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 bool 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
);
963 typedef struct Qcow2ReopenState
{
964 Qcow2Cache
*l2_table_cache
;
965 Qcow2Cache
*refcount_block_cache
;
966 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
967 bool use_lazy_refcounts
;
969 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
970 uint64_t cache_clean_interval
;
971 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
974 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
976 QDict
*options
, int flags
,
979 BDRVQcow2State
*s
= bs
->opaque
;
980 QemuOpts
*opts
= NULL
;
981 const char *opt_overlap_check
, *opt_overlap_check_template
;
982 int overlap_check_template
= 0;
983 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
985 const char *encryptfmt
;
986 QDict
*encryptopts
= NULL
;
989 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
990 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
992 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
993 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
998 /* get L2 table/refcount block cache size from command line options */
999 if (!read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
1000 &refcount_cache_size
, errp
)) {
1005 l2_cache_size
/= l2_cache_entry_size
;
1006 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
1007 l2_cache_size
= MIN_L2_CACHE_SIZE
;
1009 if (l2_cache_size
> INT_MAX
) {
1010 error_setg(errp
, "L2 cache size too big");
1015 refcount_cache_size
/= s
->cluster_size
;
1016 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
1017 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
1019 if (refcount_cache_size
> INT_MAX
) {
1020 error_setg(errp
, "Refcount cache size too big");
1025 /* alloc new L2 table/refcount block cache, flush old one */
1026 if (s
->l2_table_cache
) {
1027 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1029 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
1034 if (s
->refcount_block_cache
) {
1035 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1037 error_setg_errno(errp
, -ret
,
1038 "Failed to flush the refcount block cache");
1043 r
->l2_slice_size
= l2_cache_entry_size
/ l2_entry_size(s
);
1044 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
1045 l2_cache_entry_size
);
1046 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
1048 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
1049 error_setg(errp
, "Could not allocate metadata caches");
1054 /* New interval for cache cleanup timer */
1055 r
->cache_clean_interval
=
1056 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
1057 DEFAULT_CACHE_CLEAN_INTERVAL
);
1058 #ifndef CONFIG_LINUX
1059 if (r
->cache_clean_interval
!= 0) {
1060 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1061 " not supported on this host");
1066 if (r
->cache_clean_interval
> UINT_MAX
) {
1067 error_setg(errp
, "Cache clean interval too big");
1072 /* lazy-refcounts; flush if going from enabled to disabled */
1073 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
1074 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
1075 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
1076 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
1077 "qemu 1.1 compatibility level");
1082 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
1083 ret
= qcow2_mark_clean(bs
);
1085 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
1090 /* Overlap check options */
1091 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
1092 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1093 if (opt_overlap_check_template
&& opt_overlap_check
&&
1094 strcmp(opt_overlap_check_template
, opt_overlap_check
))
1096 error_setg(errp
, "Conflicting values for qcow2 options '"
1097 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1098 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
1102 if (!opt_overlap_check
) {
1103 opt_overlap_check
= opt_overlap_check_template
?: "cached";
1106 if (!strcmp(opt_overlap_check
, "none")) {
1107 overlap_check_template
= 0;
1108 } else if (!strcmp(opt_overlap_check
, "constant")) {
1109 overlap_check_template
= QCOW2_OL_CONSTANT
;
1110 } else if (!strcmp(opt_overlap_check
, "cached")) {
1111 overlap_check_template
= QCOW2_OL_CACHED
;
1112 } else if (!strcmp(opt_overlap_check
, "all")) {
1113 overlap_check_template
= QCOW2_OL_ALL
;
1115 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1116 "'overlap-check'. Allowed are any of the following: "
1117 "none, constant, cached, all", opt_overlap_check
);
1122 r
->overlap_check
= 0;
1123 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1124 /* overlap-check defines a template bitmask, but every flag may be
1125 * overwritten through the associated boolean option */
1127 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1128 overlap_check_template
& (1 << i
)) << i
;
1131 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1132 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1133 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1134 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1135 flags
& BDRV_O_UNMAP
);
1136 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1137 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1138 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1139 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1141 switch (s
->crypt_method_header
) {
1142 case QCOW_CRYPT_NONE
:
1144 error_setg(errp
, "No encryption in image header, but options "
1145 "specified format '%s'", encryptfmt
);
1151 case QCOW_CRYPT_AES
:
1152 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1154 "Header reported 'aes' encryption format but "
1155 "options specify '%s'", encryptfmt
);
1159 qdict_put_str(encryptopts
, "format", "qcow");
1160 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1161 if (!r
->crypto_opts
) {
1167 case QCOW_CRYPT_LUKS
:
1168 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1170 "Header reported 'luks' encryption format but "
1171 "options specify '%s'", encryptfmt
);
1175 qdict_put_str(encryptopts
, "format", "luks");
1176 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1177 if (!r
->crypto_opts
) {
1184 error_setg(errp
, "Unsupported encryption method %d",
1185 s
->crypt_method_header
);
1192 qobject_unref(encryptopts
);
1193 qemu_opts_del(opts
);
1198 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1199 Qcow2ReopenState
*r
)
1201 BDRVQcow2State
*s
= bs
->opaque
;
1204 if (s
->l2_table_cache
) {
1205 qcow2_cache_destroy(s
->l2_table_cache
);
1207 if (s
->refcount_block_cache
) {
1208 qcow2_cache_destroy(s
->refcount_block_cache
);
1210 s
->l2_table_cache
= r
->l2_table_cache
;
1211 s
->refcount_block_cache
= r
->refcount_block_cache
;
1212 s
->l2_slice_size
= r
->l2_slice_size
;
1214 s
->overlap_check
= r
->overlap_check
;
1215 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1217 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1218 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1221 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1222 cache_clean_timer_del(bs
);
1223 s
->cache_clean_interval
= r
->cache_clean_interval
;
1224 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1227 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1228 s
->crypto_opts
= r
->crypto_opts
;
1231 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1232 Qcow2ReopenState
*r
)
1234 if (r
->l2_table_cache
) {
1235 qcow2_cache_destroy(r
->l2_table_cache
);
1237 if (r
->refcount_block_cache
) {
1238 qcow2_cache_destroy(r
->refcount_block_cache
);
1240 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1243 static int qcow2_update_options(BlockDriverState
*bs
, QDict
*options
,
1244 int flags
, Error
**errp
)
1246 Qcow2ReopenState r
= {};
1249 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1251 qcow2_update_options_commit(bs
, &r
);
1253 qcow2_update_options_abort(bs
, &r
);
1259 static int validate_compression_type(BDRVQcow2State
*s
, Error
**errp
)
1261 switch (s
->compression_type
) {
1262 case QCOW2_COMPRESSION_TYPE_ZLIB
:
1264 case QCOW2_COMPRESSION_TYPE_ZSTD
:
1269 error_setg(errp
, "qcow2: unknown compression type: %u",
1270 s
->compression_type
);
1275 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1276 * the incompatible feature flag must be set
1278 if (s
->compression_type
== QCOW2_COMPRESSION_TYPE_ZLIB
) {
1279 if (s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
) {
1280 error_setg(errp
, "qcow2: Compression type incompatible feature "
1281 "bit must not be set");
1285 if (!(s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
)) {
1286 error_setg(errp
, "qcow2: Compression type incompatible feature "
1295 /* Called with s->lock held. */
1296 static int coroutine_fn
qcow2_do_open(BlockDriverState
*bs
, QDict
*options
,
1297 int flags
, Error
**errp
)
1300 BDRVQcow2State
*s
= bs
->opaque
;
1301 unsigned int len
, i
;
1305 uint64_t l1_vm_state_index
;
1306 bool update_header
= false;
1308 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
1310 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1313 header
.magic
= be32_to_cpu(header
.magic
);
1314 header
.version
= be32_to_cpu(header
.version
);
1315 header
.backing_file_offset
= be64_to_cpu(header
.backing_file_offset
);
1316 header
.backing_file_size
= be32_to_cpu(header
.backing_file_size
);
1317 header
.size
= be64_to_cpu(header
.size
);
1318 header
.cluster_bits
= be32_to_cpu(header
.cluster_bits
);
1319 header
.crypt_method
= be32_to_cpu(header
.crypt_method
);
1320 header
.l1_table_offset
= be64_to_cpu(header
.l1_table_offset
);
1321 header
.l1_size
= be32_to_cpu(header
.l1_size
);
1322 header
.refcount_table_offset
= be64_to_cpu(header
.refcount_table_offset
);
1323 header
.refcount_table_clusters
=
1324 be32_to_cpu(header
.refcount_table_clusters
);
1325 header
.snapshots_offset
= be64_to_cpu(header
.snapshots_offset
);
1326 header
.nb_snapshots
= be32_to_cpu(header
.nb_snapshots
);
1328 if (header
.magic
!= QCOW_MAGIC
) {
1329 error_setg(errp
, "Image is not in qcow2 format");
1333 if (header
.version
< 2 || header
.version
> 3) {
1334 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1339 s
->qcow_version
= header
.version
;
1341 /* Initialise cluster size */
1342 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1343 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1344 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1345 header
.cluster_bits
);
1350 s
->cluster_bits
= header
.cluster_bits
;
1351 s
->cluster_size
= 1 << s
->cluster_bits
;
1353 /* Initialise version 3 header fields */
1354 if (header
.version
== 2) {
1355 header
.incompatible_features
= 0;
1356 header
.compatible_features
= 0;
1357 header
.autoclear_features
= 0;
1358 header
.refcount_order
= 4;
1359 header
.header_length
= 72;
1361 header
.incompatible_features
=
1362 be64_to_cpu(header
.incompatible_features
);
1363 header
.compatible_features
= be64_to_cpu(header
.compatible_features
);
1364 header
.autoclear_features
= be64_to_cpu(header
.autoclear_features
);
1365 header
.refcount_order
= be32_to_cpu(header
.refcount_order
);
1366 header
.header_length
= be32_to_cpu(header
.header_length
);
1368 if (header
.header_length
< 104) {
1369 error_setg(errp
, "qcow2 header too short");
1375 if (header
.header_length
> s
->cluster_size
) {
1376 error_setg(errp
, "qcow2 header exceeds cluster size");
1381 if (header
.header_length
> sizeof(header
)) {
1382 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1383 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1384 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
1385 s
->unknown_header_fields_size
);
1387 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1393 if (header
.backing_file_offset
> s
->cluster_size
) {
1394 error_setg(errp
, "Invalid backing file offset");
1399 if (header
.backing_file_offset
) {
1400 ext_end
= header
.backing_file_offset
;
1402 ext_end
= 1 << header
.cluster_bits
;
1405 /* Handle feature bits */
1406 s
->incompatible_features
= header
.incompatible_features
;
1407 s
->compatible_features
= header
.compatible_features
;
1408 s
->autoclear_features
= header
.autoclear_features
;
1411 * Handle compression type
1412 * Older qcow2 images don't contain the compression type header.
