2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
30 #include "block/block_int.h"
31 #include "block/qdict.h"
32 #include "sysemu/block-backend.h"
33 #include "qemu/module.h"
35 #include "qemu/error-report.h"
36 #include "qapi/error.h"
37 #include "qapi/qapi-events-block-core.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qstring.h"
41 #include "qemu/option_int.h"
42 #include "qemu/cutils.h"
43 #include "qemu/bswap.h"
44 #include "qapi/qobject-input-visitor.h"
45 #include "qapi/qapi-visit-block-core.h"
47 #include "block/thread-pool.h"
50 Differences with QCOW:
52 - Support for multiple incremental snapshots.
53 - Memory management by reference counts.
54 - Clusters which have a reference count of one have the bit
55 QCOW_OFLAG_COPIED to optimize write performance.
56 - Size of compressed clusters is stored in sectors to reduce bit usage
57 in the cluster offsets.
58 - Support for storing additional data (such as the VM state) in the
60 - If a backing store is used, the cluster size is not constrained
61 (could be backported to QCOW).
62 - L2 tables have always a size of one cluster.
69 } QEMU_PACKED QCowExtension
;
71 #define QCOW2_EXT_MAGIC_END 0
72 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
73 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
74 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
75 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
77 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
79 const QCowHeader
*cow_header
= (const void *)buf
;
81 if (buf_size
>= sizeof(QCowHeader
) &&
82 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
83 be32_to_cpu(cow_header
->version
) >= 2)
90 static ssize_t
qcow2_crypto_hdr_read_func(QCryptoBlock
*block
, size_t offset
,
91 uint8_t *buf
, size_t buflen
,
92 void *opaque
, Error
**errp
)
94 BlockDriverState
*bs
= opaque
;
95 BDRVQcow2State
*s
= bs
->opaque
;
98 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
99 error_setg(errp
, "Request for data outside of extension header");
103 ret
= bdrv_pread(bs
->file
,
104 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
106 error_setg_errno(errp
, -ret
, "Could not read encryption header");
113 static ssize_t
qcow2_crypto_hdr_init_func(QCryptoBlock
*block
, size_t headerlen
,
114 void *opaque
, Error
**errp
)
116 BlockDriverState
*bs
= opaque
;
117 BDRVQcow2State
*s
= bs
->opaque
;
121 ret
= qcow2_alloc_clusters(bs
, headerlen
);
123 error_setg_errno(errp
, -ret
,
124 "Cannot allocate cluster for LUKS header size %zu",
129 s
->crypto_header
.length
= headerlen
;
130 s
->crypto_header
.offset
= ret
;
132 /* Zero fill remaining space in cluster so it has predictable
133 * content in case of future spec changes */
134 clusterlen
= size_to_clusters(s
, headerlen
) * s
->cluster_size
;
135 assert(qcow2_pre_write_overlap_check(bs
, 0, ret
, clusterlen
) == 0);
136 ret
= bdrv_pwrite_zeroes(bs
->file
,
138 clusterlen
- headerlen
, 0);
140 error_setg_errno(errp
, -ret
, "Could not zero fill encryption header");
148 static ssize_t
qcow2_crypto_hdr_write_func(QCryptoBlock
*block
, size_t offset
,
149 const uint8_t *buf
, size_t buflen
,
150 void *opaque
, Error
**errp
)
152 BlockDriverState
*bs
= opaque
;
153 BDRVQcow2State
*s
= bs
->opaque
;
156 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
157 error_setg(errp
, "Request for data outside of extension header");
161 ret
= bdrv_pwrite(bs
->file
,
162 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
164 error_setg_errno(errp
, -ret
, "Could not read encryption header");
172 * read qcow2 extension and fill bs
173 * start reading from start_offset
174 * finish reading upon magic of value 0 or when end_offset reached
175 * unknown magic is skipped (future extension this version knows nothing about)
176 * return 0 upon success, non-0 otherwise
178 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
179 uint64_t end_offset
, void **p_feature_table
,
180 int flags
, bool *need_update_header
,
183 BDRVQcow2State
*s
= bs
->opaque
;
187 Qcow2BitmapHeaderExt bitmaps_ext
;
189 if (need_update_header
!= NULL
) {
190 *need_update_header
= false;
194 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
196 offset
= start_offset
;
197 while (offset
< end_offset
) {
201 if (offset
> s
->cluster_size
)
202 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
204 printf("attempting to read extended header in offset %lu\n", offset
);
207 ret
= bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
));
209 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
210 "pread fail from offset %" PRIu64
, offset
);
213 be32_to_cpus(&ext
.magic
);
214 be32_to_cpus(&ext
.len
);
215 offset
+= sizeof(ext
);
217 printf("ext.magic = 0x%x\n", ext
.magic
);
219 if (offset
> end_offset
|| ext
.len
> end_offset
- offset
) {
220 error_setg(errp
, "Header extension too large");
225 case QCOW2_EXT_MAGIC_END
:
228 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
229 if (ext
.len
>= sizeof(bs
->backing_format
)) {
230 error_setg(errp
, "ERROR: ext_backing_format: len=%" PRIu32
231 " too large (>=%zu)", ext
.len
,
232 sizeof(bs
->backing_format
));
235 ret
= bdrv_pread(bs
->file
, offset
, bs
->backing_format
, ext
.len
);
237 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
238 "Could not read format name");
241 bs
->backing_format
[ext
.len
] = '\0';
242 s
->image_backing_format
= g_strdup(bs
->backing_format
);
244 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
248 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
249 if (p_feature_table
!= NULL
) {
250 void* feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
251 ret
= bdrv_pread(bs
->file
, offset
, feature_table
, ext
.len
);
253 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
254 "Could not read table");
258 *p_feature_table
= feature_table
;
262 case QCOW2_EXT_MAGIC_CRYPTO_HEADER
: {
263 unsigned int cflags
= 0;
264 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
265 error_setg(errp
, "CRYPTO header extension only "
266 "expected with LUKS encryption method");
269 if (ext
.len
!= sizeof(Qcow2CryptoHeaderExtension
)) {
270 error_setg(errp
, "CRYPTO header extension size %u, "
271 "but expected size %zu", ext
.len
,
272 sizeof(Qcow2CryptoHeaderExtension
));
276 ret
= bdrv_pread(bs
->file
, offset
, &s
->crypto_header
, ext
.len
);
278 error_setg_errno(errp
, -ret
,
279 "Unable to read CRYPTO header extension");
282 be64_to_cpus(&s
->crypto_header
.offset
);
283 be64_to_cpus(&s
->crypto_header
.length
);
285 if ((s
->crypto_header
.offset
% s
->cluster_size
) != 0) {
286 error_setg(errp
, "Encryption header offset '%" PRIu64
"' is "
287 "not a multiple of cluster size '%u'",
288 s
->crypto_header
.offset
, s
->cluster_size
);
292 if (flags
& BDRV_O_NO_IO
) {
293 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
295 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
296 qcow2_crypto_hdr_read_func
,
303 case QCOW2_EXT_MAGIC_BITMAPS
:
304 if (ext
.len
!= sizeof(bitmaps_ext
)) {
305 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
306 "Invalid extension length");
310 if (!(s
->autoclear_features
& QCOW2_AUTOCLEAR_BITMAPS
)) {
311 if (s
->qcow_version
< 3) {
312 /* Let's be a bit more specific */
313 warn_report("This qcow2 v2 image contains bitmaps, but "
314 "they may have been modified by a program "
315 "without persistent bitmap support; so now "
316 "they must all be considered inconsistent");
318 warn_report("a program lacking bitmap support "
319 "modified this file, so all bitmaps are now "
320 "considered inconsistent");
322 error_printf("Some clusters may be leaked, "
323 "run 'qemu-img check -r' on the image "
325 if (need_update_header
!= NULL
) {
326 /* Updating is needed to drop invalid bitmap extension. */
327 *need_update_header
= true;
332 ret
= bdrv_pread(bs
->file
, offset
, &bitmaps_ext
, ext
.len
);
334 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
335 "Could not read ext header");
339 if (bitmaps_ext
.reserved32
!= 0) {
340 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
341 "Reserved field is not zero");
345 be32_to_cpus(&bitmaps_ext
.nb_bitmaps
);
346 be64_to_cpus(&bitmaps_ext
.bitmap_directory_size
);
347 be64_to_cpus(&bitmaps_ext
.bitmap_directory_offset
);
349 if (bitmaps_ext
.nb_bitmaps
> QCOW2_MAX_BITMAPS
) {
351 "bitmaps_ext: Image has %" PRIu32
" bitmaps, "
352 "exceeding the QEMU supported maximum of %d",
353 bitmaps_ext
.nb_bitmaps
, QCOW2_MAX_BITMAPS
);
357 if (bitmaps_ext
.nb_bitmaps
== 0) {
358 error_setg(errp
, "found bitmaps extension with zero bitmaps");
362 if (bitmaps_ext
.bitmap_directory_offset
& (s
->cluster_size
- 1)) {
363 error_setg(errp
, "bitmaps_ext: "
364 "invalid bitmap directory offset");
368 if (bitmaps_ext
.bitmap_directory_size
>
369 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
) {
370 error_setg(errp
, "bitmaps_ext: "
371 "bitmap directory size (%" PRIu64
") exceeds "
372 "the maximum supported size (%d)",
373 bitmaps_ext
.bitmap_directory_size
,
374 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
);
378 s
->nb_bitmaps
= bitmaps_ext
.nb_bitmaps
;
379 s
->bitmap_directory_offset
=
380 bitmaps_ext
.bitmap_directory_offset
;
381 s
->bitmap_directory_size
=
382 bitmaps_ext
.bitmap_directory_size
;
385 printf("Qcow2: Got bitmaps extension: "
386 "offset=%" PRIu64
" nb_bitmaps=%" PRIu32
"\n",
387 s
->bitmap_directory_offset
, s
->nb_bitmaps
);
392 /* unknown magic - save it in case we need to rewrite the header */
393 /* If you add a new feature, make sure to also update the fast
394 * path of qcow2_make_empty() to deal with it. */
396 Qcow2UnknownHeaderExtension
*uext
;
398 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
399 uext
->magic
= ext
.magic
;
401 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
403 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
405 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
406 "Could not read data");
413 offset
+= ((ext
.len
+ 7) & ~7);
419 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
421 BDRVQcow2State
*s
= bs
->opaque
;
422 Qcow2UnknownHeaderExtension
*uext
, *next
;
424 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
425 QLIST_REMOVE(uext
, next
);
430 static void report_unsupported_feature(Error
**errp
, Qcow2Feature
*table
,
433 char *features
= g_strdup("");
436 while (table
&& table
->name
[0] != '\0') {
437 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
438 if (mask
& (1ULL << table
->bit
)) {
440 features
= g_strdup_printf("%s%s%.46s", old
, *old
? ", " : "",
443 mask
&= ~(1ULL << table
->bit
);
451 features
= g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64
,
452 old
, *old
? ", " : "", mask
);
456 error_setg(errp
, "Unsupported qcow2 feature(s): %s", features
);
461 * Sets the dirty bit and flushes afterwards if necessary.
463 * The incompatible_features bit is only set if the image file header was
464 * updated successfully. Therefore it is not required to check the return
465 * value of this function.
467 int qcow2_mark_dirty(BlockDriverState
*bs
)
469 BDRVQcow2State
*s
= bs
->opaque
;
473 assert(s
->qcow_version
>= 3);
475 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
476 return 0; /* already dirty */
479 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
480 ret
= bdrv_pwrite(bs
->file
, offsetof(QCowHeader
, incompatible_features
),
485 ret
= bdrv_flush(bs
->file
->bs
);
490 /* Only treat image as dirty if the header was updated successfully */
491 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
496 * Clears the dirty bit and flushes before if necessary. Only call this
497 * function when there are no pending requests, it does not guard against
498 * concurrent requests dirtying the image.
500 static int qcow2_mark_clean(BlockDriverState
*bs
)
502 BDRVQcow2State
*s
= bs
->opaque
;
504 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
507 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
509 ret
= qcow2_flush_caches(bs
);
514 return qcow2_update_header(bs
);
520 * Marks the image as corrupt.
522 int qcow2_mark_corrupt(BlockDriverState
*bs
)
524 BDRVQcow2State
*s
= bs
->opaque
;
526 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
527 return qcow2_update_header(bs
);
531 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
532 * before if necessary.
