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"
27 #include "block/qdict.h"
28 #include "sysemu/block-backend.h"
29 #include "qemu/main-loop.h"
30 #include "qemu/module.h"
32 #include "qemu/error-report.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-events-block-core.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qstring.h"
38 #include "qemu/option_int.h"
39 #include "qemu/cutils.h"
40 #include "qemu/bswap.h"
41 #include "qapi/qobject-input-visitor.h"
42 #include "qapi/qapi-visit-block-core.h"
44 #include "block/aio_task.h"
47 Differences with QCOW:
49 - Support for multiple incremental snapshots.
50 - Memory management by reference counts.
51 - Clusters which have a reference count of one have the bit
52 QCOW_OFLAG_COPIED to optimize write performance.
53 - Size of compressed clusters is stored in sectors to reduce bit usage
54 in the cluster offsets.
55 - Support for storing additional data (such as the VM state) in the
57 - If a backing store is used, the cluster size is not constrained
58 (could be backported to QCOW).
59 - L2 tables have always a size of one cluster.
66 } QEMU_PACKED QCowExtension
;
68 #define QCOW2_EXT_MAGIC_END 0
69 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca
70 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
71 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
72 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
73 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
75 static int coroutine_fn
76 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
77 uint64_t cluster_descriptor
,
83 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
85 const QCowHeader
*cow_header
= (const void *)buf
;
87 if (buf_size
>= sizeof(QCowHeader
) &&
88 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
89 be32_to_cpu(cow_header
->version
) >= 2)
96 static ssize_t
qcow2_crypto_hdr_read_func(QCryptoBlock
*block
, size_t offset
,
97 uint8_t *buf
, size_t buflen
,
98 void *opaque
, Error
**errp
)
100 BlockDriverState
*bs
= opaque
;
101 BDRVQcow2State
*s
= bs
->opaque
;
104 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
105 error_setg(errp
, "Request for data outside of extension header");
109 ret
= bdrv_pread(bs
->file
,
110 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
112 error_setg_errno(errp
, -ret
, "Could not read encryption header");
119 static ssize_t
qcow2_crypto_hdr_init_func(QCryptoBlock
*block
, size_t headerlen
,
120 void *opaque
, Error
**errp
)
122 BlockDriverState
*bs
= opaque
;
123 BDRVQcow2State
*s
= bs
->opaque
;
127 ret
= qcow2_alloc_clusters(bs
, headerlen
);
129 error_setg_errno(errp
, -ret
,
130 "Cannot allocate cluster for LUKS header size %zu",
135 s
->crypto_header
.length
= headerlen
;
136 s
->crypto_header
.offset
= ret
;
139 * Zero fill all space in cluster so it has predictable
140 * content, as we may not initialize some regions of the
141 * header (eg only 1 out of 8 key slots will be initialized)
143 clusterlen
= size_to_clusters(s
, headerlen
) * s
->cluster_size
;
144 assert(qcow2_pre_write_overlap_check(bs
, 0, ret
, clusterlen
, false) == 0);
145 ret
= bdrv_pwrite_zeroes(bs
->file
,
149 error_setg_errno(errp
, -ret
, "Could not zero fill encryption header");
157 static ssize_t
qcow2_crypto_hdr_write_func(QCryptoBlock
*block
, size_t offset
,
158 const uint8_t *buf
, size_t buflen
,
159 void *opaque
, Error
**errp
)
161 BlockDriverState
*bs
= opaque
;
162 BDRVQcow2State
*s
= bs
->opaque
;
165 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
166 error_setg(errp
, "Request for data outside of extension header");
170 ret
= bdrv_pwrite(bs
->file
,
171 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
173 error_setg_errno(errp
, -ret
, "Could not read encryption header");
180 qcow2_extract_crypto_opts(QemuOpts
*opts
, const char *fmt
, Error
**errp
)
182 QDict
*cryptoopts_qdict
;
185 /* Extract "encrypt." options into a qdict */
186 opts_qdict
= qemu_opts_to_qdict(opts
, NULL
);
187 qdict_extract_subqdict(opts_qdict
, &cryptoopts_qdict
, "encrypt.");
188 qobject_unref(opts_qdict
);
189 qdict_put_str(cryptoopts_qdict
, "format", fmt
);
190 return cryptoopts_qdict
;
194 * read qcow2 extension and fill bs
195 * start reading from start_offset
196 * finish reading upon magic of value 0 or when end_offset reached
197 * unknown magic is skipped (future extension this version knows nothing about)
198 * return 0 upon success, non-0 otherwise
200 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
201 uint64_t end_offset
, void **p_feature_table
,
202 int flags
, bool *need_update_header
,
205 BDRVQcow2State
*s
= bs
->opaque
;
209 Qcow2BitmapHeaderExt bitmaps_ext
;
211 if (need_update_header
!= NULL
) {
212 *need_update_header
= false;
216 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
218 offset
= start_offset
;
219 while (offset
< end_offset
) {
223 if (offset
> s
->cluster_size
)
224 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
226 printf("attempting to read extended header in offset %lu\n", offset
);
229 ret
= bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
));
231 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
232 "pread fail from offset %" PRIu64
, offset
);
235 ext
.magic
= be32_to_cpu(ext
.magic
);
236 ext
.len
= be32_to_cpu(ext
.len
);
237 offset
+= sizeof(ext
);
239 printf("ext.magic = 0x%x\n", ext
.magic
);
241 if (offset
> end_offset
|| ext
.len
> end_offset
- offset
) {
242 error_setg(errp
, "Header extension too large");
247 case QCOW2_EXT_MAGIC_END
:
250 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
251 if (ext
.len
>= sizeof(bs
->backing_format
)) {
252 error_setg(errp
, "ERROR: ext_backing_format: len=%" PRIu32
253 " too large (>=%zu)", ext
.len
,
254 sizeof(bs
->backing_format
));
257 ret
= bdrv_pread(bs
->file
, offset
, bs
->backing_format
, ext
.len
);
259 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
260 "Could not read format name");
263 bs
->backing_format
[ext
.len
] = '\0';
264 s
->image_backing_format
= g_strdup(bs
->backing_format
);
266 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
270 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
271 if (p_feature_table
!= NULL
) {
272 void* feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
273 ret
= bdrv_pread(bs
->file
, offset
, feature_table
, ext
.len
);
275 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
276 "Could not read table");
280 *p_feature_table
= feature_table
;
284 case QCOW2_EXT_MAGIC_CRYPTO_HEADER
: {
285 unsigned int cflags
= 0;
286 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
287 error_setg(errp
, "CRYPTO header extension only "
288 "expected with LUKS encryption method");
291 if (ext
.len
!= sizeof(Qcow2CryptoHeaderExtension
)) {
292 error_setg(errp
, "CRYPTO header extension size %u, "
293 "but expected size %zu", ext
.len
,
294 sizeof(Qcow2CryptoHeaderExtension
));
298 ret
= bdrv_pread(bs
->file
, offset
, &s
->crypto_header
, ext
.len
);
300 error_setg_errno(errp
, -ret
,
301 "Unable to read CRYPTO header extension");
304 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
305 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
307 if ((s
->crypto_header
.offset
% s
->cluster_size
) != 0) {
308 error_setg(errp
, "Encryption header offset '%" PRIu64
"' is "
309 "not a multiple of cluster size '%u'",
310 s
->crypto_header
.offset
, s
->cluster_size
);
314 if (flags
& BDRV_O_NO_IO
) {
315 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
317 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
318 qcow2_crypto_hdr_read_func
,
319 bs
, cflags
, QCOW2_MAX_THREADS
, errp
);
325 case QCOW2_EXT_MAGIC_BITMAPS
:
326 if (ext
.len
!= sizeof(bitmaps_ext
)) {
327 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
328 "Invalid extension length");
332 if (!(s
->autoclear_features
& QCOW2_AUTOCLEAR_BITMAPS
)) {
333 if (s
->qcow_version
< 3) {
334 /* Let's be a bit more specific */
335 warn_report("This qcow2 v2 image contains bitmaps, but "
336 "they may have been modified by a program "
337 "without persistent bitmap support; so now "
338 "they must all be considered inconsistent");
340 warn_report("a program lacking bitmap support "
341 "modified this file, so all bitmaps are now "
342 "considered inconsistent");
344 error_printf("Some clusters may be leaked, "
345 "run 'qemu-img check -r' on the image "
347 if (need_update_header
!= NULL
) {
348 /* Updating is needed to drop invalid bitmap extension. */
349 *need_update_header
= true;
354 ret
= bdrv_pread(bs
->file
, offset
, &bitmaps_ext
, ext
.len
);
356 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
357 "Could not read ext header");
361 if (bitmaps_ext
.reserved32
!= 0) {
362 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
363 "Reserved field is not zero");
367 bitmaps_ext
.nb_bitmaps
= be32_to_cpu(bitmaps_ext
.nb_bitmaps
);
368 bitmaps_ext
.bitmap_directory_size
=
369 be64_to_cpu(bitmaps_ext
.bitmap_directory_size
);
370 bitmaps_ext
.bitmap_directory_offset
=
371 be64_to_cpu(bitmaps_ext
.bitmap_directory_offset
);
373 if (bitmaps_ext
.nb_bitmaps
> QCOW2_MAX_BITMAPS
) {
375 "bitmaps_ext: Image has %" PRIu32
" bitmaps, "
376 "exceeding the QEMU supported maximum of %d",
377 bitmaps_ext
.nb_bitmaps
, QCOW2_MAX_BITMAPS
);
381 if (bitmaps_ext
.nb_bitmaps
== 0) {
382 error_setg(errp
, "found bitmaps extension with zero bitmaps");
386 if (offset_into_cluster(s
, bitmaps_ext
.bitmap_directory_offset
)) {
387 error_setg(errp
, "bitmaps_ext: "
388 "invalid bitmap directory offset");
392 if (bitmaps_ext
.bitmap_directory_size
>
393 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
) {
394 error_setg(errp
, "bitmaps_ext: "
395 "bitmap directory size (%" PRIu64
") exceeds "
396 "the maximum supported size (%d)",
397 bitmaps_ext
.bitmap_directory_size
,
398 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
);
402 s
->nb_bitmaps
= bitmaps_ext
.nb_bitmaps
;
403 s
->bitmap_directory_offset
=
404 bitmaps_ext
.bitmap_directory_offset
;
405 s
->bitmap_directory_size
=
406 bitmaps_ext
.bitmap_directory_size
;
409 printf("Qcow2: Got bitmaps extension: "
410 "offset=%" PRIu64
" nb_bitmaps=%" PRIu32
"\n",
411 s
->bitmap_directory_offset
, s
->nb_bitmaps
);
415 case QCOW2_EXT_MAGIC_DATA_FILE
:
417 s
->image_data_file
= g_malloc0(ext
.len
+ 1);
418 ret
= bdrv_pread(bs
->file
, offset
, s
->image_data_file
, ext
.len
);
420 error_setg_errno(errp
, -ret
,
421 "ERROR: Could not read data file name");
425 printf("Qcow2: Got external data file %s\n", s
->image_data_file
);
431 /* unknown magic - save it in case we need to rewrite the header */
432 /* If you add a new feature, make sure to also update the fast
433 * path of qcow2_make_empty() to deal with it. */
435 Qcow2UnknownHeaderExtension
*uext
;
437 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
438 uext
->magic
= ext
.magic
;
440 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
442 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
444 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
445 "Could not read data");
452 offset
+= ((ext
.len
+ 7) & ~7);
458 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
460 BDRVQcow2State
*s
= bs
->opaque
;
461 Qcow2UnknownHeaderExtension
*uext
, *next
;
463 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
464 QLIST_REMOVE(uext
, next
);
469 static void report_unsupported_feature(Error
**errp
, Qcow2Feature
*table
,
472 g_autoptr(GString
) features
= g_string_sized_new(60);
474 while (table
&& table
->name
[0] != '\0') {
475 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
476 if (mask
& (1ULL << table
->bit
)) {
477 if (features
->len
> 0) {
478 g_string_append(features
, ", ");
480 g_string_append_printf(features
, "%.46s", table
->name
);
481 mask
&= ~(1ULL << table
->bit
);
488 if (features
->len
> 0) {
489 g_string_append(features
, ", ");
491 g_string_append_printf(features
,
492 "Unknown incompatible feature: %" PRIx64
, mask
);
495 error_setg(errp
, "Unsupported qcow2 feature(s): %s", features
->str
);
499 * Sets the dirty bit and flushes afterwards if necessary.
501 * The incompatible_features bit is only set if the image file header was
502 * updated successfully. Therefore it is not required to check the return
503 * value of this function.
505 int qcow2_mark_dirty(BlockDriverState
*bs
)
507 BDRVQcow2State
*s
= bs
->opaque
;
511 assert(s
->qcow_version
>= 3);
513 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
514 return 0; /* already dirty */
517 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
518 ret
= bdrv_pwrite(bs
->file
, offsetof(QCowHeader
, incompatible_features
),
523 ret
= bdrv_flush(bs
->file
->bs
);
528 /* Only treat image as dirty if the header was updated successfully */
529 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
534 * Clears the dirty bit and flushes before if necessary. Only call this
535 * function when there are no pending requests, it does not guard against
536 * concurrent requests dirtying the image.
538 static int qcow2_mark_clean(BlockDriverState
*bs
)
540 BDRVQcow2State
*s
= bs
->opaque
;
542 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
545 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
547 ret
= qcow2_flush_caches(bs
);
552 return qcow2_update_header(bs
);
558 * Marks the image as corrupt.
560 int qcow2_mark_corrupt(BlockDriverState
*bs
)
562 BDRVQcow2State
*s
= bs
->opaque
;
564 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
565 return qcow2_update_header(bs
);
569 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
570 * before if necessary.