1413 * Distinguish them by the header length and use
1414 * the only valid (default) compression type in that case
1416 if (header
.header_length
> offsetof(QCowHeader
, compression_type
)) {
1417 s
->compression_type
= header
.compression_type
;
1419 s
->compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
1422 ret
= validate_compression_type(s
, errp
);
1427 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1428 void *feature_table
= NULL
;
1429 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1430 &feature_table
, flags
, NULL
, NULL
);
1431 report_unsupported_feature(errp
, feature_table
,
1432 s
->incompatible_features
&
1433 ~QCOW2_INCOMPAT_MASK
);
1435 g_free(feature_table
);
1439 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1440 /* Corrupt images may not be written to unless they are being repaired
1442 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1443 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1450 s
->subclusters_per_cluster
=
1451 has_subclusters(s
) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
: 1;
1452 s
->subcluster_size
= s
->cluster_size
/ s
->subclusters_per_cluster
;
1453 s
->subcluster_bits
= ctz32(s
->subcluster_size
);
1455 if (s
->subcluster_size
< (1 << MIN_CLUSTER_BITS
)) {
1456 error_setg(errp
, "Unsupported subcluster size: %d", s
->subcluster_size
);
1461 /* Check support for various header values */
1462 if (header
.refcount_order
> 6) {
1463 error_setg(errp
, "Reference count entry width too large; may not "
1468 s
->refcount_order
= header
.refcount_order
;
1469 s
->refcount_bits
= 1 << s
->refcount_order
;
1470 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1471 s
->refcount_max
+= s
->refcount_max
- 1;
1473 s
->crypt_method_header
= header
.crypt_method
;
1474 if (s
->crypt_method_header
) {
1475 if (bdrv_uses_whitelist() &&
1476 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1478 "Use of AES-CBC encrypted qcow2 images is no longer "
1479 "supported in system emulators");
1480 error_append_hint(errp
,
1481 "You can use 'qemu-img convert' to convert your "
1482 "image to an alternative supported format, such "
1483 "as unencrypted qcow2, or raw with the LUKS "
1484 "format instead.\n");
1489 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1490 s
->crypt_physical_offset
= false;
1492 /* Assuming LUKS and any future crypt methods we
1493 * add will all use physical offsets, due to the
1494 * fact that the alternative is insecure... */
1495 s
->crypt_physical_offset
= true;
1498 bs
->encrypted
= true;
1501 s
->l2_bits
= s
->cluster_bits
- ctz32(l2_entry_size(s
));
1502 s
->l2_size
= 1 << s
->l2_bits
;
1503 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1504 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1505 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1506 bs
->total_sectors
= header
.size
/ BDRV_SECTOR_SIZE
;
1507 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1508 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1509 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1511 s
->refcount_table_offset
= header
.refcount_table_offset
;
1512 s
->refcount_table_size
=
1513 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1515 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1516 error_setg(errp
, "Image does not contain a reference count table");
1521 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1522 header
.refcount_table_clusters
,
1523 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1524 "Reference count table", errp
);
1529 if (!(flags
& BDRV_O_CHECK
)) {
1531 * The total size in bytes of the snapshot table is checked in
1532 * qcow2_read_snapshots() because the size of each snapshot is
1533 * variable and we don't know it yet.
1534 * Here we only check the offset and number of snapshots.
1536 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1537 header
.nb_snapshots
,
1538 sizeof(QCowSnapshotHeader
),
1539 sizeof(QCowSnapshotHeader
) *
1541 "Snapshot table", errp
);
1547 /* read the level 1 table */
1548 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1549 header
.l1_size
, L1E_SIZE
,
1550 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1554 s
->l1_size
= header
.l1_size
;
1555 s
->l1_table_offset
= header
.l1_table_offset
;
1557 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1558 if (l1_vm_state_index
> INT_MAX
) {
1559 error_setg(errp
, "Image is too big");
1563 s
->l1_vm_state_index
= l1_vm_state_index
;
1565 /* the L1 table must contain at least enough entries to put
1566 header.size bytes */
1567 if (s
->l1_size
< s
->l1_vm_state_index
) {
1568 error_setg(errp
, "L1 table is too small");
1573 if (s
->l1_size
> 0) {
1574 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
, s
->l1_size
* L1E_SIZE
);
1575 if (s
->l1_table
== NULL
) {
1576 error_setg(errp
, "Could not allocate L1 table");
1580 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1581 s
->l1_size
* L1E_SIZE
);
1583 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1586 for(i
= 0;i
< s
->l1_size
; i
++) {
1587 s
->l1_table
[i
] = be64_to_cpu(s
->l1_table
[i
]);
1591 /* Parse driver-specific options */
1592 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1599 ret
= qcow2_refcount_init(bs
);
1601 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1605 QLIST_INIT(&s
->cluster_allocs
);
1606 QTAILQ_INIT(&s
->discards
);
1608 /* read qcow2 extensions */
1609 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1610 flags
, &update_header
, errp
)) {
1615 /* Open external data file */
1616 s
->data_file
= bdrv_open_child(NULL
, options
, "data-file", bs
,
1617 &child_of_bds
, BDRV_CHILD_DATA
,
1624 if (s
->incompatible_features
& QCOW2_INCOMPAT_DATA_FILE
) {
1625 if (!s
->data_file
&& s
->image_data_file
) {
1626 s
->data_file
= bdrv_open_child(s
->image_data_file
, options
,
1627 "data-file", bs
, &child_of_bds
,
1628 BDRV_CHILD_DATA
, false, errp
);
1629 if (!s
->data_file
) {
1634 if (!s
->data_file
) {
1635 error_setg(errp
, "'data-file' is required for this image");
1641 bs
->file
->role
&= ~BDRV_CHILD_DATA
;
1643 /* Must succeed because we have given up permissions if anything */
1644 bdrv_child_refresh_perms(bs
, bs
->file
, &error_abort
);
1647 error_setg(errp
, "'data-file' can only be set for images with an "
1648 "external data file");
1653 s
->data_file
= bs
->file
;
1655 if (data_file_is_raw(bs
)) {
1656 error_setg(errp
, "data-file-raw requires a data file");
1662 /* qcow2_read_extension may have set up the crypto context
1663 * if the crypt method needs a header region, some methods
1664 * don't need header extensions, so must check here
1666 if (s
->crypt_method_header
&& !s
->crypto
) {
1667 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1668 unsigned int cflags
= 0;
1669 if (flags
& BDRV_O_NO_IO
) {
1670 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1672 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1674 QCOW2_MAX_THREADS
, errp
);
1679 } else if (!(flags
& BDRV_O_NO_IO
)) {
1680 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1681 s
->crypt_method_header
);
1687 /* read the backing file name */
1688 if (header
.backing_file_offset
!= 0) {
1689 len
= header
.backing_file_size
;
1690 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1691 len
>= sizeof(bs
->backing_file
)) {
1692 error_setg(errp
, "Backing file name too long");
1696 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
1697 bs
->auto_backing_file
, len
);
1699 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1702 bs
->auto_backing_file
[len
] = '\0';
1703 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
),
1704 bs
->auto_backing_file
);
1705 s
->image_backing_file
= g_strdup(bs
->auto_backing_file
);
1709 * Internal snapshots; skip reading them in check mode, because
1710 * we do not need them then, and we do not want to abort because
1711 * of a broken table.
1713 if (!(flags
& BDRV_O_CHECK
)) {
1714 s
->snapshots_offset
= header
.snapshots_offset
;
1715 s
->nb_snapshots
= header
.nb_snapshots
;
1717 ret
= qcow2_read_snapshots(bs
, errp
);
1723 /* Clear unknown autoclear feature bits */
1724 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1726 update_header
&& !bs
->read_only
&& !(flags
& BDRV_O_INACTIVE
);
1727 if (update_header
) {
1728 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1731 /* == Handle persistent dirty bitmaps ==
1733 * We want load dirty bitmaps in three cases:
1735 * 1. Normal open of the disk in active mode, not related to invalidation
1738 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1739 * bitmaps are _not_ migrating through migration channel, i.e.
1740 * 'dirty-bitmaps' capability is disabled.
1742 * 3. Invalidation of source vm after failed or canceled migration.
1743 * This is a very interesting case. There are two possible types of
1746 * A. Stored on inactivation and removed. They should be loaded from the
1749 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1750 * the migration channel (with dirty-bitmaps capability).
1752 * On the other hand, there are two possible sub-cases:
1754 * 3.1 disk was changed by somebody else while were inactive. In this
1755 * case all in-RAM dirty bitmaps (both persistent and not) are
1756 * definitely invalid. And we don't have any method to determine
1759 * Simple and safe thing is to just drop all the bitmaps of type B on
1760 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1762 * On the other hand, resuming source vm, if disk was already changed
1763 * is a bad thing anyway: not only bitmaps, the whole vm state is
1764 * out of sync with disk.
1766 * This means, that user or management tool, who for some reason
1767 * decided to resume source vm, after disk was already changed by
1768 * target vm, should at least drop all dirty bitmaps by hand.
1770 * So, we can ignore this case for now, but TODO: "generation"
1771 * extension for qcow2, to determine, that image was changed after
1772 * last inactivation. And if it is changed, we will drop (or at least
1773 * mark as 'invalid' all the bitmaps of type B, both persistent
1776 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1777 * to disk ('dirty-bitmaps' capability disabled), or not saved
1778 * ('dirty-bitmaps' capability enabled), but we don't need to care
1779 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1780 * and not stored has flag IN_USE=1 in the image and will be skipped
1783 * One remaining possible case when we don't want load bitmaps:
1785 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1786 * will be loaded on invalidation, no needs try loading them before)
1789 if (!(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
)) {
1790 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1791 bool header_updated
;
1792 if (!qcow2_load_dirty_bitmaps(bs
, &header_updated
, errp
)) {
1797 update_header
= update_header
&& !header_updated
;
1800 if (update_header
) {
1801 ret
= qcow2_update_header(bs
);
1803 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1808 bs
->supported_zero_flags
= header
.version
>= 3 ?
1809 BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
: 0;
1810 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
1812 /* Repair image if dirty */
1813 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INACTIVE
)) && !bs
->read_only
&&
1814 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1815 BdrvCheckResult result
= {0};
1817 ret
= qcow2_co_check_locked(bs
, &result
,
1818 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1819 if (ret
< 0 || result
.check_errors
) {
1823 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1830 BdrvCheckResult result
= {0};
1831 qcow2_check_refcounts(bs
, &result
, 0);
1835 qemu_co_queue_init(&s
->thread_task_queue
);
1840 g_free(s
->image_data_file
);
1841 if (has_data_file(bs
)) {
1842 bdrv_unref_child(bs
, s
->data_file
);
1843 s
->data_file
= NULL
;
1845 g_free(s
->unknown_header_fields
);
1846 cleanup_unknown_header_ext(bs
);
1847 qcow2_free_snapshots(bs
);
1848 qcow2_refcount_close(bs
);
1849 qemu_vfree(s
->l1_table
);
1850 /* else pre-write overlap checks in cache_destroy may crash */
1852 cache_clean_timer_del(bs
);
1853 if (s
->l2_table_cache
) {
1854 qcow2_cache_destroy(s
->l2_table_cache
);
1856 if (s
->refcount_block_cache
) {
1857 qcow2_cache_destroy(s
->refcount_block_cache
);
1859 qcrypto_block_free(s
->crypto
);
1860 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1864 typedef struct QCow2OpenCo
{
1865 BlockDriverState
*bs
;
1872 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1874 QCow2OpenCo
*qoc
= opaque
;
1875 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1877 qemu_co_mutex_lock(&s
->lock
);
1878 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, qoc
->errp
);
1879 qemu_co_mutex_unlock(&s
->lock
);
1882 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1885 BDRVQcow2State
*s
= bs
->opaque
;
1894 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_of_bds
,
1895 BDRV_CHILD_IMAGE
, false, errp
);
1900 /* Initialise locks */
1901 qemu_co_mutex_init(&s
->lock
);
1903 if (qemu_in_coroutine()) {
1904 /* From bdrv_co_create. */
1905 qcow2_open_entry(&qoc
);
1907 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1908 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1909 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1914 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1916 BDRVQcow2State
*s
= bs
->opaque
;
1918 if (bs
->encrypted
) {
1919 /* Encryption works on a sector granularity */
1920 bs
->bl
.request_alignment
= qcrypto_block_get_sector_size(s
->crypto
);
1922 bs
->bl
.pwrite_zeroes_alignment
= s
->subcluster_size
;
1923 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1926 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1927 BlockReopenQueue
*queue
, Error
**errp
)
1929 Qcow2ReopenState
*r
;
1932 r
= g_new0(Qcow2ReopenState
, 1);
1935 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1936 state
->flags
, errp
);
1941 /* We need to write out any unwritten data if we reopen read-only. */
1942 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1943 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1948 ret
= bdrv_flush(state
->bs
);
1953 ret
= qcow2_mark_clean(state
->bs
);
1962 qcow2_update_options_abort(state
->bs
, r
);
1967 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1969 qcow2_update_options_commit(state
->bs
, state
->opaque
);
1970 g_free(state
->opaque
);
1973 static void qcow2_reopen_commit_post(BDRVReopenState
*state
)
1975 if (state
->flags
& BDRV_O_RDWR
) {
1976 Error
*local_err
= NULL
;
1978 if (qcow2_reopen_bitmaps_rw(state
->bs
, &local_err
) < 0) {
1980 * This is not fatal, bitmaps just left read-only, so all following
1981 * writes will fail. User can remove read-only bitmaps to unblock
1982 * writes or retry reopen.
