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
, false) == 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
);
401 case QCOW2_EXT_MAGIC_DATA_FILE
:
403 s
->image_data_file
= g_malloc0(ext
.len
+ 1);
404 ret
= bdrv_pread(bs
->file
, offset
, s
->image_data_file
, ext
.len
);
406 error_setg_errno(errp
, -ret
,
407 "ERROR: Could not read data file name");
411 printf("Qcow2: Got external data file %s\n", s
->image_data_file
);
417 /* unknown magic - save it in case we need to rewrite the header */
418 /* If you add a new feature, make sure to also update the fast
419 * path of qcow2_make_empty() to deal with it. */
421 Qcow2UnknownHeaderExtension
*uext
;
423 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
424 uext
->magic
= ext
.magic
;
426 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
428 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
430 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
431 "Could not read data");
438 offset
+= ((ext
.len
+ 7) & ~7);
444 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
446 BDRVQcow2State
*s
= bs
->opaque
;
447 Qcow2UnknownHeaderExtension
*uext
, *next
;
449 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
450 QLIST_REMOVE(uext
, next
);
455 static void report_unsupported_feature(Error
**errp
, Qcow2Feature
*table
,
458 char *features
= g_strdup("");
461 while (table
&& table
->name
[0] != '\0') {
462 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
463 if (mask
& (1ULL << table
->bit
)) {
465 features
= g_strdup_printf("%s%s%.46s", old
, *old
? ", " : "",
468 mask
&= ~(1ULL << table
->bit
);
476 features
= g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64
,
477 old
, *old
? ", " : "", mask
);
481 error_setg(errp
, "Unsupported qcow2 feature(s): %s", features
);
486 * Sets the dirty bit and flushes afterwards if necessary.
488 * The incompatible_features bit is only set if the image file header was
489 * updated successfully. Therefore it is not required to check the return
490 * value of this function.
492 int qcow2_mark_dirty(BlockDriverState
*bs
)
494 BDRVQcow2State
*s
= bs
->opaque
;
498 assert(s
->qcow_version
>= 3);
500 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
501 return 0; /* already dirty */
504 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
505 ret
= bdrv_pwrite(bs
->file
, offsetof(QCowHeader
, incompatible_features
),
510 ret
= bdrv_flush(bs
->file
->bs
);
515 /* Only treat image as dirty if the header was updated successfully */
516 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
521 * Clears the dirty bit and flushes before if necessary. Only call this
522 * function when there are no pending requests, it does not guard against
523 * concurrent requests dirtying the image.
525 static int qcow2_mark_clean(BlockDriverState
*bs
)
527 BDRVQcow2State
*s
= bs
->opaque
;
529 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
532 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
534 ret
= qcow2_flush_caches(bs
);
539 return qcow2_update_header(bs
);
545 * Marks the image as corrupt.
547 int qcow2_mark_corrupt(BlockDriverState
*bs
)
549 BDRVQcow2State
*s
= bs
->opaque
;
551 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
552 return qcow2_update_header(bs
);
556 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
557 * before if necessary.
559 int qcow2_mark_consistent(BlockDriverState
*bs
)
561 BDRVQcow2State
*s
= bs
->opaque
;
563 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
564 int ret
= qcow2_flush_caches(bs
);
569 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
570 return qcow2_update_header(bs
);
575 static int coroutine_fn
qcow2_co_check_locked(BlockDriverState
*bs
,
576 BdrvCheckResult
*result
,
579 int ret
= qcow2_check_refcounts(bs
, result
, fix
);
584 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
585 ret
= qcow2_mark_clean(bs
);
589 return qcow2_mark_consistent(bs
);
594 static int coroutine_fn
qcow2_co_check(BlockDriverState
*bs
,
595 BdrvCheckResult
*result
,
598 BDRVQcow2State
*s
= bs
->opaque
;
601 qemu_co_mutex_lock(&s
->lock
);
602 ret
= qcow2_co_check_locked(bs
, result
, fix
);
603 qemu_co_mutex_unlock(&s
->lock
);
607 int qcow2_validate_table(BlockDriverState
*bs
, uint64_t offset
,
608 uint64_t entries
, size_t entry_len
,
609 int64_t max_size_bytes
, const char *table_name
,
612 BDRVQcow2State
*s
= bs
->opaque
;
614 if (entries
> max_size_bytes
/ entry_len
) {
615 error_setg(errp
, "%s too large", table_name
);
619 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
620 * because values will be passed to qemu functions taking int64_t. */
621 if ((INT64_MAX
- entries
* entry_len
< offset
) ||
622 (offset_into_cluster(s
, offset
) != 0)) {
623 error_setg(errp
, "%s offset invalid", table_name
);
630 static const char *const mutable_opts
[] = {
631 QCOW2_OPT_LAZY_REFCOUNTS
,
632 QCOW2_OPT_DISCARD_REQUEST
,
633 QCOW2_OPT_DISCARD_SNAPSHOT
,
634 QCOW2_OPT_DISCARD_OTHER
,
636 QCOW2_OPT_OVERLAP_TEMPLATE
,
637 QCOW2_OPT_OVERLAP_MAIN_HEADER
,
638 QCOW2_OPT_OVERLAP_ACTIVE_L1
,
639 QCOW2_OPT_OVERLAP_ACTIVE_L2
,
640 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
641 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
642 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
643 QCOW2_OPT_OVERLAP_INACTIVE_L1
,
644 QCOW2_OPT_OVERLAP_INACTIVE_L2
,
645 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
646 QCOW2_OPT_CACHE_SIZE
,
647 QCOW2_OPT_L2_CACHE_SIZE
,
648 QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
649 QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
650 QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
654 static QemuOptsList qcow2_runtime_opts
= {
656 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
659 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
660 .type
= QEMU_OPT_BOOL
,
661 .help
= "Postpone refcount updates",
664 .name
= QCOW2_OPT_DISCARD_REQUEST
,
665 .type
= QEMU_OPT_BOOL
,
666 .help
= "Pass guest discard requests to the layer below",
669 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
670 .type
= QEMU_OPT_BOOL
,
671 .help
= "Generate discard requests when snapshot related space "
675 .name
= QCOW2_OPT_DISCARD_OTHER
,
676 .type
= QEMU_OPT_BOOL
,
677 .help
= "Generate discard requests when other clusters are freed",
680 .name
= QCOW2_OPT_OVERLAP
,
681 .type
= QEMU_OPT_STRING
,
682 .help
= "Selects which overlap checks to perform from a range of "
683 "templates (none, constant, cached, all)",
686 .name
= QCOW2_OPT_OVERLAP_TEMPLATE
,
687 .type
= QEMU_OPT_STRING
,
688 .help
= "Selects which overlap checks to perform from a range of "
689 "templates (none, constant, cached, all)",
692 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
693 .type
= QEMU_OPT_BOOL
,
694 .help
= "Check for unintended writes into the main qcow2 header",
697 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
698 .type
= QEMU_OPT_BOOL
,
699 .help
= "Check for unintended writes into the active L1 table",
702 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
703 .type
= QEMU_OPT_BOOL
,
704 .help
= "Check for unintended writes into an active L2 table",
707 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
708 .type
= QEMU_OPT_BOOL
,
709 .help
= "Check for unintended writes into the refcount table",
712 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
713 .type
= QEMU_OPT_BOOL
,
714 .help
= "Check for unintended writes into a refcount block",
717 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
718 .type
= QEMU_OPT_BOOL
,
719 .help
= "Check for unintended writes into the snapshot table",
722 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
723 .type
= QEMU_OPT_BOOL
,
724 .help
= "Check for unintended writes into an inactive L1 table",
727 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
728 .type
= QEMU_OPT_BOOL
,
729 .help
= "Check for unintended writes into an inactive L2 table",
732 .name
= QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
733 .type
= QEMU_OPT_BOOL
,
734 .help
= "Check for unintended writes into the bitmap directory",
737 .name
= QCOW2_OPT_CACHE_SIZE
,
738 .type
= QEMU_OPT_SIZE
,
739 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
743 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
744 .type
= QEMU_OPT_SIZE
,
745 .help
= "Maximum L2 table cache size",
748 .name
= QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
749 .type
= QEMU_OPT_SIZE
,
750 .help
= "Size of each entry in the L2 cache",
753 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
754 .type
= QEMU_OPT_SIZE
,
755 .help
= "Maximum refcount block cache size",
758 .name
= QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
759 .type
= QEMU_OPT_NUMBER
,
760 .help
= "Clean unused cache entries after this time (in seconds)",
762 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
763 "ID of secret providing qcow2 AES key or LUKS passphrase"),
764 { /* end of list */ }
768 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
769 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
770 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
771 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
772 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
773 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
774 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
775 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
776 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
777 [QCOW2_OL_BITMAP_DIRECTORY_BITNR
] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
780 static void cache_clean_timer_cb(void *opaque
)
782 BlockDriverState
*bs
= opaque
;
783 BDRVQcow2State
*s
= bs
->opaque
;
784 qcow2_cache_clean_unused(s
->l2_table_cache
);
785 qcow2_cache_clean_unused(s
->refcount_block_cache
);
786 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
787 (int64_t) s
->cache_clean_interval
* 1000);
790 static void cache_clean_timer_init(BlockDriverState
*bs
, AioContext
*context
)
792 BDRVQcow2State
*s
= bs
->opaque
;
793 if (s
->cache_clean_interval
> 0) {
794 s
->cache_clean_timer
= aio_timer_new(context
, QEMU_CLOCK_VIRTUAL
,
795 SCALE_MS
, cache_clean_timer_cb
,
797 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
798 (int64_t) s
->cache_clean_interval
* 1000);
802 static void cache_clean_timer_del(BlockDriverState
*bs
)
804 BDRVQcow2State
*s
= bs
->opaque
;
805 if (s
->cache_clean_timer
) {
806 timer_del(s
->cache_clean_timer
);
807 timer_free(s
->cache_clean_timer
);
808 s
->cache_clean_timer
= NULL
;
812 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
814 cache_clean_timer_del(bs
);
817 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
818 AioContext
*new_context
)
820 cache_clean_timer_init(bs
, new_context
);
823 static void read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
824 uint64_t *l2_cache_size
,
825 uint64_t *l2_cache_entry_size
,
826 uint64_t *refcount_cache_size
, Error
**errp
)
828 BDRVQcow2State
*s
= bs
->opaque
;
829 uint64_t combined_cache_size
, l2_cache_max_setting
;
830 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
831 bool l2_cache_entry_size_set
;
832 int min_refcount_cache
= MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
833 uint64_t virtual_disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
834 uint64_t max_l2_cache
= virtual_disk_size
/ (s
->cluster_size
/ 8);
836 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
837 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
838 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
839 l2_cache_entry_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
);
841 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
842 l2_cache_max_setting
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
,
843 DEFAULT_L2_CACHE_MAX_SIZE
);
844 *refcount_cache_size
= qemu_opt_get_size(opts
,
845 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
847 *l2_cache_entry_size
= qemu_opt_get_size(
848 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
850 *l2_cache_size
= MIN(max_l2_cache
, l2_cache_max_setting
);
852 if (combined_cache_size_set
) {
853 if (l2_cache_size_set
&& refcount_cache_size_set
) {
854 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
855 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
858 } else if (l2_cache_size_set
&&
859 (l2_cache_max_setting
> combined_cache_size
)) {
860 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
861 QCOW2_OPT_CACHE_SIZE
);
863 } else if (*refcount_cache_size
> combined_cache_size
) {
864 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
865 QCOW2_OPT_CACHE_SIZE
);
869 if (l2_cache_size_set
) {
870 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
871 } else if (refcount_cache_size_set
) {
872 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
874 /* Assign as much memory as possible to the L2 cache, and
875 * use the remainder for the refcount cache */
876 if (combined_cache_size
>= max_l2_cache
+ min_refcount_cache
) {
877 *l2_cache_size
= max_l2_cache
;
878 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
880 *refcount_cache_size
=
881 MIN(combined_cache_size
, min_refcount_cache
);
882 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
888 * If the L2 cache is not enough to cover the whole disk then
889 * default to 4KB entries. Smaller entries reduce the cost of
890 * loads and evictions and increase I/O performance.