572 int qcow2_mark_consistent(BlockDriverState
*bs
)
574 BDRVQcow2State
*s
= bs
->opaque
;
576 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
577 int ret
= qcow2_flush_caches(bs
);
582 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
583 return qcow2_update_header(bs
);
588 static void qcow2_add_check_result(BdrvCheckResult
*out
,
589 const BdrvCheckResult
*src
,
590 bool set_allocation_info
)
592 out
->corruptions
+= src
->corruptions
;
593 out
->leaks
+= src
->leaks
;
594 out
->check_errors
+= src
->check_errors
;
595 out
->corruptions_fixed
+= src
->corruptions_fixed
;
596 out
->leaks_fixed
+= src
->leaks_fixed
;
598 if (set_allocation_info
) {
599 out
->image_end_offset
= src
->image_end_offset
;
604 static int coroutine_fn
qcow2_co_check_locked(BlockDriverState
*bs
,
605 BdrvCheckResult
*result
,
608 BdrvCheckResult snapshot_res
= {};
609 BdrvCheckResult refcount_res
= {};
612 memset(result
, 0, sizeof(*result
));
614 ret
= qcow2_check_read_snapshot_table(bs
, &snapshot_res
, fix
);
616 qcow2_add_check_result(result
, &snapshot_res
, false);
620 ret
= qcow2_check_refcounts(bs
, &refcount_res
, fix
);
621 qcow2_add_check_result(result
, &refcount_res
, true);
623 qcow2_add_check_result(result
, &snapshot_res
, false);
627 ret
= qcow2_check_fix_snapshot_table(bs
, &snapshot_res
, fix
);
628 qcow2_add_check_result(result
, &snapshot_res
, false);
633 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
634 ret
= qcow2_mark_clean(bs
);
638 return qcow2_mark_consistent(bs
);
643 static int coroutine_fn
qcow2_co_check(BlockDriverState
*bs
,
644 BdrvCheckResult
*result
,
647 BDRVQcow2State
*s
= bs
->opaque
;
650 qemu_co_mutex_lock(&s
->lock
);
651 ret
= qcow2_co_check_locked(bs
, result
, fix
);
652 qemu_co_mutex_unlock(&s
->lock
);
656 int qcow2_validate_table(BlockDriverState
*bs
, uint64_t offset
,
657 uint64_t entries
, size_t entry_len
,
658 int64_t max_size_bytes
, const char *table_name
,
661 BDRVQcow2State
*s
= bs
->opaque
;
663 if (entries
> max_size_bytes
/ entry_len
) {
664 error_setg(errp
, "%s too large", table_name
);
668 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
669 * because values will be passed to qemu functions taking int64_t. */
670 if ((INT64_MAX
- entries
* entry_len
< offset
) ||
671 (offset_into_cluster(s
, offset
) != 0)) {
672 error_setg(errp
, "%s offset invalid", table_name
);
679 static const char *const mutable_opts
[] = {
680 QCOW2_OPT_LAZY_REFCOUNTS
,
681 QCOW2_OPT_DISCARD_REQUEST
,
682 QCOW2_OPT_DISCARD_SNAPSHOT
,
683 QCOW2_OPT_DISCARD_OTHER
,
685 QCOW2_OPT_OVERLAP_TEMPLATE
,
686 QCOW2_OPT_OVERLAP_MAIN_HEADER
,
687 QCOW2_OPT_OVERLAP_ACTIVE_L1
,
688 QCOW2_OPT_OVERLAP_ACTIVE_L2
,
689 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
690 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
691 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
692 QCOW2_OPT_OVERLAP_INACTIVE_L1
,
693 QCOW2_OPT_OVERLAP_INACTIVE_L2
,
694 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
695 QCOW2_OPT_CACHE_SIZE
,
696 QCOW2_OPT_L2_CACHE_SIZE
,
697 QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
698 QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
699 QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
703 static QemuOptsList qcow2_runtime_opts
= {
705 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
708 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
709 .type
= QEMU_OPT_BOOL
,
710 .help
= "Postpone refcount updates",
713 .name
= QCOW2_OPT_DISCARD_REQUEST
,
714 .type
= QEMU_OPT_BOOL
,
715 .help
= "Pass guest discard requests to the layer below",
718 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
719 .type
= QEMU_OPT_BOOL
,
720 .help
= "Generate discard requests when snapshot related space "
724 .name
= QCOW2_OPT_DISCARD_OTHER
,
725 .type
= QEMU_OPT_BOOL
,
726 .help
= "Generate discard requests when other clusters are freed",
729 .name
= QCOW2_OPT_OVERLAP
,
730 .type
= QEMU_OPT_STRING
,
731 .help
= "Selects which overlap checks to perform from a range of "
732 "templates (none, constant, cached, all)",
735 .name
= QCOW2_OPT_OVERLAP_TEMPLATE
,
736 .type
= QEMU_OPT_STRING
,
737 .help
= "Selects which overlap checks to perform from a range of "
738 "templates (none, constant, cached, all)",
741 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
742 .type
= QEMU_OPT_BOOL
,
743 .help
= "Check for unintended writes into the main qcow2 header",
746 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
747 .type
= QEMU_OPT_BOOL
,
748 .help
= "Check for unintended writes into the active L1 table",
751 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
752 .type
= QEMU_OPT_BOOL
,
753 .help
= "Check for unintended writes into an active L2 table",
756 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
757 .type
= QEMU_OPT_BOOL
,
758 .help
= "Check for unintended writes into the refcount table",
761 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
762 .type
= QEMU_OPT_BOOL
,
763 .help
= "Check for unintended writes into a refcount block",
766 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
767 .type
= QEMU_OPT_BOOL
,
768 .help
= "Check for unintended writes into the snapshot table",
771 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
772 .type
= QEMU_OPT_BOOL
,
773 .help
= "Check for unintended writes into an inactive L1 table",
776 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
777 .type
= QEMU_OPT_BOOL
,
778 .help
= "Check for unintended writes into an inactive L2 table",
781 .name
= QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
782 .type
= QEMU_OPT_BOOL
,
783 .help
= "Check for unintended writes into the bitmap directory",
786 .name
= QCOW2_OPT_CACHE_SIZE
,
787 .type
= QEMU_OPT_SIZE
,
788 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
792 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
793 .type
= QEMU_OPT_SIZE
,
794 .help
= "Maximum L2 table cache size",
797 .name
= QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
798 .type
= QEMU_OPT_SIZE
,
799 .help
= "Size of each entry in the L2 cache",
802 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
803 .type
= QEMU_OPT_SIZE
,
804 .help
= "Maximum refcount block cache size",
807 .name
= QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
808 .type
= QEMU_OPT_NUMBER
,
809 .help
= "Clean unused cache entries after this time (in seconds)",
811 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
812 "ID of secret providing qcow2 AES key or LUKS passphrase"),
813 { /* end of list */ }
817 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
818 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
819 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
820 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
821 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
822 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
823 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
824 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
825 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
826 [QCOW2_OL_BITMAP_DIRECTORY_BITNR
] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
829 static void cache_clean_timer_cb(void *opaque
)
831 BlockDriverState
*bs
= opaque
;
832 BDRVQcow2State
*s
= bs
->opaque
;
833 qcow2_cache_clean_unused(s
->l2_table_cache
);
834 qcow2_cache_clean_unused(s
->refcount_block_cache
);
835 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
836 (int64_t) s
->cache_clean_interval
* 1000);
839 static void cache_clean_timer_init(BlockDriverState
*bs
, AioContext
*context
)
841 BDRVQcow2State
*s
= bs
->opaque
;
842 if (s
->cache_clean_interval
> 0) {
843 s
->cache_clean_timer
= aio_timer_new(context
, QEMU_CLOCK_VIRTUAL
,
844 SCALE_MS
, cache_clean_timer_cb
,
846 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
847 (int64_t) s
->cache_clean_interval
* 1000);
851 static void cache_clean_timer_del(BlockDriverState
*bs
)
853 BDRVQcow2State
*s
= bs
->opaque
;
854 if (s
->cache_clean_timer
) {
855 timer_del(s
->cache_clean_timer
);
856 timer_free(s
->cache_clean_timer
);
857 s
->cache_clean_timer
= NULL
;
861 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
863 cache_clean_timer_del(bs
);
866 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
867 AioContext
*new_context
)
869 cache_clean_timer_init(bs
, new_context
);
872 static void read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
873 uint64_t *l2_cache_size
,
874 uint64_t *l2_cache_entry_size
,
875 uint64_t *refcount_cache_size
, Error
**errp
)
877 BDRVQcow2State
*s
= bs
->opaque
;
878 uint64_t combined_cache_size
, l2_cache_max_setting
;
879 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
880 bool l2_cache_entry_size_set
;
881 int min_refcount_cache
= MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
882 uint64_t virtual_disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
883 uint64_t max_l2_entries
= DIV_ROUND_UP(virtual_disk_size
, s
->cluster_size
);
884 /* An L2 table is always one cluster in size so the max cache size
885 * should be a multiple of the cluster size. */
886 uint64_t max_l2_cache
= ROUND_UP(max_l2_entries
* l2_entry_size(s
),
889 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
890 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
891 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
892 l2_cache_entry_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
);
894 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
895 l2_cache_max_setting
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
,
896 DEFAULT_L2_CACHE_MAX_SIZE
);
897 *refcount_cache_size
= qemu_opt_get_size(opts
,
898 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
900 *l2_cache_entry_size
= qemu_opt_get_size(
901 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
903 *l2_cache_size
= MIN(max_l2_cache
, l2_cache_max_setting
);
905 if (combined_cache_size_set
) {
906 if (l2_cache_size_set
&& refcount_cache_size_set
) {
907 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
908 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
911 } else if (l2_cache_size_set
&&
912 (l2_cache_max_setting
> combined_cache_size
)) {
913 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
914 QCOW2_OPT_CACHE_SIZE
);
916 } else if (*refcount_cache_size
> combined_cache_size
) {
917 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
918 QCOW2_OPT_CACHE_SIZE
);
922 if (l2_cache_size_set
) {
923 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
924 } else if (refcount_cache_size_set
) {
925 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
927 /* Assign as much memory as possible to the L2 cache, and
928 * use the remainder for the refcount cache */
929 if (combined_cache_size
>= max_l2_cache
+ min_refcount_cache
) {
930 *l2_cache_size
= max_l2_cache
;
931 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
933 *refcount_cache_size
=
934 MIN(combined_cache_size
, min_refcount_cache
);
935 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
941 * If the L2 cache is not enough to cover the whole disk then
942 * default to 4KB entries. Smaller entries reduce the cost of
943 * loads and evictions and increase I/O performance.
945 if (*l2_cache_size
< max_l2_cache
&& !l2_cache_entry_size_set
) {
946 *l2_cache_entry_size
= MIN(s
->cluster_size
, 4096);
949 /* l2_cache_size and refcount_cache_size are ensured to have at least
950 * their minimum values in qcow2_update_options_prepare() */
952 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
953 *l2_cache_entry_size
> s
->cluster_size
||
954 !is_power_of_2(*l2_cache_entry_size
)) {
955 error_setg(errp
, "L2 cache entry size must be a power of two "
956 "between %d and the cluster size (%d)",
957 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
962 typedef struct Qcow2ReopenState
{
963 Qcow2Cache
*l2_table_cache
;
964 Qcow2Cache
*refcount_block_cache
;
965 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
966 bool use_lazy_refcounts
;
968 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
969 uint64_t cache_clean_interval
;
970 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
973 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
975 QDict
*options
, int flags
,
978 BDRVQcow2State
*s
= bs
->opaque
;
979 QemuOpts
*opts
= NULL
;
980 const char *opt_overlap_check
, *opt_overlap_check_template
;
981 int overlap_check_template
= 0;
982 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
984 const char *encryptfmt
;
985 QDict
*encryptopts
= NULL
;
986 Error
*local_err
= NULL
;
989 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
990 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
992 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
993 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
998 /* get L2 table/refcount block cache size from command line options */
999 read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
1000 &refcount_cache_size
, &local_err
);
1002 error_propagate(errp
, local_err
);
1007 l2_cache_size
/= l2_cache_entry_size
;
1008 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
1009 l2_cache_size
= MIN_L2_CACHE_SIZE
;
1011 if (l2_cache_size
> INT_MAX
) {
1012 error_setg(errp
, "L2 cache size too big");
1017 refcount_cache_size
/= s
->cluster_size
;
1018 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
1019 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
1021 if (refcount_cache_size
> INT_MAX
) {
1022 error_setg(errp
, "Refcount cache size too big");
1027 /* alloc new L2 table/refcount block cache, flush old one */
1028 if (s
->l2_table_cache
) {
1029 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1031 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
1036 if (s
->refcount_block_cache
) {
1037 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1039 error_setg_errno(errp
, -ret
,
1040 "Failed to flush the refcount block cache");
1045 r
->l2_slice_size
= l2_cache_entry_size
/ l2_entry_size(s
);
1046 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
1047 l2_cache_entry_size
);
1048 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
1050 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
1051 error_setg(errp
, "Could not allocate metadata caches");
1056 /* New interval for cache cleanup timer */
1057 r
->cache_clean_interval
=
1058 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
1059 DEFAULT_CACHE_CLEAN_INTERVAL
);
1060 #ifndef CONFIG_LINUX
1061 if (r
->cache_clean_interval
!= 0) {
1062 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1063 " not supported on this host");
1068 if (r
->cache_clean_interval
> UINT_MAX
) {
1069 error_setg(errp
, "Cache clean interval too big");
1074 /* lazy-refcounts; flush if going from enabled to disabled */
1075 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
1076 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
1077 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
1078 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
1079 "qemu 1.1 compatibility level");
1084 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
1085 ret
= qcow2_mark_clean(bs
);
1087 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
1092 /* Overlap check options */
1093 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
1094 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1095 if (opt_overlap_check_template
&& opt_overlap_check
&&
1096 strcmp(opt_overlap_check_template
, opt_overlap_check
))
1098 error_setg(errp
, "Conflicting values for qcow2 options '"
1099 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1100 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
1104 if (!opt_overlap_check
) {
1105 opt_overlap_check
= opt_overlap_check_template
?: "cached";
1108 if (!strcmp(opt_overlap_check
, "none")) {
1109 overlap_check_template
= 0;
1110 } else if (!strcmp(opt_overlap_check
, "constant")) {
1111 overlap_check_template
= QCOW2_OL_CONSTANT
;
1112 } else if (!strcmp(opt_overlap_check
, "cached")) {
1113 overlap_check_template
= QCOW2_OL_CACHED
;
1114 } else if (!strcmp(opt_overlap_check
, "all")) {
1115 overlap_check_template
= QCOW2_OL_ALL
;
1117 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1118 "'overlap-check'. Allowed are any of the following: "
1119 "none, constant, cached, all", opt_overlap_check
);
1124 r
->overlap_check
= 0;
1125 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1126 /* overlap-check defines a template bitmask, but every flag may be
1127 * overwritten through the associated boolean option */
1129 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1130 overlap_check_template
& (1 << i
)) << i
;
1133 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1134 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1135 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1136 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1137 flags
& BDRV_O_UNMAP
);
1138 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1139 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1140 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1141 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1143 switch (s
->crypt_method_header
) {
1144 case QCOW_CRYPT_NONE
:
1146 error_setg(errp
, "No encryption in image header, but options "
1147 "specified format '%s'", encryptfmt
);
1153 case QCOW_CRYPT_AES
:
1154 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1156 "Header reported 'aes' encryption format but "
1157 "options specify '%s'", encryptfmt
);
1161 qdict_put_str(encryptopts
, "format", "qcow");
1162 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1165 case QCOW_CRYPT_LUKS
:
1166 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1168 "Header reported 'luks' encryption format but "
1169 "options specify '%s'", encryptfmt
);
1173 qdict_put_str(encryptopts
, "format", "luks");
1174 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1178 error_setg(errp
, "Unsupported encryption method %d",
1179 s
->crypt_method_header
);
1182 if (s
->crypt_method_header
!= QCOW_CRYPT_NONE
&& !r
->crypto_opts
) {
1189 qobject_unref(encryptopts
);
1190 qemu_opts_del(opts
);
1195 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1196 Qcow2ReopenState
*r
)
1198 BDRVQcow2State
*s
= bs
->opaque
;
1201 if (s
->l2_table_cache
) {
1202 qcow2_cache_destroy(s
->l2_table_cache
);
1204 if (s
->refcount_block_cache
) {
1205 qcow2_cache_destroy(s
->refcount_block_cache
);
1207 s
->l2_table_cache
= r
->l2_table_cache
;
1208 s
->refcount_block_cache
= r
->refcount_block_cache
;
1209 s
->l2_slice_size
= r
->l2_slice_size
;
1211 s
->overlap_check
= r
->overlap_check
;
1212 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1214 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1215 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1218 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1219 cache_clean_timer_del(bs
);
1220 s
->cache_clean_interval
= r
->cache_clean_interval
;
1221 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1224 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1225 s
->crypto_opts
= r
->crypto_opts
;
1228 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1229 Qcow2ReopenState
*r
)
1231 if (r
->l2_table_cache
) {
1232 qcow2_cache_destroy(r
->l2_table_cache
);
1234 if (r
->refcount_block_cache
) {
1235 qcow2_cache_destroy(r
->refcount_block_cache
);
1237 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1240 static int qcow2_update_options(BlockDriverState
*bs
, QDict
*options
,
1241 int flags
, Error
**errp
)
1243 Qcow2ReopenState r
= {};
1246 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1248 qcow2_update_options_commit(bs
, &r
);
1250 qcow2_update_options_abort(bs
, &r
);
1256 static int validate_compression_type(BDRVQcow2State
*s
, Error
**errp
)
1258 switch (s
->compression_type
) {
1259 case QCOW2_COMPRESSION_TYPE_ZLIB
:
1261 case QCOW2_COMPRESSION_TYPE_ZSTD
:
1266 error_setg(errp
, "qcow2: unknown compression type: %u",
1267 s
->compression_type
);
1272 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1273 * the incompatible feature flag must be set
1275 if (s
->compression_type
== QCOW2_COMPRESSION_TYPE_ZLIB
) {
1276 if (s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
) {
1277 error_setg(errp
, "qcow2: Compression type incompatible feature "
1278 "bit must not be set");
1282 if (!(s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
)) {
1283 error_setg(errp
, "qcow2: Compression type incompatible feature "
1292 /* Called with s->lock held. */
1293 static int coroutine_fn
qcow2_do_open(BlockDriverState
*bs
, QDict
*options
,
1294 int flags
, Error
**errp
)
1296 BDRVQcow2State
*s
= bs
->opaque
;
1297 unsigned int len
, i
;
1300 Error
*local_err
= NULL
;
1302 uint64_t l1_vm_state_index
;
1303 bool update_header
= false;
1305 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
1307 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1310 header
.magic
= be32_to_cpu(header
.magic
);
1311 header
.version
= be32_to_cpu(header
.version
);
1312 header
.backing_file_offset
= be64_to_cpu(header
.backing_file_offset
);
1313 header
.backing_file_size
= be32_to_cpu(header
.backing_file_size
);
1314 header
.size
= be64_to_cpu(header
.size
);
1315 header
.cluster_bits
= be32_to_cpu(header
.cluster_bits
);
1316 header
.crypt_method
= be32_to_cpu(header
.crypt_method
);
1317 header
.l1_table_offset
= be64_to_cpu(header
.l1_table_offset
);
1318 header
.l1_size
= be32_to_cpu(header
.l1_size
);
1319 header
.refcount_table_offset
= be64_to_cpu(header
.refcount_table_offset
);
1320 header
.refcount_table_clusters
=
1321 be32_to_cpu(header
.refcount_table_clusters
);
1322 header
.snapshots_offset
= be64_to_cpu(header
.snapshots_offset
);
1323 header
.nb_snapshots
= be32_to_cpu(header
.nb_snapshots
);
1325 if (header
.magic
!= QCOW_MAGIC
) {
1326 error_setg(errp
, "Image is not in qcow2 format");
1330 if (header
.version
< 2 || header
.version
> 3) {
1331 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1336 s
->qcow_version
= header
.version
;
1338 /* Initialise cluster size */
1339 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1340 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1341 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1342 header
.cluster_bits
);
1347 s
->cluster_bits
= header
.cluster_bits
;
1348 s
->cluster_size
= 1 << s
->cluster_bits
;
1350 /* Initialise version 3 header fields */
1351 if (header
.version
== 2) {
1352 header
.incompatible_features
= 0;
1353 header
.compatible_features
= 0;
1354 header
.autoclear_features
= 0;
1355 header
.refcount_order
= 4;
1356 header
.header_length
= 72;
1358 header
.incompatible_features
=
1359 be64_to_cpu(header
.incompatible_features
);
1360 header
.compatible_features
= be64_to_cpu(header
.compatible_features
);
1361 header
.autoclear_features
= be64_to_cpu(header
.autoclear_features
);
1362 header
.refcount_order
= be32_to_cpu(header
.refcount_order
);
1363 header
.header_length
= be32_to_cpu(header
.header_length
);
1365 if (header
.header_length
< 104) {
1366 error_setg(errp
, "qcow2 header too short");
1372 if (header
.header_length
> s
->cluster_size
) {
1373 error_setg(errp
, "qcow2 header exceeds cluster size");
1378 if (header
.header_length
> sizeof(header
)) {
1379 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1380 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1381 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
1382 s
->unknown_header_fields_size
);
1384 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1390 if (header
.backing_file_offset
> s
->cluster_size
) {
1391 error_setg(errp
, "Invalid backing file offset");
1396 if (header
.backing_file_offset
) {
1397 ext_end
= header
.backing_file_offset
;
1399 ext_end
= 1 << header
.cluster_bits
;
1402 /* Handle feature bits */
1403 s
->incompatible_features
= header
.incompatible_features
;
1404 s
->compatible_features
= header
.compatible_features
;
1405 s
->autoclear_features
= header
.autoclear_features
;
1408 * Handle compression type
1409 * Older qcow2 images don't contain the compression type header.
