bytes
<= s
->cluster_size
);
3251 /* check whether remainder of cluster already reads as zero */
3252 if (!(is_zero(bs
, offset
- head
, head
) &&
3253 is_zero(bs
, offset
+ bytes
,
3254 tail
? s
->cluster_size
- tail
: 0))) {
3258 qemu_co_mutex_lock(&s
->lock
);
3259 /* We can have new write after previous check */
3260 offset
= QEMU_ALIGN_DOWN(offset
, s
->cluster_size
);
3261 bytes
= s
->cluster_size
;
3262 nr
= s
->cluster_size
;
3263 ret
= qcow2_get_cluster_offset(bs
, offset
, &nr
, &off
);
3264 if (ret
!= QCOW2_CLUSTER_UNALLOCATED
&&
3265 ret
!= QCOW2_CLUSTER_ZERO_PLAIN
&&
3266 ret
!= QCOW2_CLUSTER_ZERO_ALLOC
) {
3267 qemu_co_mutex_unlock(&s
->lock
);
3271 qemu_co_mutex_lock(&s
->lock
);
3274 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
3276 /* Whatever is left can use real zero clusters */
3277 ret
= qcow2_cluster_zeroize(bs
, offset
, bytes
, flags
);
3278 qemu_co_mutex_unlock(&s
->lock
);
3283 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
3284 int64_t offset
, int bytes
)
3287 BDRVQcow2State
*s
= bs
->opaque
;
3289 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
3290 assert(bytes
< s
->cluster_size
);
3291 /* Ignore partial clusters, except for the special case of the
3292 * complete partial cluster at the end of an unaligned file */
3293 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
3294 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3299 qemu_co_mutex_lock(&s
->lock
);
3300 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
3302 qemu_co_mutex_unlock(&s
->lock
);
3306 static int coroutine_fn
3307 qcow2_co_copy_range_from(BlockDriverState
*bs
,
3308 BdrvChild
*src
, uint64_t src_offset
,
3309 BdrvChild
*dst
, uint64_t dst_offset
,
3310 uint64_t bytes
, BdrvRequestFlags read_flags
,
3311 BdrvRequestFlags write_flags
)
3313 BDRVQcow2State
*s
= bs
->opaque
;
3315 unsigned int cur_bytes
; /* number of bytes in current iteration */
3316 BdrvChild
*child
= NULL
;
3317 BdrvRequestFlags cur_write_flags
;
3319 assert(!bs
->encrypted
);
3320 qemu_co_mutex_lock(&s
->lock
);
3322 while (bytes
!= 0) {
3323 uint64_t copy_offset
= 0;
3324 /* prepare next request */
3325 cur_bytes
= MIN(bytes
, INT_MAX
);
3326 cur_write_flags
= write_flags
;
3328 ret
= qcow2_get_cluster_offset(bs
, src_offset
, &cur_bytes
, ©_offset
);
3334 case QCOW2_CLUSTER_UNALLOCATED
:
3335 if (bs
->backing
&& bs
->backing
->bs
) {
3336 int64_t backing_length
= bdrv_getlength(bs
->backing
->bs
);
3337 if (src_offset
>= backing_length
) {
3338 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3340 child
= bs
->backing
;
3341 cur_bytes
= MIN(cur_bytes
, backing_length
- src_offset
);
3342 copy_offset
= src_offset
;
3345 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3349 case QCOW2_CLUSTER_ZERO_PLAIN
:
3350 case QCOW2_CLUSTER_ZERO_ALLOC
:
3351 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3354 case QCOW2_CLUSTER_COMPRESSED
:
3358 case QCOW2_CLUSTER_NORMAL
:
3360 copy_offset
+= offset_into_cluster(s
, src_offset
);
3361 if ((copy_offset
& 511) != 0) {
3370 qemu_co_mutex_unlock(&s
->lock
);
3371 ret
= bdrv_co_copy_range_from(child
,
3374 cur_bytes
, read_flags
, cur_write_flags
);
3375 qemu_co_mutex_lock(&s
->lock
);
3381 src_offset
+= cur_bytes
;
3382 dst_offset
+= cur_bytes
;
3387 qemu_co_mutex_unlock(&s
->lock
);
3391 static int coroutine_fn
3392 qcow2_co_copy_range_to(BlockDriverState
*bs
,
3393 BdrvChild
*src
, uint64_t src_offset
,
3394 BdrvChild
*dst
, uint64_t dst_offset
,
3395 uint64_t bytes
, BdrvRequestFlags read_flags
,
3396 BdrvRequestFlags write_flags
)
3398 BDRVQcow2State
*s
= bs
->opaque
;
3399 int offset_in_cluster
;
3401 unsigned int cur_bytes
; /* number of sectors in current iteration */
3402 uint64_t cluster_offset
;
3403 QCowL2Meta
*l2meta
= NULL
;
3405 assert(!bs
->encrypted
);
3407 qemu_co_mutex_lock(&s
->lock
);
3409 while (bytes
!= 0) {
3413 offset_in_cluster
= offset_into_cluster(s
, dst_offset
);
3414 cur_bytes
= MIN(bytes
, INT_MAX
);
3417 * If src->bs == dst->bs, we could simply copy by incrementing
3418 * the refcnt, without copying user data.