892 if (*l2_cache_size
< max_l2_cache
&& !l2_cache_entry_size_set
) {
893 *l2_cache_entry_size
= MIN(s
->cluster_size
, 4096);
896 /* l2_cache_size and refcount_cache_size are ensured to have at least
897 * their minimum values in qcow2_update_options_prepare() */
899 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
900 *l2_cache_entry_size
> s
->cluster_size
||
901 !is_power_of_2(*l2_cache_entry_size
)) {
902 error_setg(errp
, "L2 cache entry size must be a power of two "
903 "between %d and the cluster size (%d)",
904 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
909 typedef struct Qcow2ReopenState
{
910 Qcow2Cache
*l2_table_cache
;
911 Qcow2Cache
*refcount_block_cache
;
912 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
913 bool use_lazy_refcounts
;
915 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
916 uint64_t cache_clean_interval
;
917 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
920 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
922 QDict
*options
, int flags
,
925 BDRVQcow2State
*s
= bs
->opaque
;
926 QemuOpts
*opts
= NULL
;
927 const char *opt_overlap_check
, *opt_overlap_check_template
;
928 int overlap_check_template
= 0;
929 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
931 const char *encryptfmt
;
932 QDict
*encryptopts
= NULL
;
933 Error
*local_err
= NULL
;
936 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
937 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
939 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
940 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
942 error_propagate(errp
, local_err
);
947 /* get L2 table/refcount block cache size from command line options */
948 read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
949 &refcount_cache_size
, &local_err
);
951 error_propagate(errp
, local_err
);
956 l2_cache_size
/= l2_cache_entry_size
;
957 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
958 l2_cache_size
= MIN_L2_CACHE_SIZE
;
960 if (l2_cache_size
> INT_MAX
) {
961 error_setg(errp
, "L2 cache size too big");
966 refcount_cache_size
/= s
->cluster_size
;
967 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
968 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
970 if (refcount_cache_size
> INT_MAX
) {
971 error_setg(errp
, "Refcount cache size too big");
976 /* alloc new L2 table/refcount block cache, flush old one */
977 if (s
->l2_table_cache
) {
978 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
980 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
985 if (s
->refcount_block_cache
) {
986 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
988 error_setg_errno(errp
, -ret
,
989 "Failed to flush the refcount block cache");
994 r
->l2_slice_size
= l2_cache_entry_size
/ sizeof(uint64_t);
995 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
996 l2_cache_entry_size
);
997 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
999 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
1000 error_setg(errp
, "Could not allocate metadata caches");
1005 /* New interval for cache cleanup timer */
1006 r
->cache_clean_interval
=
1007 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
1008 DEFAULT_CACHE_CLEAN_INTERVAL
);
1009 #ifndef CONFIG_LINUX
1010 if (r
->cache_clean_interval
!= 0) {
1011 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1012 " not supported on this host");
1017 if (r
->cache_clean_interval
> UINT_MAX
) {
1018 error_setg(errp
, "Cache clean interval too big");
1023 /* lazy-refcounts; flush if going from enabled to disabled */
1024 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
1025 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
1026 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
1027 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
1028 "qemu 1.1 compatibility level");
1033 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
1034 ret
= qcow2_mark_clean(bs
);
1036 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
1041 /* Overlap check options */
1042 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
1043 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1044 if (opt_overlap_check_template
&& opt_overlap_check
&&
1045 strcmp(opt_overlap_check_template
, opt_overlap_check
))
1047 error_setg(errp
, "Conflicting values for qcow2 options '"
1048 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1049 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
1053 if (!opt_overlap_check
) {
1054 opt_overlap_check
= opt_overlap_check_template
?: "cached";
1057 if (!strcmp(opt_overlap_check
, "none")) {
1058 overlap_check_template
= 0;
1059 } else if (!strcmp(opt_overlap_check
, "constant")) {
1060 overlap_check_template
= QCOW2_OL_CONSTANT
;
1061 } else if (!strcmp(opt_overlap_check
, "cached")) {
1062 overlap_check_template
= QCOW2_OL_CACHED
;
1063 } else if (!strcmp(opt_overlap_check
, "all")) {
1064 overlap_check_template
= QCOW2_OL_ALL
;
1066 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1067 "'overlap-check'. Allowed are any of the following: "
1068 "none, constant, cached, all", opt_overlap_check
);
1073 r
->overlap_check
= 0;
1074 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1075 /* overlap-check defines a template bitmask, but every flag may be
1076 * overwritten through the associated boolean option */
1078 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1079 overlap_check_template
& (1 << i
)) << i
;
1082 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1083 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1084 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1085 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1086 flags
& BDRV_O_UNMAP
);
1087 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1088 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1089 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1090 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1092 switch (s
->crypt_method_header
) {
1093 case QCOW_CRYPT_NONE
:
1095 error_setg(errp
, "No encryption in image header, but options "
1096 "specified format '%s'", encryptfmt
);
1102 case QCOW_CRYPT_AES
:
1103 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1105 "Header reported 'aes' encryption format but "
1106 "options specify '%s'", encryptfmt
);
1110 qdict_put_str(encryptopts
, "format", "qcow");
1111 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1114 case QCOW_CRYPT_LUKS
:
1115 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1117 "Header reported 'luks' encryption format but "
1118 "options specify '%s'", encryptfmt
);
1122 qdict_put_str(encryptopts
, "format", "luks");
1123 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1127 error_setg(errp
, "Unsupported encryption method %d",
1128 s
->crypt_method_header
);
1131 if (s
->crypt_method_header
!= QCOW_CRYPT_NONE
&& !r
->crypto_opts
) {
1138 qobject_unref(encryptopts
);
1139 qemu_opts_del(opts
);
1144 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1145 Qcow2ReopenState
*r
)
1147 BDRVQcow2State
*s
= bs
->opaque
;
1150 if (s
->l2_table_cache
) {
1151 qcow2_cache_destroy(s
->l2_table_cache
);
1153 if (s
->refcount_block_cache
) {
1154 qcow2_cache_destroy(s
->refcount_block_cache
);
1156 s
->l2_table_cache
= r
->l2_table_cache
;
1157 s
->refcount_block_cache
= r
->refcount_block_cache
;
1158 s
->l2_slice_size
= r
->l2_slice_size
;
1160 s
->overlap_check
= r
->overlap_check
;
1161 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1163 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1164 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1167 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1168 cache_clean_timer_del(bs
);
1169 s
->cache_clean_interval
= r
->cache_clean_interval
;
1170 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1173 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1174 s
->crypto_opts
= r
->crypto_opts
;
1177 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1178 Qcow2ReopenState
*r
)
1180 if (r
->l2_table_cache
) {
1181 qcow2_cache_destroy(r
->l2_table_cache
);
1183 if (r
->refcount_block_cache
) {
1184 qcow2_cache_destroy(r
->refcount_block_cache
);
1186 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1189 static int qcow2_update_options(BlockDriverState
*bs
, QDict
*options
,
1190 int flags
, Error
**errp
)
1192 Qcow2ReopenState r
= {};
1195 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1197 qcow2_update_options_commit(bs
, &r
);
1199 qcow2_update_options_abort(bs
, &r
);
1205 /* Called with s->lock held. */
1206 static int coroutine_fn
qcow2_do_open(BlockDriverState
*bs
, QDict
*options
,
1207 int flags
, Error
**errp
)
1209 BDRVQcow2State
*s
= bs
->opaque
;
1210 unsigned int len
, i
;
1213 Error
*local_err
= NULL
;
1215 uint64_t l1_vm_state_index
;
1216 bool update_header
= false;
1218 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
1220 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1223 header
.magic
= be32_to_cpu(header
.magic
);
1224 header
.version
= be32_to_cpu(header
.version
);
1225 header
.backing_file_offset
= be64_to_cpu(header
.backing_file_offset
);
1226 header
.backing_file_size
= be32_to_cpu(header
.backing_file_size
);
1227 header
.size
= be64_to_cpu(header
.size
);
1228 header
.cluster_bits
= be32_to_cpu(header
.cluster_bits
);
1229 header
.crypt_method
= be32_to_cpu(header
.crypt_method
);
1230 header
.l1_table_offset
= be64_to_cpu(header
.l1_table_offset
);
1231 header
.l1_size
= be32_to_cpu(header
.l1_size
);
1232 header
.refcount_table_offset
= be64_to_cpu(header
.refcount_table_offset
);
1233 header
.refcount_table_clusters
=
1234 be32_to_cpu(header
.refcount_table_clusters
);
1235 header
.snapshots_offset
= be64_to_cpu(header
.snapshots_offset
);
1236 header
.nb_snapshots
= be32_to_cpu(header
.nb_snapshots
);
1238 if (header
.magic
!= QCOW_MAGIC
) {
1239 error_setg(errp
, "Image is not in qcow2 format");
1243 if (header
.version
< 2 || header
.version
> 3) {
1244 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1249 s
->qcow_version
= header
.version
;
1251 /* Initialise cluster size */
1252 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1253 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1254 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1255 header
.cluster_bits
);
1260 s
->cluster_bits
= header
.cluster_bits
;
1261 s
->cluster_size
= 1 << s
->cluster_bits
;
1262 s
->cluster_sectors
= 1 << (s
->cluster_bits
- BDRV_SECTOR_BITS
);
1264 /* Initialise version 3 header fields */
1265 if (header
.version
== 2) {
1266 header
.incompatible_features
= 0;
1267 header
.compatible_features
= 0;
1268 header
.autoclear_features
= 0;
1269 header
.refcount_order
= 4;
1270 header
.header_length
= 72;
1272 header
.incompatible_features
=
1273 be64_to_cpu(header
.incompatible_features
);
1274 header
.compatible_features
= be64_to_cpu(header
.compatible_features
);
1275 header
.autoclear_features
= be64_to_cpu(header
.autoclear_features
);
1276 header
.refcount_order
= be32_to_cpu(header
.refcount_order
);
1277 header
.header_length
= be32_to_cpu(header
.header_length
);
1279 if (header
.header_length
< 104) {
1280 error_setg(errp
, "qcow2 header too short");
1286 if (header
.header_length
> s
->cluster_size
) {
1287 error_setg(errp
, "qcow2 header exceeds cluster size");
1292 if (header
.header_length
> sizeof(header
)) {
1293 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1294 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1295 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
1296 s
->unknown_header_fields_size
);
1298 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1304 if (header
.backing_file_offset
> s
->cluster_size
) {
1305 error_setg(errp
, "Invalid backing file offset");
1310 if (header
.backing_file_offset
) {
1311 ext_end
= header
.backing_file_offset
;
1313 ext_end
= 1 << header
.cluster_bits
;
1316 /* Handle feature bits */
1317 s
->incompatible_features
= header
.incompatible_features
;
1318 s
->compatible_features
= header
.compatible_features
;
1319 s
->autoclear_features
= header
.autoclear_features
;
1321 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1322 void *feature_table
= NULL
;
1323 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1324 &feature_table
, flags
, NULL
, NULL
);
1325 report_unsupported_feature(errp
, feature_table
,
1326 s
->incompatible_features
&
1327 ~QCOW2_INCOMPAT_MASK
);
1329 g_free(feature_table
);
1333 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1334 /* Corrupt images may not be written to unless they are being repaired
1336 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1337 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1344 /* Check support for various header values */
1345 if (header
.refcount_order
> 6) {
1346 error_setg(errp
, "Reference count entry width too large; may not "
1351 s
->refcount_order
= header
.refcount_order
;
1352 s
->refcount_bits
= 1 << s
->refcount_order
;
1353 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1354 s
->refcount_max
+= s
->refcount_max
- 1;
1356 s
->crypt_method_header
= header
.crypt_method
;
1357 if (s
->crypt_method_header
) {
1358 if (bdrv_uses_whitelist() &&
1359 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1361 "Use of AES-CBC encrypted qcow2 images is no longer "
1362 "supported in system emulators");
1363 error_append_hint(errp
,
1364 "You can use 'qemu-img convert' to convert your "
1365 "image to an alternative supported format, such "
1366 "as unencrypted qcow2, or raw with the LUKS "
1367 "format instead.\n");
1372 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1373 s
->crypt_physical_offset
= false;
1375 /* Assuming LUKS and any future crypt methods we
1376 * add will all use physical offsets, due to the
1377 * fact that the alternative is insecure... */
1378 s
->crypt_physical_offset
= true;
1381 bs
->encrypted
= true;
1384 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
1385 s
->l2_size
= 1 << s
->l2_bits
;
1386 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1387 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1388 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1389 bs
->total_sectors
= header
.size
/ BDRV_SECTOR_SIZE
;
1390 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1391 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1392 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1394 s
->refcount_table_offset
= header
.refcount_table_offset
;
1395 s
->refcount_table_size
=
1396 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1398 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1399 error_setg(errp
, "Image does not contain a reference count table");
1404 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1405 header
.refcount_table_clusters
,
1406 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1407 "Reference count table", errp
);
1412 /* The total size in bytes of the snapshot table is checked in
1413 * qcow2_read_snapshots() because the size of each snapshot is
1414 * variable and we don't know it yet.
1415 * Here we only check the offset and number of snapshots. */
1416 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1417 header
.nb_snapshots
,
1418 sizeof(QCowSnapshotHeader
),
1419 sizeof(QCowSnapshotHeader
) * QCOW_MAX_SNAPSHOTS
,
1420 "Snapshot table", errp
);
1425 /* read the level 1 table */
1426 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1427 header
.l1_size
, sizeof(uint64_t),
1428 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1432 s
->l1_size
= header
.l1_size
;
1433 s
->l1_table_offset
= header
.l1_table_offset
;
1435 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1436 if (l1_vm_state_index
> INT_MAX
) {
1437 error_setg(errp
, "Image is too big");
1441 s
->l1_vm_state_index
= l1_vm_state_index
;
1443 /* the L1 table must contain at least enough entries to put
1444 header.size bytes */
1445 if (s
->l1_size
< s
->l1_vm_state_index
) {
1446 error_setg(errp
, "L1 table is too small");
1451 if (s
->l1_size
> 0) {
1452 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
,
1453 ROUND_UP(s
->l1_size
* sizeof(uint64_t), 512));
1454 if (s
->l1_table
== NULL
) {
1455 error_setg(errp
, "Could not allocate L1 table");
1459 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1460 s
->l1_size
* sizeof(uint64_t));
1462 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1465 for(i
= 0;i
< s
->l1_size
; i
++) {
1466 s
->l1_table
[i
] = be64_to_cpu(s
->l1_table
[i
]);
1470 /* Parse driver-specific options */
1471 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1478 ret
= qcow2_refcount_init(bs
);
1480 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1484 QLIST_INIT(&s
->cluster_allocs
);
1485 QTAILQ_INIT(&s
->discards
);
1487 /* read qcow2 extensions */
1488 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1489 flags
, &update_header
, &local_err
)) {
1490 error_propagate(errp
, local_err
);
1495 /* Open external data file */
1496 s
->data_file
= bdrv_open_child(NULL
, options
, "data-file", bs
, &child_file
,
1499 error_propagate(errp
, local_err
);
1504 if (s
->incompatible_features
& QCOW2_INCOMPAT_DATA_FILE
) {
1505 if (!s
->data_file
&& s
->image_data_file
) {
1506 s
->data_file
= bdrv_open_child(s
->image_data_file
, options
,
1507 "data-file", bs
, &child_file
,
1509 if (!s
->data_file
) {
1514 if (!s
->data_file
) {
1515 error_setg(errp
, "'data-file' is required for this image");
1521 error_setg(errp
, "'data-file' can only be set for images with an "
1522 "external data file");
1527 s
->data_file
= bs
->file
;
1529 if (data_file_is_raw(bs
)) {
1530 error_setg(errp
, "data-file-raw requires a data file");
1536 /* qcow2_read_extension may have set up the crypto context
1537 * if the crypt method needs a header region, some methods
1538 * don't need header extensions, so must check here
1540 if (s
->crypt_method_header
&& !s
->crypto
) {
1541 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1542 unsigned int cflags
= 0;
1543 if (flags
& BDRV_O_NO_IO
) {
1544 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1546 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1547 NULL
, NULL
, cflags
, 1, errp
);
1552 } else if (!(flags
& BDRV_O_NO_IO
)) {
1553 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1554 s
->crypt_method_header
);
1560 /* read the backing file name */
1561 if (header
.backing_file_offset
!= 0) {
1562 len
= header
.backing_file_size
;
1563 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1564 len
>= sizeof(bs
->backing_file
)) {
1565 error_setg(errp
, "Backing file name too long");
1569 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
1570 bs
->auto_backing_file
, len
);
1572 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1575 bs
->auto_backing_file
[len
] = '\0';
1576 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
),
1577 bs
->auto_backing_file
);
1578 s
->image_backing_file
= g_strdup(bs
->auto_backing_file
);
1581 /* Internal snapshots */
1582 s
->snapshots_offset
= header
.snapshots_offset
;
1583 s
->nb_snapshots
= header
.nb_snapshots
;
1585 ret
= qcow2_read_snapshots(bs
);
1587 error_setg_errno(errp
, -ret
, "Could not read snapshots");
1591 /* Clear unknown autoclear feature bits */
1592 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1594 update_header
&& !bs
->read_only
&& !(flags
& BDRV_O_INACTIVE
);
1595 if (update_header
) {
1596 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1599 /* == Handle persistent dirty bitmaps ==
1601 * We want load dirty bitmaps in three cases:
1603 * 1. Normal open of the disk in active mode, not related to invalidation
1606 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1607 * bitmaps are _not_ migrating through migration channel, i.e.
1608 * 'dirty-bitmaps' capability is disabled.
1610 * 3. Invalidation of source vm after failed or canceled migration.
1611 * This is a very interesting case. There are two possible types of
1614 * A. Stored on inactivation and removed. They should be loaded from the
1617 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1618 * the migration channel (with dirty-bitmaps capability).
1620 * On the other hand, there are two possible sub-cases:
1622 * 3.1 disk was changed by somebody else while were inactive. In this
1623 * case all in-RAM dirty bitmaps (both persistent and not) are
1624 * definitely invalid. And we don't have any method to determine
1627 * Simple and safe thing is to just drop all the bitmaps of type B on
1628 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1630 * On the other hand, resuming source vm, if disk was already changed
1631 * is a bad thing anyway: not only bitmaps, the whole vm state is
1632 * out of sync with disk.
