2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "block/block_int.h"
27 #include "sysemu/block-backend.h"
28 #include "qemu/module.h"
30 #include "block/qcow2.h"
31 #include "qemu/error-report.h"
32 #include "qapi/error.h"
33 #include "qapi/qapi-events-block-core.h"
34 #include "qapi/qmp/qdict.h"
35 #include "qapi/qmp/qstring.h"
37 #include "qemu/option_int.h"
38 #include "qemu/cutils.h"
39 #include "qemu/bswap.h"
40 #include "qapi/qobject-input-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "block/crypto.h"
45 Differences with QCOW:
47 - Support for multiple incremental snapshots.
48 - Memory management by reference counts.
49 - Clusters which have a reference count of one have the bit
50 QCOW_OFLAG_COPIED to optimize write performance.
51 - Size of compressed clusters is stored in sectors to reduce bit usage
52 in the cluster offsets.
53 - Support for storing additional data (such as the VM state) in the
55 - If a backing store is used, the cluster size is not constrained
56 (could be backported to QCOW).
57 - L2 tables have always a size of one cluster.
64 } QEMU_PACKED QCowExtension
;
66 #define QCOW2_EXT_MAGIC_END 0
67 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
68 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
69 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
70 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
72 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
74 const QCowHeader
*cow_header
= (const void *)buf
;
76 if (buf_size
>= sizeof(QCowHeader
) &&
77 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
78 be32_to_cpu(cow_header
->version
) >= 2)
85 static ssize_t
qcow2_crypto_hdr_read_func(QCryptoBlock
*block
, size_t offset
,
86 uint8_t *buf
, size_t buflen
,
87 void *opaque
, Error
**errp
)
89 BlockDriverState
*bs
= opaque
;
90 BDRVQcow2State
*s
= bs
->opaque
;
93 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
94 error_setg(errp
, "Request for data outside of extension header");
98 ret
= bdrv_pread(bs
->file
,
99 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
101 error_setg_errno(errp
, -ret
, "Could not read encryption header");
108 static ssize_t
qcow2_crypto_hdr_init_func(QCryptoBlock
*block
, size_t headerlen
,
109 void *opaque
, Error
**errp
)
111 BlockDriverState
*bs
= opaque
;
112 BDRVQcow2State
*s
= bs
->opaque
;
116 ret
= qcow2_alloc_clusters(bs
, headerlen
);
118 error_setg_errno(errp
, -ret
,
119 "Cannot allocate cluster for LUKS header size %zu",
124 s
->crypto_header
.length
= headerlen
;
125 s
->crypto_header
.offset
= ret
;
127 /* Zero fill remaining space in cluster so it has predictable
128 * content in case of future spec changes */
129 clusterlen
= size_to_clusters(s
, headerlen
) * s
->cluster_size
;
130 assert(qcow2_pre_write_overlap_check(bs
, 0, ret
, clusterlen
) == 0);
131 ret
= bdrv_pwrite_zeroes(bs
->file
,
133 clusterlen
- headerlen
, 0);
135 error_setg_errno(errp
, -ret
, "Could not zero fill encryption header");
143 static ssize_t
qcow2_crypto_hdr_write_func(QCryptoBlock
*block
, size_t offset
,
144 const uint8_t *buf
, size_t buflen
,
145 void *opaque
, Error
**errp
)
147 BlockDriverState
*bs
= opaque
;
148 BDRVQcow2State
*s
= bs
->opaque
;
151 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
152 error_setg(errp
, "Request for data outside of extension header");
156 ret
= bdrv_pwrite(bs
->file
,
157 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
159 error_setg_errno(errp
, -ret
, "Could not read encryption header");
167 * read qcow2 extension and fill bs
168 * start reading from start_offset
169 * finish reading upon magic of value 0 or when end_offset reached
170 * unknown magic is skipped (future extension this version knows nothing about)
171 * return 0 upon success, non-0 otherwise
173 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
174 uint64_t end_offset
, void **p_feature_table
,
175 int flags
, bool *need_update_header
,
178 BDRVQcow2State
*s
= bs
->opaque
;
182 Qcow2BitmapHeaderExt bitmaps_ext
;
184 if (need_update_header
!= NULL
) {
185 *need_update_header
= false;
189 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
191 offset
= start_offset
;
192 while (offset
< end_offset
) {
196 if (offset
> s
->cluster_size
)
197 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
199 printf("attempting to read extended header in offset %lu\n", offset
);
202 ret
= bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
));
204 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
205 "pread fail from offset %" PRIu64
, offset
);
208 be32_to_cpus(&ext
.magic
);
209 be32_to_cpus(&ext
.len
);
210 offset
+= sizeof(ext
);
212 printf("ext.magic = 0x%x\n", ext
.magic
);
214 if (offset
> end_offset
|| ext
.len
> end_offset
- offset
) {
215 error_setg(errp
, "Header extension too large");
220 case QCOW2_EXT_MAGIC_END
:
223 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
224 if (ext
.len
>= sizeof(bs
->backing_format
)) {
225 error_setg(errp
, "ERROR: ext_backing_format: len=%" PRIu32
226 " too large (>=%zu)", ext
.len
,
227 sizeof(bs
->backing_format
));
230 ret
= bdrv_pread(bs
->file
, offset
, bs
->backing_format
, ext
.len
);
232 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
233 "Could not read format name");
236 bs
->backing_format
[ext
.len
] = '\0';
237 s
->image_backing_format
= g_strdup(bs
->backing_format
);
239 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
243 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
244 if (p_feature_table
!= NULL
) {
245 void* feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
246 ret
= bdrv_pread(bs
->file
, offset
, feature_table
, ext
.len
);
248 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
249 "Could not read table");
253 *p_feature_table
= feature_table
;
257 case QCOW2_EXT_MAGIC_CRYPTO_HEADER
: {
258 unsigned int cflags
= 0;
259 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
260 error_setg(errp
, "CRYPTO header extension only "
261 "expected with LUKS encryption method");
264 if (ext
.len
!= sizeof(Qcow2CryptoHeaderExtension
)) {
265 error_setg(errp
, "CRYPTO header extension size %u, "
266 "but expected size %zu", ext
.len
,
267 sizeof(Qcow2CryptoHeaderExtension
));
271 ret
= bdrv_pread(bs
->file
, offset
, &s
->crypto_header
, ext
.len
);
273 error_setg_errno(errp
, -ret
,
274 "Unable to read CRYPTO header extension");
277 be64_to_cpus(&s
->crypto_header
.offset
);
278 be64_to_cpus(&s
->crypto_header
.length
);
280 if ((s
->crypto_header
.offset
% s
->cluster_size
) != 0) {
281 error_setg(errp
, "Encryption header offset '%" PRIu64
"' is "
282 "not a multiple of cluster size '%u'",
283 s
->crypto_header
.offset
, s
->cluster_size
);
287 if (flags
& BDRV_O_NO_IO
) {
288 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
290 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
291 qcow2_crypto_hdr_read_func
,
298 case QCOW2_EXT_MAGIC_BITMAPS
:
299 if (ext
.len
!= sizeof(bitmaps_ext
)) {
300 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
301 "Invalid extension length");
305 if (!(s
->autoclear_features
& QCOW2_AUTOCLEAR_BITMAPS
)) {
306 if (s
->qcow_version
< 3) {
307 /* Let's be a bit more specific */
308 warn_report("This qcow2 v2 image contains bitmaps, but "
309 "they may have been modified by a program "
310 "without persistent bitmap support; so now "
311 "they must all be considered inconsistent");
313 warn_report("a program lacking bitmap support "
314 "modified this file, so all bitmaps are now "
315 "considered inconsistent");
317 error_printf("Some clusters may be leaked, "
318 "run 'qemu-img check -r' on the image "
320 if (need_update_header
!= NULL
) {
321 /* Updating is needed to drop invalid bitmap extension. */
322 *need_update_header
= true;
327 ret
= bdrv_pread(bs
->file
, offset
, &bitmaps_ext
, ext
.len
);
329 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
330 "Could not read ext header");
334 if (bitmaps_ext
.reserved32
!= 0) {
335 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
336 "Reserved field is not zero");
340 be32_to_cpus(&bitmaps_ext
.nb_bitmaps
);
341 be64_to_cpus(&bitmaps_ext
.bitmap_directory_size
);
342 be64_to_cpus(&bitmaps_ext
.bitmap_directory_offset
);
344 if (bitmaps_ext
.nb_bitmaps
> QCOW2_MAX_BITMAPS
) {
346 "bitmaps_ext: Image has %" PRIu32
" bitmaps, "
347 "exceeding the QEMU supported maximum of %d",
348 bitmaps_ext
.nb_bitmaps
, QCOW2_MAX_BITMAPS
);
352 if (bitmaps_ext
.nb_bitmaps
== 0) {
353 error_setg(errp
, "found bitmaps extension with zero bitmaps");
357 if (bitmaps_ext
.bitmap_directory_offset
& (s
->cluster_size
- 1)) {
358 error_setg(errp
, "bitmaps_ext: "
359 "invalid bitmap directory offset");
363 if (bitmaps_ext
.bitmap_directory_size
>
364 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
) {
365 error_setg(errp
, "bitmaps_ext: "
366 "bitmap directory size (%" PRIu64
") exceeds "
367 "the maximum supported size (%d)",
368 bitmaps_ext
.bitmap_directory_size
,
369 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
);
373 s
->nb_bitmaps
= bitmaps_ext
.nb_bitmaps
;
374 s
->bitmap_directory_offset
=
375 bitmaps_ext
.bitmap_directory_offset
;
376 s
->bitmap_directory_size
=
377 bitmaps_ext
.bitmap_directory_size
;
380 printf("Qcow2: Got bitmaps extension: "
381 "offset=%" PRIu64
" nb_bitmaps=%" PRIu32
"\n",
382 s
->bitmap_directory_offset
, s
->nb_bitmaps
);
387 /* unknown magic - save it in case we need to rewrite the header */
388 /* If you add a new feature, make sure to also update the fast
389 * path of qcow2_make_empty() to deal with it. */
391 Qcow2UnknownHeaderExtension
*uext
;
393 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
394 uext
->magic
= ext
.magic
;
396 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
398 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
400 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
401 "Could not read data");
408 offset
+= ((ext
.len
+ 7) & ~7);
414 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
416 BDRVQcow2State
*s
= bs
->opaque
;
417 Qcow2UnknownHeaderExtension
*uext
, *next
;
419 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
420 QLIST_REMOVE(uext
, next
);
425 static void report_unsupported_feature(Error
**errp
, Qcow2Feature
*table
,
428 char *features
= g_strdup("");
431 while (table
&& table
->name
[0] != '\0') {
432 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
433 if (mask
& (1ULL << table
->bit
)) {
435 features
= g_strdup_printf("%s%s%.46s", old
, *old
? ", " : "",
438 mask
&= ~(1ULL << table
->bit
);
446 features
= g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64
,
447 old
, *old
? ", " : "", mask
);
451 error_setg(errp
, "Unsupported qcow2 feature(s): %s", features
);
456 * Sets the dirty bit and flushes afterwards if necessary.
458 * The incompatible_features bit is only set if the image file header was
459 * updated successfully. Therefore it is not required to check the return
460 * value of this function.
462 int qcow2_mark_dirty(BlockDriverState
*bs
)
464 BDRVQcow2State
*s
= bs
->opaque
;
468 assert(s
->qcow_version
>= 3);
470 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
471 return 0; /* already dirty */
474 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
475 ret
= bdrv_pwrite(bs
->file
, offsetof(QCowHeader
, incompatible_features
),
480 ret
= bdrv_flush(bs
->file
->bs
);
485 /* Only treat image as dirty if the header was updated successfully */
486 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
491 * Clears the dirty bit and flushes before if necessary. Only call this
492 * function when there are no pending requests, it does not guard against
493 * concurrent requests dirtying the image.
495 static int qcow2_mark_clean(BlockDriverState
*bs
)
497 BDRVQcow2State
*s
= bs
->opaque
;
499 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
502 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
504 ret
= qcow2_flush_caches(bs
);
509 return qcow2_update_header(bs
);
515 * Marks the image as corrupt.
517 int qcow2_mark_corrupt(BlockDriverState
*bs
)
519 BDRVQcow2State
*s
= bs
->opaque
;
521 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
522 return qcow2_update_header(bs
);
526 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
527 * before if necessary.
