2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "block/block_int.h"
27 #include "sysemu/block-backend.h"
28 #include "qemu/module.h"
30 #include "block/qcow2.h"
31 #include "qemu/error-report.h"
32 #include "qapi/error.h"
33 #include "qapi/qapi-events-block-core.h"
34 #include "qapi/qmp/qdict.h"
35 #include "qapi/qmp/qstring.h"
37 #include "qemu/option_int.h"
38 #include "qemu/cutils.h"
39 #include "qemu/bswap.h"
40 #include "qapi/opts-visitor.h"
41 #include "block/crypto.h"
44 Differences with QCOW:
46 - Support for multiple incremental snapshots.
47 - Memory management by reference counts.
48 - Clusters which have a reference count of one have the bit
49 QCOW_OFLAG_COPIED to optimize write performance.
50 - Size of compressed clusters is stored in sectors to reduce bit usage
51 in the cluster offsets.
52 - Support for storing additional data (such as the VM state) in the
54 - If a backing store is used, the cluster size is not constrained
55 (could be backported to QCOW).
56 - L2 tables have always a size of one cluster.
63 } QEMU_PACKED QCowExtension
;
65 #define QCOW2_EXT_MAGIC_END 0
66 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
67 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
68 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
69 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
71 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
73 const QCowHeader
*cow_header
= (const void *)buf
;
75 if (buf_size
>= sizeof(QCowHeader
) &&
76 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
77 be32_to_cpu(cow_header
->version
) >= 2)
84 static ssize_t
qcow2_crypto_hdr_read_func(QCryptoBlock
*block
, size_t offset
,
85 uint8_t *buf
, size_t buflen
,
86 void *opaque
, Error
**errp
)
88 BlockDriverState
*bs
= opaque
;
89 BDRVQcow2State
*s
= bs
->opaque
;
92 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
93 error_setg(errp
, "Request for data outside of extension header");
97 ret
= bdrv_pread(bs
->file
,
98 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
100 error_setg_errno(errp
, -ret
, "Could not read encryption header");
107 static ssize_t
qcow2_crypto_hdr_init_func(QCryptoBlock
*block
, size_t headerlen
,
108 void *opaque
, Error
**errp
)
110 BlockDriverState
*bs
= opaque
;
111 BDRVQcow2State
*s
= bs
->opaque
;
115 ret
= qcow2_alloc_clusters(bs
, headerlen
);
117 error_setg_errno(errp
, -ret
,
118 "Cannot allocate cluster for LUKS header size %zu",
123 s
->crypto_header
.length
= headerlen
;
124 s
->crypto_header
.offset
= ret
;
126 /* Zero fill remaining space in cluster so it has predictable
127 * content in case of future spec changes */
128 clusterlen
= size_to_clusters(s
, headerlen
) * s
->cluster_size
;
129 assert(qcow2_pre_write_overlap_check(bs
, 0, ret
, clusterlen
) == 0);
130 ret
= bdrv_pwrite_zeroes(bs
->file
,
132 clusterlen
- headerlen
, 0);
134 error_setg_errno(errp
, -ret
, "Could not zero fill encryption header");
142 static ssize_t
qcow2_crypto_hdr_write_func(QCryptoBlock
*block
, size_t offset
,
143 const uint8_t *buf
, size_t buflen
,
144 void *opaque
, Error
**errp
)
146 BlockDriverState
*bs
= opaque
;
147 BDRVQcow2State
*s
= bs
->opaque
;
150 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
151 error_setg(errp
, "Request for data outside of extension header");
155 ret
= bdrv_pwrite(bs
->file
,
156 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
158 error_setg_errno(errp
, -ret
, "Could not read encryption header");
166 * read qcow2 extension and fill bs
167 * start reading from start_offset
168 * finish reading upon magic of value 0 or when end_offset reached
169 * unknown magic is skipped (future extension this version knows nothing about)
170 * return 0 upon success, non-0 otherwise
172 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
173 uint64_t end_offset
, void **p_feature_table
,
174 int flags
, bool *need_update_header
,
177 BDRVQcow2State
*s
= bs
->opaque
;
181 Qcow2BitmapHeaderExt bitmaps_ext
;
183 if (need_update_header
!= NULL
) {
184 *need_update_header
= false;
188 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
190 offset
= start_offset
;
191 while (offset
< end_offset
) {
195 if (offset
> s
->cluster_size
)
196 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
198 printf("attempting to read extended header in offset %lu\n", offset
);
201 ret
= bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
));
203 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
204 "pread fail from offset %" PRIu64
, offset
);
207 be32_to_cpus(&ext
.magic
);
208 be32_to_cpus(&ext
.len
);
209 offset
+= sizeof(ext
);
211 printf("ext.magic = 0x%x\n", ext
.magic
);
213 if (offset
> end_offset
|| ext
.len
> end_offset
- offset
) {
214 error_setg(errp
, "Header extension too large");
219 case QCOW2_EXT_MAGIC_END
:
222 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
223 if (ext
.len
>= sizeof(bs
->backing_format
)) {
224 error_setg(errp
, "ERROR: ext_backing_format: len=%" PRIu32
225 " too large (>=%zu)", ext
.len
,
226 sizeof(bs
->backing_format
));
229 ret
= bdrv_pread(bs
->file
, offset
, bs
->backing_format
, ext
.len
);
231 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
232 "Could not read format name");
235 bs
->backing_format
[ext
.len
] = '\0';
236 s
->image_backing_format
= g_strdup(bs
->backing_format
);
238 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
242 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
243 if (p_feature_table
!= NULL
) {
244 void* feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
245 ret
= bdrv_pread(bs
->file
, offset
, feature_table
, ext
.len
);
247 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
248 "Could not read table");
252 *p_feature_table
= feature_table
;
256 case QCOW2_EXT_MAGIC_CRYPTO_HEADER
: {
257 unsigned int cflags
= 0;
258 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
259 error_setg(errp
, "CRYPTO header extension only "
260 "expected with LUKS encryption method");
263 if (ext
.len
!= sizeof(Qcow2CryptoHeaderExtension
)) {
264 error_setg(errp
, "CRYPTO header extension size %u, "
265 "but expected size %zu", ext
.len
,
266 sizeof(Qcow2CryptoHeaderExtension
));
270 ret
= bdrv_pread(bs
->file
, offset
, &s
->crypto_header
, ext
.len
);
272 error_setg_errno(errp
, -ret
,
273 "Unable to read CRYPTO header extension");
276 be64_to_cpus(&s
->crypto_header
.offset
);
277 be64_to_cpus(&s
->crypto_header
.length
);
279 if ((s
->crypto_header
.offset
% s
->cluster_size
) != 0) {
280 error_setg(errp
, "Encryption header offset '%" PRIu64
"' is "
281 "not a multiple of cluster size '%u'",
282 s
->crypto_header
.offset
, s
->cluster_size
);
286 if (flags
& BDRV_O_NO_IO
) {
287 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
289 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
290 qcow2_crypto_hdr_read_func
,
297 case QCOW2_EXT_MAGIC_BITMAPS
:
298 if (ext
.len
!= sizeof(bitmaps_ext
)) {
299 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
300 "Invalid extension length");
304 if (!(s
->autoclear_features
& QCOW2_AUTOCLEAR_BITMAPS
)) {
305 if (s
->qcow_version
< 3) {
306 /* Let's be a bit more specific */
307 warn_report("This qcow2 v2 image contains bitmaps, but "
308 "they may have been modified by a program "
309 "without persistent bitmap support; so now "
310 "they must all be considered inconsistent");
312 warn_report("a program lacking bitmap support "
313 "modified this file, so all bitmaps are now "
314 "considered inconsistent");
316 error_printf("Some clusters may be leaked, "
317 "run 'qemu-img check -r' on the image "
319 if (need_update_header
!= NULL
) {
320 /* Updating is needed to drop invalid bitmap extension. */
321 *need_update_header
= true;
326 ret
= bdrv_pread(bs
->file
, offset
, &bitmaps_ext
, ext
.len
);
328 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
329 "Could not read ext header");
333 if (bitmaps_ext
.reserved32
!= 0) {
334 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
335 "Reserved field is not zero");
339 be32_to_cpus(&bitmaps_ext
.nb_bitmaps
);
340 be64_to_cpus(&bitmaps_ext
.bitmap_directory_size
);
341 be64_to_cpus(&bitmaps_ext
.bitmap_directory_offset
);
343 if (bitmaps_ext
.nb_bitmaps
> QCOW2_MAX_BITMAPS
) {
345 "bitmaps_ext: Image has %" PRIu32
" bitmaps, "
346 "exceeding the QEMU supported maximum of %d",
347 bitmaps_ext
.nb_bitmaps
, QCOW2_MAX_BITMAPS
);
351 if (bitmaps_ext
.nb_bitmaps
== 0) {
352 error_setg(errp
, "found bitmaps extension with zero bitmaps");
356 if (bitmaps_ext
.bitmap_directory_offset
& (s
->cluster_size
- 1)) {
357 error_setg(errp
, "bitmaps_ext: "
358 "invalid bitmap directory offset");
362 if (bitmaps_ext
.bitmap_directory_size
>
363 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
) {
364 error_setg(errp
, "bitmaps_ext: "
365 "bitmap directory size (%" PRIu64
") exceeds "
366 "the maximum supported size (%d)",
367 bitmaps_ext
.bitmap_directory_size
,
368 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
);
372 s
->nb_bitmaps
= bitmaps_ext
.nb_bitmaps
;
373 s
->bitmap_directory_offset
=
374 bitmaps_ext
.bitmap_directory_offset
;
375 s
->bitmap_directory_size
=
376 bitmaps_ext
.bitmap_directory_size
;
379 printf("Qcow2: Got bitmaps extension: "
380 "offset=%" PRIu64
" nb_bitmaps=%" PRIu32
"\n",
381 s
->bitmap_directory_offset
, s
->nb_bitmaps
);
386 /* unknown magic - save it in case we need to rewrite the header */
387 /* If you add a new feature, make sure to also update the fast
388 * path of qcow2_make_empty() to deal with it. */
390 Qcow2UnknownHeaderExtension
*uext
;
392 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
393 uext
->magic
= ext
.magic
;
395 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
397 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
399 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
400 "Could not read data");
407 offset
+= ((ext
.len
+ 7) & ~7);
413 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
415 BDRVQcow2State
*s
= bs
->opaque
;
416 Qcow2UnknownHeaderExtension
*uext
, *next
;
418 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
419 QLIST_REMOVE(uext
, next
);
424 static void report_unsupported_feature(Error
**errp
, Qcow2Feature
*table
,
427 char *features
= g_strdup("");
430 while (table
&& table
->name
[0] != '\0') {
431 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
432 if (mask
& (1ULL << table
->bit
)) {
434 features
= g_strdup_printf("%s%s%.46s", old
, *old
? ", " : "",
437 mask
&= ~(1ULL << table
->bit
);
445 features
= g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64
,
446 old
, *old
? ", " : "", mask
);
450 error_setg(errp
, "Unsupported qcow2 feature(s): %s", features
);
455 * Sets the dirty bit and flushes afterwards if necessary.
457 * The incompatible_features bit is only set if the image file header was
458 * updated successfully. Therefore it is not required to check the return
459 * value of this function.
461 int qcow2_mark_dirty(BlockDriverState
*bs
)
463 BDRVQcow2State
*s
= bs
->opaque
;
467 assert(s
->qcow_version
>= 3);
469 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
470 return 0; /* already dirty */
473 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
474 ret
= bdrv_pwrite(bs
->file
, offsetof(QCowHeader
, incompatible_features
),
479 ret
= bdrv_flush(bs
->file
->bs
);
484 /* Only treat image as dirty if the header was updated successfully */
485 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
490 * Clears the dirty bit and flushes before if necessary. Only call this
491 * function when there are no pending requests, it does not guard against
492 * concurrent requests dirtying the image.
494 static int qcow2_mark_clean(BlockDriverState
*bs
)
496 BDRVQcow2State
*s
= bs
->opaque
;
498 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
501 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
503 ret
= qcow2_flush_caches(bs
);
508 return qcow2_update_header(bs
);
514 * Marks the image as corrupt.
516 int qcow2_mark_corrupt(BlockDriverState
*bs
)
518 BDRVQcow2State
*s
= bs
->opaque
;
520 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
521 return qcow2_update_header(bs
);
525 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
526 * before if necessary.