3419 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
3420 ret
= qcow2_alloc_cluster_offset(bs
, dst_offset
, &cur_bytes
,
3421 &cluster_offset
, &l2meta
);
3426 assert((cluster_offset
& 511) == 0);
3428 ret
= qcow2_pre_write_overlap_check(bs
, 0,
3429 cluster_offset
+ offset_in_cluster
, cur_bytes
);
3434 qemu_co_mutex_unlock(&s
->lock
);
3435 ret
= bdrv_co_copy_range_to(src
, src_offset
,
3437 cluster_offset
+ offset_in_cluster
,
3438 cur_bytes
, read_flags
, write_flags
);
3439 qemu_co_mutex_lock(&s
->lock
);
3444 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
3450 src_offset
+= cur_bytes
;
3451 dst_offset
+= cur_bytes
;
3456 qcow2_handle_l2meta(bs
, &l2meta
, false);
3458 qemu_co_mutex_unlock(&s
->lock
);
3460 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
3465 static int coroutine_fn
qcow2_co_truncate(BlockDriverState
*bs
, int64_t offset
,
3466 PreallocMode prealloc
, Error
**errp
)
3468 BDRVQcow2State
*s
= bs
->opaque
;
3469 uint64_t old_length
;
3470 int64_t new_l1_size
;
3474 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
3475 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
3477 error_setg(errp
, "Unsupported preallocation mode '%s'",
3478 PreallocMode_str(prealloc
));
3483 error_setg(errp
, "The new size must be a multiple of 512");
3487 qemu_co_mutex_lock(&s
->lock
);
3489 /* cannot proceed if image has snapshots */
3490 if (s
->nb_snapshots
) {
3491 error_setg(errp
, "Can't resize an image which has snapshots");
3496 /* cannot proceed if image has bitmaps */
3497 if (s
->nb_bitmaps
) {
3498 /* TODO: resize bitmaps in the image */
3499 error_setg(errp
, "Can't resize an image which has bitmaps");
3504 old_length
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3505 new_l1_size
= size_to_l1(s
, offset
);
3507 if (offset
< old_length
) {
3508 int64_t last_cluster
, old_file_size
;
3509 if (prealloc
!= PREALLOC_MODE_OFF
) {
3511 "Preallocation can't be used for shrinking an image");
3516 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
3517 old_length
- ROUND_UP(offset
,
3519 QCOW2_DISCARD_ALWAYS
, true);
3521 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
3525 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
3527 error_setg_errno(errp
, -ret
,
3528 "Failed to reduce the number of L2 tables");
3532 ret
= qcow2_shrink_reftable(bs
);
3534 error_setg_errno(errp
, -ret
,
3535 "Failed to discard unused refblocks");
3539 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3540 if (old_file_size
< 0) {
3541 error_setg_errno(errp
, -old_file_size
,
3542 "Failed to inquire current file length");
3543 ret
= old_file_size
;
3546 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
3547 if (last_cluster
< 0) {
3548 error_setg_errno(errp
, -last_cluster
,
3549 "Failed to find the last cluster");
3553 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
3554 Error
*local_err
= NULL
;
3556 bdrv_co_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
3557 PREALLOC_MODE_OFF
, &local_err
);
3559 warn_reportf_err(local_err
,
3560 "Failed to truncate the tail of the image: ");
3564 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
3566 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
3572 case PREALLOC_MODE_OFF
:
3575 case PREALLOC_MODE_METADATA
:
3576 ret
= preallocate_co(bs
, old_length
, offset
);
3578 error_setg_errno(errp
, -ret
, "Preallocation failed");
3583 case PREALLOC_MODE_FALLOC
:
3584 case PREALLOC_MODE_FULL
:
3586 int64_t allocation_start
, host_offset
, guest_offset
;
3587 int64_t clusters_allocated
;
3588 int64_t old_file_size
, new_file_size
;
3589 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
3591 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3592 if (old_file_size
< 0) {
3593 error_setg_errno(errp
, -old_file_size
,
3594 "Failed to inquire current file length");
3595 ret
= old_file_size
;
3598 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
3600 nb_new_data_clusters
= DIV_ROUND_UP(offset
- old_length
,
3603 /* This is an overestimation; we will not actually allocate space for
3604 * these in the file but just make sure the new refcount structures are
3605 * able to cover them so we will not have to allocate new refblocks
3606 * while entering the data blocks in the potentially new L2 tables.
3607 * (We do not actually care where the L2 tables are placed. Maybe they
3608 * are already allocated or they can be placed somewhere before
3609 * @old_file_size. It does not matter because they will be fully
3610 * allocated automatically, so they do not need to be covered by the
3611 * preallocation. All that matters is that we will not have to allocate
3612 * new refcount structures for them.) */
3613 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
3614 s
->cluster_size
/ sizeof(uint64_t));
3615 /* The cluster range may not be aligned to L2 boundaries, so add one L2
3616 * table for a potential head/tail */
3619 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
3620 nb_new_data_clusters
+
3623 if (allocation_start
< 0) {
3624 error_setg_errno(errp
, -allocation_start
,
3625 "Failed to resize refcount structures");
3626 ret
= allocation_start
;
3630 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
3631 nb_new_data_clusters
);
3632 if (clusters_allocated
< 0) {
3633 error_setg_errno(errp
, -clusters_allocated
,
3634 "Failed to allocate data clusters");
3635 ret
= clusters_allocated
;
3639 assert(clusters_allocated
== nb_new_data_clusters
);
3641 /* Allocate the data area */
3642 new_file_size
= allocation_start
+
3643 nb_new_data_clusters
* s
->cluster_size
;
3644 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, prealloc
, errp
);
3646 error_prepend(errp
, "Failed to resize underlying file: ");
3647 qcow2_free_clusters(bs
, allocation_start
,
3648 nb_new_data_clusters
* s
->cluster_size
,
3649 QCOW2_DISCARD_OTHER
);
3653 /* Create the necessary L2 entries */
3654 host_offset
= allocation_start
;
3655 guest_offset
= old_length
;
3656 while (nb_new_data_clusters
) {
3657 int64_t nb_clusters
= MIN(
3658 nb_new_data_clusters
,
3659 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
3660 QCowL2Meta allocation
= {
3661 .offset
= guest_offset
,
3662 .alloc_offset
= host_offset
,
3663 .nb_clusters
= nb_clusters
,
