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_free(s
->cache_clean_timer
);
856 s
->cache_clean_timer
= NULL
;
860 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
862 cache_clean_timer_del(bs
);
865 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
866 AioContext
*new_context
)
868 cache_clean_timer_init(bs
, new_context
);
871 static void read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
872 uint64_t *l2_cache_size
,
873 uint64_t *l2_cache_entry_size
,
874 uint64_t *refcount_cache_size
, Error
**errp
)
876 BDRVQcow2State
*s
= bs
->opaque
;
877 uint64_t combined_cache_size
, l2_cache_max_setting
;
878 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
879 bool l2_cache_entry_size_set
;
880 int min_refcount_cache
= MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
881 uint64_t virtual_disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
882 uint64_t max_l2_entries
= DIV_ROUND_UP(virtual_disk_size
, s
->cluster_size
);
883 /* An L2 table is always one cluster in size so the max cache size
884 * should be a multiple of the cluster size. */
885 uint64_t max_l2_cache
= ROUND_UP(max_l2_entries
* l2_entry_size(s
),
888 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
889 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
890 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
891 l2_cache_entry_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
);
893 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
894 l2_cache_max_setting
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
,
895 DEFAULT_L2_CACHE_MAX_SIZE
);
896 *refcount_cache_size
= qemu_opt_get_size(opts
,
897 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
899 *l2_cache_entry_size
= qemu_opt_get_size(
900 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
902 *l2_cache_size
= MIN(max_l2_cache
, l2_cache_max_setting
);
904 if (combined_cache_size_set
) {
905 if (l2_cache_size_set
&& refcount_cache_size_set
) {
906 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
907 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
910 } else if (l2_cache_size_set
&&
911 (l2_cache_max_setting
> combined_cache_size
)) {
912 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
913 QCOW2_OPT_CACHE_SIZE
);
915 } else if (*refcount_cache_size
> combined_cache_size
) {
916 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
917 QCOW2_OPT_CACHE_SIZE
);
921 if (l2_cache_size_set
) {
922 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
923 } else if (refcount_cache_size_set
) {
924 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
926 /* Assign as much memory as possible to the L2 cache, and
927 * use the remainder for the refcount cache */
928 if (combined_cache_size
>= max_l2_cache
+ min_refcount_cache
) {
929 *l2_cache_size
= max_l2_cache
;
930 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
932 *refcount_cache_size
=
933 MIN(combined_cache_size
, min_refcount_cache
);
934 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
940 * If the L2 cache is not enough to cover the whole disk then
941 * default to 4KB entries. Smaller entries reduce the cost of
942 * loads and evictions and increase I/O performance.
944 if (*l2_cache_size
< max_l2_cache
&& !l2_cache_entry_size_set
) {
945 *l2_cache_entry_size
= MIN(s
->cluster_size
, 4096);
948 /* l2_cache_size and refcount_cache_size are ensured to have at least
949 * their minimum values in qcow2_update_options_prepare() */
951 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
952 *l2_cache_entry_size
> s
->cluster_size
||
953 !is_power_of_2(*l2_cache_entry_size
)) {
954 error_setg(errp
, "L2 cache entry size must be a power of two "
955 "between %d and the cluster size (%d)",
956 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
961 typedef struct Qcow2ReopenState
{
962 Qcow2Cache
*l2_table_cache
;
963 Qcow2Cache
*refcount_block_cache
;
964 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
965 bool use_lazy_refcounts
;
967 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
968 uint64_t cache_clean_interval
;
969 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
972 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
974 QDict
*options
, int flags
,
977 BDRVQcow2State
*s
= bs
->opaque
;
978 QemuOpts
*opts
= NULL
;
979 const char *opt_overlap_check
, *opt_overlap_check_template
;
980 int overlap_check_template
= 0;
981 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
983 const char *encryptfmt
;
984 QDict
*encryptopts
= NULL
;
985 Error
*local_err
= NULL
;
988 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
989 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
991 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
992 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
997 /* get L2 table/refcount block cache size from command line options */
998 read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
999 &refcount_cache_size
, &local_err
);
1001 error_propagate(errp
, local_err
);
1006 l2_cache_size
/= l2_cache_entry_size
;
1007 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
1008 l2_cache_size
= MIN_L2_CACHE_SIZE
;
1010 if (l2_cache_size
> INT_MAX
) {
1011 error_setg(errp
, "L2 cache size too big");
1016 refcount_cache_size
/= s
->cluster_size
;
1017 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
1018 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
1020 if (refcount_cache_size
> INT_MAX
) {
1021 error_setg(errp
, "Refcount cache size too big");
1026 /* alloc new L2 table/refcount block cache, flush old one */
1027 if (s
->l2_table_cache
) {
1028 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1030 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
1035 if (s
->refcount_block_cache
) {
1036 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1038 error_setg_errno(errp
, -ret
,
1039 "Failed to flush the refcount block cache");
1044 r
->l2_slice_size
= l2_cache_entry_size
/ l2_entry_size(s
);
1045 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
1046 l2_cache_entry_size
);
1047 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
1049 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
1050 error_setg(errp
, "Could not allocate metadata caches");
1055 /* New interval for cache cleanup timer */
1056 r
->cache_clean_interval
=
1057 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
1058 DEFAULT_CACHE_CLEAN_INTERVAL
);
1059 #ifndef CONFIG_LINUX
1060 if (r
->cache_clean_interval
!= 0) {
1061 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1062 " not supported on this host");
1067 if (r
->cache_clean_interval
> UINT_MAX
) {
1068 error_setg(errp
, "Cache clean interval too big");
1073 /* lazy-refcounts; flush if going from enabled to disabled */
1074 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
1075 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
1076 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
1077 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
1078 "qemu 1.1 compatibility level");
1083 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
1084 ret
= qcow2_mark_clean(bs
);
1086 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
1091 /* Overlap check options */
1092 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
1093 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1094 if (opt_overlap_check_template
&& opt_overlap_check
&&
1095 strcmp(opt_overlap_check_template
, opt_overlap_check
))
1097 error_setg(errp
, "Conflicting values for qcow2 options '"
1098 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1099 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
1103 if (!opt_overlap_check
) {
1104 opt_overlap_check
= opt_overlap_check_template
?: "cached";
1107 if (!strcmp(opt_overlap_check
, "none")) {
1108 overlap_check_template
= 0;
1109 } else if (!strcmp(opt_overlap_check
, "constant")) {
1110 overlap_check_template
= QCOW2_OL_CONSTANT
;
1111 } else if (!strcmp(opt_overlap_check
, "cached")) {
1112 overlap_check_template
= QCOW2_OL_CACHED
;
1113 } else if (!strcmp(opt_overlap_check
, "all")) {
1114 overlap_check_template
= QCOW2_OL_ALL
;
1116 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1117 "'overlap-check'. Allowed are any of the following: "
1118 "none, constant, cached, all", opt_overlap_check
);
1123 r
->overlap_check
= 0;
1124 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1125 /* overlap-check defines a template bitmask, but every flag may be
1126 * overwritten through the associated boolean option */
1128 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1129 overlap_check_template
& (1 << i
)) << i
;
1132 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1133 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1134 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1135 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1136 flags
& BDRV_O_UNMAP
);
1137 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1138 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1139 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1140 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1142 switch (s
->crypt_method_header
) {
1143 case QCOW_CRYPT_NONE
:
1145 error_setg(errp
, "No encryption in image header, but options "
1146 "specified format '%s'", encryptfmt
);
1152 case QCOW_CRYPT_AES
:
1153 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1155 "Header reported 'aes' encryption format but "
1156 "options specify '%s'", encryptfmt
);
1160 qdict_put_str(encryptopts
, "format", "qcow");
1161 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1164 case QCOW_CRYPT_LUKS
:
1165 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1167 "Header reported 'luks' encryption format but "
1168 "options specify '%s'", encryptfmt
);
1172 qdict_put_str(encryptopts
, "format", "luks");
1173 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1177 error_setg(errp
, "Unsupported encryption method %d",
1178 s
->crypt_method_header
);
1181 if (s
->crypt_method_header
!= QCOW_CRYPT_NONE
&& !r
->crypto_opts
) {
1188 qobject_unref(encryptopts
);
1189 qemu_opts_del(opts
);
1194 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1195 Qcow2ReopenState
*r
)
1197 BDRVQcow2State
*s
= bs
->opaque
;
1200 if (s
->l2_table_cache
) {
1201 qcow2_cache_destroy(s
->l2_table_cache
);
1203 if (s
->refcount_block_cache
) {
1204 qcow2_cache_destroy(s
->refcount_block_cache
);
1206 s
->l2_table_cache
= r
->l2_table_cache
;
1207 s
->refcount_block_cache
= r
->refcount_block_cache
;
1208 s
->l2_slice_size
= r
->l2_slice_size
;
1210 s
->overlap_check
= r
->overlap_check
;
1211 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1213 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1214 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1217 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1218 cache_clean_timer_del(bs
);
1219 s
->cache_clean_interval
= r
->cache_clean_interval
;
1220 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1223 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1224 s
->crypto_opts
= r
->crypto_opts
;
1227 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1228 Qcow2ReopenState
*r
)
1230 if (r
->l2_table_cache
) {
1231 qcow2_cache_destroy(r
->l2_table_cache
);
1233 if (r
->refcount_block_cache
) {
1234 qcow2_cache_destroy(r
->refcount_block_cache
);
1236 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1239 static int qcow2_update_options(BlockDriverState
*bs
, QDict
*options
,
1240 int flags
, Error
**errp
)
1242 Qcow2ReopenState r
= {};
1245 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1247 qcow2_update_options_commit(bs
, &r
);
1249 qcow2_update_options_abort(bs
, &r
);
1255 static int validate_compression_type(BDRVQcow2State
*s
, Error
**errp
)
1257 switch (s
->compression_type
) {
1258 case QCOW2_COMPRESSION_TYPE_ZLIB
:
1260 case QCOW2_COMPRESSION_TYPE_ZSTD
:
1265 error_setg(errp
, "qcow2: unknown compression type: %u",
1266 s
->compression_type
);
1271 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1272 * the incompatible feature flag must be set
1274 if (s
->compression_type
== QCOW2_COMPRESSION_TYPE_ZLIB
) {
1275 if (s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
) {
1276 error_setg(errp
, "qcow2: Compression type incompatible feature "
1277 "bit must not be set");
1281 if (!(s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
)) {
1282 error_setg(errp
, "qcow2: Compression type incompatible feature "
1291 /* Called with s->lock held. */
1292 static int coroutine_fn
qcow2_do_open(BlockDriverState
*bs
, QDict
*options
,
1293 int flags
, Error
**errp
)
1295 BDRVQcow2State
*s
= bs
->opaque
;
1296 unsigned int len
, i
;
1299 Error
*local_err
= NULL
;
1301 uint64_t l1_vm_state_index
;
1302 bool update_header
= false;
1304 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
1306 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1309 header
.magic
= be32_to_cpu(header
.magic
);
1310 header
.version
= be32_to_cpu(header
.version
);
1311 header
.backing_file_offset
= be64_to_cpu(header
.backing_file_offset
);
1312 header
.backing_file_size
= be32_to_cpu(header
.backing_file_size
);
1313 header
.size
= be64_to_cpu(header
.size
);
1314 header
.cluster_bits
= be32_to_cpu(header
.cluster_bits
);
1315 header
.crypt_method
= be32_to_cpu(header
.crypt_method
);
1316 header
.l1_table_offset
= be64_to_cpu(header
.l1_table_offset
);
1317 header
.l1_size
= be32_to_cpu(header
.l1_size
);
1318 header
.refcount_table_offset
= be64_to_cpu(header
.refcount_table_offset
);
1319 header
.refcount_table_clusters
=
1320 be32_to_cpu(header
.refcount_table_clusters
);
1321 header
.snapshots_offset
= be64_to_cpu(header
.snapshots_offset
);
1322 header
.nb_snapshots
= be32_to_cpu(header
.nb_snapshots
);
1324 if (header
.magic
!= QCOW_MAGIC
) {
1325 error_setg(errp
, "Image is not in qcow2 format");
1329 if (header
.version
< 2 || header
.version
> 3) {
1330 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1335 s
->qcow_version
= header
.version
;
1337 /* Initialise cluster size */
1338 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1339 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1340 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1341 header
.cluster_bits
);
1346 s
->cluster_bits
= header
.cluster_bits
;
1347 s
->cluster_size
= 1 << s
->cluster_bits
;
1349 /* Initialise version 3 header fields */
1350 if (header
.version
== 2) {
1351 header
.incompatible_features
= 0;
1352 header
.compatible_features
= 0;
1353 header
.autoclear_features
= 0;
1354 header
.refcount_order
= 4;
1355 header
.header_length
= 72;
1357 header
.incompatible_features
=
1358 be64_to_cpu(header
.incompatible_features
);
1359 header
.compatible_features
= be64_to_cpu(header
.compatible_features
);
1360 header
.autoclear_features
= be64_to_cpu(header
.autoclear_features
);
1361 header
.refcount_order
= be32_to_cpu(header
.refcount_order
);
1362 header
.header_length
= be32_to_cpu(header
.header_length
);
1364 if (header
.header_length
< 104) {
1365 error_setg(errp
, "qcow2 header too short");
1371 if (header
.header_length
> s
->cluster_size
) {
1372 error_setg(errp
, "qcow2 header exceeds cluster size");
1377 if (header
.header_length
> sizeof(header
)) {
1378 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1379 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1380 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
1381 s
->unknown_header_fields_size
);
1383 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1389 if (header
.backing_file_offset
> s
->cluster_size
) {
1390 error_setg(errp
, "Invalid backing file offset");
1395 if (header
.backing_file_offset
) {
1396 ext_end
= header
.backing_file_offset
;
1398 ext_end
= 1 << header
.cluster_bits
;
1401 /* Handle feature bits */
1402 s
->incompatible_features
= header
.incompatible_features
;
1403 s
->compatible_features
= header
.compatible_features
;
1404 s
->autoclear_features
= header
.autoclear_features
;
1407 * Handle compression type
1408 * Older qcow2 images don't contain the compression type header.
1409 * Distinguish them by the header length and use
1410 * the only valid (default) compression type in that case
1412 if (header
.header_length
> offsetof(QCowHeader
, compression_type
)) {
1413 s
->compression_type
= header
.compression_type
;
1415 s
->compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
1418 ret
= validate_compression_type(s
, errp
);
1423 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1424 void *feature_table
= NULL
;
1425 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1426 &feature_table
, flags
, NULL
, NULL
);
1427 report_unsupported_feature(errp
, feature_table
,
1428 s
->incompatible_features
&
1429 ~QCOW2_INCOMPAT_MASK
);
1431 g_free(feature_table
);
1435 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1436 /* Corrupt images may not be written to unless they are being repaired
1438 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1439 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1446 s
->subclusters_per_cluster
=
1447 has_subclusters(s
) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
: 1;
1448 s
->subcluster_size
= s
->cluster_size
/ s
->subclusters_per_cluster
;
1449 s
->subcluster_bits
= ctz32(s
->subcluster_size
);
1451 if (s
->subcluster_size
< (1 << MIN_CLUSTER_BITS
)) {
1452 error_setg(errp
, "Unsupported subcluster size: %d", s
->subcluster_size
);
1457 /* Check support for various header values */
1458 if (header
.refcount_order
> 6) {
1459 error_setg(errp
, "Reference count entry width too large; may not "
1464 s
->refcount_order
= header
.refcount_order
;
1465 s
->refcount_bits
= 1 << s
->refcount_order
;
1466 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1467 s
->refcount_max
+= s
->refcount_max
- 1;
1469 s
->crypt_method_header
= header
.crypt_method
;
1470 if (s
->crypt_method_header
) {
1471 if (bdrv_uses_whitelist() &&
1472 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1474 "Use of AES-CBC encrypted qcow2 images is no longer "
1475 "supported in system emulators");
1476 error_append_hint(errp
,
1477 "You can use 'qemu-img convert' to convert your "
1478 "image to an alternative supported format, such "
1479 "as unencrypted qcow2, or raw with the LUKS "
1480 "format instead.\n");
1485 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1486 s
->crypt_physical_offset
= false;
1488 /* Assuming LUKS and any future crypt methods we
1489 * add will all use physical offsets, due to the
1490 * fact that the alternative is insecure... */
1491 s
->crypt_physical_offset
= true;
1494 bs
->encrypted
= true;
1497 s
->l2_bits
= s
->cluster_bits
- ctz32(l2_entry_size(s
));
1498 s
->l2_size
= 1 << s
->l2_bits
;
1499 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1500 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1501 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1502 bs
->total_sectors
= header
.size
/ BDRV_SECTOR_SIZE
;
1503 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1504 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1505 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1507 s
->refcount_table_offset
= header
.refcount_table_offset
;
1508 s
->refcount_table_size
=
1509 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1511 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1512 error_setg(errp
, "Image does not contain a reference count table");
1517 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1518 header
.refcount_table_clusters
,
1519 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1520 "Reference count table", errp
);
1525 if (!(flags
& BDRV_O_CHECK
)) {
1527 * The total size in bytes of the snapshot table is checked in
1528 * qcow2_read_snapshots() because the size of each snapshot is
1529 * variable and we don't know it yet.