534 int qcow2_mark_consistent(BlockDriverState
*bs
)
536 BDRVQcow2State
*s
= bs
->opaque
;
538 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
539 int ret
= qcow2_flush_caches(bs
);
544 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
545 return qcow2_update_header(bs
);
550 static int coroutine_fn
qcow2_co_check_locked(BlockDriverState
*bs
,
551 BdrvCheckResult
*result
,
554 int ret
= qcow2_check_refcounts(bs
, result
, fix
);
559 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
560 ret
= qcow2_mark_clean(bs
);
564 return qcow2_mark_consistent(bs
);
569 static int coroutine_fn
qcow2_co_check(BlockDriverState
*bs
,
570 BdrvCheckResult
*result
,
573 BDRVQcow2State
*s
= bs
->opaque
;
576 qemu_co_mutex_lock(&s
->lock
);
577 ret
= qcow2_co_check_locked(bs
, result
, fix
);
578 qemu_co_mutex_unlock(&s
->lock
);
582 int qcow2_validate_table(BlockDriverState
*bs
, uint64_t offset
,
583 uint64_t entries
, size_t entry_len
,
584 int64_t max_size_bytes
, const char *table_name
,
587 BDRVQcow2State
*s
= bs
->opaque
;
589 if (entries
> max_size_bytes
/ entry_len
) {
590 error_setg(errp
, "%s too large", table_name
);
594 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
595 * because values will be passed to qemu functions taking int64_t. */
596 if ((INT64_MAX
- entries
* entry_len
< offset
) ||
597 (offset_into_cluster(s
, offset
) != 0)) {
598 error_setg(errp
, "%s offset invalid", table_name
);
605 static QemuOptsList qcow2_runtime_opts
= {
607 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
610 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
611 .type
= QEMU_OPT_BOOL
,
612 .help
= "Postpone refcount updates",
615 .name
= QCOW2_OPT_DISCARD_REQUEST
,
616 .type
= QEMU_OPT_BOOL
,
617 .help
= "Pass guest discard requests to the layer below",
620 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
621 .type
= QEMU_OPT_BOOL
,
622 .help
= "Generate discard requests when snapshot related space "
626 .name
= QCOW2_OPT_DISCARD_OTHER
,
627 .type
= QEMU_OPT_BOOL
,
628 .help
= "Generate discard requests when other clusters are freed",
631 .name
= QCOW2_OPT_OVERLAP
,
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_TEMPLATE
,
638 .type
= QEMU_OPT_STRING
,
639 .help
= "Selects which overlap checks to perform from a range of "
640 "templates (none, constant, cached, all)",
643 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
644 .type
= QEMU_OPT_BOOL
,
645 .help
= "Check for unintended writes into the main qcow2 header",
648 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
649 .type
= QEMU_OPT_BOOL
,
650 .help
= "Check for unintended writes into the active L1 table",
653 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
654 .type
= QEMU_OPT_BOOL
,
655 .help
= "Check for unintended writes into an active L2 table",
658 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
659 .type
= QEMU_OPT_BOOL
,
660 .help
= "Check for unintended writes into the refcount table",
663 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
664 .type
= QEMU_OPT_BOOL
,
665 .help
= "Check for unintended writes into a refcount block",
668 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
669 .type
= QEMU_OPT_BOOL
,
670 .help
= "Check for unintended writes into the snapshot table",
673 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
674 .type
= QEMU_OPT_BOOL
,
675 .help
= "Check for unintended writes into an inactive L1 table",
678 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
679 .type
= QEMU_OPT_BOOL
,
680 .help
= "Check for unintended writes into an inactive L2 table",
683 .name
= QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
684 .type
= QEMU_OPT_BOOL
,
685 .help
= "Check for unintended writes into the bitmap directory",
688 .name
= QCOW2_OPT_CACHE_SIZE
,
689 .type
= QEMU_OPT_SIZE
,
690 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
694 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
695 .type
= QEMU_OPT_SIZE
,
696 .help
= "Maximum L2 table cache size",
699 .name
= QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
700 .type
= QEMU_OPT_SIZE
,
701 .help
= "Size of each entry in the L2 cache",
704 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
705 .type
= QEMU_OPT_SIZE
,
706 .help
= "Maximum refcount block cache size",
709 .name
= QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
710 .type
= QEMU_OPT_NUMBER
,
711 .help
= "Clean unused cache entries after this time (in seconds)",
713 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
714 "ID of secret providing qcow2 AES key or LUKS passphrase"),
715 { /* end of list */ }
719 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
720 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
721 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
722 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
723 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
724 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
725 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
726 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
727 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
728 [QCOW2_OL_BITMAP_DIRECTORY_BITNR
] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
731 static void cache_clean_timer_cb(void *opaque
)
733 BlockDriverState
*bs
= opaque
;
734 BDRVQcow2State
*s
= bs
->opaque
;
735 qcow2_cache_clean_unused(s
->l2_table_cache
);
736 qcow2_cache_clean_unused(s
->refcount_block_cache
);
737 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
738 (int64_t) s
->cache_clean_interval
* 1000);
741 static void cache_clean_timer_init(BlockDriverState
*bs
, AioContext
*context
)
743 BDRVQcow2State
*s
= bs
->opaque
;
744 if (s
->cache_clean_interval
> 0) {
745 s
->cache_clean_timer
= aio_timer_new(context
, QEMU_CLOCK_VIRTUAL
,
746 SCALE_MS
, cache_clean_timer_cb
,
748 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
749 (int64_t) s
->cache_clean_interval
* 1000);
753 static void cache_clean_timer_del(BlockDriverState
*bs
)
755 BDRVQcow2State
*s
= bs
->opaque
;
756 if (s
->cache_clean_timer
) {
757 timer_del(s
->cache_clean_timer
);
758 timer_free(s
->cache_clean_timer
);
759 s
->cache_clean_timer
= NULL
;
763 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
765 cache_clean_timer_del(bs
);
768 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
769 AioContext
*new_context
)
771 cache_clean_timer_init(bs
, new_context
);
774 static void read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
775 uint64_t *l2_cache_size
,
776 uint64_t *l2_cache_entry_size
,
777 uint64_t *refcount_cache_size
, Error
**errp
)
779 BDRVQcow2State
*s
= bs
->opaque
;
780 uint64_t combined_cache_size
, l2_cache_max_setting
;
781 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
782 int min_refcount_cache
= MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
783 uint64_t virtual_disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
784 uint64_t max_l2_cache
= virtual_disk_size
/ (s
->cluster_size
/ 8);
786 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
787 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
788 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
790 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
791 l2_cache_max_setting
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
,
792 DEFAULT_L2_CACHE_MAX_SIZE
);
793 *refcount_cache_size
= qemu_opt_get_size(opts
,
794 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
796 *l2_cache_entry_size
= qemu_opt_get_size(
797 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
799 *l2_cache_size
= MIN(max_l2_cache
, l2_cache_max_setting
);
801 if (combined_cache_size_set
) {
802 if (l2_cache_size_set
&& refcount_cache_size_set
) {
803 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
804 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
807 } else if (l2_cache_size_set
&&
808 (l2_cache_max_setting
> combined_cache_size
)) {
809 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
810 QCOW2_OPT_CACHE_SIZE
);
812 } else if (*refcount_cache_size
> combined_cache_size
) {
813 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
814 QCOW2_OPT_CACHE_SIZE
);
818 if (l2_cache_size_set
) {
819 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
820 } else if (refcount_cache_size_set
) {
821 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
823 /* Assign as much memory as possible to the L2 cache, and
824 * use the remainder for the refcount cache */
825 if (combined_cache_size
>= max_l2_cache
+ min_refcount_cache
) {
826 *l2_cache_size
= max_l2_cache
;
827 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
829 *refcount_cache_size
=
830 MIN(combined_cache_size
, min_refcount_cache
);
831 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
835 /* l2_cache_size and refcount_cache_size are ensured to have at least
836 * their minimum values in qcow2_update_options_prepare() */
838 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
839 *l2_cache_entry_size
> s
->cluster_size
||
840 !is_power_of_2(*l2_cache_entry_size
)) {
841 error_setg(errp
, "L2 cache entry size must be a power of two "
842 "between %d and the cluster size (%d)",
843 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
848 typedef struct Qcow2ReopenState
{
849 Qcow2Cache
*l2_table_cache
;
850 Qcow2Cache
*refcount_block_cache
;
851 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
852 bool use_lazy_refcounts
;
854 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
855 uint64_t cache_clean_interval
;
856 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
859 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
861 QDict
*options
, int flags
,
864 BDRVQcow2State
*s
= bs
->opaque
;
865 QemuOpts
*opts
= NULL
;
866 const char *opt_overlap_check
, *opt_overlap_check_template
;
867 int overlap_check_template
= 0;
868 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
870 const char *encryptfmt
;
871 QDict
*encryptopts
= NULL
;
872 Error
*local_err
= NULL
;
875 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
876 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
878 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
879 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
881 error_propagate(errp
, local_err
);
886 /* get L2 table/refcount block cache size from command line options */
887 read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
888 &refcount_cache_size
, &local_err
);
890 error_propagate(errp
, local_err
);
895 l2_cache_size
/= l2_cache_entry_size
;
896 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
897 l2_cache_size
= MIN_L2_CACHE_SIZE
;
899 if (l2_cache_size
> INT_MAX
) {
900 error_setg(errp
, "L2 cache size too big");
905 refcount_cache_size
/= s
->cluster_size
;
906 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
907 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
909 if (refcount_cache_size
> INT_MAX
) {
910 error_setg(errp
, "Refcount cache size too big");
915 /* alloc new L2 table/refcount block cache, flush old one */
916 if (s
->l2_table_cache
) {
917 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
919 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
924 if (s
->refcount_block_cache
) {
925 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
927 error_setg_errno(errp
, -ret
,
928 "Failed to flush the refcount block cache");
933 r
->l2_slice_size
= l2_cache_entry_size
/ sizeof(uint64_t);
934 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
935 l2_cache_entry_size
);
936 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
938 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
939 error_setg(errp
, "Could not allocate metadata caches");
944 /* New interval for cache cleanup timer */
945 r
->cache_clean_interval
=
946 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
947 DEFAULT_CACHE_CLEAN_INTERVAL
);
949 if (r
->cache_clean_interval
!= 0) {
950 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
951 " not supported on this host");
956 if (r
->cache_clean_interval
> UINT_MAX
) {
957 error_setg(errp
, "Cache clean interval too big");
962 /* lazy-refcounts; flush if going from enabled to disabled */
963 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
964 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
965 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
966 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
967 "qemu 1.1 compatibility level");
972 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
973 ret
= qcow2_mark_clean(bs
);
975 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
980 /* Overlap check options */
981 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
982 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
983 if (opt_overlap_check_template
&& opt_overlap_check
&&
984 strcmp(opt_overlap_check_template
, opt_overlap_check
))
986 error_setg(errp
, "Conflicting values for qcow2 options '"
987 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
988 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
992 if (!opt_overlap_check
) {
993 opt_overlap_check
= opt_overlap_check_template
?: "cached";
996 if (!strcmp(opt_overlap_check
, "none")) {
997 overlap_check_template
= 0;
998 } else if (!strcmp(opt_overlap_check
, "constant")) {
999 overlap_check_template
= QCOW2_OL_CONSTANT
;
1000 } else if (!strcmp(opt_overlap_check
, "cached")) {
1001 overlap_check_template
= QCOW2_OL_CACHED
;
1002 } else if (!strcmp(opt_overlap_check
, "all")) {
1003 overlap_check_template
= QCOW2_OL_ALL
;
1005 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1006 "'overlap-check'. Allowed are any of the following: "
1007 "none, constant, cached, all", opt_overlap_check
);
1012 r
->overlap_check
= 0;
1013 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1014 /* overlap-check defines a template bitmask, but every flag may be
1015 * overwritten through the associated boolean option */
1017 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1018 overlap_check_template
& (1 << i
)) << i
;
1021 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1022 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1023 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1024 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1025 flags
& BDRV_O_UNMAP
);
1026 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1027 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1028 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1029 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1031 switch (s
->crypt_method_header
) {
1032 case QCOW_CRYPT_NONE
:
1034 error_setg(errp
, "No encryption in image header, but options "
1035 "specified format '%s'", encryptfmt
);
1041 case QCOW_CRYPT_AES
:
1042 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1044 "Header reported 'aes' encryption format but "
1045 "options specify '%s'", encryptfmt
);
1049 qdict_put_str(encryptopts
, "format", "qcow");
1050 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1053 case QCOW_CRYPT_LUKS
:
1054 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1056 "Header reported 'luks' encryption format but "
1057 "options specify '%s'", encryptfmt
);
1061 qdict_put_str(encryptopts
, "format", "luks");
1062 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1066 error_setg(errp
, "Unsupported encryption method %d",
1067 s
->crypt_method_header
);
1070 if (s
->crypt_method_header
!= QCOW_CRYPT_NONE
&& !r
->crypto_opts
) {
1077 qobject_unref(encryptopts
);
1078 qemu_opts_del(opts
);
1083 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1084 Qcow2ReopenState
*r
)
1086 BDRVQcow2State
*s
= bs
->opaque
;
1089 if (s
->l2_table_cache
) {
1090 qcow2_cache_destroy(s
->l2_table_cache
);
1092 if (s
->refcount_block_cache
) {
1093 qcow2_cache_destroy(s
->refcount_block_cache
);
1095 s
->l2_table_cache
= r
->l2_table_cache
;
1096 s
->refcount_block_cache
= r
->refcount_block_cache
;
1097 s
->l2_slice_size
= r
->l2_slice_size
;
1099 s
->overlap_check
= r
->overlap_check
;
1100 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1102 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1103 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1106 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1107 cache_clean_timer_del(bs
);
1108 s
->cache_clean_interval
= r
->cache_clean_interval
;
1109 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1112 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1113 s
->crypto_opts
= r
->crypto_opts
;
1116 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1117 Qcow2ReopenState
*r
)
1119 if (r
->l2_table_cache
) {
1120 qcow2_cache_destroy(r
->l2_table_cache
);
1122 if (r
->refcount_block_cache
) {
1123 qcow2_cache_destroy(r
->refcount_block_cache
);
1125 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1128 static int qcow2_update_options(BlockDriverState
*bs
, QDict
*options
,
1129 int flags
, Error
**errp
)
1131 Qcow2ReopenState r
= {};
1134 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1136 qcow2_update_options_commit(bs
, &r
);
1138 qcow2_update_options_abort(bs
, &r
);
1144 /* Called with s->lock held. */
1145 static int coroutine_fn
qcow2_do_open(BlockDriverState
*bs
, QDict
*options
,
1146 int flags
, Error
**errp
)
1148 BDRVQcow2State
*s
= bs
->opaque
;
1149 unsigned int len
, i
;
1152 Error
*local_err
= NULL
;
1154 uint64_t l1_vm_state_index
;
1155 bool update_header
= false;
1156 bool header_updated
= false;
1158 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
1160 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1163 be32_to_cpus(&header
.magic
);
1164 be32_to_cpus(&header
.version
);
1165 be64_to_cpus(&header
.backing_file_offset
);
1166 be32_to_cpus(&header
.backing_file_size
);
1167 be64_to_cpus(&header
.size
);
1168 be32_to_cpus(&header
.cluster_bits
);
1169 be32_to_cpus(&header
.crypt_method
);
1170 be64_to_cpus(&header
.l1_table_offset
);
1171 be32_to_cpus(&header
.l1_size
);
1172 be64_to_cpus(&header
.refcount_table_offset
);
1173 be32_to_cpus(&header
.refcount_table_clusters
);
1174 be64_to_cpus(&header
.snapshots_offset
);
1175 be32_to_cpus(&header
.nb_snapshots
);
1177 if (header
.magic
!= QCOW_MAGIC
) {
1178 error_setg(errp
, "Image is not in qcow2 format");
1182 if (header
.version
< 2 || header
.version
> 3) {
1183 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1188 s
->qcow_version
= header
.version
;
1190 /* Initialise cluster size */
1191 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1192 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1193 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1194 header
.cluster_bits
);
1199 s
->cluster_bits
= header
.cluster_bits
;
1200 s
->cluster_size
= 1 << s
->cluster_bits
;
1201 s
->cluster_sectors
= 1 << (s
->cluster_bits
- BDRV_SECTOR_BITS
);
1203 /* Initialise version 3 header fields */
1204 if (header
.version
== 2) {
1205 header
.incompatible_features
= 0;
1206 header
.compatible_features
= 0;
1207 header
.autoclear_features
= 0;
1208 header
.refcount_order
= 4;
1209 header
.header_length
= 72;
1211 be64_to_cpus(&header
.incompatible_features
);
1212 be64_to_cpus(&header
.compatible_features
);
1213 be64_to_cpus(&header
.autoclear_features
);
1214 be32_to_cpus(&header
.refcount_order
);
1215 be32_to_cpus(&header
.header_length
);
1217 if (header
.header_length
< 104) {
1218 error_setg(errp
, "qcow2 header too short");
1224 if (header
.header_length
> s
->cluster_size
) {
1225 error_setg(errp
, "qcow2 header exceeds cluster size");
1230 if (header
.header_length
> sizeof(header
)) {
1231 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1232 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1233 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
1234 s
->unknown_header_fields_size
);
1236 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1242 if (header
.backing_file_offset
> s
->cluster_size
) {
1243 error_setg(errp
, "Invalid backing file offset");
1248 if (header
.backing_file_offset
) {
1249 ext_end
= header
.backing_file_offset
;
1251 ext_end
= 1 << header
.cluster_bits
;
1254 /* Handle feature bits */
1255 s
->incompatible_features
= header
.incompatible_features
;
1256 s
->compatible_features
= header
.compatible_features
;
1257 s
->autoclear_features
= header
.autoclear_features
;
1259 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1260 void *feature_table
= NULL
;
1261 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1262 &feature_table
, flags
, NULL
, NULL
);
1263 report_unsupported_feature(errp
, feature_table
,
1264 s
->incompatible_features
&
1265 ~QCOW2_INCOMPAT_MASK
);
1267 g_free(feature_table
);
1271 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1272 /* Corrupt images may not be written to unless they are being repaired
1274 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1275 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1282 /* Check support for various header values */
1283 if (header
.refcount_order
> 6) {
1284 error_setg(errp
, "Reference count entry width too large; may not "
1289 s
->refcount_order
= header
.refcount_order
;
1290 s
->refcount_bits
= 1 << s
->refcount_order
;
1291 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1292 s
->refcount_max
+= s
->refcount_max
- 1;
1294 s
->crypt_method_header
= header
.crypt_method
;
1295 if (s
->crypt_method_header
) {
1296 if (bdrv_uses_whitelist() &&
1297 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1299 "Use of AES-CBC encrypted qcow2 images is no longer "
1300 "supported in system emulators");
1301 error_append_hint(errp
,
1302 "You can use 'qemu-img convert' to convert your "
1303 "image to an alternative supported format, such "
1304 "as unencrypted qcow2, or raw with the LUKS "
1305 "format instead.\n");
1310 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1311 s
->crypt_physical_offset
= false;
1313 /* Assuming LUKS and any future crypt methods we
1314 * add will all use physical offsets, due to the
1315 * fact that the alternative is insecure... */
1316 s
->crypt_physical_offset
= true;
1319 bs
->encrypted
= true;
1322 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
1323 s
->l2_size
= 1 << s
->l2_bits
;
1324 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1325 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1326 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1327 bs
->total_sectors
= header
.size
/ BDRV_SECTOR_SIZE
;
1328 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1329 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1330 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1332 s
->refcount_table_offset
= header
.refcount_table_offset
;
1333 s
->refcount_table_size
=
1334 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1336 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1337 error_setg(errp
, "Image does not contain a reference count table");
1342 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1343 header
.refcount_table_clusters
,
1344 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1345 "Reference count table", errp
);
1350 /* The total size in bytes of the snapshot table is checked in
1351 * qcow2_read_snapshots() because the size of each snapshot is
1352 * variable and we don't know it yet.