528 int qcow2_mark_consistent(BlockDriverState
*bs
)
530 BDRVQcow2State
*s
= bs
->opaque
;
532 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
533 int ret
= qcow2_flush_caches(bs
);
538 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
539 return qcow2_update_header(bs
);
544 static int coroutine_fn
qcow2_co_check_locked(BlockDriverState
*bs
,
545 BdrvCheckResult
*result
,
548 int ret
= qcow2_check_refcounts(bs
, result
, fix
);
553 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
554 ret
= qcow2_mark_clean(bs
);
558 return qcow2_mark_consistent(bs
);
563 static int coroutine_fn
qcow2_co_check(BlockDriverState
*bs
,
564 BdrvCheckResult
*result
,
567 BDRVQcow2State
*s
= bs
->opaque
;
570 qemu_co_mutex_lock(&s
->lock
);
571 ret
= qcow2_co_check_locked(bs
, result
, fix
);
572 qemu_co_mutex_unlock(&s
->lock
);
576 int qcow2_validate_table(BlockDriverState
*bs
, uint64_t offset
,
577 uint64_t entries
, size_t entry_len
,
578 int64_t max_size_bytes
, const char *table_name
,
581 BDRVQcow2State
*s
= bs
->opaque
;
583 if (entries
> max_size_bytes
/ entry_len
) {
584 error_setg(errp
, "%s too large", table_name
);
588 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
589 * because values will be passed to qemu functions taking int64_t. */
590 if ((INT64_MAX
- entries
* entry_len
< offset
) ||
591 (offset_into_cluster(s
, offset
) != 0)) {
592 error_setg(errp
, "%s offset invalid", table_name
);
599 static QemuOptsList qcow2_runtime_opts
= {
601 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
604 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
605 .type
= QEMU_OPT_BOOL
,
606 .help
= "Postpone refcount updates",
609 .name
= QCOW2_OPT_DISCARD_REQUEST
,
610 .type
= QEMU_OPT_BOOL
,
611 .help
= "Pass guest discard requests to the layer below",
614 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
615 .type
= QEMU_OPT_BOOL
,
616 .help
= "Generate discard requests when snapshot related space "
620 .name
= QCOW2_OPT_DISCARD_OTHER
,
621 .type
= QEMU_OPT_BOOL
,
622 .help
= "Generate discard requests when other clusters are freed",
625 .name
= QCOW2_OPT_OVERLAP
,
626 .type
= QEMU_OPT_STRING
,
627 .help
= "Selects which overlap checks to perform from a range of "
628 "templates (none, constant, cached, all)",
631 .name
= QCOW2_OPT_OVERLAP_TEMPLATE
,
632 .type
= QEMU_OPT_STRING
,
633 .help
= "Selects which overlap checks to perform from a range of "
634 "templates (none, constant, cached, all)",
637 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
638 .type
= QEMU_OPT_BOOL
,
639 .help
= "Check for unintended writes into the main qcow2 header",
642 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
643 .type
= QEMU_OPT_BOOL
,
644 .help
= "Check for unintended writes into the active L1 table",
647 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
648 .type
= QEMU_OPT_BOOL
,
649 .help
= "Check for unintended writes into an active L2 table",
652 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
653 .type
= QEMU_OPT_BOOL
,
654 .help
= "Check for unintended writes into the refcount table",
657 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
658 .type
= QEMU_OPT_BOOL
,
659 .help
= "Check for unintended writes into a refcount block",
662 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
663 .type
= QEMU_OPT_BOOL
,
664 .help
= "Check for unintended writes into the snapshot table",
667 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
668 .type
= QEMU_OPT_BOOL
,
669 .help
= "Check for unintended writes into an inactive L1 table",
672 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
673 .type
= QEMU_OPT_BOOL
,
674 .help
= "Check for unintended writes into an inactive L2 table",
677 .name
= QCOW2_OPT_CACHE_SIZE
,
678 .type
= QEMU_OPT_SIZE
,
679 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
683 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
684 .type
= QEMU_OPT_SIZE
,
685 .help
= "Maximum L2 table cache size",
688 .name
= QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
689 .type
= QEMU_OPT_SIZE
,
690 .help
= "Size of each entry in the L2 cache",
693 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
694 .type
= QEMU_OPT_SIZE
,
695 .help
= "Maximum refcount block cache size",
698 .name
= QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
699 .type
= QEMU_OPT_NUMBER
,
700 .help
= "Clean unused cache entries after this time (in seconds)",
702 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
703 "ID of secret providing qcow2 AES key or LUKS passphrase"),
704 { /* end of list */ }
708 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
709 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
710 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
711 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
712 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
713 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
714 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
715 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
716 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
719 static void cache_clean_timer_cb(void *opaque
)
721 BlockDriverState
*bs
= opaque
;
722 BDRVQcow2State
*s
= bs
->opaque
;
723 qcow2_cache_clean_unused(s
->l2_table_cache
);
724 qcow2_cache_clean_unused(s
->refcount_block_cache
);
725 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
726 (int64_t) s
->cache_clean_interval
* 1000);
729 static void cache_clean_timer_init(BlockDriverState
*bs
, AioContext
*context
)
731 BDRVQcow2State
*s
= bs
->opaque
;
732 if (s
->cache_clean_interval
> 0) {
733 s
->cache_clean_timer
= aio_timer_new(context
, QEMU_CLOCK_VIRTUAL
,
734 SCALE_MS
, cache_clean_timer_cb
,
736 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
737 (int64_t) s
->cache_clean_interval
* 1000);
741 static void cache_clean_timer_del(BlockDriverState
*bs
)
743 BDRVQcow2State
*s
= bs
->opaque
;
744 if (s
->cache_clean_timer
) {
745 timer_del(s
->cache_clean_timer
);
746 timer_free(s
->cache_clean_timer
);
747 s
->cache_clean_timer
= NULL
;
751 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
753 cache_clean_timer_del(bs
);
756 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
757 AioContext
*new_context
)
759 cache_clean_timer_init(bs
, new_context
);
762 static void read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
763 uint64_t *l2_cache_size
,
764 uint64_t *l2_cache_entry_size
,
765 uint64_t *refcount_cache_size
, Error
**errp
)
767 BDRVQcow2State
*s
= bs
->opaque
;
768 uint64_t combined_cache_size
;
769 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
771 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
772 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
773 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
775 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
776 *l2_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
, 0);
777 *refcount_cache_size
= qemu_opt_get_size(opts
,
778 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
780 *l2_cache_entry_size
= qemu_opt_get_size(
781 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
783 if (combined_cache_size_set
) {
784 if (l2_cache_size_set
&& refcount_cache_size_set
) {
785 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
786 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
789 } else if (*l2_cache_size
> combined_cache_size
) {
790 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
791 QCOW2_OPT_CACHE_SIZE
);
793 } else if (*refcount_cache_size
> combined_cache_size
) {
794 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
795 QCOW2_OPT_CACHE_SIZE
);
799 if (l2_cache_size_set
) {
800 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
801 } else if (refcount_cache_size_set
) {
802 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
804 *refcount_cache_size
= combined_cache_size
805 / (DEFAULT_L2_REFCOUNT_SIZE_RATIO
+ 1);
806 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
809 if (!l2_cache_size_set
&& !refcount_cache_size_set
) {
810 *l2_cache_size
= MAX(DEFAULT_L2_CACHE_BYTE_SIZE
,
811 (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
813 *refcount_cache_size
= *l2_cache_size
814 / DEFAULT_L2_REFCOUNT_SIZE_RATIO
;
815 } else if (!l2_cache_size_set
) {
816 *l2_cache_size
= *refcount_cache_size
817 * DEFAULT_L2_REFCOUNT_SIZE_RATIO
;
818 } else if (!refcount_cache_size_set
) {
819 *refcount_cache_size
= *l2_cache_size
820 / DEFAULT_L2_REFCOUNT_SIZE_RATIO
;
824 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
825 *l2_cache_entry_size
> s
->cluster_size
||
826 !is_power_of_2(*l2_cache_entry_size
)) {
827 error_setg(errp
, "L2 cache entry size must be a power of two "
828 "between %d and the cluster size (%d)",
829 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
834 typedef struct Qcow2ReopenState
{
835 Qcow2Cache
*l2_table_cache
;
836 Qcow2Cache
*refcount_block_cache
;
837 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
838 bool use_lazy_refcounts
;
840 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
841 uint64_t cache_clean_interval
;
842 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
845 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
847 QDict
*options
, int flags
,
850 BDRVQcow2State
*s
= bs
->opaque
;
851 QemuOpts
*opts
= NULL
;
852 const char *opt_overlap_check
, *opt_overlap_check_template
;
853 int overlap_check_template
= 0;
854 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
856 const char *encryptfmt
;
857 QDict
*encryptopts
= NULL
;
858 Error
*local_err
= NULL
;
861 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
862 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
864 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
865 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
867 error_propagate(errp
, local_err
);
872 /* get L2 table/refcount block cache size from command line options */
873 read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
874 &refcount_cache_size
, &local_err
);
876 error_propagate(errp
, local_err
);
881 l2_cache_size
/= l2_cache_entry_size
;
882 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
883 l2_cache_size
= MIN_L2_CACHE_SIZE
;
885 if (l2_cache_size
> INT_MAX
) {
886 error_setg(errp
, "L2 cache size too big");
891 refcount_cache_size
/= s
->cluster_size
;
892 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
893 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
895 if (refcount_cache_size
> INT_MAX
) {
896 error_setg(errp
, "Refcount cache size too big");
901 /* alloc new L2 table/refcount block cache, flush old one */
902 if (s
->l2_table_cache
) {
903 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
905 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
910 if (s
->refcount_block_cache
) {
911 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
913 error_setg_errno(errp
, -ret
,
914 "Failed to flush the refcount block cache");
919 r
->l2_slice_size
= l2_cache_entry_size
/ sizeof(uint64_t);
920 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
921 l2_cache_entry_size
);
922 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
924 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
925 error_setg(errp
, "Could not allocate metadata caches");
930 /* New interval for cache cleanup timer */
931 r
->cache_clean_interval
=
932 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
933 s
->cache_clean_interval
);
935 if (r
->cache_clean_interval
!= 0) {
936 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
937 " not supported on this host");
942 if (r
->cache_clean_interval
> UINT_MAX
) {
943 error_setg(errp
, "Cache clean interval too big");
948 /* lazy-refcounts; flush if going from enabled to disabled */
949 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
950 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
951 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
952 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
953 "qemu 1.1 compatibility level");
958 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
959 ret
= qcow2_mark_clean(bs
);
961 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
966 /* Overlap check options */
967 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
968 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
969 if (opt_overlap_check_template
&& opt_overlap_check
&&
970 strcmp(opt_overlap_check_template
, opt_overlap_check
))
972 error_setg(errp
, "Conflicting values for qcow2 options '"
973 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
974 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
978 if (!opt_overlap_check
) {
979 opt_overlap_check
= opt_overlap_check_template
?: "cached";
982 if (!strcmp(opt_overlap_check
, "none")) {
983 overlap_check_template
= 0;
984 } else if (!strcmp(opt_overlap_check
, "constant")) {
985 overlap_check_template
= QCOW2_OL_CONSTANT
;
986 } else if (!strcmp(opt_overlap_check
, "cached")) {
987 overlap_check_template
= QCOW2_OL_CACHED
;
988 } else if (!strcmp(opt_overlap_check
, "all")) {
989 overlap_check_template
= QCOW2_OL_ALL
;
991 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
992 "'overlap-check'. Allowed are any of the following: "
993 "none, constant, cached, all", opt_overlap_check
);
998 r
->overlap_check
= 0;
999 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1000 /* overlap-check defines a template bitmask, but every flag may be
1001 * overwritten through the associated boolean option */
1003 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1004 overlap_check_template
& (1 << i
)) << i
;
1007 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1008 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1009 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1010 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1011 flags
& BDRV_O_UNMAP
);
1012 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1013 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1014 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1015 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1017 switch (s
->crypt_method_header
) {
1018 case QCOW_CRYPT_NONE
:
1020 error_setg(errp
, "No encryption in image header, but options "
1021 "specified format '%s'", encryptfmt
);
1027 case QCOW_CRYPT_AES
:
1028 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1030 "Header reported 'aes' encryption format but "
1031 "options specify '%s'", encryptfmt
);
1035 qdict_del(encryptopts
, "format");
1036 r
->crypto_opts
= block_crypto_open_opts_init(
1037 Q_CRYPTO_BLOCK_FORMAT_QCOW
, encryptopts
, errp
);
1040 case QCOW_CRYPT_LUKS
:
1041 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1043 "Header reported 'luks' encryption format but "
1044 "options specify '%s'", encryptfmt
);
1048 qdict_del(encryptopts
, "format");
1049 r
->crypto_opts
= block_crypto_open_opts_init(
1050 Q_CRYPTO_BLOCK_FORMAT_LUKS
, encryptopts
, errp
);
1054 error_setg(errp
, "Unsupported encryption method %d",
1055 s
->crypt_method_header
);
1058 if (s
->crypt_method_header
!= QCOW_CRYPT_NONE
&& !r
->crypto_opts
) {
1065 QDECREF(encryptopts
);
1066 qemu_opts_del(opts
);
1071 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1072 Qcow2ReopenState
*r
)
1074 BDRVQcow2State
*s
= bs
->opaque
;
1077 if (s
->l2_table_cache
) {
1078 qcow2_cache_destroy(s
->l2_table_cache
);
1080 if (s
->refcount_block_cache
) {
1081 qcow2_cache_destroy(s
->refcount_block_cache
);
1083 s
->l2_table_cache
= r
->l2_table_cache
;
1084 s
->refcount_block_cache
= r
->refcount_block_cache
;
1085 s
->l2_slice_size
= r
->l2_slice_size
;
1087 s
->overlap_check
= r
->overlap_check
;