1984 error_reportf_err(local_err
,
1985 "%s: Failed to make dirty bitmaps writable: ",
1986 bdrv_get_node_name(state
->bs
));
1991 static void qcow2_reopen_abort(BDRVReopenState
*state
)
1993 qcow2_update_options_abort(state
->bs
, state
->opaque
);
1994 g_free(state
->opaque
);
1997 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
1999 bool has_new_overlap_template
=
2000 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
2001 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2002 bool has_new_total_cache_size
=
2003 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
2004 bool has_all_cache_options
;
2006 /* New overlap template overrides all old overlap options */
2007 if (has_new_overlap_template
) {
2008 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
2009 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2010 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
2011 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
2012 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
2013 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
2014 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
2015 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
2016 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
2017 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
2020 /* New total cache size overrides all old options */
2021 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
2022 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
2023 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2026 qdict_join(options
, old_options
, false);
2029 * If after merging all cache size options are set, an old total size is
2030 * overwritten. Do keep all options, however, if all three are new. The
2031 * resulting error message is what we want to happen.
2033 has_all_cache_options
=
2034 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
2035 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
2036 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2038 if (has_all_cache_options
&& !has_new_total_cache_size
) {
2039 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
2043 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
2045 int64_t offset
, int64_t count
,
2046 int64_t *pnum
, int64_t *map
,
2047 BlockDriverState
**file
)
2049 BDRVQcow2State
*s
= bs
->opaque
;
2050 uint64_t host_offset
;
2052 QCow2SubclusterType type
;
2053 int ret
, status
= 0;
2055 qemu_co_mutex_lock(&s
->lock
);
2057 if (!s
->metadata_preallocation_checked
) {
2058 ret
= qcow2_detect_metadata_preallocation(bs
);
2059 s
->metadata_preallocation
= (ret
== 1);
2060 s
->metadata_preallocation_checked
= true;
2063 bytes
= MIN(INT_MAX
, count
);
2064 ret
= qcow2_get_host_offset(bs
, offset
, &bytes
, &host_offset
, &type
);
2065 qemu_co_mutex_unlock(&s
->lock
);
2072 if ((type
== QCOW2_SUBCLUSTER_NORMAL
||
2073 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2074 type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) && !s
->crypto
) {
2076 *file
= s
->data_file
->bs
;
2077 status
|= BDRV_BLOCK_OFFSET_VALID
;
2079 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2080 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
) {
2081 status
|= BDRV_BLOCK_ZERO
;
2082 } else if (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
2083 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) {
2084 status
|= BDRV_BLOCK_DATA
;
2086 if (s
->metadata_preallocation
&& (status
& BDRV_BLOCK_DATA
) &&
2087 (status
& BDRV_BLOCK_OFFSET_VALID
))
2089 status
|= BDRV_BLOCK_RECURSE
;
2094 static coroutine_fn
int qcow2_handle_l2meta(BlockDriverState
*bs
,
2095 QCowL2Meta
**pl2meta
,
2099 QCowL2Meta
*l2meta
= *pl2meta
;
2101 while (l2meta
!= NULL
) {
2105 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
2110 qcow2_alloc_cluster_abort(bs
, l2meta
);
2113 /* Take the request off the list of running requests */
2114 QLIST_REMOVE(l2meta
, next_in_flight
);
2116 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2118 next
= l2meta
->next
;
2127 static coroutine_fn
int
2128 qcow2_co_preadv_encrypted(BlockDriverState
*bs
,
2129 uint64_t host_offset
,
2133 uint64_t qiov_offset
)
2136 BDRVQcow2State
*s
= bs
->opaque
;
2139 assert(bs
->encrypted
&& s
->crypto
);
2140 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2143 * For encrypted images, read everything into a temporary
2144 * contiguous buffer on which the AES functions can work.
2145 * Also, decryption in a separate buffer is better as it
2146 * prevents the guest from learning information about the
2147 * encrypted nature of the virtual disk.
2150 buf
= qemu_try_blockalign(s
->data_file
->bs
, bytes
);
2155 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2156 ret
= bdrv_co_pread(s
->data_file
, host_offset
, bytes
, buf
, 0);
2161 if (qcow2_co_decrypt(bs
, host_offset
, offset
, buf
, bytes
) < 0)
2166 qemu_iovec_from_buf(qiov
, qiov_offset
, buf
, bytes
);
2174 typedef struct Qcow2AioTask
{
2177 BlockDriverState
*bs
;
2178 QCow2SubclusterType subcluster_type
; /* only for read */
2179 uint64_t host_offset
; /* or full descriptor in compressed clusters */
2183 uint64_t qiov_offset
;
2184 QCowL2Meta
*l2meta
; /* only for write */
2187 static coroutine_fn
int qcow2_co_preadv_task_entry(AioTask
*task
);
2188 static coroutine_fn
int qcow2_add_task(BlockDriverState
*bs
,
2191 QCow2SubclusterType subcluster_type
,
2192 uint64_t host_offset
,
2199 Qcow2AioTask local_task
;
2200 Qcow2AioTask
*task
= pool
? g_new(Qcow2AioTask
, 1) : &local_task
;
2202 *task
= (Qcow2AioTask
) {
2205 .subcluster_type
= subcluster_type
,
2207 .host_offset
= host_offset
,
2210 .qiov_offset
= qiov_offset
,
2214 trace_qcow2_add_task(qemu_coroutine_self(), bs
, pool
,
2215 func
== qcow2_co_preadv_task_entry
? "read" : "write",
2216 subcluster_type
, host_offset
, offset
, bytes
,
2220 return func(&task
->task
);
2223 aio_task_pool_start_task(pool
, &task
->task
);
2228 static coroutine_fn
int qcow2_co_preadv_task(BlockDriverState
*bs
,
2229 QCow2SubclusterType subc_type
,
2230 uint64_t host_offset
,
2231 uint64_t offset
, uint64_t bytes
,
2235 BDRVQcow2State
*s
= bs
->opaque
;
2237 switch (subc_type
) {
2238 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
2239 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
2240 /* Both zero types are handled in qcow2_co_preadv_part */
2241 g_assert_not_reached();
2243 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
2244 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
2245 assert(bs
->backing
); /* otherwise handled in qcow2_co_preadv_part */
2247 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
2248 return bdrv_co_preadv_part(bs
->backing
, offset
, bytes
,
2249 qiov
, qiov_offset
, 0);
2251 case QCOW2_SUBCLUSTER_COMPRESSED
:
2252 return qcow2_co_preadv_compressed(bs
, host_offset
,
2253 offset
, bytes
, qiov
, qiov_offset
);
2255 case QCOW2_SUBCLUSTER_NORMAL
:
2256 if (bs
->encrypted
) {
2257 return qcow2_co_preadv_encrypted(bs
, host_offset
,
2258 offset
, bytes
, qiov
, qiov_offset
);
2261 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2262 return bdrv_co_preadv_part(s
->data_file
, host_offset
,
2263 bytes
, qiov
, qiov_offset
, 0);
2266 g_assert_not_reached();
2269 g_assert_not_reached();
2272 static coroutine_fn
int qcow2_co_preadv_task_entry(AioTask
*task
)
2274 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2278 return qcow2_co_preadv_task(t
->bs
, t
->subcluster_type
,
2279 t
->host_offset
, t
->offset
, t
->bytes
,
2280 t
->qiov
, t
->qiov_offset
);
2283 static coroutine_fn
int qcow2_co_preadv_part(BlockDriverState
*bs
,
2284 uint64_t offset
, uint64_t bytes
,
2286 size_t qiov_offset
, int flags
)
2288 BDRVQcow2State
*s
= bs
->opaque
;
2290 unsigned int cur_bytes
; /* number of bytes in current iteration */
2291 uint64_t host_offset
= 0;
2292 QCow2SubclusterType type
;
2293 AioTaskPool
*aio
= NULL
;
2295 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2296 /* prepare next request */
2297 cur_bytes
= MIN(bytes
, INT_MAX
);
2299 cur_bytes
= MIN(cur_bytes
,
2300 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2303 qemu_co_mutex_lock(&s
->lock
);
2304 ret
= qcow2_get_host_offset(bs
, offset
, &cur_bytes
,
2305 &host_offset
, &type
);
2306 qemu_co_mutex_unlock(&s
->lock
);
2311 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2312 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2313 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&& !bs
->backing
) ||
2314 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&& !bs
->backing
))
2316 qemu_iovec_memset(qiov
, qiov_offset
, 0, cur_bytes
);
2318 if (!aio
&& cur_bytes
!= bytes
) {
2319 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2321 ret
= qcow2_add_task(bs
, aio
, qcow2_co_preadv_task_entry
, type
,
2322 host_offset
, offset
, cur_bytes
,
2323 qiov
, qiov_offset
, NULL
);
2330 offset
+= cur_bytes
;
2331 qiov_offset
+= cur_bytes
;
2336 aio_task_pool_wait_all(aio
);
2338 ret
= aio_task_pool_status(aio
);
2346 /* Check if it's possible to merge a write request with the writing of
2347 * the data from the COW regions */
2348 static bool merge_cow(uint64_t offset
, unsigned bytes
,
2349 QEMUIOVector
*qiov
, size_t qiov_offset
,
2354 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2355 /* If both COW regions are empty then there's nothing to merge */
2356 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
2360 /* If COW regions are handled already, skip this too */
2366 * The write request should start immediately after the first
2367 * COW region. This does not always happen because the area
2368 * touched by the request can be larger than the one defined
2369 * by @m (a single request can span an area consisting of a
2370 * mix of previously unallocated and allocated clusters, that
2371 * is why @l2meta is a list).
2373 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
2374 /* In this case the request starts before this region */
2375 assert(offset
< l2meta_cow_start(m
));
2376 assert(m
->cow_start
.nb_bytes
== 0);
2380 /* The write request should end immediately before the second
2381 * COW region (see above for why it does not always happen) */
2382 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
2383 assert(offset
+ bytes
> m
->offset
+ m
->cow_end
.offset
);
2384 assert(m
->cow_end
.nb_bytes
== 0);
2388 /* Make sure that adding both COW regions to the QEMUIOVector
2389 * does not exceed IOV_MAX */
2390 if (qemu_iovec_subvec_niov(qiov
, qiov_offset
, bytes
) > IOV_MAX
- 2) {
2394 m
->data_qiov
= qiov
;
2395 m
->data_qiov_offset
= qiov_offset
;
2403 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error.
2404 * Note that returning 0 does not guarantee non-zero data.
2406 static int is_zero_cow(BlockDriverState
*bs
, QCowL2Meta
*m
)
2409 * This check is designed for optimization shortcut so it must be
2411 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is
2412 * faster (but not as accurate and can result in false negatives).