529 int qcow2_mark_consistent(BlockDriverState
*bs
)
531 BDRVQcow2State
*s
= bs
->opaque
;
533 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
534 int ret
= qcow2_flush_caches(bs
);
539 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
540 return qcow2_update_header(bs
);
545 static int coroutine_fn
qcow2_co_check_locked(BlockDriverState
*bs
,
546 BdrvCheckResult
*result
,
549 int ret
= qcow2_check_refcounts(bs
, result
, fix
);
554 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
555 ret
= qcow2_mark_clean(bs
);
559 return qcow2_mark_consistent(bs
);
564 static int coroutine_fn
qcow2_co_check(BlockDriverState
*bs
,
565 BdrvCheckResult
*result
,
568 BDRVQcow2State
*s
= bs
->opaque
;
571 qemu_co_mutex_lock(&s
->lock
);
572 ret
= qcow2_co_check_locked(bs
, result
, fix
);
573 qemu_co_mutex_unlock(&s
->lock
);
577 int qcow2_validate_table(BlockDriverState
*bs
, uint64_t offset
,
578 uint64_t entries
, size_t entry_len
,
579 int64_t max_size_bytes
, const char *table_name
,
582 BDRVQcow2State
*s
= bs
->opaque
;
584 if (entries
> max_size_bytes
/ entry_len
) {
585 error_setg(errp
, "%s too large", table_name
);
589 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
590 * because values will be passed to qemu functions taking int64_t. */
591 if ((INT64_MAX
- entries
* entry_len
< offset
) ||
592 (offset_into_cluster(s
, offset
) != 0)) {
593 error_setg(errp
, "%s offset invalid", table_name
);
600 static QemuOptsList qcow2_runtime_opts
= {
602 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
605 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
606 .type
= QEMU_OPT_BOOL
,
607 .help
= "Postpone refcount updates",
610 .name
= QCOW2_OPT_DISCARD_REQUEST
,
611 .type
= QEMU_OPT_BOOL
,
612 .help
= "Pass guest discard requests to the layer below",
615 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
616 .type
= QEMU_OPT_BOOL
,
617 .help
= "Generate discard requests when snapshot related space "
621 .name
= QCOW2_OPT_DISCARD_OTHER
,
622 .type
= QEMU_OPT_BOOL
,
623 .help
= "Generate discard requests when other clusters are freed",
626 .name
= QCOW2_OPT_OVERLAP
,
627 .type
= QEMU_OPT_STRING
,
628 .help
= "Selects which overlap checks to perform from a range of "
629 "templates (none, constant, cached, all)",
632 .name
= QCOW2_OPT_OVERLAP_TEMPLATE
,
633 .type
= QEMU_OPT_STRING
,
634 .help
= "Selects which overlap checks to perform from a range of "
635 "templates (none, constant, cached, all)",
638 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
639 .type
= QEMU_OPT_BOOL
,
640 .help
= "Check for unintended writes into the main qcow2 header",
643 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
644 .type
= QEMU_OPT_BOOL
,
645 .help
= "Check for unintended writes into the active L1 table",
648 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
649 .type
= QEMU_OPT_BOOL
,
650 .help
= "Check for unintended writes into an active L2 table",
653 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
654 .type
= QEMU_OPT_BOOL
,
655 .help
= "Check for unintended writes into the refcount table",
658 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
659 .type
= QEMU_OPT_BOOL
,
660 .help
= "Check for unintended writes into a refcount block",
663 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
664 .type
= QEMU_OPT_BOOL
,
665 .help
= "Check for unintended writes into the snapshot table",
668 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
669 .type
= QEMU_OPT_BOOL
,
670 .help
= "Check for unintended writes into an inactive L1 table",
673 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
674 .type
= QEMU_OPT_BOOL
,
675 .help
= "Check for unintended writes into an inactive L2 table",
678 .name
= QCOW2_OPT_CACHE_SIZE
,
679 .type
= QEMU_OPT_SIZE
,
680 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
684 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
685 .type
= QEMU_OPT_SIZE
,
686 .help
= "Maximum L2 table cache size",
689 .name
= QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
690 .type
= QEMU_OPT_SIZE
,
691 .help
= "Size of each entry in the L2 cache",
694 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
695 .type
= QEMU_OPT_SIZE
,
696 .help
= "Maximum refcount block cache size",
699 .name
= QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
700 .type
= QEMU_OPT_NUMBER
,
701 .help
= "Clean unused cache entries after this time (in seconds)",
703 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
704 "ID of secret providing qcow2 AES key or LUKS passphrase"),
705 { /* end of list */ }
709 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
710 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
711 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
712 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
713 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
714 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
715 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
716 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
717 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
720 static void cache_clean_timer_cb(void *opaque
)
722 BlockDriverState
*bs
= opaque
;
723 BDRVQcow2State
*s
= bs
->opaque
;
724 qcow2_cache_clean_unused(s
->l2_table_cache
);
725 qcow2_cache_clean_unused(s
->refcount_block_cache
);
726 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
727 (int64_t) s
->cache_clean_interval
* 1000);
730 static void cache_clean_timer_init(BlockDriverState
*bs
, AioContext
*context
)
732 BDRVQcow2State
*s
= bs
->opaque
;
733 if (s
->cache_clean_interval
> 0) {
734 s
->cache_clean_timer
= aio_timer_new(context
, QEMU_CLOCK_VIRTUAL
,
735 SCALE_MS
, cache_clean_timer_cb
,
737 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
738 (int64_t) s
->cache_clean_interval
* 1000);
742 static void cache_clean_timer_del(BlockDriverState
*bs
)
744 BDRVQcow2State
*s
= bs
->opaque
;
745 if (s
->cache_clean_timer
) {
746 timer_del(s
->cache_clean_timer
);
747 timer_free(s
->cache_clean_timer
);
748 s
->cache_clean_timer
= NULL
;
752 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
754 cache_clean_timer_del(bs
);
757 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
758 AioContext
*new_context
)
760 cache_clean_timer_init(bs
, new_context
);
763 static void read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
764 uint64_t *l2_cache_size
,
765 uint64_t *l2_cache_entry_size
,
766 uint64_t *refcount_cache_size
, Error
**errp
)
768 BDRVQcow2State
*s
= bs
->opaque
;
769 uint64_t combined_cache_size
;
770 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
772 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
773 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
774 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
776 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
777 *l2_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
, 0);
778 *refcount_cache_size
= qemu_opt_get_size(opts
,
779 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
781 *l2_cache_entry_size
= qemu_opt_get_size(
782 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
784 if (combined_cache_size_set
) {
785 if (l2_cache_size_set
&& refcount_cache_size_set
) {
786 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
787 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
790 } else if (*l2_cache_size
> combined_cache_size
) {
791 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
792 QCOW2_OPT_CACHE_SIZE
);
794 } else if (*refcount_cache_size
> combined_cache_size
) {
795 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
796 QCOW2_OPT_CACHE_SIZE
);
800 if (l2_cache_size_set
) {
801 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
802 } else if (refcount_cache_size_set
) {
803 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
805 uint64_t virtual_disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
806 uint64_t max_l2_cache
= virtual_disk_size
/ (s
->cluster_size
/ 8);
807 uint64_t min_refcount_cache
=
808 (uint64_t) MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
810 /* Assign as much memory as possible to the L2 cache, and
811 * use the remainder for the refcount cache */
812 if (combined_cache_size
>= max_l2_cache
+ min_refcount_cache
) {
813 *l2_cache_size
= max_l2_cache
;
814 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
816 *refcount_cache_size
=
817 MIN(combined_cache_size
, min_refcount_cache
);
818 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
822 if (!l2_cache_size_set
) {
823 *l2_cache_size
= MAX(DEFAULT_L2_CACHE_BYTE_SIZE
,
824 (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
827 if (!refcount_cache_size_set
) {
828 *refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
832 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
833 *l2_cache_entry_size
> s
->cluster_size
||
834 !is_power_of_2(*l2_cache_entry_size
)) {
835 error_setg(errp
, "L2 cache entry size must be a power of two "
836 "between %d and the cluster size (%d)",
837 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
842 typedef struct Qcow2ReopenState
{
843 Qcow2Cache
*l2_table_cache
;
844 Qcow2Cache
*refcount_block_cache
;
845 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
846 bool use_lazy_refcounts
;
848 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
849 uint64_t cache_clean_interval
;
850 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
853 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
855 QDict
*options
, int flags
,
858 BDRVQcow2State
*s
= bs
->opaque
;
859 QemuOpts
*opts
= NULL
;
860 const char *opt_overlap_check
, *opt_overlap_check_template
;
861 int overlap_check_template
= 0;
862 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
864 const char *encryptfmt
;
865 QDict
*encryptopts
= NULL
;
866 Error
*local_err
= NULL
;
869 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
870 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
872 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
873 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
875 error_propagate(errp
, local_err
);
880 /* get L2 table/refcount block cache size from command line options */
881 read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
882 &refcount_cache_size
, &local_err
);
884 error_propagate(errp
, local_err
);
889 l2_cache_size
/= l2_cache_entry_size
;
890 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
891 l2_cache_size
= MIN_L2_CACHE_SIZE
;
893 if (l2_cache_size
> INT_MAX
) {
894 error_setg(errp
, "L2 cache size too big");
899 refcount_cache_size
/= s
->cluster_size
;
900 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
901 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
903 if (refcount_cache_size
> INT_MAX
) {
904 error_setg(errp
, "Refcount cache size too big");
909 /* alloc new L2 table/refcount block cache, flush old one */
910 if (s
->l2_table_cache
) {
911 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
913 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
918 if (s
->refcount_block_cache
) {
919 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
921 error_setg_errno(errp
, -ret
,
922 "Failed to flush the refcount block cache");
927 r
->l2_slice_size
= l2_cache_entry_size
/ sizeof(uint64_t);
928 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
929 l2_cache_entry_size
);
930 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
932 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
933 error_setg(errp
, "Could not allocate metadata caches");
938 /* New interval for cache cleanup timer */
939 r
->cache_clean_interval
=
940 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
941 s
->cache_clean_interval
);
943 if (r
->cache_clean_interval
!= 0) {
944 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
945 " not supported on this host");
950 if (r
->cache_clean_interval
> UINT_MAX
) {
951 error_setg(errp
, "Cache clean interval too big");
956 /* lazy-refcounts; flush if going from enabled to disabled */
957 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
958 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
959 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
960 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
961 "qemu 1.1 compatibility level");
966 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
967 ret
= qcow2_mark_clean(bs
);
969 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
974 /* Overlap check options */
975 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
976 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
977 if (opt_overlap_check_template
&& opt_overlap_check
&&
978 strcmp(opt_overlap_check_template
, opt_overlap_check
))
980 error_setg(errp
, "Conflicting values for qcow2 options '"
981 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
982 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
986 if (!opt_overlap_check
) {
987 opt_overlap_check
= opt_overlap_check_template
?: "cached";
990 if (!strcmp(opt_overlap_check
, "none")) {
991 overlap_check_template
= 0;
992 } else if (!strcmp(opt_overlap_check
, "constant")) {
993 overlap_check_template
= QCOW2_OL_CONSTANT
;
994 } else if (!strcmp(opt_overlap_check
, "cached")) {
995 overlap_check_template
= QCOW2_OL_CACHED
;
996 } else if (!strcmp(opt_overlap_check
, "all")) {
997 overlap_check_template
= QCOW2_OL_ALL
;
999 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1000 "'overlap-check'. Allowed are any of the following: "
1001 "none, constant, cached, all", opt_overlap_check
);
1006 r
->overlap_check
= 0;
1007 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1008 /* overlap-check defines a template bitmask, but every flag may be
1009 * overwritten through the associated boolean option */
1011 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1012 overlap_check_template
& (1 << i
)) << i
;
1015 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1016 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1017 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1018 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1019 flags
& BDRV_O_UNMAP
);
1020 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1021 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1022 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1023 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1025 switch (s
->crypt_method_header
) {
1026 case QCOW_CRYPT_NONE
:
1028 error_setg(errp
, "No encryption in image header, but options "
1029 "specified format '%s'", encryptfmt
);
1035 case QCOW_CRYPT_AES
:
1036 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1038 "Header reported 'aes' encryption format but "
1039 "options specify '%s'", encryptfmt
);
1043 qdict_del(encryptopts
, "format");
1044 r
->crypto_opts
= block_crypto_open_opts_init(
1045 Q_CRYPTO_BLOCK_FORMAT_QCOW
, encryptopts
, errp
);
1048 case QCOW_CRYPT_LUKS
:
1049 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1051 "Header reported 'luks' encryption format but "
1052 "options specify '%s'", encryptfmt
);
1056 qdict_del(encryptopts
, "format");
1057 r
->crypto_opts
= block_crypto_open_opts_init(
1058 Q_CRYPTO_BLOCK_FORMAT_LUKS
, encryptopts
, errp
);
1062 error_setg(errp
, "Unsupported encryption method %d",
1063 s
->crypt_method_header
);
1066 if (s
->crypt_method_header
!= QCOW_CRYPT_NONE
&& !r
->crypto_opts
) {
1073 qobject_unref(encryptopts
);
1074 qemu_opts_del(opts
);
1079 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1080 Qcow2ReopenState
*r
)
1082 BDRVQcow2State
*s
= bs
->opaque
;
1085 if (s
->l2_table_cache
) {
1086 qcow2_cache_destroy(s
->l2_table_cache
);
1088 if (s
->refcount_block_cache
) {
1089 qcow2_cache_destroy(s
->refcount_block_cache
);
1091 s
->l2_table_cache
= r
->l2_table_cache
;
1092 s
->refcount_block_cache
= r
->refcount_block_cache
;
1093 s
->l2_slice_size
= r
->l2_slice_size
;
1095 s
->overlap_check
= r
->overlap_check
;
1096 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1098 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1099 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1102 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1103 cache_clean_timer_del(bs