1530 * Here we only check the offset and number of snapshots.
1532 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1533 header
.nb_snapshots
,
1534 sizeof(QCowSnapshotHeader
),
1535 sizeof(QCowSnapshotHeader
) *
1537 "Snapshot table", errp
);
1543 /* read the level 1 table */
1544 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1545 header
.l1_size
, L1E_SIZE
,
1546 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1550 s
->l1_size
= header
.l1_size
;
1551 s
->l1_table_offset
= header
.l1_table_offset
;
1553 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1554 if (l1_vm_state_index
> INT_MAX
) {
1555 error_setg(errp
, "Image is too big");
1559 s
->l1_vm_state_index
= l1_vm_state_index
;
1561 /* the L1 table must contain at least enough entries to put
1562 header.size bytes */
1563 if (s
->l1_size
< s
->l1_vm_state_index
) {
1564 error_setg(errp
, "L1 table is too small");
1569 if (s
->l1_size
> 0) {
1570 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
, s
->l1_size
* L1E_SIZE
);
1571 if (s
->l1_table
== NULL
) {
1572 error_setg(errp
, "Could not allocate L1 table");
1576 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1577 s
->l1_size
* L1E_SIZE
);
1579 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1582 for(i
= 0;i
< s
->l1_size
; i
++) {
1583 s
->l1_table
[i
] = be64_to_cpu(s
->l1_table
[i
]);
1587 /* Parse driver-specific options */
1588 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1595 ret
= qcow2_refcount_init(bs
);
1597 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1601 QLIST_INIT(&s
->cluster_allocs
);
1602 QTAILQ_INIT(&s
->discards
);
1604 /* read qcow2 extensions */
1605 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1606 flags
, &update_header
, errp
)) {
1611 /* Open external data file */
1612 s
->data_file
= bdrv_open_child(NULL
, options
, "data-file", bs
,
1613 &child_of_bds
, BDRV_CHILD_DATA
,
1616 error_propagate(errp
, local_err
);
1621 if (s
->incompatible_features
& QCOW2_INCOMPAT_DATA_FILE
) {
1622 if (!s
->data_file
&& s
->image_data_file
) {
1623 s
->data_file
= bdrv_open_child(s
->image_data_file
, options
,
1624 "data-file", bs
, &child_of_bds
,
1625 BDRV_CHILD_DATA
, false, errp
);
1626 if (!s
->data_file
) {
1631 if (!s
->data_file
) {
1632 error_setg(errp
, "'data-file' is required for this image");
1638 bs
->file
->role
&= ~BDRV_CHILD_DATA
;
1640 /* Must succeed because we have given up permissions if anything */
1641 bdrv_child_refresh_perms(bs
, bs
->file
, &error_abort
);
1644 error_setg(errp
, "'data-file' can only be set for images with an "
1645 "external data file");
1650 s
->data_file
= bs
->file
;
1652 if (data_file_is_raw(bs
)) {
1653 error_setg(errp
, "data-file-raw requires a data file");
1659 /* qcow2_read_extension may have set up the crypto context
1660 * if the crypt method needs a header region, some methods
1661 * don't need header extensions, so must check here
1663 if (s
->crypt_method_header
&& !s
->crypto
) {
1664 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1665 unsigned int cflags
= 0;
1666 if (flags
& BDRV_O_NO_IO
) {
1667 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1669 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1671 QCOW2_MAX_THREADS
, errp
);
1676 } else if (!(flags
& BDRV_O_NO_IO
)) {
1677 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1678 s
->crypt_method_header
);
1684 /* read the backing file name */
1685 if (header
.backing_file_offset
!= 0) {
1686 len
= header
.backing_file_size
;
1687 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1688 len
>= sizeof(bs
->backing_file
)) {
1689 error_setg(errp
, "Backing file name too long");
1693 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
1694 bs
->auto_backing_file
, len
);
1696 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1699 bs
->auto_backing_file
[len
] = '\0';
1700 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
),
1701 bs
->auto_backing_file
);
1702 s
->image_backing_file
= g_strdup(bs
->auto_backing_file
);
1706 * Internal snapshots; skip reading them in check mode, because
1707 * we do not need them then, and we do not want to abort because
1708 * of a broken table.
1710 if (!(flags
& BDRV_O_CHECK
)) {
1711 s
->snapshots_offset
= header
.snapshots_offset
;
1712 s
->nb_snapshots
= header
.nb_snapshots
;
1714 ret
= qcow2_read_snapshots(bs
, errp
);
1720 /* Clear unknown autoclear feature bits */
1721 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1723 update_header
&& !bs
->read_only
&& !(flags
& BDRV_O_INACTIVE
);
1724 if (update_header
) {
1725 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1728 /* == Handle persistent dirty bitmaps ==
1730 * We want load dirty bitmaps in three cases:
1732 * 1. Normal open of the disk in active mode, not related to invalidation
1735 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1736 * bitmaps are _not_ migrating through migration channel, i.e.
1737 * 'dirty-bitmaps' capability is disabled.
1739 * 3. Invalidation of source vm after failed or canceled migration.
1740 * This is a very interesting case. There are two possible types of
1743 * A. Stored on inactivation and removed. They should be loaded from the
1746 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1747 * the migration channel (with dirty-bitmaps capability).
1749 * On the other hand, there are two possible sub-cases:
1751 * 3.1 disk was changed by somebody else while were inactive. In this
1752 * case all in-RAM dirty bitmaps (both persistent and not) are
1753 * definitely invalid. And we don't have any method to determine
1756 * Simple and safe thing is to just drop all the bitmaps of type B on
1757 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1759 * On the other hand, resuming source vm, if disk was already changed
1760 * is a bad thing anyway: not only bitmaps, the whole vm state is
1761 * out of sync with disk.
1763 * This means, that user or management tool, who for some reason
1764 * decided to resume source vm, after disk was already changed by
1765 * target vm, should at least drop all dirty bitmaps by hand.
1767 * So, we can ignore this case for now, but TODO: "generation"
1768 * extension for qcow2, to determine, that image was changed after
1769 * last inactivation. And if it is changed, we will drop (or at least
1770 * mark as 'invalid' all the bitmaps of type B, both persistent
1773 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1774 * to disk ('dirty-bitmaps' capability disabled), or not saved
1775 * ('dirty-bitmaps' capability enabled), but we don't need to care
1776 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1777 * and not stored has flag IN_USE=1 in the image and will be skipped
1780 * One remaining possible case when we don't want load bitmaps:
1782 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1783 * will be loaded on invalidation, no needs try loading them before)
1786 if (!(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
)) {
1787 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1788 bool header_updated
= qcow2_load_dirty_bitmaps(bs
, &local_err
);
1789 if (local_err
!= NULL
) {
1790 error_propagate(errp
, local_err
);
1795 update_header
= update_header
&& !header_updated
;
1798 if (update_header
) {
1799 ret
= qcow2_update_header(bs
);
1801 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1806 bs
->supported_zero_flags
= header
.version
>= 3 ?
1807 BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
: 0;
1808 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
1810 /* Repair image if dirty */
1811 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INACTIVE
)) && !bs
->read_only
&&
1812 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1813 BdrvCheckResult result
= {0};
1815 ret
= qcow2_co_check_locked(bs
, &result
,
1816 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1817 if (ret
< 0 || result
.check_errors
) {
1821 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1828 BdrvCheckResult result
= {0};
1829 qcow2_check_refcounts(bs
, &result
, 0);
1833 qemu_co_queue_init(&s
->thread_task_queue
);
1838 g_free(s
->image_data_file
);
1839 if (has_data_file(bs
)) {
1840 bdrv_unref_child(bs
, s
->data_file
);
1841 s
->data_file
= NULL
;
1843 g_free(s
->unknown_header_fields
);
1844 cleanup_unknown_header_ext(bs
);
1845 qcow2_free_snapshots(bs
);
1846 qcow2_refcount_close(bs
);
1847 qemu_vfree(s
->l1_table
);
1848 /* else pre-write overlap checks in cache_destroy may crash */
1850 cache_clean_timer_del(bs
);
1851 if (s
->l2_table_cache
) {
1852 qcow2_cache_destroy(s
->l2_table_cache
);
1854 if (s
->refcount_block_cache
) {
1855 qcow2_cache_destroy(s
->refcount_block_cache
);
1857 qcrypto_block_free(s
->crypto
);
1858 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1862 typedef struct QCow2OpenCo
{
1863 BlockDriverState
*bs
;
1870 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1872 QCow2OpenCo
*qoc
= opaque
;
1873 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1875 qemu_co_mutex_lock(&s
->lock
);
1876 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, qoc
->errp
);
1877 qemu_co_mutex_unlock(&s
->lock
);
1880 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1883 BDRVQcow2State
*s
= bs
->opaque
;
1892 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_of_bds
,
1893 BDRV_CHILD_IMAGE
, false, errp
);
1898 /* Initialise locks */
1899 qemu_co_mutex_init(&s
->lock
);
1901 if (qemu_in_coroutine()) {
1902 /* From bdrv_co_create. */
1903 qcow2_open_entry(&qoc
);
1905 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1906 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1907 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1912 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1914 BDRVQcow2State
*s
= bs
->opaque
;
1916 if (bs
->encrypted
) {
1917 /* Encryption works on a sector granularity */
1918 bs
->bl
.request_alignment
= qcrypto_block_get_sector_size(s
->crypto
);
1920 bs
->bl
.pwrite_zeroes_alignment
= s
->subcluster_size
;
1921 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1924 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1925 BlockReopenQueue
*queue
, Error
**errp
)
1927 Qcow2ReopenState
*r
;
1930 r
= g_new0(Qcow2ReopenState
, 1);
1933 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1934 state
->flags
, errp
);
1939 /* We need to write out any unwritten data if we reopen read-only. */
1940 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1941 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1946 ret
= bdrv_flush(state
->bs
);
1951 ret
= qcow2_mark_clean(state
->bs
);
1960 qcow2_update_options_abort(state
->bs
, r
);
1965 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1967 qcow2_update_options_commit(state
->bs
, state
->opaque
);
1968 g_free(state
->opaque
);
1971 static void qcow2_reopen_commit_post(BDRVReopenState
*state
)
1973 if (state
->flags
& BDRV_O_RDWR
) {
1974 Error
*local_err
= NULL
;
1976 if (qcow2_reopen_bitmaps_rw(state
->bs
, &local_err
) < 0) {
1978 * This is not fatal, bitmaps just left read-only, so all following
1979 * writes will fail. User can remove read-only bitmaps to unblock
1980 * writes or retry reopen.
1982 error_reportf_err(local_err
,
1983 "%s: Failed to make dirty bitmaps writable: ",
1984 bdrv_get_node_name(state
->bs
));
1989 static void qcow2_reopen_abort(BDRVReopenState
*state
)
1991 qcow2_update_options_abort(state
->bs
, state
->opaque
);
1992 g_free(state
->opaque
);
1995 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
1997 bool has_new_overlap_template
=
1998 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
1999 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2000 bool has_new_total_cache_size
=
2001 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
2002 bool has_all_cache_options
;
2004 /* New overlap template overrides all old overlap options */
2005 if (has_new_overlap_template
) {
2006 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
2007 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2008 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
2009 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
2010 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
2011 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
2012 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
2013 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
2014 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
2015 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
2018 /* New total cache size overrides all old options */
2019 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
2020 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
2021 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2024 qdict_join(options
, old_options
, false);
2027 * If after merging all cache size options are set, an old total size is
2028 * overwritten. Do keep all options, however, if all three are new. The
2029 * resulting error message is what we want to happen.
2031 has_all_cache_options
=
2032 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
2033 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
2034 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2036 if (has_all_cache_options
&& !has_new_total_cache_size
) {
2037 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
2041 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
2043 int64_t offset
, int64_t count
,
2044 int64_t *pnum
, int64_t *map
,
2045 BlockDriverState
**file
)
2047 BDRVQcow2State
*s
= bs
->opaque
;
2048 uint64_t host_offset
;
2050 QCow2SubclusterType type
;
2051 int ret
, status
= 0;
2053 qemu_co_mutex_lock(&s
->lock
);
2055 if (!s
->metadata_preallocation_checked
) {
2056 ret
= qcow2_detect_metadata_preallocation(bs
);
2057 s
->metadata_preallocation
= (ret
== 1);
2058 s
->metadata_preallocation_checked
= true;
2061 bytes
= MIN(INT_MAX
, count
);
2062 ret
= qcow2_get_host_offset(bs
, offset
, &bytes
, &host_offset
, &type
);
2063 qemu_co_mutex_unlock(&s
->lock
);
2070 if ((type
== QCOW2_SUBCLUSTER_NORMAL
||
2071 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2072 type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) && !s
->crypto
) {
2074 *file
= s
->data_file
->bs
;
2075 status
|= BDRV_BLOCK_OFFSET_VALID
;
2077 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2078 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
) {
2079 status
|= BDRV_BLOCK_ZERO
;
2080 } else if (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
2081 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) {
2082 status
|= BDRV_BLOCK_DATA
;
2084 if (s
->metadata_preallocation
&& (status
& BDRV_BLOCK_DATA
) &&
2085 (status
& BDRV_BLOCK_OFFSET_VALID
))
2087 status
|= BDRV_BLOCK_RECURSE
;
2092 static coroutine_fn
int qcow2_handle_l2meta(BlockDriverState
*bs
,
2093 QCowL2Meta
**pl2meta
,
2097 QCowL2Meta
*l2meta
= *pl2meta
;
2099 while (l2meta
!= NULL
) {
2103 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
2108 qcow2_alloc_cluster_abort(bs
, l2meta
);
2111 /* Take the request off the list of running requests */
2112 QLIST_REMOVE(l2meta
, next_in_flight
);
2114 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2116 next
= l2meta
->next
;
2125 static coroutine_fn
int
2126 qcow2_co_preadv_encrypted(BlockDriverState
*bs
,
2127 uint64_t host_offset
,
2131 uint64_t qiov_offset
)
2134 BDRVQcow2State
*s
= bs
->opaque
;
2137 assert(bs
->encrypted
&& s
->crypto
);
2138 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2141 * For encrypted images, read everything into a temporary
2142 * contiguous buffer on which the AES functions can work.
2143 * Also, decryption in a separate buffer is better as it
2144 * prevents the guest from learning information about the
2145 * encrypted nature of the virtual disk.