2414 int ret
= bdrv_co_is_zero_fast(bs
, m
->offset
+ m
->cow_start
.offset
,
2415 m
->cow_start
.nb_bytes
);
2420 return bdrv_co_is_zero_fast(bs
, m
->offset
+ m
->cow_end
.offset
,
2421 m
->cow_end
.nb_bytes
);
2424 static int handle_alloc_space(BlockDriverState
*bs
, QCowL2Meta
*l2meta
)
2426 BDRVQcow2State
*s
= bs
->opaque
;
2429 if (!(s
->data_file
->bs
->supported_zero_flags
& BDRV_REQ_NO_FALLBACK
)) {
2433 if (bs
->encrypted
) {
2437 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2439 uint64_t start_offset
= m
->alloc_offset
+ m
->cow_start
.offset
;
2440 unsigned nb_bytes
= m
->cow_end
.offset
+ m
->cow_end
.nb_bytes
-
2441 m
->cow_start
.offset
;
2443 if (!m
->cow_start
.nb_bytes
&& !m
->cow_end
.nb_bytes
) {
2447 ret
= is_zero_cow(bs
, m
);
2450 } else if (ret
== 0) {
2455 * instead of writing zero COW buffers,
2456 * efficiently zero out the whole clusters
2459 ret
= qcow2_pre_write_overlap_check(bs
, 0, start_offset
, nb_bytes
,
2465 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC_SPACE
);
2466 ret
= bdrv_co_pwrite_zeroes(s
->data_file
, start_offset
, nb_bytes
,
2467 BDRV_REQ_NO_FALLBACK
);
2469 if (ret
!= -ENOTSUP
&& ret
!= -EAGAIN
) {
2475 trace_qcow2_skip_cow(qemu_coroutine_self(), m
->offset
, m
->nb_clusters
);
2482 * qcow2_co_pwritev_task
2483 * Called with s->lock unlocked
2484 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2485 * not use it somehow after qcow2_co_pwritev_task() call
2487 static coroutine_fn
int qcow2_co_pwritev_task(BlockDriverState
*bs
,
2488 uint64_t host_offset
,
2489 uint64_t offset
, uint64_t bytes
,
2491 uint64_t qiov_offset
,
2495 BDRVQcow2State
*s
= bs
->opaque
;
2496 void *crypt_buf
= NULL
;
2497 QEMUIOVector encrypted_qiov
;
2499 if (bs
->encrypted
) {
2501 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2502 crypt_buf
= qemu_try_blockalign(bs
->file
->bs
, bytes
);
2503 if (crypt_buf
== NULL
) {
2507 qemu_iovec_to_buf(qiov
, qiov_offset
, crypt_buf
, bytes
);
2509 if (qcow2_co_encrypt(bs
, host_offset
, offset
, crypt_buf
, bytes
) < 0) {
2514 qemu_iovec_init_buf(&encrypted_qiov
, crypt_buf
, bytes
);
2515 qiov
= &encrypted_qiov
;
2519 /* Try to efficiently initialize the physical space with zeroes */
2520 ret
= handle_alloc_space(bs
, l2meta
);
2526 * If we need to do COW, check if it's possible to merge the
2527 * writing of the guest data together with that of the COW regions.
2528 * If it's not possible (or not necessary) then write the
2531 if (!merge_cow(offset
, bytes
, qiov
, qiov_offset
, l2meta
)) {
2532 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2533 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset
);
2534 ret
= bdrv_co_pwritev_part(s
->data_file
, host_offset
,
2535 bytes
, qiov
, qiov_offset
, 0);
2541 qemu_co_mutex_lock(&s
->lock
);
2543 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
2547 qemu_co_mutex_lock(&s
->lock
);
2550 qcow2_handle_l2meta(bs
, &l2meta
, false);
2551 qemu_co_mutex_unlock(&s
->lock
);
2553 qemu_vfree(crypt_buf
);
2558 static coroutine_fn
int qcow2_co_pwritev_task_entry(AioTask
*task
)
2560 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2562 assert(!t
->subcluster_type
);
2564 return qcow2_co_pwritev_task(t
->bs
, t
->host_offset
,
2565 t
->offset
, t
->bytes
, t
->qiov
, t
->qiov_offset
,
2569 static coroutine_fn
int qcow2_co_pwritev_part(
2570 BlockDriverState
*bs
, uint64_t offset
, uint64_t bytes
,
2571 QEMUIOVector
*qiov
, size_t qiov_offset
, int flags
)
2573 BDRVQcow2State
*s
= bs
->opaque
;
2574 int offset_in_cluster
;
2576 unsigned int cur_bytes
; /* number of sectors in current iteration */
2577 uint64_t host_offset
;
2578 QCowL2Meta
*l2meta
= NULL
;
2579 AioTaskPool
*aio
= NULL
;
2581 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
2583 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2587 trace_qcow2_writev_start_part(qemu_coroutine_self());
2588 offset_in_cluster
= offset_into_cluster(s
, offset
);
2589 cur_bytes
= MIN(bytes
, INT_MAX
);
2590 if (bs
->encrypted
) {
2591 cur_bytes
= MIN(cur_bytes
,
2592 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
2593 - offset_in_cluster
);
2596 qemu_co_mutex_lock(&s
->lock
);
2598 ret
= qcow2_alloc_host_offset(bs
, offset
, &cur_bytes
,
2599 &host_offset
, &l2meta
);
2604 ret
= qcow2_pre_write_overlap_check(bs
, 0, host_offset
,
2610 qemu_co_mutex_unlock(&s
->lock
);
2612 if (!aio
&& cur_bytes
!= bytes
) {
2613 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2615 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_task_entry
, 0,
2616 host_offset
, offset
,
2617 cur_bytes
, qiov
, qiov_offset
, l2meta
);
2618 l2meta
= NULL
; /* l2meta is consumed by qcow2_co_pwritev_task() */
2624 offset
+= cur_bytes
;
2625 qiov_offset
+= cur_bytes
;
2626 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2630 qemu_co_mutex_lock(&s
->lock
);
2633 qcow2_handle_l2meta(bs
, &l2meta
, false);
2635 qemu_co_mutex_unlock(&s
->lock
);
2639 aio_task_pool_wait_all(aio
);
2641 ret
= aio_task_pool_status(aio
);
2646 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2651 static int qcow2_inactivate(BlockDriverState
*bs
)
2653 BDRVQcow2State
*s
= bs
->opaque
;
2654 int ret
, result
= 0;
2655 Error
*local_err
= NULL
;
2657 qcow2_store_persistent_dirty_bitmaps(bs
, true, &local_err
);
2658 if (local_err
!= NULL
) {
2660 error_reportf_err(local_err
, "Lost persistent bitmaps during "
2661 "inactivation of node '%s': ",
2662 bdrv_get_device_or_node_name(bs
));
2665 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2668 error_report("Failed to flush the L2 table cache: %s",
2672 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2675 error_report("Failed to flush the refcount block cache: %s",
2680 qcow2_mark_clean(bs
);
2686 static void qcow2_close(BlockDriverState
*bs
)
2688 BDRVQcow2State
*s
= bs
->opaque
;
2689 qemu_vfree(s
->l1_table
);
2690 /* else pre-write overlap checks in cache_destroy may crash */
2693 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2694 qcow2_inactivate(bs
);
2697 cache_clean_timer_del(bs
);
2698 qcow2_cache_destroy(s
->l2_table_cache
);
2699 qcow2_cache_destroy(s
->refcount_block_cache
);
2701 qcrypto_block_free(s
->crypto
);
2703 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
2705 g_free(s
->unknown_header_fields
);
2706 cleanup_unknown_header_ext(bs
);
2708 g_free(s
->image_data_file
);
2709 g_free(s
->image_backing_file
);
2710 g_free(s
->image_backing_format
);
2712 if (has_data_file(bs
)) {
2713 bdrv_unref_child(bs
, s
->data_file
);
2714 s
->data_file
= NULL
;
2717 qcow2_refcount_close(bs
);
2718 qcow2_free_snapshots(bs
);
2721 static void coroutine_fn
qcow2_co_invalidate_cache(BlockDriverState
*bs
,
2725 BDRVQcow2State
*s
= bs
->opaque
;
2726 int flags
= s
->flags
;
2727 QCryptoBlock
*crypto
= NULL
;
2732 * Backing files are read-only which makes all of their metadata immutable,
2733 * that means we don't have to worry about reopening them here.
2741 memset(s
, 0, sizeof(BDRVQcow2State
));
2742 options
= qdict_clone_shallow(bs
->options
);
2744 flags
&= ~BDRV_O_INACTIVE
;
2745 qemu_co_mutex_lock(&s
->lock
);
2746 ret
= qcow2_do_open(bs
, options
, flags
, errp
);
2747 qemu_co_mutex_unlock(&s
->lock
);
2748 qobject_unref(options
);
2750 error_prepend(errp
, "Could not reopen qcow2 layer: ");
2758 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2759 size_t len
, size_t buflen
)
2761 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2762 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2764 if (buflen
< ext_len
) {
2768 *ext_backing_fmt
= (QCowExtension
) {
2769 .magic
= cpu_to_be32(magic
),
2770 .len
= cpu_to_be32(len
),
2774 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2781 * Updates the qcow2 header, including the variable length parts of it, i.e.
2782 * the backing file name and all extensions. qcow2 was not designed to allow
2783 * such changes, so if we run out of space (we can only use the first cluster)
2784 * this function may fail.
2786 * Returns 0 on success, -errno in error cases.
2788 int qcow2_update_header(BlockDriverState
*bs
)
2790 BDRVQcow2State
*s
= bs
->opaque
;
2793 size_t buflen
= s
->cluster_size
;
2795 uint64_t total_size
;
2796 uint32_t refcount_table_clusters
;
2797 size_t header_length
;
2798 Qcow2UnknownHeaderExtension
*uext
;
2800 buf
= qemu_blockalign(bs
, buflen
);
2802 /* Header structure */
2803 header
= (QCowHeader
*) buf
;
2805 if (buflen
< sizeof(*header
)) {
2810 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2811 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2812 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2814 ret
= validate_compression_type(s
, NULL
);
2819 *header
= (QCowHeader
) {
2820 /* Version 2 fields */
2821 .magic
= cpu_to_be32(QCOW_MAGIC
),
2822 .version
= cpu_to_be32(s
->qcow_version
),
2823 .backing_file_offset
= 0,
2824 .backing_file_size
= 0,
2825 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2826 .size
= cpu_to_be64(total_size
),
2827 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2828 .l1_size
= cpu_to_be32(s
->l1_size
),
2829 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2830 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2831 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2832 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2833 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2835 /* Version 3 fields */
2836 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2837 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2838 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2839 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2840 .header_length
= cpu_to_be32(header_length
),
2841 .compression_type
= s
->compression_type
,
2844 /* For older versions, write a shorter header */
2845 switch (s
->qcow_version
) {
2847 ret
= offsetof(QCowHeader
, incompatible_features
);
2850 ret
= sizeof(*header
);
2859 memset(buf
, 0, buflen
);
2861 /* Preserve any unknown field in the header */
2862 if (s
->unknown_header_fields_size
) {
2863 if (buflen
< s
->unknown_header_fields_size
) {
2868 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2869 buf
+= s
->unknown_header_fields_size
;
2870 buflen
-= s
->unknown_header_fields_size
;
2873 /* Backing file format header extension */
2874 if (s
->image_backing_format
) {
2875 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2876 s
->image_backing_format
,
2877 strlen(s
->image_backing_format
),
2887 /* External data file header extension */
2888 if (has_data_file(bs
) && s
->image_data_file
) {
2889 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_DATA_FILE
,
2890 s
->image_data_file
, strlen(s
->image_data_file
),
2900 /* Full disk encryption header pointer extension */
2901 if (s
->crypto_header
.offset
!= 0) {
2902 s
->crypto_header
.offset
= cpu_to_be64(s
->crypto_header
.offset
);
2903 s
->crypto_header
.length
= cpu_to_be64(s
->crypto_header
.length
);
2904 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2905 &s
->crypto_header
, sizeof(s
->crypto_header
),
2907 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
2908 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
2917 * Feature table. A mere 8 feature names occupies 392 bytes, and
2918 * when coupled with the v3 minimum header of 104 bytes plus the
2919 * 8-byte end-of-extension marker, that would leave only 8 bytes
2920 * for a backing file name in an image with 512-byte clusters.