3665 qemu_co_queue_init(&allocation
.dependent_requests
);
3667 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
3669 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
3670 qcow2_free_clusters(bs
, host_offset
,
3671 nb_new_data_clusters
* s
->cluster_size
,
3672 QCOW2_DISCARD_OTHER
);
3676 guest_offset
+= nb_clusters
* s
->cluster_size
;
3677 host_offset
+= nb_clusters
* s
->cluster_size
;
3678 nb_new_data_clusters
-= nb_clusters
;
3684 g_assert_not_reached();
3687 if (prealloc
!= PREALLOC_MODE_OFF
) {
3688 /* Flush metadata before actually changing the image size */
3689 ret
= qcow2_write_caches(bs
);
3691 error_setg_errno(errp
, -ret
,
3692 "Failed to flush the preallocated area to disk");
3697 bs
->total_sectors
= offset
/ BDRV_SECTOR_SIZE
;
3699 /* write updated header.size */
3700 offset
= cpu_to_be64(offset
);
3701 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
3702 &offset
, sizeof(uint64_t));
3704 error_setg_errno(errp
, -ret
, "Failed to update the image size");
3708 s
->l1_vm_state_index
= new_l1_size
;
3710 /* Update cache sizes */
3711 options
= qdict_clone_shallow(bs
->options
);
3712 ret
= qcow2_update_options(bs
, options
, s
->flags
, errp
);
3713 qobject_unref(options
);
3719 qemu_co_mutex_unlock(&s
->lock
);
3726 * @dest - destination buffer, @dest_size bytes
3727 * @src - source buffer, @src_size bytes
3729 * Returns: compressed size on success
3730 * -1 destination buffer is not enough to store compressed data
3731 * -2 on any other error
3733 static ssize_t
qcow2_compress(void *dest
, size_t dest_size
,
3734 const void *src
, size_t src_size
)
3739 /* best compression, small window, no zlib header */
3740 memset(&strm
, 0, sizeof(strm
));
3741 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
, Z_DEFLATED
,
3742 -12, 9, Z_DEFAULT_STRATEGY
);
3747 /* strm.next_in is not const in old zlib versions, such as those used on
3748 * OpenBSD/NetBSD, so cast the const away */
3749 strm
.avail_in
= src_size
;
3750 strm
.next_in
= (void *) src
;
3751 strm
.avail_out
= dest_size
;
3752 strm
.next_out
= dest
;
3754 ret
= deflate(&strm
, Z_FINISH
);
3755 if (ret
== Z_STREAM_END
) {
3756 ret
= dest_size
- strm
.avail_out
;
3758 ret
= (ret
== Z_OK
? -1 : -2);
3767 * qcow2_decompress()
3769 * Decompress some data (not more than @src_size bytes) to produce exactly
3772 * @dest - destination buffer, @dest_size bytes
3773 * @src - source buffer, @src_size bytes
3775 * Returns: 0 on success
3778 static ssize_t
qcow2_decompress(void *dest
, size_t dest_size
,
3779 const void *src
, size_t src_size
)
3784 memset(&strm
, 0, sizeof(strm
));
3785 strm
.avail_in
= src_size
;
3786 strm
.next_in
= (void *) src
;
3787 strm
.avail_out
= dest_size
;
3788 strm
.next_out
= dest
;
3790 ret
= inflateInit2(&strm
, -12);
3795 ret
= inflate(&strm
, Z_FINISH
);
3796 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) || strm
.avail_out
!= 0) {
3797 /* We approve Z_BUF_ERROR because we need @dest buffer to be filled, but
3798 * @src buffer may be processed partly (because in qcow2 we know size of
3799 * compressed data with precision of one sector) */
3808 #define MAX_COMPRESS_THREADS 4
3810 typedef ssize_t (*Qcow2CompressFunc
)(void *dest
, size_t dest_size
,
3811 const void *src
, size_t src_size
);
3812 typedef struct Qcow2CompressData
{
3819 Qcow2CompressFunc func
;
3820 } Qcow2CompressData
;
3822 static int qcow2_compress_pool_func(void *opaque
)
3824 Qcow2CompressData
*data
= opaque
;
3826 data
->ret
= data
->func(data
->dest
, data
->dest_size
,
3827 data
->src
, data
->src_size
);
3832 static void qcow2_compress_complete(void *opaque
, int ret
)
3834 qemu_coroutine_enter(opaque
);
3837 static ssize_t coroutine_fn
3838 qcow2_co_do_compress(BlockDriverState
*bs
, void *dest
, size_t dest_size
,
3839 const void *src
, size_t src_size
, Qcow2CompressFunc func
)
3841 BDRVQcow2State
*s
= bs
->opaque
;
3843 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
3844 Qcow2CompressData arg
= {
3846 .dest_size
= dest_size
,
3848 .src_size
= src_size
,
3852 while (s
->nb_compress_threads
>= MAX_COMPRESS_THREADS
) {
3853 qemu_co_queue_wait(&s
->compress_wait_queue
, NULL
);
3856 s
->nb_compress_threads
++;
3857 acb
= thread_pool_submit_aio(pool
, qcow2_compress_pool_func
, &arg
,
3858 qcow2_compress_complete
,
3859 qemu_coroutine_self());
3862 s
->nb_compress_threads
--;
3865 qemu_coroutine_yield();
3866 s
->nb_compress_threads
--;
3867 qemu_co_queue_next(&s
->compress_wait_queue
);
3872 static ssize_t coroutine_fn
3873 qcow2_co_compress(BlockDriverState
*bs
, void *dest
, size_t dest_size
,
3874 const void *src
, size_t src_size
)
3876 return qcow2_co_do_compress(bs
, dest
, dest_size
, src
, src_size
,
3880 static ssize_t coroutine_fn
3881 qcow2_co_decompress(BlockDriverState
*bs
, void *dest
, size_t dest_size
,
3882 const void *src
, size_t src_size
)
3884 return qcow2_co_do_compress(bs
, dest
, dest_size
, src
, src_size
,
3888 /* XXX: put compressed sectors first, then all the cluster aligned
3889 tables to avoid losing bytes in alignment */
3890 static coroutine_fn
int
3891 qcow2_co_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
3892 uint64_t bytes
, QEMUIOVector
*qiov
)
3894 BDRVQcow2State
*s
= bs
->opaque
;
3895 QEMUIOVector hd_qiov
;
3898 uint8_t *buf
, *out_buf
;
3899 int64_t cluster_offset
;
3902 /* align end of file to a sector boundary to ease reading with
3903 sector based I/Os */
3904 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
3905 if (cluster_offset
< 0) {
3906 return cluster_offset
;
3908 return bdrv_co_truncate(bs
->file
, cluster_offset
, PREALLOC_MODE_OFF
,
3912 if (offset_into_cluster(s
, offset
)) {
3916 buf
= qemu_blockalign(bs
, s
->cluster_size
);
3917 if (bytes
!= s
->cluster_size
) {
3918 if (bytes
> s
->cluster_size
||
3919 offset
+ bytes
!= bs
->total_sectors
<< BDRV_SECTOR_BITS
)
3924 /* Zero-pad last write if image size is not cluster aligned */
3925 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
3927 qemu_iovec_to_buf(qiov
, 0, buf
, bytes
);
3929 out_buf
= g_malloc(s
->cluster_size
);
3931 out_len
= qcow2_co_compress(bs
, out_buf
, s
->cluster_size
- 1,
3932 buf
, s
->cluster_size
);
3933 if (out_len
== -2) {
3936 } else if (out_len
== -1) {