1088 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1090 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1091 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1094 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1095 cache_clean_timer_del(bs
);
1096 s
->cache_clean_interval
= r
->cache_clean_interval
;
1097 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1100 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1101 s
->crypto_opts
= r
->crypto_opts
;
1104 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1105 Qcow2ReopenState
*r
)
1107 if (r
->l2_table_cache
) {
1108 qcow2_cache_destroy(r
->l2_table_cache
);
1110 if (r
->refcount_block_cache
) {
1111 qcow2_cache_destroy(r
->refcount_block_cache
);
1113 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1116 static int qcow2_update_options(BlockDriverState
*bs
, QDict
*options
,
1117 int flags
, Error
**errp
)
1119 Qcow2ReopenState r
= {};
1122 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1124 qcow2_update_options_commit(bs
, &r
);
1126 qcow2_update_options_abort(bs
, &r
);
1132 /* Called with s->lock held. */
1133 static int coroutine_fn
qcow2_do_open(BlockDriverState
*bs
, QDict
*options
,
1134 int flags
, Error
**errp
)
1136 BDRVQcow2State
*s
= bs
->opaque
;
1137 unsigned int len
, i
;
1140 Error
*local_err
= NULL
;
1142 uint64_t l1_vm_state_index
;
1143 bool update_header
= false;
1145 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
1147 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1150 be32_to_cpus(&header
.magic
);
1151 be32_to_cpus(&header
.version
);
1152 be64_to_cpus(&header
.backing_file_offset
);
1153 be32_to_cpus(&header
.backing_file_size
);
1154 be64_to_cpus(&header
.size
);
1155 be32_to_cpus(&header
.cluster_bits
);
1156 be32_to_cpus(&header
.crypt_method
);
1157 be64_to_cpus(&header
.l1_table_offset
);
1158 be32_to_cpus(&header
.l1_size
);
1159 be64_to_cpus(&header
.refcount_table_offset
);
1160 be32_to_cpus(&header
.refcount_table_clusters
);
1161 be64_to_cpus(&header
.snapshots_offset
);
1162 be32_to_cpus(&header
.nb_snapshots
);
1164 if (header
.magic
!= QCOW_MAGIC
) {
1165 error_setg(errp
, "Image is not in qcow2 format");
1169 if (header
.version
< 2 || header
.version
> 3) {
1170 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1175 s
->qcow_version
= header
.version
;
1177 /* Initialise cluster size */
1178 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1179 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1180 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1181 header
.cluster_bits
);
1186 s
->cluster_bits
= header
.cluster_bits
;
1187 s
->cluster_size
= 1 << s
->cluster_bits
;
1188 s
->cluster_sectors
= 1 << (s
->cluster_bits
- BDRV_SECTOR_BITS
);
1190 /* Initialise version 3 header fields */
1191 if (header
.version
== 2) {
1192 header
.incompatible_features
= 0;
1193 header
.compatible_features
= 0;
1194 header
.autoclear_features
= 0;
1195 header
.refcount_order
= 4;
1196 header
.header_length
= 72;
1198 be64_to_cpus(&header
.incompatible_features
);
1199 be64_to_cpus(&header
.compatible_features
);
1200 be64_to_cpus(&header
.autoclear_features
);
1201 be32_to_cpus(&header
.refcount_order
);
1202 be32_to_cpus(&header
.header_length
);
1204 if (header
.header_length
< 104) {
1205 error_setg(errp
, "qcow2 header too short");
1211 if (header
.header_length
> s
->cluster_size
) {
1212 error_setg(errp
, "qcow2 header exceeds cluster size");
1217 if (header
.header_length
> sizeof(header
)) {
1218 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1219 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1220 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
1221 s
->unknown_header_fields_size
);
1223 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1229 if (header
.backing_file_offset
> s
->cluster_size
) {
1230 error_setg(errp
, "Invalid backing file offset");
1235 if (header
.backing_file_offset
) {
1236 ext_end
= header
.backing_file_offset
;
1238 ext_end
= 1 << header
.cluster_bits
;
1241 /* Handle feature bits */
1242 s
->incompatible_features
= header
.incompatible_features
;
1243 s
->compatible_features
= header
.compatible_features
;
1244 s
->autoclear_features
= header
.autoclear_features
;
1246 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1247 void *feature_table
= NULL
;
1248 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1249 &feature_table
, flags
, NULL
, NULL
);
1250 report_unsupported_feature(errp
, feature_table
,
1251 s
->incompatible_features
&
1252 ~QCOW2_INCOMPAT_MASK
);
1254 g_free(feature_table
);
1258 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1259 /* Corrupt images may not be written to unless they are being repaired
1261 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1262 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1269 /* Check support for various header values */
1270 if (header
.refcount_order
> 6) {
1271 error_setg(errp
, "Reference count entry width too large; may not "
1276 s
->refcount_order
= header
.refcount_order
;
1277 s
->refcount_bits
= 1 << s
->refcount_order
;
1278 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1279 s
->refcount_max
+= s
->refcount_max
- 1;
1281 s
->crypt_method_header
= header
.crypt_method
;
1282 if (s
->crypt_method_header
) {
1283 if (bdrv_uses_whitelist() &&
1284 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1286 "Use of AES-CBC encrypted qcow2 images is no longer "
1287 "supported in system emulators");
1288 error_append_hint(errp
,
1289 "You can use 'qemu-img convert' to convert your "
1290 "image to an alternative supported format, such "
1291 "as unencrypted qcow2, or raw with the LUKS "
1292 "format instead.\n");
1297 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1298 s
->crypt_physical_offset
= false;
1300 /* Assuming LUKS and any future crypt methods we
1301 * add will all use physical offsets, due to the
1302 * fact that the alternative is insecure... */
1303 s
->crypt_physical_offset
= true;
1306 bs
->encrypted
= true;
1309 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
1310 s
->l2_size
= 1 << s
->l2_bits
;
1311 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1312 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1313 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1314 bs
->total_sectors
= header
.size
/ 512;
1315 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1316 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1317 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1319 s
->refcount_table_offset
= header
.refcount_table_offset
;
1320 s
->refcount_table_size
=
1321 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1323 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1324 error_setg(errp
, "Image does not contain a reference count table");
1329 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1330 header
.refcount_table_clusters
,
1331 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1332 "Reference count table", errp
);
1337 /* The total size in bytes of the snapshot table is checked in
1338 * qcow2_read_snapshots() because the size of each snapshot is
1339 * variable and we don't know it yet.
1340 * Here we only check the offset and number of snapshots. */
1341 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1342 header
.nb_snapshots
,
1343 sizeof(QCowSnapshotHeader
),
1344 sizeof(QCowSnapshotHeader
) * QCOW_MAX_SNAPSHOTS
,
1345 "Snapshot table", errp
);
1350 /* read the level 1 table */
1351 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1352 header
.l1_size
, sizeof(uint64_t),
1353 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1357 s
->l1_size
= header
.l1_size
;
1358 s
->l1_table_offset
= header
.l1_table_offset
;
1360 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1361 if (l1_vm_state_index
> INT_MAX
) {
1362 error_setg(errp
, "Image is too big");
1366 s
->l1_vm_state_index
= l1_vm_state_index
;
1368 /* the L1 table must contain at least enough entries to put
1369 header.size bytes */
1370 if (s
->l1_size
< s
->l1_vm_state_index
) {
1371 error_setg(errp
, "L1 table is too small");
1376 if (s
->l1_size
> 0) {
1377 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
,
1378 ROUND_UP(s
->l1_size
* sizeof(uint64_t), 512));
1379 if (s
->l1_table
== NULL
) {
1380 error_setg(errp
, "Could not allocate L1 table");
1384 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1385 s
->l1_size
* sizeof(uint64_t));
1387 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1390 for(i
= 0;i
< s
->l1_size
; i
++) {
1391 be64_to_cpus(&s
->l1_table
[i
]);
1395 /* Parse driver-specific options */
1396 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1401 s
->cluster_cache_offset
= -1;
1404 ret
= qcow2_refcount_init(bs
);
1406 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1410 QLIST_INIT(&s
->cluster_allocs
);
1411 QTAILQ_INIT(&s
->discards
);
1413 /* read qcow2 extensions */
1414 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1415 flags
, &update_header
, &local_err
)) {
1416 error_propagate(errp
, local_err
);
1421 /* qcow2_read_extension may have set up the crypto context
1422 * if the crypt method needs a header region, some methods
1423 * don't need header extensions, so must check here
1425 if (s
->crypt_method_header
&& !s
->crypto
) {
1426 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1427 unsigned int cflags
= 0;
1428 if (flags
& BDRV_O_NO_IO
) {
1429 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1431 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1432 NULL
, NULL
, cflags
, errp
);
1437 } else if (!(flags
& BDRV_O_NO_IO
)) {
1438 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1439 s
->crypt_method_header
);
1445 /* read the backing file name */
1446 if (header
.backing_file_offset
!= 0) {
1447 len
= header
.backing_file_size
;
1448 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1449 len
>= sizeof(bs
->backing_file
)) {
1450 error_setg(errp
, "Backing file name too long");
1454 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
1455 bs
->backing_file
, len
);
1457 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1460 bs
->backing_file
[len
] = '\0';
1461 s
->image_backing_file
= g_strdup(bs
->backing_file
);
1464 /* Internal snapshots */
1465 s
->snapshots_offset
= header
.snapshots_offset
;
1466 s
->nb_snapshots
= header
.nb_snapshots
;
1468 ret
= qcow2_read_snapshots(bs
);
1470 error_setg_errno(errp
, -ret
, "Could not read snapshots");
1474 /* Clear unknown autoclear feature bits */
1475 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1477 update_header
&& !bs
->read_only
&& !(flags
& BDRV_O_INACTIVE
);
1478 if (update_header
) {
1479 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1482 if (qcow2_load_dirty_bitmaps(bs
, &local_err
)) {
1483 update_header
= false;
1485 if (local_err
!= NULL
) {
1486 error_propagate(errp
, local_err
);
1491 if (update_header
) {
1492 ret
= qcow2_update_header(bs
);
1494 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1499 bs
->supported_zero_flags
= header
.version
>= 3 ? BDRV_REQ_MAY_UNMAP
: 0;
1501 /* Repair image if dirty */
1502 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INACTIVE
)) && !bs
->read_only
&&
1503 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1504 BdrvCheckResult result
= {0};
1506 ret
= qcow2_co_check_locked(bs
, &result
,
1507 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1508 if (ret
< 0 || result
.check_errors
) {
1512 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1519 BdrvCheckResult result
= {0};
1520 qcow2_check_refcounts(bs
, &result
, 0);
1526 g_free(s
->unknown_header_fields
);
1527 cleanup_unknown_header_ext(bs
);
1528 qcow2_free_snapshots(bs
);
1529 qcow2_refcount_close(bs
);
1530 qemu_vfree(s
->l1_table
);
1531 /* else pre-write overlap checks in cache_destroy may crash */
1533 cache_clean_timer_del(bs
);
1534 if (s
->l2_table_cache
) {
1535 qcow2_cache_destroy(s
->l2_table_cache
);
1537 if (s
->refcount_block_cache
) {
1538 qcow2_cache_destroy(s
->refcount_block_cache
);
1540 qcrypto_block_free(s
->crypto
);
1541 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1545 typedef struct QCow2OpenCo
{
1546 BlockDriverState
*bs
;
1553 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1555 QCow2OpenCo
*qoc
= opaque
;
1556 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1558 qemu_co_mutex_lock(&s
->lock
);
1559 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, qoc
->errp
);
1560 qemu_co_mutex_unlock(&s
->lock
);
1563 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1566 BDRVQcow2State
*s
= bs
->opaque
;
1575 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_file
,
1581 /* Initialise locks */
1582 qemu_co_mutex_init(&s
->lock
);
1584 if (qemu_in_coroutine()) {
1585 /* From bdrv_co_create. */
1586 qcow2_open_entry(&qoc
);
1588 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1589 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1594 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1596 BDRVQcow2State
*s
= bs
->opaque
;
1598 if (bs
->encrypted
) {
1599 /* Encryption works on a sector granularity */
1600 bs
->bl
.request_alignment
= BDRV_SECTOR_SIZE
;
1602 bs
->bl
.pwrite_zeroes_alignment
= s
->cluster_size
;
1603 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1606 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1607 BlockReopenQueue
*queue
, Error
**errp
)
1609 Qcow2ReopenState
*r
;
1612 r
= g_new0(Qcow2ReopenState
, 1);
1615 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1616 state
->flags
, errp
);
1621 /* We need to write out any unwritten data if we reopen read-only. */
1622 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1623 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1628 ret
= bdrv_flush(state
->bs
);
1633 ret
= qcow2_mark_clean(state
->bs
);
1642 qcow2_update_options_abort(state
->bs
, r
);
1647 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1649 qcow2_update_options_commit(state
->bs
, state
->opaque
);
1650 g_free(state
->opaque
);
1653 static void qcow2_reopen_abort(BDRVReopenState
*state
)
1655 qcow2_update_options_abort(state
->bs
, state
->opaque
);
1656 g_free(state
->opaque
);
1659 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
1661 bool has_new_overlap_template
=
1662 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
1663 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1664 bool has_new_total_cache_size
=
1665 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
1666 bool has_all_cache_options
;
1668 /* New overlap template overrides all old overlap options */
1669 if (has_new_overlap_template
) {
1670 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
1671 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1672 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
1673 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
1674 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
1675 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
1676 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
1677 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
1678 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
1679 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
1682 /* New total cache size overrides all old options */
1683 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
1684 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
1685 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1688 qdict_join(options
, old_options
, false);
1691 * If after merging all cache size options are set, an old total size is
1692 * overwritten. Do keep all options, however, if all three are new. The
1693 * resulting error message is what we want to happen.