2148 buf
= qemu_try_blockalign(s
->data_file
->bs
, bytes
);
2153 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2154 ret
= bdrv_co_pread(s
->data_file
, host_offset
, bytes
, buf
, 0);
2159 if (qcow2_co_decrypt(bs
, host_offset
, offset
, buf
, bytes
) < 0)
2164 qemu_iovec_from_buf(qiov
, qiov_offset
, buf
, bytes
);
2172 typedef struct Qcow2AioTask
{
2175 BlockDriverState
*bs
;
2176 QCow2SubclusterType subcluster_type
; /* only for read */
2177 uint64_t host_offset
; /* or full descriptor in compressed clusters */
2181 uint64_t qiov_offset
;
2182 QCowL2Meta
*l2meta
; /* only for write */
2185 static coroutine_fn
int qcow2_co_preadv_task_entry(AioTask
*task
);
2186 static coroutine_fn
int qcow2_add_task(BlockDriverState
*bs
,
2189 QCow2SubclusterType subcluster_type
,
2190 uint64_t host_offset
,
2197 Qcow2AioTask local_task
;
2198 Qcow2AioTask
*task
= pool
? g_new(Qcow2AioTask
, 1) : &local_task
;
2200 *task
= (Qcow2AioTask
) {
2203 .subcluster_type
= subcluster_type
,
2205 .host_offset
= host_offset
,
2208 .qiov_offset
= qiov_offset
,
2212 trace_qcow2_add_task(qemu_coroutine_self(), bs
, pool
,
2213 func
== qcow2_co_preadv_task_entry
? "read" : "write",
2214 subcluster_type
, host_offset
, offset
, bytes
,
2218 return func(&task
->task
);
2221 aio_task_pool_start_task(pool
, &task
->task
);
2226 static coroutine_fn
int qcow2_co_preadv_task(BlockDriverState
*bs
,
2227 QCow2SubclusterType subc_type
,
2228 uint64_t host_offset
,
2229 uint64_t offset
, uint64_t bytes
,
2233 BDRVQcow2State
*s
= bs
->opaque
;
2235 switch (subc_type
) {
2236 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
2237 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
2238 /* Both zero types are handled in qcow2_co_preadv_part */
2239 g_assert_not_reached();
2241 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
2242 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
2243 assert(bs
->backing
); /* otherwise handled in qcow2_co_preadv_part */
2245 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
2246 return bdrv_co_preadv_part(bs
->backing
, offset
, bytes
,
2247 qiov
, qiov_offset
, 0);
2249 case QCOW2_SUBCLUSTER_COMPRESSED
:
2250 return qcow2_co_preadv_compressed(bs
, host_offset
,
2251 offset
, bytes
, qiov
, qiov_offset
);
2253 case QCOW2_SUBCLUSTER_NORMAL
:
2254 if (bs
->encrypted
) {
2255 return qcow2_co_preadv_encrypted(bs
, host_offset
,
2256 offset
, bytes
, qiov
, qiov_offset
);
2259 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2260 return bdrv_co_preadv_part(s
->data_file
, host_offset
,
2261 bytes
, qiov
, qiov_offset
, 0);
2264 g_assert_not_reached();
2267 g_assert_not_reached();
2270 static coroutine_fn
int qcow2_co_preadv_task_entry(AioTask
*task
)
2272 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2276 return qcow2_co_preadv_task(t
->bs
, t
->subcluster_type
,
2277 t
->host_offset
, t
->offset
, t
->bytes
,
2278 t
->qiov
, t
->qiov_offset
);
2281 static coroutine_fn
int qcow2_co_preadv_part(BlockDriverState
*bs
,
2282 uint64_t offset
, uint64_t bytes
,
2284 size_t qiov_offset
, int flags
)
2286 BDRVQcow2State
*s
= bs
->opaque
;
2288 unsigned int cur_bytes
; /* number of bytes in current iteration */
2289 uint64_t host_offset
= 0;
2290 QCow2SubclusterType type
;
2291 AioTaskPool
*aio
= NULL
;
2293 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2294 /* prepare next request */
2295 cur_bytes
= MIN(bytes
, INT_MAX
);
2297 cur_bytes
= MIN(cur_bytes
,
2298 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2301 qemu_co_mutex_lock(&s
->lock
);
2302 ret
= qcow2_get_host_offset(bs
, offset
, &cur_bytes
,
2303 &host_offset
, &type
);
2304 qemu_co_mutex_unlock(&s
->lock
);
2309 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2310 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2311 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&& !bs
->backing
) ||
2312 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&& !bs
->backing
))
2314 qemu_iovec_memset(qiov
, qiov_offset
, 0, cur_bytes
);
2316 if (!aio
&& cur_bytes
!= bytes
) {
2317 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2319 ret
= qcow2_add_task(bs
, aio
, qcow2_co_preadv_task_entry
, type
,
2320 host_offset
, offset
, cur_bytes
,
2321 qiov
, qiov_offset
, NULL
);
2328 offset
+= cur_bytes
;
2329 qiov_offset
+= cur_bytes
;
2334 aio_task_pool_wait_all(aio
);
2336 ret
= aio_task_pool_status(aio
);
2344 /* Check if it's possible to merge a write request with the writing of
2345 * the data from the COW regions */
2346 static bool merge_cow(uint64_t offset
, unsigned bytes
,
2347 QEMUIOVector
*qiov
, size_t qiov_offset
,
2352 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2353 /* If both COW regions are empty then there's nothing to merge */
2354 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
2358 /* If COW regions are handled already, skip this too */
2364 * The write request should start immediately after the first
2365 * COW region. This does not always happen because the area
2366 * touched by the request can be larger than the one defined
2367 * by @m (a single request can span an area consisting of a
2368 * mix of previously unallocated and allocated clusters, that
2369 * is why @l2meta is a list).
2371 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
2372 /* In this case the request starts before this region */
2373 assert(offset
< l2meta_cow_start(m
));
2374 assert(m
->cow_start
.nb_bytes
== 0);
2378 /* The write request should end immediately before the second
2379 * COW region (see above for why it does not always happen) */
2380 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
2381 assert(offset
+ bytes
> m
->offset
+ m
->cow_end
.offset
);
2382 assert(m
->cow_end
.nb_bytes
== 0);
2386 /* Make sure that adding both COW regions to the QEMUIOVector
2387 * does not exceed IOV_MAX */
2388 if (qemu_iovec_subvec_niov(qiov
, qiov_offset
, bytes
) > IOV_MAX
- 2) {
2392 m
->data_qiov
= qiov
;
2393 m
->data_qiov_offset
= qiov_offset
;
2401 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error.
2402 * Note that returning 0 does not guarantee non-zero data.
2404 static int is_zero_cow(BlockDriverState
*bs
, QCowL2Meta
*m
)
2407 * This check is designed for optimization shortcut so it must be
2409 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is
2410 * faster (but not as accurate and can result in false negatives).
2412 int ret
= bdrv_co_is_zero_fast(bs
, m
->offset
+ m
->cow_start
.offset
,
2413 m
->cow_start
.nb_bytes
);
2418 return bdrv_co_is_zero_fast(bs
, m
->offset
+ m
->cow_end
.offset
,
2419 m
->cow_end
.nb_bytes
);
2422 static int handle_alloc_space(BlockDriverState
*bs
, QCowL2Meta
*l2meta
)
2424 BDRVQcow2State
*s
= bs
->opaque
;
2427 if (!(s
->data_file
->bs
->supported_zero_flags
& BDRV_REQ_NO_FALLBACK
)) {
2431 if (bs
->encrypted
) {
2435 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2437 uint64_t start_offset
= m
->alloc_offset
+ m
->cow_start
.offset
;
2438 unsigned nb_bytes
= m
->cow_end
.offset
+ m
->cow_end
.nb_bytes
-
2439 m
->cow_start
.offset
;
2441 if (!m
->cow_start
.nb_bytes
&& !m
->cow_end
.nb_bytes
) {
2445 ret
= is_zero_cow(bs
, m
);
2448 } else if (ret
== 0) {
2453 * instead of writing zero COW buffers,
2454 * efficiently zero out the whole clusters
2457 ret
= qcow2_pre_write_overlap_check(bs
, 0, start_offset
, nb_bytes
,
2463 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC_SPACE
);
2464 ret
= bdrv_co_pwrite_zeroes(s
->data_file
, start_offset
, nb_bytes
,
2465 BDRV_REQ_NO_FALLBACK
);
2467 if (ret
!= -ENOTSUP
&& ret
!= -EAGAIN
) {
2473 trace_qcow2_skip_cow(qemu_coroutine_self(), m
->offset
, m
->nb_clusters
);
2480 * qcow2_co_pwritev_task
2481 * Called with s->lock unlocked
2482 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2483 * not use it somehow after qcow2_co_pwritev_task() call
2485 static coroutine_fn
int qcow2_co_pwritev_task(BlockDriverState
*bs
,
2486 uint64_t host_offset
,
2487 uint64_t offset
, uint64_t bytes
,
2489 uint64_t qiov_offset
,
2493 BDRVQcow2State
*s
= bs
->opaque
;
2494 void *crypt_buf
= NULL
;
2495 QEMUIOVector encrypted_qiov
;
2497 if (bs
->encrypted
) {
2499 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2500 crypt_buf
= qemu_try_blockalign(bs
->file
->bs
, bytes
);
2501 if (crypt_buf
== NULL
) {
2505 qemu_iovec_to_buf(qiov
, qiov_offset
, crypt_buf
, bytes
);
2507 if (qcow2_co_encrypt(bs
, host_offset
, offset
, crypt_buf
, bytes
) < 0) {
2512 qemu_iovec_init_buf(&encrypted_qiov
, crypt_buf
, bytes
);
2513 qiov
= &encrypted_qiov
;
2517 /* Try to efficiently initialize the physical space with zeroes */
2518 ret
= handle_alloc_space(bs
, l2meta
);
2524 * If we need to do COW, check if it's possible to merge the
2525 * writing of the guest data together with that of the COW regions.
2526 * If it's not possible (or not necessary) then write the
2529 if (!merge_cow(offset
, bytes
, qiov
, qiov_offset
, l2meta
)) {
2530 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2531 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset
);
2532 ret
= bdrv_co_pwritev_part(s
->data_file
, host_offset
,
2533 bytes
, qiov
, qiov_offset
, 0);
2539 qemu_co_mutex_lock(&s
->lock
);
2541 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
2545 qemu_co_mutex_lock(&s
->lock
);
2548 qcow2_handle_l2meta(bs
, &l2meta
, false);
2549 qemu_co_mutex_unlock(&s
->lock
);
2551 qemu_vfree(crypt_buf
);
2556 static coroutine_fn
int qcow2_co_pwritev_task_entry(AioTask
*task
)
2558 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2560 assert(!t
->subcluster_type
);
2562 return qcow2_co_pwritev_task(t
->bs
, t
->host_offset
,
2563 t
->offset
, t
->bytes
, t
->qiov
, t
->qiov_offset
,
2567 static coroutine_fn
int qcow2_co_pwritev_part(
2568 BlockDriverState
*bs
, uint64_t offset
, uint64_t bytes
,
2569 QEMUIOVector
*qiov
, size_t qiov_offset
, int flags
)
2571 BDRVQcow2State
*s
= bs
->opaque
;
2572 int offset_in_cluster
;
2574 unsigned int cur_bytes
; /* number of sectors in current iteration */
2575 uint64_t host_offset
;
2576 QCowL2Meta
*l2meta
= NULL
;
2577 AioTaskPool
*aio
= NULL
;
2579 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
2581 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2585 trace_qcow2_writev_start_part(qemu_coroutine_self());
2586 offset_in_cluster
= offset_into_cluster(s
, offset
);
2587 cur_bytes
= MIN(bytes
, INT_MAX
);
2588 if (bs
->encrypted
) {
2589 cur_bytes
= MIN(cur_bytes
,
2590 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
2591 - offset_in_cluster
);
2594 qemu_co_mutex_lock(&s
->lock
);
2596 ret
= qcow2_alloc_host_offset(bs
, offset
, &cur_bytes
,
2597 &host_offset
, &l2meta
);
2602 ret
= qcow2_pre_write_overlap_check(bs
, 0, host_offset
,
2608 qemu_co_mutex_unlock(&s
->lock
);
2610 if (!aio
&& cur_bytes
!= bytes
) {
2611 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2613 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_task_entry
, 0,
2614 host_offset
, offset
,
2615 cur_bytes
, qiov
, qiov_offset
, l2meta
);
2616 l2meta
= NULL
; /* l2meta is consumed by qcow2_co_pwritev_task() */
2622 offset
+= cur_bytes
;
2623 qiov_offset
+= cur_bytes
;
2624 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2628 qemu_co_mutex_lock(&s
->lock
);
2631 qcow2_handle_l2meta(bs
, &l2meta
, false);
2633 qemu_co_mutex_unlock(&s
->lock
);
2637 aio_task_pool_wait_all(aio
);
2639 ret
= aio_task_pool_status(aio
);
2644 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2649 static int qcow2_inactivate(BlockDriverState
*bs
)
2651 BDRVQcow2State
*s
= bs
->opaque
;
2652 int ret
, result
= 0;
2653 Error
*local_err
= NULL
;
2655 qcow2_store_persistent_dirty_bitmaps(bs
, true, &local_err
);
2656 if (local_err
!= NULL
) {
2658 error_reportf_err(local_err
, "Lost persistent bitmaps during "
2659 "inactivation of node '%s': ",
2660 bdrv_get_device_or_node_name(bs
));
2663 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2666 error_report("Failed to flush the L2 table cache: %s",
2670 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2673 error_report("Failed to flush the refcount block cache: %s",
2678 qcow2_mark_clean(bs
);
2684 static void qcow2_close(BlockDriverState
*bs
)
2686 BDRVQcow2State
*s
= bs
->opaque
;
2687 qemu_vfree(s
->l1_table
);
2688 /* else pre-write overlap checks in cache_destroy may crash */
2691 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2692 qcow2_inactivate(bs
);
2695 cache_clean_timer_del(bs
);
2696 qcow2_cache_destroy(s
->l2_table_cache
);
2697 qcow2_cache_destroy(s
->refcount_block_cache
);
2699 qcrypto_block_free(s
->crypto
);
2701 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
2703 g_free(s
->unknown_header_fields
);
2704 cleanup_unknown_header_ext(bs
);
2706 g_free(s
->image_data_file
);
2707 g_free(s
->image_backing_file
);
2708 g_free(s
->image_backing_format
);
2710 if (has_data_file(bs
)) {
2711 bdrv_unref_child(bs
, s
->data_file
);
2712 s
->data_file
= NULL
;
2715 qcow2_refcount_close(bs
);
2716 qcow2_free_snapshots(bs
);
2719 static void coroutine_fn
qcow2_co_invalidate_cache(BlockDriverState
*bs
,
2722 BDRVQcow2State
*s
= bs
->opaque
;
2723 int flags
= s
->flags
;
2724 QCryptoBlock
*crypto
= NULL
;
2726 Error
*local_err
= NULL
;
2730 * Backing files are read-only which makes all of their metadata immutable,
2731 * that means we don't have to worry about reopening them here.
2739 memset(s
, 0, sizeof(BDRVQcow2State
));
2740 options
= qdict_clone_shallow(bs
->options
);
2742 flags
&= ~BDRV_O_INACTIVE
;
2743 qemu_co_mutex_lock(&s
->lock
);
2744 ret
= qcow2_do_open(bs
, options
, flags
, &local_err
);
2745 qemu_co_mutex_unlock(&s
->lock
);
2746 qobject_unref(options
);
2748 error_propagate_prepend(errp
, local_err
,
2749 "Could not reopen qcow2 layer: ");
2752 } else if (ret
< 0) {
2753 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
2761 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2762 size_t len
, size_t buflen
)
2764 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2765 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2767 if (buflen
< ext_len
) {
2771 *ext_backing_fmt
= (QCowExtension
) {
2772 .magic
= cpu_to_be32(magic
),
2773 .len
= cpu_to_be32(len
),
2777 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2784 * Updates the qcow2 header, including the variable length parts of it, i.e.
2785 * the backing file name and all extensions. qcow2 was not designed to allow
2786 * such changes, so if we run out of space (we can only use the first cluster)
2787 * this function may fail.
2789 * Returns 0 on success, -errno in error cases.