2921 * Thus, we choose to omit this header for cluster sizes 4k and
2924 if (s
->qcow_version
>= 3 && s
->cluster_size
> 4096) {
2925 static const Qcow2Feature features
[] = {
2927 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2928 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2929 .name
= "dirty bit",
2932 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2933 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
2934 .name
= "corrupt bit",
2937 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2938 .bit
= QCOW2_INCOMPAT_DATA_FILE_BITNR
,
2939 .name
= "external data file",
2942 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2943 .bit
= QCOW2_INCOMPAT_COMPRESSION_BITNR
,
2944 .name
= "compression type",
2947 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2948 .bit
= QCOW2_INCOMPAT_EXTL2_BITNR
,
2949 .name
= "extended L2 entries",
2952 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
2953 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
2954 .name
= "lazy refcounts",
2957 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
2958 .bit
= QCOW2_AUTOCLEAR_BITMAPS_BITNR
,
2962 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
2963 .bit
= QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR
,
2964 .name
= "raw external data",
2968 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
2969 features
, sizeof(features
), buflen
);
2977 /* Bitmap extension */
2978 if (s
->nb_bitmaps
> 0) {
2979 Qcow2BitmapHeaderExt bitmaps_header
= {
2980 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
2981 .bitmap_directory_size
=
2982 cpu_to_be64(s
->bitmap_directory_size
),
2983 .bitmap_directory_offset
=
2984 cpu_to_be64(s
->bitmap_directory_offset
)
2986 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
2987 &bitmaps_header
, sizeof(bitmaps_header
),
2996 /* Keep unknown header extensions */
2997 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
2998 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
3007 /* End of header extensions */
3008 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
3016 /* Backing file name */
3017 if (s
->image_backing_file
) {
3018 size_t backing_file_len
= strlen(s
->image_backing_file
);
3020 if (buflen
< backing_file_len
) {
3025 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3026 strncpy(buf
, s
->image_backing_file
, buflen
);
3028 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
3029 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
3032 /* Write the new header */
3033 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
3044 static int qcow2_change_backing_file(BlockDriverState
*bs
,
3045 const char *backing_file
, const char *backing_fmt
)
3047 BDRVQcow2State
*s
= bs
->opaque
;
3049 /* Adding a backing file means that the external data file alone won't be
3050 * enough to make sense of the content */
3051 if (backing_file
&& data_file_is_raw(bs
)) {
3055 if (backing_file
&& strlen(backing_file
) > 1023) {
3059 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3060 backing_file
?: "");
3061 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3062 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3064 g_free(s
->image_backing_file
);
3065 g_free(s
->image_backing_format
);
3067 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
3068 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
3070 return qcow2_update_header(bs
);
3073 static int qcow2_set_up_encryption(BlockDriverState
*bs
,
3074 QCryptoBlockCreateOptions
*cryptoopts
,
3077 BDRVQcow2State
*s
= bs
->opaque
;
3078 QCryptoBlock
*crypto
= NULL
;
3081 switch (cryptoopts
->format
) {
3082 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
3083 fmt
= QCOW_CRYPT_LUKS
;
3085 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
3086 fmt
= QCOW_CRYPT_AES
;
3089 error_setg(errp
, "Crypto format not supported in qcow2");
3093 s
->crypt_method_header
= fmt
;
3095 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
3096 qcow2_crypto_hdr_init_func
,
3097 qcow2_crypto_hdr_write_func
,
3103 ret
= qcow2_update_header(bs
);
3105 error_setg_errno(errp
, -ret
, "Could not write encryption header");
3111 qcrypto_block_free(crypto
);
3116 * Preallocates metadata structures for data clusters between @offset (in the
3117 * guest disk) and @new_length (which is thus generally the new guest disk
3120 * Returns: 0 on success, -errno on failure.
3122 static int coroutine_fn
preallocate_co(BlockDriverState
*bs
, uint64_t offset
,
3123 uint64_t new_length
, PreallocMode mode
,
3126 BDRVQcow2State
*s
= bs
->opaque
;
3128 uint64_t host_offset
= 0;
3129 int64_t file_length
;
3130 unsigned int cur_bytes
;
3132 QCowL2Meta
*meta
= NULL
, *m
;
3134 assert(offset
<= new_length
);
3135 bytes
= new_length
- offset
;
3138 cur_bytes
= MIN(bytes
, QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
));
3139 ret
= qcow2_alloc_host_offset(bs
, offset
, &cur_bytes
,
3140 &host_offset
, &meta
);
3142 error_setg_errno(errp
, -ret
, "Allocating clusters failed");
3146 for (m
= meta
; m
!= NULL
; m
= m
->next
) {
3150 ret
= qcow2_handle_l2meta(bs
, &meta
, true);
3152 error_setg_errno(errp
, -ret
, "Mapping clusters failed");
3156 /* TODO Preallocate data if requested */
3159 offset
+= cur_bytes
;
3163 * It is expected that the image file is large enough to actually contain
3164 * all of the allocated clusters (otherwise we get failing reads after
3165 * EOF). Extend the image to the last allocated sector.
3167 file_length
= bdrv_getlength(s
->data_file
->bs
);
3168 if (file_length
< 0) {
3169 error_setg_errno(errp
, -file_length
, "Could not get file size");
3174 if (host_offset
+ cur_bytes
> file_length
) {
3175 if (mode
== PREALLOC_MODE_METADATA
) {
3176 mode
= PREALLOC_MODE_OFF
;
3178 ret
= bdrv_co_truncate(s
->data_file
, host_offset
+ cur_bytes
, false,
3188 qcow2_handle_l2meta(bs
, &meta
, false);
3192 /* qcow2_refcount_metadata_size:
3193 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3194 * @cluster_size: size of a cluster, in bytes
3195 * @refcount_order: refcount bits power-of-2 exponent
3196 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3199 * Returns: Number of bytes required for refcount blocks and table metadata.
3201 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
3202 int refcount_order
, bool generous_increase
,
3203 uint64_t *refblock_count
)
3206 * Every host cluster is reference-counted, including metadata (even
3207 * refcount metadata is recursively included).
3209 * An accurate formula for the size of refcount metadata size is difficult
3210 * to derive. An easier method of calculation is finding the fixed point
3211 * where no further refcount blocks or table clusters are required to
3212 * reference count every cluster.
3214 int64_t blocks_per_table_cluster
= cluster_size
/ REFTABLE_ENTRY_SIZE
;
3215 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
3216 int64_t table
= 0; /* number of refcount table clusters */
3217 int64_t blocks
= 0; /* number of refcount block clusters */
3223 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
3224 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
3225 n
= clusters
+ blocks
+ table
;
3227 if (n
== last
&& generous_increase
) {
3228 clusters
+= DIV_ROUND_UP(table
, 2);
3229 n
= 0; /* force another loop */
3230 generous_increase
= false;
3232 } while (n
!= last
);
3234 if (refblock_count
) {
3235 *refblock_count
= blocks
;
3238 return (blocks
+ table
) * cluster_size
;
3242 * qcow2_calc_prealloc_size:
3243 * @total_size: virtual disk size in bytes
3244 * @cluster_size: cluster size in bytes
3245 * @refcount_order: refcount bits power-of-2 exponent
3246 * @extended_l2: true if the image has extended L2 entries
3248 * Returns: Total number of bytes required for the fully allocated image
3249 * (including metadata).
3251 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
3252 size_t cluster_size
,
3256 int64_t meta_size
= 0;
3257 uint64_t nl1e
, nl2e
;
3258 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
3259 size_t l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
3261 /* header: 1 cluster */
3262 meta_size
+= cluster_size
;
3264 /* total size of L2 tables */
3265 nl2e
= aligned_total_size
/ cluster_size
;
3266 nl2e
= ROUND_UP(nl2e
, cluster_size
/ l2e_size
);
3267 meta_size
+= nl2e
* l2e_size
;
3269 /* total size of L1 tables */
3270 nl1e
= nl2e
* l2e_size
/ cluster_size
;
3271 nl1e
= ROUND_UP(nl1e
, cluster_size
/ L1E_SIZE
);
3272 meta_size
+= nl1e
* L1E_SIZE
;
3274 /* total size of refcount table and blocks */
3275 meta_size
+= qcow2_refcount_metadata_size(
3276 (meta_size
+ aligned_total_size
) / cluster_size
,
3277 cluster_size
, refcount_order
, false, NULL
);
3279 return meta_size
+ aligned_total_size
;
3282 static bool validate_cluster_size(size_t cluster_size
, bool extended_l2
,
3285 int cluster_bits
= ctz32(cluster_size
);
3286 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
3287 (1 << cluster_bits
) != cluster_size
)
3289 error_setg(errp
, "Cluster size must be a power of two between %d and "
3290 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
3295 unsigned min_cluster_size
=
3296 (1 << MIN_CLUSTER_BITS
) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
;
3297 if (cluster_size
< min_cluster_size
) {
3298 error_setg(errp
, "Extended L2 entries are only supported with "
3299 "cluster sizes of at least %u bytes", min_cluster_size
);
3307 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, bool extended_l2
,
3310 size_t cluster_size
;
3312 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
3313 DEFAULT_CLUSTER_SIZE
);
3314 if (!validate_cluster_size(cluster_size
, extended_l2
, errp
)) {
3317 return cluster_size
;
3320 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
3325 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
3327 ret
= 3; /* default */
3328 } else if (!strcmp(buf
, "0.10")) {
3330 } else if (!strcmp(buf
, "1.1")) {
3333 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
3340 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
3343 uint64_t refcount_bits
;
3345 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
3346 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
3347 error_setg(errp
, "Refcount width must be a power of two and may not "
3352 if (version
< 3 && refcount_bits
!= 16) {
3353 error_setg(errp
, "Different refcount widths than 16 bits require "
3354 "compatibility level 1.1 or above (use compat=1.1 or "
3359 return refcount_bits
;
3362 static int coroutine_fn
3363 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
3365 BlockdevCreateOptionsQcow2
*qcow2_opts
;
3369 * Open the image file and write a minimal qcow2 header.
3371 * We keep things simple and start with a zero-sized image. We also
3372 * do without refcount blocks or a L1 table for now. We'll fix the
3373 * inconsistency later.
3375 * We do need a refcount table because growing the refcount table means
3376 * allocating two new refcount blocks - the second of which would be at
3377 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3378 * size for any qcow2 image.