3937 /* could not compress: write normal cluster */
3938 ret
= qcow2_co_pwritev(bs
, offset
, bytes
, qiov
, 0);
3945 qemu_co_mutex_lock(&s
->lock
);
3947 qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
);
3948 if (!cluster_offset
) {
3949 qemu_co_mutex_unlock(&s
->lock
);
3953 cluster_offset
&= s
->cluster_offset_mask
;
3955 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
);
3956 qemu_co_mutex_unlock(&s
->lock
);
3961 qemu_iovec_init_buf(&hd_qiov
, out_buf
, out_len
);
3963 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
3964 ret
= bdrv_co_pwritev(bs
->file
, cluster_offset
, out_len
, &hd_qiov
, 0);
3976 static int coroutine_fn
3977 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
3978 uint64_t file_cluster_offset
,
3983 BDRVQcow2State
*s
= bs
->opaque
;
3984 int ret
= 0, csize
, nb_csectors
;
3986 uint8_t *buf
, *out_buf
;
3987 QEMUIOVector local_qiov
;
3988 int offset_in_cluster
= offset_into_cluster(s
, offset
);
3990 coffset
= file_cluster_offset
& s
->cluster_offset_mask
;
3991 nb_csectors
= ((file_cluster_offset
>> s
->csize_shift
) & s
->csize_mask
) + 1;
3992 csize
= nb_csectors
* 512 - (coffset
& 511);
3994 buf
= g_try_malloc(csize
);
3998 qemu_iovec_init_buf(&local_qiov
, buf
, csize
);
4000 out_buf
= qemu_blockalign(bs
, s
->cluster_size
);
4002 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
4003 ret
= bdrv_co_preadv(bs
->file
, coffset
, csize
, &local_qiov
, 0);
4008 if (qcow2_co_decompress(bs
, out_buf
, s
->cluster_size
, buf
, csize
) < 0) {
4013 qemu_iovec_from_buf(qiov
, 0, out_buf
+ offset_in_cluster
, bytes
);
4016 qemu_vfree(out_buf
);
4022 static int make_completely_empty(BlockDriverState
*bs
)
4024 BDRVQcow2State
*s
= bs
->opaque
;
4025 Error
*local_err
= NULL
;
4026 int ret
, l1_clusters
;
4028 uint64_t *new_reftable
= NULL
;
4029 uint64_t rt_entry
, l1_size2
;
4032 uint64_t reftable_offset
;
4033 uint32_t reftable_clusters
;
4034 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
4036 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
4041 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
4046 /* Refcounts will be broken utterly */
4047 ret
= qcow2_mark_dirty(bs
);
4052 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4054 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
4055 l1_size2
= (uint64_t)s
->l1_size
* sizeof(uint64_t);
4057 /* After this call, neither the in-memory nor the on-disk refcount
4058 * information accurately describe the actual references */
4060 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
4061 l1_clusters
* s
->cluster_size
, 0);
4063 goto fail_broken_refcounts
;
4065 memset(s
->l1_table
, 0, l1_size2
);
4067 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
4069 /* Overwrite enough clusters at the beginning of the sectors to place
4070 * the refcount table, a refcount block and the L1 table in; this may
4071 * overwrite parts of the existing refcount and L1 table, which is not
4072 * an issue because the dirty flag is set, complete data loss is in fact
4073 * desired and partial data loss is consequently fine as well */
4074 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
4075 (2 + l1_clusters
) * s
->cluster_size
, 0);
4076 /* This call (even if it failed overall) may have overwritten on-disk
4077 * refcount structures; in that case, the in-memory refcount information
4078 * will probably differ from the on-disk information which makes the BDS
4081 goto fail_broken_refcounts
;
4084 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4085 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
4087 /* "Create" an empty reftable (one cluster) directly after the image
4088 * header and an empty L1 table three clusters after the image header;
4089 * the cluster between those two will be used as the first refblock */
4090 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
4091 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
4092 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
4093 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
4094 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
4096 goto fail_broken_refcounts
;
4099 s
->l1_table_offset
= 3 * s
->cluster_size
;
4101 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ sizeof(uint64_t));
4102 if (!new_reftable
) {
4104 goto fail_broken_refcounts
;
4107 s
->refcount_table_offset
= s
->cluster_size
;
4108 s
->refcount_table_size
= s
->cluster_size
/ sizeof(uint64_t);
4109 s
->max_refcount_table_index
= 0;
4111 g_free(s
->refcount_table
);
4112 s
->refcount_table
= new_reftable
;
4113 new_reftable
= NULL
;
4115 /* Now the in-memory refcount information again corresponds to the on-disk
4116 * information (reftable is empty and no refblocks (the refblock cache is
4117 * empty)); however, this means some clusters (e.g. the image header) are
4118 * referenced, but not refcounted, but the normal qcow2 code assumes that
4119 * the in-memory information is always correct */
4121 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
4123 /* Enter the first refblock into the reftable */
4124 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
4125 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
4126 &rt_entry
, sizeof(rt_entry
));
4128 goto fail_broken_refcounts
;
4130 s
->refcount_table
[0] = 2 * s
->cluster_size
;
4132 s
->free_cluster_index
= 0;
4133 assert(3 + l1_clusters
<= s
->refcount_block_size
);
4134 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
4137 goto fail_broken_refcounts
;
4138 } else if (offset
> 0) {
4139 error_report("First cluster in emptied image is in use");
4143 /* Now finally the in-memory information corresponds to the on-disk
4144 * structures and is correct */
4145 ret
= qcow2_mark_clean(bs
);
4150 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
,
4151 PREALLOC_MODE_OFF
, &local_err
);
4153 error_report_err(local_err
);
4159 fail_broken_refcounts
:
4160 /* The BDS is unusable at this point. If we wanted to make it usable, we
4161 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4162 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4163 * again. However, because the functions which could have caused this error
4164 * path to be taken are used by those functions as well, it's very likely