2791 int qcow2_update_header(BlockDriverState
*bs
)
2793 BDRVQcow2State
*s
= bs
->opaque
;
2796 size_t buflen
= s
->cluster_size
;
2798 uint64_t total_size
;
2799 uint32_t refcount_table_clusters
;
2800 size_t header_length
;
2801 Qcow2UnknownHeaderExtension
*uext
;
2803 buf
= qemu_blockalign(bs
, buflen
);
2805 /* Header structure */
2806 header
= (QCowHeader
*) buf
;
2808 if (buflen
< sizeof(*header
)) {
2813 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2814 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2815 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2817 ret
= validate_compression_type(s
, NULL
);
2822 *header
= (QCowHeader
) {
2823 /* Version 2 fields */
2824 .magic
= cpu_to_be32(QCOW_MAGIC
),
2825 .version
= cpu_to_be32(s
->qcow_version
),
2826 .backing_file_offset
= 0,
2827 .backing_file_size
= 0,
2828 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2829 .size
= cpu_to_be64(total_size
),
2830 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2831 .l1_size
= cpu_to_be32(s
->l1_size
),
2832 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2833 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2834 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2835 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2836 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2838 /* Version 3 fields */
2839 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2840 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2841 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2842 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2843 .header_length
= cpu_to_be32(header_length
),
2844 .compression_type
= s
->compression_type
,
2847 /* For older versions, write a shorter header */
2848 switch (s
->qcow_version
) {
2850 ret
= offsetof(QCowHeader
, incompatible_features
);
2853 ret
= sizeof(*header
);
2862 memset(buf
, 0, buflen
);
2864 /* Preserve any unknown field in the header */
2865 if (s
->unknown_header_fields_size
) {
2866 if (buflen
< s
->unknown_header_fields_size
) {
2871 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2872 buf
+= s
->unknown_header_fields_size
;
2873 buflen
-= s
->unknown_header_fields_size
;
2876 /* Backing file format header extension */
2877 if (s
->image_backing_format
) {
2878 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2879 s
->image_backing_format
,
2880 strlen(s
->image_backing_format
),
2890 /* External data file header extension */
2891 if (has_data_file(bs
) && s
->image_data_file
) {
2892 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_DATA_FILE
,
2893 s
->image_data_file
, strlen(s
->image_data_file
),
2903 /* Full disk encryption header pointer extension */
2904 if (s
->crypto_header
.offset
!= 0) {
2905 s
->crypto_header
.offset
= cpu_to_be64(s
->crypto_header
.offset
);
2906 s
->crypto_header
.length
= cpu_to_be64(s
->crypto_header
.length
);
2907 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2908 &s
->crypto_header
, sizeof(s
->crypto_header
),
2910 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
2911 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
2920 * Feature table. A mere 8 feature names occupies 392 bytes, and
2921 * when coupled with the v3 minimum header of 104 bytes plus the
2922 * 8-byte end-of-extension marker, that would leave only 8 bytes
2923 * for a backing file name in an image with 512-byte clusters.
2924 * Thus, we choose to omit this header for cluster sizes 4k and
2927 if (s
->qcow_version
>= 3 && s
->cluster_size
> 4096) {
2928 static const Qcow2Feature features
[] = {
2930 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2931 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2932 .name
= "dirty bit",
2935 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2936 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
2937 .name
= "corrupt bit",
2940 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2941 .bit
= QCOW2_INCOMPAT_DATA_FILE_BITNR
,
2942 .name
= "external data file",
2945 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2946 .bit
= QCOW2_INCOMPAT_COMPRESSION_BITNR
,
2947 .name
= "compression type",
2950 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2951 .bit
= QCOW2_INCOMPAT_EXTL2_BITNR
,
2952 .name
= "extended L2 entries",
2955 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
2956 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
2957 .name
= "lazy refcounts",
2960 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
2961 .bit
= QCOW2_AUTOCLEAR_BITMAPS_BITNR
,
2965 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
2966 .bit
= QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR
,
2967 .name
= "raw external data",
2971 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
2972 features
, sizeof(features
), buflen
);
2980 /* Bitmap extension */
2981 if (s
->nb_bitmaps
> 0) {
2982 Qcow2BitmapHeaderExt bitmaps_header
= {
2983 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
2984 .bitmap_directory_size
=
2985 cpu_to_be64(s
->bitmap_directory_size
),
2986 .bitmap_directory_offset
=
2987 cpu_to_be64(s
->bitmap_directory_offset
)
2989 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
2990 &bitmaps_header
, sizeof(bitmaps_header
),
2999 /* Keep unknown header extensions */
3000 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
3001 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
3010 /* End of header extensions */
3011 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
3019 /* Backing file name */
3020 if (s
->image_backing_file
) {
3021 size_t backing_file_len
= strlen(s
->image_backing_file
);
3023 if (buflen
< backing_file_len
) {
3028 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3029 strncpy(buf
, s
->image_backing_file
, buflen
);
3031 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
3032 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
3035 /* Write the new header */
3036 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
3047 static int qcow2_change_backing_file(BlockDriverState
*bs
,
3048 const char *backing_file
, const char *backing_fmt
)
3050 BDRVQcow2State
*s
= bs
->opaque
;
3052 /* Adding a backing file means that the external data file alone won't be
3053 * enough to make sense of the content */
3054 if (backing_file
&& data_file_is_raw(bs
)) {
3058 if (backing_file
&& strlen(backing_file
) > 1023) {
3062 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3063 backing_file
?: "");
3064 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3065 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3067 g_free(s
->image_backing_file
);
3068 g_free(s
->image_backing_format
);
3070 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
3071 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
3073 return qcow2_update_header(bs
);
3076 static int qcow2_set_up_encryption(BlockDriverState
*bs
,
3077 QCryptoBlockCreateOptions
*cryptoopts
,
3080 BDRVQcow2State
*s
= bs
->opaque
;
3081 QCryptoBlock
*crypto
= NULL
;
3084 switch (cryptoopts
->format
) {
3085 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
3086 fmt
= QCOW_CRYPT_LUKS
;
3088 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
3089 fmt
= QCOW_CRYPT_AES
;
3092 error_setg(errp
, "Crypto format not supported in qcow2");
3096 s
->crypt_method_header
= fmt
;
3098 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
3099 qcow2_crypto_hdr_init_func
,
3100 qcow2_crypto_hdr_write_func
,
3106 ret
= qcow2_update_header(bs
);
3108 error_setg_errno(errp
, -ret
, "Could not write encryption header");
3114 qcrypto_block_free(crypto
);
3119 * Preallocates metadata structures for data clusters between @offset (in the
3120 * guest disk) and @new_length (which is thus generally the new guest disk
3123 * Returns: 0 on success, -errno on failure.
3125 static int coroutine_fn
preallocate_co(BlockDriverState
*bs
, uint64_t offset
,
3126 uint64_t new_length
, PreallocMode mode
,
3129 BDRVQcow2State
*s
= bs
->opaque
;
3131 uint64_t host_offset
= 0;
3132 int64_t file_length
;
3133 unsigned int cur_bytes
;
3135 QCowL2Meta
*meta
= NULL
, *m
;
3137 assert(offset
<= new_length
);
3138 bytes
= new_length
- offset
;
3141 cur_bytes
= MIN(bytes
, QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
));
3142 ret
= qcow2_alloc_host_offset(bs
, offset
, &cur_bytes
,
3143 &host_offset
, &meta
);
3145 error_setg_errno(errp
, -ret
, "Allocating clusters failed");
3149 for (m
= meta
; m
!= NULL
; m
= m
->next
) {
3153 ret
= qcow2_handle_l2meta(bs
, &meta
, true);
3155 error_setg_errno(errp
, -ret
, "Mapping clusters failed");
3159 /* TODO Preallocate data if requested */
3162 offset
+= cur_bytes
;
3166 * It is expected that the image file is large enough to actually contain
3167 * all of the allocated clusters (otherwise we get failing reads after
3168 * EOF). Extend the image to the last allocated sector.
3170 file_length
= bdrv_getlength(s
->data_file
->bs
);
3171 if (file_length
< 0) {
3172 error_setg_errno(errp
, -file_length
, "Could not get file size");
3177 if (host_offset
+ cur_bytes
> file_length
) {
3178 if (mode
== PREALLOC_MODE_METADATA
) {
3179 mode
= PREALLOC_MODE_OFF
;
3181 ret
= bdrv_co_truncate(s
->data_file
, host_offset
+ cur_bytes
, false,
3191 qcow2_handle_l2meta(bs
, &meta
, false);
3195 /* qcow2_refcount_metadata_size:
3196 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3197 * @cluster_size: size of a cluster, in bytes
3198 * @refcount_order: refcount bits power-of-2 exponent
3199 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3202 * Returns: Number of bytes required for refcount blocks and table metadata.
3204 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
3205 int refcount_order
, bool generous_increase
,
3206 uint64_t *refblock_count
)
3209 * Every host cluster is reference-counted, including metadata (even
3210 * refcount metadata is recursively included).
3212 * An accurate formula for the size of refcount metadata size is difficult
3213 * to derive. An easier method of calculation is finding the fixed point
3214 * where no further refcount blocks or table clusters are required to
3215 * reference count every cluster.
3217 int64_t blocks_per_table_cluster
= cluster_size
/ REFTABLE_ENTRY_SIZE
;
3218 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
3219 int64_t table
= 0; /* number of refcount table clusters */
3220 int64_t blocks
= 0; /* number of refcount block clusters */
3226 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
3227 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
3228 n
= clusters
+ blocks
+ table
;
3230 if (n
== last
&& generous_increase
) {
3231 clusters
+= DIV_ROUND_UP(table
, 2);
3232 n
= 0; /* force another loop */
3233 generous_increase
= false;
3235 } while (n
!= last
);
3237 if (refblock_count
) {
3238 *refblock_count
= blocks
;
3241 return (blocks
+ table
) * cluster_size
;
3245 * qcow2_calc_prealloc_size:
3246 * @total_size: virtual disk size in bytes
3247 * @cluster_size: cluster size in bytes
3248 * @refcount_order: refcount bits power-of-2 exponent
3249 * @extended_l2: true if the image has extended L2 entries
3251 * Returns: Total number of bytes required for the fully allocated image
3252 * (including metadata).
3254 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
3255 size_t cluster_size
,
3259 int64_t meta_size
= 0;
3260 uint64_t nl1e
, nl2e
;
3261 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
3262 size_t l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
3264 /* header: 1 cluster */
3265 meta_size
+= cluster_size
;
3267 /* total size of L2 tables */
3268 nl2e
= aligned_total_size
/ cluster_size
;
3269 nl2e
= ROUND_UP(nl2e
, cluster_size
/ l2e_size
);
3270 meta_size
+= nl2e
* l2e_size
;
3272 /* total size of L1 tables */
3273 nl1e
= nl2e
* l2e_size
/ cluster_size
;
3274 nl1e
= ROUND_UP(nl1e
, cluster_size
/ L1E_SIZE
);
3275 meta_size
+= nl1e
* L1E_SIZE
;
3277 /* total size of refcount table and blocks */
3278 meta_size
+= qcow2_refcount_metadata_size(
3279 (meta_size
+ aligned_total_size
) / cluster_size
,
3280 cluster_size
, refcount_order
, false, NULL
);
3282 return meta_size
+ aligned_total_size
;
3285 static bool validate_cluster_size(size_t cluster_size
, bool extended_l2
,
3288 int cluster_bits
= ctz32(cluster_size
);
3289 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
3290 (1 << cluster_bits
) != cluster_size
)
3292 error_setg(errp
, "Cluster size must be a power of two between %d and "
3293 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
3298 unsigned min_cluster_size
=
3299 (1 << MIN_CLUSTER_BITS
) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
;
3300 if (cluster_size
< min_cluster_size
) {
3301 error_setg(errp
, "Extended L2 entries are only supported with "
3302 "cluster sizes of at least %u bytes", min_cluster_size
);
3310 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, bool extended_l2
,
3313 size_t cluster_size
;
3315 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
3316 DEFAULT_CLUSTER_SIZE
);
3317 if (!validate_cluster_size(cluster_size
, extended_l2
, errp
)) {
3320 return cluster_size
;
3323 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
3328 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
3330 ret
= 3; /* default */
3331 } else if (!strcmp(buf
, "0.10")) {
3333 } else if (!strcmp(buf
, "1.1")) {
3336 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
3343 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
3346 uint64_t refcount_bits
;
3348 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
3349 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
3350 error_setg(errp
, "Refcount width must be a power of two and may not "
3355 if (version
< 3 && refcount_bits
!= 16) {
3356 error_setg(errp
, "Different refcount widths than 16 bits require "
3357 "compatibility level 1.1 or above (use compat=1.1 or "
3362 return refcount_bits
;
3365 static int coroutine_fn
3366 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
3368 BlockdevCreateOptionsQcow2
*qcow2_opts
;
3372 * Open the image file and write a minimal qcow2 header.
3374 * We keep things simple and start with a zero-sized image. We also
3375 * do without refcount blocks or a L1 table for now. We'll fix the
3376 * inconsistency later.
3378 * We do need a refcount table because growing the refcount table means
3379 * allocating two new refcount blocks - the second of which would be at
3380 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3381 * size for any qcow2 image.