1353 * Here we only check the offset and number of snapshots. */
1354 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1355 header
.nb_snapshots
,
1356 sizeof(QCowSnapshotHeader
),
1357 sizeof(QCowSnapshotHeader
) * QCOW_MAX_SNAPSHOTS
,
1358 "Snapshot table", errp
);
1363 /* read the level 1 table */
1364 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1365 header
.l1_size
, sizeof(uint64_t),
1366 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1370 s
->l1_size
= header
.l1_size
;
1371 s
->l1_table_offset
= header
.l1_table_offset
;
1373 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1374 if (l1_vm_state_index
> INT_MAX
) {
1375 error_setg(errp
, "Image is too big");
1379 s
->l1_vm_state_index
= l1_vm_state_index
;
1381 /* the L1 table must contain at least enough entries to put
1382 header.size bytes */
1383 if (s
->l1_size
< s
->l1_vm_state_index
) {
1384 error_setg(errp
, "L1 table is too small");
1389 if (s
->l1_size
> 0) {
1390 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
,
1391 ROUND_UP(s
->l1_size
* sizeof(uint64_t), 512));
1392 if (s
->l1_table
== NULL
) {
1393 error_setg(errp
, "Could not allocate L1 table");
1397 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1398 s
->l1_size
* sizeof(uint64_t));
1400 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1403 for(i
= 0;i
< s
->l1_size
; i
++) {
1404 be64_to_cpus(&s
->l1_table
[i
]);
1408 /* Parse driver-specific options */
1409 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1414 s
->cluster_cache_offset
= -1;
1417 ret
= qcow2_refcount_init(bs
);
1419 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1423 QLIST_INIT(&s
->cluster_allocs
);
1424 QTAILQ_INIT(&s
->discards
);
1426 /* read qcow2 extensions */
1427 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1428 flags
, &update_header
, &local_err
)) {
1429 error_propagate(errp
, local_err
);
1434 /* qcow2_read_extension may have set up the crypto context
1435 * if the crypt method needs a header region, some methods
1436 * don't need header extensions, so must check here
1438 if (s
->crypt_method_header
&& !s
->crypto
) {
1439 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1440 unsigned int cflags
= 0;
1441 if (flags
& BDRV_O_NO_IO
) {
1442 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1444 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1445 NULL
, NULL
, cflags
, errp
);
1450 } else if (!(flags
& BDRV_O_NO_IO
)) {
1451 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1452 s
->crypt_method_header
);
1458 /* read the backing file name */
1459 if (header
.backing_file_offset
!= 0) {
1460 len
= header
.backing_file_size
;
1461 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1462 len
>= sizeof(bs
->backing_file
)) {
1463 error_setg(errp
, "Backing file name too long");
1467 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
1468 bs
->backing_file
, len
);
1470 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1473 bs
->backing_file
[len
] = '\0';
1474 s
->image_backing_file
= g_strdup(bs
->backing_file
);
1477 /* Internal snapshots */
1478 s
->snapshots_offset
= header
.snapshots_offset
;
1479 s
->nb_snapshots
= header
.nb_snapshots
;
1481 ret
= qcow2_read_snapshots(bs
);
1483 error_setg_errno(errp
, -ret
, "Could not read snapshots");
1487 /* Clear unknown autoclear feature bits */
1488 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1490 update_header
&& !bs
->read_only
&& !(flags
& BDRV_O_INACTIVE
);
1491 if (update_header
) {
1492 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1495 if (s
->dirty_bitmaps_loaded
) {
1496 /* It's some kind of reopen. There are no known cases where we need to
1497 * reload bitmaps in such a situation, so it's safer to skip them.
1499 * Moreover, if we have some readonly bitmaps and we are reopening for
1500 * rw we should reopen bitmaps correspondingly.
1502 if (bdrv_has_readonly_bitmaps(bs
) &&
1503 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
))
1505 qcow2_reopen_bitmaps_rw_hint(bs
, &header_updated
, &local_err
);
1508 header_updated
= qcow2_load_dirty_bitmaps(bs
, &local_err
);
1509 s
->dirty_bitmaps_loaded
= true;
1511 update_header
= update_header
&& !header_updated
;
1512 if (local_err
!= NULL
) {
1513 error_propagate(errp
, local_err
);
1518 if (update_header
) {
1519 ret
= qcow2_update_header(bs
);
1521 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1526 bs
->supported_zero_flags
= header
.version
>= 3 ? BDRV_REQ_MAY_UNMAP
: 0;
1528 /* Repair image if dirty */
1529 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INACTIVE
)) && !bs
->read_only
&&
1530 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1531 BdrvCheckResult result
= {0};
1533 ret
= qcow2_co_check_locked(bs
, &result
,
1534 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1535 if (ret
< 0 || result
.check_errors
) {
1539 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1546 BdrvCheckResult result
= {0};
1547 qcow2_check_refcounts(bs
, &result
, 0);
1551 qemu_co_queue_init(&s
->compress_wait_queue
);
1556 g_free(s
->unknown_header_fields
);
1557 cleanup_unknown_header_ext(bs
);
1558 qcow2_free_snapshots(bs
);
1559 qcow2_refcount_close(bs
);
1560 qemu_vfree(s
->l1_table
);
1561 /* else pre-write overlap checks in cache_destroy may crash */
1563 cache_clean_timer_del(bs
);
1564 if (s
->l2_table_cache
) {
1565 qcow2_cache_destroy(s
->l2_table_cache
);
1567 if (s
->refcount_block_cache
) {
1568 qcow2_cache_destroy(s
->refcount_block_cache
);
1570 qcrypto_block_free(s
->crypto
);
1571 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1575 typedef struct QCow2OpenCo
{
1576 BlockDriverState
*bs
;
1583 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1585 QCow2OpenCo
*qoc
= opaque
;
1586 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1588 qemu_co_mutex_lock(&s
->lock
);
1589 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, qoc
->errp
);
1590 qemu_co_mutex_unlock(&s
->lock
);
1593 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1596 BDRVQcow2State
*s
= bs
->opaque
;
1605 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_file
,
1611 /* Initialise locks */
1612 qemu_co_mutex_init(&s
->lock
);
1614 if (qemu_in_coroutine()) {
1615 /* From bdrv_co_create. */
1616 qcow2_open_entry(&qoc
);
1618 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1619 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1624 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1626 BDRVQcow2State
*s
= bs
->opaque
;
1628 if (bs
->encrypted
) {
1629 /* Encryption works on a sector granularity */
1630 bs
->bl
.request_alignment
= BDRV_SECTOR_SIZE
;
1632 bs
->bl
.pwrite_zeroes_alignment
= s
->cluster_size
;
1633 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1636 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1637 BlockReopenQueue
*queue
, Error
**errp
)
1639 Qcow2ReopenState
*r
;
1642 r
= g_new0(Qcow2ReopenState
, 1);
1645 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1646 state
->flags
, errp
);
1651 /* We need to write out any unwritten data if we reopen read-only. */
1652 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1653 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1658 ret
= bdrv_flush(state
->bs
);
1663 ret
= qcow2_mark_clean(state
->bs
);
1672 qcow2_update_options_abort(state
->bs
, r
);
1677 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1679 qcow2_update_options_commit(state
->bs
, state
->opaque
);
1680 g_free(state
->opaque
);
1683 static void qcow2_reopen_abort(BDRVReopenState
*state
)
1685 qcow2_update_options_abort(state
->bs
, state
->opaque
);
1686 g_free(state
->opaque
);
1689 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
1691 bool has_new_overlap_template
=
1692 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
1693 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1694 bool has_new_total_cache_size
=
1695 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
1696 bool has_all_cache_options
;
1698 /* New overlap template overrides all old overlap options */
1699 if (has_new_overlap_template
) {
1700 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
1701 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1702 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
1703 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
1704 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
1705 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
1706 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
1707 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
1708 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
1709 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
1712 /* New total cache size overrides all old options */
1713 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
1714 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
1715 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1718 qdict_join(options
, old_options
, false);
1721 * If after merging all cache size options are set, an old total size is
1722 * overwritten. Do keep all options, however, if all three are new. The
1723 * resulting error message is what we want to happen.
1725 has_all_cache_options
=
1726 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
1727 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
1728 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1730 if (has_all_cache_options
&& !has_new_total_cache_size
) {
1731 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
1735 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
1737 int64_t offset
, int64_t count
,
1738 int64_t *pnum
, int64_t *map
,
1739 BlockDriverState
**file
)
1741 BDRVQcow2State
*s
= bs
->opaque
;
1742 uint64_t cluster_offset
;
1743 int index_in_cluster
, ret
;
1747 bytes
= MIN(INT_MAX
, count
);
1748 qemu_co_mutex_lock(&s
->lock
);
1749 ret
= qcow2_get_cluster_offset(bs
, offset
, &bytes
, &cluster_offset
);
1750 qemu_co_mutex_unlock(&s
->lock
);
1757 if (cluster_offset
!= 0 && ret
!= QCOW2_CLUSTER_COMPRESSED
&&
1759 index_in_cluster
= offset
& (s
->cluster_size
- 1);
1760 *map
= cluster_offset
| index_in_cluster
;
1761 *file
= bs
->file
->bs
;
1762 status
|= BDRV_BLOCK_OFFSET_VALID
;
1764 if (ret
== QCOW2_CLUSTER_ZERO_PLAIN
|| ret
== QCOW2_CLUSTER_ZERO_ALLOC
) {
1765 status
|= BDRV_BLOCK_ZERO
;
1766 } else if (ret
!= QCOW2_CLUSTER_UNALLOCATED
) {
1767 status
|= BDRV_BLOCK_DATA
;
1772 static coroutine_fn
int qcow2_handle_l2meta(BlockDriverState
*bs
,
1773 QCowL2Meta
**pl2meta
,
1777 QCowL2Meta
*l2meta
= *pl2meta
;
1779 while (l2meta
!= NULL
) {
1783 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
1788 qcow2_alloc_cluster_abort(bs
, l2meta
);
1791 /* Take the request off the list of running requests */
1792 if (l2meta
->nb_clusters
!= 0) {
1793 QLIST_REMOVE(l2meta
, next_in_flight
);
1796 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
1798 next
= l2meta
->next
;
1807 static coroutine_fn
int qcow2_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1808 uint64_t bytes
, QEMUIOVector
*qiov
,
1811 BDRVQcow2State
*s
= bs
->opaque
;
1812 int offset_in_cluster
;
1814 unsigned int cur_bytes
; /* number of bytes in current iteration */
1815 uint64_t cluster_offset
= 0;
1816 uint64_t bytes_done
= 0;
1817 QEMUIOVector hd_qiov
;
1818 uint8_t *cluster_data
= NULL
;
1820 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1822 qemu_co_mutex_lock(&s
->lock
);
1824 while (bytes
!= 0) {
1826 /* prepare next request */
1827 cur_bytes
= MIN(bytes
, INT_MAX
);
1829 cur_bytes
= MIN(cur_bytes
,
1830 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1833 ret
= qcow2_get_cluster_offset(bs
, offset
, &cur_bytes
, &cluster_offset
);
1838 offset_in_cluster
= offset_into_cluster(s
, offset
);
1840 qemu_iovec_reset(&hd_qiov
);
1841 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
1844 case QCOW2_CLUSTER_UNALLOCATED
:
1847 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
1848 qemu_co_mutex_unlock(&s
->lock
);
1849 ret
= bdrv_co_preadv(bs
->backing
, offset
, cur_bytes
,
1851 qemu_co_mutex_lock(&s
->lock
);
1856 /* Note: in this case, no need to wait */
1857 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1861 case QCOW2_CLUSTER_ZERO_PLAIN
:
1862 case QCOW2_CLUSTER_ZERO_ALLOC
:
1863 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1866 case QCOW2_CLUSTER_COMPRESSED
:
1867 /* add AIO support for compressed blocks ? */
1868 ret
= qcow2_decompress_cluster(bs
, cluster_offset
);
1873 qemu_iovec_from_buf(&hd_qiov
, 0,
1874 s
->cluster_cache
+ offset_in_cluster
,
1878 case QCOW2_CLUSTER_NORMAL
:
1879 if ((cluster_offset
& 511) != 0) {
1884 if (bs
->encrypted
) {
1888 * For encrypted images, read everything into a temporary
1889 * contiguous buffer on which the AES functions can work.