1634 * This means, that user or management tool, who for some reason
1635 * decided to resume source vm, after disk was already changed by
1636 * target vm, should at least drop all dirty bitmaps by hand.
1638 * So, we can ignore this case for now, but TODO: "generation"
1639 * extension for qcow2, to determine, that image was changed after
1640 * last inactivation. And if it is changed, we will drop (or at least
1641 * mark as 'invalid' all the bitmaps of type B, both persistent
1644 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1645 * to disk ('dirty-bitmaps' capability disabled), or not saved
1646 * ('dirty-bitmaps' capability enabled), but we don't need to care
1647 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1648 * and not stored has flag IN_USE=1 in the image and will be skipped
1651 * One remaining possible case when we don't want load bitmaps:
1653 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1654 * will be loaded on invalidation, no needs try loading them before)
1657 if (!(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
)) {
1658 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1659 bool header_updated
= qcow2_load_dirty_bitmaps(bs
, &local_err
);
1661 update_header
= update_header
&& !header_updated
;
1663 if (local_err
!= NULL
) {
1664 error_propagate(errp
, local_err
);
1669 if (update_header
) {
1670 ret
= qcow2_update_header(bs
);
1672 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1677 bs
->supported_zero_flags
= header
.version
>= 3 ? BDRV_REQ_MAY_UNMAP
: 0;
1679 /* Repair image if dirty */
1680 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INACTIVE
)) && !bs
->read_only
&&
1681 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1682 BdrvCheckResult result
= {0};
1684 ret
= qcow2_co_check_locked(bs
, &result
,
1685 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1686 if (ret
< 0 || result
.check_errors
) {
1690 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1697 BdrvCheckResult result
= {0};
1698 qcow2_check_refcounts(bs
, &result
, 0);
1702 qemu_co_queue_init(&s
->compress_wait_queue
);
1707 g_free(s
->image_data_file
);
1708 if (has_data_file(bs
)) {
1709 bdrv_unref_child(bs
, s
->data_file
);
1711 g_free(s
->unknown_header_fields
);
1712 cleanup_unknown_header_ext(bs
);
1713 qcow2_free_snapshots(bs
);
1714 qcow2_refcount_close(bs
);
1715 qemu_vfree(s
->l1_table
);
1716 /* else pre-write overlap checks in cache_destroy may crash */
1718 cache_clean_timer_del(bs
);
1719 if (s
->l2_table_cache
) {
1720 qcow2_cache_destroy(s
->l2_table_cache
);
1722 if (s
->refcount_block_cache
) {
1723 qcow2_cache_destroy(s
->refcount_block_cache
);
1725 qcrypto_block_free(s
->crypto
);
1726 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1730 typedef struct QCow2OpenCo
{
1731 BlockDriverState
*bs
;
1738 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1740 QCow2OpenCo
*qoc
= opaque
;
1741 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1743 qemu_co_mutex_lock(&s
->lock
);
1744 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, qoc
->errp
);
1745 qemu_co_mutex_unlock(&s
->lock
);
1748 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1751 BDRVQcow2State
*s
= bs
->opaque
;
1760 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_file
,
1766 /* Initialise locks */
1767 qemu_co_mutex_init(&s
->lock
);
1769 if (qemu_in_coroutine()) {
1770 /* From bdrv_co_create. */
1771 qcow2_open_entry(&qoc
);
1773 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1774 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1775 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1780 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1782 BDRVQcow2State
*s
= bs
->opaque
;
1784 if (bs
->encrypted
) {
1785 /* Encryption works on a sector granularity */
1786 bs
->bl
.request_alignment
= qcrypto_block_get_sector_size(s
->crypto
);
1788 bs
->bl
.pwrite_zeroes_alignment
= s
->cluster_size
;
1789 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1792 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1793 BlockReopenQueue
*queue
, Error
**errp
)
1795 Qcow2ReopenState
*r
;
1798 r
= g_new0(Qcow2ReopenState
, 1);
1801 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1802 state
->flags
, errp
);
1807 /* We need to write out any unwritten data if we reopen read-only. */
1808 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1809 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1814 ret
= bdrv_flush(state
->bs
);
1819 ret
= qcow2_mark_clean(state
->bs
);
1828 qcow2_update_options_abort(state
->bs
, r
);
1833 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1835 qcow2_update_options_commit(state
->bs
, state
->opaque
);
1836 g_free(state
->opaque
);
1839 static void qcow2_reopen_abort(BDRVReopenState
*state
)
1841 qcow2_update_options_abort(state
->bs
, state
->opaque
);
1842 g_free(state
->opaque
);
1845 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
1847 bool has_new_overlap_template
=
1848 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
1849 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1850 bool has_new_total_cache_size
=
1851 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
1852 bool has_all_cache_options
;
1854 /* New overlap template overrides all old overlap options */
1855 if (has_new_overlap_template
) {
1856 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
1857 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1858 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
1859 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
1860 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
1861 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
1862 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
1863 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
1864 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
1865 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
1868 /* New total cache size overrides all old options */
1869 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
1870 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
1871 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1874 qdict_join(options
, old_options
, false);
1877 * If after merging all cache size options are set, an old total size is
1878 * overwritten. Do keep all options, however, if all three are new. The
1879 * resulting error message is what we want to happen.
1881 has_all_cache_options
=
1882 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
1883 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
1884 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1886 if (has_all_cache_options
&& !has_new_total_cache_size
) {
1887 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
1891 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
1893 int64_t offset
, int64_t count
,
1894 int64_t *pnum
, int64_t *map
,
1895 BlockDriverState
**file
)
1897 BDRVQcow2State
*s
= bs
->opaque
;
1898 uint64_t cluster_offset
;
1899 int index_in_cluster
, ret
;
1903 bytes
= MIN(INT_MAX
, count
);
1904 qemu_co_mutex_lock(&s
->lock
);
1905 ret
= qcow2_get_cluster_offset(bs
, offset
, &bytes
, &cluster_offset
);
1906 qemu_co_mutex_unlock(&s
->lock
);
1913 if ((ret
== QCOW2_CLUSTER_NORMAL
|| ret
== QCOW2_CLUSTER_ZERO_ALLOC
) &&
1915 index_in_cluster
= offset
& (s
->cluster_size
- 1);
1916 *map
= cluster_offset
| index_in_cluster
;
1917 *file
= s
->data_file
->bs
;
1918 status
|= BDRV_BLOCK_OFFSET_VALID
;
1920 if (ret
== QCOW2_CLUSTER_ZERO_PLAIN
|| ret
== QCOW2_CLUSTER_ZERO_ALLOC
) {
1921 status
|= BDRV_BLOCK_ZERO
;
1922 } else if (ret
!= QCOW2_CLUSTER_UNALLOCATED
) {
1923 status
|= BDRV_BLOCK_DATA
;
1928 static coroutine_fn
int qcow2_handle_l2meta(BlockDriverState
*bs
,
1929 QCowL2Meta
**pl2meta
,
1933 QCowL2Meta
*l2meta
= *pl2meta
;
1935 while (l2meta
!= NULL
) {
1939 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
1944 qcow2_alloc_cluster_abort(bs
, l2meta
);
1947 /* Take the request off the list of running requests */
1948 if (l2meta
->nb_clusters
!= 0) {
1949 QLIST_REMOVE(l2meta
, next_in_flight
);
1952 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
1954 next
= l2meta
->next
;
1963 static coroutine_fn
int qcow2_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1964 uint64_t bytes
, QEMUIOVector
*qiov
,
1967 BDRVQcow2State
*s
= bs
->opaque
;
1968 int offset_in_cluster
;
1970 unsigned int cur_bytes
; /* number of bytes in current iteration */
1971 uint64_t cluster_offset
= 0;
1972 uint64_t bytes_done
= 0;
1973 QEMUIOVector hd_qiov
;
1974 uint8_t *cluster_data
= NULL
;
1976 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1978 qemu_co_mutex_lock(&s
->lock
);
1980 while (bytes
!= 0) {
1982 /* prepare next request */
1983 cur_bytes
= MIN(bytes
, INT_MAX
);
1985 cur_bytes
= MIN(cur_bytes
,
1986 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1989 ret
= qcow2_get_cluster_offset(bs
, offset
, &cur_bytes
, &cluster_offset
);
1994 offset_in_cluster
= offset_into_cluster(s
, offset
);
1996 qemu_iovec_reset(&hd_qiov
);
1997 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
2000 case QCOW2_CLUSTER_UNALLOCATED
:
2003 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
2004 qemu_co_mutex_unlock(&s
->lock
);
2005 ret
= bdrv_co_preadv(bs
->backing
, offset
, cur_bytes
,
2007 qemu_co_mutex_lock(&s
->lock
);
2012 /* Note: in this case, no need to wait */
2013 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
2017 case QCOW2_CLUSTER_ZERO_PLAIN
:
2018 case QCOW2_CLUSTER_ZERO_ALLOC
:
2019 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
2022 case QCOW2_CLUSTER_COMPRESSED
:
2023 qemu_co_mutex_unlock(&s
->lock
);
2024 ret
= qcow2_co_preadv_compressed(bs
, cluster_offset
,
2027 qemu_co_mutex_lock(&s
->lock
);
2034 case QCOW2_CLUSTER_NORMAL
:
2035 if ((cluster_offset
& 511) != 0) {
2040 if (bs
->encrypted
) {
2044 * For encrypted images, read everything into a temporary
2045 * contiguous buffer on which the AES functions can work.
2047 if (!cluster_data
) {
2049 qemu_try_blockalign(s
->data_file
->bs
,
2050 QCOW_MAX_CRYPT_CLUSTERS
2052 if (cluster_data
== NULL
) {
2058 assert(cur_bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2059 qemu_iovec_reset(&hd_qiov
);
2060 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
2063 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2064 qemu_co_mutex_unlock(&s
->lock
);
2065 ret
= bdrv_co_preadv(s
->data_file
,
2066 cluster_offset
+ offset_in_cluster
,
2067 cur_bytes
, &hd_qiov
, 0);
2068 qemu_co_mutex_lock(&s
->lock
);
2072 if (bs
->encrypted
) {
2074 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
2075 assert((cur_bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
2076 if (qcrypto_block_decrypt(s
->crypto
,
2077 (s
->crypt_physical_offset
?
2078 cluster_offset
+ offset_in_cluster
:
2086 qemu_iovec_from_buf(qiov
, bytes_done
, cluster_data
, cur_bytes
);
2091 g_assert_not_reached();
2097 offset
+= cur_bytes
;
2098 bytes_done
+= cur_bytes
;
2103 qemu_co_mutex_unlock(&s
->lock
);
2105 qemu_iovec_destroy(&hd_qiov
);
2106 qemu_vfree(cluster_data
);
2111 /* Check if it's possible to merge a write request with the writing of
2112 * the data from the COW regions */
2113 static bool merge_cow(uint64_t offset
, unsigned bytes
,
2114 QEMUIOVector
*hd_qiov
, QCowL2Meta
*l2meta
)
2118 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2119 /* If both COW regions are empty then there's nothing to merge */
2120 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
2124 /* The data (middle) region must be immediately after the
2126 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
2130 /* The end region must be immediately after the data (middle)
2132 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
2136 /* Make sure that adding both COW regions to the QEMUIOVector
2137 * does not exceed IOV_MAX */
2138 if (hd_qiov
->niov
> IOV_MAX
- 2) {
2142 m
->data_qiov
= hd_qiov
;
2149 static coroutine_fn
int qcow2_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
2150 uint64_t bytes
, QEMUIOVector
*qiov
,
2153 BDRVQcow2State
*s
= bs
->opaque
;
2154 int offset_in_cluster
;
2156 unsigned int cur_bytes
; /* number of sectors in current iteration */
2157 uint64_t cluster_offset
;
2158 QEMUIOVector hd_qiov
;
2159 uint64_t bytes_done
= 0;
2160 uint8_t *cluster_data
= NULL
;
2161 QCowL2Meta
*l2meta
= NULL
;
2163 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
2165 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
2167 qemu_co_mutex_lock(&s
->lock
);
2169 while (bytes
!= 0) {
2173 trace_qcow2_writev_start_part(qemu_coroutine_self());
2174 offset_in_cluster
= offset_into_cluster(s
, offset
);
2175 cur_bytes
= MIN(bytes
, INT_MAX
);
2176 if (bs
->encrypted
) {
2177 cur_bytes
= MIN(cur_bytes
,
2178 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
2179 - offset_in_cluster
);
2182 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2183 &cluster_offset
, &l2meta
);
2188 assert((cluster_offset
& 511) == 0);
2190 qemu_iovec_reset(&hd_qiov
);
2191 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
2193 if (bs
->encrypted
) {
2195 if (!cluster_data
) {
2196 cluster_data
= qemu_try_blockalign(bs
->file
->bs
,
2197 QCOW_MAX_CRYPT_CLUSTERS
2199 if (cluster_data
== NULL
) {
2205 assert(hd_qiov
.size
<=
2206 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2207 qemu_iovec_to_buf(&hd_qiov
, 0, cluster_data
, hd_qiov
.size
);
2209 if (qcrypto_block_encrypt(s
->crypto
,
2210 (s
->crypt_physical_offset
?
2211 cluster_offset
+ offset_in_cluster
:
2214 cur_bytes
, NULL
) < 0) {
2219 qemu_iovec_reset(&hd_qiov
);
2220 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
2223 ret
= qcow2_pre_write_overlap_check(bs
, 0,
2224 cluster_offset
+ offset_in_cluster
, cur_bytes
, true);
2229 /* If we need to do COW, check if it's possible to merge the
2230 * writing of the guest data together with that of the COW regions.