3383 BlockBackend
*blk
= NULL
;
3384 BlockDriverState
*bs
= NULL
;
3385 BlockDriverState
*data_bs
= NULL
;
3387 size_t cluster_size
;
3390 uint64_t *refcount_table
;
3392 uint8_t compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
3394 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
3395 qcow2_opts
= &create_options
->u
.qcow2
;
3397 bs
= bdrv_open_blockdev_ref(qcow2_opts
->file
, errp
);
3402 /* Validate options and set default values */
3403 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
3404 error_setg(errp
, "Image size must be a multiple of %u bytes",
3405 (unsigned) BDRV_SECTOR_SIZE
);
3410 if (qcow2_opts
->has_version
) {
3411 switch (qcow2_opts
->version
) {
3412 case BLOCKDEV_QCOW2_VERSION_V2
:
3415 case BLOCKDEV_QCOW2_VERSION_V3
:
3419 g_assert_not_reached();
3425 if (qcow2_opts
->has_cluster_size
) {
3426 cluster_size
= qcow2_opts
->cluster_size
;
3428 cluster_size
= DEFAULT_CLUSTER_SIZE
;
3431 if (!qcow2_opts
->has_extended_l2
) {
3432 qcow2_opts
->extended_l2
= false;
3434 if (qcow2_opts
->extended_l2
) {
3436 error_setg(errp
, "Extended L2 entries are only supported with "
3437 "compatibility level 1.1 and above (use version=v3 or "
3444 if (!validate_cluster_size(cluster_size
, qcow2_opts
->extended_l2
, errp
)) {
3449 if (!qcow2_opts
->has_preallocation
) {
3450 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
3452 if (qcow2_opts
->has_backing_file
&&
3453 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
&&
3454 !qcow2_opts
->extended_l2
)
3456 error_setg(errp
, "Backing file and preallocation can only be used at "
3457 "the same time if extended_l2 is on");
3461 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->has_backing_file
) {
3462 error_setg(errp
, "Backing format cannot be used without backing file");
3467 if (!qcow2_opts
->has_lazy_refcounts
) {
3468 qcow2_opts
->lazy_refcounts
= false;
3470 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
3471 error_setg(errp
, "Lazy refcounts only supported with compatibility "
3472 "level 1.1 and above (use version=v3 or greater)");
3477 if (!qcow2_opts
->has_refcount_bits
) {
3478 qcow2_opts
->refcount_bits
= 16;
3480 if (qcow2_opts
->refcount_bits
> 64 ||
3481 !is_power_of_2(qcow2_opts
->refcount_bits
))
3483 error_setg(errp
, "Refcount width must be a power of two and may not "
3488 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
3489 error_setg(errp
, "Different refcount widths than 16 bits require "
3490 "compatibility level 1.1 or above (use version=v3 or "
3495 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
3497 if (qcow2_opts
->data_file_raw
&& !qcow2_opts
->data_file
) {
3498 error_setg(errp
, "data-file-raw requires data-file");
3502 if (qcow2_opts
->data_file_raw
&& qcow2_opts
->has_backing_file
) {
3503 error_setg(errp
, "Backing file and data-file-raw cannot be used at "
3509 if (qcow2_opts
->data_file
) {
3511 error_setg(errp
, "External data files are only supported with "
3512 "compatibility level 1.1 and above (use version=v3 or "
3517 data_bs
= bdrv_open_blockdev_ref(qcow2_opts
->data_file
, errp
);
3518 if (data_bs
== NULL
) {
3524 if (qcow2_opts
->has_compression_type
&&
3525 qcow2_opts
->compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3530 error_setg(errp
, "Non-zlib compression type is only supported with "
3531 "compatibility level 1.1 and above (use version=v3 or "
3536 switch (qcow2_opts
->compression_type
) {
3538 case QCOW2_COMPRESSION_TYPE_ZSTD
:
3542 error_setg(errp
, "Unknown compression type");
3546 compression_type
= qcow2_opts
->compression_type
;
3549 /* Create BlockBackend to write to the image */
3550 blk
= blk_new_with_bs(bs
, BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
,
3556 blk_set_allow_write_beyond_eof(blk
, true);
3558 /* Write the header */
3559 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
3560 header
= g_malloc0(cluster_size
);
3561 *header
= (QCowHeader
) {
3562 .magic
= cpu_to_be32(QCOW_MAGIC
),
3563 .version
= cpu_to_be32(version
),
3564 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
3565 .size
= cpu_to_be64(0),
3566 .l1_table_offset
= cpu_to_be64(0),
3567 .l1_size
= cpu_to_be32(0),
3568 .refcount_table_offset
= cpu_to_be64(cluster_size
),
3569 .refcount_table_clusters
= cpu_to_be32(1),
3570 .refcount_order
= cpu_to_be32(refcount_order
),
3571 /* don't deal with endianness since compression_type is 1 byte long */
3572 .compression_type
= compression_type
,
3573 .header_length
= cpu_to_be32(sizeof(*header
)),
3576 /* We'll update this to correct value later */
3577 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
3579 if (qcow2_opts
->lazy_refcounts
) {
3580 header
->compatible_features
|=
3581 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
3584 header
->incompatible_features
|=
3585 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE
);
3587 if (qcow2_opts
->data_file_raw
) {
3588 header
->autoclear_features
|=
3589 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW
);
3591 if (compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3592 header
->incompatible_features
|=
3593 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION
);
3596 if (qcow2_opts
->extended_l2
) {
3597 header
->incompatible_features
|=
3598 cpu_to_be64(QCOW2_INCOMPAT_EXTL2
);
3601 ret
= blk_pwrite(blk
, 0, header
, cluster_size
, 0);
3604 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
3608 /* Write a refcount table with one refcount block */
3609 refcount_table
= g_malloc0(2 * cluster_size
);
3610 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
3611 ret
= blk_pwrite(blk
, cluster_size
, refcount_table
, 2 * cluster_size
, 0);
3612 g_free(refcount_table
);
3615 error_setg_errno(errp
, -ret
, "Could not write refcount table");
3623 * And now open the image and make it consistent first (i.e. increase the
3624 * refcount of the cluster that is occupied by the header and the refcount
3627 options
= qdict_new();
3628 qdict_put_str(options
, "driver", "qcow2");
3629 qdict_put_str(options
, "file", bs
->node_name
);
3631 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3633 blk
= blk_new_open(NULL
, NULL
, options
,
3634 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
3641 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
3643 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
3644 "header and refcount table");
3647 } else if (ret
!= 0) {
3648 error_report("Huh, first cluster in empty image is already in use?");
3652 /* Set the external data file if necessary */
3654 BDRVQcow2State
*s
= blk_bs(blk
)->opaque
;
3655 s
->image_data_file
= g_strdup(data_bs
->filename
);
3658 /* Create a full header (including things like feature table) */
3659 ret
= qcow2_update_header(blk_bs(blk
));
3661 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
3665 /* Okay, now that we have a valid image, let's give it the right size */
3666 ret
= blk_truncate(blk
, qcow2_opts
->size
, false, qcow2_opts
->preallocation
,
3669 error_prepend(errp
, "Could not resize image: ");
3673 /* Want a backing file? There you go. */
3674 if (qcow2_opts
->has_backing_file
) {
3675 const char *backing_format
= NULL
;
3677 if (qcow2_opts
->has_backing_fmt
) {
3678 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
3681 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
3682 backing_format
, false);
3684 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
3685 "with format '%s'", qcow2_opts
->backing_file
,
3691 /* Want encryption? There you go. */
3692 if (qcow2_opts
->has_encrypt
) {
3693 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3702 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3703 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3704 * have to setup decryption context. We're not doing any I/O on the top
3705 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3708 options
= qdict_new();
3709 qdict_put_str(options
, "driver", "qcow2");
3710 qdict_put_str(options
, "file", bs
->node_name
);
3712 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3714 blk
= blk_new_open(NULL
, NULL
, options
,
3715 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3726 bdrv_unref(data_bs
);
3730 static int coroutine_fn
qcow2_co_create_opts(BlockDriver
*drv
,
3731 const char *filename
,
3735 BlockdevCreateOptions
*create_options
= NULL
;
3738 BlockDriverState
*bs
= NULL
;
3739 BlockDriverState
*data_bs
= NULL
;
3743 /* Only the keyval visitor supports the dotted syntax needed for
3744 * encryption, so go through a QDict before getting a QAPI type. Ignore
3745 * options meant for the protocol layer so that the visitor doesn't
3747 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3750 /* Handle encryption options */
3751 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3752 if (val
&& !strcmp(val
, "on")) {
3753 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3754 } else if (val
&& !strcmp(val
, "off")) {
3755 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3758 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3759 if (val
&& !strcmp(val
, "aes")) {
3760 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3763 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3764 * version=v2/v3 below. */
3765 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3766 if (val
&& !strcmp(val
, "0.10")) {
3767 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3768 } else if (val
&& !strcmp(val
, "1.1")) {
3769 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3772 /* Change legacy command line options into QMP ones */
3773 static const QDictRenames opt_renames
[] = {
3774 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3775 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3776 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3777 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3778 { BLOCK_OPT_EXTL2
, "extended-l2" },
3779 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3780 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3781 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3782 { BLOCK_OPT_DATA_FILE_RAW
, "data-file-raw" },
3783 { BLOCK_OPT_COMPRESSION_TYPE
, "compression-type" },
3787 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3792 /* Create and open the file (protocol layer) */
3793 ret
= bdrv_create_file(filename
, opts
, errp
);
3798 bs
= bdrv_open(filename
, NULL
, NULL
,
3799 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3805 /* Create and open an external data file (protocol layer) */
3806 val
= qdict_get_try_str(qdict
, BLOCK_OPT_DATA_FILE
);
3808 ret
= bdrv_create_file(val
, opts
, errp
);
3813 data_bs
= bdrv_open(val
, NULL
, NULL
,
3814 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
,
3816 if (data_bs
== NULL
) {
3821 qdict_del(qdict
, BLOCK_OPT_DATA_FILE
);
3822 qdict_put_str(qdict
, "data-file", data_bs
->node_name
);
3825 /* Set 'driver' and 'node' options */
3826 qdict_put_str(qdict
, "driver", "qcow2");
3827 qdict_put_str(qdict
, "file", bs
->node_name
);
3829 /* Now get the QAPI type BlockdevCreateOptions */
3830 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
3836 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, errp
);
3838 if (!create_options
) {
3843 /* Silently round up size */
3844 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3847 /* Create the qcow2 image (format layer) */
3848 ret
= qcow2_co_create(create_options
, errp
);
3851 bdrv_co_delete_file_noerr(bs
);
3852 bdrv_co_delete_file_noerr(data_bs
);
3857 qobject_unref(qdict
);
3859 bdrv_unref(data_bs
);
3860 qapi_free_BlockdevCreateOptions(create_options
);
3865 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3870 /* Clamp to image length, before checking status of underlying sectors */
3871 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3872 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3880 * bdrv_block_status_above doesn't merge different types of zeros, for
3881 * example, zeros which come from the region which is unallocated in
3882 * the whole backing chain, and zeros which come because of a short
3883 * backing file. So, we need a loop.
3886 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3889 } while (res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
&& bytes
);
3891 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && bytes
== 0;
3894 static coroutine_fn
int qcow2_co_pwrite_zeroes(BlockDriverState
*bs
,
3895 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3898 BDRVQcow2State
*s
= bs
->opaque
;
3900 uint32_t head
= offset_into_subcluster(s
, offset
);
3901 uint32_t tail
= ROUND_UP(offset
+ bytes
, s
->subcluster_size
) -
3904 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3905 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3912 QCow2SubclusterType type
;
3914 assert(head
+ bytes
+ tail
<= s
->subcluster_size
);
3916 /* check whether remainder of cluster already reads as zero */
3917 if (!(is_zero(bs
, offset
- head
, head
) &&
3918 is_zero(bs
, offset
+ bytes
, tail
))) {
3922 qemu_co_mutex_lock(&s
->lock
);
3923 /* We can have new write after previous check */
3925 bytes
= s
->subcluster_size
;
3926 nr
= s
->subcluster_size
;
3927 ret
= qcow2_get_host_offset(bs
, offset
, &nr
, &off
, &type
);
3929 (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
3930 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&&
3931 type
!= QCOW2_SUBCLUSTER_ZERO_PLAIN
&&
3932 type
!= QCOW2_SUBCLUSTER_ZERO_ALLOC
)) {
3933 qemu_co_mutex_unlock(&s
->lock
);
3934 return ret
< 0 ? ret
: -ENOTSUP
;
3937 qemu_co_mutex_lock(&s
->lock
);
3940 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
3942 /* Whatever is left can use real zero subclusters */
3943 ret
= qcow2_subcluster_zeroize(bs
, offset
, bytes
, flags
);
3944 qemu_co_mutex_unlock(&s
->lock
);
3949 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
3950 int64_t offset
, int bytes
)
3953 BDRVQcow2State
*s
= bs
->opaque
;
3955 /* If the image does not support QCOW_OFLAG_ZERO then discarding
3956 * clusters could expose stale data from the backing file. */
3957 if (s
->qcow_version
< 3 && bs
->backing
) {
3961 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
3962 assert(bytes
< s
->cluster_size
);
3963 /* Ignore partial clusters, except for the special case of the
3964 * complete partial cluster at the end of an unaligned file */
3965 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
3966 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3971 qemu_co_mutex_lock(&s
->lock
);
3972 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
3974 qemu_co_mutex_unlock(&s
->lock
);
3978 static int coroutine_fn
3979 qcow2_co_copy_range_from(BlockDriverState
*bs
,
3980 BdrvChild
*src
, uint64_t src_offset
,
3981 BdrvChild
*dst
, uint64_t dst_offset
,
3982 uint64_t bytes
, BdrvRequestFlags read_flags
,
3983 BdrvRequestFlags write_flags
)
3985 BDRVQcow2State
*s
= bs
->opaque
;
3987 unsigned int cur_bytes
; /* number of bytes in current iteration */
3988 BdrvChild
*child
= NULL
;
3989 BdrvRequestFlags cur_write_flags
;
3991 assert(!bs
->encrypted
);
3992 qemu_co_mutex_lock(&s
->lock
);
3994 while (bytes
!= 0) {
3995 uint64_t copy_offset
= 0;
3996 QCow2SubclusterType type
;
3997 /* prepare next request */
3998 cur_bytes
= MIN(bytes
, INT_MAX
);
3999 cur_write_flags
= write_flags
;
4001 ret
= qcow2_get_host_offset(bs
, src_offset
, &cur_bytes
,
4002 ©_offset
, &type
);
4008 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
4009 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
4010 if (bs
->backing
&& bs
->backing
->bs
) {
4011 int64_t backing_length
= bdrv_getlength(bs
->backing
->bs
);
4012 if (src_offset
>= backing_length
) {
4013 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4015 child
= bs
->backing
;
4016 cur_bytes
= MIN(cur_bytes
, backing_length
- src_offset
);
4017 copy_offset
= src_offset
;
4020 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4024 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
4025 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
4026 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4029 case QCOW2_SUBCLUSTER_COMPRESSED
:
4033 case QCOW2_SUBCLUSTER_NORMAL
:
4034 child
= s
->data_file
;
4040 qemu_co_mutex_unlock(&s
->lock
);
4041 ret
= bdrv_co_copy_range_from(child
,
4044 cur_bytes
, read_flags
, cur_write_flags
);
4045 qemu_co_mutex_lock(&s
->lock
);
4051 src_offset
+= cur_bytes
;
4052 dst_offset
+= cur_bytes
;
4057 qemu_co_mutex_unlock(&s
->lock
);
4061 static int coroutine_fn
4062 qcow2_co_copy_range_to(BlockDriverState
*bs
,
4063 BdrvChild
*src
, uint64_t src_offset
,
4064 BdrvChild
*dst
, uint64_t dst_offset
,
4065 uint64_t bytes
, BdrvRequestFlags read_flags
,
4066 BdrvRequestFlags write_flags
)
4068 BDRVQcow2State
*s
= bs
->opaque
;
4070 unsigned int cur_bytes
; /* number of sectors in current iteration */
4071 uint64_t host_offset
;
4072 QCowL2Meta
*l2meta
= NULL
;
4074 assert(!bs
->encrypted
);
4076 qemu_co_mutex_lock(&s
->lock
);
4078 while (bytes
!= 0) {
4082 cur_bytes
= MIN(bytes
, INT_MAX
);
4085 * If src->bs == dst->bs, we could simply copy by incrementing
4086 * the refcnt, without copying user data.
4087 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4088 ret
= qcow2_alloc_host_offset(bs
, dst_offset
, &cur_bytes
,
4089 &host_offset
, &l2meta
);
4094 ret
= qcow2_pre_write_overlap_check(bs
, 0, host_offset
, cur_bytes
,
4100 qemu_co_mutex_unlock(&s
->lock
);
4101 ret
= bdrv_co_copy_range_to(src
, src_offset
, s
->data_file
, host_offset
,
4102 cur_bytes
, read_flags
, write_flags
);
4103 qemu_co_mutex_lock(&s
->lock
);
4108 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
4114 src_offset
+= cur_bytes
;
4115 dst_offset
+= cur_bytes
;
4120 qcow2_handle_l2meta(bs
, &l2meta
, false);
4122 qemu_co_mutex_unlock(&s
->lock
);
4124 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
4129 static int coroutine_fn
qcow2_co_truncate(BlockDriverState
*bs
, int64_t offset
,
4130 bool exact
, PreallocMode prealloc
,
4131 BdrvRequestFlags flags
, Error
**errp
)
4133 BDRVQcow2State
*s
= bs
->opaque
;
4134 uint64_t old_length
;
4135 int64_t new_l1_size
;
4139 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
4140 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
4142 error_setg(errp
, "Unsupported preallocation mode '%s'",
4143 PreallocMode_str(prealloc
));
4147 if (!QEMU_IS_ALIGNED(offset
, BDRV_SECTOR_SIZE
)) {
4148 error_setg(errp
, "The new size must be a multiple of %u",
4149 (unsigned) BDRV_SECTOR_SIZE
);
4153 qemu_co_mutex_lock(&s
->lock
);
4156 * Even though we store snapshot size for all images, it was not
4157 * required until v3, so it is not safe to proceed for v2.