);
1104 s
->cache_clean_interval
= r
->cache_clean_interval
;
1105 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1108 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1109 s
->crypto_opts
= r
->crypto_opts
;
1112 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1113 Qcow2ReopenState
*r
)
1115 if (r
->l2_table_cache
) {
1116 qcow2_cache_destroy(r
->l2_table_cache
);
1118 if (r
->refcount_block_cache
) {
1119 qcow2_cache_destroy(r
->refcount_block_cache
);
1121 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1124 static int qcow2_update_options(BlockDriverState
*bs
, QDict
*options
,
1125 int flags
, Error
**errp
)
1127 Qcow2ReopenState r
= {};
1130 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1132 qcow2_update_options_commit(bs
, &r
);
1134 qcow2_update_options_abort(bs
, &r
);
1140 /* Called with s->lock held. */
1141 static int coroutine_fn
qcow2_do_open(BlockDriverState
*bs
, QDict
*options
,
1142 int flags
, Error
**errp
)
1144 BDRVQcow2State
*s
= bs
->opaque
;
1145 unsigned int len
, i
;
1148 Error
*local_err
= NULL
;
1150 uint64_t l1_vm_state_index
;
1151 bool update_header
= false;
1152 bool header_updated
= false;
1154 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
1156 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1159 be32_to_cpus(&header
.magic
);
1160 be32_to_cpus(&header
.version
);
1161 be64_to_cpus(&header
.backing_file_offset
);
1162 be32_to_cpus(&header
.backing_file_size
);
1163 be64_to_cpus(&header
.size
);
1164 be32_to_cpus(&header
.cluster_bits
);
1165 be32_to_cpus(&header
.crypt_method
);
1166 be64_to_cpus(&header
.l1_table_offset
);
1167 be32_to_cpus(&header
.l1_size
);
1168 be64_to_cpus(&header
.refcount_table_offset
);
1169 be32_to_cpus(&header
.refcount_table_clusters
);
1170 be64_to_cpus(&header
.snapshots_offset
);
1171 be32_to_cpus(&header
.nb_snapshots
);
1173 if (header
.magic
!= QCOW_MAGIC
) {
1174 error_setg(errp
, "Image is not in qcow2 format");
1178 if (header
.version
< 2 || header
.version
> 3) {
1179 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1184 s
->qcow_version
= header
.version
;
1186 /* Initialise cluster size */
1187 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1188 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1189 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1190 header
.cluster_bits
);
1195 s
->cluster_bits
= header
.cluster_bits
;
1196 s
->cluster_size
= 1 << s
->cluster_bits
;
1197 s
->cluster_sectors
= 1 << (s
->cluster_bits
- BDRV_SECTOR_BITS
);
1199 /* Initialise version 3 header fields */
1200 if (header
.version
== 2) {
1201 header
.incompatible_features
= 0;
1202 header
.compatible_features
= 0;
1203 header
.autoclear_features
= 0;
1204 header
.refcount_order
= 4;
1205 header
.header_length
= 72;
1207 be64_to_cpus(&header
.incompatible_features
);
1208 be64_to_cpus(&header
.compatible_features
);
1209 be64_to_cpus(&header
.autoclear_features
);
1210 be32_to_cpus(&header
.refcount_order
);
1211 be32_to_cpus(&header
.header_length
);
1213 if (header
.header_length
< 104) {
1214 error_setg(errp
, "qcow2 header too short");
1220 if (header
.header_length
> s
->cluster_size
) {
1221 error_setg(errp
, "qcow2 header exceeds cluster size");
1226 if (header
.header_length
> sizeof(header
)) {
1227 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1228 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1229 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
1230 s
->unknown_header_fields_size
);
1232 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1238 if (header
.backing_file_offset
> s
->cluster_size
) {
1239 error_setg(errp
, "Invalid backing file offset");
1244 if (header
.backing_file_offset
) {
1245 ext_end
= header
.backing_file_offset
;
1247 ext_end
= 1 << header
.cluster_bits
;
1250 /* Handle feature bits */
1251 s
->incompatible_features
= header
.incompatible_features
;
1252 s
->compatible_features
= header
.compatible_features
;
1253 s
->autoclear_features
= header
.autoclear_features
;
1255 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1256 void *feature_table
= NULL
;
1257 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1258 &feature_table
, flags
, NULL
, NULL
);
1259 report_unsupported_feature(errp
, feature_table
,
1260 s
->incompatible_features
&
1261 ~QCOW2_INCOMPAT_MASK
);
1263 g_free(feature_table
);
1267 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1268 /* Corrupt images may not be written to unless they are being repaired
1270 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1271 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1278 /* Check support for various header values */
1279 if (header
.refcount_order
> 6) {
1280 error_setg(errp
, "Reference count entry width too large; may not "
1285 s
->refcount_order
= header
.refcount_order
;
1286 s
->refcount_bits
= 1 << s
->refcount_order
;
1287 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1288 s
->refcount_max
+= s
->refcount_max
- 1;
1290 s
->crypt_method_header
= header
.crypt_method
;
1291 if (s
->crypt_method_header
) {
1292 if (bdrv_uses_whitelist() &&
1293 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1295 "Use of AES-CBC encrypted qcow2 images is no longer "
1296 "supported in system emulators");
1297 error_append_hint(errp
,
1298 "You can use 'qemu-img convert' to convert your "
1299 "image to an alternative supported format, such "
1300 "as unencrypted qcow2, or raw with the LUKS "
1301 "format instead.\n");
1306 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1307 s
->crypt_physical_offset
= false;
1309 /* Assuming LUKS and any future crypt methods we
1310 * add will all use physical offsets, due to the
1311 * fact that the alternative is insecure... */
1312 s
->crypt_physical_offset
= true;
1315 bs
->encrypted
= true;
1318 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
1319 s
->l2_size
= 1 << s
->l2_bits
;
1320 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1321 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1322 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1323 bs
->total_sectors
= header
.size
/ 512;
1324 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1325 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1326 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1328 s
->refcount_table_offset
= header
.refcount_table_offset
;
1329 s
->refcount_table_size
=
1330 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1332 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1333 error_setg(errp
, "Image does not contain a reference count table");
1338 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1339 header
.refcount_table_clusters
,
1340 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1341 "Reference count table", errp
);
1346 /* The total size in bytes of the snapshot table is checked in
1347 * qcow2_read_snapshots() because the size of each snapshot is
1348 * variable and we don't know it yet.
1349 * Here we only check the offset and number of snapshots. */
1350 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1351 header
.nb_snapshots
,
1352 sizeof(QCowSnapshotHeader
),
1353 sizeof(QCowSnapshotHeader
) * QCOW_MAX_SNAPSHOTS
,
1354 "Snapshot table", errp
);
1359 /* read the level 1 table */
1360 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1361 header
.l1_size
, sizeof(uint64_t),
1362 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1366 s
->l1_size
= header
.l1_size
;
1367 s
->l1_table_offset
= header
.l1_table_offset
;
1369 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1370 if (l1_vm_state_index
> INT_MAX
) {
1371 error_setg(errp
, "Image is too big");
1375 s
->l1_vm_state_index
= l1_vm_state_index
;
1377 /* the L1 table must contain at least enough entries to put
1378 header.size bytes */
1379 if (s
->l1_size
< s
->l1_vm_state_index
) {
1380 error_setg(errp
, "L1 table is too small");
1385 if (s
->l1_size
> 0) {
1386 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
,
1387 ROUND_UP(s
->l1_size
* sizeof(uint64_t), 512));
1388 if (s
->l1_table
== NULL
) {
1389 error_setg(errp
, "Could not allocate L1 table");
1393 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1394 s
->l1_size
* sizeof(uint64_t));
1396 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1399 for(i
= 0;i
< s
->l1_size
; i
++) {
1400 be64_to_cpus(&s
->l1_table
[i
]);
1404 /* Parse driver-specific options */
1405 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1410 s
->cluster_cache_offset
= -1;
1413 ret
= qcow2_refcount_init(bs
);
1415 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1419 QLIST_INIT(&s
->cluster_allocs
);
1420 QTAILQ_INIT(&s
->discards
);
1422 /* read qcow2 extensions */
1423 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1424 flags
, &update_header
, &local_err
)) {
1425 error_propagate(errp
, local_err
);
1430 /* qcow2_read_extension may have set up the crypto context
1431 * if the crypt method needs a header region, some methods
1432 * don't need header extensions, so must check here
1434 if (s
->crypt_method_header
&& !s
->crypto
) {
1435 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1436 unsigned int cflags
= 0;
1437 if (flags
& BDRV_O_NO_IO
) {
1438 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1440 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1441 NULL
, NULL
, cflags
, errp
);
1446 } else if (!(flags
& BDRV_O_NO_IO
)) {
1447 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1448 s
->crypt_method_header
);
1454 /* read the backing file name */
1455 if (header
.backing_file_offset
!= 0) {
1456 len
= header
.backing_file_size
;
1457 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1458 len
>= sizeof(bs
->backing_file
)) {
1459 error_setg(errp
, "Backing file name too long");
1463 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
1464 bs
->backing_file
, len
);
1466 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1469 bs
->backing_file
[len
] = '\0';
1470 s
->image_backing_file
= g_strdup(bs
->backing_file
);
1473 /* Internal snapshots */
1474 s
->snapshots_offset
= header
.snapshots_offset
;
1475 s
->nb_snapshots
= header
.nb_snapshots
;
1477 ret
= qcow2_read_snapshots(bs
);
1479 error_setg_errno(errp
, -ret
, "Could not read snapshots");
1483 /* Clear unknown autoclear feature bits */
1484 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1486 update_header
&& !bs
->read_only
&& !(flags
& BDRV_O_INACTIVE
);
1487 if (update_header
) {
1488 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1491 if (s
->dirty_bitmaps_loaded
) {
1492 /* It's some kind of reopen. There are no known cases where we need to
1493 * reload bitmaps in such a situation, so it's safer to skip them.
1495 * Moreover, if we have some readonly bitmaps and we are reopening for
1496 * rw we should reopen bitmaps correspondingly.
1498 if (bdrv_has_readonly_bitmaps(bs
) &&
1499 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
))
1501 qcow2_reopen_bitmaps_rw_hint(bs
, &header_updated
, &local_err
);
1504 header_updated
= qcow2_load_dirty_bitmaps(bs
, &local_err
);
1505 s
->dirty_bitmaps_loaded
= true;
1507 update_header
= update_header
&& !header_updated
;
1508 if (local_err
!= NULL
) {
1509 error_propagate(errp
, local_err
);
1514 if (update_header
) {
1515 ret
= qcow2_update_header(bs
);
1517 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1522 bs
->supported_zero_flags
= header
.version
>= 3 ? BDRV_REQ_MAY_UNMAP
: 0;
1524 /* Repair image if dirty */
1525 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INACTIVE
)) && !bs
->read_only
&&
1526 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1527 BdrvCheckResult result
= {0};
1529 ret
= qcow2_co_check_locked(bs
, &result
,
1530 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1531 if (ret
< 0 || result
.check_errors
) {
1535 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1542 BdrvCheckResult result
= {0};
1543 qcow2_check_refcounts(bs
, &result
, 0);
1549 g_free(s
->unknown_header_fields
);
1550 cleanup_unknown_header_ext(bs
);
1551 qcow2_free_snapshots(bs
);
1552 qcow2_refcount_close(bs
);
1553 qemu_vfree(s
->l1_table
);
1554 /* else pre-write overlap checks in cache_destroy may crash */
1556 cache_clean_timer_del(bs
);
1557 if (s
->l2_table_cache
) {
1558 qcow2_cache_destroy(s
->l2_table_cache
);
1560 if (s
->refcount_block_cache
) {
1561 qcow2_cache_destroy(s
->refcount_block_cache
);
1563 qcrypto_block_free(s
->crypto
);
1564 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1568 typedef struct QCow2OpenCo
{
1569 BlockDriverState
*bs
;
1576 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1578 QCow2OpenCo
*qoc
= opaque
;
1579 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1581 qemu_co_mutex_lock(&s
->lock
);
1582 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, qoc
->errp
);
1583 qemu_co_mutex_unlock(&s
->lock
);
1586 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1589 BDRVQcow2State
*s
= bs
->opaque
;
1598 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_file
,
1604 /* Initialise locks */
1605 qemu_co_mutex_init(&s
->lock
);
1607 if (qemu_in_coroutine()) {
1608 /* From bdrv_co_create. */
1609 qcow2_open_entry(&qoc
);
1611 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1612 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1617 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1619 BDRVQcow2State
*s
= bs
->opaque
;
1621 if (bs
->encrypted
) {
1622 /* Encryption works on a sector granularity */
1623 bs
->bl
.request_alignment
= BDRV_SECTOR_SIZE
;
1625 bs
->bl
.pwrite_zeroes_alignment
= s
->cluster_size
;
1626 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1629 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1630 BlockReopenQueue
*queue
, Error
**errp
)
1632 Qcow2ReopenState
*r
;
1635 r
= g_new0(Qcow2ReopenState
, 1);
1638 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1639 state
->flags
, errp
);
1644 /* We need to write out any unwritten data if we reopen read-only. */
1645 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1646 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1651 ret
= bdrv_flush(state
->bs
);
1656 ret
= qcow2_mark_clean(state
->bs
);
1665 qcow2_update_options_abort(state
->bs
, r
);
1670 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1672 qcow2_update_options_commit(state
->bs
, state
->opaque
);
1673 g_free(state
->opaque
);
1676 static void qcow2_reopen_abort(BDRVReopenState
*state
)
1678 qcow2_update_options_abort(state
->bs
, state
->opaque
);
1679 g_free(state
->opaque
);
1682 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
1684 bool has_new_overlap_template
=
1685 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
1686 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1687 bool has_new_total_cache_size
=
1688 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
1689 bool has_all_cache_options
;
1691 /* New overlap template overrides all old overlap options */
1692 if (has_new_overlap_template
) {
1693 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
1694 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1695 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
1696 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
1697 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
1698 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
1699 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
1700 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
1701 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
1702 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
1705 /* New total cache size overrides all old options */
1706 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
1707 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
1708 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1711 qdict_join(options
, old_options
, false);
1714 * If after merging all cache size options are set, an old total size is
1715 * overwritten. Do keep all options, however, if all three are new. The
1716 * resulting error message is what we want to happen.