1891 if (!cluster_data
) {
1893 qemu_try_blockalign(bs
->file
->bs
,
1894 QCOW_MAX_CRYPT_CLUSTERS
1896 if (cluster_data
== NULL
) {
1902 assert(cur_bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1903 qemu_iovec_reset(&hd_qiov
);
1904 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
1907 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1908 qemu_co_mutex_unlock(&s
->lock
);
1909 ret
= bdrv_co_preadv(bs
->file
,
1910 cluster_offset
+ offset_in_cluster
,
1911 cur_bytes
, &hd_qiov
, 0);
1912 qemu_co_mutex_lock(&s
->lock
);
1916 if (bs
->encrypted
) {
1918 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1919 assert((cur_bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1920 if (qcrypto_block_decrypt(s
->crypto
,
1921 (s
->crypt_physical_offset
?
1922 cluster_offset
+ offset_in_cluster
:
1930 qemu_iovec_from_buf(qiov
, bytes_done
, cluster_data
, cur_bytes
);
1935 g_assert_not_reached();
1941 offset
+= cur_bytes
;
1942 bytes_done
+= cur_bytes
;
1947 qemu_co_mutex_unlock(&s
->lock
);
1949 qemu_iovec_destroy(&hd_qiov
);
1950 qemu_vfree(cluster_data
);
1955 /* Check if it's possible to merge a write request with the writing of
1956 * the data from the COW regions */
1957 static bool merge_cow(uint64_t offset
, unsigned bytes
,
1958 QEMUIOVector
*hd_qiov
, QCowL2Meta
*l2meta
)
1962 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
1963 /* If both COW regions are empty then there's nothing to merge */
1964 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
1968 /* The data (middle) region must be immediately after the
1970 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
1974 /* The end region must be immediately after the data (middle)
1976 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
1980 /* Make sure that adding both COW regions to the QEMUIOVector
1981 * does not exceed IOV_MAX */
1982 if (hd_qiov
->niov
> IOV_MAX
- 2) {
1986 m
->data_qiov
= hd_qiov
;
1993 static coroutine_fn
int qcow2_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1994 uint64_t bytes
, QEMUIOVector
*qiov
,
1997 BDRVQcow2State
*s
= bs
->opaque
;
1998 int offset_in_cluster
;
2000 unsigned int cur_bytes
; /* number of sectors in current iteration */
2001 uint64_t cluster_offset
;
2002 QEMUIOVector hd_qiov
;
2003 uint64_t bytes_done
= 0;
2004 uint8_t *cluster_data
= NULL
;
2005 QCowL2Meta
*l2meta
= NULL
;
2007 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
2009 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
2011 s
->cluster_cache_offset
= -1; /* disable compressed cache */
2013 qemu_co_mutex_lock(&s
->lock
);
2015 while (bytes
!= 0) {
2019 trace_qcow2_writev_start_part(qemu_coroutine_self());
2020 offset_in_cluster
= offset_into_cluster(s
, offset
);
2021 cur_bytes
= MIN(bytes
, INT_MAX
);
2022 if (bs
->encrypted
) {
2023 cur_bytes
= MIN(cur_bytes
,
2024 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
2025 - offset_in_cluster
);
2028 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2029 &cluster_offset
, &l2meta
);
2034 assert((cluster_offset
& 511) == 0);
2036 qemu_iovec_reset(&hd_qiov
);
2037 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
2039 if (bs
->encrypted
) {
2041 if (!cluster_data
) {
2042 cluster_data
= qemu_try_blockalign(bs
->file
->bs
,
2043 QCOW_MAX_CRYPT_CLUSTERS
2045 if (cluster_data
== NULL
) {
2051 assert(hd_qiov
.size
<=
2052 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2053 qemu_iovec_to_buf(&hd_qiov
, 0, cluster_data
, hd_qiov
.size
);
2055 if (qcrypto_block_encrypt(s
->crypto
,
2056 (s
->crypt_physical_offset
?
2057 cluster_offset
+ offset_in_cluster
:
2060 cur_bytes
, NULL
) < 0) {
2065 qemu_iovec_reset(&hd_qiov
);
2066 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
2069 ret
= qcow2_pre_write_overlap_check(bs
, 0,
2070 cluster_offset
+ offset_in_cluster
, cur_bytes
);
2075 /* If we need to do COW, check if it's possible to merge the
2076 * writing of the guest data together with that of the COW regions.
2077 * If it's not possible (or not necessary) then write the
2078 * guest data now. */
2079 if (!merge_cow(offset
, cur_bytes
, &hd_qiov
, l2meta
)) {
2080 qemu_co_mutex_unlock(&s
->lock
);
2081 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2082 trace_qcow2_writev_data(qemu_coroutine_self(),
2083 cluster_offset
+ offset_in_cluster
);
2084 ret
= bdrv_co_pwritev(bs
->file
,
2085 cluster_offset
+ offset_in_cluster
,
2086 cur_bytes
, &hd_qiov
, 0);
2087 qemu_co_mutex_lock(&s
->lock
);
2093 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
2099 offset
+= cur_bytes
;
2100 bytes_done
+= cur_bytes
;
2101 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2106 qcow2_handle_l2meta(bs
, &l2meta
, false);
2108 qemu_co_mutex_unlock(&s
->lock
);
2110 qemu_iovec_destroy(&hd_qiov
);
2111 qemu_vfree(cluster_data
);
2112 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2117 static int qcow2_inactivate(BlockDriverState
*bs
)
2119 BDRVQcow2State
*s
= bs
->opaque
;
2120 int ret
, result
= 0;
2121 Error
*local_err
= NULL
;
2123 qcow2_store_persistent_dirty_bitmaps(bs
, &local_err
);
2124 if (local_err
!= NULL
) {
2126 error_report_err(local_err
);
2127 error_report("Persistent bitmaps are lost for node '%s'",
2128 bdrv_get_device_or_node_name(bs
));
2131 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2134 error_report("Failed to flush the L2 table cache: %s",
2138 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2141 error_report("Failed to flush the refcount block cache: %s",
2146 qcow2_mark_clean(bs
);
2152 static void qcow2_close(BlockDriverState
*bs
)
2154 BDRVQcow2State
*s
= bs
->opaque
;
2155 qemu_vfree(s
->l1_table
);
2156 /* else pre-write overlap checks in cache_destroy may crash */
2159 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2160 qcow2_inactivate(bs
);
2163 cache_clean_timer_del(bs
);
2164 qcow2_cache_destroy(s
->l2_table_cache
);
2165 qcow2_cache_destroy(s
->refcount_block_cache
);
2167 qcrypto_block_free(s
->crypto
);
2170 g_free(s
->unknown_header_fields
);
2171 cleanup_unknown_header_ext(bs
);
2173 g_free(s
->image_backing_file
);
2174 g_free(s
->image_backing_format
);
2176 g_free(s
->cluster_cache
);
2177 qemu_vfree(s
->cluster_data
);
2178 qcow2_refcount_close(bs
);
2179 qcow2_free_snapshots(bs
);
2182 static void coroutine_fn
qcow2_co_invalidate_cache(BlockDriverState
*bs
,
2185 BDRVQcow2State
*s
= bs
->opaque
;
2186 int flags
= s
->flags
;
2187 QCryptoBlock
*crypto
= NULL
;
2189 Error
*local_err
= NULL
;
2193 * Backing files are read-only which makes all of their metadata immutable,
2194 * that means we don't have to worry about reopening them here.
2202 memset(s
, 0, sizeof(BDRVQcow2State
));
2203 options
= qdict_clone_shallow(bs
->options
);
2205 flags
&= ~BDRV_O_INACTIVE
;
2206 qemu_co_mutex_lock(&s
->lock
);
2207 ret
= qcow2_do_open(bs
, options
, flags
, &local_err
);
2208 qemu_co_mutex_unlock(&s
->lock
);
2209 qobject_unref(options
);
2211 error_propagate_prepend(errp
, local_err
,
2212 "Could not reopen qcow2 layer: ");
2215 } else if (ret
< 0) {
2216 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
2224 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2225 size_t len
, size_t buflen
)
2227 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2228 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2230 if (buflen
< ext_len
) {
2234 *ext_backing_fmt
= (QCowExtension
) {
2235 .magic
= cpu_to_be32(magic
),
2236 .len
= cpu_to_be32(len
),
2240 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2247 * Updates the qcow2 header, including the variable length parts of it, i.e.
2248 * the backing file name and all extensions. qcow2 was not designed to allow
2249 * such changes, so if we run out of space (we can only use the first cluster)
2250 * this function may fail.
2252 * Returns 0 on success, -errno in error cases.
2254 int qcow2_update_header(BlockDriverState
*bs
)
2256 BDRVQcow2State
*s
= bs
->opaque
;
2259 size_t buflen
= s
->cluster_size
;
2261 uint64_t total_size
;
2262 uint32_t refcount_table_clusters
;
2263 size_t header_length
;
2264 Qcow2UnknownHeaderExtension
*uext
;
2266 buf
= qemu_blockalign(bs
, buflen
);
2268 /* Header structure */
2269 header
= (QCowHeader
*) buf
;
2271 if (buflen
< sizeof(*header
)) {
2276 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2277 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2278 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2280 *header
= (QCowHeader
) {
2281 /* Version 2 fields */
2282 .magic
= cpu_to_be32(QCOW_MAGIC
),
2283 .version
= cpu_to_be32(s
->qcow_version
),
2284 .backing_file_offset
= 0,
2285 .backing_file_size
= 0,
2286 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2287 .size
= cpu_to_be64(total_size
),
2288 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2289 .l1_size
= cpu_to_be32(s
->l1_size
),
2290 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2291 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2292 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2293 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2294 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2296 /* Version 3 fields */
2297 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2298 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2299 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2300 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2301 .header_length
= cpu_to_be32(header_length
),
2304 /* For older versions, write a shorter header */
2305 switch (s
->qcow_version
) {
2307 ret
= offsetof(QCowHeader
, incompatible_features
);
2310 ret
= sizeof(*header
);
2319 memset(buf
, 0, buflen
);
2321 /* Preserve any unknown field in the header */
2322 if (s
->unknown_header_fields_size
) {
2323 if (buflen
< s
->unknown_header_fields_size
) {
2328 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2329 buf
+= s
->unknown_header_fields_size
;
2330 buflen
-= s
->unknown_header_fields_size
;
2333 /* Backing file format header extension */
2334 if (s
->image_backing_format
) {
2335 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2336 s
->image_backing_format
,
2337 strlen(s
->image_backing_format
),
2347 /* Full disk encryption header pointer extension */
2348 if (s
->crypto_header
.offset
!= 0) {
2349 cpu_to_be64s(&s
->crypto_header
.offset
);
2350 cpu_to_be64s(&s
->crypto_header
.length
);
2351 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2352 &s
->crypto_header
, sizeof(s
->crypto_header
),
2354 be64_to_cpus(&s
->crypto_header
.offset
);
2355 be64_to_cpus(&s
->crypto_header
.length
);
2364 if (s
->qcow_version
>= 3) {
2365 Qcow2Feature features
[] = {
2367 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2368 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2369 .name
= "dirty bit",
2372 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2373 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
2374 .name
= "corrupt bit",
2377 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
2378 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
2379 .name
= "lazy refcounts",
2383 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
2384 features
, sizeof(features
), buflen
);
2392 /* Bitmap extension */
2393 if (s
->nb_bitmaps
> 0) {
2394 Qcow2BitmapHeaderExt bitmaps_header
= {
2395 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
2396 .bitmap_directory_size
=
2397 cpu_to_be64(s
->bitmap_directory_size
),
2398 .bitmap_directory_offset
=
2399 cpu_to_be64(s
->bitmap_directory_offset
)
2401 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
2402 &bitmaps_header
, sizeof(bitmaps_header
),
2411 /* Keep unknown header extensions */
2412 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
2413 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
2422 /* End of header extensions */
2423 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
2431 /* Backing file name */
2432 if (s
->image_backing_file
) {
2433 size_t backing_file_len
= strlen(s
->image_backing_file
);
2435 if (buflen
< backing_file_len
) {
2440 /* Using strncpy is ok here, since buf is not NUL-terminated. */
2441 strncpy(buf
, s
->image_backing_file
, buflen
);
2443 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
2444 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
2447 /* Write the new header */
2448 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
2459 static int qcow2_change_backing_file(BlockDriverState
*bs
,
2460 const char *backing_file
, const char *backing_fmt
)
2462 BDRVQcow2State
*s
= bs
->opaque
;
2464 if (backing_file
&& strlen(backing_file
) > 1023) {
2468 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
2469 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
2471 g_free(s
->image_backing_file
);
2472 g_free(s
->image_backing_format
);
2474 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
2475 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
2477 return qcow2_update_header(bs
);
2480 static int qcow2_crypt_method_from_format(const char *encryptfmt
)
2482 if (g_str_equal(encryptfmt
, "luks")) {
2483 return QCOW_CRYPT_LUKS
;
2484 } else if (g_str_equal(encryptfmt
, "aes")) {
2485 return QCOW_CRYPT_AES
;
2491 static int qcow2_set_up_encryption(BlockDriverState
*bs
,
2492 QCryptoBlockCreateOptions
*cryptoopts
,
2495 BDRVQcow2State
*s
= bs
->opaque
;
2496 QCryptoBlock
*crypto
= NULL
;
2499 switch (cryptoopts
->format
) {
2500 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
2501 fmt
= QCOW_CRYPT_LUKS
;
2503 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
2504 fmt
= QCOW_CRYPT_AES
;
2507 error_setg(errp
, "Crypto format not supported in qcow2");
2511 s
->crypt_method_header
= fmt
;
2513 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
2514 qcow2_crypto_hdr_init_func
,
2515 qcow2_crypto_hdr_write_func
,
2521 ret
= qcow2_update_header(bs
);
2523 error_setg_errno(errp
, -ret
, "Could not write encryption header");
2529 qcrypto_block_free(crypto
);
2534 * Preallocates metadata structures for data clusters between @offset (in the
2535 * guest disk) and @new_length (which is thus generally the new guest disk
2538 * Returns: 0 on success, -errno on failure.