1410 * Distinguish them by the header length and use
1411 * the only valid (default) compression type in that case
1413 if (header
.header_length
> offsetof(QCowHeader
, compression_type
)) {
1414 s
->compression_type
= header
.compression_type
;
1416 s
->compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
1419 ret
= validate_compression_type(s
, errp
);
1424 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1425 void *feature_table
= NULL
;
1426 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1427 &feature_table
, flags
, NULL
, NULL
);
1428 report_unsupported_feature(errp
, feature_table
,
1429 s
->incompatible_features
&
1430 ~QCOW2_INCOMPAT_MASK
);
1432 g_free(feature_table
);
1436 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1437 /* Corrupt images may not be written to unless they are being repaired
1439 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1440 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1447 s
->subclusters_per_cluster
=
1448 has_subclusters(s
) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
: 1;
1449 s
->subcluster_size
= s
->cluster_size
/ s
->subclusters_per_cluster
;
1450 s
->subcluster_bits
= ctz32(s
->subcluster_size
);
1452 /* Check support for various header values */
1453 if (header
.refcount_order
> 6) {
1454 error_setg(errp
, "Reference count entry width too large; may not "
1459 s
->refcount_order
= header
.refcount_order
;
1460 s
->refcount_bits
= 1 << s
->refcount_order
;
1461 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1462 s
->refcount_max
+= s
->refcount_max
- 1;
1464 s
->crypt_method_header
= header
.crypt_method
;
1465 if (s
->crypt_method_header
) {
1466 if (bdrv_uses_whitelist() &&
1467 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1469 "Use of AES-CBC encrypted qcow2 images is no longer "
1470 "supported in system emulators");
1471 error_append_hint(errp
,
1472 "You can use 'qemu-img convert' to convert your "
1473 "image to an alternative supported format, such "
1474 "as unencrypted qcow2, or raw with the LUKS "
1475 "format instead.\n");
1480 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1481 s
->crypt_physical_offset
= false;
1483 /* Assuming LUKS and any future crypt methods we
1484 * add will all use physical offsets, due to the
1485 * fact that the alternative is insecure... */
1486 s
->crypt_physical_offset
= true;
1489 bs
->encrypted
= true;
1492 s
->l2_bits
= s
->cluster_bits
- ctz32(l2_entry_size(s
));
1493 s
->l2_size
= 1 << s
->l2_bits
;
1494 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1495 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1496 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1497 bs
->total_sectors
= header
.size
/ BDRV_SECTOR_SIZE
;
1498 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1499 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1500 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1502 s
->refcount_table_offset
= header
.refcount_table_offset
;
1503 s
->refcount_table_size
=
1504 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1506 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1507 error_setg(errp
, "Image does not contain a reference count table");
1512 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1513 header
.refcount_table_clusters
,
1514 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1515 "Reference count table", errp
);
1520 if (!(flags
& BDRV_O_CHECK
)) {
1522 * The total size in bytes of the snapshot table is checked in
1523 * qcow2_read_snapshots() because the size of each snapshot is
1524 * variable and we don't know it yet.
1525 * Here we only check the offset and number of snapshots.
1527 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1528 header
.nb_snapshots
,
1529 sizeof(QCowSnapshotHeader
),
1530 sizeof(QCowSnapshotHeader
) *
1532 "Snapshot table", errp
);
1538 /* read the level 1 table */
1539 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1540 header
.l1_size
, sizeof(uint64_t),
1541 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1545 s
->l1_size
= header
.l1_size
;
1546 s
->l1_table_offset
= header
.l1_table_offset
;
1548 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1549 if (l1_vm_state_index
> INT_MAX
) {
1550 error_setg(errp
, "Image is too big");
1554 s
->l1_vm_state_index
= l1_vm_state_index
;
1556 /* the L1 table must contain at least enough entries to put
1557 header.size bytes */
1558 if (s
->l1_size
< s
->l1_vm_state_index
) {
1559 error_setg(errp
, "L1 table is too small");
1564 if (s
->l1_size
> 0) {
1565 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
,
1566 s
->l1_size
* sizeof(uint64_t));
1567 if (s
->l1_table
== NULL
) {
1568 error_setg(errp
, "Could not allocate L1 table");
1572 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1573 s
->l1_size
* sizeof(uint64_t));
1575 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1578 for(i
= 0;i
< s
->l1_size
; i
++) {
1579 s
->l1_table
[i
] = be64_to_cpu(s
->l1_table
[i
]);
1583 /* Parse driver-specific options */
1584 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1591 ret
= qcow2_refcount_init(bs
);
1593 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1597 QLIST_INIT(&s
->cluster_allocs
);
1598 QTAILQ_INIT(&s
->discards
);
1600 /* read qcow2 extensions */
1601 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1602 flags
, &update_header
, errp
)) {
1607 /* Open external data file */
1608 s
->data_file
= bdrv_open_child(NULL
, options
, "data-file", bs
,
1609 &child_of_bds
, BDRV_CHILD_DATA
,
1612 error_propagate(errp
, local_err
);
1617 if (s
->incompatible_features
& QCOW2_INCOMPAT_DATA_FILE
) {
1618 if (!s
->data_file
&& s
->image_data_file
) {
1619 s
->data_file
= bdrv_open_child(s
->image_data_file
, options
,
1620 "data-file", bs
, &child_of_bds
,
1621 BDRV_CHILD_DATA
, false, errp
);
1622 if (!s
->data_file
) {
1627 if (!s
->data_file
) {
1628 error_setg(errp
, "'data-file' is required for this image");
1634 bs
->file
->role
&= ~BDRV_CHILD_DATA
;
1636 /* Must succeed because we have given up permissions if anything */
1637 bdrv_child_refresh_perms(bs
, bs
->file
, &error_abort
);
1640 error_setg(errp
, "'data-file' can only be set for images with an "
1641 "external data file");
1646 s
->data_file
= bs
->file
;
1648 if (data_file_is_raw(bs
)) {
1649 error_setg(errp
, "data-file-raw requires a data file");
1655 /* qcow2_read_extension may have set up the crypto context
1656 * if the crypt method needs a header region, some methods
1657 * don't need header extensions, so must check here
1659 if (s
->crypt_method_header
&& !s
->crypto
) {
1660 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1661 unsigned int cflags
= 0;
1662 if (flags
& BDRV_O_NO_IO
) {
1663 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1665 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1667 QCOW2_MAX_THREADS
, errp
);
1672 } else if (!(flags
& BDRV_O_NO_IO
)) {
1673 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1674 s
->crypt_method_header
);
1680 /* read the backing file name */
1681 if (header
.backing_file_offset
!= 0) {
1682 len
= header
.backing_file_size
;
1683 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1684 len
>= sizeof(bs
->backing_file
)) {
1685 error_setg(errp
, "Backing file name too long");
1689 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
1690 bs
->auto_backing_file
, len
);
1692 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1695 bs
->auto_backing_file
[len
] = '\0';
1696 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
),
1697 bs
->auto_backing_file
);
1698 s
->image_backing_file
= g_strdup(bs
->auto_backing_file
);
1702 * Internal snapshots; skip reading them in check mode, because
1703 * we do not need them then, and we do not want to abort because
1704 * of a broken table.
1706 if (!(flags
& BDRV_O_CHECK
)) {
1707 s
->snapshots_offset
= header
.snapshots_offset
;
1708 s
->nb_snapshots
= header
.nb_snapshots
;
1710 ret
= qcow2_read_snapshots(bs
, errp
);
1716 /* Clear unknown autoclear feature bits */
1717 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1719 update_header
&& !bs
->read_only
&& !(flags
& BDRV_O_INACTIVE
);
1720 if (update_header
) {
1721 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1724 /* == Handle persistent dirty bitmaps ==
1726 * We want load dirty bitmaps in three cases:
1728 * 1. Normal open of the disk in active mode, not related to invalidation
1731 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1732 * bitmaps are _not_ migrating through migration channel, i.e.
1733 * 'dirty-bitmaps' capability is disabled.
1735 * 3. Invalidation of source vm after failed or canceled migration.
1736 * This is a very interesting case. There are two possible types of
1739 * A. Stored on inactivation and removed. They should be loaded from the
1742 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1743 * the migration channel (with dirty-bitmaps capability).
1745 * On the other hand, there are two possible sub-cases:
1747 * 3.1 disk was changed by somebody else while were inactive. In this
1748 * case all in-RAM dirty bitmaps (both persistent and not) are
1749 * definitely invalid. And we don't have any method to determine
1752 * Simple and safe thing is to just drop all the bitmaps of type B on
1753 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1755 * On the other hand, resuming source vm, if disk was already changed
1756 * is a bad thing anyway: not only bitmaps, the whole vm state is
1757 * out of sync with disk.
1759 * This means, that user or management tool, who for some reason
1760 * decided to resume source vm, after disk was already changed by
1761 * target vm, should at least drop all dirty bitmaps by hand.
1763 * So, we can ignore this case for now, but TODO: "generation"
1764 * extension for qcow2, to determine, that image was changed after
1765 * last inactivation. And if it is changed, we will drop (or at least
1766 * mark as 'invalid' all the bitmaps of type B, both persistent
1769 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1770 * to disk ('dirty-bitmaps' capability disabled), or not saved
1771 * ('dirty-bitmaps' capability enabled), but we don't need to care
1772 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1773 * and not stored has flag IN_USE=1 in the image and will be skipped
1776 * One remaining possible case when we don't want load bitmaps:
1778 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1779 * will be loaded on invalidation, no needs try loading them before)
1782 if (!(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
)) {
1783 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1784 bool header_updated
= qcow2_load_dirty_bitmaps(bs
, &local_err
);
1785 if (local_err
!= NULL
) {
1786 error_propagate(errp
, local_err
);
1791 update_header
= update_header
&& !header_updated
;
1794 if (update_header
) {
1795 ret
= qcow2_update_header(bs
);
1797 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1802 bs
->supported_zero_flags
= header
.version
>= 3 ?
1803 BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
: 0;
1804 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
1806 /* Repair image if dirty */
1807 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INACTIVE
)) && !bs
->read_only
&&
1808 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1809 BdrvCheckResult result
= {0};
1811 ret
= qcow2_co_check_locked(bs
, &result
,
1812 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1813 if (ret
< 0 || result
.check_errors
) {
1817 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1824 BdrvCheckResult result
= {0};
1825 qcow2_check_refcounts(bs
, &result
, 0);
1829 qemu_co_queue_init(&s
->thread_task_queue
);
1834 g_free(s
->image_data_file
);
1835 if (has_data_file(bs
)) {
1836 bdrv_unref_child(bs
, s
->data_file
);
1837 s
->data_file
= NULL
;
1839 g_free(s
->unknown_header_fields
);
1840 cleanup_unknown_header_ext(bs
);
1841 qcow2_free_snapshots(bs
);
1842 qcow2_refcount_close(bs
);
1843 qemu_vfree(s
->l1_table
);
1844 /* else pre-write overlap checks in cache_destroy may crash */
1846 cache_clean_timer_del(bs
);
1847 if (s
->l2_table_cache
) {
1848 qcow2_cache_destroy(s
->l2_table_cache
);
1850 if (s
->refcount_block_cache
) {
1851 qcow2_cache_destroy(s
->refcount_block_cache
);
1853 qcrypto_block_free(s
->crypto
);
1854 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1858 typedef struct QCow2OpenCo
{
1859 BlockDriverState
*bs
;
1866 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1868 QCow2OpenCo
*qoc
= opaque
;
1869 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1871 qemu_co_mutex_lock(&s
->lock
);
1872 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, qoc
->errp
);
1873 qemu_co_mutex_unlock(&s
->lock
);
1876 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1879 BDRVQcow2State
*s
= bs
->opaque
;
1888 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_of_bds
,
1889 BDRV_CHILD_IMAGE
, false, errp
);
1894 /* Initialise locks */
1895 qemu_co_mutex_init(&s
->lock
);
1897 if (qemu_in_coroutine()) {
1898 /* From bdrv_co_create. */
1899 qcow2_open_entry(&qoc
);
1901 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1902 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1903 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1908 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1910 BDRVQcow2State
*s
= bs
->opaque
;
1912 if (bs
->encrypted
) {
1913 /* Encryption works on a sector granularity */
1914 bs
->bl
.request_alignment
= qcrypto_block_get_sector_size(s
->crypto
);
1916 bs
->bl
.pwrite_zeroes_alignment
= s
->subcluster_size
;
1917 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1920 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1921 BlockReopenQueue
*queue
, Error
**errp
)
1923 Qcow2ReopenState
*r
;
1926 r
= g_new0(Qcow2ReopenState
, 1);
1929 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1930 state
->flags
, errp
);
1935 /* We need to write out any unwritten data if we reopen read-only. */
1936 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1937 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1942 ret
= bdrv_flush(state
->bs
);
1947 ret
= qcow2_mark_clean(state
->bs
);
1956 qcow2_update_options_abort(state
->bs
, r
);
1961 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1963 qcow2_update_options_commit(state
->bs
, state
->opaque
);
1964 g_free(state
->opaque
);
1967 static void qcow2_reopen_commit_post(BDRVReopenState
*state
)
1969 if (state
->flags
& BDRV_O_RDWR
) {
1970 Error
*local_err
= NULL
;
1972 if (qcow2_reopen_bitmaps_rw(state
->bs
, &local_err
) < 0) {
1974 * This is not fatal, bitmaps just left read-only, so all following
1975 * writes will fail. User can remove read-only bitmaps to unblock
1976 * writes or retry reopen.
1978 error_reportf_err(local_err
,
1979 "%s: Failed to make dirty bitmaps writable: ",
1980 bdrv_get_node_name(state
->bs
));
1985 static void qcow2_reopen_abort(BDRVReopenState
*state
)
1987 qcow2_update_options_abort(state
->bs
, state
->opaque
);
1988 g_free(state
->opaque
);
1991 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
1993 bool has_new_overlap_template
=
1994 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
1995 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1996 bool has_new_total_cache_size
=
1997 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
1998 bool has_all_cache_options
;
2000 /* New overlap template overrides all old overlap options */
2001 if (has_new_overlap_template
) {
2002 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
2003 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2004 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
2005 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
2006 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
2007 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
2008 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
2009 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
2010 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
2011 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
2014 /* New total cache size overrides all old options */
2015 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
2016 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
2017 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2020 qdict_join(options
, old_options
, false);
2023 * If after merging all cache size options are set, an old total size is
2024 * overwritten. Do keep all options, however, if all three are new. The
2025 * resulting error message is what we want to happen.
2027 has_all_cache_options
=
2028 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
2029 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
2030 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2032 if (has_all_cache_options
&& !has_new_total_cache_size
) {
2033 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
2037 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
2039 int64_t offset
, int64_t count
,
2040 int64_t *pnum
, int64_t *map
,
2041 BlockDriverState
**file
)
2043 BDRVQcow2State
*s
= bs
->opaque
;
2044 uint64_t host_offset
;
2046 QCow2SubclusterType type
;
2047 int ret
, status
= 0;
2049 qemu_co_mutex_lock(&s
->lock
);
2051 if (!s
->metadata_preallocation_checked
) {
2052 ret
= qcow2_detect_metadata_preallocation(bs
);
2053 s
->metadata_preallocation
= (ret
== 1);
2054 s
->metadata_preallocation_checked
= true;
2057 bytes
= MIN(INT_MAX
, count
);
2058 ret
= qcow2_get_host_offset(bs
, offset
, &bytes
, &host_offset
, &type
);
2059 qemu_co_mutex_unlock(&s
->lock
);
2066 if ((type
== QCOW2_SUBCLUSTER_NORMAL
||
2067 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2068 type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) && !s
->crypto
) {
2070 *file
= s
->data_file
->bs
;
2071 status
|= BDRV_BLOCK_OFFSET_VALID
;
2073 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2074 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
) {
2075 status
|= BDRV_BLOCK_ZERO
;
2076 } else if (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
2077 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) {
2078 status
|= BDRV_BLOCK_DATA
;
2080 if (s
->metadata_preallocation
&& (status
& BDRV_BLOCK_DATA
) &&
2081 (status
& BDRV_BLOCK_OFFSET_VALID
))
2083 status
|= BDRV_BLOCK_RECURSE
;
2088 static coroutine_fn
int qcow2_handle_l2meta(BlockDriverState
*bs
,
2089 QCowL2Meta
**pl2meta
,
2093 QCowL2Meta
*l2meta
= *pl2meta
;
2095 while (l2meta
!= NULL
) {
2099 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
2104 qcow2_alloc_cluster_abort(bs
, l2meta
);
2107 /* Take the request off the list of running requests */
2108 if (l2meta
->nb_clusters
!= 0) {
2109 QLIST_REMOVE(l2meta
, next_in_flight
);
2112 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2114 next
= l2meta
->next
;
2123 static coroutine_fn
int
2124 qcow2_co_preadv_encrypted(BlockDriverState
*bs
,
2125 uint64_t host_offset
,
2129 uint64_t qiov_offset
)
2132 BDRVQcow2State
*s
= bs
->opaque
;
2135 assert(bs
->encrypted
&& s
->crypto
);
2136 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2139 * For encrypted images, read everything into a temporary
2140 * contiguous buffer on which the AES functions can work.
2141 * Also, decryption in a separate buffer is better as it
2142 * prevents the guest from learning information about the
2143 * encrypted nature of the virtual disk.