2231 * If it's not possible (or not necessary) then write the
2232 * guest data now. */
2233 if (!merge_cow(offset
, cur_bytes
, &hd_qiov
, l2meta
)) {
2234 qemu_co_mutex_unlock(&s
->lock
);
2235 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2236 trace_qcow2_writev_data(qemu_coroutine_self(),
2237 cluster_offset
+ offset_in_cluster
);
2238 ret
= bdrv_co_pwritev(s
->data_file
,
2239 cluster_offset
+ offset_in_cluster
,
2240 cur_bytes
, &hd_qiov
, 0);
2241 qemu_co_mutex_lock(&s
->lock
);
2247 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
2253 offset
+= cur_bytes
;
2254 bytes_done
+= cur_bytes
;
2255 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2260 qcow2_handle_l2meta(bs
, &l2meta
, false);
2262 qemu_co_mutex_unlock(&s
->lock
);
2264 qemu_iovec_destroy(&hd_qiov
);
2265 qemu_vfree(cluster_data
);
2266 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2271 static int qcow2_inactivate(BlockDriverState
*bs
)
2273 BDRVQcow2State
*s
= bs
->opaque
;
2274 int ret
, result
= 0;
2275 Error
*local_err
= NULL
;
2277 qcow2_store_persistent_dirty_bitmaps(bs
, &local_err
);
2278 if (local_err
!= NULL
) {
2280 error_reportf_err(local_err
, "Lost persistent bitmaps during "
2281 "inactivation of node '%s': ",
2282 bdrv_get_device_or_node_name(bs
));
2285 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2288 error_report("Failed to flush the L2 table cache: %s",
2292 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2295 error_report("Failed to flush the refcount block cache: %s",
2300 qcow2_mark_clean(bs
);
2306 static void qcow2_close(BlockDriverState
*bs
)
2308 BDRVQcow2State
*s
= bs
->opaque
;
2309 qemu_vfree(s
->l1_table
);
2310 /* else pre-write overlap checks in cache_destroy may crash */
2313 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2314 qcow2_inactivate(bs
);
2317 cache_clean_timer_del(bs
);
2318 qcow2_cache_destroy(s
->l2_table_cache
);
2319 qcow2_cache_destroy(s
->refcount_block_cache
);
2321 qcrypto_block_free(s
->crypto
);
2324 g_free(s
->unknown_header_fields
);
2325 cleanup_unknown_header_ext(bs
);
2327 g_free(s
->image_data_file
);
2328 g_free(s
->image_backing_file
);
2329 g_free(s
->image_backing_format
);
2331 if (has_data_file(bs
)) {
2332 bdrv_unref_child(bs
, s
->data_file
);
2335 qcow2_refcount_close(bs
);
2336 qcow2_free_snapshots(bs
);
2339 static void coroutine_fn
qcow2_co_invalidate_cache(BlockDriverState
*bs
,
2342 BDRVQcow2State
*s
= bs
->opaque
;
2343 int flags
= s
->flags
;
2344 QCryptoBlock
*crypto
= NULL
;
2346 Error
*local_err
= NULL
;
2350 * Backing files are read-only which makes all of their metadata immutable,
2351 * that means we don't have to worry about reopening them here.
2359 memset(s
, 0, sizeof(BDRVQcow2State
));
2360 options
= qdict_clone_shallow(bs
->options
);
2362 flags
&= ~BDRV_O_INACTIVE
;
2363 qemu_co_mutex_lock(&s
->lock
);
2364 ret
= qcow2_do_open(bs
, options
, flags
, &local_err
);
2365 qemu_co_mutex_unlock(&s
->lock
);
2366 qobject_unref(options
);
2368 error_propagate_prepend(errp
, local_err
,
2369 "Could not reopen qcow2 layer: ");
2372 } else if (ret
< 0) {
2373 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
2381 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2382 size_t len
, size_t buflen
)
2384 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2385 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2387 if (buflen
< ext_len
) {
2391 *ext_backing_fmt
= (QCowExtension
) {
2392 .magic
= cpu_to_be32(magic
),
2393 .len
= cpu_to_be32(len
),
2397 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2404 * Updates the qcow2 header, including the variable length parts of it, i.e.
2405 * the backing file name and all extensions. qcow2 was not designed to allow
2406 * such changes, so if we run out of space (we can only use the first cluster)
2407 * this function may fail.
2409 * Returns 0 on success, -errno in error cases.
2411 int qcow2_update_header(BlockDriverState
*bs
)
2413 BDRVQcow2State
*s
= bs
->opaque
;
2416 size_t buflen
= s
->cluster_size
;
2418 uint64_t total_size
;
2419 uint32_t refcount_table_clusters
;
2420 size_t header_length
;
2421 Qcow2UnknownHeaderExtension
*uext
;
2423 buf
= qemu_blockalign(bs
, buflen
);
2425 /* Header structure */
2426 header
= (QCowHeader
*) buf
;
2428 if (buflen
< sizeof(*header
)) {
2433 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2434 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2435 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2437 *header
= (QCowHeader
) {
2438 /* Version 2 fields */
2439 .magic
= cpu_to_be32(QCOW_MAGIC
),
2440 .version
= cpu_to_be32(s
->qcow_version
),
2441 .backing_file_offset
= 0,
2442 .backing_file_size
= 0,
2443 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2444 .size
= cpu_to_be64(total_size
),
2445 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2446 .l1_size
= cpu_to_be32(s
->l1_size
),
2447 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2448 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2449 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2450 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2451 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2453 /* Version 3 fields */
2454 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2455 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2456 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2457 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2458 .header_length
= cpu_to_be32(header_length
),
2461 /* For older versions, write a shorter header */
2462 switch (s
->qcow_version
) {
2464 ret
= offsetof(QCowHeader
, incompatible_features
);
2467 ret
= sizeof(*header
);
2476 memset(buf
, 0, buflen
);
2478 /* Preserve any unknown field in the header */
2479 if (s
->unknown_header_fields_size
) {
2480 if (buflen
< s
->unknown_header_fields_size
) {
2485 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2486 buf
+= s
->unknown_header_fields_size
;
2487 buflen
-= s
->unknown_header_fields_size
;
2490 /* Backing file format header extension */
2491 if (s
->image_backing_format
) {
2492 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2493 s
->image_backing_format
,
2494 strlen(s
->image_backing_format
),
2504 /* External data file header extension */
2505 if (has_data_file(bs
) && s
->image_data_file
) {
2506 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_DATA_FILE
,
2507 s
->image_data_file
, strlen(s
->image_data_file
),
2517 /* Full disk encryption header pointer extension */
2518 if (s
->crypto_header
.offset
!= 0) {
2519 s
->crypto_header
.offset
= cpu_to_be64(s
->crypto_header
.offset
);
2520 s
->crypto_header
.length
= cpu_to_be64(s
->crypto_header
.length
);
2521 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2522 &s
->crypto_header
, sizeof(s
->crypto_header
),
2524 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
2525 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
2534 if (s
->qcow_version
>= 3) {
2535 Qcow2Feature features
[] = {
2537 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2538 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2539 .name
= "dirty bit",
2542 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2543 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
2544 .name
= "corrupt bit",
2547 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2548 .bit
= QCOW2_INCOMPAT_DATA_FILE_BITNR
,
2549 .name
= "external data file",
2552 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
2553 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
2554 .name
= "lazy refcounts",
2558 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
2559 features
, sizeof(features
), buflen
);
2567 /* Bitmap extension */
2568 if (s
->nb_bitmaps
> 0) {
2569 Qcow2BitmapHeaderExt bitmaps_header
= {
2570 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
2571 .bitmap_directory_size
=
2572 cpu_to_be64(s
->bitmap_directory_size
),
2573 .bitmap_directory_offset
=
2574 cpu_to_be64(s
->bitmap_directory_offset
)
2576 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
2577 &bitmaps_header
, sizeof(bitmaps_header
),
2586 /* Keep unknown header extensions */
2587 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
2588 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
2597 /* End of header extensions */
2598 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
2606 /* Backing file name */
2607 if (s
->image_backing_file
) {
2608 size_t backing_file_len
= strlen(s
->image_backing_file
);
2610 if (buflen
< backing_file_len
) {
2615 /* Using strncpy is ok here, since buf is not NUL-terminated. */
2616 strncpy(buf
, s
->image_backing_file
, buflen
);
2618 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
2619 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
2622 /* Write the new header */
2623 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
2634 static int qcow2_change_backing_file(BlockDriverState
*bs
,
2635 const char *backing_file
, const char *backing_fmt
)
2637 BDRVQcow2State
*s
= bs
->opaque
;
2639 /* Adding a backing file means that the external data file alone won't be
2640 * enough to make sense of the content */
2641 if (backing_file
&& data_file_is_raw(bs
)) {
2645 if (backing_file
&& strlen(backing_file
) > 1023) {
2649 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2650 backing_file
?: "");
2651 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
2652 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
2654 g_free(s
->image_backing_file
);
2655 g_free(s
->image_backing_format
);
2657 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
2658 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
2660 return qcow2_update_header(bs
);
2663 static int qcow2_crypt_method_from_format(const char *encryptfmt
)
2665 if (g_str_equal(encryptfmt
, "luks")) {
2666 return QCOW_CRYPT_LUKS
;
2667 } else if (g_str_equal(encryptfmt
, "aes")) {
2668 return QCOW_CRYPT_AES
;
2674 static int qcow2_set_up_encryption(BlockDriverState
*bs
,
2675 QCryptoBlockCreateOptions
*cryptoopts
,
2678 BDRVQcow2State
*s
= bs
->opaque
;
2679 QCryptoBlock
*crypto
= NULL
;
2682 switch (cryptoopts
->format
) {
2683 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
2684 fmt
= QCOW_CRYPT_LUKS
;
2686 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
2687 fmt
= QCOW_CRYPT_AES
;
2690 error_setg(errp
, "Crypto format not supported in qcow2");
2694 s
->crypt_method_header
= fmt
;
2696 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
2697 qcow2_crypto_hdr_init_func
,
2698 qcow2_crypto_hdr_write_func
,
2704 ret
= qcow2_update_header(bs
);
2706 error_setg_errno(errp
, -ret
, "Could not write encryption header");
2712 qcrypto_block_free(crypto
);
2717 * Preallocates metadata structures for data clusters between @offset (in the
2718 * guest disk) and @new_length (which is thus generally the new guest disk
2721 * Returns: 0 on success, -errno on failure.
2723 static int coroutine_fn
preallocate_co(BlockDriverState
*bs
, uint64_t offset
,
2724 uint64_t new_length
)
2727 uint64_t host_offset
= 0;
2728 unsigned int cur_bytes
;
2732 assert(offset
<= new_length
);
2733 bytes
= new_length
- offset
;
2736 cur_bytes
= MIN(bytes
, INT_MAX
);
2737 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2738 &host_offset
, &meta
);
2744 QCowL2Meta
*next
= meta
->next
;
2746 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
2748 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
2749 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
2753 /* There are no dependent requests, but we need to remove our
2754 * request from the list of in-flight requests */
2755 QLIST_REMOVE(meta
, next_in_flight
);
2761 /* TODO Preallocate data if requested */
2764 offset
+= cur_bytes
;
2768 * It is expected that the image file is large enough to actually contain
2769 * all of the allocated clusters (otherwise we get failing reads after
2770 * EOF). Extend the image to the last allocated sector.
2772 if (host_offset
!= 0) {
2774 ret
= bdrv_pwrite(bs
->file
, (host_offset
+ cur_bytes
) - 1,
2784 /* qcow2_refcount_metadata_size:
2785 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2786 * @cluster_size: size of a cluster, in bytes
2787 * @refcount_order: refcount bits power-of-2 exponent
2788 * @generous_increase: allow for the refcount table to be 1.5x as large as it
2791 * Returns: Number of bytes required for refcount blocks and table metadata.
2793 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
2794 int refcount_order
, bool generous_increase
,
2795 uint64_t *refblock_count
)
2798 * Every host cluster is reference-counted, including metadata (even
2799 * refcount metadata is recursively included).
2801 * An accurate formula for the size of refcount metadata size is difficult
2802 * to derive. An easier method of calculation is finding the fixed point
2803 * where no further refcount blocks or table clusters are required to
2804 * reference count every cluster.
2806 int64_t blocks_per_table_cluster
= cluster_size
/ sizeof(uint64_t);
2807 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
2808 int64_t table
= 0; /* number of refcount table clusters */
2809 int64_t blocks
= 0; /* number of refcount block clusters */
2815 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
2816 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
2817 n
= clusters
+ blocks
+ table
;
2819 if (n
== last
&& generous_increase
) {
2820 clusters
+= DIV_ROUND_UP(table
, 2);
2821 n
= 0; /* force another loop */
2822 generous_increase
= false;
2824 } while (n
!= last
);
2826 if (refblock_count
) {
2827 *refblock_count
= blocks
;
2830 return (blocks
+ table
) * cluster_size
;
2834 * qcow2_calc_prealloc_size:
2835 * @total_size: virtual disk size in bytes
2836 * @cluster_size: cluster size in bytes
2837 * @refcount_order: refcount bits power-of-2 exponent
2839 * Returns: Total number of bytes required for the fully allocated image
2840 * (including metadata).
2842 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
2843 size_t cluster_size
,
2846 int64_t meta_size
= 0;
2847 uint64_t nl1e
, nl2e
;
2848 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
2850 /* header: 1 cluster */
2851 meta_size
+= cluster_size
;
2853 /* total size of L2 tables */
2854 nl2e
= aligned_total_size
/ cluster_size
;
2855 nl2e
= ROUND_UP(nl2e
, cluster_size
/ sizeof(uint64_t));
2856 meta_size
+= nl2e
* sizeof(uint64_t);
2858 /* total size of L1 tables */
2859 nl1e
= nl2e
* sizeof(uint64_t) / cluster_size
;
2860 nl1e
= ROUND_UP(nl1e
, cluster_size
/ sizeof(uint64_t));
2861 meta_size
+= nl1e
* sizeof(uint64_t);
2863 /* total size of refcount table and blocks */
2864 meta_size
+= qcow2_refcount_metadata_size(
2865 (meta_size
+ aligned_total_size
) / cluster_size
,
2866 cluster_size
, refcount_order
, false, NULL
);
2868 return meta_size
+ aligned_total_size
;
2871 static bool validate_cluster_size(size_t cluster_size
, Error
**errp
)
2873 int cluster_bits
= ctz32(cluster_size
);
2874 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
2875 (1 << cluster_bits
) != cluster_size
)
2877 error_setg(errp
, "Cluster size must be a power of two between %d and "
2878 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
2884 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, Error
**errp
)
2886 size_t cluster_size
;
2888 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
2889 DEFAULT_CLUSTER_SIZE
);
2890 if (!validate_cluster_size(cluster_size
, errp
)) {
2893 return cluster_size
;
2896 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
2901 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
2903 ret
= 3; /* default */
2904 } else if (!strcmp(buf
, "0.10")) {
2906 } else if (!strcmp(buf
, "1.1")) {
2909 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
2916 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
2919 uint64_t refcount_bits
;
2921 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
2922 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
2923 error_setg(errp
, "Refcount width must be a power of two and may not "
2928 if (version
< 3 && refcount_bits
!= 16) {
2929 error_setg(errp
, "Different refcount widths than 16 bits require "
2930 "compatibility level 1.1 or above (use compat=1.1 or "
2935 return refcount_bits
;
2938 static int coroutine_fn
2939 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
2941 BlockdevCreateOptionsQcow2
*qcow2_opts
;
2945 * Open the image file and write a minimal qcow2 header.
2947 * We keep things simple and start with a zero-sized image. We also
2948 * do without refcount blocks or a L1 table for now. We'll fix the
2949 * inconsistency later.
2951 * We do need a refcount table because growing the refcount table means
2952 * allocating two new refcount blocks - the seconds of which would be at
2953 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2954 * size for any qcow2 image.