2540 static int coroutine_fn
preallocate_co(BlockDriverState
*bs
, uint64_t offset
,
2541 uint64_t new_length
)
2544 uint64_t host_offset
= 0;
2545 unsigned int cur_bytes
;
2549 assert(offset
<= new_length
);
2550 bytes
= new_length
- offset
;
2553 cur_bytes
= MIN(bytes
, INT_MAX
);
2554 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2555 &host_offset
, &meta
);
2561 QCowL2Meta
*next
= meta
->next
;
2563 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
2565 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
2566 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
2570 /* There are no dependent requests, but we need to remove our
2571 * request from the list of in-flight requests */
2572 QLIST_REMOVE(meta
, next_in_flight
);
2578 /* TODO Preallocate data if requested */
2581 offset
+= cur_bytes
;
2585 * It is expected that the image file is large enough to actually contain
2586 * all of the allocated clusters (otherwise we get failing reads after
2587 * EOF). Extend the image to the last allocated sector.
2589 if (host_offset
!= 0) {
2591 ret
= bdrv_pwrite(bs
->file
, (host_offset
+ cur_bytes
) - 1,
2601 /* qcow2_refcount_metadata_size:
2602 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2603 * @cluster_size: size of a cluster, in bytes
2604 * @refcount_order: refcount bits power-of-2 exponent
2605 * @generous_increase: allow for the refcount table to be 1.5x as large as it
2608 * Returns: Number of bytes required for refcount blocks and table metadata.
2610 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
2611 int refcount_order
, bool generous_increase
,
2612 uint64_t *refblock_count
)
2615 * Every host cluster is reference-counted, including metadata (even
2616 * refcount metadata is recursively included).
2618 * An accurate formula for the size of refcount metadata size is difficult
2619 * to derive. An easier method of calculation is finding the fixed point
2620 * where no further refcount blocks or table clusters are required to
2621 * reference count every cluster.
2623 int64_t blocks_per_table_cluster
= cluster_size
/ sizeof(uint64_t);
2624 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
2625 int64_t table
= 0; /* number of refcount table clusters */
2626 int64_t blocks
= 0; /* number of refcount block clusters */
2632 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
2633 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
2634 n
= clusters
+ blocks
+ table
;
2636 if (n
== last
&& generous_increase
) {
2637 clusters
+= DIV_ROUND_UP(table
, 2);
2638 n
= 0; /* force another loop */
2639 generous_increase
= false;
2641 } while (n
!= last
);
2643 if (refblock_count
) {
2644 *refblock_count
= blocks
;
2647 return (blocks
+ table
) * cluster_size
;
2651 * qcow2_calc_prealloc_size:
2652 * @total_size: virtual disk size in bytes
2653 * @cluster_size: cluster size in bytes
2654 * @refcount_order: refcount bits power-of-2 exponent
2656 * Returns: Total number of bytes required for the fully allocated image
2657 * (including metadata).
2659 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
2660 size_t cluster_size
,
2663 int64_t meta_size
= 0;
2664 uint64_t nl1e
, nl2e
;
2665 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
2667 /* header: 1 cluster */
2668 meta_size
+= cluster_size
;
2670 /* total size of L2 tables */
2671 nl2e
= aligned_total_size
/ cluster_size
;
2672 nl2e
= ROUND_UP(nl2e
, cluster_size
/ sizeof(uint64_t));
2673 meta_size
+= nl2e
* sizeof(uint64_t);
2675 /* total size of L1 tables */
2676 nl1e
= nl2e
* sizeof(uint64_t) / cluster_size
;
2677 nl1e
= ROUND_UP(nl1e
, cluster_size
/ sizeof(uint64_t));
2678 meta_size
+= nl1e
* sizeof(uint64_t);
2680 /* total size of refcount table and blocks */
2681 meta_size
+= qcow2_refcount_metadata_size(
2682 (meta_size
+ aligned_total_size
) / cluster_size
,
2683 cluster_size
, refcount_order
, false, NULL
);
2685 return meta_size
+ aligned_total_size
;
2688 static bool validate_cluster_size(size_t cluster_size
, Error
**errp
)
2690 int cluster_bits
= ctz32(cluster_size
);
2691 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
2692 (1 << cluster_bits
) != cluster_size
)
2694 error_setg(errp
, "Cluster size must be a power of two between %d and "
2695 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
2701 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, Error
**errp
)
2703 size_t cluster_size
;
2705 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
2706 DEFAULT_CLUSTER_SIZE
);
2707 if (!validate_cluster_size(cluster_size
, errp
)) {
2710 return cluster_size
;
2713 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
2718 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
2720 ret
= 3; /* default */
2721 } else if (!strcmp(buf
, "0.10")) {
2723 } else if (!strcmp(buf
, "1.1")) {
2726 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
2733 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
2736 uint64_t refcount_bits
;
2738 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
2739 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
2740 error_setg(errp
, "Refcount width must be a power of two and may not "
2745 if (version
< 3 && refcount_bits
!= 16) {
2746 error_setg(errp
, "Different refcount widths than 16 bits require "
2747 "compatibility level 1.1 or above (use compat=1.1 or "
2752 return refcount_bits
;
2755 static int coroutine_fn
2756 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
2758 BlockdevCreateOptionsQcow2
*qcow2_opts
;
2762 * Open the image file and write a minimal qcow2 header.
2764 * We keep things simple and start with a zero-sized image. We also
2765 * do without refcount blocks or a L1 table for now. We'll fix the
2766 * inconsistency later.
2768 * We do need a refcount table because growing the refcount table means
2769 * allocating two new refcount blocks - the seconds of which would be at
2770 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2771 * size for any qcow2 image.
2773 BlockBackend
*blk
= NULL
;
2774 BlockDriverState
*bs
= NULL
;
2776 size_t cluster_size
;
2779 uint64_t* refcount_table
;
2780 Error
*local_err
= NULL
;
2783 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
2784 qcow2_opts
= &create_options
->u
.qcow2
;
2786 bs
= bdrv_open_blockdev_ref(qcow2_opts
->file
, errp
);
2791 /* Validate options and set default values */
2792 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
2793 error_setg(errp
, "Image size must be a multiple of 512 bytes");
2798 if (qcow2_opts
->has_version
) {
2799 switch (qcow2_opts
->version
) {
2800 case BLOCKDEV_QCOW2_VERSION_V2
:
2803 case BLOCKDEV_QCOW2_VERSION_V3
:
2807 g_assert_not_reached();
2813 if (qcow2_opts
->has_cluster_size
) {
2814 cluster_size
= qcow2_opts
->cluster_size
;
2816 cluster_size
= DEFAULT_CLUSTER_SIZE
;
2819 if (!validate_cluster_size(cluster_size
, errp
)) {
2824 if (!qcow2_opts
->has_preallocation
) {
2825 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
2827 if (qcow2_opts
->has_backing_file
&&
2828 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
)
2830 error_setg(errp
, "Backing file and preallocation cannot be used at "
2835 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->has_backing_file
) {
2836 error_setg(errp
, "Backing format cannot be used without backing file");
2841 if (!qcow2_opts
->has_lazy_refcounts
) {
2842 qcow2_opts
->lazy_refcounts
= false;
2844 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
2845 error_setg(errp
, "Lazy refcounts only supported with compatibility "
2846 "level 1.1 and above (use version=v3 or greater)");
2851 if (!qcow2_opts
->has_refcount_bits
) {
2852 qcow2_opts
->refcount_bits
= 16;
2854 if (qcow2_opts
->refcount_bits
> 64 ||
2855 !is_power_of_2(qcow2_opts
->refcount_bits
))
2857 error_setg(errp
, "Refcount width must be a power of two and may not "
2862 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
2863 error_setg(errp
, "Different refcount widths than 16 bits require "
2864 "compatibility level 1.1 or above (use version=v3 or "
2869 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
2872 /* Create BlockBackend to write to the image */
2873 blk
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
);
2874 ret
= blk_insert_bs(blk
, bs
, errp
);
2878 blk_set_allow_write_beyond_eof(blk
, true);
2880 /* Clear the protocol layer and preallocate it if necessary */
2881 ret
= blk_truncate(blk
, 0, PREALLOC_MODE_OFF
, errp
);
2886 if (qcow2_opts
->preallocation
== PREALLOC_MODE_FULL
||
2887 qcow2_opts
->preallocation
== PREALLOC_MODE_FALLOC
)
2889 int64_t prealloc_size
=
2890 qcow2_calc_prealloc_size(qcow2_opts
->size
, cluster_size
,
2893 ret
= blk_truncate(blk
, prealloc_size
, qcow2_opts
->preallocation
, errp
);
2899 /* Write the header */
2900 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
2901 header
= g_malloc0(cluster_size
);
2902 *header
= (QCowHeader
) {
2903 .magic
= cpu_to_be32(QCOW_MAGIC
),
2904 .version
= cpu_to_be32(version
),
2905 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
2906 .size
= cpu_to_be64(0),
2907 .l1_table_offset
= cpu_to_be64(0),
2908 .l1_size
= cpu_to_be32(0),
2909 .refcount_table_offset
= cpu_to_be64(cluster_size
),
2910 .refcount_table_clusters
= cpu_to_be32(1),
2911 .refcount_order
= cpu_to_be32(refcount_order
),
2912 .header_length
= cpu_to_be32(sizeof(*header
)),
2915 /* We'll update this to correct value later */
2916 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
2918 if (qcow2_opts
->lazy_refcounts
) {
2919 header
->compatible_features
|=
2920 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
2923 ret
= blk_pwrite(blk
, 0, header
, cluster_size
, 0);
2926 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
2930 /* Write a refcount table with one refcount block */
2931 refcount_table
= g_malloc0(2 * cluster_size
);
2932 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
2933 ret
= blk_pwrite(blk
, cluster_size
, refcount_table
, 2 * cluster_size
, 0);
2934 g_free(refcount_table
);
2937 error_setg_errno(errp
, -ret
, "Could not write refcount table");
2945 * And now open the image and make it consistent first (i.e. increase the
2946 * refcount of the cluster that is occupied by the header and the refcount
2949 options
= qdict_new();
2950 qdict_put_str(options
, "driver", "qcow2");
2951 qdict_put_str(options
, "file", bs
->node_name
);
2952 blk
= blk_new_open(NULL
, NULL
, options
,
2953 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
2956 error_propagate(errp
, local_err
);
2961 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
2963 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
2964 "header and refcount table");
2967 } else if (ret
!= 0) {
2968 error_report("Huh, first cluster in empty image is already in use?");
2972 /* Create a full header (including things like feature table) */
2973 ret
= qcow2_update_header(blk_bs(blk
));
2975 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
2979 /* Okay, now that we have a valid image, let's give it the right size */
2980 ret
= blk_truncate(blk
, qcow2_opts
->size
, PREALLOC_MODE_OFF
, errp
);
2982 error_prepend(errp
, "Could not resize image: ");
2986 /* Want a backing file? There you go.*/
2987 if (qcow2_opts
->has_backing_file
) {
2988 const char *backing_format
= NULL
;
2990 if (qcow2_opts
->has_backing_fmt
) {
2991 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
2994 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
2997 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
2998 "with format '%s'", qcow2_opts
->backing_file
,
3004 /* Want encryption? There you go. */
3005 if (qcow2_opts
->has_encrypt
) {
3006 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3012 /* And if we're supposed to preallocate metadata, do that now */
3013 if (qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
) {
3014 BDRVQcow2State
*s
= blk_bs(blk
)->opaque
;
3015 qemu_co_mutex_lock(&s
->lock
);
3016 ret
= preallocate_co(blk_bs(blk
), 0, qcow2_opts
->size
);
3017 qemu_co_mutex_unlock(&s
->lock
);
3020 error_setg_errno(errp
, -ret
, "Could not preallocate metadata");
3028 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3029 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3030 * have to setup decryption context. We're not doing any I/O on the top
3031 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3034 options
= qdict_new();
3035 qdict_put_str(options
, "driver", "qcow2");
3036 qdict_put_str(options
, "file", bs
->node_name
);
3037 blk
= blk_new_open(NULL
, NULL
, options
,
3038 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3041 error_propagate(errp
, local_err
);
3053 static int coroutine_fn
qcow2_co_create_opts(const char *filename
, QemuOpts
*opts
,
3056 BlockdevCreateOptions
*create_options
= NULL
;
3059 BlockDriverState
*bs
= NULL
;
3060 Error
*local_err
= NULL
;
3064 /* Only the keyval visitor supports the dotted syntax needed for
3065 * encryption, so go through a QDict before getting a QAPI type. Ignore
3066 * options meant for the protocol layer so that the visitor doesn't
3068 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3071 /* Handle encryption options */
3072 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3073 if (val
&& !strcmp(val
, "on")) {
3074 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3075 } else if (val
&& !strcmp(val
, "off")) {
3076 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3079 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3080 if (val
&& !strcmp(val
, "aes")) {
3081 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3084 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3085 * version=v2/v3 below. */
3086 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3087 if (val
&& !strcmp(val
, "0.10")) {
3088 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3089 } else if (val
&& !strcmp(val
, "1.1")) {
3090 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3093 /* Change legacy command line options into QMP ones */
3094 static const QDictRenames opt_renames
[] = {
3095 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3096 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3097 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3098 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3099 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3100 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3101 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3105 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3110 /* Create and open the file (protocol layer) */
3111 ret
= bdrv_create_file(filename
, opts
, errp
);
3116 bs
= bdrv_open(filename
, NULL
, NULL
,
3117 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3123 /* Set 'driver' and 'node' options */
3124 qdict_put_str(qdict