2146 buf
= qemu_try_blockalign(s
->data_file
->bs
, bytes
);
2151 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2152 ret
= bdrv_co_pread(s
->data_file
, host_offset
, bytes
, buf
, 0);
2157 if (qcow2_co_decrypt(bs
, host_offset
, offset
, buf
, bytes
) < 0)
2162 qemu_iovec_from_buf(qiov
, qiov_offset
, buf
, bytes
);
2170 typedef struct Qcow2AioTask
{
2173 BlockDriverState
*bs
;
2174 QCow2SubclusterType subcluster_type
; /* only for read */
2175 uint64_t host_offset
; /* or full descriptor in compressed clusters */
2179 uint64_t qiov_offset
;
2180 QCowL2Meta
*l2meta
; /* only for write */
2183 static coroutine_fn
int qcow2_co_preadv_task_entry(AioTask
*task
);
2184 static coroutine_fn
int qcow2_add_task(BlockDriverState
*bs
,
2187 QCow2SubclusterType subcluster_type
,
2188 uint64_t host_offset
,
2195 Qcow2AioTask local_task
;
2196 Qcow2AioTask
*task
= pool
? g_new(Qcow2AioTask
, 1) : &local_task
;
2198 *task
= (Qcow2AioTask
) {
2201 .subcluster_type
= subcluster_type
,
2203 .host_offset
= host_offset
,
2206 .qiov_offset
= qiov_offset
,
2210 trace_qcow2_add_task(qemu_coroutine_self(), bs
, pool
,
2211 func
== qcow2_co_preadv_task_entry
? "read" : "write",
2212 subcluster_type
, host_offset
, offset
, bytes
,
2216 return func(&task
->task
);
2219 aio_task_pool_start_task(pool
, &task
->task
);
2224 static coroutine_fn
int qcow2_co_preadv_task(BlockDriverState
*bs
,
2225 QCow2SubclusterType subc_type
,
2226 uint64_t host_offset
,
2227 uint64_t offset
, uint64_t bytes
,
2231 BDRVQcow2State
*s
= bs
->opaque
;
2233 switch (subc_type
) {
2234 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
2235 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
2236 /* Both zero types are handled in qcow2_co_preadv_part */
2237 g_assert_not_reached();
2239 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
2240 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
2241 assert(bs
->backing
); /* otherwise handled in qcow2_co_preadv_part */
2243 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
2244 return bdrv_co_preadv_part(bs
->backing
, offset
, bytes
,
2245 qiov
, qiov_offset
, 0);
2247 case QCOW2_SUBCLUSTER_COMPRESSED
:
2248 return qcow2_co_preadv_compressed(bs
, host_offset
,
2249 offset
, bytes
, qiov
, qiov_offset
);
2251 case QCOW2_SUBCLUSTER_NORMAL
:
2252 if (bs
->encrypted
) {
2253 return qcow2_co_preadv_encrypted(bs
, host_offset
,
2254 offset
, bytes
, qiov
, qiov_offset
);
2257 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2258 return bdrv_co_preadv_part(s
->data_file
, host_offset
,
2259 bytes
, qiov
, qiov_offset
, 0);
2262 g_assert_not_reached();
2265 g_assert_not_reached();
2268 static coroutine_fn
int qcow2_co_preadv_task_entry(AioTask
*task
)
2270 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2274 return qcow2_co_preadv_task(t
->bs
, t
->subcluster_type
,
2275 t
->host_offset
, t
->offset
, t
->bytes
,
2276 t
->qiov
, t
->qiov_offset
);
2279 static coroutine_fn
int qcow2_co_preadv_part(BlockDriverState
*bs
,
2280 uint64_t offset
, uint64_t bytes
,
2282 size_t qiov_offset
, int flags
)
2284 BDRVQcow2State
*s
= bs
->opaque
;
2286 unsigned int cur_bytes
; /* number of bytes in current iteration */
2287 uint64_t host_offset
= 0;
2288 QCow2SubclusterType type
;
2289 AioTaskPool
*aio
= NULL
;
2291 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2292 /* prepare next request */
2293 cur_bytes
= MIN(bytes
, INT_MAX
);
2295 cur_bytes
= MIN(cur_bytes
,
2296 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2299 qemu_co_mutex_lock(&s
->lock
);
2300 ret
= qcow2_get_host_offset(bs
, offset
, &cur_bytes
,
2301 &host_offset
, &type
);
2302 qemu_co_mutex_unlock(&s
->lock
);
2307 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2308 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2309 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&& !bs
->backing
) ||
2310 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&& !bs
->backing
))
2312 qemu_iovec_memset(qiov
, qiov_offset
, 0, cur_bytes
);
2314 if (!aio
&& cur_bytes
!= bytes
) {
2315 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2317 ret
= qcow2_add_task(bs
, aio
, qcow2_co_preadv_task_entry
, type
,
2318 host_offset
, offset
, cur_bytes
,
2319 qiov
, qiov_offset
, NULL
);
2326 offset
+= cur_bytes
;
2327 qiov_offset
+= cur_bytes
;
2332 aio_task_pool_wait_all(aio
);
2334 ret
= aio_task_pool_status(aio
);
2342 /* Check if it's possible to merge a write request with the writing of
2343 * the data from the COW regions */
2344 static bool merge_cow(uint64_t offset
, unsigned bytes
,
2345 QEMUIOVector
*qiov
, size_t qiov_offset
,
2350 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2351 /* If both COW regions are empty then there's nothing to merge */
2352 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
2356 /* If COW regions are handled already, skip this too */
2361 /* The data (middle) region must be immediately after the
2363 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
2367 /* The end region must be immediately after the data (middle)
2369 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
2373 /* Make sure that adding both COW regions to the QEMUIOVector
2374 * does not exceed IOV_MAX */
2375 if (qemu_iovec_subvec_niov(qiov
, qiov_offset
, bytes
) > IOV_MAX
- 2) {
2379 m
->data_qiov
= qiov
;
2380 m
->data_qiov_offset
= qiov_offset
;
2387 static bool is_unallocated(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
2391 (!bdrv_is_allocated_above(bs
, NULL
, false, offset
, bytes
, &nr
) &&
2395 static bool is_zero_cow(BlockDriverState
*bs
, QCowL2Meta
*m
)
2398 * This check is designed for optimization shortcut so it must be
2400 * Instead of is_zero(), use is_unallocated() as it is faster (but not
2401 * as accurate and can result in false negatives).
2403 return is_unallocated(bs
, m
->offset
+ m
->cow_start
.offset
,
2404 m
->cow_start
.nb_bytes
) &&
2405 is_unallocated(bs
, m
->offset
+ m
->cow_end
.offset
,
2406 m
->cow_end
.nb_bytes
);
2409 static int handle_alloc_space(BlockDriverState
*bs
, QCowL2Meta
*l2meta
)
2411 BDRVQcow2State
*s
= bs
->opaque
;
2414 if (!(s
->data_file
->bs
->supported_zero_flags
& BDRV_REQ_NO_FALLBACK
)) {
2418 if (bs
->encrypted
) {
2422 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2424 uint64_t start_offset
= m
->alloc_offset
+ m
->cow_start
.offset
;
2425 unsigned nb_bytes
= m
->cow_end
.offset
+ m
->cow_end
.nb_bytes
-
2426 m
->cow_start
.offset
;
2428 if (!m
->cow_start
.nb_bytes
&& !m
->cow_end
.nb_bytes
) {
2432 if (!is_zero_cow(bs
, m
)) {
2437 * instead of writing zero COW buffers,
2438 * efficiently zero out the whole clusters
2441 ret
= qcow2_pre_write_overlap_check(bs
, 0, start_offset
, nb_bytes
,
2447 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC_SPACE
);
2448 ret
= bdrv_co_pwrite_zeroes(s
->data_file
, start_offset
, nb_bytes
,
2449 BDRV_REQ_NO_FALLBACK
);
2451 if (ret
!= -ENOTSUP
&& ret
!= -EAGAIN
) {
2457 trace_qcow2_skip_cow(qemu_coroutine_self(), m
->offset
, m
->nb_clusters
);
2464 * qcow2_co_pwritev_task
2465 * Called with s->lock unlocked
2466 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2467 * not use it somehow after qcow2_co_pwritev_task() call
2469 static coroutine_fn
int qcow2_co_pwritev_task(BlockDriverState
*bs
,
2470 uint64_t host_offset
,
2471 uint64_t offset
, uint64_t bytes
,
2473 uint64_t qiov_offset
,
2477 BDRVQcow2State
*s
= bs
->opaque
;
2478 void *crypt_buf
= NULL
;
2479 QEMUIOVector encrypted_qiov
;
2481 if (bs
->encrypted
) {
2483 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2484 crypt_buf
= qemu_try_blockalign(bs
->file
->bs
, bytes
);
2485 if (crypt_buf
== NULL
) {
2489 qemu_iovec_to_buf(qiov
, qiov_offset
, crypt_buf
, bytes
);
2491 if (qcow2_co_encrypt(bs
, host_offset
, offset
, crypt_buf
, bytes
) < 0) {
2496 qemu_iovec_init_buf(&encrypted_qiov
, crypt_buf
, bytes
);
2497 qiov
= &encrypted_qiov
;
2501 /* Try to efficiently initialize the physical space with zeroes */
2502 ret
= handle_alloc_space(bs
, l2meta
);
2508 * If we need to do COW, check if it's possible to merge the
2509 * writing of the guest data together with that of the COW regions.
2510 * If it's not possible (or not necessary) then write the
2513 if (!merge_cow(offset
, bytes
, qiov
, qiov_offset
, l2meta
)) {
2514 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2515 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset
);
2516 ret
= bdrv_co_pwritev_part(s
->data_file
, host_offset
,
2517 bytes
, qiov
, qiov_offset
, 0);
2523 qemu_co_mutex_lock(&s
->lock
);
2525 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
2529 qemu_co_mutex_lock(&s
->lock
);
2532 qcow2_handle_l2meta(bs
, &l2meta
, false);
2533 qemu_co_mutex_unlock(&s
->lock
);
2535 qemu_vfree(crypt_buf
);
2540 static coroutine_fn
int qcow2_co_pwritev_task_entry(AioTask
*task
)
2542 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2544 assert(!t
->subcluster_type
);
2546 return qcow2_co_pwritev_task(t
->bs
, t
->host_offset
,
2547 t
->offset
, t
->bytes
, t
->qiov
, t
->qiov_offset
,
2551 static coroutine_fn
int qcow2_co_pwritev_part(
2552 BlockDriverState
*bs
, uint64_t offset
, uint64_t bytes
,
2553 QEMUIOVector
*qiov
, size_t qiov_offset
, int flags
)
2555 BDRVQcow2State
*s
= bs
->opaque
;
2556 int offset_in_cluster
;
2558 unsigned int cur_bytes
; /* number of sectors in current iteration */
2559 uint64_t cluster_offset
;
2560 QCowL2Meta
*l2meta
= NULL
;
2561 AioTaskPool
*aio
= NULL
;
2563 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
2565 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2569 trace_qcow2_writev_start_part(qemu_coroutine_self());
2570 offset_in_cluster
= offset_into_cluster(s
, offset
);
2571 cur_bytes
= MIN(bytes
, INT_MAX
);
2572 if (bs
->encrypted
) {
2573 cur_bytes
= MIN(cur_bytes
,
2574 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
2575 - offset_in_cluster
);
2578 qemu_co_mutex_lock(&s
->lock
);
2580 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2581 &cluster_offset
, &l2meta
);
2586 assert(offset_into_cluster(s
, cluster_offset
) == 0);
2588 ret
= qcow2_pre_write_overlap_check(bs
, 0,
2589 cluster_offset
+ offset_in_cluster
,
2595 qemu_co_mutex_unlock(&s
->lock
);
2597 if (!aio
&& cur_bytes
!= bytes
) {
2598 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2600 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_task_entry
, 0,
2601 cluster_offset
+ offset_in_cluster
, offset
,
2602 cur_bytes
, qiov
, qiov_offset
, l2meta
);
2603 l2meta
= NULL
; /* l2meta is consumed by qcow2_co_pwritev_task() */
2609 offset
+= cur_bytes
;
2610 qiov_offset
+= cur_bytes
;
2611 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2615 qemu_co_mutex_lock(&s
->lock
);
2618 qcow2_handle_l2meta(bs
, &l2meta
, false);
2620 qemu_co_mutex_unlock(&s
->lock
);
2624 aio_task_pool_wait_all(aio
);
2626 ret
= aio_task_pool_status(aio
);
2631 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2636 static int qcow2_inactivate(BlockDriverState
*bs
)
2638 BDRVQcow2State
*s
= bs
->opaque
;
2639 int ret
, result
= 0;
2640 Error
*local_err
= NULL
;
2642 qcow2_store_persistent_dirty_bitmaps(bs
, true, &local_err
);
2643 if (local_err
!= NULL
) {
2645 error_reportf_err(local_err
, "Lost persistent bitmaps during "
2646 "inactivation of node '%s': ",
2647 bdrv_get_device_or_node_name(bs
));
2650 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2653 error_report("Failed to flush the L2 table cache: %s",
2657 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2660 error_report("Failed to flush the refcount block cache: %s",
2665 qcow2_mark_clean(bs
);
2671 static void qcow2_close(BlockDriverState
*bs
)
2673 BDRVQcow2State
*s
= bs
->opaque
;
2674 qemu_vfree(s
->l1_table
);
2675 /* else pre-write overlap checks in cache_destroy may crash */
2678 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2679 qcow2_inactivate(bs
);
2682 cache_clean_timer_del(bs
);
2683 qcow2_cache_destroy(s
->l2_table_cache
);
2684 qcow2_cache_destroy(s
->refcount_block_cache
);
2686 qcrypto_block_free(s
->crypto
);
2688 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
2690 g_free(s
->unknown_header_fields
);
2691 cleanup_unknown_header_ext(bs
);
2693 g_free(s
->image_data_file
);
2694 g_free(s
->image_backing_file
);
2695 g_free(s
->image_backing_format
);
2697 if (has_data_file(bs
)) {
2698 bdrv_unref_child(bs
, s
->data_file
);
2699 s
->data_file
= NULL
;
2702 qcow2_refcount_close(bs
);
2703 qcow2_free_snapshots(bs
);
2706 static void coroutine_fn
qcow2_co_invalidate_cache(BlockDriverState
*bs
,
2709 BDRVQcow2State
*s
= bs
->opaque
;
2710 int flags
= s
->flags
;
2711 QCryptoBlock
*crypto
= NULL
;
2713 Error
*local_err
= NULL
;
2717 * Backing files are read-only which makes all of their metadata immutable,
2718 * that means we don't have to worry about reopening them here.
2726 memset(s
, 0, sizeof(BDRVQcow2State
));
2727 options
= qdict_clone_shallow(bs
->options
);
2729 flags
&= ~BDRV_O_INACTIVE
;
2730 qemu_co_mutex_lock(&s
->lock
);
2731 ret
= qcow2_do_open(bs
, options
, flags
, &local_err
);
2732 qemu_co_mutex_unlock(&s
->lock
);
2733 qobject_unref(options
);
2735 error_propagate_prepend(errp
, local_err
,
2736 "Could not reopen qcow2 layer: ");
2739 } else if (ret
< 0) {
2740 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
2748 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2749 size_t len
, size_t buflen
)
2751 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2752 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2754 if (buflen
< ext_len
) {
2758 *ext_backing_fmt
= (QCowExtension
) {
2759 .magic
= cpu_to_be32(magic
),
2760 .len
= cpu_to_be32(len
),
2764 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2771 * Updates the qcow2 header, including the variable length parts of it, i.e.
2772 * the backing file name and all extensions. qcow2 was not designed to allow
2773 * such changes, so if we run out of space (we can only use the first cluster)
2774 * this function may fail.
2776 * Returns 0 on success, -errno in error cases.
2778 int qcow2_update_header(BlockDriverState
*bs
)
2780 BDRVQcow2State
*s
= bs
->opaque
;
2783 size_t buflen
= s
->cluster_size
;
2785 uint64_t total_size
;
2786 uint32_t refcount_table_clusters
;
2787 size_t header_length
;
2788 Qcow2UnknownHeaderExtension
*uext
;
2790 buf
= qemu_blockalign(bs
, buflen
);
2792 /* Header structure */
2793 header
= (QCowHeader
*) buf
;
2795 if (buflen
< sizeof(*header
)) {
2800 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2801 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2802 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2804 ret
= validate_compression_type(s
, NULL
);
2809 *header
= (QCowHeader
) {
2810 /* Version 2 fields */
2811 .magic
= cpu_to_be32(QCOW_MAGIC
),
2812 .version
= cpu_to_be32(s
->qcow_version
),
2813 .backing_file_offset
= 0,
2814 .backing_file_size
= 0,
2815 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2816 .size
= cpu_to_be64(total_size
),
2817 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2818 .l1_size
= cpu_to_be32(s
->l1_size
),
2819 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2820 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2821 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2822 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2823 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2825 /* Version 3 fields */
2826 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2827 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2828 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2829 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2830 .header_length
= cpu_to_be32(header_length
),
2831 .compression_type
= s
->compression_type
,
2834 /* For older versions, write a shorter header */
2835 switch (s
->qcow_version
) {
2837 ret
= offsetof(QCowHeader
, incompatible_features
);
2840 ret
= sizeof(*header
);
2849 memset(buf
, 0, buflen
);
2851 /* Preserve any unknown field in the header */
2852 if (s
->unknown_header_fields_size
) {
2853 if (buflen
< s
->unknown_header_fields_size
) {
2858 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2859 buf
+= s
->unknown_header_fields_size
;
2860 buflen
-= s
->unknown_header_fields_size
;
2863 /* Backing file format header extension */
2864 if (s
->image_backing_format
) {
2865 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2866 s
->image_backing_format
,
2867 strlen(s
->image_backing_format
),
2877 /* External data file header extension */
2878 if (has_data_file(bs
) && s
->image_data_file
) {
2879 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_DATA_FILE
,
2880 s
->image_data_file
, strlen(s
->image_data_file
),
2890 /* Full disk encryption header pointer extension */
2891 if (s
->crypto_header
.offset
!= 0) {
2892 s
->crypto_header
.offset
= cpu_to_be64(s
->crypto_header
.offset
);
2893 s
->crypto_header
.length
= cpu_to_be64(s
->crypto_header
.length
);
2894 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2895 &s
->crypto_header
, sizeof(s
->crypto_header
),
2897 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
2898 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
2907 * Feature table. A mere 8 feature names occupies 392 bytes, and
2908 * when coupled with the v3 minimum header of 104 bytes plus the
2909 * 8-byte end-of-extension marker, that would leave only 8 bytes
2910 * for a backing file name in an image with 512-byte clusters.