4165 * that that sequence will fail as well. Therefore, just eject the BDS. */
4169 g_free(new_reftable
);
4173 static int qcow2_make_empty(BlockDriverState
*bs
)
4175 BDRVQcow2State
*s
= bs
->opaque
;
4176 uint64_t offset
, end_offset
;
4177 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
4178 int l1_clusters
, ret
= 0;
4180 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
4182 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
4183 3 + l1_clusters
<= s
->refcount_block_size
&&
4184 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
4185 /* The following function only works for qcow2 v3 images (it
4186 * requires the dirty flag) and only as long as there are no
4187 * features that reserve extra clusters (such as snapshots,
4188 * LUKS header, or persistent bitmaps), because it completely
4189 * empties the image. Furthermore, the L1 table and three
4190 * additional clusters (image header, refcount table, one
4191 * refcount block) have to fit inside one refcount block. */
4192 return make_completely_empty(bs
);
4195 /* This fallback code simply discards every active cluster; this is slow,
4196 * but works in all cases */
4197 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4198 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
4199 /* As this function is generally used after committing an external
4200 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4201 * default action for this kind of discard is to pass the discard,
4202 * which will ideally result in an actually smaller image file, as
4203 * is probably desired. */
4204 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
4205 QCOW2_DISCARD_SNAPSHOT
, true);
4214 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
4216 BDRVQcow2State
*s
= bs
->opaque
;
4219 qemu_co_mutex_lock(&s
->lock
);
4220 ret
= qcow2_write_caches(bs
);
4221 qemu_co_mutex_unlock(&s
->lock
);
4226 static ssize_t
qcow2_measure_crypto_hdr_init_func(QCryptoBlock
*block
,
4227 size_t headerlen
, void *opaque
, Error
**errp
)
4229 size_t *headerlenp
= opaque
;
4231 /* Stash away the payload size */
4232 *headerlenp
= headerlen
;
4236 static ssize_t
qcow2_measure_crypto_hdr_write_func(QCryptoBlock
*block
,
4237 size_t offset
, const uint8_t *buf
, size_t buflen
,
4238 void *opaque
, Error
**errp
)
4240 /* Discard the bytes, we're not actually writing to an image */
4244 /* Determine the number of bytes for the LUKS payload */
4245 static bool qcow2_measure_luks_headerlen(QemuOpts
*opts
, size_t *len
,
4249 QDict
*cryptoopts_qdict
;
4250 QCryptoBlockCreateOptions
*cryptoopts
;
4251 QCryptoBlock
*crypto
;
4253 /* Extract "encrypt." options into a qdict */
4254 opts_qdict
= qemu_opts_to_qdict(opts
, NULL
);
4255 qdict_extract_subqdict(opts_qdict
, &cryptoopts_qdict
, "encrypt.");
4256 qobject_unref(opts_qdict
);
4258 /* Build QCryptoBlockCreateOptions object from qdict */
4259 qdict_put_str(cryptoopts_qdict
, "format", "luks");
4260 cryptoopts
= block_crypto_create_opts_init(cryptoopts_qdict
, errp
);
4261 qobject_unref(cryptoopts_qdict
);
4266 /* Fake LUKS creation in order to determine the payload size */
4267 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
4268 qcow2_measure_crypto_hdr_init_func
,
4269 qcow2_measure_crypto_hdr_write_func
,
4271 qapi_free_QCryptoBlockCreateOptions(cryptoopts
);
4276 qcrypto_block_free(crypto
);
4280 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
4283 Error
*local_err
= NULL
;
4284 BlockMeasureInfo
*info
;
4285 uint64_t required
= 0; /* bytes that contribute to required size */
4286 uint64_t virtual_size
; /* disk size as seen by guest */
4287 uint64_t refcount_bits
;
4289 uint64_t luks_payload_size
= 0;
4290 size_t cluster_size
;
4293 PreallocMode prealloc
;
4294 bool has_backing_file
;
4297 /* Parse image creation options */
4298 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, &local_err
);
4303 version
= qcow2_opt_get_version_del(opts
, &local_err
);
4308 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
4313 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
4314 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
4315 PREALLOC_MODE_OFF
, &local_err
);
4321 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
4322 has_backing_file
= !!optstr
;
4325 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
4326 has_luks
= optstr
&& strcmp(optstr
, "luks") == 0;
4332 if (!qcow2_measure_luks_headerlen(opts
, &headerlen
, &local_err
)) {
4336 luks_payload_size
= ROUND_UP(headerlen
, cluster_size
);
4339 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
4340 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
4342 /* Check that virtual disk size is valid */
4343 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
4344 cluster_size
/ sizeof(uint64_t));
4345 if (l2_tables
* sizeof(uint64_t) > QCOW_MAX_L1_SIZE
) {
4346 error_setg(&local_err
, "The image size is too large "
4347 "(try using a larger cluster size)");
4351 /* Account for input image */
4353 int64_t ssize
= bdrv_getlength(in_bs
);
4355 error_setg_errno(&local_err
, -ssize
,
4356 "Unable to get image virtual_size");
4360 virtual_size
= ROUND_UP(ssize
, cluster_size
);
4362 if (has_backing_file
) {
4363 /* We don't how much of the backing chain is shared by the input
4364 * image and the new image file. In the worst case the new image's
4365 * backing file has nothing in common with the input image. Be
4366 * conservative and assume all clusters need to be written.
4368 required
= virtual_size
;
4373 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
4376 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
4377 ssize
- offset
, &pnum
, NULL
,
4380 error_setg_errno(&local_err
, -ret
,
4381 "Unable to get block status");
4385 if (ret
& BDRV_BLOCK_ZERO
) {
4386 /* Skip zero regions (safe with no backing file) */
4387 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
4388 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
4389 /* Extend pnum to end of cluster for next iteration */
4390 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
4392 /* Count clusters we've seen */
4393 required
+= offset
% cluster_size
+ pnum
;
4399 /* Take into account preallocation. Nothing special is needed for
4400 * PREALLOC_MODE_METADATA since metadata is always counted.