1718 has_all_cache_options
=
1719 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
1720 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
1721 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1723 if (has_all_cache_options
&& !has_new_total_cache_size
) {
1724 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
1728 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
1730 int64_t offset
, int64_t count
,
1731 int64_t *pnum
, int64_t *map
,
1732 BlockDriverState
**file
)
1734 BDRVQcow2State
*s
= bs
->opaque
;
1735 uint64_t cluster_offset
;
1736 int index_in_cluster
, ret
;
1740 bytes
= MIN(INT_MAX
, count
);
1741 qemu_co_mutex_lock(&s
->lock
);
1742 ret
= qcow2_get_cluster_offset(bs
, offset
, &bytes
, &cluster_offset
);
1743 qemu_co_mutex_unlock(&s
->lock
);
1750 if (cluster_offset
!= 0 && ret
!= QCOW2_CLUSTER_COMPRESSED
&&
1752 index_in_cluster
= offset
& (s
->cluster_size
- 1);
1753 *map
= cluster_offset
| index_in_cluster
;
1754 *file
= bs
->file
->bs
;
1755 status
|= BDRV_BLOCK_OFFSET_VALID
;
1757 if (ret
== QCOW2_CLUSTER_ZERO_PLAIN
|| ret
== QCOW2_CLUSTER_ZERO_ALLOC
) {
1758 status
|= BDRV_BLOCK_ZERO
;
1759 } else if (ret
!= QCOW2_CLUSTER_UNALLOCATED
) {
1760 status
|= BDRV_BLOCK_DATA
;
1765 static coroutine_fn
int qcow2_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1766 uint64_t bytes
, QEMUIOVector
*qiov
,
1769 BDRVQcow2State
*s
= bs
->opaque
;
1770 int offset_in_cluster
;
1772 unsigned int cur_bytes
; /* number of bytes in current iteration */
1773 uint64_t cluster_offset
= 0;
1774 uint64_t bytes_done
= 0;
1775 QEMUIOVector hd_qiov
;
1776 uint8_t *cluster_data
= NULL
;
1778 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1780 qemu_co_mutex_lock(&s
->lock
);
1782 while (bytes
!= 0) {
1784 /* prepare next request */
1785 cur_bytes
= MIN(bytes
, INT_MAX
);
1787 cur_bytes
= MIN(cur_bytes
,
1788 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1791 ret
= qcow2_get_cluster_offset(bs
, offset
, &cur_bytes
, &cluster_offset
);
1796 offset_in_cluster
= offset_into_cluster(s
, offset
);
1798 qemu_iovec_reset(&hd_qiov
);
1799 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
1802 case QCOW2_CLUSTER_UNALLOCATED
:
1805 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
1806 qemu_co_mutex_unlock(&s
->lock
);
1807 ret
= bdrv_co_preadv(bs
->backing
, offset
, cur_bytes
,
1809 qemu_co_mutex_lock(&s
->lock
);
1814 /* Note: in this case, no need to wait */
1815 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1819 case QCOW2_CLUSTER_ZERO_PLAIN
:
1820 case QCOW2_CLUSTER_ZERO_ALLOC
:
1821 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1824 case QCOW2_CLUSTER_COMPRESSED
:
1825 /* add AIO support for compressed blocks ? */
1826 ret
= qcow2_decompress_cluster(bs
, cluster_offset
);
1831 qemu_iovec_from_buf(&hd_qiov
, 0,
1832 s
->cluster_cache
+ offset_in_cluster
,
1836 case QCOW2_CLUSTER_NORMAL
:
1837 if ((cluster_offset
& 511) != 0) {
1842 if (bs
->encrypted
) {
1846 * For encrypted images, read everything into a temporary
1847 * contiguous buffer on which the AES functions can work.
1849 if (!cluster_data
) {
1851 qemu_try_blockalign(bs
->file
->bs
,
1852 QCOW_MAX_CRYPT_CLUSTERS
1854 if (cluster_data
== NULL
) {
1860 assert(cur_bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1861 qemu_iovec_reset(&hd_qiov
);
1862 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
1865 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1866 qemu_co_mutex_unlock(&s
->lock
);
1867 ret
= bdrv_co_preadv(bs
->file
,
1868 cluster_offset
+ offset_in_cluster
,
1869 cur_bytes
, &hd_qiov
, 0);
1870 qemu_co_mutex_lock(&s
->lock
);
1874 if (bs
->encrypted
) {
1876 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1877 assert((cur_bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1878 if (qcrypto_block_decrypt(s
->crypto
,
1879 (s
->crypt_physical_offset
?
1880 cluster_offset
+ offset_in_cluster
:
1888 qemu_iovec_from_buf(qiov
, bytes_done
, cluster_data
, cur_bytes
);
1893 g_assert_not_reached();
1899 offset
+= cur_bytes
;
1900 bytes_done
+= cur_bytes
;
1905 qemu_co_mutex_unlock(&s
->lock
);
1907 qemu_iovec_destroy(&hd_qiov
);
1908 qemu_vfree(cluster_data
);
1913 /* Check if it's possible to merge a write request with the writing of
1914 * the data from the COW regions */
1915 static bool merge_cow(uint64_t offset
, unsigned bytes
,
1916 QEMUIOVector
*hd_qiov
, QCowL2Meta
*l2meta
)
1920 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
1921 /* If both COW regions are empty then there's nothing to merge */
1922 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
1926 /* The data (middle) region must be immediately after the
1928 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
1932 /* The end region must be immediately after the data (middle)
1934 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
1938 /* Make sure that adding both COW regions to the QEMUIOVector
1939 * does not exceed IOV_MAX */
1940 if (hd_qiov
->niov
> IOV_MAX
- 2) {
1944 m
->data_qiov
= hd_qiov
;
1951 static coroutine_fn
int qcow2_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1952 uint64_t bytes
, QEMUIOVector
*qiov
,
1955 BDRVQcow2State
*s
= bs
->opaque
;
1956 int offset_in_cluster
;
1958 unsigned int cur_bytes
; /* number of sectors in current iteration */
1959 uint64_t cluster_offset
;
1960 QEMUIOVector hd_qiov
;
1961 uint64_t bytes_done
= 0;
1962 uint8_t *cluster_data
= NULL
;
1963 QCowL2Meta
*l2meta
= NULL
;
1965 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
1967 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1969 s
->cluster_cache_offset
= -1; /* disable compressed cache */
1971 qemu_co_mutex_lock(&s
->lock
);
1973 while (bytes
!= 0) {
1977 trace_qcow2_writev_start_part(qemu_coroutine_self());
1978 offset_in_cluster
= offset_into_cluster(s
, offset
);
1979 cur_bytes
= MIN(bytes
, INT_MAX
);
1980 if (bs
->encrypted
) {
1981 cur_bytes
= MIN(cur_bytes
,
1982 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
1983 - offset_in_cluster
);
1986 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
1987 &cluster_offset
, &l2meta
);
1992 assert((cluster_offset
& 511) == 0);
1994 qemu_iovec_reset(&hd_qiov
);
1995 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
1997 if (bs
->encrypted
) {
1999 if (!cluster_data
) {
2000 cluster_data
= qemu_try_blockalign(bs
->file
->bs
,
2001 QCOW_MAX_CRYPT_CLUSTERS
2003 if (cluster_data
== NULL
) {
2009 assert(hd_qiov
.size
<=
2010 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2011 qemu_iovec_to_buf(&hd_qiov
, 0, cluster_data
, hd_qiov
.size
);
2013 if (qcrypto_block_encrypt(s
->crypto
,
2014 (s
->crypt_physical_offset
?
2015 cluster_offset
+ offset_in_cluster
:
2018 cur_bytes
, NULL
) < 0) {
2023 qemu_iovec_reset(&hd_qiov
);
2024 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
2027 ret
= qcow2_pre_write_overlap_check(bs
, 0,
2028 cluster_offset
+ offset_in_cluster
, cur_bytes
);
2033 /* If we need to do COW, check if it's possible to merge the
2034 * writing of the guest data together with that of the COW regions.
2035 * If it's not possible (or not necessary) then write the
2036 * guest data now. */
2037 if (!merge_cow(offset
, cur_bytes
, &hd_qiov
, l2meta
)) {
2038 qemu_co_mutex_unlock(&s
->lock
);
2039 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2040 trace_qcow2_writev_data(qemu_coroutine_self(),
2041 cluster_offset
+ offset_in_cluster
);
2042 ret
= bdrv_co_pwritev(bs
->file
,
2043 cluster_offset
+ offset_in_cluster
,
2044 cur_bytes
, &hd_qiov
, 0);
2045 qemu_co_mutex_lock(&s
->lock
);
2051 while (l2meta
!= NULL
) {
2054 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
2059 /* Take the request off the list of running requests */
2060 if (l2meta
->nb_clusters
!= 0) {
2061 QLIST_REMOVE(l2meta
, next_in_flight
);
2064 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2066 next
= l2meta
->next
;
2072 offset
+= cur_bytes
;
2073 bytes_done
+= cur_bytes
;
2074 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2079 while (l2meta
!= NULL
) {
2082 if (l2meta
->nb_clusters
!= 0) {
2083 QLIST_REMOVE(l2meta
, next_in_flight
);
2085 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2087 next
= l2meta
->next
;
2092 qemu_co_mutex_unlock(&s
->lock
);
2094 qemu_iovec_destroy(&hd_qiov
);
2095 qemu_vfree(cluster_data
);
2096 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2101 static int qcow2_inactivate(BlockDriverState
*bs
)
2103 BDRVQcow2State
*s
= bs
->opaque
;
2104 int ret
, result
= 0;
2105 Error
*local_err
= NULL
;
2107 qcow2_store_persistent_dirty_bitmaps(bs
, &local_err
);
2108 if (local_err
!= NULL
) {
2110 error_report_err(local_err
);
2111 error_report("Persistent bitmaps are lost for node '%s'",
2112 bdrv_get_device_or_node_name(bs
));
2115 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2118 error_report("Failed to flush the L2 table cache: %s",
2122 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2125 error_report("Failed to flush the refcount block cache: %s",
2130 qcow2_mark_clean(bs
);
2136 static void qcow2_close(BlockDriverState
*bs
)
2138 BDRVQcow2State
*s
= bs
->opaque
;
2139 qemu_vfree(s
->l1_table
);
2140 /* else pre-write overlap checks in cache_destroy may crash */
2143 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2144 qcow2_inactivate(bs
);
2147 cache_clean_timer_del(bs
);
2148 qcow2_cache_destroy(s
->l2_table_cache
);
2149 qcow2_cache_destroy(s
->refcount_block_cache
);
2151 qcrypto_block_free(s
->crypto
);
2154 g_free(s
->unknown_header_fields
);
2155 cleanup_unknown_header_ext(bs
);
2157 g_free(s
->image_backing_file
);
2158 g_free(s
->image_backing_format
);
2160 g_free(s
->cluster_cache
);
2161 qemu_vfree(s
->cluster_data
);
2162 qcow2_refcount_close(bs
);
2163 qcow2_free_snapshots(bs
);
2166 static void coroutine_fn
qcow2_co_invalidate_cache(BlockDriverState
*bs
,
2169 BDRVQcow2State
*s
= bs
->opaque
;
2170 int flags
= s
->flags
;
2171 QCryptoBlock
*crypto
= NULL
;
2173 Error
*local_err
= NULL
;
2177 * Backing files are read-only which makes all of their metadata immutable,
2178 * that means we don't have to worry about reopening them here.
2186 memset(s
, 0, sizeof(BDRVQcow2State
));
2187 options
= qdict_clone_shallow(bs
->options
);
2189 flags
&= ~BDRV_O_INACTIVE
;
2190 qemu_co_mutex_lock(&s
->lock
);
2191 ret
= qcow2_do_open(bs
, options
, flags
, &local_err
);
2192 qemu_co_mutex_unlock(&s
->lock
);
2193 qobject_unref(options
);
2195 error_propagate(errp
, local_err
);
2196 error_prepend(errp
, "Could not reopen qcow2 layer: ");
2199 } else if (ret
< 0) {
2200 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
2208 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2209 size_t len
, size_t buflen
)
2211 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2212 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2214 if (buflen
< ext_len
) {
2218 *ext_backing_fmt
= (QCowExtension
) {
2219 .magic
= cpu_to_be32(magic
),
2220 .len
= cpu_to_be32(len
),
2224 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2231 * Updates the qcow2 header, including the variable length parts of it, i.e.
2232 * the backing file name and all extensions. qcow2 was not designed to allow
2233 * such changes, so if we run out of space (we can only use the first cluster)
2234 * this function may fail.
2236 * Returns 0 on success, -errno in error cases.