2911 * Thus, we choose to omit this header for cluster sizes 4k and
2914 if (s
->qcow_version
>= 3 && s
->cluster_size
> 4096) {
2915 static const Qcow2Feature features
[] = {
2917 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2918 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2919 .name
= "dirty bit",
2922 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2923 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
2924 .name
= "corrupt bit",
2927 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2928 .bit
= QCOW2_INCOMPAT_DATA_FILE_BITNR
,
2929 .name
= "external data file",
2932 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2933 .bit
= QCOW2_INCOMPAT_COMPRESSION_BITNR
,
2934 .name
= "compression type",
2937 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
2938 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
2939 .name
= "lazy refcounts",
2942 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
2943 .bit
= QCOW2_AUTOCLEAR_BITMAPS_BITNR
,
2947 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
2948 .bit
= QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR
,
2949 .name
= "raw external data",
2953 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
2954 features
, sizeof(features
), buflen
);
2962 /* Bitmap extension */
2963 if (s
->nb_bitmaps
> 0) {
2964 Qcow2BitmapHeaderExt bitmaps_header
= {
2965 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
2966 .bitmap_directory_size
=
2967 cpu_to_be64(s
->bitmap_directory_size
),
2968 .bitmap_directory_offset
=
2969 cpu_to_be64(s
->bitmap_directory_offset
)
2971 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
2972 &bitmaps_header
, sizeof(bitmaps_header
),
2981 /* Keep unknown header extensions */
2982 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
2983 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
2992 /* End of header extensions */
2993 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
3001 /* Backing file name */
3002 if (s
->image_backing_file
) {
3003 size_t backing_file_len
= strlen(s
->image_backing_file
);
3005 if (buflen
< backing_file_len
) {
3010 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3011 strncpy(buf
, s
->image_backing_file
, buflen
);
3013 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
3014 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
3017 /* Write the new header */
3018 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
3029 static int qcow2_change_backing_file(BlockDriverState
*bs
,
3030 const char *backing_file
, const char *backing_fmt
)
3032 BDRVQcow2State
*s
= bs
->opaque
;
3034 /* Adding a backing file means that the external data file alone won't be
3035 * enough to make sense of the content */
3036 if (backing_file
&& data_file_is_raw(bs
)) {
3040 if (backing_file
&& strlen(backing_file
) > 1023) {
3044 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3045 backing_file
?: "");
3046 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3047 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3049 g_free(s
->image_backing_file
);
3050 g_free(s
->image_backing_format
);
3052 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
3053 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
3055 return qcow2_update_header(bs
);
3058 static int qcow2_set_up_encryption(BlockDriverState
*bs
,
3059 QCryptoBlockCreateOptions
*cryptoopts
,
3062 BDRVQcow2State
*s
= bs
->opaque
;
3063 QCryptoBlock
*crypto
= NULL
;
3066 switch (cryptoopts
->format
) {
3067 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
3068 fmt
= QCOW_CRYPT_LUKS
;
3070 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
3071 fmt
= QCOW_CRYPT_AES
;
3074 error_setg(errp
, "Crypto format not supported in qcow2");
3078 s
->crypt_method_header
= fmt
;
3080 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
3081 qcow2_crypto_hdr_init_func
,
3082 qcow2_crypto_hdr_write_func
,
3088 ret
= qcow2_update_header(bs
);
3090 error_setg_errno(errp
, -ret
, "Could not write encryption header");
3096 qcrypto_block_free(crypto
);
3101 * Preallocates metadata structures for data clusters between @offset (in the
3102 * guest disk) and @new_length (which is thus generally the new guest disk
3105 * Returns: 0 on success, -errno on failure.
3107 static int coroutine_fn
preallocate_co(BlockDriverState
*bs
, uint64_t offset
,
3108 uint64_t new_length
, PreallocMode mode
,
3111 BDRVQcow2State
*s
= bs
->opaque
;
3113 uint64_t host_offset
= 0;
3114 int64_t file_length
;
3115 unsigned int cur_bytes
;
3119 assert(offset
<= new_length
);
3120 bytes
= new_length
- offset
;
3123 cur_bytes
= MIN(bytes
, QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
));
3124 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
3125 &host_offset
, &meta
);
3127 error_setg_errno(errp
, -ret
, "Allocating clusters failed");
3132 QCowL2Meta
*next
= meta
->next
;
3134 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
3136 error_setg_errno(errp
, -ret
, "Mapping clusters failed");
3137 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
3138 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
3142 /* There are no dependent requests, but we need to remove our
3143 * request from the list of in-flight requests */
3144 QLIST_REMOVE(meta
, next_in_flight
);
3150 /* TODO Preallocate data if requested */
3153 offset
+= cur_bytes
;
3157 * It is expected that the image file is large enough to actually contain
3158 * all of the allocated clusters (otherwise we get failing reads after
3159 * EOF). Extend the image to the last allocated sector.
3161 file_length
= bdrv_getlength(s
->data_file
->bs
);
3162 if (file_length
< 0) {
3163 error_setg_errno(errp
, -file_length
, "Could not get file size");
3167 if (host_offset
+ cur_bytes
> file_length
) {
3168 if (mode
== PREALLOC_MODE_METADATA
) {
3169 mode
= PREALLOC_MODE_OFF
;
3171 ret
= bdrv_co_truncate(s
->data_file
, host_offset
+ cur_bytes
, false,
3181 /* qcow2_refcount_metadata_size:
3182 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3183 * @cluster_size: size of a cluster, in bytes
3184 * @refcount_order: refcount bits power-of-2 exponent
3185 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3188 * Returns: Number of bytes required for refcount blocks and table metadata.
3190 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
3191 int refcount_order
, bool generous_increase
,
3192 uint64_t *refblock_count
)
3195 * Every host cluster is reference-counted, including metadata (even
3196 * refcount metadata is recursively included).
3198 * An accurate formula for the size of refcount metadata size is difficult
3199 * to derive. An easier method of calculation is finding the fixed point
3200 * where no further refcount blocks or table clusters are required to
3201 * reference count every cluster.
3203 int64_t blocks_per_table_cluster
= cluster_size
/ sizeof(uint64_t);
3204 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
3205 int64_t table
= 0; /* number of refcount table clusters */
3206 int64_t blocks
= 0; /* number of refcount block clusters */
3212 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
3213 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
3214 n
= clusters
+ blocks
+ table
;
3216 if (n
== last
&& generous_increase
) {
3217 clusters
+= DIV_ROUND_UP(table
, 2);
3218 n
= 0; /* force another loop */
3219 generous_increase
= false;
3221 } while (n
!= last
);
3223 if (refblock_count
) {
3224 *refblock_count
= blocks
;
3227 return (blocks
+ table
) * cluster_size
;
3231 * qcow2_calc_prealloc_size:
3232 * @total_size: virtual disk size in bytes
3233 * @cluster_size: cluster size in bytes
3234 * @refcount_order: refcount bits power-of-2 exponent
3235 * @extended_l2: true if the image has extended L2 entries
3237 * Returns: Total number of bytes required for the fully allocated image
3238 * (including metadata).
3240 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
3241 size_t cluster_size
,
3245 int64_t meta_size
= 0;
3246 uint64_t nl1e
, nl2e
;
3247 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
3248 size_t l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
3250 /* header: 1 cluster */
3251 meta_size
+= cluster_size
;
3253 /* total size of L2 tables */
3254 nl2e
= aligned_total_size
/ cluster_size
;
3255 nl2e
= ROUND_UP(nl2e
, cluster_size
/ l2e_size
);
3256 meta_size
+= nl2e
* l2e_size
;
3258 /* total size of L1 tables */
3259 nl1e
= nl2e
* l2e_size
/ cluster_size
;
3260 nl1e
= ROUND_UP(nl1e
, cluster_size
/ sizeof(uint64_t));
3261 meta_size
+= nl1e
* sizeof(uint64_t);
3263 /* total size of refcount table and blocks */
3264 meta_size
+= qcow2_refcount_metadata_size(
3265 (meta_size
+ aligned_total_size
) / cluster_size
,
3266 cluster_size
, refcount_order
, false, NULL
);
3268 return meta_size
+ aligned_total_size
;
3271 static bool validate_cluster_size(size_t cluster_size
, Error
**errp
)
3273 int cluster_bits
= ctz32(cluster_size
);
3274 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
3275 (1 << cluster_bits
) != cluster_size
)
3277 error_setg(errp
, "Cluster size must be a power of two between %d and "
3278 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
3284 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, Error
**errp
)
3286 size_t cluster_size
;
3288 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
3289 DEFAULT_CLUSTER_SIZE
);
3290 if (!validate_cluster_size(cluster_size
, errp
)) {
3293 return cluster_size
;
3296 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
3301 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
3303 ret
= 3; /* default */
3304 } else if (!strcmp(buf
, "0.10")) {
3306 } else if (!strcmp(buf
, "1.1")) {
3309 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
3316 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
3319 uint64_t refcount_bits
;
3321 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
3322 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
3323 error_setg(errp
, "Refcount width must be a power of two and may not "
3328 if (version
< 3 && refcount_bits
!= 16) {
3329 error_setg(errp
, "Different refcount widths than 16 bits require "
3330 "compatibility level 1.1 or above (use compat=1.1 or "
3335 return refcount_bits
;
3338 static int coroutine_fn
3339 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
3341 BlockdevCreateOptionsQcow2
*qcow2_opts
;
3345 * Open the image file and write a minimal qcow2 header.
3347 * We keep things simple and start with a zero-sized image. We also
3348 * do without refcount blocks or a L1 table for now. We'll fix the
3349 * inconsistency later.
3351 * We do need a refcount table because growing the refcount table means
3352 * allocating two new refcount blocks - the second of which would be at
3353 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3354 * size for any qcow2 image.
3356 BlockBackend
*blk
= NULL
;
3357 BlockDriverState
*bs
= NULL
;
3358 BlockDriverState
*data_bs
= NULL
;
3360 size_t cluster_size
;
3363 uint64_t* refcount_table
;
3365 uint8_t compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
3367 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
3368 qcow2_opts
= &create_options
->u
.qcow2
;
3370 bs
= bdrv_open_blockdev_ref(qcow2_opts
->file
, errp
);
3375 /* Validate options and set default values */
3376 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
3377 error_setg(errp
, "Image size must be a multiple of %u bytes",
3378 (unsigned) BDRV_SECTOR_SIZE
);
3383 if (qcow2_opts
->has_version
) {
3384 switch (qcow2_opts
->version
) {
3385 case BLOCKDEV_QCOW2_VERSION_V2
:
3388 case BLOCKDEV_QCOW2_VERSION_V3
:
3392 g_assert_not_reached();
3398 if (qcow2_opts
->has_cluster_size
) {
3399 cluster_size
= qcow2_opts
->cluster_size
;
3401 cluster_size
= DEFAULT_CLUSTER_SIZE
;
3404 if (!validate_cluster_size(cluster_size
, errp
)) {
3409 if (!qcow2_opts
->has_preallocation
) {
3410 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
3412 if (qcow2_opts
->has_backing_file
&&
3413 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
)
3415 error_setg(errp
, "Backing file and preallocation cannot be used at "
3420 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->has_backing_file
) {
3421 error_setg(errp
, "Backing format cannot be used without backing file");
3426 if (!qcow2_opts
->has_lazy_refcounts
) {
3427 qcow2_opts
->lazy_refcounts
= false;
3429 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
3430 error_setg(errp
, "Lazy refcounts only supported with compatibility "
3431 "level 1.1 and above (use version=v3 or greater)");
3436 if (!qcow2_opts
->has_refcount_bits
) {
3437 qcow2_opts
->refcount_bits
= 16;
3439 if (qcow2_opts
->refcount_bits
> 64 ||
3440 !is_power_of_2(qcow2_opts
->refcount_bits
))
3442 error_setg(errp
, "Refcount width must be a power of two and may not "
3447 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
3448 error_setg(errp
, "Different refcount widths than 16 bits require "
3449 "compatibility level 1.1 or above (use version=v3 or "
3454 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
3456 if (qcow2_opts
->data_file_raw
&& !qcow2_opts
->data_file
) {
3457 error_setg(errp
, "data-file-raw requires data-file");
3461 if (qcow2_opts
->data_file_raw
&& qcow2_opts
->has_backing_file
) {
3462 error_setg(errp
, "Backing file and data-file-raw cannot be used at "
3468 if (qcow2_opts
->data_file
) {
3470 error_setg(errp
, "External data files are only supported with "
3471 "compatibility level 1.1 and above (use version=v3 or "
3476 data_bs
= bdrv_open_blockdev_ref(qcow2_opts
->data_file
, errp
);
3477 if (data_bs
== NULL
) {
3483 if (qcow2_opts
->has_compression_type
&&
3484 qcow2_opts
->compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3489 error_setg(errp
, "Non-zlib compression type is only supported with "
3490 "compatibility level 1.1 and above (use version=v3 or "
3495 switch (qcow2_opts
->compression_type
) {
3497 case QCOW2_COMPRESSION_TYPE_ZSTD
:
3501 error_setg(errp
, "Unknown compression type");
3505 compression_type
= qcow2_opts
->compression_type
;
3508 /* Create BlockBackend to write to the image */
3509 blk
= blk_new_with_bs(bs
, BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
,
3515 blk_set_allow_write_beyond_eof(blk
, true);
3517 /* Write the header */
3518 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
3519 header
= g_malloc0(cluster_size
);
3520 *header
= (QCowHeader
) {
3521 .magic
= cpu_to_be32(QCOW_MAGIC
),
3522 .version
= cpu_to_be32(version
),
3523 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
3524 .size
= cpu_to_be64(0),
3525 .l1_table_offset
= cpu_to_be64(0),
3526 .l1_size
= cpu_to_be32(0),
3527 .refcount_table_offset
= cpu_to_be64(cluster_size
),
3528 .refcount_table_clusters
= cpu_to_be32(1),
3529 .refcount_order
= cpu_to_be32(refcount_order
),
3530 /* don't deal with endianness since compression_type is 1 byte long */
3531 .compression_type
= compression_type
,
3532 .header_length
= cpu_to_be32(sizeof(*header
)),
3535 /* We'll update this to correct value later */
3536 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
3538 if (qcow2_opts
->lazy_refcounts
) {
3539 header
->compatible_features
|=
3540 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
3543 header
->incompatible_features
|=
3544 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE
);
3546 if (qcow2_opts
->data_file_raw
) {
3547 header
->autoclear_features
|=
3548 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW
);
3550 if (compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3551 header
->incompatible_features
|=
3552 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION
);
3555 ret
= blk_pwrite(blk
, 0, header
, cluster_size
, 0);
3558 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
3562 /* Write a refcount table with one refcount block */
3563 refcount_table
= g_malloc0(2 * cluster_size
);
3564 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
3565 ret
= blk_pwrite(blk
, cluster_size
, refcount_table
, 2 * cluster_size
, 0);
3566 g_free(refcount_table
);
3569 error_setg_errno(errp
, -ret
, "Could not write refcount table");
3577 * And now open the image and make it consistent first (i.e. increase the
3578 * refcount of the cluster that is occupied by the header and the refcount
3581 options
= qdict_new();
3582 qdict_put_str(options
, "driver", "qcow2");
3583 qdict_put_str(options
, "file", bs
->node_name
);
3585 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3587 blk
= blk_new_open(NULL
, NULL
, options
,
3588 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
3595 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
3597 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
3598 "header and refcount table");
3601 } else if (ret
!= 0) {
3602 error_report("Huh, first cluster in empty image is already in use?");
3606 /* Set the external data file if necessary */
3608 BDRVQcow2State
*s
= blk_bs(blk
)->opaque
;
3609 s
->image_data_file
= g_strdup(data_bs
->filename
);
3612 /* Create a full header (including things like feature table) */
3613 ret
= qcow2_update_header(blk_bs(blk
));
3615 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
3619 /* Okay, now that we have a valid image, let's give it the right size */
3620 ret
= blk_truncate(blk
, qcow2_opts
->size
, false, qcow2_opts
->preallocation
,
3623 error_prepend(errp
, "Could not resize image: ");
3627 /* Want a backing file? There you go. */
3628 if (qcow2_opts
->has_backing_file
) {
3629 const char *backing_format
= NULL
;
3631 if (qcow2_opts
->has_backing_fmt
) {
3632 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
3635 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
3636 backing_format
, false);
3638 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
3639 "with format '%s'", qcow2_opts
->backing_file
,
3645 /* Want encryption? There you go. */
3646 if (qcow2_opts
->has_encrypt
) {
3647 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3656 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3657 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3658 * have to setup decryption context. We're not doing any I/O on the top
3659 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3662 options
= qdict_new();
3663 qdict_put_str(options
, "driver", "qcow2");
3664 qdict_put_str(options
, "file", bs
->node_name
);
3666 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3668 blk
= blk_new_open(NULL
, NULL
, options
,
3669 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3680 bdrv_unref(data_bs
);
3684 static int coroutine_fn
qcow2_co_create_opts(BlockDriver
*drv
,
3685 const char *filename
,
3689 BlockdevCreateOptions
*create_options
= NULL
;
3692 BlockDriverState
*bs
= NULL
;
3693 BlockDriverState
*data_bs
= NULL
;
3697 /* Only the keyval visitor supports the dotted syntax needed for
3698 * encryption, so go through a QDict before getting a QAPI type. Ignore
3699 * options meant for the protocol layer so that the visitor doesn't
3701 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3704 /* Handle encryption options */
3705 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3706 if (val
&& !strcmp(val
, "on")) {
3707 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3708 } else if (val
&& !strcmp(val
, "off")) {
3709 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3712 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3713 if (val
&& !strcmp(val
, "aes")) {
3714 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3717 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3718 * version=v2/v3 below. */
3719 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3720 if (val
&& !strcmp(val
, "0.10")) {
3721 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3722 } else if (val
&& !strcmp(val
, "1.1")) {
3723 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3726 /* Change legacy command line options into QMP ones */
3727 static const QDictRenames opt_renames
[] = {
3728 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3729 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3730 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3731 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3732 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3733 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3734 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3735 { BLOCK_OPT_DATA_FILE_RAW
, "data-file-raw" },
3736 { BLOCK_OPT_COMPRESSION_TYPE
, "compression-type" },
3740 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3745 /* Create and open the file (protocol layer) */
3746 ret
= bdrv_create_file(filename
, opts
, errp
);
3751 bs
= bdrv_open(filename
, NULL
, NULL
,
3752 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3758 /* Create and open an external data file (protocol layer) */
3759 val
= qdict_get_try_str(qdict
, BLOCK_OPT_DATA_FILE
);
3761 ret
= bdrv_create_file(val
, opts
, errp
);
3766 data_bs
= bdrv_open(val
, NULL
, NULL
,
3767 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
,
3769 if (data_bs
== NULL
) {
3774 qdict_del(qdict
, BLOCK_OPT_DATA_FILE
);
3775 qdict_put_str(qdict
, "data-file", data_bs
->node_name
);
3778 /* Set 'driver' and 'node' options */
3779 qdict_put_str(qdict
, "driver", "qcow2");
3780 qdict_put_str(qdict
, "file", bs
->node_name
);
3782 /* Now get the QAPI type BlockdevCreateOptions */
3783 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
3789 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, errp
);
3791 if (!create_options
) {
3796 /* Silently round up size */
3797 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3800 /* Create the qcow2 image (format layer) */
3801 ret
= qcow2_co_create(create_options
, errp
);
3808 qobject_unref(qdict
);
3810 bdrv_unref(data_bs
);
3811 qapi_free_BlockdevCreateOptions(create_options
);
3816 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3821 /* Clamp to image length, before checking status of underlying sectors */
3822 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3823 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3829 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3830 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
== bytes
;
3833 static coroutine_fn
int qcow2_co_pwrite_zeroes(BlockDriverState
*bs
,
3834 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3837 BDRVQcow2State
*s
= bs
->opaque
;
3839 uint32_t head
= offset_into_subcluster(s
, offset
);
3840 uint32_t tail
= ROUND_UP(offset
+ bytes
, s
->subcluster_size
) -
3843 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3844 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3851 QCow2SubclusterType type
;
3853 assert(head
+ bytes
+ tail
<= s
->subcluster_size
);
3855 /* check whether remainder of cluster already reads as zero */
3856 if (!(is_zero(bs
, offset
- head
, head
) &&
3857 is_zero(bs
, offset
+ bytes
, tail
))) {
3861 qemu_co_mutex_lock(&s
->lock
);
3862 /* We can have new write after previous check */
3864 bytes
= s
->subcluster_size
;
3865 nr
= s
->subcluster_size
;
3866 ret
= qcow2_get_host_offset(bs
, offset
, &nr
, &off
, &type
);
3868 (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
3869 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&&
3870 type
!= QCOW2_SUBCLUSTER_ZERO_PLAIN
&&
3871 type
!= QCOW2_SUBCLUSTER_ZERO_ALLOC
)) {
3872 qemu_co_mutex_unlock(&s
->lock
);
3876 qemu_co_mutex_lock(&s
->lock
);
3879 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
3881 /* Whatever is left can use real zero subclusters */
3882 ret
= qcow2_subcluster_zeroize(bs
, offset
, bytes
, flags
);
3883 qemu_co_mutex_unlock(&s
->lock
);
3888 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
3889 int64_t offset
, int bytes
)
3892 BDRVQcow2State
*s
= bs
->opaque
;
3894 /* If the image does not support QCOW_OFLAG_ZERO then discarding
3895 * clusters could expose stale data from the backing file. */
3896 if (s
->qcow_version
< 3 && bs
->backing
) {
3900 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
3901 assert(bytes
< s
->cluster_size
);
3902 /* Ignore partial clusters, except for the special case of the
3903 * complete partial cluster at the end of an unaligned file */
3904 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
3905 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3910 qemu_co_mutex_lock(&s
->lock
);
3911 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
3913 qemu_co_mutex_unlock(&s
->lock
);
3917 static int coroutine_fn
3918 qcow2_co_copy_range_from(BlockDriverState
*bs
,
3919 BdrvChild
*src
, uint64_t src_offset
,
3920 BdrvChild
*dst
, uint64_t dst_offset
,
3921 uint64_t bytes
, BdrvRequestFlags read_flags
,
3922 BdrvRequestFlags write_flags
)
3924 BDRVQcow2State
*s
= bs
->opaque
;
3926 unsigned int cur_bytes
; /* number of bytes in current iteration */
3927 BdrvChild
*child
= NULL
;
3928 BdrvRequestFlags cur_write_flags
;
3930 assert(!bs
->encrypted
);
3931 qemu_co_mutex_lock(&s
->lock
);
3933 while (bytes
!= 0) {
3934 uint64_t copy_offset
= 0;
3935 QCow2SubclusterType type
;
3936 /* prepare next request */
3937 cur_bytes
= MIN(bytes
, INT_MAX
);
3938 cur_write_flags
= write_flags
;
3940 ret
= qcow2_get_host_offset(bs
, src_offset
, &cur_bytes
,
3941 ©_offset
, &type
);
3947 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
3948 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
3949 if (bs
->backing
&& bs
->backing
->bs
) {
3950 int64_t backing_length
= bdrv_getlength(bs
->backing
->bs
);
3951 if (src_offset
>= backing_length
) {
3952 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3954 child
= bs
->backing
;
3955 cur_bytes
= MIN(cur_bytes
, backing_length
- src_offset
);
3956 copy_offset
= src_offset
;
3959 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3963 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
3964 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
3965 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3968 case QCOW2_SUBCLUSTER_COMPRESSED
:
3972 case QCOW2_SUBCLUSTER_NORMAL
:
3973 child
= s
->data_file
;
3979 qemu_co_mutex_unlock(&s
->lock
);
3980 ret
= bdrv_co_copy_range_from(child
,
3983 cur_bytes
, read_flags
, cur_write_flags
);
3984 qemu_co_mutex_lock(&s
->lock
);
3990 src_offset
+= cur_bytes
;
3991 dst_offset
+= cur_bytes
;
3996 qemu_co_mutex_unlock(&s
->lock
);
4000 static int coroutine_fn
4001 qcow2_co_copy_range_to(BlockDriverState
*bs
,
4002 BdrvChild
*src
, uint64_t src_offset
,
4003 BdrvChild
*dst
, uint64_t dst_offset
,
4004 uint64_t bytes
, BdrvRequestFlags read_flags
,
4005 BdrvRequestFlags write_flags
)
4007 BDRVQcow2State
*s
= bs
->opaque
;
4008 int offset_in_cluster
;
4010 unsigned int cur_bytes
; /* number of sectors in current iteration */
4011 uint64_t cluster_offset
;
4012 QCowL2Meta
*l2meta
= NULL
;
4014 assert(!bs
->encrypted
);
4016 qemu_co_mutex_lock(&s
->lock
);
4018 while (bytes
!= 0) {
4022 offset_in_cluster
= offset_into_cluster(s
, dst_offset
);
4023 cur_bytes
= MIN(bytes
, INT_MAX
);
4026 * If src->bs == dst->bs, we could simply copy by incrementing
4027 * the refcnt, without copying user data.