1695 has_all_cache_options
=
1696 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
1697 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
1698 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1700 if (has_all_cache_options
&& !has_new_total_cache_size
) {
1701 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
1705 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
1707 int64_t offset
, int64_t count
,
1708 int64_t *pnum
, int64_t *map
,
1709 BlockDriverState
**file
)
1711 BDRVQcow2State
*s
= bs
->opaque
;
1712 uint64_t cluster_offset
;
1713 int index_in_cluster
, ret
;
1717 bytes
= MIN(INT_MAX
, count
);
1718 qemu_co_mutex_lock(&s
->lock
);
1719 ret
= qcow2_get_cluster_offset(bs
, offset
, &bytes
, &cluster_offset
);
1720 qemu_co_mutex_unlock(&s
->lock
);
1727 if (cluster_offset
!= 0 && ret
!= QCOW2_CLUSTER_COMPRESSED
&&
1729 index_in_cluster
= offset
& (s
->cluster_size
- 1);
1730 *map
= cluster_offset
| index_in_cluster
;
1731 *file
= bs
->file
->bs
;
1732 status
|= BDRV_BLOCK_OFFSET_VALID
;
1734 if (ret
== QCOW2_CLUSTER_ZERO_PLAIN
|| ret
== QCOW2_CLUSTER_ZERO_ALLOC
) {
1735 status
|= BDRV_BLOCK_ZERO
;
1736 } else if (ret
!= QCOW2_CLUSTER_UNALLOCATED
) {
1737 status
|= BDRV_BLOCK_DATA
;
1742 static coroutine_fn
int qcow2_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1743 uint64_t bytes
, QEMUIOVector
*qiov
,
1746 BDRVQcow2State
*s
= bs
->opaque
;
1747 int offset_in_cluster
;
1749 unsigned int cur_bytes
; /* number of bytes in current iteration */
1750 uint64_t cluster_offset
= 0;
1751 uint64_t bytes_done
= 0;
1752 QEMUIOVector hd_qiov
;
1753 uint8_t *cluster_data
= NULL
;
1755 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1757 qemu_co_mutex_lock(&s
->lock
);
1759 while (bytes
!= 0) {
1761 /* prepare next request */
1762 cur_bytes
= MIN(bytes
, INT_MAX
);
1764 cur_bytes
= MIN(cur_bytes
,
1765 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1768 ret
= qcow2_get_cluster_offset(bs
, offset
, &cur_bytes
, &cluster_offset
);
1773 offset_in_cluster
= offset_into_cluster(s
, offset
);
1775 qemu_iovec_reset(&hd_qiov
);
1776 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
1779 case QCOW2_CLUSTER_UNALLOCATED
:
1782 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
1783 qemu_co_mutex_unlock(&s
->lock
);
1784 ret
= bdrv_co_preadv(bs
->backing
, offset
, cur_bytes
,
1786 qemu_co_mutex_lock(&s
->lock
);
1791 /* Note: in this case, no need to wait */
1792 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1796 case QCOW2_CLUSTER_ZERO_PLAIN
:
1797 case QCOW2_CLUSTER_ZERO_ALLOC
:
1798 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1801 case QCOW2_CLUSTER_COMPRESSED
:
1802 /* add AIO support for compressed blocks ? */
1803 ret
= qcow2_decompress_cluster(bs
, cluster_offset
);
1808 qemu_iovec_from_buf(&hd_qiov
, 0,
1809 s
->cluster_cache
+ offset_in_cluster
,
1813 case QCOW2_CLUSTER_NORMAL
:
1814 if ((cluster_offset
& 511) != 0) {
1819 if (bs
->encrypted
) {
1823 * For encrypted images, read everything into a temporary
1824 * contiguous buffer on which the AES functions can work.
1826 if (!cluster_data
) {
1828 qemu_try_blockalign(bs
->file
->bs
,
1829 QCOW_MAX_CRYPT_CLUSTERS
1831 if (cluster_data
== NULL
) {
1837 assert(cur_bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1838 qemu_iovec_reset(&hd_qiov
);
1839 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
1842 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1843 qemu_co_mutex_unlock(&s
->lock
);
1844 ret
= bdrv_co_preadv(bs
->file
,
1845 cluster_offset
+ offset_in_cluster
,
1846 cur_bytes
, &hd_qiov
, 0);
1847 qemu_co_mutex_lock(&s
->lock
);
1851 if (bs
->encrypted
) {
1853 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1854 assert((cur_bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1855 if (qcrypto_block_decrypt(s
->crypto
,
1856 (s
->crypt_physical_offset
?
1857 cluster_offset
+ offset_in_cluster
:
1865 qemu_iovec_from_buf(qiov
, bytes_done
, cluster_data
, cur_bytes
);
1870 g_assert_not_reached();
1876 offset
+= cur_bytes
;
1877 bytes_done
+= cur_bytes
;
1882 qemu_co_mutex_unlock(&s
->lock
);
1884 qemu_iovec_destroy(&hd_qiov
);
1885 qemu_vfree(cluster_data
);
1890 /* Check if it's possible to merge a write request with the writing of
1891 * the data from the COW regions */
1892 static bool merge_cow(uint64_t offset
, unsigned bytes
,
1893 QEMUIOVector
*hd_qiov
, QCowL2Meta
*l2meta
)
1897 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
1898 /* If both COW regions are empty then there's nothing to merge */
1899 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
1903 /* The data (middle) region must be immediately after the
1905 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
1909 /* The end region must be immediately after the data (middle)
1911 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
1915 /* Make sure that adding both COW regions to the QEMUIOVector
1916 * does not exceed IOV_MAX */
1917 if (hd_qiov
->niov
> IOV_MAX
- 2) {
1921 m
->data_qiov
= hd_qiov
;
1928 static coroutine_fn
int qcow2_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1929 uint64_t bytes
, QEMUIOVector
*qiov
,
1932 BDRVQcow2State
*s
= bs
->opaque
;
1933 int offset_in_cluster
;
1935 unsigned int cur_bytes
; /* number of sectors in current iteration */
1936 uint64_t cluster_offset
;
1937 QEMUIOVector hd_qiov
;
1938 uint64_t bytes_done
= 0;
1939 uint8_t *cluster_data
= NULL
;
1940 QCowL2Meta
*l2meta
= NULL
;
1942 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
1944 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1946 s
->cluster_cache_offset
= -1; /* disable compressed cache */
1948 qemu_co_mutex_lock(&s
->lock
);
1950 while (bytes
!= 0) {
1954 trace_qcow2_writev_start_part(qemu_coroutine_self());
1955 offset_in_cluster
= offset_into_cluster(s
, offset
);
1956 cur_bytes
= MIN(bytes
, INT_MAX
);
1957 if (bs
->encrypted
) {
1958 cur_bytes
= MIN(cur_bytes
,
1959 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
1960 - offset_in_cluster
);
1963 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
1964 &cluster_offset
, &l2meta
);
1969 assert((cluster_offset
& 511) == 0);
1971 qemu_iovec_reset(&hd_qiov
);
1972 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
1974 if (bs
->encrypted
) {
1976 if (!cluster_data
) {
1977 cluster_data
= qemu_try_blockalign(bs
->file
->bs
,
1978 QCOW_MAX_CRYPT_CLUSTERS
1980 if (cluster_data
== NULL
) {
1986 assert(hd_qiov
.size
<=
1987 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1988 qemu_iovec_to_buf(&hd_qiov
, 0, cluster_data
, hd_qiov
.size
);
1990 if (qcrypto_block_encrypt(s
->crypto
,
1991 (s
->crypt_physical_offset
?
1992 cluster_offset
+ offset_in_cluster
:
1995 cur_bytes
, NULL
) < 0) {
2000 qemu_iovec_reset(&hd_qiov
);
2001 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
2004 ret
= qcow2_pre_write_overlap_check(bs
, 0,
2005 cluster_offset
+ offset_in_cluster
, cur_bytes
);
2010 /* If we need to do COW, check if it's possible to merge the
2011 * writing of the guest data together with that of the COW regions.
2012 * If it's not possible (or not necessary) then write the
2013 * guest data now. */
2014 if (!merge_cow(offset
, cur_bytes
, &hd_qiov
, l2meta
)) {
2015 qemu_co_mutex_unlock(&s
->lock
);
2016 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2017 trace_qcow2_writev_data(qemu_coroutine_self(),
2018 cluster_offset
+ offset_in_cluster
);
2019 ret
= bdrv_co_pwritev(bs
->file
,
2020 cluster_offset
+ offset_in_cluster
,
2021 cur_bytes
, &hd_qiov
, 0);
2022 qemu_co_mutex_lock(&s
->lock
);
2028 while (l2meta
!= NULL
) {
2031 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
2036 /* Take the request off the list of running requests */
2037 if (l2meta
->nb_clusters
!= 0) {
2038 QLIST_REMOVE(l2meta
, next_in_flight
);
2041 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2043 next
= l2meta
->next
;
2049 offset
+= cur_bytes
;
2050 bytes_done
+= cur_bytes
;
2051 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2056 while (l2meta
!= NULL
) {
2059 if (l2meta
->nb_clusters
!= 0) {
2060 QLIST_REMOVE(l2meta
, next_in_flight
);
2062 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2064 next
= l2meta
->next
;
2069 qemu_co_mutex_unlock(&s
->lock
);
2071 qemu_iovec_destroy(&hd_qiov
);
2072 qemu_vfree(cluster_data
);
2073 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2078 static int qcow2_inactivate(BlockDriverState
*bs
)
2080 BDRVQcow2State
*s
= bs
->opaque
;
2081 int ret
, result
= 0;
2082 Error
*local_err
= NULL
;
2084 qcow2_store_persistent_dirty_bitmaps(bs
, &local_err
);
2085 if (local_err
!= NULL
) {
2087 error_report_err(local_err
);
2088 error_report("Persistent bitmaps are lost for node '%s'",
2089 bdrv_get_device_or_node_name(bs
));
2092 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2095 error_report("Failed to flush the L2 table cache: %s",
2099 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2102 error_report("Failed to flush the refcount block cache: %s",
2107 qcow2_mark_clean(bs
);
2113 static void qcow2_close(BlockDriverState
*bs
)
2115 BDRVQcow2State
*s
= bs
->opaque
;
2116 qemu_vfree(s
->l1_table
);
2117 /* else pre-write overlap checks in cache_destroy may crash */
2120 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2121 qcow2_inactivate(bs
);
2124 cache_clean_timer_del(bs
);
2125 qcow2_cache_destroy(s
->l2_table_cache
);
2126 qcow2_cache_destroy(s
->refcount_block_cache
);
2128 qcrypto_block_free(s
->crypto
);
2131 g_free(s
->unknown_header_fields
);
2132 cleanup_unknown_header_ext(bs
);
2134 g_free(s
->image_backing_file
);
2135 g_free(s
->image_backing_format
);
2137 g_free(s
->cluster_cache
);
2138 qemu_vfree(s
->cluster_data
);
2139 qcow2_refcount_close(bs
);
2140 qcow2_free_snapshots(bs
);
2143 static void coroutine_fn
qcow2_co_invalidate_cache(BlockDriverState
*bs
,
2146 BDRVQcow2State
*s
= bs
->opaque
;
2147 int flags
= s
->flags
;
2148 QCryptoBlock
*crypto
= NULL
;
2150 Error
*local_err
= NULL
;
2154 * Backing files are read-only which makes all of their metadata immutable,
2155 * that means we don't have to worry about reopening them here.
2163 memset(s
, 0, sizeof(BDRVQcow2State
));
2164 options
= qdict_clone_shallow(bs
->options
);
2166 flags
&= ~BDRV_O_INACTIVE
;
2167 qemu_co_mutex_lock(&s
->lock
);
2168 ret
= qcow2_do_open(bs
, options
, flags
, &local_err
);
2169 qemu_co_mutex_unlock(&s
->lock
);
2172 error_propagate(errp
, local_err
);
2173 error_prepend(errp
, "Could not reopen qcow2 layer: ");
2176 } else if (ret
< 0) {
2177 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
2185 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2186 size_t len
, size_t buflen
)
2188 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2189 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2191 if (buflen
< ext_len
) {
2195 *ext_backing_fmt
= (QCowExtension
) {
2196 .magic
= cpu_to_be32(magic
),
2197 .len
= cpu_to_be32(len
),
2201 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2208 * Updates the qcow2 header, including the variable length parts of it, i.e.
2209 * the backing file name and all extensions. qcow2 was not designed to allow
2210 * such changes, so if we run out of space (we can only use the first cluster)
2211 * this function may fail.
2213 * Returns 0 on success, -errno in error cases.