4159 if (s
->nb_snapshots
&& s
->qcow_version
< 3) {
4160 error_setg(errp
, "Can't resize a v2 image which has snapshots");
4165 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4166 if (qcow2_truncate_bitmaps_check(bs
, errp
)) {
4171 old_length
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4172 new_l1_size
= size_to_l1(s
, offset
);
4174 if (offset
< old_length
) {
4175 int64_t last_cluster
, old_file_size
;
4176 if (prealloc
!= PREALLOC_MODE_OFF
) {
4178 "Preallocation can't be used for shrinking an image");
4183 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
4184 old_length
- ROUND_UP(offset
,
4186 QCOW2_DISCARD_ALWAYS
, true);
4188 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
4192 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
4194 error_setg_errno(errp
, -ret
,
4195 "Failed to reduce the number of L2 tables");
4199 ret
= qcow2_shrink_reftable(bs
);
4201 error_setg_errno(errp
, -ret
,
4202 "Failed to discard unused refblocks");
4206 old_file_size
= bdrv_getlength(bs
->file
->bs
);
4207 if (old_file_size
< 0) {
4208 error_setg_errno(errp
, -old_file_size
,
4209 "Failed to inquire current file length");
4210 ret
= old_file_size
;
4213 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4214 if (last_cluster
< 0) {
4215 error_setg_errno(errp
, -last_cluster
,
4216 "Failed to find the last cluster");
4220 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
4221 Error
*local_err
= NULL
;
4224 * Do not pass @exact here: It will not help the user if
4225 * we get an error here just because they wanted to shrink
4226 * their qcow2 image (on a block device) with qemu-img.
4227 * (And on the qcow2 layer, the @exact requirement is
4228 * always fulfilled, so there is no need to pass it on.)
4230 bdrv_co_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
4231 false, PREALLOC_MODE_OFF
, 0, &local_err
);
4233 warn_reportf_err(local_err
,
4234 "Failed to truncate the tail of the image: ");
4238 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
4240 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
4246 case PREALLOC_MODE_OFF
:
4247 if (has_data_file(bs
)) {
4249 * If the caller wants an exact resize, the external data
4250 * file should be resized to the exact target size, too,
4251 * so we pass @exact here.
4253 ret
= bdrv_co_truncate(s
->data_file
, offset
, exact
, prealloc
, 0,
4261 case PREALLOC_MODE_METADATA
:
4262 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4268 case PREALLOC_MODE_FALLOC
:
4269 case PREALLOC_MODE_FULL
:
4271 int64_t allocation_start
, host_offset
, guest_offset
;
4272 int64_t clusters_allocated
;
4273 int64_t old_file_size
, last_cluster
, new_file_size
;
4274 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
4275 bool subclusters_need_allocation
= false;
4277 /* With a data file, preallocation means just allocating the metadata
4278 * and forwarding the truncate request to the data file */
4279 if (has_data_file(bs
)) {
4280 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4287 old_file_size
= bdrv_getlength(bs
->file
->bs
);
4288 if (old_file_size
< 0) {
4289 error_setg_errno(errp
, -old_file_size
,
4290 "Failed to inquire current file length");
4291 ret
= old_file_size
;
4295 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4296 if (last_cluster
>= 0) {
4297 old_file_size
= (last_cluster
+ 1) * s
->cluster_size
;
4299 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
4302 nb_new_data_clusters
= (ROUND_UP(offset
, s
->cluster_size
) -
4303 start_of_cluster(s
, old_length
)) >> s
->cluster_bits
;
4305 /* This is an overestimation; we will not actually allocate space for
4306 * these in the file but just make sure the new refcount structures are
4307 * able to cover them so we will not have to allocate new refblocks
4308 * while entering the data blocks in the potentially new L2 tables.
4309 * (We do not actually care where the L2 tables are placed. Maybe they
4310 * are already allocated or they can be placed somewhere before
4311 * @old_file_size. It does not matter because they will be fully
4312 * allocated automatically, so they do not need to be covered by the
4313 * preallocation. All that matters is that we will not have to allocate
4314 * new refcount structures for them.) */
4315 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
4316 s
->cluster_size
/ l2_entry_size(s
));
4317 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4318 * table for a potential head/tail */
4321 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
4322 nb_new_data_clusters
+
4325 if (allocation_start
< 0) {
4326 error_setg_errno(errp
, -allocation_start
,
4327 "Failed to resize refcount structures");
4328 ret
= allocation_start
;
4332 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
4333 nb_new_data_clusters
);
4334 if (clusters_allocated
< 0) {
4335 error_setg_errno(errp
, -clusters_allocated
,
4336 "Failed to allocate data clusters");
4337 ret
= clusters_allocated
;
4341 assert(clusters_allocated
== nb_new_data_clusters
);
4343 /* Allocate the data area */
4344 new_file_size
= allocation_start
+
4345 nb_new_data_clusters
* s
->cluster_size
;
4347 * Image file grows, so @exact does not matter.
4349 * If we need to zero out the new area, try first whether the protocol
4350 * driver can already take care of this.
4352 if (flags
& BDRV_REQ_ZERO_WRITE
) {
4353 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
,
4354 BDRV_REQ_ZERO_WRITE
, NULL
);
4356 flags
&= ~BDRV_REQ_ZERO_WRITE
;
4357 /* Ensure that we read zeroes and not backing file data */
4358 subclusters_need_allocation
= true;
4364 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
, 0,
4368 error_prepend(errp
, "Failed to resize underlying file: ");
4369 qcow2_free_clusters(bs
, allocation_start
,
4370 nb_new_data_clusters
* s
->cluster_size
,
4371 QCOW2_DISCARD_OTHER
);
4375 /* Create the necessary L2 entries */
4376 host_offset
= allocation_start
;
4377 guest_offset
= old_length
;
4378 while (nb_new_data_clusters
) {
4379 int64_t nb_clusters
= MIN(
4380 nb_new_data_clusters
,
4381 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
4382 unsigned cow_start_length
= offset_into_cluster(s
, guest_offset
);
4383 QCowL2Meta allocation
;
4384 guest_offset
= start_of_cluster(s
, guest_offset
);
4385 allocation
= (QCowL2Meta
) {
4386 .offset
= guest_offset
,
4387 .alloc_offset
= host_offset
,
4388 .nb_clusters
= nb_clusters
,
4391 .nb_bytes
= cow_start_length
,
4394 .offset
= nb_clusters
<< s
->cluster_bits
,
4397 .prealloc
= !subclusters_need_allocation
,
4399 qemu_co_queue_init(&allocation
.dependent_requests
);
4401 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
4403 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
4404 qcow2_free_clusters(bs
, host_offset
,
4405 nb_new_data_clusters
* s
->cluster_size
,
4406 QCOW2_DISCARD_OTHER
);
4410 guest_offset
+= nb_clusters
* s
->cluster_size
;
4411 host_offset
+= nb_clusters
* s
->cluster_size
;
4412 nb_new_data_clusters
-= nb_clusters
;
4418 g_assert_not_reached();
4421 if ((flags
& BDRV_REQ_ZERO_WRITE
) && offset
> old_length
) {
4422 uint64_t zero_start
= QEMU_ALIGN_UP(old_length
, s
->subcluster_size
);
4425 * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4426 * requires a subcluster-aligned start. The end may be unaligned if
4427 * it is at the end of the image (which it is here).
4429 if (offset
> zero_start
) {
4430 ret
= qcow2_subcluster_zeroize(bs
, zero_start
, offset
- zero_start
,
4433 error_setg_errno(errp
, -ret
, "Failed to zero out new clusters");
4438 /* Write explicit zeros for the unaligned head */
4439 if (zero_start
> old_length
) {
4440 uint64_t len
= MIN(zero_start
, offset
) - old_length
;
4441 uint8_t *buf
= qemu_blockalign0(bs
, len
);
4443 qemu_iovec_init_buf(&qiov
, buf
, len
);
4445 qemu_co_mutex_unlock(&s
->lock
);
4446 ret
= qcow2_co_pwritev_part(bs
, old_length
, len
, &qiov
, 0, 0);
4447 qemu_co_mutex_lock(&s
->lock
);
4451 error_setg_errno(errp
, -ret
, "Failed to zero out the new area");
4457 if (prealloc
!= PREALLOC_MODE_OFF
) {
4458 /* Flush metadata before actually changing the image size */
4459 ret
= qcow2_write_caches(bs
);
4461 error_setg_errno(errp
, -ret
,
4462 "Failed to flush the preallocated area to disk");
4467 bs
->total_sectors
= offset
/ BDRV_SECTOR_SIZE
;
4469 /* write updated header.size */
4470 offset
= cpu_to_be64(offset
);
4471 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
4472 &offset
, sizeof(offset
));
4474 error_setg_errno(errp
, -ret
, "Failed to update the image size");
4478 s
->l1_vm_state_index
= new_l1_size
;
4480 /* Update cache sizes */
4481 options
= qdict_clone_shallow(bs
->options
);
4482 ret
= qcow2_update_options(bs
, options
, s
->flags
, errp
);
4483 qobject_unref(options
);
4489 qemu_co_mutex_unlock(&s
->lock
);
4493 static coroutine_fn
int
4494 qcow2_co_pwritev_compressed_task(BlockDriverState
*bs
,
4495 uint64_t offset
, uint64_t bytes
,
4496 QEMUIOVector
*qiov
, size_t qiov_offset
)
4498 BDRVQcow2State
*s
= bs
->opaque
;
4501 uint8_t *buf
, *out_buf
;
4502 uint64_t cluster_offset
;
4504 assert(bytes
== s
->cluster_size
|| (bytes
< s
->cluster_size
&&
4505 (offset
+ bytes
== bs
->total_sectors
<< BDRV_SECTOR_BITS
)));
4507 buf
= qemu_blockalign(bs
, s
->cluster_size
);
4508 if (bytes
< s
->cluster_size
) {
4509 /* Zero-pad last write if image size is not cluster aligned */
4510 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
4512 qemu_iovec_to_buf(qiov
, qiov_offset
, buf
, bytes
);
4514 out_buf
= g_malloc(s
->cluster_size
);
4516 out_len
= qcow2_co_compress(bs
, out_buf
, s
->cluster_size
- 1,
4517 buf
, s
->cluster_size
);
4518 if (out_len
== -ENOMEM
) {
4519 /* could not compress: write normal cluster */
4520 ret
= qcow2_co_pwritev_part(bs
, offset
, bytes
, qiov
, qiov_offset
, 0);
4525 } else if (out_len
< 0) {
4530 qemu_co_mutex_lock(&s
->lock
);
4531 ret
= qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
,
4534 qemu_co_mutex_unlock(&s
->lock
);
4538 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
, true);
4539 qemu_co_mutex_unlock(&s
->lock
);
4544 BLKDBG_EVENT(s
->data_file
, BLKDBG_WRITE_COMPRESSED
);
4545 ret
= bdrv_co_pwrite(s
->data_file
, cluster_offset
, out_len
, out_buf
, 0);
4557 static coroutine_fn
int qcow2_co_pwritev_compressed_task_entry(AioTask
*task
)
4559 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
4561 assert(!t
->subcluster_type
&& !t
->l2meta
);
4563 return qcow2_co_pwritev_compressed_task(t
->bs
, t
->offset
, t
->bytes
, t
->qiov
,
4568 * XXX: put compressed sectors first, then all the cluster aligned
4569 * tables to avoid losing bytes in alignment
4571 static coroutine_fn
int
4572 qcow2_co_pwritev_compressed_part(BlockDriverState
*bs
,
4573 uint64_t offset
, uint64_t bytes
,
4574 QEMUIOVector
*qiov
, size_t qiov_offset
)
4576 BDRVQcow2State
*s
= bs
->opaque
;
4577 AioTaskPool
*aio
= NULL
;
4580 if (has_data_file(bs
)) {
4586 * align end of file to a sector boundary to ease reading with
4589 int64_t len
= bdrv_getlength(bs
->file
->bs
);
4593 return bdrv_co_truncate(bs
->file
, len
, false, PREALLOC_MODE_OFF
, 0,
4597 if (offset_into_cluster(s
, offset
)) {
4601 if (offset_into_cluster(s
, bytes
) &&
4602 (offset
+ bytes
) != (bs
->total_sectors
<< BDRV_SECTOR_BITS
)) {
4606 while (bytes
&& aio_task_pool_status(aio
) == 0) {
4607 uint64_t chunk_size
= MIN(bytes
, s
->cluster_size
);
4609 if (!aio
&& chunk_size
!= bytes
) {
4610 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
4613 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_compressed_task_entry
,
4614 0, 0, offset
, chunk_size
, qiov
, qiov_offset
, NULL
);
4618 qiov_offset
+= chunk_size
;
4619 offset
+= chunk_size
;
4620 bytes
-= chunk_size
;
4624 aio_task_pool_wait_all(aio
);
4626 ret
= aio_task_pool_status(aio
);
4634 static int coroutine_fn
4635 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
4636 uint64_t cluster_descriptor
,
4642 BDRVQcow2State
*s
= bs
->opaque
;
4643 int ret
= 0, csize
, nb_csectors
;
4645 uint8_t *buf
, *out_buf
;
4646 int offset_in_cluster
= offset_into_cluster(s
, offset
);
4648 coffset
= cluster_descriptor
& s
->cluster_offset_mask
;
4649 nb_csectors
= ((cluster_descriptor
>> s
->csize_shift
) & s
->csize_mask
) + 1;
4650 csize
= nb_csectors
* QCOW2_COMPRESSED_SECTOR_SIZE
-
4651 (coffset
& ~QCOW2_COMPRESSED_SECTOR_MASK
);
4653 buf
= g_try_malloc(csize
);
4658 out_buf
= qemu_blockalign(bs
, s
->cluster_size
);
4660 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
4661 ret
= bdrv_co_pread(bs
->file
, coffset
, csize
, buf
, 0);
4666 if (qcow2_co_decompress(bs
, out_buf
, s
->cluster_size
, buf
, csize
) < 0) {
4671 qemu_iovec_from_buf(qiov
, qiov_offset
, out_buf
+ offset_in_cluster
, bytes
);
4674 qemu_vfree(out_buf
);
4680 static int make_completely_empty(BlockDriverState
*bs
)
4682 BDRVQcow2State
*s
= bs
->opaque
;
4683 Error
*local_err
= NULL
;
4684 int ret
, l1_clusters
;
4686 uint64_t *new_reftable
= NULL
;
4687 uint64_t rt_entry
, l1_size2
;
4690 uint64_t reftable_offset
;
4691 uint32_t reftable_clusters
;
4692 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
4694 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
4699 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
4704 /* Refcounts will be broken utterly */
4705 ret
= qcow2_mark_dirty(bs
);
4710 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4712 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ L1E_SIZE
);
4713 l1_size2
= (uint64_t)s
->l1_size
* L1E_SIZE
;
4715 /* After this call, neither the in-memory nor the on-disk refcount
4716 * information accurately describe the actual references */
4718 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
4719 l1_clusters
* s
->cluster_size
, 0);
4721 goto fail_broken_refcounts
;
4723 memset(s
->l1_table
, 0, l1_size2
);
4725 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
4727 /* Overwrite enough clusters at the beginning of the sectors to place
4728 * the refcount table, a refcount block and the L1 table in; this may
4729 * overwrite parts of the existing refcount and L1 table, which is not
4730 * an issue because the dirty flag is set, complete data loss is in fact
4731 * desired and partial data loss is consequently fine as well */
4732 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
4733 (2 + l1_clusters
) * s
->cluster_size
, 0);
4734 /* This call (even if it failed overall) may have overwritten on-disk
4735 * refcount structures; in that case, the in-memory refcount information
4736 * will probably differ from the on-disk information which makes the BDS
4739 goto fail_broken_refcounts
;
4742 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4743 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
4745 /* "Create" an empty reftable (one cluster) directly after the image
4746 * header and an empty L1 table three clusters after the image header;
4747 * the cluster between those two will be used as the first refblock */
4748 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
4749 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
4750 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
4751 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
4752 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
4754 goto fail_broken_refcounts
;
4757 s
->l1_table_offset
= 3 * s
->cluster_size
;
4759 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ REFTABLE_ENTRY_SIZE