, "driver", "qcow2");
3125 qdict_put_str(qdict
, "file", bs
->node_name
);
3127 /* Now get the QAPI type BlockdevCreateOptions */
3128 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
3134 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, &local_err
);
3138 error_propagate(errp
, local_err
);
3143 /* Silently round up size */
3144 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3147 /* Create the qcow2 image (format layer) */
3148 ret
= qcow2_co_create(create_options
, errp
);
3155 qobject_unref(qdict
);
3157 qapi_free_BlockdevCreateOptions(create_options
);
3162 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3167 /* Clamp to image length, before checking status of underlying sectors */
3168 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3169 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3175 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3176 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
== bytes
;
3179 static coroutine_fn
int qcow2_co_pwrite_zeroes(BlockDriverState
*bs
,
3180 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3183 BDRVQcow2State
*s
= bs
->opaque
;
3185 uint32_t head
= offset
% s
->cluster_size
;
3186 uint32_t tail
= (offset
+ bytes
) % s
->cluster_size
;
3188 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3189 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3197 assert(head
+ bytes
<= s
->cluster_size
);
3199 /* check whether remainder of cluster already reads as zero */
3200 if (!(is_zero(bs
, offset
- head
, head
) &&
3201 is_zero(bs
, offset
+ bytes
,
3202 tail
? s
->cluster_size
- tail
: 0))) {
3206 qemu_co_mutex_lock(&s
->lock
);
3207 /* We can have new write after previous check */
3208 offset
= QEMU_ALIGN_DOWN(offset
, s
->cluster_size
);
3209 bytes
= s
->cluster_size
;
3210 nr
= s
->cluster_size
;
3211 ret
= qcow2_get_cluster_offset(bs
, offset
, &nr
, &off
);
3212 if (ret
!= QCOW2_CLUSTER_UNALLOCATED
&&
3213 ret
!= QCOW2_CLUSTER_ZERO_PLAIN
&&
3214 ret
!= QCOW2_CLUSTER_ZERO_ALLOC
) {
3215 qemu_co_mutex_unlock(&s
->lock
);
3219 qemu_co_mutex_lock(&s
->lock
);
3222 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
3224 /* Whatever is left can use real zero clusters */
3225 ret
= qcow2_cluster_zeroize(bs
, offset
, bytes
, flags
);
3226 qemu_co_mutex_unlock(&s
->lock
);
3231 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
3232 int64_t offset
, int bytes
)
3235 BDRVQcow2State
*s
= bs
->opaque
;
3237 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
3238 assert(bytes
< s
->cluster_size
);
3239 /* Ignore partial clusters, except for the special case of the
3240 * complete partial cluster at the end of an unaligned file */
3241 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
3242 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3247 qemu_co_mutex_lock(&s
->lock
);
3248 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
3250 qemu_co_mutex_unlock(&s
->lock
);
3254 static int coroutine_fn
3255 qcow2_co_copy_range_from(BlockDriverState
*bs
,
3256 BdrvChild
*src
, uint64_t src_offset
,
3257 BdrvChild
*dst
, uint64_t dst_offset
,
3258 uint64_t bytes
, BdrvRequestFlags read_flags
,
3259 BdrvRequestFlags write_flags
)
3261 BDRVQcow2State
*s
= bs
->opaque
;
3263 unsigned int cur_bytes
; /* number of bytes in current iteration */
3264 BdrvChild
*child
= NULL
;
3265 BdrvRequestFlags cur_write_flags
;
3267 assert(!bs
->encrypted
);
3268 qemu_co_mutex_lock(&s
->lock
);
3270 while (bytes
!= 0) {
3271 uint64_t copy_offset
= 0;
3272 /* prepare next request */
3273 cur_bytes
= MIN(bytes
, INT_MAX
);
3274 cur_write_flags
= write_flags
;
3276 ret
= qcow2_get_cluster_offset(bs
, src_offset
, &cur_bytes
, ©_offset
);
3282 case QCOW2_CLUSTER_UNALLOCATED
:
3283 if (bs
->backing
&& bs
->backing
->bs
) {
3284 int64_t backing_length
= bdrv_getlength(bs
->backing
->bs
);
3285 if (src_offset
>= backing_length
) {
3286 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3288 child
= bs
->backing
;
3289 cur_bytes
= MIN(cur_bytes
, backing_length
- src_offset
);
3290 copy_offset
= src_offset
;
3293 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3297 case QCOW2_CLUSTER_ZERO_PLAIN
:
3298 case QCOW2_CLUSTER_ZERO_ALLOC
:
3299 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3302 case QCOW2_CLUSTER_COMPRESSED
:
3306 case QCOW2_CLUSTER_NORMAL
:
3308 copy_offset
+= offset_into_cluster(s
, src_offset
);
3309 if ((copy_offset
& 511) != 0) {
3318 qemu_co_mutex_unlock(&s
->lock
);
3319 ret
= bdrv_co_copy_range_from(child
,
3322 cur_bytes
, read_flags
, cur_write_flags
);
3323 qemu_co_mutex_lock(&s
->lock
);
3329 src_offset
+= cur_bytes
;
3330 dst_offset
+= cur_bytes
;
3335 qemu_co_mutex_unlock(&s
->lock
);
3339 static int coroutine_fn
3340 qcow2_co_copy_range_to(BlockDriverState
*bs
,
3341 BdrvChild
*src
, uint64_t src_offset
,
3342 BdrvChild
*dst
, uint64_t dst_offset
,
3343 uint64_t bytes
, BdrvRequestFlags read_flags
,
3344 BdrvRequestFlags write_flags
)
3346 BDRVQcow2State
*s
= bs
->opaque
;
3347 int offset_in_cluster
;
3349 unsigned int cur_bytes
; /* number of sectors in current iteration */
3350 uint64_t cluster_offset
;
3351 QCowL2Meta
*l2meta
= NULL
;
3353 assert(!bs
->encrypted
);
3354 s
->cluster_cache_offset
= -1; /* disable compressed cache */
3356 qemu_co_mutex_lock(&s
->lock
);
3358 while (bytes
!= 0) {
3362 offset_in_cluster
= offset_into_cluster(s
, dst_offset
);
3363 cur_bytes
= MIN(bytes
, INT_MAX
);
3366 * If src->bs == dst->bs, we could simply copy by incrementing
3367 * the refcnt, without copying user data.
3368 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
3369 ret
= qcow2_alloc_cluster_offset(bs
, dst_offset
, &cur_bytes
,
3370 &cluster_offset
, &l2meta
);
3375 assert((cluster_offset
& 511) == 0);
3377 ret
= qcow2_pre_write_overlap_check(bs
, 0,
3378 cluster_offset
+ offset_in_cluster
, cur_bytes
);
3383 qemu_co_mutex_unlock(&s
->lock
);
3384 ret
= bdrv_co_copy_range_to(src
, src_offset
,
3386 cluster_offset
+ offset_in_cluster
,
3387 cur_bytes
, read_flags
, write_flags
);
3388 qemu_co_mutex_lock(&s
->lock
);
3393 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
3399 src_offset
+= cur_bytes
;
3400 dst_offset
+= cur_bytes
;
3405 qcow2_handle_l2meta(bs
, &l2meta
, false);
3407 qemu_co_mutex_unlock(&s
->lock
);
3409 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
3414 static int coroutine_fn
qcow2_co_truncate(BlockDriverState
*bs
, int64_t offset
,
3415 PreallocMode prealloc
, Error
**errp
)
3417 BDRVQcow2State
*s
= bs
->opaque
;
3418 uint64_t old_length
;
3419 int64_t new_l1_size
;
3423 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
3424 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
3426 error_setg(errp
, "Unsupported preallocation mode '%s'",
3427 PreallocMode_str(prealloc
));
3432 error_setg(errp
, "The new size must be a multiple of 512");
3436 qemu_co_mutex_lock(&s
->lock
);
3438 /* cannot proceed if image has snapshots */
3439 if (s
->nb_snapshots
) {
3440 error_setg(errp
, "Can't resize an image which has snapshots");
3445 /* cannot proceed if image has bitmaps */
3446 if (s
->nb_bitmaps
) {
3447 /* TODO: resize bitmaps in the image */
3448 error_setg(errp
, "Can't resize an image which has bitmaps");
3453 old_length
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3454 new_l1_size
= size_to_l1(s
, offset
);
3456 if (offset
< old_length
) {
3457 int64_t last_cluster
, old_file_size
;
3458 if (prealloc
!= PREALLOC_MODE_OFF
) {
3460 "Preallocation can't be used for shrinking an image");
3465 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
3466 old_length
- ROUND_UP(offset
,
3468 QCOW2_DISCARD_ALWAYS
, true);
3470 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
3474 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
3476 error_setg_errno(errp
, -ret
,
3477 "Failed to reduce the number of L2 tables");
3481 ret
= qcow2_shrink_reftable(bs
);
3483 error_setg_errno(errp
, -ret
,
3484 "Failed to discard unused refblocks");
3488 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3489 if (old_file_size
< 0) {
3490 error_setg_errno(errp
, -old_file_size
,
3491 "Failed to inquire current file length");
3492 ret
= old_file_size
;
3495 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
3496 if (last_cluster
< 0) {
3497 error_setg_errno(errp
, -last_cluster
,
3498 "Failed to find the last cluster");
3502 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
3503 Error
*local_err
= NULL
;
3505 bdrv_co_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
3506 PREALLOC_MODE_OFF
, &local_err
);
3508 warn_reportf_err(local_err
,
3509 "Failed to truncate the tail of the image: ");
3513 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
3515 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
3521 case PREALLOC_MODE_OFF
:
3524 case PREALLOC_MODE_METADATA
:
3525 ret
= preallocate_co(bs
, old_length
, offset
);
3527 error_setg_errno(errp
, -ret
, "Preallocation failed");
3532 case PREALLOC_MODE_FALLOC
:
3533 case PREALLOC_MODE_FULL
:
3535 int64_t allocation_start
, host_offset
, guest_offset
;
3536 int64_t clusters_allocated
;
3537 int64_t old_file_size
, new_file_size
;
3538 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
3540 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3541 if (old_file_size
< 0) {
3542 error_setg_errno(errp
, -old_file_size
,
3543 "Failed to inquire current file length");
3544 ret
= old_file_size
;
3547 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
3549 nb_new_data_clusters
= DIV_ROUND_UP(offset
- old_length
,
3552 /* This is an overestimation; we will not actually allocate space for
3553 * these in the file but just make sure the new refcount structures are
3554 * able to cover them so we will not have to allocate new refblocks
3555 * while entering the data blocks in the potentially new L2 tables.
3556 * (We do not actually care where the L2 tables are placed. Maybe they
3557 * are already allocated or they can be placed somewhere before
3558 * @old_file_size. It does not matter because they will be fully
3559 * allocated automatically, so they do not need to be covered by the
3560 * preallocation. All that matters is that we will not have to allocate
3561 * new refcount structures for them.) */
3562 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
3563 s
->cluster_size
/ sizeof(uint64_t));
3564 /* The cluster range may not be aligned to L2 boundaries, so add one L2
3565 * table for a potential head/tail */
3568 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
3569 nb_new_data_clusters
+
3572 if (allocation_start
< 0) {
3573 error_setg_errno(errp
, -allocation_start
,
3574 "Failed to resize refcount structures");
3575 ret
= allocation_start
;
3579 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
3580 nb_new_data_clusters
);
3581 if (clusters_allocated
< 0) {
3582 error_setg_errno(errp
, -clusters_allocated
,
3583 "Failed to allocate data clusters");
3584 ret
= clusters_allocated
;
3588 assert(clusters_allocated
== nb_new_data_clusters
);
3590 /* Allocate the data area */
3591 new_file_size
= allocation_start
+
3592 nb_new_data_clusters
* s
->cluster_size
;
3593 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, prealloc
, errp
);
3595 error_prepend(errp
, "Failed to resize underlying file: ");
3596 qcow2_free_clusters(bs
, allocation_start
,
3597 nb_new_data_clusters
* s
->cluster_size
,
3598 QCOW2_DISCARD_OTHER
);
3602 /* Create the necessary L2 entries */
3603 host_offset
= allocation_start
;
3604 guest_offset
= old_length
;
3605 while (nb_new_data_clusters
) {
3606 int64_t nb_clusters
= MIN(
3607 nb_new_data_clusters
,
3608 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
3609 QCowL2Meta allocation
= {
3610 .offset
= guest_offset
,
3611 .alloc_offset
= host_offset
,
3612 .nb_clusters
= nb_clusters
,
3614 qemu_co_queue_init(&allocation
.dependent_requests
);
3616 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
3618 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
3619 qcow2_free_clusters(bs
, host_offset
,
3620 nb_new_data_clusters
* s
->cluster_size
,
3621 QCOW2_DISCARD_OTHER
);
3625 guest_offset
+= nb_clusters
* s
->cluster_size
;
3626 host_offset
+= nb_clusters
* s
->cluster_size
;
3627 nb_new_data_clusters
-= nb_clusters
;
3633 g_assert_not_reached();
3636 if (prealloc
!= PREALLOC_MODE_OFF
) {
3637 /* Flush metadata before actually changing the image size */
3638 ret
= qcow2_write_caches(bs
);
3640 error_setg_errno(errp
, -ret
,
3641 "Failed to flush the preallocated area to disk");
3646 bs
->total_sectors
= offset
/ BDRV_SECTOR_SIZE
;
3648 /* write updated header.size */
3649 offset
= cpu_to_be64(offset
);
3650 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
3651 &offset
, sizeof(uint64_t));
3653 error_setg_errno(errp
, -ret
, "Failed to update the image size");
3657 s
->l1_vm_state_index
= new_l1_size
;
3659 /* Update cache sizes */
3660 options
= qdict_clone_shallow(bs
->options
);
3661 ret
= qcow2_update_options(bs
, options
, s
->flags
, errp
);
3662 qobject_unref(options
);
3668 qemu_co_mutex_unlock(&s
->lock
);
3675 * @dest - destination buffer, at least of @size-1 bytes
3676 * @src - source buffer, @size bytes
3678 * Returns: compressed size on success
3679 * -1 if compression is inefficient
3680 * -2 on any other error
3682 static ssize_t
qcow2_compress(void *dest
, const void *src
, size_t size
)
3687 /* best compression, small window, no zlib header */
3688 memset(&strm
, 0, sizeof(strm
));
3689 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
, Z_DEFLATED
,
3690 -12, 9, Z_DEFAULT_STRATEGY
);
3695 /* strm.next_in is not const in old zlib versions, such as those used on
3696 * OpenBSD/NetBSD, so cast the const away */
3697 strm
.avail_in
= size
;
3698 strm
.next_in
= (void *) src
;
3699 strm
.avail_out
= size
- 1;
3700 strm
.next_out
= dest
;
3702 ret
= deflate(&strm
, Z_FINISH
);
3703 if (ret
== Z_STREAM_END
) {
3704 ret
= size
- 1 - strm
.avail_out
;
3706 ret
= (ret
== Z_OK
? -1 : -2);
3714 #define MAX_COMPRESS_THREADS 4
3716 typedef struct Qcow2CompressData
{
3721 } Qcow2CompressData
;
3723 static int qcow2_compress_pool_func(void *opaque
)
3725 Qcow2CompressData
*data
= opaque
;
3727 data
->ret
= qcow2_compress(data
->dest
, data
->src
, data
->size
);
3732 static void qcow2_compress_complete(void *opaque
, int ret
)
3734 qemu_coroutine_enter(opaque
);
3737 /* See qcow2_compress definition for parameters description */
3738 static ssize_t
qcow2_co_compress(BlockDriverState
*bs
,
3739 void *dest
, const void *src
, size_t size
)
3741 BDRVQcow2State
*s
= bs
->opaque
;
3743 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
3744 Qcow2CompressData arg
= {
3750 while (s
->nb_compress_threads
>= MAX_COMPRESS_THREADS
) {
3751 qemu_co_queue_wait(&s
->compress_wait_queue
, NULL
);