2215 int qcow2_update_header(BlockDriverState
*bs
)
2217 BDRVQcow2State
*s
= bs
->opaque
;
2220 size_t buflen
= s
->cluster_size
;
2222 uint64_t total_size
;
2223 uint32_t refcount_table_clusters
;
2224 size_t header_length
;
2225 Qcow2UnknownHeaderExtension
*uext
;
2227 buf
= qemu_blockalign(bs
, buflen
);
2229 /* Header structure */
2230 header
= (QCowHeader
*) buf
;
2232 if (buflen
< sizeof(*header
)) {
2237 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2238 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2239 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2241 *header
= (QCowHeader
) {
2242 /* Version 2 fields */
2243 .magic
= cpu_to_be32(QCOW_MAGIC
),
2244 .version
= cpu_to_be32(s
->qcow_version
),
2245 .backing_file_offset
= 0,
2246 .backing_file_size
= 0,
2247 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2248 .size
= cpu_to_be64(total_size
),
2249 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2250 .l1_size
= cpu_to_be32(s
->l1_size
),
2251 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2252 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2253 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2254 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2255 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2257 /* Version 3 fields */
2258 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2259 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2260 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2261 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2262 .header_length
= cpu_to_be32(header_length
),
2265 /* For older versions, write a shorter header */
2266 switch (s
->qcow_version
) {
2268 ret
= offsetof(QCowHeader
, incompatible_features
);
2271 ret
= sizeof(*header
);
2280 memset(buf
, 0, buflen
);
2282 /* Preserve any unknown field in the header */
2283 if (s
->unknown_header_fields_size
) {
2284 if (buflen
< s
->unknown_header_fields_size
) {
2289 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2290 buf
+= s
->unknown_header_fields_size
;
2291 buflen
-= s
->unknown_header_fields_size
;
2294 /* Backing file format header extension */
2295 if (s
->image_backing_format
) {
2296 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2297 s
->image_backing_format
,
2298 strlen(s
->image_backing_format
),
2308 /* Full disk encryption header pointer extension */
2309 if (s
->crypto_header
.offset
!= 0) {
2310 cpu_to_be64s(&s
->crypto_header
.offset
);
2311 cpu_to_be64s(&s
->crypto_header
.length
);
2312 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2313 &s
->crypto_header
, sizeof(s
->crypto_header
),
2315 be64_to_cpus(&s
->crypto_header
.offset
);
2316 be64_to_cpus(&s
->crypto_header
.length
);
2325 if (s
->qcow_version
>= 3) {
2326 Qcow2Feature features
[] = {
2328 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2329 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2330 .name
= "dirty bit",
2333 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2334 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
2335 .name
= "corrupt bit",
2338 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
2339 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
2340 .name
= "lazy refcounts",
2344 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
2345 features
, sizeof(features
), buflen
);
2353 /* Bitmap extension */
2354 if (s
->nb_bitmaps
> 0) {
2355 Qcow2BitmapHeaderExt bitmaps_header
= {
2356 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
2357 .bitmap_directory_size
=
2358 cpu_to_be64(s
->bitmap_directory_size
),
2359 .bitmap_directory_offset
=
2360 cpu_to_be64(s
->bitmap_directory_offset
)
2362 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
2363 &bitmaps_header
, sizeof(bitmaps_header
),
2372 /* Keep unknown header extensions */
2373 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
2374 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
2383 /* End of header extensions */
2384 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
2392 /* Backing file name */
2393 if (s
->image_backing_file
) {
2394 size_t backing_file_len
= strlen(s
->image_backing_file
);
2396 if (buflen
< backing_file_len
) {
2401 /* Using strncpy is ok here, since buf is not NUL-terminated. */
2402 strncpy(buf
, s
->image_backing_file
, buflen
);
2404 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
2405 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
2408 /* Write the new header */
2409 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
2420 static int qcow2_change_backing_file(BlockDriverState
*bs
,
2421 const char *backing_file
, const char *backing_fmt
)
2423 BDRVQcow2State
*s
= bs
->opaque
;
2425 if (backing_file
&& strlen(backing_file
) > 1023) {
2429 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
2430 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
2432 g_free(s
->image_backing_file
);
2433 g_free(s
->image_backing_format
);
2435 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
2436 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
2438 return qcow2_update_header(bs
);
2441 static int qcow2_crypt_method_from_format(const char *encryptfmt
)
2443 if (g_str_equal(encryptfmt
, "luks")) {
2444 return QCOW_CRYPT_LUKS
;
2445 } else if (g_str_equal(encryptfmt
, "aes")) {
2446 return QCOW_CRYPT_AES
;
2452 static int qcow2_set_up_encryption(BlockDriverState
*bs
, const char *encryptfmt
,
2453 QemuOpts
*opts
, Error
**errp
)
2455 BDRVQcow2State
*s
= bs
->opaque
;
2456 QCryptoBlockCreateOptions
*cryptoopts
= NULL
;
2457 QCryptoBlock
*crypto
= NULL
;
2459 QDict
*options
, *encryptopts
;
2462 options
= qemu_opts_to_qdict(opts
, NULL
);
2463 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
2466 fmt
= qcow2_crypt_method_from_format(encryptfmt
);
2469 case QCOW_CRYPT_LUKS
:
2470 cryptoopts
= block_crypto_create_opts_init(
2471 Q_CRYPTO_BLOCK_FORMAT_LUKS
, encryptopts
, errp
);
2473 case QCOW_CRYPT_AES
:
2474 cryptoopts
= block_crypto_create_opts_init(
2475 Q_CRYPTO_BLOCK_FORMAT_QCOW
, encryptopts
, errp
);
2478 error_setg(errp
, "Unknown encryption format '%s'", encryptfmt
);
2485 s
->crypt_method_header
= fmt
;
2487 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
2488 qcow2_crypto_hdr_init_func
,
2489 qcow2_crypto_hdr_write_func
,
2496 ret
= qcow2_update_header(bs
);
2498 error_setg_errno(errp
, -ret
, "Could not write encryption header");
2503 QDECREF(encryptopts
);
2504 qcrypto_block_free(crypto
);
2505 qapi_free_QCryptoBlockCreateOptions(cryptoopts
);
2510 typedef struct PreallocCo
{
2511 BlockDriverState
*bs
;
2513 uint64_t new_length
;
2519 * Preallocates metadata structures for data clusters between @offset (in the
2520 * guest disk) and @new_length (which is thus generally the new guest disk
2523 * Returns: 0 on success, -errno on failure.
2525 static void coroutine_fn
preallocate_co(void *opaque
)
2527 PreallocCo
*params
= opaque
;
2528 BlockDriverState
*bs
= params
->bs
;
2529 uint64_t offset
= params
->offset
;
2530 uint64_t new_length
= params
->new_length
;
2531 BDRVQcow2State
*s
= bs
->opaque
;
2533 uint64_t host_offset
= 0;
2534 unsigned int cur_bytes
;
2538 qemu_co_mutex_lock(&s
->lock
);
2540 assert(offset
<= new_length
);
2541 bytes
= new_length
- offset
;
2544 cur_bytes
= MIN(bytes
, INT_MAX
);
2545 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2546 &host_offset
, &meta
);
2552 QCowL2Meta
*next
= meta
->next
;
2554 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
2556 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
2557 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
2561 /* There are no dependent requests, but we need to remove our
2562 * request from the list of in-flight requests */
2563 QLIST_REMOVE(meta
, next_in_flight
);
2569 /* TODO Preallocate data if requested */
2572 offset
+= cur_bytes
;
2576 * It is expected that the image file is large enough to actually contain
2577 * all of the allocated clusters (otherwise we get failing reads after
2578 * EOF). Extend the image to the last allocated sector.
2580 if (host_offset
!= 0) {
2582 ret
= bdrv_pwrite(bs
->file
, (host_offset
+ cur_bytes
) - 1,
2592 qemu_co_mutex_unlock(&s
->lock
);
2596 static int preallocate(BlockDriverState
*bs
,
2597 uint64_t offset
, uint64_t new_length
)
2599 PreallocCo params
= {
2602 .new_length
= new_length
,
2603 .ret
= -EINPROGRESS
,
2606 if (qemu_in_coroutine()) {
2607 preallocate_co(¶ms
);
2609 Coroutine
*co
= qemu_coroutine_create(preallocate_co
, ¶ms
);
2610 bdrv_coroutine_enter(bs
, co
);
2611 BDRV_POLL_WHILE(bs
, params
.ret
== -EINPROGRESS
);
2616 /* qcow2_refcount_metadata_size:
2617 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2618 * @cluster_size: size of a cluster, in bytes
2619 * @refcount_order: refcount bits power-of-2 exponent
2620 * @generous_increase: allow for the refcount table to be 1.5x as large as it
2623 * Returns: Number of bytes required for refcount blocks and table metadata.
2625 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
2626 int refcount_order
, bool generous_increase
,
2627 uint64_t *refblock_count
)
2630 * Every host cluster is reference-counted, including metadata (even
2631 * refcount metadata is recursively included).
2633 * An accurate formula for the size of refcount metadata size is difficult
2634 * to derive. An easier method of calculation is finding the fixed point
2635 * where no further refcount blocks or table clusters are required to
2636 * reference count every cluster.
2638 int64_t blocks_per_table_cluster
= cluster_size
/ sizeof(uint64_t);
2639 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
2640 int64_t table
= 0; /* number of refcount table clusters */
2641 int64_t blocks
= 0; /* number of refcount block clusters */
2647 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
2648 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
2649 n
= clusters
+ blocks
+ table
;
2651 if (n
== last
&& generous_increase
) {
2652 clusters
+= DIV_ROUND_UP(table
, 2);
2653 n
= 0; /* force another loop */
2654 generous_increase
= false;
2656 } while (n
!= last
);
2658 if (refblock_count
) {
2659 *refblock_count
= blocks
;
2662 return (blocks
+ table
) * cluster_size
;
2666 * qcow2_calc_prealloc_size:
2667 * @total_size: virtual disk size in bytes
2668 * @cluster_size: cluster size in bytes
2669 * @refcount_order: refcount bits power-of-2 exponent
2671 * Returns: Total number of bytes required for the fully allocated image
2672 * (including metadata).
2674 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
2675 size_t cluster_size
,
2678 int64_t meta_size
= 0;
2679 uint64_t nl1e
, nl2e
;
2680 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
2682 /* header: 1 cluster */
2683 meta_size
+= cluster_size
;
2685 /* total size of L2 tables */
2686 nl2e
= aligned_total_size
/ cluster_size
;
2687 nl2e
= ROUND_UP(nl2e
, cluster_size
/ sizeof(uint64_t));
2688 meta_size
+= nl2e
* sizeof(uint64_t);
2690 /* total size of L1 tables */
2691 nl1e
= nl2e
* sizeof(uint64_t) / cluster_size
;
2692 nl1e
= ROUND_UP(nl1e
, cluster_size
/ sizeof(uint64_t));
2693 meta_size
+= nl1e
* sizeof(uint64_t);
2695 /* total size of refcount table and blocks */
2696 meta_size
+= qcow2_refcount_metadata_size(
2697 (meta_size
+ aligned_total_size
) / cluster_size
,
2698 cluster_size
, refcount_order
, false, NULL
);
2700 return meta_size
+ aligned_total_size
;
2703 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, Error
**errp
)
2705 size_t cluster_size
;
2708 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
2709 DEFAULT_CLUSTER_SIZE
);
2710 cluster_bits
= ctz32(cluster_size
);
2711 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
2712 (1 << cluster_bits
) != cluster_size
)
2714 error_setg(errp
, "Cluster size must be a power of two between %d and "
2715 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
2718 return cluster_size
;
2721 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
2726 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
2728 ret
= 3; /* default */
2729 } else if (!strcmp(buf
, "0.10")) {
2731 } else if (!strcmp(buf
, "1.1")) {
2734 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
2741 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
2744 uint64_t refcount_bits
;
2746 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
2747 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
2748 error_setg(errp
, "Refcount width must be a power of two and may not "
2753 if (version
< 3 && refcount_bits
!= 16) {
2754 error_setg(errp
, "Different refcount widths than 16 bits require "
2755 "compatibility level 1.1 or above (use compat=1.1 or "
2760 return refcount_bits
;
2763 static int coroutine_fn
2764 qcow2_co_create(BlockDriverState
*bs
, int64_t total_size
,
2765 const char *backing_file
, const char *backing_format
,
2766 int flags
, size_t cluster_size
, PreallocMode prealloc
,
2767 QemuOpts
*opts
, int version
, int refcount_order
,
2768 const char *encryptfmt
, Error
**errp
)
2773 * Open the image file and write a minimal qcow2 header.
2775 * We keep things simple and start with a zero-sized image. We also
2776 * do without refcount blocks or a L1 table for now. We'll fix the
2777 * inconsistency later.
2779 * We do need a refcount table because growing the refcount table means
2780 * allocating two new refcount blocks - the seconds of which would be at
2781 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2782 * size for any qcow2 image.