2956 BlockBackend
*blk
= NULL
;
2957 BlockDriverState
*bs
= NULL
;
2958 BlockDriverState
*data_bs
= NULL
;
2960 size_t cluster_size
;
2963 uint64_t* refcount_table
;
2964 Error
*local_err
= NULL
;
2967 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
2968 qcow2_opts
= &create_options
->u
.qcow2
;
2970 bs
= bdrv_open_blockdev_ref(qcow2_opts
->file
, errp
);
2975 /* Validate options and set default values */
2976 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
2977 error_setg(errp
, "Image size must be a multiple of 512 bytes");
2982 if (qcow2_opts
->has_version
) {
2983 switch (qcow2_opts
->version
) {
2984 case BLOCKDEV_QCOW2_VERSION_V2
:
2987 case BLOCKDEV_QCOW2_VERSION_V3
:
2991 g_assert_not_reached();
2997 if (qcow2_opts
->has_cluster_size
) {
2998 cluster_size
= qcow2_opts
->cluster_size
;
3000 cluster_size
= DEFAULT_CLUSTER_SIZE
;
3003 if (!validate_cluster_size(cluster_size
, errp
)) {
3008 if (!qcow2_opts
->has_preallocation
) {
3009 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
3011 if (qcow2_opts
->has_backing_file
&&
3012 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
)
3014 error_setg(errp
, "Backing file and preallocation cannot be used at "
3019 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->has_backing_file
) {
3020 error_setg(errp
, "Backing format cannot be used without backing file");
3025 if (!qcow2_opts
->has_lazy_refcounts
) {
3026 qcow2_opts
->lazy_refcounts
= false;
3028 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
3029 error_setg(errp
, "Lazy refcounts only supported with compatibility "
3030 "level 1.1 and above (use version=v3 or greater)");
3035 if (!qcow2_opts
->has_refcount_bits
) {
3036 qcow2_opts
->refcount_bits
= 16;
3038 if (qcow2_opts
->refcount_bits
> 64 ||
3039 !is_power_of_2(qcow2_opts
->refcount_bits
))
3041 error_setg(errp
, "Refcount width must be a power of two and may not "
3046 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
3047 error_setg(errp
, "Different refcount widths than 16 bits require "
3048 "compatibility level 1.1 or above (use version=v3 or "
3053 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
3055 if (qcow2_opts
->data_file_raw
&& !qcow2_opts
->data_file
) {
3056 error_setg(errp
, "data-file-raw requires data-file");
3060 if (qcow2_opts
->data_file_raw
&& qcow2_opts
->has_backing_file
) {
3061 error_setg(errp
, "Backing file and data-file-raw cannot be used at "
3067 if (qcow2_opts
->data_file
) {
3069 error_setg(errp
, "External data files are only supported with "
3070 "compatibility level 1.1 and above (use version=v3 or "
3075 data_bs
= bdrv_open_blockdev_ref(qcow2_opts
->data_file
, errp
);
3076 if (data_bs
== NULL
) {
3082 /* Create BlockBackend to write to the image */
3083 blk
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
);
3084 ret
= blk_insert_bs(blk
, bs
, errp
);
3088 blk_set_allow_write_beyond_eof(blk
, true);
3090 /* Clear the protocol layer and preallocate it if necessary */
3091 ret
= blk_truncate(blk
, 0, PREALLOC_MODE_OFF
, errp
);
3096 /* Write the header */
3097 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
3098 header
= g_malloc0(cluster_size
);
3099 *header
= (QCowHeader
) {
3100 .magic
= cpu_to_be32(QCOW_MAGIC
),
3101 .version
= cpu_to_be32(version
),
3102 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
3103 .size
= cpu_to_be64(0),
3104 .l1_table_offset
= cpu_to_be64(0),
3105 .l1_size
= cpu_to_be32(0),
3106 .refcount_table_offset
= cpu_to_be64(cluster_size
),
3107 .refcount_table_clusters
= cpu_to_be32(1),
3108 .refcount_order
= cpu_to_be32(refcount_order
),
3109 .header_length
= cpu_to_be32(sizeof(*header
)),
3112 /* We'll update this to correct value later */
3113 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
3115 if (qcow2_opts
->lazy_refcounts
) {
3116 header
->compatible_features
|=
3117 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
3120 header
->incompatible_features
|=
3121 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE
);
3123 if (qcow2_opts
->data_file_raw
) {
3124 header
->autoclear_features
|=
3125 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW
);
3128 ret
= blk_pwrite(blk
, 0, header
, cluster_size
, 0);
3131 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
3135 /* Write a refcount table with one refcount block */
3136 refcount_table
= g_malloc0(2 * cluster_size
);
3137 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
3138 ret
= blk_pwrite(blk
, cluster_size
, refcount_table
, 2 * cluster_size
, 0);
3139 g_free(refcount_table
);
3142 error_setg_errno(errp
, -ret
, "Could not write refcount table");
3150 * And now open the image and make it consistent first (i.e. increase the
3151 * refcount of the cluster that is occupied by the header and the refcount
3154 options
= qdict_new();
3155 qdict_put_str(options
, "driver", "qcow2");
3156 qdict_put_str(options
, "file", bs
->node_name
);
3158 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3160 blk
= blk_new_open(NULL
, NULL
, options
,
3161 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
3164 error_propagate(errp
, local_err
);
3169 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
3171 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
3172 "header and refcount table");
3175 } else if (ret
!= 0) {
3176 error_report("Huh, first cluster in empty image is already in use?");
3180 /* Set the external data file if necessary */
3182 BDRVQcow2State
*s
= blk_bs(blk
)->opaque
;
3183 s
->image_data_file
= g_strdup(data_bs
->filename
);
3186 /* Create a full header (including things like feature table) */
3187 ret
= qcow2_update_header(blk_bs(blk
));
3189 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
3193 /* Okay, now that we have a valid image, let's give it the right size */
3194 ret
= blk_truncate(blk
, qcow2_opts
->size
, qcow2_opts
->preallocation
, errp
);
3196 error_prepend(errp
, "Could not resize image: ");
3200 /* Want a backing file? There you go.*/
3201 if (qcow2_opts
->has_backing_file
) {
3202 const char *backing_format
= NULL
;
3204 if (qcow2_opts
->has_backing_fmt
) {
3205 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
3208 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
3211 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
3212 "with format '%s'", qcow2_opts
->backing_file
,
3218 /* Want encryption? There you go. */
3219 if (qcow2_opts
->has_encrypt
) {
3220 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3229 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3230 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3231 * have to setup decryption context. We're not doing any I/O on the top
3232 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3235 options
= qdict_new();
3236 qdict_put_str(options
, "driver", "qcow2");
3237 qdict_put_str(options
, "file", bs
->node_name
);
3239 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3241 blk
= blk_new_open(NULL
, NULL
, options
,
3242 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3245 error_propagate(errp
, local_err
);
3254 bdrv_unref(data_bs
);
3258 static int coroutine_fn
qcow2_co_create_opts(const char *filename
, QemuOpts
*opts
,
3261 BlockdevCreateOptions
*create_options
= NULL
;
3264 BlockDriverState
*bs
= NULL
;
3265 BlockDriverState
*data_bs
= NULL
;
3266 Error
*local_err
= NULL
;
3270 /* Only the keyval visitor supports the dotted syntax needed for
3271 * encryption, so go through a QDict before getting a QAPI type. Ignore
3272 * options meant for the protocol layer so that the visitor doesn't
3274 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3277 /* Handle encryption options */
3278 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3279 if (val
&& !strcmp(val
, "on")) {
3280 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3281 } else if (val
&& !strcmp(val
, "off")) {
3282 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3285 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3286 if (val
&& !strcmp(val
, "aes")) {
3287 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3290 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3291 * version=v2/v3 below. */
3292 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3293 if (val
&& !strcmp(val
, "0.10")) {
3294 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3295 } else if (val
&& !strcmp(val
, "1.1")) {
3296 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3299 /* Change legacy command line options into QMP ones */
3300 static const QDictRenames opt_renames
[] = {
3301 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3302 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3303 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3304 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3305 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3306 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3307 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3308 { BLOCK_OPT_DATA_FILE_RAW
, "data-file-raw" },
3312 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3317 /* Create and open the file (protocol layer) */
3318 ret
= bdrv_create_file(filename
, opts
, errp
);
3323 bs
= bdrv_open(filename
, NULL
, NULL
,
3324 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3330 /* Create and open an external data file (protocol layer) */
3331 val
= qdict_get_try_str(qdict
, BLOCK_OPT_DATA_FILE
);
3333 ret
= bdrv_create_file(val
, opts
, errp
);
3338 data_bs
= bdrv_open(val
, NULL
, NULL
,
3339 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
,
3341 if (data_bs
== NULL
) {
3346 qdict_del(qdict
, BLOCK_OPT_DATA_FILE
);
3347 qdict_put_str(qdict
, "data-file", data_bs
->node_name
);
3350 /* Set 'driver' and 'node' options */
3351 qdict_put_str(qdict
, "driver", "qcow2");
3352 qdict_put_str(qdict
, "file", bs
->node_name
);
3354 /* Now get the QAPI type BlockdevCreateOptions */
3355 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
3361 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, &local_err
);
3365 error_propagate(errp
, local_err
);
3370 /* Silently round up size */
3371 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3374 /* Create the qcow2 image (format layer) */
3375 ret
= qcow2_co_create(create_options
, errp
);
3382 qobject_unref(qdict
);
3384 bdrv_unref(data_bs
);
3385 qapi_free_BlockdevCreateOptions(create_options
);
3390 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3395 /* Clamp to image length, before checking status of underlying sectors */
3396 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3397 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3403 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3404 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
== bytes
;
3407 static coroutine_fn
int qcow2_co_pwrite_zeroes(BlockDriverState
*bs
,
3408 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3411 BDRVQcow2State
*s
= bs
->opaque
;
3413 uint32_t head
= offset
% s
->cluster_size
;
3414 uint32_t tail
= (offset
+ bytes
) % s
->cluster_size
;
3416 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3417 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3425 assert(head
+ bytes
<= s
->cluster_size
);
3427 /* check whether remainder of cluster already reads as zero */
3428 if (!(is_zero(bs
, offset
- head
, head
) &&
3429 is_zero(bs
, offset
+ bytes
,
3430 tail
? s
->cluster_size
- tail
: 0))) {
3434 qemu_co_mutex_lock(&s
->lock
);
3435 /* We can have new write after previous check */
3436 offset
= QEMU_ALIGN_DOWN(offset
, s
->cluster_size
);
3437 bytes
= s
->cluster_size
;
3438 nr
= s
->cluster_size
;
3439 ret
= qcow2_get_cluster_offset(bs
, offset
, &nr
, &off
);
3440 if (ret
!= QCOW2_CLUSTER_UNALLOCATED
&&
3441 ret
!= QCOW2_CLUSTER_ZERO_PLAIN
&&
3442 ret
!= QCOW2_CLUSTER_ZERO_ALLOC
) {
3443 qemu_co_mutex_unlock(&s
->lock
);
3447 qemu_co_mutex_lock(&s
->lock
);
3450 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
3452 /* Whatever is left can use real zero clusters */
3453 ret
= qcow2_cluster_zeroize(bs
, offset
, bytes
, flags
);
3454 qemu_co_mutex_unlock(&s
->lock
);
3459 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
3460 int64_t offset
, int bytes
)
3463 BDRVQcow2State
*s
= bs
->opaque
;
3465 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
3466 assert(bytes
< s
->cluster_size
);
3467 /* Ignore partial clusters, except for the special case of the
3468 * complete partial cluster at the end of an unaligned file */
3469 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
3470 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3475 qemu_co_mutex_lock(&s
->lock
);
3476 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
3478 qemu_co_mutex_unlock(&s
->lock
);
3482 static int coroutine_fn
3483 qcow2_co_copy_range_from(BlockDriverState
*bs
,
3484 BdrvChild
*src
, uint64_t src_offset
,
3485 BdrvChild
*dst
, uint64_t dst_offset
,
3486 uint64_t bytes
, BdrvRequestFlags read_flags
,
3487 BdrvRequestFlags write_flags
)
3489 BDRVQcow2State
*s
= bs
->opaque
;
3491 unsigned int cur_bytes
; /* number of bytes in current iteration */
3492 BdrvChild
*child
= NULL
;
3493 BdrvRequestFlags cur_write_flags
;
3495 assert(!bs
->encrypted
);
3496 qemu_co_mutex_lock(&s
->lock
);
3498 while (bytes
!= 0) {
3499 uint64_t copy_offset
= 0;
3500 /* prepare next request */
3501 cur_bytes
= MIN(bytes
, INT_MAX
);
3502 cur_write_flags
= write_flags
;
3504 ret
= qcow2_get_cluster_offset(bs
, src_offset
, &cur_bytes
, ©_offset
);
3510 case QCOW2_CLUSTER_UNALLOCATED
:
3511 if (bs
->backing
&& bs
->backing
->bs
) {
3512 int64_t backing_length
= bdrv_getlength(bs
->backing
->bs
);
3513 if (src_offset
>= backing_length
) {
3514 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3516 child
= bs
->backing
;
3517 cur_bytes
= MIN(cur_bytes
, backing_length
- src_offset
);
3518 copy_offset
= src_offset
;
3521 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3525 case QCOW2_CLUSTER_ZERO_PLAIN
:
3526 case QCOW2_CLUSTER_ZERO_ALLOC
:
3527 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3530 case QCOW2_CLUSTER_COMPRESSED
:
3534 case QCOW2_CLUSTER_NORMAL
:
3535 child
= s
->data_file
;
3536 copy_offset
+= offset_into_cluster(s
, src_offset
);
3537 if ((copy_offset
& 511) != 0) {
3546 qemu_co_mutex_unlock(&s
->lock
);
3547 ret
= bdrv_co_copy_range_from(child
,
3550 cur_bytes
, read_flags
, cur_write_flags
);
3551 qemu_co_mutex_lock(&s
->lock
);
3557 src_offset
+= cur_bytes
;
3558 dst_offset
+= cur_bytes
;
3563 qemu_co_mutex_unlock(&s
->lock
);
3567 static int coroutine_fn
3568 qcow2_co_copy_range_to(BlockDriverState
*bs
,
3569 BdrvChild
*src
, uint64_t src_offset
,
3570 BdrvChild
*dst
, uint64_t dst_offset
,
3571 uint64_t bytes
, BdrvRequestFlags read_flags
,
3572 BdrvRequestFlags write_flags
)
3574 BDRVQcow2State
*s
= bs
->opaque
;
3575 int offset_in_cluster
;
3577 unsigned int cur_bytes
; /* number of sectors in current iteration */
3578 uint64_t cluster_offset
;
3579 QCowL2Meta
*l2meta
= NULL
;
3581 assert(!bs
->encrypted
);
3583 qemu_co_mutex_lock(&s
->lock
);
3585 while (bytes
!= 0) {
3589 offset_in_cluster
= offset_into_cluster(s
, dst_offset
);
3590 cur_bytes
= MIN(bytes
, INT_MAX
);
3593 * If src->bs == dst->bs, we could simply copy by incrementing
3594 * the refcnt, without copying user data.