3754 s
->nb_compress_threads
++;
3755 acb
= thread_pool_submit_aio(pool
, qcow2_compress_pool_func
, &arg
,
3756 qcow2_compress_complete
,
3757 qemu_coroutine_self());
3760 s
->nb_compress_threads
--;
3763 qemu_coroutine_yield();
3764 s
->nb_compress_threads
--;
3765 qemu_co_queue_next(&s
->compress_wait_queue
);
3770 /* XXX: put compressed sectors first, then all the cluster aligned
3771 tables to avoid losing bytes in alignment */
3772 static coroutine_fn
int
3773 qcow2_co_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
3774 uint64_t bytes
, QEMUIOVector
*qiov
)
3776 BDRVQcow2State
*s
= bs
->opaque
;
3777 QEMUIOVector hd_qiov
;
3781 uint8_t *buf
, *out_buf
;
3782 int64_t cluster_offset
;
3785 /* align end of file to a sector boundary to ease reading with
3786 sector based I/Os */
3787 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
3788 if (cluster_offset
< 0) {
3789 return cluster_offset
;
3791 return bdrv_co_truncate(bs
->file
, cluster_offset
, PREALLOC_MODE_OFF
,
3795 if (offset_into_cluster(s
, offset
)) {
3799 buf
= qemu_blockalign(bs
, s
->cluster_size
);
3800 if (bytes
!= s
->cluster_size
) {
3801 if (bytes
> s
->cluster_size
||
3802 offset
+ bytes
!= bs
->total_sectors
<< BDRV_SECTOR_BITS
)
3807 /* Zero-pad last write if image size is not cluster aligned */
3808 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
3810 qemu_iovec_to_buf(qiov
, 0, buf
, bytes
);
3812 out_buf
= g_malloc(s
->cluster_size
);
3814 out_len
= qcow2_co_compress(bs
, out_buf
, buf
, s
->cluster_size
);
3815 if (out_len
== -2) {
3818 } else if (out_len
== -1) {
3819 /* could not compress: write normal cluster */
3820 ret
= qcow2_co_pwritev(bs
, offset
, bytes
, qiov
, 0);
3827 qemu_co_mutex_lock(&s
->lock
);
3829 qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
);
3830 if (!cluster_offset
) {
3831 qemu_co_mutex_unlock(&s
->lock
);
3835 cluster_offset
&= s
->cluster_offset_mask
;
3837 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
);
3838 qemu_co_mutex_unlock(&s
->lock
);
3843 iov
= (struct iovec
) {
3844 .iov_base
= out_buf
,
3847 qemu_iovec_init_external(&hd_qiov
, &iov
, 1);
3849 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
3850 ret
= bdrv_co_pwritev(bs
->file
, cluster_offset
, out_len
, &hd_qiov
, 0);
3862 static int make_completely_empty(BlockDriverState
*bs
)
3864 BDRVQcow2State
*s
= bs
->opaque
;
3865 Error
*local_err
= NULL
;
3866 int ret
, l1_clusters
;
3868 uint64_t *new_reftable
= NULL
;
3869 uint64_t rt_entry
, l1_size2
;
3872 uint64_t reftable_offset
;
3873 uint32_t reftable_clusters
;
3874 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
3876 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
3881 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
3886 /* Refcounts will be broken utterly */
3887 ret
= qcow2_mark_dirty(bs
);
3892 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
3894 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
3895 l1_size2
= (uint64_t)s
->l1_size
* sizeof(uint64_t);
3897 /* After this call, neither the in-memory nor the on-disk refcount
3898 * information accurately describe the actual references */
3900 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
3901 l1_clusters
* s
->cluster_size
, 0);
3903 goto fail_broken_refcounts
;
3905 memset(s
->l1_table
, 0, l1_size2
);
3907 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
3909 /* Overwrite enough clusters at the beginning of the sectors to place
3910 * the refcount table, a refcount block and the L1 table in; this may
3911 * overwrite parts of the existing refcount and L1 table, which is not
3912 * an issue because the dirty flag is set, complete data loss is in fact
3913 * desired and partial data loss is consequently fine as well */
3914 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
3915 (2 + l1_clusters
) * s
->cluster_size
, 0);
3916 /* This call (even if it failed overall) may have overwritten on-disk
3917 * refcount structures; in that case, the in-memory refcount information
3918 * will probably differ from the on-disk information which makes the BDS
3921 goto fail_broken_refcounts
;
3924 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
3925 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
3927 /* "Create" an empty reftable (one cluster) directly after the image
3928 * header and an empty L1 table three clusters after the image header;
3929 * the cluster between those two will be used as the first refblock */
3930 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
3931 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
3932 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
3933 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
3934 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
3936 goto fail_broken_refcounts
;
3939 s
->l1_table_offset
= 3 * s
->cluster_size
;
3941 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ sizeof(uint64_t));
3942 if (!new_reftable
) {
3944 goto fail_broken_refcounts
;
3947 s
->refcount_table_offset
= s
->cluster_size
;
3948 s
->refcount_table_size
= s
->cluster_size
/ sizeof(uint64_t);
3949 s
->max_refcount_table_index
= 0;
3951 g_free(s
->refcount_table
);
3952 s
->refcount_table
= new_reftable
;
3953 new_reftable
= NULL
;
3955 /* Now the in-memory refcount information again corresponds to the on-disk
3956 * information (reftable is empty and no refblocks (the refblock cache is
3957 * empty)); however, this means some clusters (e.g. the image header) are
3958 * referenced, but not refcounted, but the normal qcow2 code assumes that
3959 * the in-memory information is always correct */
3961 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
3963 /* Enter the first refblock into the reftable */
3964 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
3965 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
3966 &rt_entry
, sizeof(rt_entry
));
3968 goto fail_broken_refcounts
;
3970 s
->refcount_table
[0] = 2 * s
->cluster_size
;
3972 s
->free_cluster_index
= 0;
3973 assert(3 + l1_clusters
<= s
->refcount_block_size
);
3974 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
3977 goto fail_broken_refcounts
;
3978 } else if (offset
> 0) {
3979 error_report("First cluster in emptied image is in use");
3983 /* Now finally the in-memory information corresponds to the on-disk
3984 * structures and is correct */
3985 ret
= qcow2_mark_clean(bs
);
3990 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
,
3991 PREALLOC_MODE_OFF
, &local_err
);
3993 error_report_err(local_err
);
3999 fail_broken_refcounts
:
4000 /* The BDS is unusable at this point. If we wanted to make it usable, we
4001 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4002 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4003 * again. However, because the functions which could have caused this error
4004 * path to be taken are used by those functions as well, it's very likely
4005 * that that sequence will fail as well. Therefore, just eject the BDS. */
4009 g_free(new_reftable
);
4013 static int qcow2_make_empty(BlockDriverState
*bs
)
4015 BDRVQcow2State
*s
= bs
->opaque
;
4016 uint64_t offset
, end_offset
;
4017 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
4018 int l1_clusters
, ret
= 0;
4020 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
4022 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
4023 3 + l1_clusters
<= s
->refcount_block_size
&&
4024 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
4025 /* The following function only works for qcow2 v3 images (it
4026 * requires the dirty flag) and only as long as there are no
4027 * features that reserve extra clusters (such as snapshots,
4028 * LUKS header, or persistent bitmaps), because it completely
4029 * empties the image. Furthermore, the L1 table and three
4030 * additional clusters (image header, refcount table, one
4031 * refcount block) have to fit inside one refcount block. */
4032 return make_completely_empty(bs
);
4035 /* This fallback code simply discards every active cluster; this is slow,
4036 * but works in all cases */
4037 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4038 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
4039 /* As this function is generally used after committing an external
4040 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4041 * default action for this kind of discard is to pass the discard,
4042 * which will ideally result in an actually smaller image file, as
4043 * is probably desired. */
4044 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
4045 QCOW2_DISCARD_SNAPSHOT
, true);
4054 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
4056 BDRVQcow2State
*s
= bs
->opaque
;
4059 qemu_co_mutex_lock(&s
->lock
);
4060 ret
= qcow2_write_caches(bs
);
4061 qemu_co_mutex_unlock(&s
->lock
);
4066 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
4069 Error
*local_err
= NULL
;
4070 BlockMeasureInfo
*info
;
4071 uint64_t required
= 0; /* bytes that contribute to required size */
4072 uint64_t virtual_size
; /* disk size as seen by guest */
4073 uint64_t refcount_bits
;
4075 size_t cluster_size
;
4078 PreallocMode prealloc
;
4079 bool has_backing_file
;
4081 /* Parse image creation options */
4082 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, &local_err
);
4087 version
= qcow2_opt_get_version_del(opts
, &local_err
);
4092 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
4097 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
4098 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
4099 PREALLOC_MODE_OFF
, &local_err
);
4105 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
4106 has_backing_file
= !!optstr
;
4109 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
4110 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
4112 /* Check that virtual disk size is valid */
4113 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
4114 cluster_size
/ sizeof(uint64_t));
4115 if (l2_tables
* sizeof(uint64_t) > QCOW_MAX_L1_SIZE
) {
4116 error_setg(&local_err
, "The image size is too large "
4117 "(try using a larger cluster size)");
4121 /* Account for input image */
4123 int64_t ssize
= bdrv_getlength(in_bs
);
4125 error_setg_errno(&local_err
, -ssize
,
4126 "Unable to get image virtual_size");
4130 virtual_size
= ROUND_UP(ssize
, cluster_size
);
4132 if (has_backing_file
) {
4133 /* We don't how much of the backing chain is shared by the input
4134 * image and the new image file. In the worst case the new image's
4135 * backing file has nothing in common with the input image. Be
4136 * conservative and assume all clusters need to be written.
4138 required
= virtual_size
;
4143 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
4146 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
4147 ssize
- offset
, &pnum
, NULL
,
4150 error_setg_errno(&local_err
, -ret
,
4151 "Unable to get block status");
4155 if (ret
& BDRV_BLOCK_ZERO
) {
4156 /* Skip zero regions (safe with no backing file) */
4157 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
4158 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
4159 /* Extend pnum to end of cluster for next iteration */
4160 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
4162 /* Count clusters we've seen */
4163 required
+= offset
% cluster_size
+ pnum
;
4169 /* Take into account preallocation. Nothing special is needed for
4170 * PREALLOC_MODE_METADATA since metadata is always counted.
4172 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
4173 required
= virtual_size
;
4176 info
= g_new(BlockMeasureInfo
, 1);
4177 info
->fully_allocated
=
4178 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
4179 ctz32(refcount_bits
));
4181 /* Remove data clusters that are not required. This overestimates the
4182 * required size because metadata needed for the fully allocated file is
4185 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
4189 error_propagate(errp
, local_err
);
4193 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4195 BDRVQcow2State
*s
= bs
->opaque
;
4196 bdi
->unallocated_blocks_are_zero
= true;
4197 bdi
->cluster_size
= s
->cluster_size
;
4198 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
4202 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
)
4204 BDRVQcow2State
*s
= bs
->opaque
;
4205 ImageInfoSpecific
*spec_info
;
4206 QCryptoBlockInfo
*encrypt_info
= NULL
;
4208 if (s
->crypto
!= NULL
) {
4209 encrypt_info
= qcrypto_block_get_info(s
->crypto
, &error_abort
);
4212 spec_info
= g_new(ImageInfoSpecific
, 1);
4213 *spec_info
= (ImageInfoSpecific
){
4214 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
4215 .u
.qcow2
.data
= g_new(ImageInfoSpecificQCow2
, 1),
4217 if (s
->qcow_version
== 2) {
4218 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
4219 .compat
= g_strdup("0.10"),
4220 .refcount_bits
= s
->refcount_bits
,
4222 } else if (s
->qcow_version
== 3) {
4223 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
4224 .compat
= g_strdup("1.1"),
4225 .lazy_refcounts
= s
->compatible_features
&
4226 QCOW2_COMPAT_LAZY_REFCOUNTS
,
4227 .has_lazy_refcounts
= true,
4228 .corrupt
= s
->incompatible_features
&
4229 QCOW2_INCOMPAT_CORRUPT
,
4230 .has_corrupt
= true,
4231 .refcount_bits
= s
->refcount_bits
,
4234 /* if this assertion fails, this probably means a new version was
4235 * added without having it covered here */
4240 ImageInfoSpecificQCow2Encryption
*qencrypt
=
4241 g_new(ImageInfoSpecificQCow2Encryption
, 1);
4242 switch (encrypt_info
->format
) {
4243 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
4244 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
4246 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
4247 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
4248 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
4253 /* Since we did shallow copy above, erase any pointers
4254 * in the original info */
4255 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
4256 qapi_free_QCryptoBlockInfo(encrypt_info
);
4258 spec_info
->u
.qcow2
.data
->has_encrypt
= true;
4259 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
4265 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
4268 BDRVQcow2State
*s
= bs
->opaque
;
4270 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
4271 return bs
->drv
->bdrv_co_pwritev(bs
, qcow2_vm_state_offset(s
) + pos
,
4272 qiov
->size
, qiov
, 0);
4275 static int qcow2_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
4278 BDRVQcow2State
*s
= bs
->opaque
;
4280 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
4281 return bs
->drv
->bdrv_co_preadv(bs
, qcow2_vm_state_offset(s
) + pos
,
4282 qiov
->size
, qiov
, 0);
4286 * Downgrades an image's version. To achieve this, any incompatible features
4287 * have to be removed.