4028 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4029 ret
= qcow2_alloc_cluster_offset(bs
, dst_offset
, &cur_bytes
,
4030 &cluster_offset
, &l2meta
);
4035 assert(offset_into_cluster(s
, cluster_offset
) == 0);
4037 ret
= qcow2_pre_write_overlap_check(bs
, 0,
4038 cluster_offset
+ offset_in_cluster
, cur_bytes
, true);
4043 qemu_co_mutex_unlock(&s
->lock
);
4044 ret
= bdrv_co_copy_range_to(src
, src_offset
,
4046 cluster_offset
+ offset_in_cluster
,
4047 cur_bytes
, read_flags
, write_flags
);
4048 qemu_co_mutex_lock(&s
->lock
);
4053 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
4059 src_offset
+= cur_bytes
;
4060 dst_offset
+= cur_bytes
;
4065 qcow2_handle_l2meta(bs
, &l2meta
, false);
4067 qemu_co_mutex_unlock(&s
->lock
);
4069 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
4074 static int coroutine_fn
qcow2_co_truncate(BlockDriverState
*bs
, int64_t offset
,
4075 bool exact
, PreallocMode prealloc
,
4076 BdrvRequestFlags flags
, Error
**errp
)
4078 BDRVQcow2State
*s
= bs
->opaque
;
4079 uint64_t old_length
;
4080 int64_t new_l1_size
;
4084 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
4085 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
4087 error_setg(errp
, "Unsupported preallocation mode '%s'",
4088 PreallocMode_str(prealloc
));
4092 if (!QEMU_IS_ALIGNED(offset
, BDRV_SECTOR_SIZE
)) {
4093 error_setg(errp
, "The new size must be a multiple of %u",
4094 (unsigned) BDRV_SECTOR_SIZE
);
4098 qemu_co_mutex_lock(&s
->lock
);
4101 * Even though we store snapshot size for all images, it was not
4102 * required until v3, so it is not safe to proceed for v2.
4104 if (s
->nb_snapshots
&& s
->qcow_version
< 3) {
4105 error_setg(errp
, "Can't resize a v2 image which has snapshots");
4110 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4111 if (qcow2_truncate_bitmaps_check(bs
, errp
)) {
4116 old_length
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4117 new_l1_size
= size_to_l1(s
, offset
);
4119 if (offset
< old_length
) {
4120 int64_t last_cluster
, old_file_size
;
4121 if (prealloc
!= PREALLOC_MODE_OFF
) {
4123 "Preallocation can't be used for shrinking an image");
4128 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
4129 old_length
- ROUND_UP(offset
,
4131 QCOW2_DISCARD_ALWAYS
, true);
4133 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
4137 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
4139 error_setg_errno(errp
, -ret
,
4140 "Failed to reduce the number of L2 tables");
4144 ret
= qcow2_shrink_reftable(bs
);
4146 error_setg_errno(errp
, -ret
,
4147 "Failed to discard unused refblocks");
4151 old_file_size
= bdrv_getlength(bs
->file
->bs
);
4152 if (old_file_size
< 0) {
4153 error_setg_errno(errp
, -old_file_size
,
4154 "Failed to inquire current file length");
4155 ret
= old_file_size
;
4158 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4159 if (last_cluster
< 0) {
4160 error_setg_errno(errp
, -last_cluster
,
4161 "Failed to find the last cluster");
4165 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
4166 Error
*local_err
= NULL
;
4169 * Do not pass @exact here: It will not help the user if
4170 * we get an error here just because they wanted to shrink
4171 * their qcow2 image (on a block device) with qemu-img.
4172 * (And on the qcow2 layer, the @exact requirement is
4173 * always fulfilled, so there is no need to pass it on.)
4175 bdrv_co_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
4176 false, PREALLOC_MODE_OFF
, 0, &local_err
);
4178 warn_reportf_err(local_err
,
4179 "Failed to truncate the tail of the image: ");
4183 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
4185 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
4191 case PREALLOC_MODE_OFF
:
4192 if (has_data_file(bs
)) {
4194 * If the caller wants an exact resize, the external data
4195 * file should be resized to the exact target size, too,
4196 * so we pass @exact here.
4198 ret
= bdrv_co_truncate(s
->data_file
, offset
, exact
, prealloc
, 0,
4206 case PREALLOC_MODE_METADATA
:
4207 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4213 case PREALLOC_MODE_FALLOC
:
4214 case PREALLOC_MODE_FULL
:
4216 int64_t allocation_start
, host_offset
, guest_offset
;
4217 int64_t clusters_allocated
;
4218 int64_t old_file_size
, last_cluster
, new_file_size
;
4219 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
4221 /* With a data file, preallocation means just allocating the metadata
4222 * and forwarding the truncate request to the data file */
4223 if (has_data_file(bs
)) {
4224 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4231 old_file_size
= bdrv_getlength(bs
->file
->bs
);
4232 if (old_file_size
< 0) {
4233 error_setg_errno(errp
, -old_file_size
,
4234 "Failed to inquire current file length");
4235 ret
= old_file_size
;
4239 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4240 if (last_cluster
>= 0) {
4241 old_file_size
= (last_cluster
+ 1) * s
->cluster_size
;
4243 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
4246 nb_new_data_clusters
= (ROUND_UP(offset
, s
->cluster_size
) -
4247 start_of_cluster(s
, old_length
)) >> s
->cluster_bits
;
4249 /* This is an overestimation; we will not actually allocate space for
4250 * these in the file but just make sure the new refcount structures are
4251 * able to cover them so we will not have to allocate new refblocks
4252 * while entering the data blocks in the potentially new L2 tables.
4253 * (We do not actually care where the L2 tables are placed. Maybe they
4254 * are already allocated or they can be placed somewhere before
4255 * @old_file_size. It does not matter because they will be fully
4256 * allocated automatically, so they do not need to be covered by the
4257 * preallocation. All that matters is that we will not have to allocate
4258 * new refcount structures for them.) */
4259 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
4260 s
->cluster_size
/ l2_entry_size(s
));
4261 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4262 * table for a potential head/tail */
4265 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
4266 nb_new_data_clusters
+
4269 if (allocation_start
< 0) {
4270 error_setg_errno(errp
, -allocation_start
,
4271 "Failed to resize refcount structures");
4272 ret
= allocation_start
;
4276 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
4277 nb_new_data_clusters
);
4278 if (clusters_allocated
< 0) {
4279 error_setg_errno(errp
, -clusters_allocated
,
4280 "Failed to allocate data clusters");
4281 ret
= clusters_allocated
;
4285 assert(clusters_allocated
== nb_new_data_clusters
);
4287 /* Allocate the data area */
4288 new_file_size
= allocation_start
+
4289 nb_new_data_clusters
* s
->cluster_size
;
4291 * Image file grows, so @exact does not matter.
4293 * If we need to zero out the new area, try first whether the protocol
4294 * driver can already take care of this.
4296 if (flags
& BDRV_REQ_ZERO_WRITE
) {
4297 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
,
4298 BDRV_REQ_ZERO_WRITE
, NULL
);
4300 flags
&= ~BDRV_REQ_ZERO_WRITE
;
4306 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
, 0,
4310 error_prepend(errp
, "Failed to resize underlying file: ");
4311 qcow2_free_clusters(bs
, allocation_start
,
4312 nb_new_data_clusters
* s
->cluster_size
,
4313 QCOW2_DISCARD_OTHER
);
4317 /* Create the necessary L2 entries */
4318 host_offset
= allocation_start
;
4319 guest_offset
= old_length
;
4320 while (nb_new_data_clusters
) {
4321 int64_t nb_clusters
= MIN(
4322 nb_new_data_clusters
,
4323 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
4324 unsigned cow_start_length
= offset_into_cluster(s
, guest_offset
);
4325 QCowL2Meta allocation
;
4326 guest_offset
= start_of_cluster(s
, guest_offset
);
4327 allocation
= (QCowL2Meta
) {
4328 .offset
= guest_offset
,
4329 .alloc_offset
= host_offset
,
4330 .nb_clusters
= nb_clusters
,
4333 .nb_bytes
= cow_start_length
,
4336 .offset
= nb_clusters
<< s
->cluster_bits
,
4340 qemu_co_queue_init(&allocation
.dependent_requests
);
4342 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
4344 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
4345 qcow2_free_clusters(bs
, host_offset
,
4346 nb_new_data_clusters
* s
->cluster_size
,
4347 QCOW2_DISCARD_OTHER
);
4351 guest_offset
+= nb_clusters
* s
->cluster_size
;
4352 host_offset
+= nb_clusters
* s
->cluster_size
;
4353 nb_new_data_clusters
-= nb_clusters
;
4359 g_assert_not_reached();
4362 if ((flags
& BDRV_REQ_ZERO_WRITE
) && offset
> old_length
) {
4363 uint64_t zero_start
= QEMU_ALIGN_UP(old_length
, s
->subcluster_size
);
4366 * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4367 * requires a subcluster-aligned start. The end may be unaligned if
4368 * it is at the end of the image (which it is here).