);
4760 if (!new_reftable
) {
4762 goto fail_broken_refcounts
;
4765 s
->refcount_table_offset
= s
->cluster_size
;
4766 s
->refcount_table_size
= s
->cluster_size
/ REFTABLE_ENTRY_SIZE
;
4767 s
->max_refcount_table_index
= 0;
4769 g_free(s
->refcount_table
);
4770 s
->refcount_table
= new_reftable
;
4771 new_reftable
= NULL
;
4773 /* Now the in-memory refcount information again corresponds to the on-disk
4774 * information (reftable is empty and no refblocks (the refblock cache is
4775 * empty)); however, this means some clusters (e.g. the image header) are
4776 * referenced, but not refcounted, but the normal qcow2 code assumes that
4777 * the in-memory information is always correct */
4779 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
4781 /* Enter the first refblock into the reftable */
4782 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
4783 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
4784 &rt_entry
, sizeof(rt_entry
));
4786 goto fail_broken_refcounts
;
4788 s
->refcount_table
[0] = 2 * s
->cluster_size
;
4790 s
->free_cluster_index
= 0;
4791 assert(3 + l1_clusters
<= s
->refcount_block_size
);
4792 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
4795 goto fail_broken_refcounts
;
4796 } else if (offset
> 0) {
4797 error_report("First cluster in emptied image is in use");
4801 /* Now finally the in-memory information corresponds to the on-disk
4802 * structures and is correct */
4803 ret
= qcow2_mark_clean(bs
);
4808 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
, false,
4809 PREALLOC_MODE_OFF
, 0, &local_err
);
4811 error_report_err(local_err
);
4817 fail_broken_refcounts
:
4818 /* The BDS is unusable at this point. If we wanted to make it usable, we
4819 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4820 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4821 * again. However, because the functions which could have caused this error
4822 * path to be taken are used by those functions as well, it's very likely
4823 * that that sequence will fail as well. Therefore, just eject the BDS. */
4827 g_free(new_reftable
);
4831 static int qcow2_make_empty(BlockDriverState
*bs
)
4833 BDRVQcow2State
*s
= bs
->opaque
;
4834 uint64_t offset
, end_offset
;
4835 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
4836 int l1_clusters
, ret
= 0;
4838 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ L1E_SIZE
);
4840 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
4841 3 + l1_clusters
<= s
->refcount_block_size
&&
4842 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
&&
4843 !has_data_file(bs
)) {
4844 /* The following function only works for qcow2 v3 images (it
4845 * requires the dirty flag) and only as long as there are no
4846 * features that reserve extra clusters (such as snapshots,
4847 * LUKS header, or persistent bitmaps), because it completely
4848 * empties the image. Furthermore, the L1 table and three
4849 * additional clusters (image header, refcount table, one
4850 * refcount block) have to fit inside one refcount block. It
4851 * only resets the image file, i.e. does not work with an
4852 * external data file. */
4853 return make_completely_empty(bs
);
4856 /* This fallback code simply discards every active cluster; this is slow,
4857 * but works in all cases */
4858 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4859 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
4860 /* As this function is generally used after committing an external
4861 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4862 * default action for this kind of discard is to pass the discard,
4863 * which will ideally result in an actually smaller image file, as
4864 * is probably desired. */
4865 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
4866 QCOW2_DISCARD_SNAPSHOT
, true);
4875 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
4877 BDRVQcow2State
*s
= bs
->opaque
;
4880 qemu_co_mutex_lock(&s
->lock
);
4881 ret
= qcow2_write_caches(bs
);
4882 qemu_co_mutex_unlock(&s
->lock
);
4887 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
4890 Error
*local_err
= NULL
;
4891 BlockMeasureInfo
*info
;
4892 uint64_t required
= 0; /* bytes that contribute to required size */
4893 uint64_t virtual_size
; /* disk size as seen by guest */
4894 uint64_t refcount_bits
;
4896 uint64_t luks_payload_size
= 0;
4897 size_t cluster_size
;
4900 PreallocMode prealloc
;
4901 bool has_backing_file
;
4906 /* Parse image creation options */
4907 extended_l2
= qemu_opt_get_bool_del(opts
, BLOCK_OPT_EXTL2
, false);
4909 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, extended_l2
,
4915 version
= qcow2_opt_get_version_del(opts
, &local_err
);
4920 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
4925 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
4926 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
4927 PREALLOC_MODE_OFF
, &local_err
);
4933 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
4934 has_backing_file
= !!optstr
;
4937 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
4938 has_luks
= optstr
&& strcmp(optstr
, "luks") == 0;
4942 g_autoptr(QCryptoBlockCreateOptions
) create_opts
= NULL
;
4943 QDict
*cryptoopts
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
4946 create_opts
= block_crypto_create_opts_init(cryptoopts
, errp
);
4947 qobject_unref(cryptoopts
);
4952 if (!qcrypto_block_calculate_payload_offset(create_opts
,
4959 luks_payload_size
= ROUND_UP(headerlen
, cluster_size
);
4962 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
4963 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
4965 /* Check that virtual disk size is valid */
4966 l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
4967 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
4968 cluster_size
/ l2e_size
);
4969 if (l2_tables
* L1E_SIZE
> QCOW_MAX_L1_SIZE
) {
4970 error_setg(&local_err
, "The image size is too large "
4971 "(try using a larger cluster size)");
4975 /* Account for input image */
4977 int64_t ssize
= bdrv_getlength(in_bs
);
4979 error_setg_errno(&local_err
, -ssize
,
4980 "Unable to get image virtual_size");
4984 virtual_size
= ROUND_UP(ssize
, cluster_size
);
4986 if (has_backing_file
) {
4987 /* We don't how much of the backing chain is shared by the input
4988 * image and the new image file. In the worst case the new image's
4989 * backing file has nothing in common with the input image. Be
4990 * conservative and assume all clusters need to be written.
4992 required
= virtual_size
;
4997 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
5000 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
5001 ssize
- offset
, &pnum
, NULL
,
5004 error_setg_errno(&local_err
, -ret
,
5005 "Unable to get block status");
5009 if (ret
& BDRV_BLOCK_ZERO
) {
5010 /* Skip zero regions (safe with no backing file) */
5011 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
5012 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
5013 /* Extend pnum to end of cluster for next iteration */
5014 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
5016 /* Count clusters we've seen */
5017 required
+= offset
% cluster_size
+ pnum
;
5023 /* Take into account preallocation. Nothing special is needed for
5024 * PREALLOC_MODE_METADATA since metadata is always counted.
5026 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
5027 required
= virtual_size
;
5030 info
= g_new0(BlockMeasureInfo
, 1);
5031 info
->fully_allocated
= luks_payload_size
+
5032 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
5033 ctz32(refcount_bits
), extended_l2
);
5036 * Remove data clusters that are not required. This overestimates the
5037 * required size because metadata needed for the fully allocated file is
5038 * still counted. Show bitmaps only if both source and destination
5039 * would support them.
5041 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
5042 info
->has_bitmaps
= version
>= 3 && in_bs
&&
5043 bdrv_supports_persistent_dirty_bitmap(in_bs
);
5044 if (info
->has_bitmaps
) {
5045 info
->bitmaps
= qcow2_get_persistent_dirty_bitmap_size(in_bs
,
5051 error_propagate(errp
, local_err
);
5055 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5057 BDRVQcow2State
*s
= bs
->opaque
;
5058 bdi
->cluster_size
= s
->cluster_size
;
5059 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
5063 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
,
5066 BDRVQcow2State
*s
= bs
->opaque
;
5067 ImageInfoSpecific
*spec_info
;
5068 QCryptoBlockInfo
*encrypt_info
= NULL
;
5069 Error
*local_err
= NULL
;
5071 if (s
->crypto
!= NULL
) {
5072 encrypt_info
= qcrypto_block_get_info(s
->crypto
, &local_err
);
5074 error_propagate(errp
, local_err
);
5079 spec_info
= g_new(ImageInfoSpecific
, 1);
5080 *spec_info
= (ImageInfoSpecific
){
5081 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
5082 .u
.qcow2
.data
= g_new0(ImageInfoSpecificQCow2
, 1),
5084 if (s
->qcow_version
== 2) {
5085 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5086 .compat
= g_strdup("0.10"),
5087 .refcount_bits
= s
->refcount_bits
,
5089 } else if (s
->qcow_version
== 3) {
5090 Qcow2BitmapInfoList
*bitmaps
;
5091 bitmaps
= qcow2_get_bitmap_info_list(bs
, &local_err
);
5093 error_propagate(errp
, local_err
);
5094 qapi_free_ImageInfoSpecific(spec_info
);
5095 qapi_free_QCryptoBlockInfo(encrypt_info
);
5098 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5099 .compat
= g_strdup("1.1"),
5100 .lazy_refcounts
= s
->compatible_features
&
5101 QCOW2_COMPAT_LAZY_REFCOUNTS
,
5102 .has_lazy_refcounts
= true,
5103 .corrupt
= s
->incompatible_features
&
5104 QCOW2_INCOMPAT_CORRUPT
,
5105 .has_corrupt
= true,
5106 .has_extended_l2
= true,
5107 .extended_l2
= has_subclusters(s
),
5108 .refcount_bits
= s
->refcount_bits
,
5109 .has_bitmaps
= !!bitmaps
,
5111 .has_data_file
= !!s
->image_data_file
,
5112 .data_file
= g_strdup(s
->image_data_file
),
5113 .has_data_file_raw
= has_data_file(bs
),
5114 .data_file_raw
= data_file_is_raw(bs
),
5115 .compression_type
= s
->compression_type
,
5118 /* if this assertion fails, this probably means a new version was
5119 * added without having it covered here */
5124 ImageInfoSpecificQCow2Encryption
*qencrypt
=
5125 g_new(ImageInfoSpecificQCow2Encryption
, 1);
5126 switch (encrypt_info
->format
) {
5127 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
5128 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
5130 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
5131 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
5132 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
5137 /* Since we did shallow copy above, erase any pointers
5138 * in the original info */
5139 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
5140 qapi_free_QCryptoBlockInfo(encrypt_info
);
5142 spec_info
->u
.qcow2
.data
->has_encrypt
= true;
5143 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
5149 static int qcow2_has_zero_init(BlockDriverState
*bs
)
5151 BDRVQcow2State
*s
= bs
->opaque
;
5154 if (qemu_in_coroutine()) {
5155 qemu_co_mutex_lock(&s
->lock
);
5158 * Check preallocation status: Preallocated images have all L2
5159 * tables allocated, nonpreallocated images have none. It is
5160 * therefore enough to check the first one.
5162 preallocated
= s
->l1_size
> 0 && s
->l1_table
[0] != 0;
5163 if (qemu_in_coroutine()) {
5164 qemu_co_mutex_unlock(&s
->lock
);
5167 if (!preallocated
) {
5169 } else if (bs
->encrypted
) {
5172 return bdrv_has_zero_init(s
->data_file
->bs
);
5176 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
5179 BDRVQcow2State
*s
= bs
->opaque
;
5181 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
5182 return bs
->drv
->bdrv_co_pwritev_part(bs
, qcow2_vm_state_offset(s
) + pos
,
5183 qiov
->size
, qiov
, 0, 0);
5186 static int qcow2_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
5189 BDRVQcow2State
*s
= bs
->opaque
;
5191 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
5192 return bs
->drv
->bdrv_co_preadv_part(bs
, qcow2_vm_state_offset(s
) + pos
,
5193 qiov
->size
, qiov
, 0, 0);
5197 * Downgrades an image's version. To achieve this, any incompatible features
5198 * have to be removed.
5200 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
5201 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5204 BDRVQcow2State
*s
= bs
->opaque
;
5205 int current_version
= s
->qcow_version
;
5209 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5210 assert(target_version
< current_version
);
5212 /* There are no other versions (now) that you can downgrade to */
5213 assert(target_version
== 2);
5215 if (s
->refcount_order
!= 4) {
5216 error_setg(errp
, "compat=0.10 requires refcount_bits=16");
5220 if (has_data_file(bs
)) {
5221 error_setg(errp
, "Cannot downgrade an image with a data file");
5226 * If any internal snapshot has a different size than the current
5227 * image size, or VM state size that exceeds 32 bits, downgrading
5228 * is unsafe. Even though we would still use v3-compliant output
5229 * to preserve that data, other v2 programs might not realize
5230 * those optional fields are important.
5232 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5233 if (s
->snapshots
[i
].vm_state_size
> UINT32_MAX
||
5234 s
->snapshots
[i
].disk_size
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
5235 error_setg(errp
, "Internal snapshots prevent downgrade of image");
5240 /* clear incompatible features */
5241 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
5242 ret
= qcow2_mark_clean(bs
);
5244 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5249 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5250 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5251 * best thing to do anyway */
5253 if (s
->incompatible_features
) {
5254 error_setg(errp
, "Cannot downgrade an image with incompatible features "
5255 "%#" PRIx64
" set", s
->incompatible_features
);
5259 /* since we can ignore compatible features, we can set them to 0 as well */
5260 s
->compatible_features
= 0;
5261 /* if lazy refcounts have been used, they have already been fixed through
5262 * clearing the dirty flag */
5264 /* clearing autoclear features is trivial */
5265 s
->autoclear_features
= 0;
5267 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
5269 error_setg_errno(errp
, -ret
, "Failed to turn zero into data clusters");
5273 s
->qcow_version
= target_version
;
5274 ret
= qcow2_update_header(bs
);
5276 s
->qcow_version
= current_version
;
5277 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5284 * Upgrades an image's version. While newer versions encompass all
5285 * features of older versions, some things may have to be presented
5288 static int qcow2_upgrade(BlockDriverState
*bs
, int target_version
,
5289 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5292 BDRVQcow2State
*s
= bs
->opaque
;
5293 bool need_snapshot_update
;
5294 int current_version
= s
->qcow_version
;
5298 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5299 assert(target_version
> current_version
);
5301 /* There are no other versions (yet) that you can upgrade to */
5302 assert(target_version
== 3);
5304 status_cb(bs
, 0, 2, cb_opaque
);
5307 * In v2, snapshots do not need to have extra data. v3 requires
5308 * the 64-bit VM state size and the virtual disk size to be
5310 * qcow2_write_snapshots() will always write the list in the
5311 * v3-compliant format.