4289 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
4290 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
4293 BDRVQcow2State
*s
= bs
->opaque
;
4294 int current_version
= s
->qcow_version
;
4297 /* This is qcow2_downgrade(), not qcow2_upgrade() */
4298 assert(target_version
< current_version
);
4300 /* There are no other versions (now) that you can downgrade to */
4301 assert(target_version
== 2);
4303 if (s
->refcount_order
!= 4) {
4304 error_setg(errp
, "compat=0.10 requires refcount_bits=16");
4308 /* clear incompatible features */
4309 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
4310 ret
= qcow2_mark_clean(bs
);
4312 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
4317 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
4318 * the first place; if that happens nonetheless, returning -ENOTSUP is the
4319 * best thing to do anyway */
4321 if (s
->incompatible_features
) {
4322 error_setg(errp
, "Cannot downgrade an image with incompatible features "
4323 "%#" PRIx64
" set", s
->incompatible_features
);
4327 /* since we can ignore compatible features, we can set them to 0 as well */
4328 s
->compatible_features
= 0;
4329 /* if lazy refcounts have been used, they have already been fixed through
4330 * clearing the dirty flag */
4332 /* clearing autoclear features is trivial */
4333 s
->autoclear_features
= 0;
4335 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
4337 error_setg_errno(errp
, -ret
, "Failed to turn zero into data clusters");
4341 s
->qcow_version
= target_version
;
4342 ret
= qcow2_update_header(bs
);
4344 s
->qcow_version
= current_version
;
4345 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4351 typedef enum Qcow2AmendOperation
{
4352 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
4353 * statically initialized to so that the helper CB can discern the first
4354 * invocation from an operation change */
4355 QCOW2_NO_OPERATION
= 0,
4357 QCOW2_CHANGING_REFCOUNT_ORDER
,
4359 } Qcow2AmendOperation
;
4361 typedef struct Qcow2AmendHelperCBInfo
{
4362 /* The code coordinating the amend operations should only modify
4363 * these four fields; the rest will be managed by the CB */
4364 BlockDriverAmendStatusCB
*original_status_cb
;
4365 void *original_cb_opaque
;
4367 Qcow2AmendOperation current_operation
;
4369 /* Total number of operations to perform (only set once) */
4370 int total_operations
;
4372 /* The following fields are managed by the CB */
4374 /* Number of operations completed */
4375 int operations_completed
;
4377 /* Cumulative offset of all completed operations */
4378 int64_t offset_completed
;
4380 Qcow2AmendOperation last_operation
;
4381 int64_t last_work_size
;
4382 } Qcow2AmendHelperCBInfo
;
4384 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
4385 int64_t operation_offset
,
4386 int64_t operation_work_size
, void *opaque
)
4388 Qcow2AmendHelperCBInfo
*info
= opaque
;
4389 int64_t current_work_size
;
4390 int64_t projected_work_size
;
4392 if (info
->current_operation
!= info
->last_operation
) {
4393 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
4394 info
->offset_completed
+= info
->last_work_size
;
4395 info
->operations_completed
++;
4398 info
->last_operation
= info
->current_operation
;
4401 assert(info
->total_operations
> 0);
4402 assert(info
->operations_completed
< info
->total_operations
);
4404 info
->last_work_size
= operation_work_size
;
4406 current_work_size
= info
->offset_completed
+ operation_work_size
;
4408 /* current_work_size is the total work size for (operations_completed + 1)
4409 * operations (which includes this one), so multiply it by the number of
4410 * operations not covered and divide it by the number of operations
4411 * covered to get a projection for the operations not covered */
4412 projected_work_size
= current_work_size
* (info
->total_operations
-
4413 info
->operations_completed
- 1)
4414 / (info
->operations_completed
+ 1);
4416 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
4417 current_work_size
+ projected_work_size
,
4418 info
->original_cb_opaque
);
4421 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4422 BlockDriverAmendStatusCB
*status_cb
,
4426 BDRVQcow2State
*s
= bs
->opaque
;
4427 int old_version
= s
->qcow_version
, new_version
= old_version
;
4428 uint64_t new_size
= 0;
4429 const char *backing_file
= NULL
, *backing_format
= NULL
;
4430 bool lazy_refcounts
= s
->use_lazy_refcounts
;
4431 const char *compat
= NULL
;
4432 uint64_t cluster_size
= s
->cluster_size
;
4435 int refcount_bits
= s
->refcount_bits
;
4437 QemuOptDesc
*desc
= opts
->list
->desc
;
4438 Qcow2AmendHelperCBInfo helper_cb_info
;
4440 while (desc
&& desc
->name
) {
4441 if (!qemu_opt_find(opts
, desc
->name
)) {
4442 /* only change explicitly defined options */
4447 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
4448 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
4450 /* preserve default */
4451 } else if (!strcmp(compat
, "0.10")) {
4453 } else if (!strcmp(compat
, "1.1")) {
4456 error_setg(errp
, "Unknown compatibility level %s", compat
);
4459 } else if (!strcmp(desc
->name
, BLOCK_OPT_PREALLOC
)) {
4460 error_setg(errp
, "Cannot change preallocation mode");
4462 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
4463 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
4464 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
4465 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4466 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
4467 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4468 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT
)) {
4469 encrypt
= qemu_opt_get_bool(opts
, BLOCK_OPT_ENCRYPT
,
4472 if (encrypt
!= !!s
->crypto
) {
4474 "Changing the encryption flag is not supported");
4477 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT_FORMAT
)) {
4478 encformat
= qcow2_crypt_method_from_format(
4479 qemu_opt_get(opts
, BLOCK_OPT_ENCRYPT_FORMAT
));
4481 if (encformat
!= s
->crypt_method_header
) {
4483 "Changing the encryption format is not supported");
4486 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
4488 "Changing the encryption parameters is not supported");
4490 } else if (!strcmp(desc
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
4491 cluster_size
= qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
,
4493 if (cluster_size
!= s
->cluster_size
) {
4494 error_setg(errp
, "Changing the cluster size is not supported");
4497 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
4498 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
4500 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
4501 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
4504 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
4505 !is_power_of_2(refcount_bits
))
4507 error_setg(errp
, "Refcount width must be a power of two and "
4508 "may not exceed 64 bits");
4512 /* if this point is reached, this probably means a new option was
4513 * added without having it covered here */
4520 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
4521 .original_status_cb
= status_cb
,
4522 .original_cb_opaque
= cb_opaque
,
4523 .total_operations
= (new_version
< old_version
)
4524 + (s
->refcount_bits
!= refcount_bits
)
4527 /* Upgrade first (some features may require compat=1.1) */
4528 if (new_version
> old_version
) {
4529 s
->qcow_version
= new_version
;
4530 ret
= qcow2_update_header(bs
);
4532 s
->qcow_version
= old_version
;
4533 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4538 if (s
->refcount_bits
!= refcount_bits
) {
4539 int refcount_order
= ctz32(refcount_bits
);
4541 if (new_version
< 3 && refcount_bits
!= 16) {
4542 error_setg(errp
, "Refcount widths other than 16 bits require "
4543 "compatibility level 1.1 or above (use compat=1.1 or "
4548 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
4549 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
4550 &qcow2_amend_helper_cb
,
4551 &helper_cb_info
, errp
);
4557 if (backing_file
|| backing_format
) {
4558 ret
= qcow2_change_backing_file(bs
,
4559 backing_file
?: s
->image_backing_file
,
4560 backing_format
?: s
->image_backing_format
);
4562 error_setg_errno(errp
, -ret
, "Failed to change the backing file");
4567 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
4568 if (lazy_refcounts
) {
4569 if (new_version
< 3) {
4570 error_setg(errp
, "Lazy refcounts only supported with "
4571 "compatibility level 1.1 and above (use compat=1.1 "
4575 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
4576 ret
= qcow2_update_header(bs
);
4578 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
4579 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4582 s
->use_lazy_refcounts
= true;
4584 /* make image clean first */
4585 ret
= qcow2_mark_clean(bs
);
4587 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
4590 /* now disallow lazy refcounts */
4591 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
4592 ret
= qcow2_update_header(bs
);
4594 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
4595 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4598 s
->use_lazy_refcounts
= false;
4603 BlockBackend
*blk
= blk_new(BLK_PERM_RESIZE
, BLK_PERM_ALL
);
4604 ret
= blk_insert_bs(blk
, bs
, errp
);
4610 ret
= blk_truncate(blk
, new_size
, PREALLOC_MODE_OFF
, errp
);
4617 /* Downgrade last (so unsupported features can be removed before) */
4618 if (new_version
< old_version
) {
4619 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
4620 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
4621 &helper_cb_info
, errp
);
4631 * If offset or size are negative, respectively, they will not be included in
4632 * the BLOCK_IMAGE_CORRUPTED event emitted.
4633 * fatal will be ignored for read-only BDS; corruptions found there will always
4634 * be considered non-fatal.
4636 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
4637 int64_t size
, const char *message_format
, ...)
4639 BDRVQcow2State
*s
= bs
->opaque
;
4640 const char *node_name
;
4644 fatal
= fatal
&& bdrv_is_writable(bs
);
4646 if (s
->signaled_corruption
&&
4647 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
4652 va_start(ap
, message_format
);
4653 message
= g_strdup_vprintf(message_format
, ap
);
4657 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
4658 "corruption events will be suppressed\n", message
);
4660 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
4661 "corruption events will be suppressed\n", message
);
4664 node_name
= bdrv_get_node_name(bs
);
4665 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
4666 *node_name
!= '\0', node_name
,
4667 message
, offset
>= 0, offset
,
4673 qcow2_mark_corrupt(bs
);
4674 bs
->drv
= NULL
; /* make BDS unusable */
4677 s
->signaled_corruption
= true;
4680 static QemuOptsList qcow2_create_opts
= {
4681 .name
= "qcow2-create-opts",
4682 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
4685 .name
= BLOCK_OPT_SIZE
,
4686 .type
= QEMU_OPT_SIZE
,
4687 .help
= "Virtual disk size"
4690 .name
= BLOCK_OPT_COMPAT_LEVEL
,
4691 .type
= QEMU_OPT_STRING
,
4692 .help
= "Compatibility level (0.10 or 1.1)"
4695 .name
= BLOCK_OPT_BACKING_FILE
,
4696 .type
= QEMU_OPT_STRING
,
4697 .help
= "File name of a base image"
4700 .name
= BLOCK_OPT_BACKING_FMT
,
4701 .type
= QEMU_OPT_STRING
,
4702 .help
= "Image format of the base image"
4705 .name
= BLOCK_OPT_ENCRYPT
,
4706 .type
= QEMU_OPT_BOOL
,
4707 .help
= "Encrypt the image with format 'aes'. (Deprecated "
4708 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)",
4711 .name
= BLOCK_OPT_ENCRYPT_FORMAT
,
4712 .type
= QEMU_OPT_STRING
,
4713 .help
= "Encrypt the image, format choices: 'aes', 'luks'",
4715 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
4716 "ID of secret providing qcow AES key or LUKS passphrase"),
4717 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
4718 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
4719 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
4720 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
4721 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
4722 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
4724 .name
= BLOCK_OPT_CLUSTER_SIZE
,
4725 .type
= QEMU_OPT_SIZE
,
4726 .help
= "qcow2 cluster size",
4727 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
)
4730 .name
= BLOCK_OPT_PREALLOC
,
4731 .type
= QEMU_OPT_STRING
,
4732 .help
= "Preallocation mode (allowed values: off, metadata, "
4736 .name
= BLOCK_OPT_LAZY_REFCOUNTS
,
4737 .type
= QEMU_OPT_BOOL
,
4738 .help
= "Postpone refcount updates",
4739 .def_value_str
= "off"
4742 .name
= BLOCK_OPT_REFCOUNT_BITS
,
4743 .type
= QEMU_OPT_NUMBER
,
4744 .help
= "Width of a reference count entry in bits",
4745 .def_value_str
= "16"
4747 { /* end of list */ }
4751 BlockDriver bdrv_qcow2
= {
4752 .format_name
= "qcow2",
4753 .instance_size
= sizeof(BDRVQcow2State
),
4754 .bdrv_probe
= qcow2_probe
,
4755 .bdrv_open
= qcow2_open
,
4756 .bdrv_close
= qcow2_close
,
4757 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
4758 .bdrv_reopen_commit
= qcow2_reopen_commit
,
4759 .bdrv_reopen_abort
= qcow2_reopen_abort
,
4760 .bdrv_join_options
= qcow2_join_options
,
4761 .bdrv_child_perm
= bdrv_format_default_perms
,
4762 .bdrv_co_create_opts
= qcow2_co_create_opts
,
4763 .bdrv_co_create
= qcow2_co_create
,
4764 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
4765 .bdrv_co_block_status
= qcow2_co_block_status
,
4767 .bdrv_co_preadv
= qcow2_co_preadv
,
4768 .bdrv_co_pwritev
= qcow2_co_pwritev
,
4769 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
4771 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
4772 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
4773 .bdrv_co_copy_range_from
= qcow2_co_copy_range_from
,
4774 .bdrv_co_copy_range_to
= qcow2_co_copy_range_to
,
4775 .bdrv_co_truncate
= qcow2_co_truncate
,
4776 .bdrv_co_pwritev_compressed
= qcow2_co_pwritev_compressed
,
4777 .bdrv_make_empty
= qcow2_make_empty
,
4779 .bdrv_snapshot_create
= qcow2_snapshot_create
,
4780 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
4781 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
4782 .bdrv_snapshot_list
= qcow2_snapshot_list
,
4783 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
4784 .bdrv_measure
= qcow2_measure
,
4785 .bdrv_get_info
= qcow2_get_info
,
4786 .bdrv_get_specific_info
= qcow2_get_specific_info
,
4788 .bdrv_save_vmstate
= qcow2_save_vmstate
,
4789 .bdrv_load_vmstate
= qcow2_load_vmstate
,
4791 .supports_backing
= true,
4792 .bdrv_change_backing_file
= qcow2_change_backing_file
,
4794 .bdrv_refresh_limits
= qcow2_refresh_limits
,
4795 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
4796 .bdrv_inactivate
= qcow2_inactivate
,
4798 .create_opts
= &qcow2_create_opts
,
4799 .bdrv_co_check
= qcow2_co_check
,
4800 .bdrv_amend_options
= qcow2_amend_options
,
4802 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
4803 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
4805 .bdrv_reopen_bitmaps_rw
= qcow2_reopen_bitmaps_rw
,
4806 .bdrv_can_store_new_dirty_bitmap
= qcow2_can_store_new_dirty_bitmap
,
4807 .bdrv_remove_persistent_dirty_bitmap
= qcow2_remove_persistent_dirty_bitmap
,
4810 static void bdrv_qcow2_init(void)
4812 bdrv_register(&bdrv_qcow2
);
4815 block_init(bdrv_qcow2_init
);