2786 uint64_t* refcount_table
;
2787 Error
*local_err
= NULL
;
2790 blk
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
);
2791 ret
= blk_insert_bs(blk
, bs
, errp
);
2795 blk_set_allow_write_beyond_eof(blk
, true);
2797 /* Write the header */
2798 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
2799 header
= g_malloc0(cluster_size
);
2800 *header
= (QCowHeader
) {
2801 .magic
= cpu_to_be32(QCOW_MAGIC
),
2802 .version
= cpu_to_be32(version
),
2803 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
2804 .size
= cpu_to_be64(0),
2805 .l1_table_offset
= cpu_to_be64(0),
2806 .l1_size
= cpu_to_be32(0),
2807 .refcount_table_offset
= cpu_to_be64(cluster_size
),
2808 .refcount_table_clusters
= cpu_to_be32(1),
2809 .refcount_order
= cpu_to_be32(refcount_order
),
2810 .header_length
= cpu_to_be32(sizeof(*header
)),
2813 /* We'll update this to correct value later */
2814 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
2816 if (flags
& BLOCK_FLAG_LAZY_REFCOUNTS
) {
2817 header
->compatible_features
|=
2818 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
2821 ret
= blk_pwrite(blk
, 0, header
, cluster_size
, 0);
2824 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
2828 /* Write a refcount table with one refcount block */
2829 refcount_table
= g_malloc0(2 * cluster_size
);
2830 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
2831 ret
= blk_pwrite(blk
, cluster_size
, refcount_table
, 2 * cluster_size
, 0);
2832 g_free(refcount_table
);
2835 error_setg_errno(errp
, -ret
, "Could not write refcount table");
2843 * And now open the image and make it consistent first (i.e. increase the
2844 * refcount of the cluster that is occupied by the header and the refcount
2847 options
= qdict_new();
2848 qdict_put_str(options
, "driver", "qcow2");
2849 qdict_put_str(options
, "file", bs
->node_name
);
2850 blk
= blk_new_open(NULL
, NULL
, options
,
2851 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
2854 error_propagate(errp
, local_err
);
2859 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
2861 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
2862 "header and refcount table");
2865 } else if (ret
!= 0) {
2866 error_report("Huh, first cluster in empty image is already in use?");
2870 /* Create a full header (including things like feature table) */
2871 ret
= qcow2_update_header(blk_bs(blk
));
2873 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
2877 /* Okay, now that we have a valid image, let's give it the right size */
2878 ret
= blk_truncate(blk
, total_size
, PREALLOC_MODE_OFF
, errp
);
2880 error_prepend(errp
, "Could not resize image: ");
2884 /* Want a backing file? There you go.*/
2886 ret
= bdrv_change_backing_file(blk_bs(blk
), backing_file
, backing_format
);
2888 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
2889 "with format '%s'", backing_file
, backing_format
);
2894 /* Want encryption? There you go. */
2896 ret
= qcow2_set_up_encryption(blk_bs(blk
), encryptfmt
, opts
, errp
);
2902 /* And if we're supposed to preallocate metadata, do that now */
2903 if (prealloc
!= PREALLOC_MODE_OFF
) {
2904 ret
= preallocate(blk_bs(blk
), 0, total_size
);
2906 error_setg_errno(errp
, -ret
, "Could not preallocate metadata");
2914 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
2915 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
2916 * have to setup decryption context. We're not doing any I/O on the top
2917 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
2920 options
= qdict_new();
2921 qdict_put_str(options
, "driver", "qcow2");
2922 qdict_put_str(options
, "file", bs
->node_name
);
2923 blk
= blk_new_open(NULL
, NULL
, options
,
2924 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
2927 error_propagate(errp
, local_err
);
2940 static int coroutine_fn
qcow2_co_create_opts(const char *filename
, QemuOpts
*opts
,
2943 char *backing_file
= NULL
;
2944 char *backing_fmt
= NULL
;
2948 size_t cluster_size
= DEFAULT_CLUSTER_SIZE
;
2949 PreallocMode prealloc
;
2951 uint64_t refcount_bits
;
2953 char *encryptfmt
= NULL
;
2954 BlockDriverState
*bs
= NULL
;
2955 Error
*local_err
= NULL
;
2958 /* Read out options */
2959 size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2961 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
2962 backing_fmt
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FMT
);
2963 encryptfmt
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
2965 if (qemu_opt_get(opts
, BLOCK_OPT_ENCRYPT
)) {
2966 error_setg(errp
, "Options " BLOCK_OPT_ENCRYPT
" and "
2967 BLOCK_OPT_ENCRYPT_FORMAT
" are mutually exclusive");
2971 } else if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_ENCRYPT
, false)) {
2972 encryptfmt
= g_strdup("aes");
2974 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, &local_err
);
2976 error_propagate(errp
, local_err
);
2980 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2981 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2982 PREALLOC_MODE_OFF
, &local_err
);
2984 error_propagate(errp
, local_err
);
2989 version
= qcow2_opt_get_version_del(opts
, &local_err
);
2991 error_propagate(errp
, local_err
);
2996 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_LAZY_REFCOUNTS
, false)) {
2997 flags
|= BLOCK_FLAG_LAZY_REFCOUNTS
;
3000 if (backing_file
&& prealloc
!= PREALLOC_MODE_OFF
) {
3001 error_setg(errp
, "Backing file and preallocation cannot be used at "
3007 if (version
< 3 && (flags
& BLOCK_FLAG_LAZY_REFCOUNTS
)) {
3008 error_setg(errp
, "Lazy refcounts only supported with compatibility "
3009 "level 1.1 and above (use compat=1.1 or greater)");
3014 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
3016 error_propagate(errp
, local_err
);
3021 refcount_order
= ctz32(refcount_bits
);
3023 /* Create and open the file (protocol layer) */
3024 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
3025 int64_t prealloc_size
=
3026 qcow2_calc_prealloc_size(size
, cluster_size
, refcount_order
);
3027 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, prealloc_size
, &error_abort
);
3028 qemu_opt_set(opts
, BLOCK_OPT_PREALLOC
, PreallocMode_str(prealloc
),
3032 ret
= bdrv_create_file(filename
, opts
, errp
);
3037 bs
= bdrv_open(filename
, NULL
, NULL
,
3038 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3044 /* Create the qcow2 image (format layer) */
3045 ret
= qcow2_co_create(bs
, size
, backing_file
, backing_fmt
, flags
,
3046 cluster_size
, prealloc
, opts
, version
, refcount_order
,
3055 g_free(backing_file
);
3056 g_free(backing_fmt
);
3063 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3068 /* Clamp to image length, before checking status of underlying sectors */
3069 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3070 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3076 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3077 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
== bytes
;
3080 static coroutine_fn
int qcow2_co_pwrite_zeroes(BlockDriverState
*bs
,
3081 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3084 BDRVQcow2State
*s
= bs
->opaque
;
3086 uint32_t head
= offset
% s
->cluster_size
;
3087 uint32_t tail
= (offset
+ bytes
) % s
->cluster_size
;
3089 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3090 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3098 assert(head
+ bytes
<= s
->cluster_size
);
3100 /* check whether remainder of cluster already reads as zero */
3101 if (!(is_zero(bs
, offset
- head
, head
) &&
3102 is_zero(bs
, offset
+ bytes
,
3103 tail
? s
->cluster_size
- tail
: 0))) {
3107 qemu_co_mutex_lock(&s
->lock
);
3108 /* We can have new write after previous check */
3109 offset
= QEMU_ALIGN_DOWN(offset
, s
->cluster_size
);
3110 bytes
= s
->cluster_size
;
3111 nr
= s
->cluster_size
;
3112 ret
= qcow2_get_cluster_offset(bs
, offset
, &nr
, &off
);
3113 if (ret
!= QCOW2_CLUSTER_UNALLOCATED
&&
3114 ret
!= QCOW2_CLUSTER_ZERO_PLAIN
&&
3115 ret
!= QCOW2_CLUSTER_ZERO_ALLOC
) {
3116 qemu_co_mutex_unlock(&s
->lock
);
3120 qemu_co_mutex_lock(&s
->lock
);
3123 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
3125 /* Whatever is left can use real zero clusters */
3126 ret
= qcow2_cluster_zeroize(bs
, offset
, bytes
, flags
);
3127 qemu_co_mutex_unlock(&s
->lock
);
3132 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
3133 int64_t offset
, int bytes
)
3136 BDRVQcow2State
*s
= bs
->opaque
;
3138 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
3139 assert(bytes
< s
->cluster_size
);
3140 /* Ignore partial clusters, except for the special case of the
3141 * complete partial cluster at the end of an unaligned file */
3142 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
3143 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3148 qemu_co_mutex_lock(&s
->lock
);
3149 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
3151 qemu_co_mutex_unlock(&s
->lock
);
3155 static int qcow2_truncate(BlockDriverState
*bs
, int64_t offset
,
3156 PreallocMode prealloc
, Error
**errp
)
3158 BDRVQcow2State
*s
= bs
->opaque
;
3159 uint64_t old_length
;
3160 int64_t new_l1_size
;
3163 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
3164 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
3166 error_setg(errp
, "Unsupported preallocation mode '%s'",
3167 PreallocMode_str(prealloc
));
3172 error_setg(errp
, "The new size must be a multiple of 512");
3176 /* cannot proceed if image has snapshots */
3177 if (s
->nb_snapshots
) {
3178 error_setg(errp
, "Can't resize an image which has snapshots");
3182 /* cannot proceed if image has bitmaps */
3183 if (s
->nb_bitmaps
) {
3184 /* TODO: resize bitmaps in the image */
3185 error_setg(errp
, "Can't resize an image which has bitmaps");
3189 old_length
= bs
->total_sectors
* 512;
3190 new_l1_size
= size_to_l1(s
, offset
);
3192 if (offset
< old_length
) {
3193 int64_t last_cluster
, old_file_size
;
3194 if (prealloc
!= PREALLOC_MODE_OFF
) {
3196 "Preallocation can't be used for shrinking an image");
3200 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
3201 old_length
- ROUND_UP(offset
,
3203 QCOW2_DISCARD_ALWAYS
, true);
3205 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
3209 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
3211 error_setg_errno(errp
, -ret
,
3212 "Failed to reduce the number of L2 tables");
3216 ret
= qcow2_shrink_reftable(bs
);
3218 error_setg_errno(errp
, -ret
,
3219 "Failed to discard unused refblocks");
3223 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3224 if (old_file_size
< 0) {
3225 error_setg_errno(errp
, -old_file_size
,
3226 "Failed to inquire current file length");
3227 return old_file_size
;
3229 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
3230 if (last_cluster
< 0) {
3231 error_setg_errno(errp
, -last_cluster
,
3232 "Failed to find the last cluster");
3233 return last_cluster
;
3235 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
3236 Error
*local_err
= NULL
;
3238 bdrv_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
3239 PREALLOC_MODE_OFF
, &local_err
);
3241 warn_reportf_err(local_err
,
3242 "Failed to truncate the tail of the image: ");
3246 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
3248 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
3254 case PREALLOC_MODE_OFF
:
3257 case PREALLOC_MODE_METADATA
:
3258 ret
= preallocate(bs
, old_length
, offset
);
3260 error_setg_errno(errp
, -ret
, "Preallocation failed");
3265 case PREALLOC_MODE_FALLOC
:
3266 case PREALLOC_MODE_FULL
:
3268 int64_t allocation_start
, host_offset
, guest_offset
;
3269 int64_t clusters_allocated
;
3270 int64_t old_file_size
, new_file_size
;
3271 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
3273 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3274 if (old_file_size
< 0) {
3275 error_setg_errno(errp
, -old_file_size
,
3276 "Failed to inquire current file length");
3277 return old_file_size
;
3279 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
3281 nb_new_data_clusters
= DIV_ROUND_UP(offset
- old_length
,
3284 /* This is an overestimation; we will not actually allocate space for
3285 * these in the file but just make sure the new refcount structures are
3286 * able to cover them so we will not have to allocate new refblocks
3287 * while entering the data blocks in the potentially new L2 tables.