5313 need_snapshot_update
= false;
5314 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5315 if (s
->snapshots
[i
].extra_data_size
<
5316 sizeof_field(QCowSnapshotExtraData
, vm_state_size_large
) +
5317 sizeof_field(QCowSnapshotExtraData
, disk_size
))
5319 need_snapshot_update
= true;
5323 if (need_snapshot_update
) {
5324 ret
= qcow2_write_snapshots(bs
);
5326 error_setg_errno(errp
, -ret
, "Failed to update the snapshot table");
5330 status_cb(bs
, 1, 2, cb_opaque
);
5332 s
->qcow_version
= target_version
;
5333 ret
= qcow2_update_header(bs
);
5335 s
->qcow_version
= current_version
;
5336 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5339 status_cb(bs
, 2, 2, cb_opaque
);
5344 typedef enum Qcow2AmendOperation
{
5345 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5346 * statically initialized to so that the helper CB can discern the first
5347 * invocation from an operation change */
5348 QCOW2_NO_OPERATION
= 0,
5351 QCOW2_UPDATING_ENCRYPTION
,
5352 QCOW2_CHANGING_REFCOUNT_ORDER
,
5354 } Qcow2AmendOperation
;
5356 typedef struct Qcow2AmendHelperCBInfo
{
5357 /* The code coordinating the amend operations should only modify
5358 * these four fields; the rest will be managed by the CB */
5359 BlockDriverAmendStatusCB
*original_status_cb
;
5360 void *original_cb_opaque
;
5362 Qcow2AmendOperation current_operation
;
5364 /* Total number of operations to perform (only set once) */
5365 int total_operations
;
5367 /* The following fields are managed by the CB */
5369 /* Number of operations completed */
5370 int operations_completed
;
5372 /* Cumulative offset of all completed operations */
5373 int64_t offset_completed
;
5375 Qcow2AmendOperation last_operation
;
5376 int64_t last_work_size
;
5377 } Qcow2AmendHelperCBInfo
;
5379 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
5380 int64_t operation_offset
,
5381 int64_t operation_work_size
, void *opaque
)
5383 Qcow2AmendHelperCBInfo
*info
= opaque
;
5384 int64_t current_work_size
;
5385 int64_t projected_work_size
;
5387 if (info
->current_operation
!= info
->last_operation
) {
5388 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
5389 info
->offset_completed
+= info
->last_work_size
;
5390 info
->operations_completed
++;
5393 info
->last_operation
= info
->current_operation
;
5396 assert(info
->total_operations
> 0);
5397 assert(info
->operations_completed
< info
->total_operations
);
5399 info
->last_work_size
= operation_work_size
;
5401 current_work_size
= info
->offset_completed
+ operation_work_size
;
5403 /* current_work_size is the total work size for (operations_completed + 1)
5404 * operations (which includes this one), so multiply it by the number of
5405 * operations not covered and divide it by the number of operations
5406 * covered to get a projection for the operations not covered */
5407 projected_work_size
= current_work_size
* (info
->total_operations
-
5408 info
->operations_completed
- 1)
5409 / (info
->operations_completed
+ 1);
5411 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
5412 current_work_size
+ projected_work_size
,
5413 info
->original_cb_opaque
);
5416 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5417 BlockDriverAmendStatusCB
*status_cb
,
5422 BDRVQcow2State
*s
= bs
->opaque
;
5423 int old_version
= s
->qcow_version
, new_version
= old_version
;
5424 uint64_t new_size
= 0;
5425 const char *backing_file
= NULL
, *backing_format
= NULL
, *data_file
= NULL
;
5426 bool lazy_refcounts
= s
->use_lazy_refcounts
;
5427 bool data_file_raw
= data_file_is_raw(bs
);
5428 const char *compat
= NULL
;
5429 int refcount_bits
= s
->refcount_bits
;
5431 QemuOptDesc
*desc
= opts
->list
->desc
;
5432 Qcow2AmendHelperCBInfo helper_cb_info
;
5433 bool encryption_update
= false;
5435 while (desc
&& desc
->name
) {
5436 if (!qemu_opt_find(opts
, desc
->name
)) {
5437 /* only change explicitly defined options */
5442 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
5443 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
5445 /* preserve default */
5446 } else if (!strcmp(compat
, "0.10") || !strcmp(compat
, "v2")) {
5448 } else if (!strcmp(compat
, "1.1") || !strcmp(compat
, "v3")) {
5451 error_setg(errp
, "Unknown compatibility level %s", compat
);
5454 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
5455 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
5456 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
5457 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5458 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
5459 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5460 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
5463 "Can't amend encryption options - encryption not present");
5466 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5468 "Only LUKS encryption options can be amended");
5471 encryption_update
= true;
5472 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
5473 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
5475 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
5476 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
5479 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
5480 !is_power_of_2(refcount_bits
))
5482 error_setg(errp
, "Refcount width must be a power of two and "
5483 "may not exceed 64 bits");
5486 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE
)) {
5487 data_file
= qemu_opt_get(opts
, BLOCK_OPT_DATA_FILE
);
5488 if (data_file
&& !has_data_file(bs
)) {
5489 error_setg(errp
, "data-file can only be set for images that "
5490 "use an external data file");
5493 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE_RAW
)) {
5494 data_file_raw
= qemu_opt_get_bool(opts
, BLOCK_OPT_DATA_FILE_RAW
,
5496 if (data_file_raw
&& !data_file_is_raw(bs
)) {
5497 error_setg(errp
, "data-file-raw cannot be set on existing "
5502 /* if this point is reached, this probably means a new option was
5503 * added without having it covered here */
5510 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
5511 .original_status_cb
= status_cb
,
5512 .original_cb_opaque
= cb_opaque
,
5513 .total_operations
= (new_version
!= old_version
)
5514 + (s
->refcount_bits
!= refcount_bits
) +
5515 (encryption_update
== true)
5518 /* Upgrade first (some features may require compat=1.1) */
5519 if (new_version
> old_version
) {
5520 helper_cb_info
.current_operation
= QCOW2_UPGRADING
;
5521 ret
= qcow2_upgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5522 &helper_cb_info
, errp
);
5528 if (encryption_update
) {
5529 QDict
*amend_opts_dict
;
5530 QCryptoBlockAmendOptions
*amend_opts
;
5532 helper_cb_info
.current_operation
= QCOW2_UPDATING_ENCRYPTION
;
5533 amend_opts_dict
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
5534 if (!amend_opts_dict
) {
5537 amend_opts
= block_crypto_amend_opts_init(amend_opts_dict
, errp
);
5538 qobject_unref(amend_opts_dict
);
5542 ret
= qcrypto_block_amend_options(s
->crypto
,
5543 qcow2_crypto_hdr_read_func
,
5544 qcow2_crypto_hdr_write_func
,
5549 qapi_free_QCryptoBlockAmendOptions(amend_opts
);
5555 if (s
->refcount_bits
!= refcount_bits
) {
5556 int refcount_order
= ctz32(refcount_bits
);
5558 if (new_version
< 3 && refcount_bits
!= 16) {
5559 error_setg(errp
, "Refcount widths other than 16 bits require "
5560 "compatibility level 1.1 or above (use compat=1.1 or "
5565 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
5566 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
5567 &qcow2_amend_helper_cb
,
5568 &helper_cb_info
, errp
);
5574 /* data-file-raw blocks backing files, so clear it first if requested */
5575 if (data_file_raw
) {
5576 s
->autoclear_features
|= QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5578 s
->autoclear_features
&= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5582 g_free(s
->image_data_file
);
5583 s
->image_data_file
= *data_file
? g_strdup(data_file
) : NULL
;
5586 ret
= qcow2_update_header(bs
);
5588 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5592 if (backing_file
|| backing_format
) {
5593 if (g_strcmp0(backing_file
, s
->image_backing_file
) ||
5594 g_strcmp0(backing_format
, s
->image_backing_format
)) {
5595 warn_report("Deprecated use of amend to alter the backing file; "
5596 "use qemu-img rebase instead");
5598 ret
= qcow2_change_backing_file(bs
,
5599 backing_file
?: s
->image_backing_file
,
5600 backing_format
?: s
->image_backing_format
);
5602 error_setg_errno(errp
, -ret
, "Failed to change the backing file");
5607 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
5608 if (lazy_refcounts
) {
5609 if (new_version
< 3) {
5610 error_setg(errp
, "Lazy refcounts only supported with "
5611 "compatibility level 1.1 and above (use compat=1.1 "
5615 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5616 ret
= qcow2_update_header(bs
);
5618 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5619 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5622 s
->use_lazy_refcounts
= true;
5624 /* make image clean first */
5625 ret
= qcow2_mark_clean(bs
);
5627 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5630 /* now disallow lazy refcounts */
5631 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5632 ret
= qcow2_update_header(bs
);
5634 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5635 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5638 s
->use_lazy_refcounts
= false;
5643 BlockBackend
*blk
= blk_new_with_bs(bs
, BLK_PERM_RESIZE
, BLK_PERM_ALL
,
5650 * Amending image options should ensure that the image has
5651 * exactly the given new values, so pass exact=true here.
5653 ret
= blk_truncate(blk
, new_size
, true, PREALLOC_MODE_OFF
, 0, errp
);
5660 /* Downgrade last (so unsupported features can be removed before) */
5661 if (new_version
< old_version
) {
5662 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
5663 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5664 &helper_cb_info
, errp
);
5673 static int coroutine_fn
qcow2_co_amend(BlockDriverState
*bs
,
5674 BlockdevAmendOptions
*opts
,
5678 BlockdevAmendOptionsQcow2
*qopts
= &opts
->u
.qcow2
;
5679 BDRVQcow2State
*s
= bs
->opaque
;
5682 if (qopts
->has_encrypt
) {
5684 error_setg(errp
, "image is not encrypted, can't amend");
5688 if (qopts
->encrypt
->format
!= Q_CRYPTO_BLOCK_FORMAT_LUKS
) {
5690 "Amend can't be used to change the qcow2 encryption format");
5694 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5696 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5700 ret
= qcrypto_block_amend_options(s
->crypto
,
5701 qcow2_crypto_hdr_read_func
,
5702 qcow2_crypto_hdr_write_func
,
5712 * If offset or size are negative, respectively, they will not be included in
5713 * the BLOCK_IMAGE_CORRUPTED event emitted.
5714 * fatal will be ignored for read-only BDS; corruptions found there will always
5715 * be considered non-fatal.
5717 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
5718 int64_t size
, const char *message_format
, ...)
5720 BDRVQcow2State
*s
= bs
->opaque
;
5721 const char *node_name
;
5725 fatal
= fatal
&& bdrv_is_writable(bs
);
5727 if (s
->signaled_corruption
&&
5728 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
5733 va_start(ap
, message_format
);
5734 message
= g_strdup_vprintf(message_format
, ap
);
5738 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
5739 "corruption events will be suppressed\n", message
);
5741 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
5742 "corruption events will be suppressed\n", message
);
5745 node_name
= bdrv_get_node_name(bs
);
5746 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
5747 *node_name
!= '\0', node_name
,
5748 message
, offset
>= 0, offset
,
5754 qcow2_mark_corrupt(bs
);
5755 bs
->drv
= NULL
; /* make BDS unusable */
5758 s
->signaled_corruption
= true;
5761 #define QCOW_COMMON_OPTIONS \
5763 .name = BLOCK_OPT_SIZE, \
5764 .type = QEMU_OPT_SIZE, \
5765 .help = "Virtual disk size" \
5768 .name = BLOCK_OPT_COMPAT_LEVEL, \
5769 .type = QEMU_OPT_STRING, \
5770 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
5773 .name = BLOCK_OPT_BACKING_FILE, \
5774 .type = QEMU_OPT_STRING, \
5775 .help = "File name of a base image" \
5778 .name = BLOCK_OPT_BACKING_FMT, \
5779 .type = QEMU_OPT_STRING, \
5780 .help = "Image format of the base image" \
5783 .name = BLOCK_OPT_DATA_FILE, \
5784 .type = QEMU_OPT_STRING, \
5785 .help = "File name of an external data file" \
5788 .name = BLOCK_OPT_DATA_FILE_RAW, \
5789 .type = QEMU_OPT_BOOL, \
5790 .help = "The external data file must stay valid " \
5794 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
5795 .type = QEMU_OPT_BOOL, \
5796 .help = "Postpone refcount updates", \
5797 .def_value_str = "off" \
5800 .name = BLOCK_OPT_REFCOUNT_BITS, \
5801 .type = QEMU_OPT_NUMBER, \
5802 .help = "Width of a reference count entry in bits", \
5803 .def_value_str = "16" \
5806 static QemuOptsList qcow2_create_opts
= {
5807 .name
= "qcow2-create-opts",
5808 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
5811 .name
= BLOCK_OPT_ENCRYPT
, \
5812 .type
= QEMU_OPT_BOOL
, \
5813 .help
= "Encrypt the image with format 'aes'. (Deprecated " \
5814 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)", \
5817 .name
= BLOCK_OPT_ENCRYPT_FORMAT
, \
5818 .type
= QEMU_OPT_STRING
, \
5819 .help
= "Encrypt the image, format choices: 'aes', 'luks'", \
5821 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
5822 "ID of secret providing qcow AES key or LUKS passphrase"), \
5823 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
5824 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
5825 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
5826 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
5827 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
5828 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
5830 .name
= BLOCK_OPT_CLUSTER_SIZE
, \
5831 .type
= QEMU_OPT_SIZE
, \
5832 .help
= "qcow2 cluster size", \
5833 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
) \
5836 .name
= BLOCK_OPT_EXTL2
, \
5837 .type
= QEMU_OPT_BOOL
, \
5838 .help
= "Extended L2 tables", \
5839 .def_value_str
= "off" \
5842 .name
= BLOCK_OPT_PREALLOC
, \
5843 .type
= QEMU_OPT_STRING
, \
5844 .help
= "Preallocation mode (allowed values: off, " \
5845 "metadata, falloc, full)" \
5848 .name
= BLOCK_OPT_COMPRESSION_TYPE
, \
5849 .type
= QEMU_OPT_STRING
, \
5850 .help
= "Compression method used for image cluster " \
5852 .def_value_str
= "zlib" \
5854 QCOW_COMMON_OPTIONS
,
5855 { /* end of list */ }
5859 static QemuOptsList qcow2_amend_opts
= {
5860 .name
= "qcow2-amend-opts",
5861 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts
.head
),
5863 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
5864 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
5865 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
5866 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
5867 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
5868 QCOW_COMMON_OPTIONS
,
5869 { /* end of list */ }
5873 static const char *const qcow2_strong_runtime_opts
[] = {
5874 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET
,
5879 BlockDriver bdrv_qcow2
= {
5880 .format_name
= "qcow2",
5881 .instance_size
= sizeof(BDRVQcow2State
),
5882 .bdrv_probe
= qcow2_probe
,
5883 .bdrv_open
= qcow2_open
,
5884 .bdrv_close
= qcow2_close
,
5885 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
5886 .bdrv_reopen_commit
= qcow2_reopen_commit
,
5887 .bdrv_reopen_commit_post
= qcow2_reopen_commit_post
,
5888 .bdrv_reopen_abort
= qcow2_reopen_abort
,
5889 .bdrv_join_options
= qcow2_join_options
,
5890 .bdrv_child_perm
= bdrv_default_perms
,
5891 .bdrv_co_create_opts
= qcow2_co_create_opts
,
5892 .bdrv_co_create
= qcow2_co_create
,
5893 .bdrv_has_zero_init
= qcow2_has_zero_init
,
5894 .bdrv_co_block_status
= qcow2_co_block_status
,
5896 .bdrv_co_preadv_part
= qcow2_co_preadv_part
,
5897 .bdrv_co_pwritev_part
= qcow2_co_pwritev_part
,
5898 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
5900 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
5901 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
5902 .bdrv_co_copy_range_from
= qcow2_co_copy_range_from
,
5903 .bdrv_co_copy_range_to
= qcow2_co_copy_range_to
,
5904 .bdrv_co_truncate
= qcow2_co_truncate
,
5905 .bdrv_co_pwritev_compressed_part
= qcow2_co_pwritev_compressed_part
,
5906 .bdrv_make_empty
= qcow2_make_empty
,
5908 .bdrv_snapshot_create
= qcow2_snapshot_create
,
5909 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
5910 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
5911 .bdrv_snapshot_list
= qcow2_snapshot_list
,
5912 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
5913 .bdrv_measure
= qcow2_measure
,
5914 .bdrv_get_info
= qcow2_get_info
,
5915 .bdrv_get_specific_info
= qcow2_get_specific_info
,
5917 .bdrv_save_vmstate
= qcow2_save_vmstate
,
5918 .bdrv_load_vmstate
= qcow2_load_vmstate
,
5921 .supports_backing
= true,
5922 .bdrv_change_backing_file
= qcow2_change_backing_file
,
5924 .bdrv_refresh_limits
= qcow2_refresh_limits
,
5925 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
5926 .bdrv_inactivate
= qcow2_inactivate
,
5928 .create_opts
= &qcow2_create_opts
,
5929 .amend_opts
= &qcow2_amend_opts
,
5930 .strong_runtime_opts
= qcow2_strong_runtime_opts
,
5931 .mutable_opts
= mutable_opts
,
5932 .bdrv_co_check
= qcow2_co_check
,
5933 .bdrv_amend_options
= qcow2_amend_options
,
5934 .bdrv_co_amend
= qcow2_co_amend
,
5936 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
5937 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
5939 .bdrv_supports_persistent_dirty_bitmap
=
5940 qcow2_supports_persistent_dirty_bitmap
,
5941 .bdrv_co_can_store_new_dirty_bitmap
= qcow2_co_can_store_new_dirty_bitmap
,
5942 .bdrv_co_remove_persistent_dirty_bitmap
=
5943 qcow2_co_remove_persistent_dirty_bitmap
,
5946 static void bdrv_qcow2_init(void)
5948 bdrv_register(&bdrv_qcow2
);
5951 block_init(bdrv_qcow2_init
);