3380 BlockBackend
*blk
= NULL
;
3381 BlockDriverState
*bs
= NULL
;
3382 BlockDriverState
*data_bs
= NULL
;
3384 size_t cluster_size
;
3387 uint64_t *refcount_table
;
3389 uint8_t compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
3391 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
3392 qcow2_opts
= &create_options
->u
.qcow2
;
3394 bs
= bdrv_open_blockdev_ref(qcow2_opts
->file
, errp
);
3399 /* Validate options and set default values */
3400 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
3401 error_setg(errp
, "Image size must be a multiple of %u bytes",
3402 (unsigned) BDRV_SECTOR_SIZE
);
3407 if (qcow2_opts
->has_version
) {
3408 switch (qcow2_opts
->version
) {
3409 case BLOCKDEV_QCOW2_VERSION_V2
:
3412 case BLOCKDEV_QCOW2_VERSION_V3
:
3416 g_assert_not_reached();
3422 if (qcow2_opts
->has_cluster_size
) {
3423 cluster_size
= qcow2_opts
->cluster_size
;
3425 cluster_size
= DEFAULT_CLUSTER_SIZE
;
3428 if (!qcow2_opts
->has_extended_l2
) {
3429 qcow2_opts
->extended_l2
= false;
3431 if (qcow2_opts
->extended_l2
) {
3433 error_setg(errp
, "Extended L2 entries are only supported with "
3434 "compatibility level 1.1 and above (use version=v3 or "
3441 if (!validate_cluster_size(cluster_size
, qcow2_opts
->extended_l2
, errp
)) {
3446 if (!qcow2_opts
->has_preallocation
) {
3447 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
3449 if (qcow2_opts
->has_backing_file
&&
3450 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
&&
3451 !qcow2_opts
->extended_l2
)
3453 error_setg(errp
, "Backing file and preallocation can only be used at "
3454 "the same time if extended_l2 is on");
3458 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->has_backing_file
) {
3459 error_setg(errp
, "Backing format cannot be used without backing file");
3464 if (!qcow2_opts
->has_lazy_refcounts
) {
3465 qcow2_opts
->lazy_refcounts
= false;
3467 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
3468 error_setg(errp
, "Lazy refcounts only supported with compatibility "
3469 "level 1.1 and above (use version=v3 or greater)");
3474 if (!qcow2_opts
->has_refcount_bits
) {
3475 qcow2_opts
->refcount_bits
= 16;
3477 if (qcow2_opts
->refcount_bits
> 64 ||
3478 !is_power_of_2(qcow2_opts
->refcount_bits
))
3480 error_setg(errp
, "Refcount width must be a power of two and may not "
3485 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
3486 error_setg(errp
, "Different refcount widths than 16 bits require "
3487 "compatibility level 1.1 or above (use version=v3 or "
3492 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
3494 if (qcow2_opts
->data_file_raw
&& !qcow2_opts
->data_file
) {
3495 error_setg(errp
, "data-file-raw requires data-file");
3499 if (qcow2_opts
->data_file_raw
&& qcow2_opts
->has_backing_file
) {
3500 error_setg(errp
, "Backing file and data-file-raw cannot be used at "
3506 if (qcow2_opts
->data_file
) {
3508 error_setg(errp
, "External data files are only supported with "
3509 "compatibility level 1.1 and above (use version=v3 or "
3514 data_bs
= bdrv_open_blockdev_ref(qcow2_opts
->data_file
, errp
);
3515 if (data_bs
== NULL
) {
3521 if (qcow2_opts
->has_compression_type
&&
3522 qcow2_opts
->compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3527 error_setg(errp
, "Non-zlib compression type is only supported with "
3528 "compatibility level 1.1 and above (use version=v3 or "
3533 switch (qcow2_opts
->compression_type
) {
3535 case QCOW2_COMPRESSION_TYPE_ZSTD
:
3539 error_setg(errp
, "Unknown compression type");
3543 compression_type
= qcow2_opts
->compression_type
;
3546 /* Create BlockBackend to write to the image */
3547 blk
= blk_new_with_bs(bs
, BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
,
3553 blk_set_allow_write_beyond_eof(blk
, true);
3555 /* Write the header */
3556 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
3557 header
= g_malloc0(cluster_size
);
3558 *header
= (QCowHeader
) {
3559 .magic
= cpu_to_be32(QCOW_MAGIC
),
3560 .version
= cpu_to_be32(version
),
3561 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
3562 .size
= cpu_to_be64(0),
3563 .l1_table_offset
= cpu_to_be64(0),
3564 .l1_size
= cpu_to_be32(0),
3565 .refcount_table_offset
= cpu_to_be64(cluster_size
),
3566 .refcount_table_clusters
= cpu_to_be32(1),
3567 .refcount_order
= cpu_to_be32(refcount_order
),
3568 /* don't deal with endianness since compression_type is 1 byte long */
3569 .compression_type
= compression_type
,
3570 .header_length
= cpu_to_be32(sizeof(*header
)),
3573 /* We'll update this to correct value later */
3574 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
3576 if (qcow2_opts
->lazy_refcounts
) {
3577 header
->compatible_features
|=
3578 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
3581 header
->incompatible_features
|=
3582 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE
);
3584 if (qcow2_opts
->data_file_raw
) {
3585 header
->autoclear_features
|=
3586 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW
);
3588 if (compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3589 header
->incompatible_features
|=
3590 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION
);
3593 if (qcow2_opts
->extended_l2
) {
3594 header
->incompatible_features
|=
3595 cpu_to_be64(QCOW2_INCOMPAT_EXTL2
);
3598 ret
= blk_pwrite(blk
, 0, header
, cluster_size
, 0);
3601 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
3605 /* Write a refcount table with one refcount block */
3606 refcount_table
= g_malloc0(2 * cluster_size
);
3607 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
3608 ret
= blk_pwrite(blk
, cluster_size
, refcount_table
, 2 * cluster_size
, 0);
3609 g_free(refcount_table
);
3612 error_setg_errno(errp
, -ret
, "Could not write refcount table");
3620 * And now open the image and make it consistent first (i.e. increase the
3621 * refcount of the cluster that is occupied by the header and the refcount
3624 options
= qdict_new();
3625 qdict_put_str(options
, "driver", "qcow2");
3626 qdict_put_str(options
, "file", bs
->node_name
);
3628 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3630 blk
= blk_new_open(NULL
, NULL
, options
,
3631 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
3638 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
3640 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
3641 "header and refcount table");
3644 } else if (ret
!= 0) {
3645 error_report("Huh, first cluster in empty image is already in use?");
3649 /* Set the external data file if necessary */
3651 BDRVQcow2State
*s
= blk_bs(blk
)->opaque
;
3652 s
->image_data_file
= g_strdup(data_bs
->filename
);
3655 /* Create a full header (including things like feature table) */
3656 ret
= qcow2_update_header(blk_bs(blk
));
3658 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
3662 /* Okay, now that we have a valid image, let's give it the right size */
3663 ret
= blk_truncate(blk
, qcow2_opts
->size
, false, qcow2_opts
->preallocation
,
3666 error_prepend(errp
, "Could not resize image: ");
3670 /* Want a backing file? There you go. */
3671 if (qcow2_opts
->has_backing_file
) {
3672 const char *backing_format
= NULL
;
3674 if (qcow2_opts
->has_backing_fmt
) {
3675 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
3678 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
3679 backing_format
, false);
3681 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
3682 "with format '%s'", qcow2_opts
->backing_file
,
3688 /* Want encryption? There you go. */
3689 if (qcow2_opts
->has_encrypt
) {
3690 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3699 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3700 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3701 * have to setup decryption context. We're not doing any I/O on the top
3702 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3705 options
= qdict_new();
3706 qdict_put_str(options
, "driver", "qcow2");
3707 qdict_put_str(options
, "file", bs
->node_name
);
3709 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3711 blk
= blk_new_open(NULL
, NULL
, options
,
3712 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3723 bdrv_unref(data_bs
);
3727 static int coroutine_fn
qcow2_co_create_opts(BlockDriver
*drv
,
3728 const char *filename
,
3732 BlockdevCreateOptions
*create_options
= NULL
;
3735 BlockDriverState
*bs
= NULL
;
3736 BlockDriverState
*data_bs
= NULL
;
3740 /* Only the keyval visitor supports the dotted syntax needed for
3741 * encryption, so go through a QDict before getting a QAPI type. Ignore
3742 * options meant for the protocol layer so that the visitor doesn't
3744 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3747 /* Handle encryption options */
3748 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3749 if (val
&& !strcmp(val
, "on")) {
3750 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3751 } else if (val
&& !strcmp(val
, "off")) {
3752 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3755 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3756 if (val
&& !strcmp(val
, "aes")) {
3757 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3760 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3761 * version=v2/v3 below. */
3762 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3763 if (val
&& !strcmp(val
, "0.10")) {
3764 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3765 } else if (val
&& !strcmp(val
, "1.1")) {
3766 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3769 /* Change legacy command line options into QMP ones */
3770 static const QDictRenames opt_renames
[] = {
3771 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3772 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3773 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3774 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3775 { BLOCK_OPT_EXTL2
, "extended-l2" },
3776 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3777 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3778 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3779 { BLOCK_OPT_DATA_FILE_RAW
, "data-file-raw" },
3780 { BLOCK_OPT_COMPRESSION_TYPE
, "compression-type" },
3784 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3789 /* Create and open the file (protocol layer) */
3790 ret
= bdrv_create_file(filename
, opts
, errp
);
3795 bs
= bdrv_open(filename
, NULL
, NULL
,
3796 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3802 /* Create and open an external data file (protocol layer) */
3803 val
= qdict_get_try_str(qdict
, BLOCK_OPT_DATA_FILE
);
3805 ret
= bdrv_create_file(val
, opts
, errp
);
3810 data_bs
= bdrv_open(val
, NULL
, NULL
,
3811 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
,
3813 if (data_bs
== NULL
) {
3818 qdict_del(qdict
, BLOCK_OPT_DATA_FILE
);
3819 qdict_put_str(qdict
, "data-file", data_bs
->node_name
);
3822 /* Set 'driver' and 'node' options */
3823 qdict_put_str(qdict
, "driver", "qcow2");
3824 qdict_put_str(qdict
, "file", bs
->node_name
);
3826 /* Now get the QAPI type BlockdevCreateOptions */
3827 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
3833 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, errp
);
3835 if (!create_options
) {
3840 /* Silently round up size */
3841 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3844 /* Create the qcow2 image (format layer) */
3845 ret
= qcow2_co_create(create_options
, errp
);
3848 bdrv_co_delete_file_noerr(bs
);
3849 bdrv_co_delete_file_noerr(data_bs
);
3854 qobject_unref(qdict
);
3856 bdrv_unref(data_bs
);
3857 qapi_free_BlockdevCreateOptions(create_options
);
3862 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3867 /* Clamp to image length, before checking status of underlying sectors */
3868 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3869 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3877 * bdrv_block_status_above doesn't merge different types of zeros, for
3878 * example, zeros which come from the region which is unallocated in
3879 * the whole backing chain, and zeros which come because of a short
3880 * backing file. So, we need a loop.