2238 int qcow2_update_header(BlockDriverState
*bs
)
2240 BDRVQcow2State
*s
= bs
->opaque
;
2243 size_t buflen
= s
->cluster_size
;
2245 uint64_t total_size
;
2246 uint32_t refcount_table_clusters
;
2247 size_t header_length
;
2248 Qcow2UnknownHeaderExtension
*uext
;
2250 buf
= qemu_blockalign(bs
, buflen
);
2252 /* Header structure */
2253 header
= (QCowHeader
*) buf
;
2255 if (buflen
< sizeof(*header
)) {
2260 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2261 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2262 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2264 *header
= (QCowHeader
) {
2265 /* Version 2 fields */
2266 .magic
= cpu_to_be32(QCOW_MAGIC
),
2267 .version
= cpu_to_be32(s
->qcow_version
),
2268 .backing_file_offset
= 0,
2269 .backing_file_size
= 0,
2270 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2271 .size
= cpu_to_be64(total_size
),
2272 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2273 .l1_size
= cpu_to_be32(s
->l1_size
),
2274 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2275 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2276 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2277 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2278 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2280 /* Version 3 fields */
2281 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2282 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2283 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2284 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2285 .header_length
= cpu_to_be32(header_length
),
2288 /* For older versions, write a shorter header */
2289 switch (s
->qcow_version
) {
2291 ret
= offsetof(QCowHeader
, incompatible_features
);
2294 ret
= sizeof(*header
);
2303 memset(buf
, 0, buflen
);
2305 /* Preserve any unknown field in the header */
2306 if (s
->unknown_header_fields_size
) {
2307 if (buflen
< s
->unknown_header_fields_size
) {
2312 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2313 buf
+= s
->unknown_header_fields_size
;
2314 buflen
-= s
->unknown_header_fields_size
;
2317 /* Backing file format header extension */
2318 if (s
->image_backing_format
) {
2319 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2320 s
->image_backing_format
,
2321 strlen(s
->image_backing_format
),
2331 /* Full disk encryption header pointer extension */
2332 if (s
->crypto_header
.offset
!= 0) {
2333 cpu_to_be64s(&s
->crypto_header
.offset
);
2334 cpu_to_be64s(&s
->crypto_header
.length
);
2335 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2336 &s
->crypto_header
, sizeof(s
->crypto_header
),
2338 be64_to_cpus(&s
->crypto_header
.offset
);
2339 be64_to_cpus(&s
->crypto_header
.length
);
2348 if (s
->qcow_version
>= 3) {
2349 Qcow2Feature features
[] = {
2351 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2352 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2353 .name
= "dirty bit",
2356 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2357 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
2358 .name
= "corrupt bit",
2361 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
2362 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
2363 .name
= "lazy refcounts",
2367 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
2368 features
, sizeof(features
), buflen
);
2376 /* Bitmap extension */
2377 if (s
->nb_bitmaps
> 0) {
2378 Qcow2BitmapHeaderExt bitmaps_header
= {
2379 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
2380 .bitmap_directory_size
=
2381 cpu_to_be64(s
->bitmap_directory_size
),
2382 .bitmap_directory_offset
=
2383 cpu_to_be64(s
->bitmap_directory_offset
)
2385 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
2386 &bitmaps_header
, sizeof(bitmaps_header
),
2395 /* Keep unknown header extensions */
2396 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
2397 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
2406 /* End of header extensions */
2407 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
2415 /* Backing file name */
2416 if (s
->image_backing_file
) {
2417 size_t backing_file_len
= strlen(s
->image_backing_file
);
2419 if (buflen
< backing_file_len
) {
2424 /* Using strncpy is ok here, since buf is not NUL-terminated. */
2425 strncpy(buf
, s
->image_backing_file
, buflen
);
2427 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
2428 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
2431 /* Write the new header */
2432 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
2443 static int qcow2_change_backing_file(BlockDriverState
*bs
,
2444 const char *backing_file
, const char *backing_fmt
)
2446 BDRVQcow2State
*s
= bs
->opaque
;
2448 if (backing_file
&& strlen(backing_file
) > 1023) {
2452 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
2453 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
2455 g_free(s
->image_backing_file
);
2456 g_free(s
->image_backing_format
);
2458 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
2459 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
2461 return qcow2_update_header(bs
);
2464 static int qcow2_crypt_method_from_format(const char *encryptfmt
)
2466 if (g_str_equal(encryptfmt
, "luks")) {
2467 return QCOW_CRYPT_LUKS
;
2468 } else if (g_str_equal(encryptfmt
, "aes")) {
2469 return QCOW_CRYPT_AES
;
2475 static int qcow2_set_up_encryption(BlockDriverState
*bs
,
2476 QCryptoBlockCreateOptions
*cryptoopts
,
2479 BDRVQcow2State
*s
= bs
->opaque
;
2480 QCryptoBlock
*crypto
= NULL
;
2483 switch (cryptoopts
->format
) {
2484 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
2485 fmt
= QCOW_CRYPT_LUKS
;
2487 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
2488 fmt
= QCOW_CRYPT_AES
;
2491 error_setg(errp
, "Crypto format not supported in qcow2");
2495 s
->crypt_method_header
= fmt
;
2497 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
2498 qcow2_crypto_hdr_init_func
,
2499 qcow2_crypto_hdr_write_func
,
2505 ret
= qcow2_update_header(bs
);
2507 error_setg_errno(errp
, -ret
, "Could not write encryption header");
2513 qcrypto_block_free(crypto
);
2518 typedef struct PreallocCo
{
2519 BlockDriverState
*bs
;
2521 uint64_t new_length
;
2527 * Preallocates metadata structures for data clusters between @offset (in the
2528 * guest disk) and @new_length (which is thus generally the new guest disk
2531 * Returns: 0 on success, -errno on failure.
2533 static void coroutine_fn
preallocate_co(void *opaque
)
2535 PreallocCo
*params
= opaque
;
2536 BlockDriverState
*bs
= params
->bs
;
2537 uint64_t offset
= params
->offset
;
2538 uint64_t new_length
= params
->new_length
;
2539 BDRVQcow2State
*s
= bs
->opaque
;
2541 uint64_t host_offset
= 0;
2542 unsigned int cur_bytes
;
2546 qemu_co_mutex_lock(&s
->lock
);
2548 assert(offset
<= new_length
);
2549 bytes
= new_length
- offset
;
2552 cur_bytes
= MIN(bytes
, INT_MAX
);
2553 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2554 &host_offset
, &meta
);
2560 QCowL2Meta
*next
= meta
->next
;
2562 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
2564 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
2565 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
2569 /* There are no dependent requests, but we need to remove our
2570 * request from the list of in-flight requests */
2571 QLIST_REMOVE(meta
, next_in_flight
);
2577 /* TODO Preallocate data if requested */
2580 offset
+= cur_bytes
;
2584 * It is expected that the image file is large enough to actually contain
2585 * all of the allocated clusters (otherwise we get failing reads after
2586 * EOF). Extend the image to the last allocated sector.
2588 if (host_offset
!= 0) {
2590 ret
= bdrv_pwrite(bs
->file
, (host_offset
+ cur_bytes
) - 1,
2600 qemu_co_mutex_unlock(&s
->lock
);
2604 static int preallocate(BlockDriverState
*bs
,
2605 uint64_t offset
, uint64_t new_length
)
2607 PreallocCo params
= {
2610 .new_length
= new_length
,
2611 .ret
= -EINPROGRESS
,
2614 if (qemu_in_coroutine()) {
2615 preallocate_co(¶ms
);
2617 Coroutine
*co
= qemu_coroutine_create(preallocate_co
, ¶ms
);
2618 bdrv_coroutine_enter(bs
, co
);
2619 BDRV_POLL_WHILE(bs
, params
.ret
== -EINPROGRESS
);
2624 /* qcow2_refcount_metadata_size:
2625 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2626 * @cluster_size: size of a cluster, in bytes
2627 * @refcount_order: refcount bits power-of-2 exponent
2628 * @generous_increase: allow for the refcount table to be 1.5x as large as it
2631 * Returns: Number of bytes required for refcount blocks and table metadata.
2633 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
2634 int refcount_order
, bool generous_increase
,
2635 uint64_t *refblock_count
)
2638 * Every host cluster is reference-counted, including metadata (even
2639 * refcount metadata is recursively included).
2641 * An accurate formula for the size of refcount metadata size is difficult
2642 * to derive. An easier method of calculation is finding the fixed point
2643 * where no further refcount blocks or table clusters are required to
2644 * reference count every cluster.
2646 int64_t blocks_per_table_cluster
= cluster_size
/ sizeof(uint64_t);
2647 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
2648 int64_t table
= 0; /* number of refcount table clusters */
2649 int64_t blocks
= 0; /* number of refcount block clusters */
2655 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
2656 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
2657 n
= clusters
+ blocks
+ table
;
2659 if (n
== last
&& generous_increase
) {
2660 clusters
+= DIV_ROUND_UP(table
, 2);
2661 n
= 0; /* force another loop */
2662 generous_increase
= false;
2664 } while (n
!= last
);
2666 if (refblock_count
) {
2667 *refblock_count
= blocks
;
2670 return (blocks
+ table
) * cluster_size
;
2674 * qcow2_calc_prealloc_size:
2675 * @total_size: virtual disk size in bytes
2676 * @cluster_size: cluster size in bytes
2677 * @refcount_order: refcount bits power-of-2 exponent
2679 * Returns: Total number of bytes required for the fully allocated image
2680 * (including metadata).
2682 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
2683 size_t cluster_size
,
2686 int64_t meta_size
= 0;
2687 uint64_t nl1e
, nl2e
;
2688 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
2690 /* header: 1 cluster */
2691 meta_size
+= cluster_size
;
2693 /* total size of L2 tables */
2694 nl2e
= aligned_total_size
/ cluster_size
;
2695 nl2e
= ROUND_UP(nl2e
, cluster_size
/ sizeof(uint64_t));
2696 meta_size
+= nl2e
* sizeof(uint64_t);
2698 /* total size of L1 tables */
2699 nl1e
= nl2e
* sizeof(uint64_t) / cluster_size
;
2700 nl1e
= ROUND_UP(nl1e
, cluster_size
/ sizeof(uint64_t));
2701 meta_size
+= nl1e
* sizeof(uint64_t);
2703 /* total size of refcount table and blocks */
2704 meta_size
+= qcow2_refcount_metadata_size(
2705 (meta_size
+ aligned_total_size
) / cluster_size
,
2706 cluster_size
, refcount_order
, false, NULL
);
2708 return meta_size
+ aligned_total_size
;
2711 static bool validate_cluster_size(size_t cluster_size
, Error
**errp
)
2713 int cluster_bits
= ctz32(cluster_size
);
2714 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
2715 (1 << cluster_bits
) != cluster_size
)
2717 error_setg(errp
, "Cluster size must be a power of two between %d and "
2718 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
2724 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, Error
**errp
)
2726 size_t cluster_size
;
2728 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
2729 DEFAULT_CLUSTER_SIZE
);
2730 if (!validate_cluster_size(cluster_size
, errp
)) {
2733 return cluster_size
;
2736 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
2741 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
2743 ret
= 3; /* default */
2744 } else if (!strcmp(buf
, "0.10")) {
2746 } else if (!strcmp(buf
, "1.1")) {
2749 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
2756 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
2759 uint64_t refcount_bits
;
2761 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
2762 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
2763 error_setg(errp
, "Refcount width must be a power of two and may not "
2768 if (version
< 3 && refcount_bits
!= 16) {
2769 error_setg(errp
, "Different refcount widths than 16 bits require "
2770 "compatibility level 1.1 or above (use compat=1.1 or "
2775 return refcount_bits
;
2778 static int coroutine_fn
2779 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
2781 BlockdevCreateOptionsQcow2
*qcow2_opts
;
2785 * Open the image file and write a minimal qcow2 header.
2787 * We keep things simple and start with a zero-sized image. We also
2788 * do without refcount blocks or a L1 table for now. We'll fix the
2789 * inconsistency later.
2791 * We do need a refcount table because growing the refcount table means
2792 * allocating two new refcount blocks - the seconds of which would be at
2793 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2794 * size for any qcow2 image.