4370 if (offset
> zero_start
) {
4371 ret
= qcow2_subcluster_zeroize(bs
, zero_start
, offset
- zero_start
,
4374 error_setg_errno(errp
, -ret
, "Failed to zero out new clusters");
4379 /* Write explicit zeros for the unaligned head */
4380 if (zero_start
> old_length
) {
4381 uint64_t len
= MIN(zero_start
, offset
) - old_length
;
4382 uint8_t *buf
= qemu_blockalign0(bs
, len
);
4384 qemu_iovec_init_buf(&qiov
, buf
, len
);
4386 qemu_co_mutex_unlock(&s
->lock
);
4387 ret
= qcow2_co_pwritev_part(bs
, old_length
, len
, &qiov
, 0, 0);
4388 qemu_co_mutex_lock(&s
->lock
);
4392 error_setg_errno(errp
, -ret
, "Failed to zero out the new area");
4398 if (prealloc
!= PREALLOC_MODE_OFF
) {
4399 /* Flush metadata before actually changing the image size */
4400 ret
= qcow2_write_caches(bs
);
4402 error_setg_errno(errp
, -ret
,
4403 "Failed to flush the preallocated area to disk");
4408 bs
->total_sectors
= offset
/ BDRV_SECTOR_SIZE
;
4410 /* write updated header.size */
4411 offset
= cpu_to_be64(offset
);
4412 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
4413 &offset
, sizeof(uint64_t));
4415 error_setg_errno(errp
, -ret
, "Failed to update the image size");
4419 s
->l1_vm_state_index
= new_l1_size
;
4421 /* Update cache sizes */
4422 options
= qdict_clone_shallow(bs
->options
);
4423 ret
= qcow2_update_options(bs
, options
, s
->flags
, errp
);
4424 qobject_unref(options
);
4430 qemu_co_mutex_unlock(&s
->lock
);
4434 static coroutine_fn
int
4435 qcow2_co_pwritev_compressed_task(BlockDriverState
*bs
,
4436 uint64_t offset
, uint64_t bytes
,
4437 QEMUIOVector
*qiov
, size_t qiov_offset
)
4439 BDRVQcow2State
*s
= bs
->opaque
;
4442 uint8_t *buf
, *out_buf
;
4443 uint64_t cluster_offset
;
4445 assert(bytes
== s
->cluster_size
|| (bytes
< s
->cluster_size
&&
4446 (offset
+ bytes
== bs
->total_sectors
<< BDRV_SECTOR_BITS
)));
4448 buf
= qemu_blockalign(bs
, s
->cluster_size
);
4449 if (bytes
< s
->cluster_size
) {
4450 /* Zero-pad last write if image size is not cluster aligned */
4451 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
4453 qemu_iovec_to_buf(qiov
, qiov_offset
, buf
, bytes
);
4455 out_buf
= g_malloc(s
->cluster_size
);
4457 out_len
= qcow2_co_compress(bs
, out_buf
, s
->cluster_size
- 1,
4458 buf
, s
->cluster_size
);
4459 if (out_len
== -ENOMEM
) {
4460 /* could not compress: write normal cluster */
4461 ret
= qcow2_co_pwritev_part(bs
, offset
, bytes
, qiov
, qiov_offset
, 0);
4466 } else if (out_len
< 0) {
4471 qemu_co_mutex_lock(&s
->lock
);
4472 ret
= qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
,
4475 qemu_co_mutex_unlock(&s
->lock
);
4479 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
, true);
4480 qemu_co_mutex_unlock(&s
->lock
);
4485 BLKDBG_EVENT(s
->data_file
, BLKDBG_WRITE_COMPRESSED
);
4486 ret
= bdrv_co_pwrite(s
->data_file
, cluster_offset
, out_len
, out_buf
, 0);
4498 static coroutine_fn
int qcow2_co_pwritev_compressed_task_entry(AioTask
*task
)
4500 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
4502 assert(!t
->subcluster_type
&& !t
->l2meta
);
4504 return qcow2_co_pwritev_compressed_task(t
->bs
, t
->offset
, t
->bytes
, t
->qiov
,
4509 * XXX: put compressed sectors first, then all the cluster aligned
4510 * tables to avoid losing bytes in alignment
4512 static coroutine_fn
int
4513 qcow2_co_pwritev_compressed_part(BlockDriverState
*bs
,
4514 uint64_t offset
, uint64_t bytes
,
4515 QEMUIOVector
*qiov
, size_t qiov_offset
)
4517 BDRVQcow2State
*s
= bs
->opaque
;
4518 AioTaskPool
*aio
= NULL
;
4521 if (has_data_file(bs
)) {
4527 * align end of file to a sector boundary to ease reading with
4530 int64_t len
= bdrv_getlength(bs
->file
->bs
);
4534 return bdrv_co_truncate(bs
->file
, len
, false, PREALLOC_MODE_OFF
, 0,
4538 if (offset_into_cluster(s
, offset
)) {
4542 if (offset_into_cluster(s
, bytes
) &&
4543 (offset
+ bytes
) != (bs
->total_sectors
<< BDRV_SECTOR_BITS
)) {
4547 while (bytes
&& aio_task_pool_status(aio
) == 0) {
4548 uint64_t chunk_size
= MIN(bytes
, s
->cluster_size
);
4550 if (!aio
&& chunk_size
!= bytes
) {
4551 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
4554 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_compressed_task_entry
,
4555 0, 0, offset
, chunk_size
, qiov
, qiov_offset
, NULL
);
4559 qiov_offset
+= chunk_size
;
4560 offset
+= chunk_size
;
4561 bytes
-= chunk_size
;
4565 aio_task_pool_wait_all(aio
);
4567 ret
= aio_task_pool_status(aio
);
4575 static int coroutine_fn
4576 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
4577 uint64_t cluster_descriptor
,
4583 BDRVQcow2State
*s
= bs
->opaque
;
4584 int ret
= 0, csize
, nb_csectors
;
4586 uint8_t *buf
, *out_buf
;
4587 int offset_in_cluster
= offset_into_cluster(s
, offset
);
4589 coffset
= cluster_descriptor
& s
->cluster_offset_mask
;
4590 nb_csectors
= ((cluster_descriptor
>> s
->csize_shift
) & s
->csize_mask
) + 1;
4591 csize
= nb_csectors
* QCOW2_COMPRESSED_SECTOR_SIZE
-
4592 (coffset
& ~QCOW2_COMPRESSED_SECTOR_MASK
);
4594 buf
= g_try_malloc(csize
);
4599 out_buf
= qemu_blockalign(bs
, s
->cluster_size
);
4601 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
4602 ret
= bdrv_co_pread(bs
->file
, coffset
, csize
, buf
, 0);
4607 if (qcow2_co_decompress(bs
, out_buf
, s
->cluster_size
, buf
, csize
) < 0) {
4612 qemu_iovec_from_buf(qiov
, qiov_offset
, out_buf
+ offset_in_cluster
, bytes
);
4615 qemu_vfree(out_buf
);
4621 static int make_completely_empty(BlockDriverState
*bs
)
4623 BDRVQcow2State
*s
= bs
->opaque
;
4624 Error
*local_err
= NULL
;
4625 int ret
, l1_clusters
;
4627 uint64_t *new_reftable
= NULL
;
4628 uint64_t rt_entry
, l1_size2
;
4631 uint64_t reftable_offset
;
4632 uint32_t reftable_clusters
;
4633 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
4635 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
4640 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
4645 /* Refcounts will be broken utterly */
4646 ret
= qcow2_mark_dirty(bs
);
4651 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4653 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
4654 l1_size2
= (uint64_t)s
->l1_size
* sizeof(uint64_t);
4656 /* After this call, neither the in-memory nor the on-disk refcount
4657 * information accurately describe the actual references */
4659 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
4660 l1_clusters
* s
->cluster_size
, 0);
4662 goto fail_broken_refcounts
;
4664 memset(s
->l1_table
, 0, l1_size2
);
4666 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
4668 /* Overwrite enough clusters at the beginning of the sectors to place
4669 * the refcount table, a refcount block and the L1 table in; this may
4670 * overwrite parts of the existing refcount and L1 table, which is not
4671 * an issue because the dirty flag is set, complete data loss is in fact
4672 * desired and partial data loss is consequently fine as well */
4673 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
4674 (2 + l1_clusters
) * s
->cluster_size
, 0);
4675 /* This call (even if it failed overall) may have overwritten on-disk
4676 * refcount structures; in that case, the in-memory refcount information
4677 * will probably differ from the on-disk information which makes the BDS
4680 goto fail_broken_refcounts
;
4683 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4684 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
4686 /* "Create" an empty reftable (one cluster) directly after the image
4687 * header and an empty L1 table three clusters after the image header;
4688 * the cluster between those two will be used as the first refblock */
4689 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
4690 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
4691 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
4692 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
4693 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
4695 goto fail_broken_refcounts
;
4698 s
->l1_table_offset
= 3 * s
->cluster_size
;
4700 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ sizeof(uint64_t));
4701 if (!new_reftable
) {
4703 goto fail_broken_refcounts
;
4706 s
->refcount_table_offset
= s
->cluster_size
;
4707 s
->refcount_table_size
= s
->cluster_size
/ sizeof(uint64_t);
4708 s
->max_refcount_table_index
= 0;
4710 g_free(s
->refcount_table
);
4711 s
->refcount_table
= new_reftable
;
4712 new_reftable
= NULL
;
4714 /* Now the in-memory refcount information again corresponds to the on-disk
4715 * information (reftable is empty and no refblocks (the refblock cache is
4716 * empty)); however, this means some clusters (e.g. the image header) are
4717 * referenced, but not refcounted, but the normal qcow2 code assumes that
4718 * the in-memory information is always correct */
4720 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
4722 /* Enter the first refblock into the reftable */
4723 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
4724 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
4725 &rt_entry
, sizeof(rt_entry
));
4727 goto fail_broken_refcounts
;
4729 s
->refcount_table
[0] = 2 * s
->cluster_size
;
4731 s
->free_cluster_index
= 0;
4732 assert(3 + l1_clusters
<= s
->refcount_block_size
);
4733 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
4736 goto fail_broken_refcounts
;
4737 } else if (offset
> 0) {
4738 error_report("First cluster in emptied image is in use");
4742 /* Now finally the in-memory information corresponds to the on-disk
4743 * structures and is correct */
4744 ret
= qcow2_mark_clean(bs
);
4749 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
, false,
4750 PREALLOC_MODE_OFF
, 0, &local_err
);
4752 error_report_err(local_err
);
4758 fail_broken_refcounts
:
4759 /* The BDS is unusable at this point. If we wanted to make it usable, we
4760 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4761 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4762 * again. However, because the functions which could have caused this error
4763 * path to be taken are used by those functions as well, it's very likely
4764 * that that sequence will fail as well. Therefore, just eject the BDS. */
4768 g_free(new_reftable
);
4772 static int qcow2_make_empty(BlockDriverState
*bs
)
4774 BDRVQcow2State
*s
= bs
->opaque
;
4775 uint64_t offset
, end_offset
;
4776 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
4777 int l1_clusters
, ret
= 0;
4779 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
4781 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
4782 3 + l1_clusters
<= s
->refcount_block_size
&&
4783 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
&&
4784 !has_data_file(bs
)) {
4785 /* The following function only works for qcow2 v3 images (it
4786 * requires the dirty flag) and only as long as there are no
4787 * features that reserve extra clusters (such as snapshots,
4788 * LUKS header, or persistent bitmaps), because it completely
4789 * empties the image. Furthermore, the L1 table and three
4790 * additional clusters (image header, refcount table, one
4791 * refcount block) have to fit inside one refcount block. It
4792 * only resets the image file, i.e. does not work with an
4793 * external data file. */
4794 return make_completely_empty(bs
);
4797 /* This fallback code simply discards every active cluster; this is slow,
4798 * but works in all cases */
4799 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4800 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
4801 /* As this function is generally used after committing an external
4802 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4803 * default action for this kind of discard is to pass the discard,
4804 * which will ideally result in an actually smaller image file, as
4805 * is probably desired. */
4806 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
4807 QCOW2_DISCARD_SNAPSHOT
, true);
4816 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
4818 BDRVQcow2State
*s
= bs
->opaque
;
4821 qemu_co_mutex_lock(&s
->lock
);
4822 ret
= qcow2_write_caches(bs
);
4823 qemu_co_mutex_unlock(&s
->lock
);
4828 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
4831 Error
*local_err
= NULL
;
4832 BlockMeasureInfo
*info
;
4833 uint64_t required
= 0; /* bytes that contribute to required size */
4834 uint64_t virtual_size
; /* disk size as seen by guest */
4835 uint64_t refcount_bits
;
4837 uint64_t luks_payload_size
= 0;
4838 size_t cluster_size
;
4841 PreallocMode prealloc
;
4842 bool has_backing_file
;
4844 bool extended_l2
= false; /* Set to false until the option is added */
4847 /* Parse image creation options */
4848 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, &local_err
);
4853 version
= qcow2_opt_get_version_del(opts
, &local_err
);
4858 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
4863 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
4864 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
4865 PREALLOC_MODE_OFF
, &local_err
);
4871 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
4872 has_backing_file
= !!optstr
;
4875 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
4876 has_luks
= optstr
&& strcmp(optstr
, "luks") == 0;
4880 g_autoptr(QCryptoBlockCreateOptions
) create_opts
= NULL
;
4881 QDict
*cryptoopts
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
4884 create_opts
= block_crypto_create_opts_init(cryptoopts
, errp
);
4885 qobject_unref(cryptoopts
);
4890 if (!qcrypto_block_calculate_payload_offset(create_opts
,
4897 luks_payload_size
= ROUND_UP(headerlen
, cluster_size
);
4900 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
4901 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
4903 /* Check that virtual disk size is valid */
4904 l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
4905 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
4906 cluster_size
/ l2e_size
);
4907 if (l2_tables
* sizeof(uint64_t) > QCOW_MAX_L1_SIZE
) {
4908 error_setg(&local_err
, "The image size is too large "
4909 "(try using a larger cluster size)");
4913 /* Account for input image */
4915 int64_t ssize
= bdrv_getlength(in_bs
);
4917 error_setg_errno(&local_err
, -ssize
,
4918 "Unable to get image virtual_size");
4922 virtual_size
= ROUND_UP(ssize
, cluster_size
);
4924 if (has_backing_file
) {
4925 /* We don't how much of the backing chain is shared by the input
4926 * image and the new image file. In the worst case the new image's
4927 * backing file has nothing in common with the input image. Be
4928 * conservative and assume all clusters need to be written.
4930 required
= virtual_size
;
4935 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
4938 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
4939 ssize
- offset
, &pnum
, NULL
,
4942 error_setg_errno(&local_err
, -ret
,
4943 "Unable to get block status");
4947 if (ret
& BDRV_BLOCK_ZERO
) {
4948 /* Skip zero regions (safe with no backing file) */
4949 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
4950 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
4951 /* Extend pnum to end of cluster for next iteration */
4952 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
4954 /* Count clusters we've seen */
4955 required
+= offset
% cluster_size
+ pnum
;
4961 /* Take into account preallocation. Nothing special is needed for
4962 * PREALLOC_MODE_METADATA since metadata is always counted.
4964 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
4965 required
= virtual_size
;
4968 info
= g_new0(BlockMeasureInfo
, 1);
4969 info
->fully_allocated
= luks_payload_size
+
4970 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
4971 ctz32(refcount_bits
), extended_l2
);
4974 * Remove data clusters that are not required. This overestimates the
4975 * required size because metadata needed for the fully allocated file is
4976 * still counted. Show bitmaps only if both source and destination
4977 * would support them.
4979 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
4980 info
->has_bitmaps
= version
>= 3 && in_bs
&&
4981 bdrv_supports_persistent_dirty_bitmap(in_bs
);
4982 if (info
->has_bitmaps
) {
4983 info
->bitmaps
= qcow2_get_persistent_dirty_bitmap_size(in_bs
,
4989 error_propagate(errp
, local_err
);
4993 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4995 BDRVQcow2State
*s
= bs
->opaque
;
4996 bdi
->cluster_size
= s
->cluster_size
;
4997 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
5001 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
,
5004 BDRVQcow2State
*s
= bs
->opaque
;
5005 ImageInfoSpecific
*spec_info
;
5006 QCryptoBlockInfo
*encrypt_info
= NULL
;
5007 Error
*local_err
= NULL
;
5009 if (s
->crypto
!= NULL
) {
5010 encrypt_info
= qcrypto_block_get_info(s
->crypto
, &local_err
);
5012 error_propagate(errp
, local_err
);
5017 spec_info
= g_new(ImageInfoSpecific
, 1);
5018 *spec_info
= (ImageInfoSpecific
){
5019 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
5020 .u
.qcow2
.data
= g_new0(ImageInfoSpecificQCow2
, 1),
5022 if (s
->qcow_version
== 2) {
5023 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5024 .compat
= g_strdup("0.10"),
5025 .refcount_bits
= s
->refcount_bits
,
5027 } else if (s
->qcow_version
== 3) {
5028 Qcow2BitmapInfoList
*bitmaps
;
5029 bitmaps
= qcow2_get_bitmap_info_list(bs
, &local_err
);
5031 error_propagate(errp
, local_err
);
5032 qapi_free_ImageInfoSpecific(spec_info
);
5033 qapi_free_QCryptoBlockInfo(encrypt_info
);
5036 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5037 .compat
= g_strdup("1.1"),
5038 .lazy_refcounts
= s
->compatible_features
&
5039 QCOW2_COMPAT_LAZY_REFCOUNTS
,
5040 .has_lazy_refcounts
= true,
5041 .corrupt
= s
->incompatible_features
&
5042 QCOW2_INCOMPAT_CORRUPT
,
5043 .has_corrupt
= true,
5044 .refcount_bits
= s
->refcount_bits
,
5045 .has_bitmaps
= !!bitmaps
,
5047 .has_data_file
= !!s
->image_data_file
,
5048 .data_file
= g_strdup(s
->image_data_file
),
5049 .has_data_file_raw
= has_data_file(bs
),
5050 .data_file_raw
= data_file_is_raw(bs
),
5051 .compression_type
= s
->compression_type
,
5054 /* if this assertion fails, this probably means a new version was
5055 * added without having it covered here */
5060 ImageInfoSpecificQCow2Encryption
*qencrypt
=
5061 g_new(ImageInfoSpecificQCow2Encryption
, 1);
5062 switch (encrypt_info
->format
) {
5063 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
5064 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
5066 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
5067 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
5068 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
5073 /* Since we did shallow copy above, erase any pointers
5074 * in the original info */
5075 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
5076 qapi_free_QCryptoBlockInfo(encrypt_info
);
5078 spec_info
->u
.qcow2
.data
->has_encrypt
= true;
5079 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
5085 static int qcow2_has_zero_init(BlockDriverState
*bs
)
5087 BDRVQcow2State
*s
= bs
->opaque
;
5090 if (qemu_in_coroutine()) {
5091 qemu_co_mutex_lock(&s
->lock
);
5094 * Check preallocation status: Preallocated images have all L2
5095 * tables allocated, nonpreallocated images have none. It is
5096 * therefore enough to check the first one.
5098 preallocated
= s
->l1_size
> 0 && s
->l1_table
[0] != 0;
5099 if (qemu_in_coroutine()) {
5100 qemu_co_mutex_unlock(&s
->lock
);
5103 if (!preallocated
) {
5105 } else if (bs
->encrypted
) {
5108 return bdrv_has_zero_init(s
->data_file
->bs
);
5112 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
5115 BDRVQcow2State
*s
= bs
->opaque
;
5117 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
5118 return bs
->drv
->bdrv_co_pwritev_part(bs
, qcow2_vm_state_offset(s
) + pos
,
5119 qiov
->size
, qiov
, 0, 0);
5122 static int qcow2_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
5125 BDRVQcow2State
*s
= bs
->opaque
;
5127 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
5128 return bs
->drv
->bdrv_co_preadv_part(bs
, qcow2_vm_state_offset(s
) + pos
,
5129 qiov
->size
, qiov
, 0, 0);
5133 * Downgrades an image's version. To achieve this, any incompatible features
5134 * have to be removed.
5136 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
5137 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5140 BDRVQcow2State
*s
= bs
->opaque
;
5141 int current_version
= s
->qcow_version
;
5145 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5146 assert(target_version
< current_version
);
5148 /* There are no other versions (now) that you can downgrade to */
5149 assert(target_version
== 2);
5151 if (s
->refcount_order
!= 4) {
5152 error_setg(errp
, "compat=0.10 requires refcount_bits=16");
5156 if (has_data_file(bs
)) {
5157 error_setg(errp
, "Cannot downgrade an image with a data file");
5162 * If any internal snapshot has a different size than the current
5163 * image size, or VM state size that exceeds 32 bits, downgrading
5164 * is unsafe. Even though we would still use v3-compliant output
5165 * to preserve that data, other v2 programs might not realize
5166 * those optional fields are important.
5168 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5169 if (s
->snapshots
[i
].vm_state_size
> UINT32_MAX
||
5170 s
->snapshots
[i
].disk_size
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
5171 error_setg(errp
, "Internal snapshots prevent downgrade of image");
5176 /* clear incompatible features */
5177 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
5178 ret
= qcow2_mark_clean(bs
);
5180 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5185 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5186 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5187 * best thing to do anyway */
5189 if (s
->incompatible_features
) {
5190 error_setg(errp
, "Cannot downgrade an image with incompatible features "
5191 "%#" PRIx64
" set", s
->incompatible_features
);
5195 /* since we can ignore compatible features, we can set them to 0 as well */
5196 s
->compatible_features
= 0;
5197 /* if lazy refcounts have been used, they have already been fixed through
5198 * clearing the dirty flag */
5200 /* clearing autoclear features is trivial */
5201 s
->autoclear_features
= 0;
5203 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
5205 error_setg_errno(errp
, -ret
, "Failed to turn zero into data clusters");
5209 s
->qcow_version
= target_version
;
5210 ret
= qcow2_update_header(bs
);
5212 s
->qcow_version
= current_version
;
5213 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5220 * Upgrades an image's version. While newer versions encompass all
5221 * features of older versions, some things may have to be presented
5224 static int qcow2_upgrade(BlockDriverState
*bs
, int target_version
,
5225 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5228 BDRVQcow2State
*s
= bs
->opaque
;
5229 bool need_snapshot_update
;
5230 int current_version
= s
->qcow_version
;
5234 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5235 assert(target_version
> current_version
);
5237 /* There are no other versions (yet) that you can upgrade to */
5238 assert(target_version
== 3);
5240 status_cb(bs
, 0, 2, cb_opaque
);
5243 * In v2, snapshots do not need to have extra data. v3 requires
5244 * the 64-bit VM state size and the virtual disk size to be
5246 * qcow2_write_snapshots() will always write the list in the
5247 * v3-compliant format.