3883 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3886 } while (res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
&& bytes
);
3888 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && bytes
== 0;
3891 static coroutine_fn
int qcow2_co_pwrite_zeroes(BlockDriverState
*bs
,
3892 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3895 BDRVQcow2State
*s
= bs
->opaque
;
3897 uint32_t head
= offset_into_subcluster(s
, offset
);
3898 uint32_t tail
= ROUND_UP(offset
+ bytes
, s
->subcluster_size
) -
3901 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3902 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3909 QCow2SubclusterType type
;
3911 assert(head
+ bytes
+ tail
<= s
->subcluster_size
);
3913 /* check whether remainder of cluster already reads as zero */
3914 if (!(is_zero(bs
, offset
- head
, head
) &&
3915 is_zero(bs
, offset
+ bytes
, tail
))) {
3919 qemu_co_mutex_lock(&s
->lock
);
3920 /* We can have new write after previous check */
3922 bytes
= s
->subcluster_size
;
3923 nr
= s
->subcluster_size
;
3924 ret
= qcow2_get_host_offset(bs
, offset
, &nr
, &off
, &type
);
3926 (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
3927 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&&
3928 type
!= QCOW2_SUBCLUSTER_ZERO_PLAIN
&&
3929 type
!= QCOW2_SUBCLUSTER_ZERO_ALLOC
)) {
3930 qemu_co_mutex_unlock(&s
->lock
);
3931 return ret
< 0 ? ret
: -ENOTSUP
;
3934 qemu_co_mutex_lock(&s
->lock
);
3937 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
3939 /* Whatever is left can use real zero subclusters */
3940 ret
= qcow2_subcluster_zeroize(bs
, offset
, bytes
, flags
);
3941 qemu_co_mutex_unlock(&s
->lock
);
3946 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
3947 int64_t offset
, int bytes
)
3950 BDRVQcow2State
*s
= bs
->opaque
;
3952 /* If the image does not support QCOW_OFLAG_ZERO then discarding
3953 * clusters could expose stale data from the backing file. */
3954 if (s
->qcow_version
< 3 && bs
->backing
) {
3958 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
3959 assert(bytes
< s
->cluster_size
);
3960 /* Ignore partial clusters, except for the special case of the
3961 * complete partial cluster at the end of an unaligned file */
3962 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
3963 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3968 qemu_co_mutex_lock(&s
->lock
);
3969 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
3971 qemu_co_mutex_unlock(&s
->lock
);
3975 static int coroutine_fn
3976 qcow2_co_copy_range_from(BlockDriverState
*bs
,
3977 BdrvChild
*src
, uint64_t src_offset
,
3978 BdrvChild
*dst
, uint64_t dst_offset
,
3979 uint64_t bytes
, BdrvRequestFlags read_flags
,
3980 BdrvRequestFlags write_flags
)
3982 BDRVQcow2State
*s
= bs
->opaque
;
3984 unsigned int cur_bytes
; /* number of bytes in current iteration */
3985 BdrvChild
*child
= NULL
;
3986 BdrvRequestFlags cur_write_flags
;
3988 assert(!bs
->encrypted
);
3989 qemu_co_mutex_lock(&s
->lock
);
3991 while (bytes
!= 0) {
3992 uint64_t copy_offset
= 0;
3993 QCow2SubclusterType type
;
3994 /* prepare next request */
3995 cur_bytes
= MIN(bytes
, INT_MAX
);
3996 cur_write_flags
= write_flags
;
3998 ret
= qcow2_get_host_offset(bs
, src_offset
, &cur_bytes
,
3999 ©_offset
, &type
);
4005 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
4006 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
4007 if (bs
->backing
&& bs
->backing
->bs
) {
4008 int64_t backing_length
= bdrv_getlength(bs
->backing
->bs
);
4009 if (src_offset
>= backing_length
) {
4010 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4012 child
= bs
->backing
;
4013 cur_bytes
= MIN(cur_bytes
, backing_length
- src_offset
);
4014 copy_offset
= src_offset
;
4017 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4021 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
4022 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
4023 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4026 case QCOW2_SUBCLUSTER_COMPRESSED
:
4030 case QCOW2_SUBCLUSTER_NORMAL
:
4031 child
= s
->data_file
;
4037 qemu_co_mutex_unlock(&s
->lock
);
4038 ret
= bdrv_co_copy_range_from(child
,
4041 cur_bytes
, read_flags
, cur_write_flags
);
4042 qemu_co_mutex_lock(&s
->lock
);
4048 src_offset
+= cur_bytes
;
4049 dst_offset
+= cur_bytes
;
4054 qemu_co_mutex_unlock(&s
->lock
);
4058 static int coroutine_fn
4059 qcow2_co_copy_range_to(BlockDriverState
*bs
,
4060 BdrvChild
*src
, uint64_t src_offset
,
4061 BdrvChild
*dst
, uint64_t dst_offset
,
4062 uint64_t bytes
, BdrvRequestFlags read_flags
,
4063 BdrvRequestFlags write_flags
)
4065 BDRVQcow2State
*s
= bs
->opaque
;
4067 unsigned int cur_bytes
; /* number of sectors in current iteration */
4068 uint64_t host_offset
;
4069 QCowL2Meta
*l2meta
= NULL
;
4071 assert(!bs
->encrypted
);
4073 qemu_co_mutex_lock(&s
->lock
);
4075 while (bytes
!= 0) {
4079 cur_bytes
= MIN(bytes
, INT_MAX
);
4082 * If src->bs == dst->bs, we could simply copy by incrementing
4083 * the refcnt, without copying user data.
4084 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4085 ret
= qcow2_alloc_host_offset(bs
, dst_offset
, &cur_bytes
,
4086 &host_offset
, &l2meta
);
4091 ret
= qcow2_pre_write_overlap_check(bs
, 0, host_offset
, cur_bytes
,
4097 qemu_co_mutex_unlock(&s
->lock
);
4098 ret
= bdrv_co_copy_range_to(src
, src_offset
, s
->data_file
, host_offset
,
4099 cur_bytes
, read_flags
, write_flags
);
4100 qemu_co_mutex_lock(&s
->lock
);
4105 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
4111 src_offset
+= cur_bytes
;
4112 dst_offset
+= cur_bytes
;
4117 qcow2_handle_l2meta(bs
, &l2meta
, false);
4119 qemu_co_mutex_unlock(&s
->lock
);
4121 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
4126 static int coroutine_fn
qcow2_co_truncate(BlockDriverState
*bs
, int64_t offset
,
4127 bool exact
, PreallocMode prealloc
,
4128 BdrvRequestFlags flags
, Error
**errp
)
4130 BDRVQcow2State
*s
= bs
->opaque
;
4131 uint64_t old_length
;
4132 int64_t new_l1_size
;
4136 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
4137 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
4139 error_setg(errp
, "Unsupported preallocation mode '%s'",
4140 PreallocMode_str(prealloc
));
4144 if (!QEMU_IS_ALIGNED(offset
, BDRV_SECTOR_SIZE
)) {
4145 error_setg(errp
, "The new size must be a multiple of %u",
4146 (unsigned) BDRV_SECTOR_SIZE
);
4150 qemu_co_mutex_lock(&s
->lock
);
4153 * Even though we store snapshot size for all images, it was not
4154 * required until v3, so it is not safe to proceed for v2.
4156 if (s
->nb_snapshots
&& s
->qcow_version
< 3) {
4157 error_setg(errp
, "Can't resize a v2 image which has snapshots");
4162 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4163 if (qcow2_truncate_bitmaps_check(bs
, errp
)) {
4168 old_length
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4169 new_l1_size
= size_to_l1(s
, offset
);
4171 if (offset
< old_length
) {
4172 int64_t last_cluster
, old_file_size
;
4173 if (prealloc
!= PREALLOC_MODE_OFF
) {
4175 "Preallocation can't be used for shrinking an image");
4180 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
4181 old_length
- ROUND_UP(offset
,
4183 QCOW2_DISCARD_ALWAYS
, true);
4185 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
4189 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
4191 error_setg_errno(errp
, -ret
,
4192 "Failed to reduce the number of L2 tables");
4196 ret
= qcow2_shrink_reftable(bs
);
4198 error_setg_errno(errp
, -ret
,
4199 "Failed to discard unused refblocks");
4203 old_file_size
= bdrv_getlength(bs
->file
->bs
);
4204 if (old_file_size
< 0) {
4205 error_setg_errno(errp
, -old_file_size
,
4206 "Failed to inquire current file length");
4207 ret
= old_file_size
;
4210 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4211 if (last_cluster
< 0) {
4212 error_setg_errno(errp
, -last_cluster
,
4213 "Failed to find the last cluster");
4217 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
4218 Error
*local_err
= NULL
;
4221 * Do not pass @exact here: It will not help the user if
4222 * we get an error here just because they wanted to shrink
4223 * their qcow2 image (on a block device) with qemu-img.
4224 * (And on the qcow2 layer, the @exact requirement is
4225 * always fulfilled, so there is no need to pass it on.)
4227 bdrv_co_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
4228 false, PREALLOC_MODE_OFF
, 0, &local_err
);
4230 warn_reportf_err(local_err
,
4231 "Failed to truncate the tail of the image: ");
4235 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
4237 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
4243 case PREALLOC_MODE_OFF
:
4244 if (has_data_file(bs
)) {
4246 * If the caller wants an exact resize, the external data
4247 * file should be resized to the exact target size, too,
4248 * so we pass @exact here.
4250 ret
= bdrv_co_truncate(s
->data_file
, offset
, exact
, prealloc
, 0,
4258 case PREALLOC_MODE_METADATA
:
4259 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4265 case PREALLOC_MODE_FALLOC
:
4266 case PREALLOC_MODE_FULL
:
4268 int64_t allocation_start
, host_offset
, guest_offset
;
4269 int64_t clusters_allocated
;
4270 int64_t old_file_size
, last_cluster
, new_file_size
;
4271 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
4272 bool subclusters_need_allocation
= false;
4274 /* With a data file, preallocation means just allocating the metadata
4275 * and forwarding the truncate request to the data file */
4276 if (has_data_file(bs
)) {
4277 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4284 old_file_size
= bdrv_getlength(bs
->file
->bs
);
4285 if (old_file_size
< 0) {
4286 error_setg_errno(errp
, -old_file_size
,
4287 "Failed to inquire current file length");
4288 ret
= old_file_size
;
4292 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4293 if (last_cluster
>= 0) {
4294 old_file_size
= (last_cluster
+ 1) * s
->cluster_size
;
4296 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
4299 nb_new_data_clusters
= (ROUND_UP(offset
, s
->cluster_size
) -
4300 start_of_cluster(s
, old_length
)) >> s
->cluster_bits
;
4302 /* This is an overestimation; we will not actually allocate space for
4303 * these in the file but just make sure the new refcount structures are
4304 * able to cover them so we will not have to allocate new refblocks
4305 * while entering the data blocks in the potentially new L2 tables.
4306 * (We do not actually care where the L2 tables are placed. Maybe they
4307 * are already allocated or they can be placed somewhere before
4308 * @old_file_size. It does not matter because they will be fully
4309 * allocated automatically, so they do not need to be covered by the
4310 * preallocation. All that matters is that we will not have to allocate
4311 * new refcount structures for them.) */
4312 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
4313 s
->cluster_size
/ l2_entry_size(s
));
4314 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4315 * table for a potential head/tail */
4318 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
4319 nb_new_data_clusters
+
4322 if (allocation_start
< 0) {
4323 error_setg_errno(errp
, -allocation_start
,
4324 "Failed to resize refcount structures");
4325 ret
= allocation_start
;
4329 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
4330 nb_new_data_clusters
);
4331 if (clusters_allocated
< 0) {
4332 error_setg_errno(errp
, -clusters_allocated
,
4333 "Failed to allocate data clusters");
4334 ret
= clusters_allocated
;
4338 assert(clusters_allocated
== nb_new_data_clusters
);
4340 /* Allocate the data area */
4341 new_file_size
= allocation_start
+
4342 nb_new_data_clusters
* s
->cluster_size
;
4344 * Image file grows, so @exact does not matter.
4346 * If we need to zero out the new area, try first whether the protocol
4347 * driver can already take care of this.
4349 if (flags
& BDRV_REQ_ZERO_WRITE
) {
4350 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
,
4351 BDRV_REQ_ZERO_WRITE
, NULL
);
4353 flags
&= ~BDRV_REQ_ZERO_WRITE
;
4354 /* Ensure that we read zeroes and not backing file data */
4355 subclusters_need_allocation
= true;
4361 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
, 0,
4365 error_prepend(errp
, "Failed to resize underlying file: ");
4366 qcow2_free_clusters(bs
, allocation_start
,
4367 nb_new_data_clusters
* s
->cluster_size
,
4368 QCOW2_DISCARD_OTHER
);
4372 /* Create the necessary L2 entries */
4373 host_offset
= allocation_start
;
4374 guest_offset
= old_length
;
4375 while (nb_new_data_clusters
) {
4376 int64_t nb_clusters
= MIN(
4377 nb_new_data_clusters
,
4378 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
4379 unsigned cow_start_length
= offset_into_cluster(s
, guest_offset
);
4380 QCowL2Meta allocation
;
4381 guest_offset
= start_of_cluster(s
, guest_offset
);
4382 allocation
= (QCowL2Meta
) {
4383 .offset
= guest_offset
,
4384 .alloc_offset
= host_offset
,
4385 .nb_clusters
= nb_clusters
,
4388 .nb_bytes
= cow_start_length
,
4391 .offset
= nb_clusters
<< s
->cluster_bits
,
4394 .prealloc
= !subclusters_need_allocation
,
4396 qemu_co_queue_init(&allocation
.dependent_requests
);
4398 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
4400 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
4401 qcow2_free_clusters(bs
, host_offset
,
4402 nb_new_data_clusters
* s
->cluster_size
,
4403 QCOW2_DISCARD_OTHER
);
4407 guest_offset
+= nb_clusters
* s
->cluster_size
;
4408 host_offset
+= nb_clusters
* s
->cluster_size
;
4409 nb_new_data_clusters
-= nb_clusters
;
4415 g_assert_not_reached();
4418 if ((flags
& BDRV_REQ_ZERO_WRITE
) && offset
> old_length
) {
4419 uint64_t zero_start
= QEMU_ALIGN_UP(old_length
, s
->subcluster_size
);
4422 * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4423 * requires a subcluster-aligned start. The end may be unaligned if
4424 * it is at the end of the image (which it is here).