2796 BlockBackend
*blk
= NULL
;
2797 BlockDriverState
*bs
= NULL
;
2799 size_t cluster_size
;
2802 uint64_t* refcount_table
;
2803 Error
*local_err
= NULL
;
2806 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
2807 qcow2_opts
= &create_options
->u
.qcow2
;
2809 bs
= bdrv_open_blockdev_ref(qcow2_opts
->file
, errp
);
2814 /* Validate options and set default values */
2815 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
2816 error_setg(errp
, "Image size must be a multiple of 512 bytes");
2821 if (qcow2_opts
->has_version
) {
2822 switch (qcow2_opts
->version
) {
2823 case BLOCKDEV_QCOW2_VERSION_V2
:
2826 case BLOCKDEV_QCOW2_VERSION_V3
:
2830 g_assert_not_reached();
2836 if (qcow2_opts
->has_cluster_size
) {
2837 cluster_size
= qcow2_opts
->cluster_size
;
2839 cluster_size
= DEFAULT_CLUSTER_SIZE
;
2842 if (!validate_cluster_size(cluster_size
, errp
)) {
2847 if (!qcow2_opts
->has_preallocation
) {
2848 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
2850 if (qcow2_opts
->has_backing_file
&&
2851 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
)
2853 error_setg(errp
, "Backing file and preallocation cannot be used at "
2858 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->has_backing_file
) {
2859 error_setg(errp
, "Backing format cannot be used without backing file");
2864 if (!qcow2_opts
->has_lazy_refcounts
) {
2865 qcow2_opts
->lazy_refcounts
= false;
2867 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
2868 error_setg(errp
, "Lazy refcounts only supported with compatibility "
2869 "level 1.1 and above (use version=v3 or greater)");
2874 if (!qcow2_opts
->has_refcount_bits
) {
2875 qcow2_opts
->refcount_bits
= 16;
2877 if (qcow2_opts
->refcount_bits
> 64 ||
2878 !is_power_of_2(qcow2_opts
->refcount_bits
))
2880 error_setg(errp
, "Refcount width must be a power of two and may not "
2885 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
2886 error_setg(errp
, "Different refcount widths than 16 bits require "
2887 "compatibility level 1.1 or above (use version=v3 or "
2892 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
2895 /* Create BlockBackend to write to the image */
2896 blk
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
);
2897 ret
= blk_insert_bs(blk
, bs
, errp
);
2901 blk_set_allow_write_beyond_eof(blk
, true);
2903 /* Clear the protocol layer and preallocate it if necessary */
2904 ret
= blk_truncate(blk
, 0, PREALLOC_MODE_OFF
, errp
);
2909 if (qcow2_opts
->preallocation
== PREALLOC_MODE_FULL
||
2910 qcow2_opts
->preallocation
== PREALLOC_MODE_FALLOC
)
2912 int64_t prealloc_size
=
2913 qcow2_calc_prealloc_size(qcow2_opts
->size
, cluster_size
,
2916 ret
= blk_truncate(blk
, prealloc_size
, qcow2_opts
->preallocation
, errp
);
2922 /* Write the header */
2923 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
2924 header
= g_malloc0(cluster_size
);
2925 *header
= (QCowHeader
) {
2926 .magic
= cpu_to_be32(QCOW_MAGIC
),
2927 .version
= cpu_to_be32(version
),
2928 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
2929 .size
= cpu_to_be64(0),
2930 .l1_table_offset
= cpu_to_be64(0),
2931 .l1_size
= cpu_to_be32(0),
2932 .refcount_table_offset
= cpu_to_be64(cluster_size
),
2933 .refcount_table_clusters
= cpu_to_be32(1),
2934 .refcount_order
= cpu_to_be32(refcount_order
),
2935 .header_length
= cpu_to_be32(sizeof(*header
)),
2938 /* We'll update this to correct value later */
2939 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
2941 if (qcow2_opts
->lazy_refcounts
) {
2942 header
->compatible_features
|=
2943 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
2946 ret
= blk_pwrite(blk
, 0, header
, cluster_size
, 0);
2949 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
2953 /* Write a refcount table with one refcount block */
2954 refcount_table
= g_malloc0(2 * cluster_size
);
2955 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
2956 ret
= blk_pwrite(blk
, cluster_size
, refcount_table
, 2 * cluster_size
, 0);
2957 g_free(refcount_table
);
2960 error_setg_errno(errp
, -ret
, "Could not write refcount table");
2968 * And now open the image and make it consistent first (i.e. increase the
2969 * refcount of the cluster that is occupied by the header and the refcount
2972 options
= qdict_new();
2973 qdict_put_str(options
, "driver", "qcow2");
2974 qdict_put_str(options
, "file", bs
->node_name
);
2975 blk
= blk_new_open(NULL
, NULL
, options
,
2976 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
2979 error_propagate(errp
, local_err
);
2984 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
2986 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
2987 "header and refcount table");
2990 } else if (ret
!= 0) {
2991 error_report("Huh, first cluster in empty image is already in use?");
2995 /* Create a full header (including things like feature table) */
2996 ret
= qcow2_update_header(blk_bs(blk
));
2998 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
3002 /* Okay, now that we have a valid image, let's give it the right size */
3003 ret
= blk_truncate(blk
, qcow2_opts
->size
, PREALLOC_MODE_OFF
, errp
);
3005 error_prepend(errp
, "Could not resize image: ");
3009 /* Want a backing file? There you go.*/
3010 if (qcow2_opts
->has_backing_file
) {
3011 const char *backing_format
= NULL
;
3013 if (qcow2_opts
->has_backing_fmt
) {
3014 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
3017 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
3020 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
3021 "with format '%s'", qcow2_opts
->backing_file
,
3027 /* Want encryption? There you go. */
3028 if (qcow2_opts
->has_encrypt
) {
3029 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3035 /* And if we're supposed to preallocate metadata, do that now */
3036 if (qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
) {
3037 ret
= preallocate(blk_bs(blk
), 0, qcow2_opts
->size
);
3039 error_setg_errno(errp
, -ret
, "Could not preallocate metadata");
3047 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3048 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3049 * have to setup decryption context. We're not doing any I/O on the top
3050 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3053 options
= qdict_new();
3054 qdict_put_str(options
, "driver", "qcow2");
3055 qdict_put_str(options
, "file", bs
->node_name
);
3056 blk
= blk_new_open(NULL
, NULL
, options
,
3057 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3060 error_propagate(errp
, local_err
);
3072 static int coroutine_fn
qcow2_co_create_opts(const char *filename
, QemuOpts
*opts
,
3075 BlockdevCreateOptions
*create_options
= NULL
;
3076 QDict
*qdict
= NULL
;
3079 BlockDriverState
*bs
= NULL
;
3080 Error
*local_err
= NULL
;
3084 /* Only the keyval visitor supports the dotted syntax needed for
3085 * encryption, so go through a QDict before getting a QAPI type. Ignore
3086 * options meant for the protocol layer so that the visitor doesn't
3088 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3091 /* Handle encryption options */
3092 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3093 if (val
&& !strcmp(val
, "on")) {
3094 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3095 } else if (val
&& !strcmp(val
, "off")) {
3096 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3099 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3100 if (val
&& !strcmp(val
, "aes")) {
3101 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3104 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3105 * version=v2/v3 below. */
3106 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3107 if (val
&& !strcmp(val
, "0.10")) {
3108 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3109 } else if (val
&& !strcmp(val
, "1.1")) {
3110 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3113 /* Change legacy command line options into QMP ones */
3114 static const QDictRenames opt_renames
[] = {
3115 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3116 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3117 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3118 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3119 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3120 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3121 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3125 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3130 /* Create and open the file (protocol layer) */
3131 ret
= bdrv_create_file(filename
, opts
, errp
);
3136 bs
= bdrv_open(filename
, NULL
, NULL
,
3137 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3143 /* Set 'driver' and 'node' options */
3144 qdict_put_str(qdict
, "driver", "qcow2");
3145 qdict_put_str(qdict
, "file", bs
->node_name
);
3147 /* Now get the QAPI type BlockdevCreateOptions */
3148 qobj
= qdict_crumple(qdict
, errp
);
3149 qobject_unref(qdict
);
3150 qdict
= qobject_to(QDict
, qobj
);
3151 if (qdict
== NULL
) {
3156 v
= qobject_input_visitor_new_keyval(QOBJECT(qdict
));
3157 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, &local_err
);
3161 error_propagate(errp
, local_err
);
3166 /* Silently round up size */
3167 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3170 /* Create the qcow2 image (format layer) */
3171 ret
= qcow2_co_create(create_options
, errp
);
3178 qobject_unref(qdict
);
3180 qapi_free_BlockdevCreateOptions(create_options
);
3185 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3190 /* Clamp to image length, before checking status of underlying sectors */
3191 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3192 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3198 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3199 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
== bytes
;
3202 static coroutine_fn
int qcow2_co_pwrite_zeroes(BlockDriverState
*bs
,
3203 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3206 BDRVQcow2State
*s
= bs
->opaque
;
3208 uint32_t head
= offset
% s
->cluster_size
;
3209 uint32_t tail
= (offset
+ bytes
) % s
->cluster_size
;
3211 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3212 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3220 assert(head
+ bytes
<= s
->cluster_size
);
3222 /* check whether remainder of cluster already reads as zero */
3223 if (!(is_zero(bs
, offset
- head
, head
) &&
3224 is_zero(bs
, offset
+ bytes
,
3225 tail
? s
->cluster_size
- tail
: 0))) {
3229 qemu_co_mutex_lock(&s
->lock
);
3230 /* We can have new write after previous check */
3231 offset
= QEMU_ALIGN_DOWN(offset
, s
->cluster_size
);
3232 bytes
= s
->cluster_size
;
3233 nr
= s
->cluster_size
;
3234 ret
= qcow2_get_cluster_offset(bs
, offset
, &nr
, &off
);
3235 if (ret
!= QCOW2_CLUSTER_UNALLOCATED
&&
3236 ret
!= QCOW2_CLUSTER_ZERO_PLAIN
&&
3237 ret
!= QCOW2_CLUSTER_ZERO_ALLOC
) {
3238 qemu_co_mutex_unlock(&s
->lock
);
3242 qemu_co_mutex_lock(&s
->lock
);
3245 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
3247 /* Whatever is left can use real zero clusters */
3248 ret
= qcow2_cluster_zeroize(bs
, offset
, bytes
, flags
);
3249 qemu_co_mutex_unlock(&s
->lock
);
3254 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
3255 int64_t offset
, int bytes
)
3258 BDRVQcow2State
*s
= bs
->opaque
;
3260 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
3261 assert(bytes
< s
->cluster_size
);
3262 /* Ignore partial clusters, except for the special case of the
3263 * complete partial cluster at the end of an unaligned file */
3264 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
3265 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3270 qemu_co_mutex_lock(&s
->lock
);
3271 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
3273 qemu_co_mutex_unlock(&s
->lock
);
3277 static int qcow2_truncate(BlockDriverState
*bs
, int64_t offset
,
3278 PreallocMode prealloc
, Error
**errp
)
3280 BDRVQcow2State
*s
= bs
->opaque
;
3281 uint64_t old_length
;
3282 int64_t new_l1_size
;
3285 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
3286 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
3288 error_setg(errp
, "Unsupported preallocation mode '%s'",
3289 PreallocMode_str(prealloc
));
3294 error_setg(errp
, "The new size must be a multiple of 512");
3298 /* cannot proceed if image has snapshots */
3299 if (s
->nb_snapshots
) {
3300 error_setg(errp
, "Can't resize an image which has snapshots");
3304 /* cannot proceed if image has bitmaps */
3305 if (s
->nb_bitmaps
) {
3306 /* TODO: resize bitmaps in the image */
3307 error_setg(errp
, "Can't resize an image which has bitmaps");
3311 old_length
= bs
->total_sectors
* 512;
3312 new_l1_size
= size_to_l1(s
, offset
);
3314 if (offset
< old_length
) {
3315 int64_t last_cluster
, old_file_size
;
3316 if (prealloc
!= PREALLOC_MODE_OFF
) {
3318 "Preallocation can't be used for shrinking an image");
3322 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
3323 old_length
- ROUND_UP(offset
,
3325 QCOW2_DISCARD_ALWAYS
, true);
3327 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
3331 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
3333 error_setg_errno(errp
, -ret
,
3334 "Failed to reduce the number of L2 tables");
3338 ret
= qcow2_shrink_reftable(bs
);
3340 error_setg_errno(errp
, -ret
,
3341 "Failed to discard unused refblocks");
3345 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3346 if (old_file_size
< 0) {
3347 error_setg_errno(errp
, -old_file_size
,
3348 "Failed to inquire current file length");
3349 return old_file_size
;
3351 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
3352 if (last_cluster
< 0) {
3353 error_setg_errno(errp
, -last_cluster
,
3354 "Failed to find the last cluster");
3355 return last_cluster
;
3357 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
3358 Error
*local_err
= NULL
;
3360 bdrv_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
3361 PREALLOC_MODE_OFF
, &local_err
);
3363 warn_reportf_err(local_err
,
3364 "Failed to truncate the tail of the image: ");
3368 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
3370 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
3376 case PREALLOC_MODE_OFF
:
3379 case PREALLOC_MODE_METADATA
:
3380 ret
= preallocate(bs
, old_length
, offset
);
3382 error_setg_errno(errp
, -ret
, "Preallocation failed");
3387 case PREALLOC_MODE_FALLOC
:
3388 case PREALLOC_MODE_FULL
:
3390 int64_t allocation_start
, host_offset
, guest_offset
;
3391 int64_t clusters_allocated
;
3392 int64_t old_file_size
, new_file_size
;
3393 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
3395 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3396 if (old_file_size
< 0) {
3397 error_setg_errno(errp
, -old_file_size
,
3398 "Failed to inquire current file length");
3399 return old_file_size
;
3401 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
3403 nb_new_data_clusters
= DIV_ROUND_UP(offset
- old_length
,
3406 /* This is an overestimation; we will not actually allocate space for
3407 * these in the file but just make sure the new refcount structures are
3408 * able to cover them so we will not have to allocate new refblocks
3409 * while entering the data blocks in the potentially new L2 tables.