3288 * (We do not actually care where the L2 tables are placed. Maybe they
3289 * are already allocated or they can be placed somewhere before
3290 * @old_file_size. It does not matter because they will be fully
3291 * allocated automatically, so they do not need to be covered by the
3292 * preallocation. All that matters is that we will not have to allocate
3293 * new refcount structures for them.) */
3294 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
3295 s
->cluster_size
/ sizeof(uint64_t));
3296 /* The cluster range may not be aligned to L2 boundaries, so add one L2
3297 * table for a potential head/tail */
3300 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
3301 nb_new_data_clusters
+
3304 if (allocation_start
< 0) {
3305 error_setg_errno(errp
, -allocation_start
,
3306 "Failed to resize refcount structures");
3307 return allocation_start
;
3310 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
3311 nb_new_data_clusters
);
3312 if (clusters_allocated
< 0) {
3313 error_setg_errno(errp
, -clusters_allocated
,
3314 "Failed to allocate data clusters");
3315 return -clusters_allocated
;
3318 assert(clusters_allocated
== nb_new_data_clusters
);
3320 /* Allocate the data area */
3321 new_file_size
= allocation_start
+
3322 nb_new_data_clusters
* s
->cluster_size
;
3323 ret
= bdrv_truncate(bs
->file
, new_file_size
, prealloc
, errp
);
3325 error_prepend(errp
, "Failed to resize underlying file: ");
3326 qcow2_free_clusters(bs
, allocation_start
,
3327 nb_new_data_clusters
* s
->cluster_size
,
3328 QCOW2_DISCARD_OTHER
);
3332 /* Create the necessary L2 entries */
3333 host_offset
= allocation_start
;
3334 guest_offset
= old_length
;
3335 while (nb_new_data_clusters
) {
3336 int64_t nb_clusters
= MIN(
3337 nb_new_data_clusters
,
3338 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
3339 QCowL2Meta allocation
= {
3340 .offset
= guest_offset
,
3341 .alloc_offset
= host_offset
,
3342 .nb_clusters
= nb_clusters
,
3344 qemu_co_queue_init(&allocation
.dependent_requests
);
3346 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
3348 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
3349 qcow2_free_clusters(bs
, host_offset
,
3350 nb_new_data_clusters
* s
->cluster_size
,
3351 QCOW2_DISCARD_OTHER
);
3355 guest_offset
+= nb_clusters
* s
->cluster_size
;
3356 host_offset
+= nb_clusters
* s
->cluster_size
;
3357 nb_new_data_clusters
-= nb_clusters
;
3363 g_assert_not_reached();
3366 if (prealloc
!= PREALLOC_MODE_OFF
) {
3367 /* Flush metadata before actually changing the image size */
3368 ret
= bdrv_flush(bs
);
3370 error_setg_errno(errp
, -ret
,
3371 "Failed to flush the preallocated area to disk");
3376 /* write updated header.size */
3377 offset
= cpu_to_be64(offset
);
3378 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
3379 &offset
, sizeof(uint64_t));
3381 error_setg_errno(errp
, -ret
, "Failed to update the image size");
3385 s
->l1_vm_state_index
= new_l1_size
;
3389 /* XXX: put compressed sectors first, then all the cluster aligned
3390 tables to avoid losing bytes in alignment */
3391 static coroutine_fn
int
3392 qcow2_co_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
3393 uint64_t bytes
, QEMUIOVector
*qiov
)
3395 BDRVQcow2State
*s
= bs
->opaque
;
3396 QEMUIOVector hd_qiov
;
3400 uint8_t *buf
, *out_buf
;
3401 int64_t cluster_offset
;
3404 /* align end of file to a sector boundary to ease reading with
3405 sector based I/Os */
3406 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
3407 if (cluster_offset
< 0) {
3408 return cluster_offset
;
3410 return bdrv_truncate(bs
->file
, cluster_offset
, PREALLOC_MODE_OFF
, NULL
);
3413 if (offset_into_cluster(s
, offset
)) {
3417 buf
= qemu_blockalign(bs
, s
->cluster_size
);
3418 if (bytes
!= s
->cluster_size
) {
3419 if (bytes
> s
->cluster_size
||
3420 offset
+ bytes
!= bs
->total_sectors
<< BDRV_SECTOR_BITS
)
3425 /* Zero-pad last write if image size is not cluster aligned */
3426 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
3428 qemu_iovec_to_buf(qiov
, 0, buf
, bytes
);
3430 out_buf
= g_malloc(s
->cluster_size
);
3432 /* best compression, small window, no zlib header */
3433 memset(&strm
, 0, sizeof(strm
));
3434 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
3436 9, Z_DEFAULT_STRATEGY
);
3442 strm
.avail_in
= s
->cluster_size
;
3443 strm
.next_in
= (uint8_t *)buf
;
3444 strm
.avail_out
= s
->cluster_size
;
3445 strm
.next_out
= out_buf
;
3447 ret
= deflate(&strm
, Z_FINISH
);
3448 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
3453 out_len
= strm
.next_out
- out_buf
;
3457 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
3458 /* could not compress: write normal cluster */
3459 ret
= qcow2_co_pwritev(bs
, offset
, bytes
, qiov
, 0);
3466 qemu_co_mutex_lock(&s
->lock
);
3468 qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
);
3469 if (!cluster_offset
) {
3470 qemu_co_mutex_unlock(&s
->lock
);
3474 cluster_offset
&= s
->cluster_offset_mask
;
3476 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
);
3477 qemu_co_mutex_unlock(&s
->lock
);
3482 iov
= (struct iovec
) {
3483 .iov_base
= out_buf
,
3486 qemu_iovec_init_external(&hd_qiov
, &iov
, 1);
3488 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
3489 ret
= bdrv_co_pwritev(bs
->file
, cluster_offset
, out_len
, &hd_qiov
, 0);
3501 static int make_completely_empty(BlockDriverState
*bs
)
3503 BDRVQcow2State
*s
= bs
->opaque
;
3504 Error
*local_err
= NULL
;
3505 int ret
, l1_clusters
;
3507 uint64_t *new_reftable
= NULL
;
3508 uint64_t rt_entry
, l1_size2
;
3511 uint64_t reftable_offset
;
3512 uint32_t reftable_clusters
;
3513 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
3515 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
3520 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
3525 /* Refcounts will be broken utterly */
3526 ret
= qcow2_mark_dirty(bs
);
3531 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
3533 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
3534 l1_size2
= (uint64_t)s
->l1_size
* sizeof(uint64_t);
3536 /* After this call, neither the in-memory nor the on-disk refcount
3537 * information accurately describe the actual references */
3539 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
3540 l1_clusters
* s
->cluster_size
, 0);
3542 goto fail_broken_refcounts
;
3544 memset(s
->l1_table
, 0, l1_size2
);
3546 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
3548 /* Overwrite enough clusters at the beginning of the sectors to place
3549 * the refcount table, a refcount block and the L1 table in; this may
3550 * overwrite parts of the existing refcount and L1 table, which is not
3551 * an issue because the dirty flag is set, complete data loss is in fact
3552 * desired and partial data loss is consequently fine as well */
3553 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
3554 (2 + l1_clusters
) * s
->cluster_size
, 0);
3555 /* This call (even if it failed overall) may have overwritten on-disk
3556 * refcount structures; in that case, the in-memory refcount information
3557 * will probably differ from the on-disk information which makes the BDS
3560 goto fail_broken_refcounts
;
3563 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
3564 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
3566 /* "Create" an empty reftable (one cluster) directly after the image
3567 * header and an empty L1 table three clusters after the image header;
3568 * the cluster between those two will be used as the first refblock */
3569 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
3570 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
3571 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
3572 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
3573 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
3575 goto fail_broken_refcounts
;
3578 s
->l1_table_offset
= 3 * s
->cluster_size
;
3580 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ sizeof(uint64_t));
3581 if (!new_reftable
) {
3583 goto fail_broken_refcounts
;
3586 s
->refcount_table_offset
= s
->cluster_size
;
3587 s
->refcount_table_size
= s
->cluster_size
/ sizeof(uint64_t);
3588 s
->max_refcount_table_index
= 0;
3590 g_free(s
->refcount_table
);
3591 s
->refcount_table
= new_reftable
;
3592 new_reftable
= NULL
;
3594 /* Now the in-memory refcount information again corresponds to the on-disk
3595 * information (reftable is empty and no refblocks (the refblock cache is
3596 * empty)); however, this means some clusters (e.g. the image header) are
3597 * referenced, but not refcounted, but the normal qcow2 code assumes that
3598 * the in-memory information is always correct */
3600 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
3602 /* Enter the first refblock into the reftable */
3603 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
3604 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
3605 &rt_entry
, sizeof(rt_entry
));
3607 goto fail_broken_refcounts
;
3609 s
->refcount_table
[0] = 2 * s
->cluster_size
;
3611 s
->free_cluster_index
= 0;
3612 assert(3 + l1_clusters
<= s
->refcount_block_size
);
3613 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
3616 goto fail_broken_refcounts
;
3617 } else if (offset
> 0) {
3618 error_report("First cluster in emptied image is in use");
3622 /* Now finally the in-memory information corresponds to the on-disk
3623 * structures and is correct */
3624 ret
= qcow2_mark_clean(bs
);
3629 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
,
3630 PREALLOC_MODE_OFF
, &local_err
);
3632 error_report_err(local_err
);
3638 fail_broken_refcounts
:
3639 /* The BDS is unusable at this point. If we wanted to make it usable, we
3640 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
3641 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
3642 * again. However, because the functions which could have caused this error
3643 * path to be taken are used by those functions as well, it's very likely
3644 * that that sequence will fail as well. Therefore, just eject the BDS. */
3648 g_free(new_reftable
);
3652 static int qcow2_make_empty(BlockDriverState
*bs
)
3654 BDRVQcow2State
*s
= bs
->opaque
;
3655 uint64_t offset
, end_offset
;
3656 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
3657 int l1_clusters
, ret
= 0;
3659 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
3661 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
3662 3 + l1_clusters
<= s
->refcount_block_size
&&
3663 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
3664 /* The following function only works for qcow2 v3 images (it
3665 * requires the dirty flag) and only as long as there are no
3666 * features that reserve extra clusters (such as snapshots,
3667 * LUKS header, or persistent bitmaps), because it completely
3668 * empties the image. Furthermore, the L1 table and three
3669 * additional clusters (image header, refcount table, one
3670 * refcount block) have to fit inside one refcount block. */
3671 return make_completely_empty(bs
);
3674 /* This fallback code simply discards every active cluster; this is slow,
3675 * but works in all cases */
3676 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3677 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
3678 /* As this function is generally used after committing an external
3679 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
3680 * default action for this kind of discard is to pass the discard,
3681 * which will ideally result in an actually smaller image file, as
3682 * is probably desired. */
3683 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
3684 QCOW2_DISCARD_SNAPSHOT
, true);
3693 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
3695 BDRVQcow2State
*s
= bs
->opaque
;
3698 qemu_co_mutex_lock(&s
->lock
);
3699 ret
= qcow2_write_caches(bs
);
3700 qemu_co_mutex_unlock(&s
->lock
);
3705 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
3708 Error
*local_err
= NULL
;
3709 BlockMeasureInfo
*info
;
3710 uint64_t required
= 0; /* bytes that contribute to required size */
3711 uint64_t virtual_size
; /* disk size as seen by guest */
3712 uint64_t refcount_bits
;
3714 size_t cluster_size
;
3717 PreallocMode prealloc
;
3718 bool has_backing_file
;
3720 /* Parse image creation options */
3721 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, &local_err
);
3726 version
= qcow2_opt_get_version_del(opts
, &local_err
);
3731 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
3736 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
3737 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
3738 PREALLOC_MODE_OFF
, &local_err
);
3744 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
3745 has_backing_file
= !!optstr
;
3748 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
3749 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
3751 /* Check that virtual disk size is valid */
3752 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
3753 cluster_size
/ sizeof(uint64_t));
3754 if (l2_tables
* sizeof(uint64_t) > QCOW_MAX_L1_SIZE
) {
3755 error_setg(&local_err
, "The image size is too large "
3756 "(try using a larger cluster size)");
3760 /* Account for input image */
3762 int64_t ssize
= bdrv_getlength(in_bs
);
3764 error_setg_errno(&local_err
, -ssize
,
3765 "Unable to get image virtual_size");
3769 virtual_size
= ROUND_UP(ssize
, cluster_size
);
3771 if (has_backing_file
) {
3772 /* We don't how much of the backing chain is shared by the input
3773 * image and the new image file. In the worst case the new image's
3774 * backing file has nothing in common with the input image. Be
3775 * conservative and assume all clusters need to be written.
3777 required
= virtual_size
;
3782 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
3785 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
3786 ssize
- offset
, &pnum
, NULL
,
3789 error_setg_errno(&local_err
, -ret
,
3790 "Unable to get block status");
3794 if (ret
& BDRV_BLOCK_ZERO
) {
3795 /* Skip zero regions (safe with no backing file) */
3796 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
3797 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
3798 /* Extend pnum to end of cluster for next iteration */
3799 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
3801 /* Count clusters we've seen */
3802 required
+= offset
% cluster_size
+ pnum
;
3808 /* Take into account preallocation. Nothing special is needed for
3809 * PREALLOC_MODE_METADATA since metadata is always counted.
3811 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
3812 required
= virtual_size
;
3815 info
= g_new(BlockMeasureInfo
, 1);
3816 info
->fully_allocated
=
3817 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
3818 ctz32(refcount_bits
));
3820 /* Remove data clusters that are not required. This overestimates the
3821 * required size because metadata needed for the fully allocated file is
3824 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
3828 error_propagate(errp
, local_err
);
3832 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3834 BDRVQcow2State
*s
= bs
->opaque
;
3835 bdi
->unallocated_blocks_are_zero
= true;
3836 bdi
->cluster_size
= s
->cluster_size
;
3837 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
3841 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
)
3843 BDRVQcow2State
*s
= bs
->opaque
;
3844 ImageInfoSpecific
*spec_info
;
3845 QCryptoBlockInfo
*encrypt_info
= NULL
;
3847 if (s
->crypto
!= NULL
) {
3848 encrypt_info
= qcrypto_block_get_info(s
->crypto
, &error_abort
);
3851 spec_info
= g_new(ImageInfoSpecific
, 1);
3852 *spec_info
= (ImageInfoSpecific
){
3853 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
3854 .u
.qcow2
.data
= g_new(ImageInfoSpecificQCow2
, 1),
3856 if (s
->qcow_version
== 2) {
3857 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
3858 .compat
= g_strdup("0.10"),
3859 .refcount_bits
= s
->refcount_bits
,
3861 } else if (s
->qcow_version
== 3) {
3862 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
3863 .compat
= g_strdup("1.1"),
3864 .lazy_refcounts
= s
->compatible_features
&
3865 QCOW2_COMPAT_LAZY_REFCOUNTS
,
3866 .has_lazy_refcounts
= true,
3867 .corrupt
= s
->incompatible_features
&
3868 QCOW2_INCOMPAT_CORRUPT
,
3869 .has_corrupt
= true,
3870 .refcount_bits
= s
->refcount_bits
,
3873 /* if this assertion fails, this probably means a new version was
3874 * added without having it covered here */
3879 ImageInfoSpecificQCow2Encryption
*qencrypt
=
3880 g_new(ImageInfoSpecificQCow2Encryption
, 1);
3881 switch (encrypt_info
->format
) {
3882 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
3883 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
3884 qencrypt
->u
.aes
= encrypt_info
->u
.qcow
;
3886 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
3887 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
3888 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
3893 /* Since we did shallow copy above, erase any pointers
3894 * in the original info */
3895 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
3896 qapi_free_QCryptoBlockInfo(encrypt_info
);
3898 spec_info
->u
.qcow2
.data
->has_encrypt
= true;
3899 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
3905 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
3908 BDRVQcow2State
*s
= bs
->opaque
;
3910 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
3911 return bs
->drv
->bdrv_co_pwritev(bs
, qcow2_vm_state_offset(s
) + pos
,
3912 qiov
->size
, qiov
, 0);
3915 static int qcow2_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
3918 BDRVQcow2State
*s
= bs
->opaque
;
3920 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
3921 return bs
->drv
->bdrv_co_preadv(bs
, qcow2_vm_state_offset(s
) + pos
,
3922 qiov
->size
, qiov
, 0);
3926 * Downgrades an image's version. To achieve this, any incompatible features
3927 * have to be removed.