4402 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
4403 required
= virtual_size
;
4406 info
= g_new(BlockMeasureInfo
, 1);
4407 info
->fully_allocated
=
4408 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
4409 ctz32(refcount_bits
)) + luks_payload_size
;
4411 /* Remove data clusters that are not required. This overestimates the
4412 * required size because metadata needed for the fully allocated file is
4415 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
4419 error_propagate(errp
, local_err
);
4423 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4425 BDRVQcow2State
*s
= bs
->opaque
;
4426 bdi
->unallocated_blocks_are_zero
= true;
4427 bdi
->cluster_size
= s
->cluster_size
;
4428 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
4432 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
,
4435 BDRVQcow2State
*s
= bs
->opaque
;
4436 ImageInfoSpecific
*spec_info
;
4437 QCryptoBlockInfo
*encrypt_info
= NULL
;
4438 Error
*local_err
= NULL
;
4440 if (s
->crypto
!= NULL
) {
4441 encrypt_info
= qcrypto_block_get_info(s
->crypto
, &local_err
);
4443 error_propagate(errp
, local_err
);
4448 spec_info
= g_new(ImageInfoSpecific
, 1);
4449 *spec_info
= (ImageInfoSpecific
){
4450 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
4451 .u
.qcow2
.data
= g_new0(ImageInfoSpecificQCow2
, 1),
4453 if (s
->qcow_version
== 2) {
4454 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
4455 .compat
= g_strdup("0.10"),
4456 .refcount_bits
= s
->refcount_bits
,
4458 } else if (s
->qcow_version
== 3) {
4459 Qcow2BitmapInfoList
*bitmaps
;
4460 bitmaps
= qcow2_get_bitmap_info_list(bs
, &local_err
);
4462 error_propagate(errp
, local_err
);
4463 qapi_free_ImageInfoSpecific(spec_info
);
4466 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
4467 .compat
= g_strdup("1.1"),
4468 .lazy_refcounts
= s
->compatible_features
&
4469 QCOW2_COMPAT_LAZY_REFCOUNTS
,
4470 .has_lazy_refcounts
= true,
4471 .corrupt
= s
->incompatible_features
&
4472 QCOW2_INCOMPAT_CORRUPT
,
4473 .has_corrupt
= true,
4474 .refcount_bits
= s
->refcount_bits
,
4475 .has_bitmaps
= !!bitmaps
,
4479 /* if this assertion fails, this probably means a new version was
4480 * added without having it covered here */
4485 ImageInfoSpecificQCow2Encryption
*qencrypt
=
4486 g_new(ImageInfoSpecificQCow2Encryption
, 1);
4487 switch (encrypt_info
->format
) {
4488 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
4489 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
4491 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
4492 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
4493 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
4498 /* Since we did shallow copy above, erase any pointers
4499 * in the original info */
4500 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
4501 qapi_free_QCryptoBlockInfo(encrypt_info
);
4503 spec_info
->u
.qcow2
.data
->has_encrypt
= true;
4504 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
4510 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
4513 BDRVQcow2State
*s
= bs
->opaque
;
4515 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
4516 return bs
->drv
->bdrv_co_pwritev(bs
, qcow2_vm_state_offset(s
) + pos
,
4517 qiov
->size
, qiov
, 0);
4520 static int qcow2_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
4523 BDRVQcow2State
*s
= bs
->opaque
;
4525 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
4526 return bs
->drv
->bdrv_co_preadv(bs
, qcow2_vm_state_offset(s
) + pos
,
4527 qiov
->size
, qiov
, 0);
4531 * Downgrades an image's version. To achieve this, any incompatible features
4532 * have to be removed.
4534 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
4535 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
4538 BDRVQcow2State
*s
= bs
->opaque
;
4539 int current_version
= s
->qcow_version
;
4542 /* This is qcow2_downgrade(), not qcow2_upgrade() */
4543 assert(target_version
< current_version
);
4545 /* There are no other versions (now) that you can downgrade to */
4546 assert(target_version
== 2);
4548 if (s
->refcount_order
!= 4) {
4549 error_setg(errp
, "compat=0.10 requires refcount_bits=16");
4553 /* clear incompatible features */
4554 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
4555 ret
= qcow2_mark_clean(bs
);
4557 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
4562 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
4563 * the first place; if that happens nonetheless, returning -ENOTSUP is the
4564 * best thing to do anyway */
4566 if (s
->incompatible_features
) {
4567 error_setg(errp
, "Cannot downgrade an image with incompatible features "
4568 "%#" PRIx64
" set", s
->incompatible_features
);
4572 /* since we can ignore compatible features, we can set them to 0 as well */
4573 s
->compatible_features
= 0;
4574 /* if lazy refcounts have been used, they have already been fixed through
4575 * clearing the dirty flag */
4577 /* clearing autoclear features is trivial */
4578 s
->autoclear_features
= 0;
4580 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
4582 error_setg_errno(errp
, -ret
, "Failed to turn zero into data clusters");
4586 s
->qcow_version
= target_version
;
4587 ret
= qcow2_update_header(bs
);
4589 s
->qcow_version
= current_version
;
4590 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4596 typedef enum Qcow2AmendOperation
{
4597 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
4598 * statically initialized to so that the helper CB can discern the first
4599 * invocation from an operation change */
4600 QCOW2_NO_OPERATION
= 0,
4602 QCOW2_CHANGING_REFCOUNT_ORDER
,
4604 } Qcow2AmendOperation
;
4606 typedef struct Qcow2AmendHelperCBInfo
{
4607 /* The code coordinating the amend operations should only modify
4608 * these four fields; the rest will be managed by the CB */
4609 BlockDriverAmendStatusCB
*original_status_cb
;
4610 void *original_cb_opaque
;
4612 Qcow2AmendOperation current_operation
;
4614 /* Total number of operations to perform (only set once) */
4615 int total_operations
;
4617 /* The following fields are managed by the CB */
4619 /* Number of operations completed */
4620 int operations_completed
;
4622 /* Cumulative offset of all completed operations */
4623 int64_t offset_completed
;
4625 Qcow2AmendOperation last_operation
;
4626 int64_t last_work_size
;
4627 } Qcow2AmendHelperCBInfo
;
4629 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
4630 int64_t operation_offset
,
4631 int64_t operation_work_size
, void *opaque
)
4633 Qcow2AmendHelperCBInfo
*info
= opaque
;
4634 int64_t current_work_size
;
4635 int64_t projected_work_size
;
4637 if (info
->current_operation
!= info
->last_operation
) {
4638 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
4639 info
->offset_completed
+= info
->last_work_size