3595 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
3596 ret
= qcow2_alloc_cluster_offset(bs
, dst_offset
, &cur_bytes
,
3597 &cluster_offset
, &l2meta
);
3602 assert((cluster_offset
& 511) == 0);
3604 ret
= qcow2_pre_write_overlap_check(bs
, 0,
3605 cluster_offset
+ offset_in_cluster
, cur_bytes
, true);
3610 qemu_co_mutex_unlock(&s
->lock
);
3611 ret
= bdrv_co_copy_range_to(src
, src_offset
,
3613 cluster_offset
+ offset_in_cluster
,
3614 cur_bytes
, read_flags
, write_flags
);
3615 qemu_co_mutex_lock(&s
->lock
);
3620 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
3626 src_offset
+= cur_bytes
;
3627 dst_offset
+= cur_bytes
;
3632 qcow2_handle_l2meta(bs
, &l2meta
, false);
3634 qemu_co_mutex_unlock(&s
->lock
);
3636 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
3641 static int coroutine_fn
qcow2_co_truncate(BlockDriverState
*bs
, int64_t offset
,
3642 PreallocMode prealloc
, Error
**errp
)
3644 BDRVQcow2State
*s
= bs
->opaque
;
3645 uint64_t old_length
;
3646 int64_t new_l1_size
;
3650 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
3651 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
3653 error_setg(errp
, "Unsupported preallocation mode '%s'",
3654 PreallocMode_str(prealloc
));
3659 error_setg(errp
, "The new size must be a multiple of 512");
3663 qemu_co_mutex_lock(&s
->lock
);
3665 /* cannot proceed if image has snapshots */
3666 if (s
->nb_snapshots
) {
3667 error_setg(errp
, "Can't resize an image which has snapshots");
3672 /* cannot proceed if image has bitmaps */
3673 if (qcow2_truncate_bitmaps_check(bs
, errp
)) {
3678 old_length
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3679 new_l1_size
= size_to_l1(s
, offset
);
3681 if (offset
< old_length
) {
3682 int64_t last_cluster
, old_file_size
;
3683 if (prealloc
!= PREALLOC_MODE_OFF
) {
3685 "Preallocation can't be used for shrinking an image");
3690 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
3691 old_length
- ROUND_UP(offset
,
3693 QCOW2_DISCARD_ALWAYS
, true);
3695 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
3699 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
3701 error_setg_errno(errp
, -ret
,
3702 "Failed to reduce the number of L2 tables");
3706 ret
= qcow2_shrink_reftable(bs
);
3708 error_setg_errno(errp
, -ret
,
3709 "Failed to discard unused refblocks");
3713 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3714 if (old_file_size
< 0) {
3715 error_setg_errno(errp
, -old_file_size
,
3716 "Failed to inquire current file length");
3717 ret
= old_file_size
;
3720 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
3721 if (last_cluster
< 0) {
3722 error_setg_errno(errp
, -last_cluster
,
3723 "Failed to find the last cluster");
3727 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
3728 Error
*local_err
= NULL
;
3730 bdrv_co_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
3731 PREALLOC_MODE_OFF
, &local_err
);
3733 warn_reportf_err(local_err
,
3734 "Failed to truncate the tail of the image: ");
3738 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
3740 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
3746 case PREALLOC_MODE_OFF
:
3749 case PREALLOC_MODE_METADATA
:
3750 ret
= preallocate_co(bs
, old_length
, offset
);
3752 error_setg_errno(errp
, -ret
, "Preallocation failed");
3757 case PREALLOC_MODE_FALLOC
:
3758 case PREALLOC_MODE_FULL
:
3760 int64_t allocation_start
, host_offset
, guest_offset
;
3761 int64_t clusters_allocated
;
3762 int64_t old_file_size
, new_file_size
;
3763 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
3765 /* With a data file, preallocation means just allocating the metadata
3766 * and forwarding the truncate request to the data file */
3767 if (has_data_file(bs
)) {
3768 ret
= preallocate_co(bs
, old_length
, offset
);
3770 error_setg_errno(errp
, -ret
, "Preallocation failed");
3776 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3777 if (old_file_size
< 0) {
3778 error_setg_errno(errp
, -old_file_size
,
3779 "Failed to inquire current file length");
3780 ret
= old_file_size
;
3783 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
3785 nb_new_data_clusters
= DIV_ROUND_UP(offset
- old_length
,
3788 /* This is an overestimation; we will not actually allocate space for
3789 * these in the file but just make sure the new refcount structures are
3790 * able to cover them so we will not have to allocate new refblocks
3791 * while entering the data blocks in the potentially new L2 tables.
3792 * (We do not actually care where the L2 tables are placed. Maybe they
3793 * are already allocated or they can be placed somewhere before
3794 * @old_file_size. It does not matter because they will be fully
3795 * allocated automatically, so they do not need to be covered by the
3796 * preallocation. All that matters is that we will not have to allocate
3797 * new refcount structures for them.) */
3798 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
3799 s
->cluster_size
/ sizeof(uint64_t));
3800 /* The cluster range may not be aligned to L2 boundaries, so add one L2
3801 * table for a potential head/tail */
3804 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
3805 nb_new_data_clusters
+
3808 if (allocation_start
< 0) {
3809 error_setg_errno(errp
, -allocation_start
,
3810 "Failed to resize refcount structures");
3811 ret
= allocation_start
;
3815 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
3816 nb_new_data_clusters
);
3817 if (clusters_allocated
< 0) {
3818 error_setg_errno(errp
, -clusters_allocated
,
3819 "Failed to allocate data clusters");
3820 ret
= clusters_allocated
;
3824 assert(clusters_allocated
== nb_new_data_clusters
);
3826 /* Allocate the data area */
3827 new_file_size
= allocation_start
+
3828 nb_new_data_clusters
* s
->cluster_size
;
3829 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, prealloc
, errp
);
3831 error_prepend(errp
, "Failed to resize underlying file: ");
3832 qcow2_free_clusters(bs
, allocation_start
,
3833 nb_new_data_clusters
* s
->cluster_size
,
3834 QCOW2_DISCARD_OTHER
);
3838 /* Create the necessary L2 entries */
3839 host_offset
= allocation_start
;
3840 guest_offset
= old_length
;
3841 while (nb_new_data_clusters
) {
3842 int64_t nb_clusters
= MIN(
3843 nb_new_data_clusters
,
3844 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
3845 QCowL2Meta allocation
= {
3846 .offset
= guest_offset
,
3847 .alloc_offset
= host_offset
,
3848 .nb_clusters
= nb_clusters
,
3850 qemu_co_queue_init(&allocation
.dependent_requests
);
3852 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
3854 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
3855 qcow2_free_clusters(bs
, host_offset
,
3856 nb_new_data_clusters
* s
->cluster_size
,
3857 QCOW2_DISCARD_OTHER
);
3861 guest_offset
+= nb_clusters
* s
->cluster_size
;
3862 host_offset
+= nb_clusters
* s
->cluster_size
;
3863 nb_new_data_clusters
-= nb_clusters
;
3869 g_assert_not_reached();
3872 if (prealloc
!= PREALLOC_MODE_OFF
) {
3873 /* Flush metadata before actually changing the image size */
3874 ret
= qcow2_write_caches(bs
);
3876 error_setg_errno(errp
, -ret
,
3877 "Failed to flush the preallocated area to disk");
3882 bs
->total_sectors
= offset
/ BDRV_SECTOR_SIZE
;
3884 if (has_data_file(bs
)) {
3885 if (prealloc
== PREALLOC_MODE_METADATA
) {
3886 prealloc
= PREALLOC_MODE_OFF
;
3888 ret
= bdrv_co_truncate(s
->data_file
, offset
, prealloc
, errp
);
3894 /* write updated header.size */
3895 offset
= cpu_to_be64(offset
);
3896 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
3897 &offset
, sizeof(uint64_t));
3899 error_setg_errno(errp
, -ret
, "Failed to update the image size");
3903 s
->l1_vm_state_index
= new_l1_size
;
3905 /* Update cache sizes */
3906 options
= qdict_clone_shallow(bs
->options
);
3907 ret
= qcow2_update_options(bs
, options
, s
->flags
, errp
);
3908 qobject_unref(options
);
3914 qemu_co_mutex_unlock(&s
->lock
);
3921 * @dest - destination buffer, @dest_size bytes
3922 * @src - source buffer, @src_size bytes
3924 * Returns: compressed size on success
3925 * -1 destination buffer is not enough to store compressed data
3926 * -2 on any other error
3928 static ssize_t
qcow2_compress(void *dest
, size_t dest_size
,
3929 const void *src
, size_t src_size
)
3934 /* best compression, small window, no zlib header */
3935 memset(&strm
, 0, sizeof(strm
));
3936 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
, Z_DEFLATED
,
3937 -12, 9, Z_DEFAULT_STRATEGY
);
3942 /* strm.next_in is not const in old zlib versions, such as those used on
3943 * OpenBSD/NetBSD, so cast the const away */
3944 strm
.avail_in
= src_size
;
3945 strm
.next_in
= (void *) src
;
3946 strm
.avail_out
= dest_size
;
3947 strm
.next_out
= dest
;
3949 ret
= deflate(&strm
, Z_FINISH
);
3950 if (ret
== Z_STREAM_END
) {
3951 ret
= dest_size
- strm
.avail_out
;
3953 ret
= (ret
== Z_OK
? -1 : -2);
3962 * qcow2_decompress()
3964 * Decompress some data (not more than @src_size bytes) to produce exactly
3967 * @dest - destination buffer, @dest_size bytes
3968 * @src - source buffer, @src_size bytes
3970 * Returns: 0 on success
3973 static ssize_t
qcow2_decompress(void *dest
, size_t dest_size
,
3974 const void *src
, size_t src_size
)
3979 memset(&strm
, 0, sizeof(strm
));
3980 strm
.avail_in
= src_size
;
3981 strm
.next_in
= (void *) src
;
3982 strm
.avail_out
= dest_size
;
3983 strm
.next_out
= dest
;
3985 ret
= inflateInit2(&strm
, -12);
3990 ret
= inflate(&strm
, Z_FINISH
);
3991 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) || strm
.avail_out
!= 0) {
3992 /* We approve Z_BUF_ERROR because we need @dest buffer to be filled, but
3993 * @src buffer may be processed partly (because in qcow2 we know size of
3994 * compressed data with precision of one sector) */
4003 #define MAX_COMPRESS_THREADS 4
4005 typedef ssize_t (*Qcow2CompressFunc
)(void *dest
, size_t dest_size
,
4006 const void *src
, size_t src_size
);
4007 typedef struct Qcow2CompressData
{
4014 Qcow2CompressFunc func
;
4015 } Qcow2CompressData
;
4017 static int qcow2_compress_pool_func(void *opaque
)
4019 Qcow2CompressData
*data
= opaque
;
4021 data
->ret
= data
->func(data
->dest
, data
->dest_size
,
4022 data
->src
, data
->src_size
);
4027 static void qcow2_compress_complete(void *opaque
, int ret
)
4029 qemu_coroutine_enter(opaque
);
4032 static ssize_t coroutine_fn
4033 qcow2_co_do_compress(BlockDriverState
*bs
, void *dest
, size_t dest_size
,
4034 const void *src
, size_t src_size
, Qcow2CompressFunc func
)
4036 BDRVQcow2State
*s
= bs
->opaque
;
4038 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
4039 Qcow2CompressData arg
= {
4041 .dest_size
= dest_size
,
4043 .src_size
= src_size
,
4047 while (s
->nb_compress_threads
>= MAX_COMPRESS_THREADS
) {
4048 qemu_co_queue_wait(&s
->compress_wait_queue
, NULL
);
4051 s
->nb_compress_threads
++;
4052 acb
= thread_pool_submit_aio(pool
, qcow2_compress_pool_func
, &arg
,
4053 qcow2_compress_complete
,
4054 qemu_coroutine_self());
4057 s
->nb_compress_threads
--;
4060 qemu_coroutine_yield();
4061 s
->nb_compress_threads
--;
4062 qemu_co_queue_next(&s
->compress_wait_queue
);
4067 static ssize_t coroutine_fn
4068 qcow2_co_compress(BlockDriverState
*bs
, void *dest
, size_t dest_size
,
4069 const void *src
, size_t src_size
)
4071 return qcow2_co_do_compress(bs
, dest
, dest_size
, src
, src_size
,
4075 static ssize_t coroutine_fn
4076 qcow2_co_decompress(BlockDriverState
*bs
, void *dest
, size_t dest_size
,
4077 const void *src
, size_t src_size
)
4079 return qcow2_co_do_compress(bs
, dest
, dest_size
, src
, src_size
,
4083 /* XXX: put compressed sectors first, then all the cluster aligned
4084 tables to avoid losing bytes in alignment */
4085 static coroutine_fn
int
4086 qcow2_co_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
4087 uint64_t bytes
, QEMUIOVector
*qiov
)
4089 BDRVQcow2State
*s
= bs
->opaque
;
4090 QEMUIOVector hd_qiov
;
4093 uint8_t *buf
, *out_buf
;
4094 uint64_t cluster_offset
;
4096 if (has_data_file(bs
)) {
4101 /* align end of file to a sector boundary to ease reading with
4102 sector based I/Os */
4103 int64_t len
= bdrv_getlength(bs
->file
->bs
);
4107 return bdrv_co_truncate(bs
->file
, len
, PREALLOC_MODE_OFF
, NULL
);
4110 if (offset_into_cluster(s
, offset
)) {
4114 buf
= qemu_blockalign(bs
, s
->cluster_size
);
4115 if (bytes
!= s
->cluster_size
) {
4116 if (bytes
> s
->cluster_size
||
4117 offset
+ bytes
!= bs
->total_sectors
<< BDRV_SECTOR_BITS
)
4122 /* Zero-pad last write if image size is not cluster aligned */
4123 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
4125 qemu_iovec_to_buf(qiov
, 0, buf
, bytes
);
4127 out_buf
= g_malloc(s
->cluster_size
);
4129 out_len
= qcow2_co_compress(bs
, out_buf
, s
->cluster_size
- 1,
4130 buf
, s
->cluster_size
);
4131 if (out_len
== -2) {
4134 } else if (out_len
== -1) {
4135 /* could not compress: write normal cluster */
4136 ret
= qcow2_co_pwritev(bs
, offset
, bytes
, qiov
, 0);
4143 qemu_co_mutex_lock(&s
->lock
);
4144 ret
= qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
,
4147 qemu_co_mutex_unlock(&s
->lock
);
4151 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
, true);
4152 qemu_co_mutex_unlock(&s
->lock
);
4157 qemu_iovec_init_buf(&hd_qiov
, out_buf
, out_len
);
4159 BLKDBG_EVENT(s
->data_file
, BLKDBG_WRITE_COMPRESSED
);
4160 ret
= bdrv_co_pwritev(s
->data_file
, cluster_offset
, out_len
, &hd_qiov
, 0);
4172 static int coroutine_fn
4173 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
4174 uint64_t file_cluster_offset
,
4179 BDRVQcow2State
*s
= bs
->opaque
;
4180 int ret
= 0, csize
, nb_csectors
;
4182 uint8_t *buf
, *out_buf
;
4183 QEMUIOVector local_qiov
;
4184 int offset_in_cluster
= offset_into_cluster(s
, offset
);
4186 coffset
= file_cluster_offset
& s
->cluster_offset_mask
;
4187 nb_csectors
= ((file_cluster_offset
>> s
->csize_shift
) & s
->csize_mask
) + 1;
4188 csize
= nb_csectors
* 512 - (coffset
& 511);
4190 buf
= g_try_malloc(csize
);
4194 qemu_iovec_init_buf(&local_qiov
, buf
, csize
);
4196 out_buf
= qemu_blockalign(bs
, s
->cluster_size
);
4198 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
4199 ret
= bdrv_co_preadv(bs
->file
, coffset
, csize
, &local_qiov
, 0);
4204 if (qcow2_co_decompress(bs
, out_buf
, s
->cluster_size
, buf
, csize
) < 0) {
4209 qemu_iovec_from_buf(qiov
, 0, out_buf
+ offset_in_cluster
, bytes
);
4212 qemu_vfree(out_buf
);
4218 static int make_completely_empty(BlockDriverState
*bs
)
4220 BDRVQcow2State
*s
= bs
->opaque
;
4221 Error
*local_err
= NULL
;
4222 int ret
, l1_clusters
;