3929 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
3930 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
)
3932 BDRVQcow2State
*s
= bs
->opaque
;
3933 int current_version
= s
->qcow_version
;
3936 if (target_version
== current_version
) {
3938 } else if (target_version
> current_version
) {
3940 } else if (target_version
!= 2) {
3944 if (s
->refcount_order
!= 4) {
3945 error_report("compat=0.10 requires refcount_bits=16");
3949 /* clear incompatible features */
3950 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
3951 ret
= qcow2_mark_clean(bs
);
3957 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
3958 * the first place; if that happens nonetheless, returning -ENOTSUP is the
3959 * best thing to do anyway */
3961 if (s
->incompatible_features
) {
3965 /* since we can ignore compatible features, we can set them to 0 as well */
3966 s
->compatible_features
= 0;
3967 /* if lazy refcounts have been used, they have already been fixed through
3968 * clearing the dirty flag */
3970 /* clearing autoclear features is trivial */
3971 s
->autoclear_features
= 0;
3973 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
3978 s
->qcow_version
= target_version
;
3979 ret
= qcow2_update_header(bs
);
3981 s
->qcow_version
= current_version
;
3987 typedef enum Qcow2AmendOperation
{
3988 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
3989 * statically initialized to so that the helper CB can discern the first
3990 * invocation from an operation change */
3991 QCOW2_NO_OPERATION
= 0,
3993 QCOW2_CHANGING_REFCOUNT_ORDER
,
3995 } Qcow2AmendOperation
;
3997 typedef struct Qcow2AmendHelperCBInfo
{
3998 /* The code coordinating the amend operations should only modify
3999 * these four fields; the rest will be managed by the CB */
4000 BlockDriverAmendStatusCB
*original_status_cb
;
4001 void *original_cb_opaque
;
4003 Qcow2AmendOperation current_operation
;
4005 /* Total number of operations to perform (only set once) */
4006 int total_operations
;
4008 /* The following fields are managed by the CB */
4010 /* Number of operations completed */
4011 int operations_completed
;
4013 /* Cumulative offset of all completed operations */
4014 int64_t offset_completed
;
4016 Qcow2AmendOperation last_operation
;
4017 int64_t last_work_size
;
4018 } Qcow2AmendHelperCBInfo
;
4020 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
4021 int64_t operation_offset
,
4022 int64_t operation_work_size
, void *opaque
)
4024 Qcow2AmendHelperCBInfo
*info
= opaque
;
4025 int64_t current_work_size
;
4026 int64_t projected_work_size
;
4028 if (info
->current_operation
!= info
->last_operation
) {
4029 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
4030 info
->offset_completed
+= info
->last_work_size
;
4031 info
->operations_completed
++;
4034 info
->last_operation
= info
->current_operation
;
4037 assert(info
->total_operations
> 0);
4038 assert(info
->operations_completed
< info
->total_operations
);
4040 info
->last_work_size
= operation_work_size
;
4042 current_work_size
= info
->offset_completed
+ operation_work_size
;
4044 /* current_work_size is the total work size for (operations_completed + 1)
4045 * operations (which includes this one), so multiply it by the number of
4046 * operations not covered and divide it by the number of operations
4047 * covered to get a projection for the operations not covered */
4048 projected_work_size
= current_work_size
* (info
->total_operations
-
4049 info
->operations_completed
- 1)
4050 / (info
->operations_completed
+ 1);
4052 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
4053 current_work_size
+ projected_work_size
,
4054 info
->original_cb_opaque
);
4057 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4058 BlockDriverAmendStatusCB
*status_cb
,
4061 BDRVQcow2State
*s
= bs
->opaque
;
4062 int old_version
= s
->qcow_version
, new_version
= old_version
;
4063 uint64_t new_size
= 0;
4064 const char *backing_file
= NULL
, *backing_format
= NULL
;
4065 bool lazy_refcounts
= s
->use_lazy_refcounts
;
4066 const char *compat
= NULL
;
4067 uint64_t cluster_size
= s
->cluster_size
;
4070 int refcount_bits
= s
->refcount_bits
;
4071 Error
*local_err
= NULL
;
4073 QemuOptDesc
*desc
= opts
->list
->desc
;
4074 Qcow2AmendHelperCBInfo helper_cb_info
;
4076 while (desc
&& desc
->name
) {
4077 if (!qemu_opt_find(opts
, desc
->name
)) {
4078 /* only change explicitly defined options */
4083 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
4084 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
4086 /* preserve default */
4087 } else if (!strcmp(compat
, "0.10")) {
4089 } else if (!strcmp(compat
, "1.1")) {
4092 error_report("Unknown compatibility level %s", compat
);
4095 } else if (!strcmp(desc
->name
, BLOCK_OPT_PREALLOC
)) {
4096 error_report("Cannot change preallocation mode");
4098 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
4099 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
4100 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
4101 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4102 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
4103 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4104 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT
)) {
4105 encrypt
= qemu_opt_get_bool(opts
, BLOCK_OPT_ENCRYPT
,
4108 if (encrypt
!= !!s
->crypto
) {
4109 error_report("Changing the encryption flag is not supported");
4112 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT_FORMAT
)) {
4113 encformat
= qcow2_crypt_method_from_format(
4114 qemu_opt_get(opts
, BLOCK_OPT_ENCRYPT_FORMAT
));
4116 if (encformat
!= s
->crypt_method_header
) {
4117 error_report("Changing the encryption format is not supported");
4120 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
4121 error_report("Changing the encryption parameters is not supported");
4123 } else if (!strcmp(desc
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
4124 cluster_size
= qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
,
4126 if (cluster_size
!= s
->cluster_size
) {
4127 error_report("Changing the cluster size is not supported");
4130 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
4131 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
4133 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
4134 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
4137 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
4138 !is_power_of_2(refcount_bits
))
4140 error_report("Refcount width must be a power of two and may "
4141 "not exceed 64 bits");
4145 /* if this point is reached, this probably means a new option was
4146 * added without having it covered here */
4153 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
4154 .original_status_cb
= status_cb
,
4155 .original_cb_opaque
= cb_opaque
,
4156 .total_operations
= (new_version
< old_version
)
4157 + (s
->refcount_bits
!= refcount_bits
)
4160 /* Upgrade first (some features may require compat=1.1) */
4161 if (new_version
> old_version
) {
4162 s
->qcow_version
= new_version
;
4163 ret
= qcow2_update_header(bs
);
4165 s
->qcow_version
= old_version
;
4170 if (s
->refcount_bits
!= refcount_bits
) {
4171 int refcount_order
= ctz32(refcount_bits
);
4173 if (new_version
< 3 && refcount_bits
!= 16) {
4174 error_report("Different refcount widths than 16 bits require "
4175 "compatibility level 1.1 or above (use compat=1.1 or "
4180 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
4181 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
4182 &qcow2_amend_helper_cb
,
4183 &helper_cb_info
, &local_err
);
4185 error_report_err(local_err
);
4190 if (backing_file
|| backing_format
) {
4191 ret
= qcow2_change_backing_file(bs
,
4192 backing_file
?: s
->image_backing_file
,
4193 backing_format
?: s
->image_backing_format
);
4199 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
4200 if (lazy_refcounts
) {
4201 if (new_version
< 3) {
4202 error_report("Lazy refcounts only supported with compatibility "
4203 "level 1.1 and above (use compat=1.1 or greater)");
4206 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
4207 ret
= qcow2_update_header(bs
);
4209 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
4212 s
->use_lazy_refcounts
= true;
4214 /* make image clean first */
4215 ret
= qcow2_mark_clean(bs
);
4219 /* now disallow lazy refcounts */
4220 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
4221 ret
= qcow2_update_header(bs
);
4223 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
4226 s
->use_lazy_refcounts
= false;
4231 BlockBackend
*blk
= blk_new(BLK_PERM_RESIZE
, BLK_PERM_ALL
);
4232 ret
= blk_insert_bs(blk
, bs
, &local_err
);
4234 error_report_err(local_err
);
4239 ret
= blk_truncate(blk
, new_size
, PREALLOC_MODE_OFF
, &local_err
);
4242 error_report_err(local_err
);
4247 /* Downgrade last (so unsupported features can be removed before) */
4248 if (new_version
< old_version
) {
4249 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
4250 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
4261 * If offset or size are negative, respectively, they will not be included in
4262 * the BLOCK_IMAGE_CORRUPTED event emitted.
4263 * fatal will be ignored for read-only BDS; corruptions found there will always
4264 * be considered non-fatal.
4266 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
4267 int64_t size
, const char *message_format
, ...)
4269 BDRVQcow2State
*s
= bs
->opaque
;
4270 const char *node_name
;
4274 fatal
= fatal
&& !bs
->read_only
;
4276 if (s
->signaled_corruption
&&
4277 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
4282 va_start(ap
, message_format
);
4283 message
= g_strdup_vprintf(message_format
, ap
);
4287 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
4288 "corruption events will be suppressed\n", message
);
4290 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
4291 "corruption events will be suppressed\n", message
);
4294 node_name
= bdrv_get_node_name(bs
);
4295 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
4296 *node_name
!= '\0', node_name
,
4297 message
, offset
>= 0, offset
,
4299 fatal
, &error_abort
);
4303 qcow2_mark_corrupt(bs
);
4304 bs
->drv
= NULL
; /* make BDS unusable */
4307 s
->signaled_corruption
= true;
4310 static QemuOptsList qcow2_create_opts
= {
4311 .name
= "qcow2-create-opts",
4312 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
4315 .name
= BLOCK_OPT_SIZE
,
4316 .type
= QEMU_OPT_SIZE
,
4317 .help
= "Virtual disk size"
4320 .name
= BLOCK_OPT_COMPAT_LEVEL
,
4321 .type
= QEMU_OPT_STRING
,
4322 .help
= "Compatibility level (0.10 or 1.1)"
4325 .name
= BLOCK_OPT_BACKING_FILE
,
4326 .type
= QEMU_OPT_STRING
,
4327 .help
= "File name of a base image"
4330 .name
= BLOCK_OPT_BACKING_FMT
,
4331 .type
= QEMU_OPT_STRING
,
4332 .help
= "Image format of the base image"
4335 .name
= BLOCK_OPT_ENCRYPT
,
4336 .type
= QEMU_OPT_BOOL
,
4337 .help
= "Encrypt the image with format 'aes'. (Deprecated "
4338 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)",
4341 .name
= BLOCK_OPT_ENCRYPT_FORMAT
,
4342 .type
= QEMU_OPT_STRING
,
4343 .help
= "Encrypt the image, format choices: 'aes', 'luks'",
4345 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
4346 "ID of secret providing qcow AES key or LUKS passphrase"),
4347 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
4348 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
4349 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
4350 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
4351 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
4352 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
4354 .name
= BLOCK_OPT_CLUSTER_SIZE
,
4355 .type
= QEMU_OPT_SIZE
,
4356 .help
= "qcow2 cluster size",
4357 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
)
4360 .name
= BLOCK_OPT_PREALLOC
,
4361 .type
= QEMU_OPT_STRING
,
4362 .help
= "Preallocation mode (allowed values: off, metadata, "
4366 .name
= BLOCK_OPT_LAZY_REFCOUNTS
,
4367 .type
= QEMU_OPT_BOOL
,
4368 .help
= "Postpone refcount updates",
4369 .def_value_str
= "off"
4372 .name
= BLOCK_OPT_REFCOUNT_BITS
,
4373 .type
= QEMU_OPT_NUMBER
,
4374 .help
= "Width of a reference count entry in bits",
4375 .def_value_str
= "16"
4377 { /* end of list */ }
4381 BlockDriver bdrv_qcow2
= {
4382 .format_name
= "qcow2",
4383 .instance_size
= sizeof(BDRVQcow2State
),
4384 .bdrv_probe
= qcow2_probe
,
4385 .bdrv_open
= qcow2_open
,
4386 .bdrv_close
= qcow2_close
,
4387 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
4388 .bdrv_reopen_commit
= qcow2_reopen_commit
,
4389 .bdrv_reopen_abort
= qcow2_reopen_abort
,
4390 .bdrv_join_options
= qcow2_join_options
,
4391 .bdrv_child_perm
= bdrv_format_default_perms
,
4392 .bdrv_co_create_opts
= qcow2_co_create_opts
,
4393 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
4394 .bdrv_co_block_status
= qcow2_co_block_status
,
4396 .bdrv_co_preadv
= qcow2_co_preadv
,
4397 .bdrv_co_pwritev
= qcow2_co_pwritev
,
4398 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
4400 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
4401 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
4402 .bdrv_truncate
= qcow2_truncate
,
4403 .bdrv_co_pwritev_compressed
= qcow2_co_pwritev_compressed
,
4404 .bdrv_make_empty
= qcow2_make_empty
,
4406 .bdrv_snapshot_create
= qcow2_snapshot_create
,
4407 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
4408 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
4409 .bdrv_snapshot_list
= qcow2_snapshot_list
,
4410 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
4411 .bdrv_measure
= qcow2_measure
,
4412 .bdrv_get_info
= qcow2_get_info
,
4413 .bdrv_get_specific_info
= qcow2_get_specific_info
,
4415 .bdrv_save_vmstate
= qcow2_save_vmstate
,
4416 .bdrv_load_vmstate
= qcow2_load_vmstate
,
4418 .supports_backing
= true,
4419 .bdrv_change_backing_file
= qcow2_change_backing_file
,
4421 .bdrv_refresh_limits
= qcow2_refresh_limits
,
4422 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
4423 .bdrv_inactivate
= qcow2_inactivate
,
4425 .create_opts
= &qcow2_create_opts
,
4426 .bdrv_co_check
= qcow2_co_check
,
4427 .bdrv_amend_options
= qcow2_amend_options
,
4429 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
4430 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
4432 .bdrv_reopen_bitmaps_rw
= qcow2_reopen_bitmaps_rw
,
4433 .bdrv_can_store_new_dirty_bitmap
= qcow2_can_store_new_dirty_bitmap
,
4434 .bdrv_remove_persistent_dirty_bitmap
= qcow2_remove_persistent_dirty_bitmap
,
4437 static void bdrv_qcow2_init(void)
4439 bdrv_register(&bdrv_qcow2
);
4442 block_init(bdrv_qcow2_init
);