3410 * (We do not actually care where the L2 tables are placed. Maybe they
3411 * are already allocated or they can be placed somewhere before
3412 * @old_file_size. It does not matter because they will be fully
3413 * allocated automatically, so they do not need to be covered by the
3414 * preallocation. All that matters is that we will not have to allocate
3415 * new refcount structures for them.) */
3416 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
3417 s
->cluster_size
/ sizeof(uint64_t));
3418 /* The cluster range may not be aligned to L2 boundaries, so add one L2
3419 * table for a potential head/tail */
3422 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
3423 nb_new_data_clusters
+
3426 if (allocation_start
< 0) {
3427 error_setg_errno(errp
, -allocation_start
,
3428 "Failed to resize refcount structures");
3429 return allocation_start
;
3432 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
3433 nb_new_data_clusters
);
3434 if (clusters_allocated
< 0) {
3435 error_setg_errno(errp
, -clusters_allocated
,
3436 "Failed to allocate data clusters");
3437 return -clusters_allocated
;
3440 assert(clusters_allocated
== nb_new_data_clusters
);
3442 /* Allocate the data area */
3443 new_file_size
= allocation_start
+
3444 nb_new_data_clusters
* s
->cluster_size
;
3445 ret
= bdrv_truncate(bs
->file
, new_file_size
, prealloc
, errp
);
3447 error_prepend(errp
, "Failed to resize underlying file: ");
3448 qcow2_free_clusters(bs
, allocation_start
,
3449 nb_new_data_clusters
* s
->cluster_size
,
3450 QCOW2_DISCARD_OTHER
);
3454 /* Create the necessary L2 entries */
3455 host_offset
= allocation_start
;
3456 guest_offset
= old_length
;
3457 while (nb_new_data_clusters
) {
3458 int64_t nb_clusters
= MIN(
3459 nb_new_data_clusters
,
3460 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
3461 QCowL2Meta allocation
= {
3462 .offset
= guest_offset
,
3463 .alloc_offset
= host_offset
,
3464 .nb_clusters
= nb_clusters
,
3466 qemu_co_queue_init(&allocation
.dependent_requests
);
3468 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
3470 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
3471 qcow2_free_clusters(bs
, host_offset
,
3472 nb_new_data_clusters
* s
->cluster_size
,
3473 QCOW2_DISCARD_OTHER
);
3477 guest_offset
+= nb_clusters
* s
->cluster_size
;
3478 host_offset
+= nb_clusters
* s
->cluster_size
;
3479 nb_new_data_clusters
-= nb_clusters
;
3485 g_assert_not_reached();
3488 if (prealloc
!= PREALLOC_MODE_OFF
) {
3489 /* Flush metadata before actually changing the image size */
3490 ret
= bdrv_flush(bs
);
3492 error_setg_errno(errp
, -ret
,
3493 "Failed to flush the preallocated area to disk");
3498 /* write updated header.size */
3499 offset
= cpu_to_be64(offset
);
3500 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
3501 &offset
, sizeof(uint64_t));
3503 error_setg_errno(errp
, -ret
, "Failed to update the image size");
3507 s
->l1_vm_state_index
= new_l1_size
;
3511 /* XXX: put compressed sectors first, then all the cluster aligned
3512 tables to avoid losing bytes in alignment */
3513 static coroutine_fn
int
3514 qcow2_co_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
3515 uint64_t bytes
, QEMUIOVector
*qiov
)
3517 BDRVQcow2State
*s
= bs
->opaque
;
3518 QEMUIOVector hd_qiov
;
3522 uint8_t *buf
, *out_buf
;
3523 int64_t cluster_offset
;
3526 /* align end of file to a sector boundary to ease reading with
3527 sector based I/Os */
3528 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
3529 if (cluster_offset
< 0) {
3530 return cluster_offset
;
3532 return bdrv_truncate(bs
->file
, cluster_offset
, PREALLOC_MODE_OFF
, NULL
);
3535 if (offset_into_cluster(s
, offset
)) {
3539 buf
= qemu_blockalign(bs
, s
->cluster_size
);
3540 if (bytes
!= s
->cluster_size
) {
3541 if (bytes
> s
->cluster_size
||
3542 offset
+ bytes
!= bs
->total_sectors
<< BDRV_SECTOR_BITS
)
3547 /* Zero-pad last write if image size is not cluster aligned */
3548 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
3550 qemu_iovec_to_buf(qiov
, 0, buf
, bytes
);
3552 out_buf
= g_malloc(s
->cluster_size
);
3554 /* best compression, small window, no zlib header */
3555 memset(&strm
, 0, sizeof(strm
));
3556 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
3558 9, Z_DEFAULT_STRATEGY
);
3564 strm
.avail_in
= s
->cluster_size
;
3565 strm
.next_in
= (uint8_t *)buf
;
3566 strm
.avail_out
= s
->cluster_size
;
3567 strm
.next_out
= out_buf
;
3569 ret
= deflate(&strm
, Z_FINISH
);
3570 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
3575 out_len
= strm
.next_out
- out_buf
;
3579 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
3580 /* could not compress: write normal cluster */
3581 ret
= qcow2_co_pwritev(bs
, offset
, bytes
, qiov
, 0);
3588 qemu_co_mutex_lock(&s
->lock
);
3590 qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
);
3591 if (!cluster_offset
) {
3592 qemu_co_mutex_unlock(&s
->lock
);
3596 cluster_offset
&= s
->cluster_offset_mask
;
3598 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
);
3599 qemu_co_mutex_unlock(&s
->lock
);
3604 iov
= (struct iovec
) {
3605 .iov_base
= out_buf
,
3608 qemu_iovec_init_external(&hd_qiov
, &iov
, 1);
3610 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
3611 ret
= bdrv_co_pwritev(bs
->file
, cluster_offset
, out_len
, &hd_qiov
, 0);
3623 static int make_completely_empty(BlockDriverState
*bs
)
3625 BDRVQcow2State
*s
= bs
->opaque
;
3626 Error
*local_err
= NULL
;
3627 int ret
, l1_clusters
;
3629 uint64_t *new_reftable
= NULL
;
3630 uint64_t rt_entry
, l1_size2
;
3633 uint64_t reftable_offset
;
3634 uint32_t reftable_clusters
;
3635 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
3637 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
3642 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
3647 /* Refcounts will be broken utterly */
3648 ret
= qcow2_mark_dirty(bs
);
3653 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
3655 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
3656 l1_size2
= (uint64_t)s
->l1_size
* sizeof(uint64_t);
3658 /* After this call, neither the in-memory nor the on-disk refcount
3659 * information accurately describe the actual references */
3661 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
3662 l1_clusters
* s
->cluster_size
, 0);
3664 goto fail_broken_refcounts
;
3666 memset(s
->l1_table
, 0, l1_size2
);
3668 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
3670 /* Overwrite enough clusters at the beginning of the sectors to place
3671 * the refcount table, a refcount block and the L1 table in; this may
3672 * overwrite parts of the existing refcount and L1 table, which is not
3673 * an issue because the dirty flag is set, complete data loss is in fact
3674 * desired and partial data loss is consequently fine as well */
3675 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
3676 (2 + l1_clusters
) * s
->cluster_size
, 0);
3677 /* This call (even if it failed overall) may have overwritten on-disk
3678 * refcount structures; in that case, the in-memory refcount information
3679 * will probably differ from the on-disk information which makes the BDS
3682 goto fail_broken_refcounts
;
3685 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
3686 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
3688 /* "Create" an empty reftable (one cluster) directly after the image
3689 * header and an empty L1 table three clusters after the image header;
3690 * the cluster between those two will be used as the first refblock */
3691 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
3692 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
3693 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
3694 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
3695 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
3697 goto fail_broken_refcounts
;
3700 s
->l1_table_offset
= 3 * s
->cluster_size
;
3702 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ sizeof(uint64_t));
3703 if (!new_reftable
) {
3705 goto fail_broken_refcounts
;
3708 s
->refcount_table_offset
= s
->cluster_size
;
3709 s
->refcount_table_size
= s
->cluster_size
/ sizeof(uint64_t);
3710 s
->max_refcount_table_index
= 0;
3712 g_free(s
->refcount_table
);
3713 s
->refcount_table
= new_reftable
;
3714 new_reftable
= NULL
;
3716 /* Now the in-memory refcount information again corresponds to the on-disk
3717 * information (reftable is empty and no refblocks (the refblock cache is
3718 * empty)); however, this means some clusters (e.g. the image header) are
3719 * referenced, but not refcounted, but the normal qcow2 code assumes that
3720 * the in-memory information is always correct */
3722 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
3724 /* Enter the first refblock into the reftable */
3725 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
3726 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
3727 &rt_entry
, sizeof(rt_entry
));
3729 goto fail_broken_refcounts
;
3731 s
->refcount_table
[0] = 2 * s
->cluster_size
;
3733 s
->free_cluster_index
= 0;
3734 assert(3 + l1_clusters
<= s
->refcount_block_size
);
3735 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
3738 goto fail_broken_refcounts
;
3739 } else if (offset
> 0) {
3740 error_report("First cluster in emptied image is in use");
3744 /* Now finally the in-memory information corresponds to the on-disk
3745 * structures and is correct */
3746 ret
= qcow2_mark_clean(bs
);
3751 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
,
3752 PREALLOC_MODE_OFF
, &local_err
);
3754 error_report_err(local_err
);
3760 fail_broken_refcounts
:
3761 /* The BDS is unusable at this point. If we wanted to make it usable, we
3762 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
3763 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
3764 * again. However, because the functions which could have caused this error
3765 * path to be taken are used by those functions as well, it's very likely
3766 * that that sequence will fail as well. Therefore, just eject the BDS. */
3770 g_free(new_reftable
);
3774 static int qcow2_make_empty(BlockDriverState
*bs
)
3776 BDRVQcow2State
*s
= bs
->opaque
;
3777 uint64_t offset
, end_offset
;
3778 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
3779 int l1_clusters
, ret
= 0;
3781 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
3783 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
3784 3 + l1_clusters
<= s
->refcount_block_size
&&
3785 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
3786 /* The following function only works for qcow2 v3 images (it
3787 * requires the dirty flag) and only as long as there are no
3788 * features that reserve extra clusters (such as snapshots,
3789 * LUKS header, or persistent bitmaps), because it completely
3790 * empties the image. Furthermore, the L1 table and three
3791 * additional clusters (image header, refcount table, one
3792 * refcount block) have to fit inside one refcount block. */
3793 return make_completely_empty(bs
);
3796 /* This fallback code simply discards every active cluster; this is slow,
3797 * but works in all cases */
3798 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3799 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
3800 /* As this function is generally used after committing an external
3801 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
3802 * default action for this kind of discard is to pass the discard,
3803 * which will ideally result in an actually smaller image file, as
3804 * is probably desired. */
3805 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
3806 QCOW2_DISCARD_SNAPSHOT
, true);
3815 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
3817 BDRVQcow2State
*s
= bs
->opaque
;
3820 qemu_co_mutex_lock(&s
->lock
);
3821 ret
= qcow2_write_caches(bs
);
3822 qemu_co_mutex_unlock(&s
->lock
);
3827 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
3830 Error
*local_err
= NULL
;
3831 BlockMeasureInfo
*info
;
3832 uint64_t required
= 0; /* bytes that contribute to required size */
3833 uint64_t virtual_size
; /* disk size as seen by guest */
3834 uint64_t refcount_bits
;
3836 size_t cluster_size
;
3839 PreallocMode prealloc
;
3840 bool has_backing_file
;
3842 /* Parse image creation options */
3843 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, &local_err
);
3848 version
= qcow2_opt_get_version_del(opts
, &local_err
);
3853 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
3858 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
3859 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
3860 PREALLOC_MODE_OFF
, &local_err
);
3866 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
3867 has_backing_file
= !!optstr
;
3870 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
3871 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
3873 /* Check that virtual disk size is valid */
3874 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
3875 cluster_size
/ sizeof(uint64_t));
3876 if (l2_tables
* sizeof(uint64_t) > QCOW_MAX_L1_SIZE
) {
3877 error_setg(&local_err
, "The image size is too large "
3878 "(try using a larger cluster size)");
3882 /* Account for input image */
3884 int64_t ssize
= bdrv_getlength(in_bs
);
3886 error_setg_errno(&local_err
, -ssize
,
3887 "Unable to get image virtual_size");
3891 virtual_size
= ROUND_UP(ssize
, cluster_size
);
3893 if (has_backing_file
) {
3894 /* We don't how much of the backing chain is shared by the input
3895 * image and the new image file. In the worst case the new image's
3896 * backing file has nothing in common with the input image. Be
3897 * conservative and assume all clusters need to be written.
3899 required
= virtual_size
;
3904 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
3907 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
3908 ssize
- offset
, &pnum
, NULL
,
3911 error_setg_errno(&local_err
, -ret
,
3912 "Unable to get block status");
3916 if (ret
& BDRV_BLOCK_ZERO
) {
3917 /* Skip zero regions (safe with no backing file) */
3918 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
3919 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
3920 /* Extend pnum to end of cluster for next iteration */
3921 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
3923 /* Count clusters we've seen */
3924 required
+= offset
% cluster_size
+ pnum
;
3930 /* Take into account preallocation. Nothing special is needed for
3931 * PREALLOC_MODE_METADATA since metadata is always counted.
3933 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
3934 required
= virtual_size
;
3937 info
= g_new(BlockMeasureInfo
, 1);
3938 info
->fully_allocated
=
3939 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
3940 ctz32(refcount_bits
));
3942 /* Remove data clusters that are not required. This overestimates the
3943 * required size because metadata needed for the fully allocated file is
3946 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
3950 error_propagate(errp
, local_err
);
3954 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3956 BDRVQcow2State
*s
= bs
->opaque
;
3957 bdi
->unallocated_blocks_are_zero
= true;
3958 bdi
->cluster_size
= s
->cluster_size
;
3959 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
3963 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
)
3965 BDRVQcow2State
*s
= bs
->opaque
;
3966 ImageInfoSpecific
*spec_info
;
3967 QCryptoBlockInfo
*encrypt_info
= NULL
;
3969 if (s
->crypto
!= NULL
) {
3970 encrypt_info
= qcrypto_block_get_info(s
->crypto
, &error_abort
);
3973 spec_info
= g_new(ImageInfoSpecific
, 1);
3974 *spec_info
= (ImageInfoSpecific
){
3975 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
3976 .u
.qcow2
.data
= g_new(ImageInfoSpecificQCow2
, 1),
3978 if (s
->qcow_version
== 2) {
3979 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
3980 .compat
= g_strdup("0.10"),
3981 .refcount_bits
= s
->refcount_bits
,
3983 } else if (s
->qcow_version
== 3) {
3984 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
3985 .compat
= g_strdup("1.1"),
3986 .lazy_refcounts
= s
->compatible_features
&
3987 QCOW2_COMPAT_LAZY_REFCOUNTS
,
3988 .has_lazy_refcounts
= true,
3989 .corrupt
= s
->incompatible_features
&
3990 QCOW2_INCOMPAT_CORRUPT
,
3991 .has_corrupt
= true,
3992 .refcount_bits
= s
->refcount_bits
,
3995 /* if this assertion fails, this probably means a new version was
3996 * added without having it covered here */
4001 ImageInfoSpecificQCow2Encryption
*qencrypt
=
4002 g_new(ImageInfoSpecificQCow2Encryption
, 1);
4003 switch (encrypt_info
->format
) {
4004 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
4005 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
4006 qencrypt
->u
.aes
= encrypt_info
->u
.qcow
;
4008 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
4009 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
4010 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
4015 /* Since we did shallow copy above, erase any pointers
4016 * in the original info */
4017 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
4018 qapi_free_QCryptoBlockInfo(encrypt_info
);
4020 spec_info
->u
.qcow2
.data
->has_encrypt
= true;
4021 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
4027 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
4030 BDRVQcow2State
*s
= bs
->opaque
;
4032 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
4033 return bs
->drv
->bdrv_co_pwritev(bs
, qcow2_vm_state_offset(s
) + pos
,
4034 qiov
->size
, qiov
, 0);
4037 static int qcow2_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
4040 BDRVQcow2State
*s
= bs
->opaque
;
4042 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
4043 return bs
->drv
->bdrv_co_preadv(bs
, qcow2_vm_state_offset(s
) + pos
,
4044 qiov
->size
, qiov
, 0);
4048 * Downgrades an image's version. To achieve this, any incompatible features
4049 * have to be removed.