4224 uint64_t *new_reftable
= NULL
;
4225 uint64_t rt_entry
, l1_size2
;
4228 uint64_t reftable_offset
;
4229 uint32_t reftable_clusters
;
4230 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
4232 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
4237 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
4242 /* Refcounts will be broken utterly */
4243 ret
= qcow2_mark_dirty(bs
);
4248 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4250 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
4251 l1_size2
= (uint64_t)s
->l1_size
* sizeof(uint64_t);
4253 /* After this call, neither the in-memory nor the on-disk refcount
4254 * information accurately describe the actual references */
4256 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
4257 l1_clusters
* s
->cluster_size
, 0);
4259 goto fail_broken_refcounts
;
4261 memset(s
->l1_table
, 0, l1_size2
);
4263 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
4265 /* Overwrite enough clusters at the beginning of the sectors to place
4266 * the refcount table, a refcount block and the L1 table in; this may
4267 * overwrite parts of the existing refcount and L1 table, which is not
4268 * an issue because the dirty flag is set, complete data loss is in fact
4269 * desired and partial data loss is consequently fine as well */
4270 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
4271 (2 + l1_clusters
) * s
->cluster_size
, 0);
4272 /* This call (even if it failed overall) may have overwritten on-disk
4273 * refcount structures; in that case, the in-memory refcount information
4274 * will probably differ from the on-disk information which makes the BDS
4277 goto fail_broken_refcounts
;
4280 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4281 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
4283 /* "Create" an empty reftable (one cluster) directly after the image
4284 * header and an empty L1 table three clusters after the image header;
4285 * the cluster between those two will be used as the first refblock */
4286 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
4287 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
4288 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
4289 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
4290 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
4292 goto fail_broken_refcounts
;
4295 s
->l1_table_offset
= 3 * s
->cluster_size
;
4297 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ sizeof(uint64_t));
4298 if (!new_reftable
) {
4300 goto fail_broken_refcounts
;
4303 s
->refcount_table_offset
= s
->cluster_size
;
4304 s
->refcount_table_size
= s
->cluster_size
/ sizeof(uint64_t);
4305 s
->max_refcount_table_index
= 0;
4307 g_free(s
->refcount_table
);
4308 s
->refcount_table
= new_reftable
;
4309 new_reftable
= NULL
;
4311 /* Now the in-memory refcount information again corresponds to the on-disk
4312 * information (reftable is empty and no refblocks (the refblock cache is
4313 * empty)); however, this means some clusters (e.g. the image header) are
4314 * referenced, but not refcounted, but the normal qcow2 code assumes that
4315 * the in-memory information is always correct */
4317 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
4319 /* Enter the first refblock into the reftable */
4320 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
4321 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
4322 &rt_entry
, sizeof(rt_entry
));
4324 goto fail_broken_refcounts
;
4326 s
->refcount_table
[0] = 2 * s
->cluster_size
;
4328 s
->free_cluster_index
= 0;
4329 assert(3 + l1_clusters
<= s
->refcount_block_size
);
4330 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
4333 goto fail_broken_refcounts
;
4334 } else if (offset
> 0) {
4335 error_report("First cluster in emptied image is in use");
4339 /* Now finally the in-memory information corresponds to the on-disk
4340 * structures and is correct */
4341 ret
= qcow2_mark_clean(bs
);
4346 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
,
4347 PREALLOC_MODE_OFF
, &local_err
);
4349 error_report_err(local_err
);
4355 fail_broken_refcounts
:
4356 /* The BDS is unusable at this point. If we wanted to make it usable, we
4357 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4358 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4359 * again. However, because the functions which could have caused this error
4360 * path to be taken are used by those functions as well, it's very likely
4361 * that that sequence will fail as well. Therefore, just eject the BDS. */
4365 g_free(new_reftable
);
4369 static int qcow2_make_empty(BlockDriverState
*bs
)
4371 BDRVQcow2State
*s
= bs
->opaque
;
4372 uint64_t offset
, end_offset
;
4373 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
4374 int l1_clusters
, ret
= 0;
4376 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
4378 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
4379 3 + l1_clusters
<= s
->refcount_block_size
&&
4380 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
4381 /* The following function only works for qcow2 v3 images (it
4382 * requires the dirty flag) and only as long as there are no
4383 * features that reserve extra clusters (such as snapshots,
4384 * LUKS header, or persistent bitmaps), because it completely
4385 * empties the image. Furthermore, the L1 table and three
4386 * additional clusters (image header, refcount table, one
4387 * refcount block) have to fit inside one refcount block. */
4388 return make_completely_empty(bs
);
4391 /* This fallback code simply discards every active cluster; this is slow,
4392 * but works in all cases */
4393 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4394 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
4395 /* As this function is generally used after committing an external
4396 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4397 * default action for this kind of discard is to pass the discard,
4398 * which will ideally result in an actually smaller image file, as
4399 * is probably desired. */
4400 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
4401 QCOW2_DISCARD_SNAPSHOT
, true);
4410 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
4412 BDRVQcow2State
*s
= bs
->opaque
;
4415 qemu_co_mutex_lock(&s
->lock
);
4416 ret
= qcow2_write_caches(bs
);
4417 qemu_co_mutex_unlock(&s
->lock
);
4422 static ssize_t
qcow2_measure_crypto_hdr_init_func(QCryptoBlock
*block
,
4423 size_t headerlen
, void *opaque
, Error
**errp
)
4425 size_t *headerlenp
= opaque
;
4427 /* Stash away the payload size */
4428 *headerlenp
= headerlen
;
4432 static ssize_t
qcow2_measure_crypto_hdr_write_func(QCryptoBlock
*block
,
4433 size_t offset
, const uint8_t *buf
, size_t buflen
,
4434 void *opaque
, Error
**errp
)
4436 /* Discard the bytes, we're not actually writing to an image */
4440 /* Determine the number of bytes for the LUKS payload */
4441 static bool qcow2_measure_luks_headerlen(QemuOpts
*opts
, size_t *len
,
4445 QDict
*cryptoopts_qdict
;
4446 QCryptoBlockCreateOptions
*cryptoopts
;
4447 QCryptoBlock
*crypto
;
4449 /* Extract "encrypt." options into a qdict */
4450 opts_qdict
= qemu_opts_to_qdict(opts
, NULL
);
4451 qdict_extract_subqdict(opts_qdict
, &cryptoopts_qdict
, "encrypt.");
4452 qobject_unref(opts_qdict
);
4454 /* Build QCryptoBlockCreateOptions object from qdict */
4455 qdict_put_str(cryptoopts_qdict
, "format", "luks");
4456 cryptoopts
= block_crypto_create_opts_init(cryptoopts_qdict
, errp
);
4457 qobject_unref(cryptoopts_qdict
);
4462 /* Fake LUKS creation in order to determine the payload size */
4463 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
4464 qcow2_measure_crypto_hdr_init_func
,
4465 qcow2_measure_crypto_hdr_write_func
,
4467 qapi_free_QCryptoBlockCreateOptions(cryptoopts
);
4472 qcrypto_block_free(crypto
);
4476 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
4479 Error
*local_err
= NULL
;
4480 BlockMeasureInfo
*info
;
4481 uint64_t required
= 0; /* bytes that contribute to required size */
4482 uint64_t virtual_size
; /* disk size as seen by guest */
4483 uint64_t refcount_bits
;
4485 uint64_t luks_payload_size
= 0;
4486 size_t cluster_size
;
4489 PreallocMode prealloc
;
4490 bool has_backing_file
;
4493 /* Parse image creation options */
4494 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, &local_err
);
4499 version
= qcow2_opt_get_version_del(opts
, &local_err
);
4504 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
4509 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
4510 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
4511 PREALLOC_MODE_OFF
, &local_err
);
4517 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
4518 has_backing_file
= !!optstr
;
4521 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
4522 has_luks
= optstr
&& strcmp(optstr
, "luks") == 0;
4528 if (!qcow2_measure_luks_headerlen(opts
, &headerlen
, &local_err
)) {
4532 luks_payload_size
= ROUND_UP(headerlen
, cluster_size
);
4535 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
4536 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
4538 /* Check that virtual disk size is valid */
4539 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
4540 cluster_size
/ sizeof(uint64_t));
4541 if (l2_tables
* sizeof(uint64_t) > QCOW_MAX_L1_SIZE
) {
4542 error_setg(&local_err
, "The image size is too large "
4543 "(try using a larger cluster size)");
4547 /* Account for input image */
4549 int64_t ssize
= bdrv_getlength(in_bs
);
4551 error_setg_errno(&local_err
, -ssize
,
4552 "Unable to get image virtual_size");
4556 virtual_size
= ROUND_UP(ssize
, cluster_size
);
4558 if (has_backing_file
) {
4559 /* We don't how much of the backing chain is shared by the input
4560 * image and the new image file. In the worst case the new image's
4561 * backing file has nothing in common with the input image. Be
4562 * conservative and assume all clusters need to be written.
4564 required
= virtual_size
;
4569 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
4572 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
4573 ssize
- offset
, &pnum
, NULL
,
4576 error_setg_errno(&local_err
, -ret
,
4577 "Unable to get block status");
4581 if (ret
& BDRV_BLOCK_ZERO
) {
4582 /* Skip zero regions (safe with no backing file) */
4583 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
4584 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
4585 /* Extend pnum to end of cluster for next iteration */
4586 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
4588 /* Count clusters we've seen */
4589 required
+= offset
% cluster_size
+ pnum
;
4595 /* Take into account preallocation. Nothing special is needed for
4596 * PREALLOC_MODE_METADATA since metadata is always counted.
4598 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
4599 required
= virtual_size
;
4602 info
= g_new(BlockMeasureInfo
, 1);
4603 info
->fully_allocated
=
4604 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
4605 ctz32(refcount_bits
)) + luks_payload_size
;
4607 /* Remove data clusters that are not required. This overestimates the
4608 * required size because metadata needed for the fully allocated file is
4611 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
4615 error_propagate(errp
, local_err
);
4619 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4621 BDRVQcow2State
*s
= bs
->opaque
;
4622 bdi
->unallocated_blocks_are_zero
= true;
4623 bdi
->cluster_size
= s
->cluster_size
;
4624 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
4628 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
,
4631 BDRVQcow2State
*s
= bs
->opaque
;
4632 ImageInfoSpecific
*spec_info
;
4633 QCryptoBlockInfo
*encrypt_info
= NULL
;
4634 Error
*local_err
= NULL
;
4636 if (s
->crypto
!= NULL
) {
4637 encrypt_info
= qcrypto_block_get_info(s
->crypto
, &local_err
);
4639 error_propagate(errp
, local_err
);
4644 spec_info
= g_new(ImageInfoSpecific
, 1);
4645 *spec_info
= (ImageInfoSpecific
){
4646 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
4647 .u
.qcow2
.data
= g_new0(ImageInfoSpecificQCow2
, 1),
4649 if (s
->qcow_version
== 2) {
4650 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
4651 .compat
= g_strdup("0.10"),
4652 .refcount_bits
= s
->refcount_bits
,
4654 } else if (s
->qcow_version
== 3) {
4655 Qcow2BitmapInfoList
*bitmaps
;
4656 bitmaps
= qcow2_get_bitmap_info_list(bs
, &local_err
);
4658 error_propagate(errp
, local_err
);
4659 qapi_free_ImageInfoSpecific(spec_info
);
4662 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
4663 .compat
= g_strdup("1.1"),
4664 .lazy_refcounts
= s
->compatible_features
&
4665 QCOW2_COMPAT_LAZY_REFCOUNTS
,
4666 .has_lazy_refcounts
= true,
4667 .corrupt
= s
->incompatible_features
&
4668 QCOW2_INCOMPAT_CORRUPT
,
4669 .has_corrupt
= true,
4670 .refcount_bits
= s
->refcount_bits
,
4671 .has_bitmaps
= !!bitmaps
,
4673 .has_data_file
= !!s
->image_data_file
,
4674 .data_file
= g_strdup(s
->image_data_file
),
4675 .has_data_file_raw
= has_data_file(bs
),
4676 .data_file_raw
= data_file_is_raw(bs
),
4679 /* if this assertion fails, this probably means a new version was
4680 * added without having it covered here */
4685 ImageInfoSpecificQCow2Encryption
*qencrypt
=
4686 g_new(ImageInfoSpecificQCow2Encryption
, 1);
4687 switch (encrypt_info
->format
) {
4688 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
4689 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
4691 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
4692 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
4693 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
4698 /* Since we did shallow copy above, erase any pointers
4699 * in the original info */
4700 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
4701 qapi_free_QCryptoBlockInfo(encrypt_info
);
4703 spec_info
->u
.qcow2
.data
->has_encrypt
= true;
4704 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
4710 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
4713 BDRVQcow2State
*s
= bs
->opaque
;
4715 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
4716 return bs
->drv
->bdrv_co_pwritev(bs
, qcow2_vm_state_offset(s
) + pos
,
4717 qiov
->size
, qiov
, 0);
4720 static int qcow2_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
4723 BDRVQcow2State
*s
= bs
->opaque
;
4725 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
4726 return bs
->drv
->bdrv_co_preadv(bs
, qcow2_vm_state_offset(s
) + pos
,
4727 qiov
->size
, qiov
, 0);
4731 * Downgrades an image's version. To achieve this, any incompatible features
4732 * have to be removed.