5249 need_snapshot_update
= false;
5250 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5251 if (s
->snapshots
[i
].extra_data_size
<
5252 sizeof_field(QCowSnapshotExtraData
, vm_state_size_large
) +
5253 sizeof_field(QCowSnapshotExtraData
, disk_size
))
5255 need_snapshot_update
= true;
5259 if (need_snapshot_update
) {
5260 ret
= qcow2_write_snapshots(bs
);
5262 error_setg_errno(errp
, -ret
, "Failed to update the snapshot table");
5266 status_cb(bs
, 1, 2, cb_opaque
);
5268 s
->qcow_version
= target_version
;
5269 ret
= qcow2_update_header(bs
);
5271 s
->qcow_version
= current_version
;
5272 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5275 status_cb(bs
, 2, 2, cb_opaque
);
5280 typedef enum Qcow2AmendOperation
{
5281 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5282 * statically initialized to so that the helper CB can discern the first
5283 * invocation from an operation change */
5284 QCOW2_NO_OPERATION
= 0,
5287 QCOW2_UPDATING_ENCRYPTION
,
5288 QCOW2_CHANGING_REFCOUNT_ORDER
,
5290 } Qcow2AmendOperation
;
5292 typedef struct Qcow2AmendHelperCBInfo
{
5293 /* The code coordinating the amend operations should only modify
5294 * these four fields; the rest will be managed by the CB */
5295 BlockDriverAmendStatusCB
*original_status_cb
;
5296 void *original_cb_opaque
;
5298 Qcow2AmendOperation current_operation
;
5300 /* Total number of operations to perform (only set once) */
5301 int total_operations
;
5303 /* The following fields are managed by the CB */
5305 /* Number of operations completed */
5306 int operations_completed
;
5308 /* Cumulative offset of all completed operations */
5309 int64_t offset_completed
;
5311 Qcow2AmendOperation last_operation
;
5312 int64_t last_work_size
;
5313 } Qcow2AmendHelperCBInfo
;
5315 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
5316 int64_t operation_offset
,
5317 int64_t operation_work_size
, void *opaque
)
5319 Qcow2AmendHelperCBInfo
*info
= opaque
;
5320 int64_t current_work_size
;
5321 int64_t projected_work_size
;
5323 if (info
->current_operation
!= info
->last_operation
) {
5324 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
5325 info
->offset_completed
+= info
->last_work_size
;
5326 info
->operations_completed
++;
5329 info
->last_operation
= info
->current_operation
;
5332 assert(info
->total_operations
> 0);
5333 assert(info
->operations_completed
< info
->total_operations
);
5335 info
->last_work_size
= operation_work_size
;
5337 current_work_size
= info
->offset_completed
+ operation_work_size
;
5339 /* current_work_size is the total work size for (operations_completed + 1)
5340 * operations (which includes this one), so multiply it by the number of
5341 * operations not covered and divide it by the number of operations
5342 * covered to get a projection for the operations not covered */
5343 projected_work_size
= current_work_size
* (info
->total_operations
-
5344 info
->operations_completed
- 1)
5345 / (info
->operations_completed
+ 1);
5347 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
5348 current_work_size
+ projected_work_size
,
5349 info
->original_cb_opaque
);
5352 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5353 BlockDriverAmendStatusCB
*status_cb
,
5358 BDRVQcow2State
*s
= bs
->opaque
;
5359 int old_version
= s
->qcow_version
, new_version
= old_version
;
5360 uint64_t new_size
= 0;
5361 const char *backing_file
= NULL
, *backing_format
= NULL
, *data_file
= NULL
;
5362 bool lazy_refcounts
= s
->use_lazy_refcounts
;
5363 bool data_file_raw
= data_file_is_raw(bs
);
5364 const char *compat
= NULL
;
5365 int refcount_bits
= s
->refcount_bits
;
5367 QemuOptDesc
*desc
= opts
->list
->desc
;
5368 Qcow2AmendHelperCBInfo helper_cb_info
;
5369 bool encryption_update
= false;
5371 while (desc
&& desc
->name
) {
5372 if (!qemu_opt_find(opts
, desc
->name
)) {
5373 /* only change explicitly defined options */
5378 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
5379 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
5381 /* preserve default */
5382 } else if (!strcmp(compat
, "0.10") || !strcmp(compat
, "v2")) {
5384 } else if (!strcmp(compat
, "1.1") || !strcmp(compat
, "v3")) {
5387 error_setg(errp
, "Unknown compatibility level %s", compat
);
5390 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
5391 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
5392 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
5393 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5394 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
5395 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5396 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
5399 "Can't amend encryption options - encryption not present");
5402 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5404 "Only LUKS encryption options can be amended");
5407 encryption_update
= true;
5408 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
5409 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
5411 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
5412 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
5415 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
5416 !is_power_of_2(refcount_bits
))
5418 error_setg(errp
, "Refcount width must be a power of two and "
5419 "may not exceed 64 bits");
5422 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE
)) {
5423 data_file
= qemu_opt_get(opts
, BLOCK_OPT_DATA_FILE
);
5424 if (data_file
&& !has_data_file(bs
)) {
5425 error_setg(errp
, "data-file can only be set for images that "
5426 "use an external data file");
5429 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE_RAW
)) {
5430 data_file_raw
= qemu_opt_get_bool(opts
, BLOCK_OPT_DATA_FILE_RAW
,
5432 if (data_file_raw
&& !data_file_is_raw(bs
)) {
5433 error_setg(errp
, "data-file-raw cannot be set on existing "
5438 /* if this point is reached, this probably means a new option was
5439 * added without having it covered here */
5446 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
5447 .original_status_cb
= status_cb
,
5448 .original_cb_opaque
= cb_opaque
,
5449 .total_operations
= (new_version
!= old_version
)
5450 + (s
->refcount_bits
!= refcount_bits
) +
5451 (encryption_update
== true)
5454 /* Upgrade first (some features may require compat=1.1) */
5455 if (new_version
> old_version
) {
5456 helper_cb_info
.current_operation
= QCOW2_UPGRADING
;
5457 ret
= qcow2_upgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5458 &helper_cb_info
, errp
);
5464 if (encryption_update
) {
5465 QDict
*amend_opts_dict
;
5466 QCryptoBlockAmendOptions
*amend_opts
;
5468 helper_cb_info
.current_operation
= QCOW2_UPDATING_ENCRYPTION
;
5469 amend_opts_dict
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
5470 if (!amend_opts_dict
) {
5473 amend_opts
= block_crypto_amend_opts_init(amend_opts_dict
, errp
);
5474 qobject_unref(amend_opts_dict
);
5478 ret
= qcrypto_block_amend_options(s
->crypto
,
5479 qcow2_crypto_hdr_read_func
,
5480 qcow2_crypto_hdr_write_func
,
5485 qapi_free_QCryptoBlockAmendOptions(amend_opts
);
5491 if (s
->refcount_bits
!= refcount_bits
) {
5492 int refcount_order
= ctz32(refcount_bits
);
5494 if (new_version
< 3 && refcount_bits
!= 16) {
5495 error_setg(errp
, "Refcount widths other than 16 bits require "
5496 "compatibility level 1.1 or above (use compat=1.1 or "
5501 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
5502 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
5503 &qcow2_amend_helper_cb
,
5504 &helper_cb_info
, errp
);
5510 /* data-file-raw blocks backing files, so clear it first if requested */
5511 if (data_file_raw
) {
5512 s
->autoclear_features
|= QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5514 s
->autoclear_features
&= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5518 g_free(s
->image_data_file
);
5519 s
->image_data_file
= *data_file
? g_strdup(data_file
) : NULL
;
5522 ret
= qcow2_update_header(bs
);
5524 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5528 if (backing_file
|| backing_format
) {
5529 if (g_strcmp0(backing_file
, s
->image_backing_file
) ||
5530 g_strcmp0(backing_format
, s
->image_backing_format
)) {
5531 warn_report("Deprecated use of amend to alter the backing file; "
5532 "use qemu-img rebase instead");
5534 ret
= qcow2_change_backing_file(bs
,
5535 backing_file
?: s
->image_backing_file
,
5536 backing_format
?: s
->image_backing_format
);
5538 error_setg_errno(errp
, -ret
, "Failed to change the backing file");
5543 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
5544 if (lazy_refcounts
) {
5545 if (new_version
< 3) {
5546 error_setg(errp
, "Lazy refcounts only supported with "
5547 "compatibility level 1.1 and above (use compat=1.1 "
5551 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5552 ret
= qcow2_update_header(bs
);
5554 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5555 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5558 s
->use_lazy_refcounts
= true;
5560 /* make image clean first */
5561 ret
= qcow2_mark_clean(bs
);
5563 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5566 /* now disallow lazy refcounts */
5567 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5568 ret
= qcow2_update_header(bs
);
5570 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5571 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5574 s
->use_lazy_refcounts
= false;
5579 BlockBackend
*blk
= blk_new_with_bs(bs
, BLK_PERM_RESIZE
, BLK_PERM_ALL
,
5586 * Amending image options should ensure that the image has
5587 * exactly the given new values, so pass exact=true here.
5589 ret
= blk_truncate(blk
, new_size
, true, PREALLOC_MODE_OFF
, 0, errp
);
5596 /* Downgrade last (so unsupported features can be removed before) */
5597 if (new_version
< old_version
) {
5598 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
5599 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5600 &helper_cb_info
, errp
);
5609 static int coroutine_fn
qcow2_co_amend(BlockDriverState
*bs
,
5610 BlockdevAmendOptions
*opts
,
5614 BlockdevAmendOptionsQcow2
*qopts
= &opts
->u
.qcow2
;
5615 BDRVQcow2State
*s
= bs
->opaque
;
5618 if (qopts
->has_encrypt
) {
5620 error_setg(errp
, "image is not encrypted, can't amend");
5624 if (qopts
->encrypt
->format
!= Q_CRYPTO_BLOCK_FORMAT_LUKS
) {
5626 "Amend can't be used to change the qcow2 encryption format");
5630 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5632 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5636 ret
= qcrypto_block_amend_options(s
->crypto
,
5637 qcow2_crypto_hdr_read_func
,
5638 qcow2_crypto_hdr_write_func
,
5648 * If offset or size are negative, respectively, they will not be included in
5649 * the BLOCK_IMAGE_CORRUPTED event emitted.
5650 * fatal will be ignored for read-only BDS; corruptions found there will always
5651 * be considered non-fatal.
5653 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
5654 int64_t size
, const char *message_format
, ...)
5656 BDRVQcow2State
*s
= bs
->opaque
;
5657 const char *node_name
;
5661 fatal
= fatal
&& bdrv_is_writable(bs
);
5663 if (s
->signaled_corruption
&&
5664 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
5669 va_start(ap
, message_format
);
5670 message
= g_strdup_vprintf(message_format
, ap
);
5674 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
5675 "corruption events will be suppressed\n", message
);
5677 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
5678 "corruption events will be suppressed\n", message
);
5681 node_name
= bdrv_get_node_name(bs
);
5682 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
5683 *node_name
!= '\0', node_name
,
5684 message
, offset
>= 0, offset
,
5690 qcow2_mark_corrupt(bs
);
5691 bs
->drv
= NULL
; /* make BDS unusable */
5694 s
->signaled_corruption
= true;
5697 #define QCOW_COMMON_OPTIONS \
5699 .name = BLOCK_OPT_SIZE, \
5700 .type = QEMU_OPT_SIZE, \
5701 .help = "Virtual disk size" \
5704 .name = BLOCK_OPT_COMPAT_LEVEL, \
5705 .type = QEMU_OPT_STRING, \
5706 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
5709 .name = BLOCK_OPT_BACKING_FILE, \
5710 .type = QEMU_OPT_STRING, \
5711 .help = "File name of a base image" \
5714 .name = BLOCK_OPT_BACKING_FMT, \
5715 .type = QEMU_OPT_STRING, \
5716 .help = "Image format of the base image" \
5719 .name = BLOCK_OPT_DATA_FILE, \
5720 .type = QEMU_OPT_STRING, \
5721 .help = "File name of an external data file" \
5724 .name = BLOCK_OPT_DATA_FILE_RAW, \
5725 .type = QEMU_OPT_BOOL, \
5726 .help = "The external data file must stay valid " \
5730 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
5731 .type = QEMU_OPT_BOOL, \
5732 .help = "Postpone refcount updates", \
5733 .def_value_str = "off" \
5736 .name = BLOCK_OPT_REFCOUNT_BITS, \
5737 .type = QEMU_OPT_NUMBER, \
5738 .help = "Width of a reference count entry in bits", \
5739 .def_value_str = "16" \
5742 static QemuOptsList qcow2_create_opts
= {
5743 .name
= "qcow2-create-opts",
5744 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
5747 .name
= BLOCK_OPT_ENCRYPT
, \
5748 .type
= QEMU_OPT_BOOL
, \
5749 .help
= "Encrypt the image with format 'aes'. (Deprecated " \
5750 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)", \
5753 .name
= BLOCK_OPT_ENCRYPT_FORMAT
, \
5754 .type
= QEMU_OPT_STRING
, \
5755 .help
= "Encrypt the image, format choices: 'aes', 'luks'", \
5757 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
5758 "ID of secret providing qcow AES key or LUKS passphrase"), \
5759 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
5760 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
5761 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
5762 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
5763 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
5764 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
5766 .name
= BLOCK_OPT_CLUSTER_SIZE
, \
5767 .type
= QEMU_OPT_SIZE
, \
5768 .help
= "qcow2 cluster size", \
5769 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
) \
5772 .name
= BLOCK_OPT_PREALLOC
, \
5773 .type
= QEMU_OPT_STRING
, \
5774 .help
= "Preallocation mode (allowed values: off, " \
5775 "metadata, falloc, full)" \
5778 .name
= BLOCK_OPT_COMPRESSION_TYPE
, \
5779 .type
= QEMU_OPT_STRING
, \
5780 .help
= "Compression method used for image cluster " \
5782 .def_value_str
= "zlib" \
5784 QCOW_COMMON_OPTIONS
,
5785 { /* end of list */ }
5789 static QemuOptsList qcow2_amend_opts
= {
5790 .name
= "qcow2-amend-opts",
5791 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts
.head
),
5793 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
5794 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
5795 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
5796 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
5797 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
5798 QCOW_COMMON_OPTIONS
,
5799 { /* end of list */ }
5803 static const char *const qcow2_strong_runtime_opts
[] = {
5804 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET
,
5809 BlockDriver bdrv_qcow2
= {
5810 .format_name
= "qcow2",
5811 .instance_size
= sizeof(BDRVQcow2State
),
5812 .bdrv_probe
= qcow2_probe
,
5813 .bdrv_open
= qcow2_open
,
5814 .bdrv_close
= qcow2_close
,
5815 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
5816 .bdrv_reopen_commit
= qcow2_reopen_commit
,
5817 .bdrv_reopen_commit_post
= qcow2_reopen_commit_post
,
5818 .bdrv_reopen_abort
= qcow2_reopen_abort
,
5819 .bdrv_join_options
= qcow2_join_options
,
5820 .bdrv_child_perm
= bdrv_default_perms
,
5821 .bdrv_co_create_opts
= qcow2_co_create_opts
,
5822 .bdrv_co_create
= qcow2_co_create
,
5823 .bdrv_has_zero_init
= qcow2_has_zero_init
,
5824 .bdrv_co_block_status
= qcow2_co_block_status
,
5826 .bdrv_co_preadv_part
= qcow2_co_preadv_part
,
5827 .bdrv_co_pwritev_part
= qcow2_co_pwritev_part
,
5828 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
5830 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
5831 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
5832 .bdrv_co_copy_range_from
= qcow2_co_copy_range_from
,
5833 .bdrv_co_copy_range_to
= qcow2_co_copy_range_to
,
5834 .bdrv_co_truncate
= qcow2_co_truncate
,
5835 .bdrv_co_pwritev_compressed_part
= qcow2_co_pwritev_compressed_part
,
5836 .bdrv_make_empty
= qcow2_make_empty
,
5838 .bdrv_snapshot_create
= qcow2_snapshot_create
,
5839 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
5840 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
5841 .bdrv_snapshot_list
= qcow2_snapshot_list
,
5842 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
5843 .bdrv_measure
= qcow2_measure
,
5844 .bdrv_get_info
= qcow2_get_info
,
5845 .bdrv_get_specific_info
= qcow2_get_specific_info
,
5847 .bdrv_save_vmstate
= qcow2_save_vmstate
,
5848 .bdrv_load_vmstate
= qcow2_load_vmstate
,
5851 .supports_backing
= true,
5852 .bdrv_change_backing_file
= qcow2_change_backing_file
,
5854 .bdrv_refresh_limits
= qcow2_refresh_limits
,
5855 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
5856 .bdrv_inactivate
= qcow2_inactivate
,
5858 .create_opts
= &qcow2_create_opts
,
5859 .amend_opts
= &qcow2_amend_opts
,
5860 .strong_runtime_opts
= qcow2_strong_runtime_opts
,
5861 .mutable_opts
= mutable_opts
,
5862 .bdrv_co_check
= qcow2_co_check
,
5863 .bdrv_amend_options
= qcow2_amend_options
,
5864 .bdrv_co_amend
= qcow2_co_amend
,
5866 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
5867 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
5869 .bdrv_supports_persistent_dirty_bitmap
=
5870 qcow2_supports_persistent_dirty_bitmap
,
5871 .bdrv_co_can_store_new_dirty_bitmap
= qcow2_co_can_store_new_dirty_bitmap
,
5872 .bdrv_co_remove_persistent_dirty_bitmap
=
5873 qcow2_co_remove_persistent_dirty_bitmap
,
5876 static void bdrv_qcow2_init(void)
5878 bdrv_register(&bdrv_qcow2
);
5881 block_init(bdrv_qcow2_init
);