;
4640 info
->operations_completed
++;
4643 info
->last_operation
= info
->current_operation
;
4646 assert(info
->total_operations
> 0);
4647 assert(info
->operations_completed
< info
->total_operations
);
4649 info
->last_work_size
= operation_work_size
;
4651 current_work_size
= info
->offset_completed
+ operation_work_size
;
4653 /* current_work_size is the total work size for (operations_completed + 1)
4654 * operations (which includes this one), so multiply it by the number of
4655 * operations not covered and divide it by the number of operations
4656 * covered to get a projection for the operations not covered */
4657 projected_work_size
= current_work_size
* (info
->total_operations
-
4658 info
->operations_completed
- 1)
4659 / (info
->operations_completed
+ 1);
4661 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
4662 current_work_size
+ projected_work_size
,
4663 info
->original_cb_opaque
);
4666 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4667 BlockDriverAmendStatusCB
*status_cb
,
4671 BDRVQcow2State
*s
= bs
->opaque
;
4672 int old_version
= s
->qcow_version
, new_version
= old_version
;
4673 uint64_t new_size
= 0;
4674 const char *backing_file
= NULL
, *backing_format
= NULL
;
4675 bool lazy_refcounts
= s
->use_lazy_refcounts
;
4676 const char *compat
= NULL
;
4677 uint64_t cluster_size
= s
->cluster_size
;
4680 int refcount_bits
= s
->refcount_bits
;
4682 QemuOptDesc
*desc
= opts
->list
->desc
;
4683 Qcow2AmendHelperCBInfo helper_cb_info
;
4685 while (desc
&& desc
->name
) {
4686 if (!qemu_opt_find(opts
, desc
->name
)) {
4687 /* only change explicitly defined options */
4692 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
4693 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
4695 /* preserve default */
4696 } else if (!strcmp(compat
, "0.10")) {
4698 } else if (!strcmp(compat
, "1.1")) {
4701 error_setg(errp
, "Unknown compatibility level %s", compat
);
4704 } else if (!strcmp(desc
->name
, BLOCK_OPT_PREALLOC
)) {
4705 error_setg(errp
, "Cannot change preallocation mode");
4707 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
4708 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
4709 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
4710 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4711 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
4712 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4713 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT
)) {
4714 encrypt
= qemu_opt_get_bool(opts
, BLOCK_OPT_ENCRYPT
,
4717 if (encrypt
!= !!s
->crypto
) {
4719 "Changing the encryption flag is not supported");
4722 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT_FORMAT
)) {
4723 encformat
= qcow2_crypt_method_from_format(
4724 qemu_opt_get(opts
, BLOCK_OPT_ENCRYPT_FORMAT
));
4726 if (encformat
!= s
->crypt_method_header
) {
4728 "Changing the encryption format is not supported");
4731 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
4733 "Changing the encryption parameters is not supported");
4735 } else if (!strcmp(desc
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
4736 cluster_size
= qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
,
4738 if (cluster_size
!= s
->cluster_size
) {
4739 error_setg(errp
, "Changing the cluster size is not supported");
4742 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
4743 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
4745 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
4746 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
4749 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
4750 !is_power_of_2(refcount_bits
))
4752 error_setg(errp
, "Refcount width must be a power of two and "
4753 "may not exceed 64 bits");
4757 /* if this point is reached, this probably means a new option was
4758 * added without having it covered here */
4765 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
4766 .original_status_cb
= status_cb
,
4767 .original_cb_opaque
= cb_opaque
,
4768 .total_operations
= (new_version
< old_version
)
4769 + (s
->refcount_bits
!= refcount_bits
)
4772 /* Upgrade first (some features may require compat=1.1) */
4773 if (new_version
> old_version
) {
4774 s
->qcow_version
= new_version
;
4775 ret
= qcow2_update_header(bs
);
4777 s
->qcow_version
= old_version
;
4778 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4783 if (s
->refcount_bits
!= refcount_bits
) {
4784 int refcount_order
= ctz32(refcount_bits
);
4786 if (new_version
< 3 && refcount_bits
!= 16) {
4787 error_setg(errp
, "Refcount widths other than 16 bits require "
4788 "compatibility level 1.1 or above (use compat=1.1 or "
4793 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
4794 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
4795 &qcow2_amend_helper_cb
,
4796 &helper_cb_info
, errp
);
4802 if (backing_file
|| backing_format
) {
4803 ret
= qcow2_change_backing_file(bs
,
4804 backing_file
?: s
->image_backing_file
,
4805 backing_format
?: s
->image_backing_format
);
4807 error_setg_errno(errp
, -ret
, "Failed to change the backing file");
4812 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
4813 if (lazy_refcounts
) {
4814 if (new_version
< 3) {
4815 error_setg(errp
, "Lazy refcounts only supported with "
4816 "compatibility level 1.1 and above (use compat=1.1 "
4820 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
4821 ret
= qcow2_update_header(bs
);
4823 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
4824 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4827 s
->use_lazy_refcounts
= true;
4829 /* make image clean first */
4830 ret
= qcow2_mark_clean(bs
);
4832 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
4835 /* now disallow lazy refcounts */
4836 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
4837 ret
= qcow2_update_header(bs
);
4839 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
4840 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4843 s
->use_lazy_refcounts
= false;
4848 BlockBackend
*blk
= blk_new(BLK_PERM_RESIZE
, BLK_PERM_ALL
);
4849 ret
= blk_insert_bs(blk
, bs
, errp
);
4855 ret
= blk_truncate(blk
, new_size
, PREALLOC_MODE_OFF
, errp
);
4862 /* Downgrade last (so unsupported features can be removed before) */
4863 if (new_version
< old_version
) {
4864 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
4865 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
4866 &helper_cb_info
, errp
);
4876 * If offset or size are negative, respectively, they will not be included in
4877 * the BLOCK_IMAGE_CORRUPTED event emitted.
4878 * fatal will be ignored for read-only BDS; corruptions found there will always
4879 * be considered non-fatal.
4881 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
4882 int64_t size
, const char *message_format
, ...)