4051 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
4052 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
)
4054 BDRVQcow2State
*s
= bs
->opaque
;
4055 int current_version
= s
->qcow_version
;
4058 if (target_version
== current_version
) {
4060 } else if (target_version
> current_version
) {
4062 } else if (target_version
!= 2) {
4066 if (s
->refcount_order
!= 4) {
4067 error_report("compat=0.10 requires refcount_bits=16");
4071 /* clear incompatible features */
4072 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
4073 ret
= qcow2_mark_clean(bs
);
4079 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
4080 * the first place; if that happens nonetheless, returning -ENOTSUP is the
4081 * best thing to do anyway */
4083 if (s
->incompatible_features
) {
4087 /* since we can ignore compatible features, we can set them to 0 as well */
4088 s
->compatible_features
= 0;
4089 /* if lazy refcounts have been used, they have already been fixed through
4090 * clearing the dirty flag */
4092 /* clearing autoclear features is trivial */
4093 s
->autoclear_features
= 0;
4095 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
4100 s
->qcow_version
= target_version
;
4101 ret
= qcow2_update_header(bs
);
4103 s
->qcow_version
= current_version
;
4109 typedef enum Qcow2AmendOperation
{
4110 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
4111 * statically initialized to so that the helper CB can discern the first
4112 * invocation from an operation change */
4113 QCOW2_NO_OPERATION
= 0,
4115 QCOW2_CHANGING_REFCOUNT_ORDER
,
4117 } Qcow2AmendOperation
;
4119 typedef struct Qcow2AmendHelperCBInfo
{
4120 /* The code coordinating the amend operations should only modify
4121 * these four fields; the rest will be managed by the CB */
4122 BlockDriverAmendStatusCB
*original_status_cb
;
4123 void *original_cb_opaque
;
4125 Qcow2AmendOperation current_operation
;
4127 /* Total number of operations to perform (only set once) */
4128 int total_operations
;
4130 /* The following fields are managed by the CB */
4132 /* Number of operations completed */
4133 int operations_completed
;
4135 /* Cumulative offset of all completed operations */
4136 int64_t offset_completed
;
4138 Qcow2AmendOperation last_operation
;
4139 int64_t last_work_size
;
4140 } Qcow2AmendHelperCBInfo
;
4142 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
4143 int64_t operation_offset
,
4144 int64_t operation_work_size
, void *opaque
)
4146 Qcow2AmendHelperCBInfo
*info
= opaque
;
4147 int64_t current_work_size
;
4148 int64_t projected_work_size
;
4150 if (info
->current_operation
!= info
->last_operation
) {
4151 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
4152 info
->offset_completed
+= info
->last_work_size
;
4153 info
->operations_completed
++;
4156 info
->last_operation
= info
->current_operation
;
4159 assert(info
->total_operations
> 0);
4160 assert(info
->operations_completed
< info
->total_operations
);
4162 info
->last_work_size
= operation_work_size
;
4164 current_work_size
= info
->offset_completed
+ operation_work_size
;
4166 /* current_work_size is the total work size for (operations_completed + 1)
4167 * operations (which includes this one), so multiply it by the number of
4168 * operations not covered and divide it by the number of operations
4169 * covered to get a projection for the operations not covered */
4170 projected_work_size
= current_work_size
* (info
->total_operations
-
4171 info
->operations_completed
- 1)
4172 / (info
->operations_completed
+ 1);
4174 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
4175 current_work_size
+ projected_work_size
,
4176 info
->original_cb_opaque
);
4179 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4180 BlockDriverAmendStatusCB
*status_cb
,
4183 BDRVQcow2State
*s
= bs
->opaque
;
4184 int old_version
= s
->qcow_version
, new_version
= old_version
;
4185 uint64_t new_size
= 0;
4186 const char *backing_file
= NULL
, *backing_format
= NULL
;
4187 bool lazy_refcounts
= s
->use_lazy_refcounts
;
4188 const char *compat
= NULL
;
4189 uint64_t cluster_size
= s
->cluster_size
;
4192 int refcount_bits
= s
->refcount_bits
;
4193 Error
*local_err
= NULL
;
4195 QemuOptDesc
*desc
= opts
->list
->desc
;
4196 Qcow2AmendHelperCBInfo helper_cb_info
;
4198 while (desc
&& desc
->name
) {
4199 if (!qemu_opt_find(opts
, desc
->name
)) {
4200 /* only change explicitly defined options */
4205 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
4206 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
4208 /* preserve default */
4209 } else if (!strcmp(compat
, "0.10")) {
4211 } else if (!strcmp(compat
, "1.1")) {
4214 error_report("Unknown compatibility level %s", compat
);
4217 } else if (!strcmp(desc
->name
, BLOCK_OPT_PREALLOC
)) {
4218 error_report("Cannot change preallocation mode");
4220 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
4221 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
4222 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
4223 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4224 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
4225 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4226 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT
)) {
4227 encrypt
= qemu_opt_get_bool(opts
, BLOCK_OPT_ENCRYPT
,
4230 if (encrypt
!= !!s
->crypto
) {
4231 error_report("Changing the encryption flag is not supported");
4234 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT_FORMAT
)) {
4235 encformat
= qcow2_crypt_method_from_format(
4236 qemu_opt_get(opts
, BLOCK_OPT_ENCRYPT_FORMAT
));
4238 if (encformat
!= s
->crypt_method_header
) {
4239 error_report("Changing the encryption format is not supported");
4242 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
4243 error_report("Changing the encryption parameters is not supported");
4245 } else if (!strcmp(desc
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
4246 cluster_size
= qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
,
4248 if (cluster_size
!= s
->cluster_size
) {
4249 error_report("Changing the cluster size is not supported");
4252 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
4253 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
4255 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
4256 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
4259 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
4260 !is_power_of_2(refcount_bits
))
4262 error_report("Refcount width must be a power of two and may "
4263 "not exceed 64 bits");
4267 /* if this point is reached, this probably means a new option was
4268 * added without having it covered here */
4275 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
4276 .original_status_cb
= status_cb
,
4277 .original_cb_opaque
= cb_opaque
,
4278 .total_operations
= (new_version
< old_version
)
4279 + (s
->refcount_bits
!= refcount_bits
)
4282 /* Upgrade first (some features may require compat=1.1) */
4283 if (new_version
> old_version
) {
4284 s
->qcow_version
= new_version
;
4285 ret
= qcow2_update_header(bs
);
4287 s
->qcow_version
= old_version
;
4292 if (s
->refcount_bits
!= refcount_bits
) {
4293 int refcount_order
= ctz32(refcount_bits
);
4295 if (new_version
< 3 && refcount_bits
!= 16) {
4296 error_report("Different refcount widths than 16 bits require "
4297 "compatibility level 1.1 or above (use compat=1.1 or "
4302 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
4303 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
4304 &qcow2_amend_helper_cb
,
4305 &helper_cb_info
, &local_err
);
4307 error_report_err(local_err
);
4312 if (backing_file
|| backing_format
) {
4313 ret
= qcow2_change_backing_file(bs
,
4314 backing_file
?: s
->image_backing_file
,
4315 backing_format
?: s
->image_backing_format
);
4321 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
4322 if (lazy_refcounts
) {
4323 if (new_version
< 3) {
4324 error_report("Lazy refcounts only supported with compatibility "
4325 "level 1.1 and above (use compat=1.1 or greater)");
4328 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
4329 ret
= qcow2_update_header(bs
);
4331 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
4334 s
->use_lazy_refcounts
= true;
4336 /* make image clean first */
4337 ret
= qcow2_mark_clean(bs
);
4341 /* now disallow lazy refcounts */
4342 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
4343 ret
= qcow2_update_header(bs
);
4345 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
4348 s
->use_lazy_refcounts
= false;
4353 BlockBackend
*blk
= blk_new(BLK_PERM_RESIZE
, BLK_PERM_ALL
);
4354 ret
= blk_insert_bs(blk
, bs
, &local_err
);
4356 error_report_err(local_err
);
4361 ret
= blk_truncate(blk
, new_size
, PREALLOC_MODE_OFF
, &local_err
);
4364 error_report_err(local_err
);
4369 /* Downgrade last (so unsupported features can be removed before) */
4370 if (new_version
< old_version
) {
4371 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
4372 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
4383 * If offset or size are negative, respectively, they will not be included in
4384 * the BLOCK_IMAGE_CORRUPTED event emitted.
4385 * fatal will be ignored for read-only BDS; corruptions found there will always
4386 * be considered non-fatal.
4388 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
4389 int64_t size
, const char *message_format
, ...)
4391 BDRVQcow2State
*s
= bs
->opaque
;
4392 const char *node_name
;
4396 fatal
= fatal
&& !bs
->read_only
;
4398 if (s
->signaled_corruption
&&
4399 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
4404 va_start(ap
, message_format
);
4405 message
= g_strdup_vprintf(message_format
, ap
);
4409 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
4410 "corruption events will be suppressed\n", message
);
4412 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
4413 "corruption events will be suppressed\n", message
);
4416 node_name
= bdrv_get_node_name(bs
);
4417 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
4418 *node_name
!= '\0', node_name
,
4419 message
, offset
>= 0, offset
,
4421 fatal
, &error_abort
);
4425 qcow2_mark_corrupt(bs
);
4426 bs
->drv
= NULL
; /* make BDS unusable */
4429 s
->signaled_corruption
= true;
4432 static QemuOptsList qcow2_create_opts
= {
4433 .name
= "qcow2-create-opts",
4434 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
4437 .name
= BLOCK_OPT_SIZE
,
4438 .type
= QEMU_OPT_SIZE
,
4439 .help
= "Virtual disk size"
4442 .name
= BLOCK_OPT_COMPAT_LEVEL
,
4443 .type
= QEMU_OPT_STRING
,
4444 .help
= "Compatibility level (0.10 or 1.1)"
4447 .name
= BLOCK_OPT_BACKING_FILE
,
4448 .type
= QEMU_OPT_STRING
,
4449 .help
= "File name of a base image"
4452 .name
= BLOCK_OPT_BACKING_FMT
,
4453 .type
= QEMU_OPT_STRING
,
4454 .help
= "Image format of the base image"
4457 .name
= BLOCK_OPT_ENCRYPT
,
4458 .type
= QEMU_OPT_BOOL
,
4459 .help
= "Encrypt the image with format 'aes'. (Deprecated "
4460 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)",
4463 .name
= BLOCK_OPT_ENCRYPT_FORMAT
,
4464 .type
= QEMU_OPT_STRING
,
4465 .help
= "Encrypt the image, format choices: 'aes', 'luks'",
4467 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
4468 "ID of secret providing qcow AES key or LUKS passphrase"),
4469 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
4470 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
4471 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
4472 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
4473 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
4474 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
4476 .name
= BLOCK_OPT_CLUSTER_SIZE
,
4477 .type
= QEMU_OPT_SIZE
,
4478 .help
= "qcow2 cluster size",
4479 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
)
4482 .name
= BLOCK_OPT_PREALLOC
,
4483 .type
= QEMU_OPT_STRING
,
4484 .help
= "Preallocation mode (allowed values: off, metadata, "
4488 .name
= BLOCK_OPT_LAZY_REFCOUNTS
,
4489 .type
= QEMU_OPT_BOOL
,
4490 .help
= "Postpone refcount updates",
4491 .def_value_str
= "off"
4494 .name
= BLOCK_OPT_REFCOUNT_BITS
,
4495 .type
= QEMU_OPT_NUMBER
,
4496 .help
= "Width of a reference count entry in bits",
4497 .def_value_str
= "16"
4499 { /* end of list */ }
4503 BlockDriver bdrv_qcow2
= {
4504 .format_name
= "qcow2",
4505 .instance_size
= sizeof(BDRVQcow2State
),
4506 .bdrv_probe
= qcow2_probe
,
4507 .bdrv_open
= qcow2_open
,
4508 .bdrv_close
= qcow2_close
,
4509 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
4510 .bdrv_reopen_commit
= qcow2_reopen_commit
,
4511 .bdrv_reopen_abort
= qcow2_reopen_abort
,
4512 .bdrv_join_options
= qcow2_join_options
,
4513 .bdrv_child_perm
= bdrv_format_default_perms
,
4514 .bdrv_co_create_opts
= qcow2_co_create_opts
,
4515 .bdrv_co_create
= qcow2_co_create
,
4516 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
4517 .bdrv_co_block_status
= qcow2_co_block_status
,
4519 .bdrv_co_preadv
= qcow2_co_preadv
,
4520 .bdrv_co_pwritev
= qcow2_co_pwritev
,
4521 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
4523 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
4524 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
4525 .bdrv_truncate
= qcow2_truncate
,
4526 .bdrv_co_pwritev_compressed
= qcow2_co_pwritev_compressed
,
4527 .bdrv_make_empty
= qcow2_make_empty
,
4529 .bdrv_snapshot_create
= qcow2_snapshot_create
,
4530 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
4531 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
4532 .bdrv_snapshot_list
= qcow2_snapshot_list
,
4533 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
4534 .bdrv_measure
= qcow2_measure
,
4535 .bdrv_get_info
= qcow2_get_info
,
4536 .bdrv_get_specific_info
= qcow2_get_specific_info
,
4538 .bdrv_save_vmstate
= qcow2_save_vmstate
,
4539 .bdrv_load_vmstate
= qcow2_load_vmstate
,
4541 .supports_backing
= true,
4542 .bdrv_change_backing_file
= qcow2_change_backing_file
,
4544 .bdrv_refresh_limits
= qcow2_refresh_limits
,
4545 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
4546 .bdrv_inactivate
= qcow2_inactivate
,
4548 .create_opts
= &qcow2_create_opts
,
4549 .bdrv_co_check
= qcow2_co_check
,
4550 .bdrv_amend_options
= qcow2_amend_options
,
4552 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
4553 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
4555 .bdrv_reopen_bitmaps_rw
= qcow2_reopen_bitmaps_rw
,
4556 .bdrv_can_store_new_dirty_bitmap
= qcow2_can_store_new_dirty_bitmap
,
4557 .bdrv_remove_persistent_dirty_bitmap
= qcow2_remove_persistent_dirty_bitmap
,
4560 static void bdrv_qcow2_init(void)
4562 bdrv_register(&bdrv_qcow2
);
4565 block_init(bdrv_qcow2_init
);