4734 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
4735 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
4738 BDRVQcow2State
*s
= bs
->opaque
;
4739 int current_version
= s
->qcow_version
;
4742 /* This is qcow2_downgrade(), not qcow2_upgrade() */
4743 assert(target_version
< current_version
);
4745 /* There are no other versions (now) that you can downgrade to */
4746 assert(target_version
== 2);
4748 if (s
->refcount_order
!= 4) {
4749 error_setg(errp
, "compat=0.10 requires refcount_bits=16");
4753 if (has_data_file(bs
)) {
4754 error_setg(errp
, "Cannot downgrade an image with a data file");
4758 /* clear incompatible features */
4759 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
4760 ret
= qcow2_mark_clean(bs
);
4762 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
4767 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
4768 * the first place; if that happens nonetheless, returning -ENOTSUP is the
4769 * best thing to do anyway */
4771 if (s
->incompatible_features
) {
4772 error_setg(errp
, "Cannot downgrade an image with incompatible features "
4773 "%#" PRIx64
" set", s
->incompatible_features
);
4777 /* since we can ignore compatible features, we can set them to 0 as well */
4778 s
->compatible_features
= 0;
4779 /* if lazy refcounts have been used, they have already been fixed through
4780 * clearing the dirty flag */
4782 /* clearing autoclear features is trivial */
4783 s
->autoclear_features
= 0;
4785 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
4787 error_setg_errno(errp
, -ret
, "Failed to turn zero into data clusters");
4791 s
->qcow_version
= target_version
;
4792 ret
= qcow2_update_header(bs
);
4794 s
->qcow_version
= current_version
;
4795 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4801 typedef enum Qcow2AmendOperation
{
4802 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
4803 * statically initialized to so that the helper CB can discern the first
4804 * invocation from an operation change */
4805 QCOW2_NO_OPERATION
= 0,
4807 QCOW2_CHANGING_REFCOUNT_ORDER
,
4809 } Qcow2AmendOperation
;
4811 typedef struct Qcow2AmendHelperCBInfo
{
4812 /* The code coordinating the amend operations should only modify
4813 * these four fields; the rest will be managed by the CB */
4814 BlockDriverAmendStatusCB
*original_status_cb
;
4815 void *original_cb_opaque
;
4817 Qcow2AmendOperation current_operation
;
4819 /* Total number of operations to perform (only set once) */
4820 int total_operations
;
4822 /* The following fields are managed by the CB */
4824 /* Number of operations completed */
4825 int operations_completed
;
4827 /* Cumulative offset of all completed operations */
4828 int64_t offset_completed
;
4830 Qcow2AmendOperation last_operation
;
4831 int64_t last_work_size
;
4832 } Qcow2AmendHelperCBInfo
;
4834 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
4835 int64_t operation_offset
,
4836 int64_t operation_work_size
, void *opaque
)
4838 Qcow2AmendHelperCBInfo
*info
= opaque
;
4839 int64_t current_work_size
;
4840 int64_t projected_work_size
;
4842 if (info
->current_operation
!= info
->last_operation
) {
4843 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
4844 info
->offset_completed
+= info
->last_work_size
;
4845 info
->operations_completed
++;
4848 info
->last_operation
= info
->current_operation
;
4851 assert(info
->total_operations
> 0);
4852 assert(info
->operations_completed
< info
->total_operations
);
4854 info
->last_work_size
= operation_work_size
;
4856 current_work_size
= info
->offset_completed
+ operation_work_size
;
4858 /* current_work_size is the total work size for (operations_completed + 1)
4859 * operations (which includes this one), so multiply it by the number of
4860 * operations not covered and divide it by the number of operations
4861 * covered to get a projection for the operations not covered */
4862 projected_work_size
= current_work_size
* (info
->total_operations
-
4863 info
->operations_completed
- 1)
4864 / (info
->operations_completed
+ 1);
4866 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
4867 current_work_size
+ projected_work_size
,
4868 info
->original_cb_opaque
);
4871 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4872 BlockDriverAmendStatusCB
*status_cb
,
4876 BDRVQcow2State
*s
= bs
->opaque
;
4877 int old_version
= s
->qcow_version
, new_version
= old_version
;
4878 uint64_t new_size
= 0;
4879 const char *backing_file
= NULL
, *backing_format
= NULL
, *data_file
= NULL
;
4880 bool lazy_refcounts
= s
->use_lazy_refcounts
;
4881 bool data_file_raw
= data_file_is_raw(bs
);
4882 const char *compat
= NULL
;
4883 uint64_t cluster_size
= s
->cluster_size
;
4886 int refcount_bits
= s
->refcount_bits
;
4888 QemuOptDesc
*desc
= opts
->list
->desc
;
4889 Qcow2AmendHelperCBInfo helper_cb_info
;
4891 while (desc
&& desc
->name
) {
4892 if (!qemu_opt_find(opts
, desc
->name
)) {
4893 /* only change explicitly defined options */
4898 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
4899 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
4901 /* preserve default */
4902 } else if (!strcmp(compat
, "0.10")) {
4904 } else if (!strcmp(compat
, "1.1")) {
4907 error_setg(errp
, "Unknown compatibility level %s", compat
);
4910 } else if (!strcmp(desc
->name
, BLOCK_OPT_PREALLOC
)) {
4911 error_setg(errp
, "Cannot change preallocation mode");
4913 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
4914 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
4915 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
4916 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4917 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
4918 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4919 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT
)) {
4920 encrypt
= qemu_opt_get_bool(opts
, BLOCK_OPT_ENCRYPT
,
4923 if (encrypt
!= !!s
->crypto
) {
4925 "Changing the encryption flag is not supported");
4928 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT_FORMAT
)) {
4929 encformat
= qcow2_crypt_method_from_format(
4930 qemu_opt_get(opts
, BLOCK_OPT_ENCRYPT_FORMAT
));
4932 if (encformat
!= s
->crypt_method_header
) {
4934 "Changing the encryption format is not supported");
4937 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
4939 "Changing the encryption parameters is not supported");
4941 } else if (!strcmp(desc
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
4942 cluster_size
= qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
,
4944 if (cluster_size
!= s
->cluster_size
) {
4945 error_setg(errp
, "Changing the cluster size is not supported");
4948 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
4949 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
4951 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
4952 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
4955 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
4956 !is_power_of_2(refcount_bits
))
4958 error_setg(errp
, "Refcount width must be a power of two and "
4959 "may not exceed 64 bits");
4962 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE
)) {
4963 data_file
= qemu_opt_get(opts
, BLOCK_OPT_DATA_FILE
);
4964 if (data_file
&& !has_data_file(bs
)) {
4965 error_setg(errp
, "data-file can only be set for images that "
4966 "use an external data file");
4969 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE_RAW
)) {
4970 data_file_raw
= qemu_opt_get_bool(opts
, BLOCK_OPT_DATA_FILE_RAW
,
4972 if (data_file_raw
&& !data_file_is_raw(bs
)) {
4973 error_setg(errp
, "data-file-raw cannot be set on existing "
4978 /* if this point is reached, this probably means a new option was
4979 * added without having it covered here */
4986 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
4987 .original_status_cb
= status_cb
,
4988 .original_cb_opaque
= cb_opaque
,
4989 .total_operations
= (new_version
< old_version
)
4990 + (s
->refcount_bits
!= refcount_bits
)
4993 /* Upgrade first (some features may require compat=1.1) */
4994 if (new_version
> old_version
) {
4995 s
->qcow_version
= new_version
;
4996 ret
= qcow2_update_header(bs
);
4998 s
->qcow_version
= old_version
;
4999 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5004 if (s
->refcount_bits
!= refcount_bits
) {
5005 int refcount_order
= ctz32(refcount_bits
);
5007 if (new_version
< 3 && refcount_bits
!= 16) {
5008 error_setg(errp
, "Refcount widths other than 16 bits require "
5009 "compatibility level 1.1 or above (use compat=1.1 or "
5014 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
5015 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
5016 &qcow2_amend_helper_cb
,
5017 &helper_cb_info
, errp
);
5023 /* data-file-raw blocks backing files, so clear it first if requested */
5024 if (data_file_raw
) {
5025 s
->autoclear_features
|= QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5027 s
->autoclear_features
&= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5031 g_free(s
->image_data_file
);
5032 s
->image_data_file
= *data_file
? g_strdup(data_file
) : NULL
;
5035 ret
= qcow2_update_header(bs
);
5037 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5041 if (backing_file
|| backing_format
) {
5042 ret
= qcow2_change_backing_file(bs
,
5043 backing_file
?: s
->image_backing_file
,
5044 backing_format
?: s
->image_backing_format
);
5046 error_setg_errno(errp
, -ret
, "Failed to change the backing file");
5051 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
5052 if (lazy_refcounts
) {
5053 if (new_version
< 3) {
5054 error_setg(errp
, "Lazy refcounts only supported with "
5055 "compatibility level 1.1 and above (use compat=1.1 "
5059 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5060 ret
= qcow2_update_header(bs
);
5062 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5063 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5066 s
->use_lazy_refcounts
= true;
5068 /* make image clean first */
5069 ret
= qcow2_mark_clean(bs
);
5071 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5074 /* now disallow lazy refcounts */
5075 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5076 ret
= qcow2_update_header(bs
);
5078 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5079 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5082 s
->use_lazy_refcounts
= false;
5087 BlockBackend
*blk
= blk_new(BLK_PERM_RESIZE
, BLK_PERM_ALL
);
5088 ret
= blk_insert_bs(blk
, bs
, errp
);
5094 ret
= blk_truncate(blk
, new_size
, PREALLOC_MODE_OFF
, errp
);
5101 /* Downgrade last (so unsupported features can be removed before) */
5102 if (new_version
< old_version
) {
5103 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
5104 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5105 &helper_cb_info
, errp
);
5115 * If offset or size are negative, respectively, they will not be included in
5116 * the BLOCK_IMAGE_CORRUPTED event emitted.
5117 * fatal will be ignored for read-only BDS; corruptions found there will always
5118 * be considered non-fatal.
5120 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
5121 int64_t size
, const char *message_format
, ...)
5123 BDRVQcow2State
*s
= bs
->opaque
;
5124 const char *node_name
;
5128 fatal
= fatal
&& bdrv_is_writable(bs
);
5130 if (s
->signaled_corruption
&&
5131 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
5136 va_start(ap
, message_format
);
5137 message
= g_strdup_vprintf(message_format
, ap
);
5141 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
5142 "corruption events will be suppressed\n", message
);
5144 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
5145 "corruption events will be suppressed\n", message
);
5148 node_name
= bdrv_get_node_name(bs
);
5149 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
5150 *node_name
!= '\0', node_name
,
5151 message
, offset
>= 0, offset
,
5157 qcow2_mark_corrupt(bs
);
5158 bs
->drv
= NULL
; /* make BDS unusable */
5161 s
->signaled_corruption
= true;
5164 static QemuOptsList qcow2_create_opts
= {
5165 .name
= "qcow2-create-opts",
5166 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
5169 .name
= BLOCK_OPT_SIZE
,
5170 .type
= QEMU_OPT_SIZE
,
5171 .help
= "Virtual disk size"
5174 .name
= BLOCK_OPT_COMPAT_LEVEL
,
5175 .type
= QEMU_OPT_STRING
,
5176 .help
= "Compatibility level (0.10 or 1.1)"
5179 .name
= BLOCK_OPT_BACKING_FILE
,
5180 .type
= QEMU_OPT_STRING
,
5181 .help
= "File name of a base image"
5184 .name
= BLOCK_OPT_BACKING_FMT
,
5185 .type
= QEMU_OPT_STRING
,
5186 .help
= "Image format of the base image"
5189 .name
= BLOCK_OPT_DATA_FILE
,
5190 .type
= QEMU_OPT_STRING
,
5191 .help
= "File name of an external data file"
5194 .name
= BLOCK_OPT_DATA_FILE_RAW
,
5195 .type
= QEMU_OPT_BOOL
,
5196 .help
= "The external data file must stay valid as a raw image"
5199 .name
= BLOCK_OPT_ENCRYPT
,
5200 .type
= QEMU_OPT_BOOL
,
5201 .help
= "Encrypt the image with format 'aes'. (Deprecated "
5202 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)",
5205 .name
= BLOCK_OPT_ENCRYPT_FORMAT
,
5206 .type
= QEMU_OPT_STRING
,
5207 .help
= "Encrypt the image, format choices: 'aes', 'luks'",
5209 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
5210 "ID of secret providing qcow AES key or LUKS passphrase"),
5211 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
5212 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
5213 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
5214 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
5215 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
5216 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
5218 .name
= BLOCK_OPT_CLUSTER_SIZE
,
5219 .type
= QEMU_OPT_SIZE
,
5220 .help
= "qcow2 cluster size",
5221 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
)
5224 .name
= BLOCK_OPT_PREALLOC
,
5225 .type
= QEMU_OPT_STRING
,
5226 .help
= "Preallocation mode (allowed values: off, metadata, "
5230 .name
= BLOCK_OPT_LAZY_REFCOUNTS
,
5231 .type
= QEMU_OPT_BOOL
,
5232 .help
= "Postpone refcount updates",
5233 .def_value_str
= "off"
5236 .name
= BLOCK_OPT_REFCOUNT_BITS
,
5237 .type
= QEMU_OPT_NUMBER
,
5238 .help
= "Width of a reference count entry in bits",
5239 .def_value_str
= "16"
5241 { /* end of list */ }
5245 static const char *const qcow2_strong_runtime_opts
[] = {
5246 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET
,
5251 BlockDriver bdrv_qcow2
= {
5252 .format_name
= "qcow2",
5253 .instance_size
= sizeof(BDRVQcow2State
),
5254 .bdrv_probe
= qcow2_probe
,
5255 .bdrv_open
= qcow2_open
,
5256 .bdrv_close
= qcow2_close
,
5257 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
5258 .bdrv_reopen_commit
= qcow2_reopen_commit
,
5259 .bdrv_reopen_abort
= qcow2_reopen_abort
,
5260 .bdrv_join_options
= qcow2_join_options
,
5261 .bdrv_child_perm
= bdrv_format_default_perms
,
5262 .bdrv_co_create_opts
= qcow2_co_create_opts
,
5263 .bdrv_co_create
= qcow2_co_create
,
5264 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
5265 .bdrv_co_block_status
= qcow2_co_block_status
,
5267 .bdrv_co_preadv
= qcow2_co_preadv
,
5268 .bdrv_co_pwritev
= qcow2_co_pwritev
,
5269 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
5271 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
5272 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
5273 .bdrv_co_copy_range_from
= qcow2_co_copy_range_from
,
5274 .bdrv_co_copy_range_to
= qcow2_co_copy_range_to
,
5275 .bdrv_co_truncate
= qcow2_co_truncate
,
5276 .bdrv_co_pwritev_compressed
= qcow2_co_pwritev_compressed
,
5277 .bdrv_make_empty
= qcow2_make_empty
,
5279 .bdrv_snapshot_create
= qcow2_snapshot_create
,
5280 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
5281 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
5282 .bdrv_snapshot_list
= qcow2_snapshot_list
,
5283 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
5284 .bdrv_measure
= qcow2_measure
,
5285 .bdrv_get_info
= qcow2_get_info
,
5286 .bdrv_get_specific_info
= qcow2_get_specific_info
,
5288 .bdrv_save_vmstate
= qcow2_save_vmstate
,
5289 .bdrv_load_vmstate
= qcow2_load_vmstate
,
5291 .supports_backing
= true,
5292 .bdrv_change_backing_file
= qcow2_change_backing_file
,
5294 .bdrv_refresh_limits
= qcow2_refresh_limits
,
5295 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
5296 .bdrv_inactivate
= qcow2_inactivate
,
5298 .create_opts
= &qcow2_create_opts
,
5299 .strong_runtime_opts
= qcow2_strong_runtime_opts
,
5300 .mutable_opts
= mutable_opts
,
5301 .bdrv_co_check
= qcow2_co_check
,
5302 .bdrv_amend_options
= qcow2_amend_options
,
5304 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
5305 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
5307 .bdrv_reopen_bitmaps_rw
= qcow2_reopen_bitmaps_rw
,
5308 .bdrv_can_store_new_dirty_bitmap
= qcow2_can_store_new_dirty_bitmap
,
5309 .bdrv_remove_persistent_dirty_bitmap
= qcow2_remove_persistent_dirty_bitmap
,
5312 static void bdrv_qcow2_init(void)
5314 bdrv_register(&bdrv_qcow2
);
5317 block_init(bdrv_qcow2_init
);