4884 BDRVQcow2State
*s
= bs
->opaque
;
4885 const char *node_name
;
4889 fatal
= fatal
&& bdrv_is_writable(bs
);
4891 if (s
->signaled_corruption
&&
4892 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
4897 va_start(ap
, message_format
);
4898 message
= g_strdup_vprintf(message_format
, ap
);
4902 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
4903 "corruption events will be suppressed\n", message
);
4905 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
4906 "corruption events will be suppressed\n", message
);
4909 node_name
= bdrv_get_node_name(bs
);
4910 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
4911 *node_name
!= '\0', node_name
,
4912 message
, offset
>= 0, offset
,
4918 qcow2_mark_corrupt(bs
);
4919 bs
->drv
= NULL
; /* make BDS unusable */
4922 s
->signaled_corruption
= true;
4925 static QemuOptsList qcow2_create_opts
= {
4926 .name
= "qcow2-create-opts",
4927 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
4930 .name
= BLOCK_OPT_SIZE
,
4931 .type
= QEMU_OPT_SIZE
,
4932 .help
= "Virtual disk size"
4935 .name
= BLOCK_OPT_COMPAT_LEVEL
,
4936 .type
= QEMU_OPT_STRING
,
4937 .help
= "Compatibility level (0.10 or 1.1)"
4940 .name
= BLOCK_OPT_BACKING_FILE
,
4941 .type
= QEMU_OPT_STRING
,
4942 .help
= "File name of a base image"
4945 .name
= BLOCK_OPT_BACKING_FMT
,
4946 .type
= QEMU_OPT_STRING
,
4947 .help
= "Image format of the base image"
4950 .name
= BLOCK_OPT_ENCRYPT
,
4951 .type
= QEMU_OPT_BOOL
,
4952 .help
= "Encrypt the image with format 'aes'. (Deprecated "
4953 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)",
4956 .name
= BLOCK_OPT_ENCRYPT_FORMAT
,
4957 .type
= QEMU_OPT_STRING
,
4958 .help
= "Encrypt the image, format choices: 'aes', 'luks'",
4960 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
4961 "ID of secret providing qcow AES key or LUKS passphrase"),
4962 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
4963 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
4964 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
4965 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
4966 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
4967 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
4969 .name
= BLOCK_OPT_CLUSTER_SIZE
,
4970 .type
= QEMU_OPT_SIZE
,
4971 .help
= "qcow2 cluster size",
4972 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
)
4975 .name
= BLOCK_OPT_PREALLOC
,
4976 .type
= QEMU_OPT_STRING
,
4977 .help
= "Preallocation mode (allowed values: off, metadata, "
4981 .name
= BLOCK_OPT_LAZY_REFCOUNTS
,
4982 .type
= QEMU_OPT_BOOL
,
4983 .help
= "Postpone refcount updates",
4984 .def_value_str
= "off"
4987 .name
= BLOCK_OPT_REFCOUNT_BITS
,
4988 .type
= QEMU_OPT_NUMBER
,
4989 .help
= "Width of a reference count entry in bits",
4990 .def_value_str
= "16"
4992 { /* end of list */ }
4996 static const char *const qcow2_strong_runtime_opts
[] = {
4997 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET
,
5002 BlockDriver bdrv_qcow2
= {
5003 .format_name
= "qcow2",
5004 .instance_size
= sizeof(BDRVQcow2State
),
5005 .bdrv_probe
= qcow2_probe
,
5006 .bdrv_open
= qcow2_open
,
5007 .bdrv_close
= qcow2_close
,
5008 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
5009 .bdrv_reopen_commit
= qcow2_reopen_commit
,
5010 .bdrv_reopen_abort
= qcow2_reopen_abort
,
5011 .bdrv_join_options
= qcow2_join_options
,
5012 .bdrv_child_perm
= bdrv_format_default_perms
,
5013 .bdrv_co_create_opts
= qcow2_co_create_opts
,
5014 .bdrv_co_create
= qcow2_co_create
,
5015 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
5016 .bdrv_co_block_status
= qcow2_co_block_status
,
5018 .bdrv_co_preadv
= qcow2_co_preadv
,
5019 .bdrv_co_pwritev
= qcow2_co_pwritev
,
5020 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
5022 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
5023 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
5024 .bdrv_co_copy_range_from
= qcow2_co_copy_range_from
,
5025 .bdrv_co_copy_range_to
= qcow2_co_copy_range_to
,
5026 .bdrv_co_truncate
= qcow2_co_truncate
,
5027 .bdrv_co_pwritev_compressed
= qcow2_co_pwritev_compressed
,
5028 .bdrv_make_empty
= qcow2_make_empty
,
5030 .bdrv_snapshot_create
= qcow2_snapshot_create
,
5031 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
5032 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
5033 .bdrv_snapshot_list
= qcow2_snapshot_list
,
5034 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
5035 .bdrv_measure
= qcow2_measure
,
5036 .bdrv_get_info
= qcow2_get_info
,
5037 .bdrv_get_specific_info
= qcow2_get_specific_info
,
5039 .bdrv_save_vmstate
= qcow2_save_vmstate
,
5040 .bdrv_load_vmstate
= qcow2_load_vmstate
,
5042 .supports_backing
= true,
5043 .bdrv_change_backing_file
= qcow2_change_backing_file
,
5045 .bdrv_refresh_limits
= qcow2_refresh_limits
,
5046 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
5047 .bdrv_inactivate
= qcow2_inactivate
,
5049 .create_opts
= &qcow2_create_opts
,
5050 .strong_runtime_opts
= qcow2_strong_runtime_opts
,
5051 .bdrv_co_check
= qcow2_co_check
,
5052 .bdrv_amend_options
= qcow2_amend_options
,
5054 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
5055 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
5057 .bdrv_reopen_bitmaps_rw
= qcow2_reopen_bitmaps_rw
,
5058 .bdrv_can_store_new_dirty_bitmap
= qcow2_can_store_new_dirty_bitmap
,
5059 .bdrv_remove_persistent_dirty_bitmap
= qcow2_remove_persistent_dirty_bitmap
,
5062 static void bdrv_qcow2_init(void)
5064 bdrv_register(&bdrv_qcow2
);
5067 block_init(bdrv_qcow2_init
);