4426 if (offset
> zero_start
) {
4427 ret
= qcow2_subcluster_zeroize(bs
, zero_start
, offset
- zero_start
,
4430 error_setg_errno(errp
, -ret
, "Failed to zero out new clusters");
4435 /* Write explicit zeros for the unaligned head */
4436 if (zero_start
> old_length
) {
4437 uint64_t len
= MIN(zero_start
, offset
) - old_length
;
4438 uint8_t *buf
= qemu_blockalign0(bs
, len
);
4440 qemu_iovec_init_buf(&qiov
, buf
, len
);
4442 qemu_co_mutex_unlock(&s
->lock
);
4443 ret
= qcow2_co_pwritev_part(bs
, old_length
, len
, &qiov
, 0, 0);
4444 qemu_co_mutex_lock(&s
->lock
);
4448 error_setg_errno(errp
, -ret
, "Failed to zero out the new area");
4454 if (prealloc
!= PREALLOC_MODE_OFF
) {
4455 /* Flush metadata before actually changing the image size */
4456 ret
= qcow2_write_caches(bs
);
4458 error_setg_errno(errp
, -ret
,
4459 "Failed to flush the preallocated area to disk");
4464 bs
->total_sectors
= offset
/ BDRV_SECTOR_SIZE
;
4466 /* write updated header.size */
4467 offset
= cpu_to_be64(offset
);
4468 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
4469 &offset
, sizeof(offset
));
4471 error_setg_errno(errp
, -ret
, "Failed to update the image size");
4475 s
->l1_vm_state_index
= new_l1_size
;
4477 /* Update cache sizes */
4478 options
= qdict_clone_shallow(bs
->options
);
4479 ret
= qcow2_update_options(bs
, options
, s
->flags
, errp
);
4480 qobject_unref(options
);
4486 qemu_co_mutex_unlock(&s
->lock
);
4490 static coroutine_fn
int
4491 qcow2_co_pwritev_compressed_task(BlockDriverState
*bs
,
4492 uint64_t offset
, uint64_t bytes
,
4493 QEMUIOVector
*qiov
, size_t qiov_offset
)
4495 BDRVQcow2State
*s
= bs
->opaque
;
4498 uint8_t *buf
, *out_buf
;
4499 uint64_t cluster_offset
;
4501 assert(bytes
== s
->cluster_size
|| (bytes
< s
->cluster_size
&&
4502 (offset
+ bytes
== bs
->total_sectors
<< BDRV_SECTOR_BITS
)));
4504 buf
= qemu_blockalign(bs
, s
->cluster_size
);
4505 if (bytes
< s
->cluster_size
) {
4506 /* Zero-pad last write if image size is not cluster aligned */
4507 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
4509 qemu_iovec_to_buf(qiov
, qiov_offset
, buf
, bytes
);
4511 out_buf
= g_malloc(s
->cluster_size
);
4513 out_len
= qcow2_co_compress(bs
, out_buf
, s
->cluster_size
- 1,
4514 buf
, s
->cluster_size
);
4515 if (out_len
== -ENOMEM
) {
4516 /* could not compress: write normal cluster */
4517 ret
= qcow2_co_pwritev_part(bs
, offset
, bytes
, qiov
, qiov_offset
, 0);
4522 } else if (out_len
< 0) {
4527 qemu_co_mutex_lock(&s
->lock
);
4528 ret
= qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
,
4531 qemu_co_mutex_unlock(&s
->lock
);
4535 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
, true);
4536 qemu_co_mutex_unlock(&s
->lock
);
4541 BLKDBG_EVENT(s
->data_file
, BLKDBG_WRITE_COMPRESSED
);
4542 ret
= bdrv_co_pwrite(s
->data_file
, cluster_offset
, out_len
, out_buf
, 0);
4554 static coroutine_fn
int qcow2_co_pwritev_compressed_task_entry(AioTask
*task
)
4556 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
4558 assert(!t
->subcluster_type
&& !t
->l2meta
);
4560 return qcow2_co_pwritev_compressed_task(t
->bs
, t
->offset
, t
->bytes
, t
->qiov
,
4565 * XXX: put compressed sectors first, then all the cluster aligned
4566 * tables to avoid losing bytes in alignment
4568 static coroutine_fn
int
4569 qcow2_co_pwritev_compressed_part(BlockDriverState
*bs
,
4570 uint64_t offset
, uint64_t bytes
,
4571 QEMUIOVector
*qiov
, size_t qiov_offset
)
4573 BDRVQcow2State
*s
= bs
->opaque
;
4574 AioTaskPool
*aio
= NULL
;
4577 if (has_data_file(bs
)) {
4583 * align end of file to a sector boundary to ease reading with
4586 int64_t len
= bdrv_getlength(bs
->file
->bs
);
4590 return bdrv_co_truncate(bs
->file
, len
, false, PREALLOC_MODE_OFF
, 0,
4594 if (offset_into_cluster(s
, offset
)) {
4598 if (offset_into_cluster(s
, bytes
) &&
4599 (offset
+ bytes
) != (bs
->total_sectors
<< BDRV_SECTOR_BITS
)) {
4603 while (bytes
&& aio_task_pool_status(aio
) == 0) {
4604 uint64_t chunk_size
= MIN(bytes
, s
->cluster_size
);
4606 if (!aio
&& chunk_size
!= bytes
) {
4607 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
4610 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_compressed_task_entry
,
4611 0, 0, offset
, chunk_size
, qiov
, qiov_offset
, NULL
);
4615 qiov_offset
+= chunk_size
;
4616 offset
+= chunk_size
;
4617 bytes
-= chunk_size
;
4621 aio_task_pool_wait_all(aio
);
4623 ret
= aio_task_pool_status(aio
);
4631 static int coroutine_fn
4632 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
4633 uint64_t cluster_descriptor
,
4639 BDRVQcow2State
*s
= bs
->opaque
;
4640 int ret
= 0, csize
, nb_csectors
;
4642 uint8_t *buf
, *out_buf
;
4643 int offset_in_cluster
= offset_into_cluster(s
, offset
);
4645 coffset
= cluster_descriptor
& s
->cluster_offset_mask
;
4646 nb_csectors
= ((cluster_descriptor
>> s
->csize_shift
) & s
->csize_mask
) + 1;
4647 csize
= nb_csectors
* QCOW2_COMPRESSED_SECTOR_SIZE
-
4648 (coffset
& ~QCOW2_COMPRESSED_SECTOR_MASK
);
4650 buf
= g_try_malloc(csize
);
4655 out_buf
= qemu_blockalign(bs
, s
->cluster_size
);
4657 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
4658 ret
= bdrv_co_pread(bs
->file
, coffset
, csize
, buf
, 0);
4663 if (qcow2_co_decompress(bs
, out_buf
, s
->cluster_size
, buf
, csize
) < 0) {
4668 qemu_iovec_from_buf(qiov
, qiov_offset
, out_buf
+ offset_in_cluster
, bytes
);
4671 qemu_vfree(out_buf
);
4677 static int make_completely_empty(BlockDriverState
*bs
)
4679 BDRVQcow2State
*s
= bs
->opaque
;
4680 Error
*local_err
= NULL
;
4681 int ret
, l1_clusters
;
4683 uint64_t *new_reftable
= NULL
;
4684 uint64_t rt_entry
, l1_size2
;
4687 uint64_t reftable_offset
;
4688 uint32_t reftable_clusters
;
4689 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
4691 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
4696 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
4701 /* Refcounts will be broken utterly */
4702 ret
= qcow2_mark_dirty(bs
);
4707 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4709 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ L1E_SIZE
);
4710 l1_size2
= (uint64_t)s
->l1_size
* L1E_SIZE
;
4712 /* After this call, neither the in-memory nor the on-disk refcount
4713 * information accurately describe the actual references */
4715 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
4716 l1_clusters
* s
->cluster_size
, 0);
4718 goto fail_broken_refcounts
;
4720 memset(s
->l1_table
, 0, l1_size2
);
4722 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
4724 /* Overwrite enough clusters at the beginning of the sectors to place
4725 * the refcount table, a refcount block and the L1 table in; this may
4726 * overwrite parts of the existing refcount and L1 table, which is not
4727 * an issue because the dirty flag is set, complete data loss is in fact
4728 * desired and partial data loss is consequently fine as well */
4729 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
4730 (2 + l1_clusters
) * s
->cluster_size
, 0);
4731 /* This call (even if it failed overall) may have overwritten on-disk
4732 * refcount structures; in that case, the in-memory refcount information
4733 * will probably differ from the on-disk information which makes the BDS
4736 goto fail_broken_refcounts
;
4739 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4740 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
4742 /* "Create" an empty reftable (one cluster) directly after the image
4743 * header and an empty L1 table three clusters after the image header;
4744 * the cluster between those two will be used as the first refblock */
4745 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
4746 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
4747 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
4748 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
4749 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
4751 goto fail_broken_refcounts
;
4754 s
->l1_table_offset
= 3 * s
->cluster_size
;
4756 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ REFTABLE_ENTRY_SIZE
);
4757 if (!new_reftable
) {
4759 goto fail_broken_refcounts
;
4762 s
->refcount_table_offset
= s
->cluster_size
;
4763 s
->refcount_table_size
= s
->cluster_size
/ REFTABLE_ENTRY_SIZE
;
4764 s
->max_refcount_table_index
= 0;
4766 g_free(s
->refcount_table
);
4767 s
->refcount_table
= new_reftable
;
4768 new_reftable
= NULL
;
4770 /* Now the in-memory refcount information again corresponds to the on-disk
4771 * information (reftable is empty and no refblocks (the refblock cache is
4772 * empty)); however, this means some clusters (e.g. the image header) are
4773 * referenced, but not refcounted, but the normal qcow2 code assumes that
4774 * the in-memory information is always correct */
4776 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
4778 /* Enter the first refblock into the reftable */
4779 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
4780 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
4781 &rt_entry
, sizeof(rt_entry
));
4783 goto fail_broken_refcounts
;
4785 s
->refcount_table
[0] = 2 * s
->cluster_size
;
4787 s
->free_cluster_index
= 0;
4788 assert(3 + l1_clusters
<= s
->refcount_block_size
);
4789 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
4792 goto fail_broken_refcounts
;
4793 } else if (offset
> 0) {
4794 error_report("First cluster in emptied image is in use");
4798 /* Now finally the in-memory information corresponds to the on-disk
4799 * structures and is correct */
4800 ret
= qcow2_mark_clean(bs
);
4805 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
, false,
4806 PREALLOC_MODE_OFF
, 0, &local_err
);
4808 error_report_err(local_err
);
4814 fail_broken_refcounts
:
4815 /* The BDS is unusable at this point. If we wanted to make it usable, we
4816 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4817 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4818 * again. However, because the functions which could have caused this error
4819 * path to be taken are used by those functions as well, it's very likely
4820 * that that sequence will fail as well. Therefore, just eject the BDS. */
4824 g_free(new_reftable
);
4828 static int qcow2_make_empty(BlockDriverState
*bs
)
4830 BDRVQcow2State
*s
= bs
->opaque
;
4831 uint64_t offset
, end_offset
;
4832 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
4833 int l1_clusters
, ret
= 0;
4835 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ L1E_SIZE
);
4837 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
4838 3 + l1_clusters
<= s
->refcount_block_size
&&
4839 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
&&
4840 !has_data_file(bs
)) {
4841 /* The following function only works for qcow2 v3 images (it
4842 * requires the dirty flag) and only as long as there are no
4843 * features that reserve extra clusters (such as snapshots,
4844 * LUKS header, or persistent bitmaps), because it completely
4845 * empties the image. Furthermore, the L1 table and three
4846 * additional clusters (image header, refcount table, one
4847 * refcount block) have to fit inside one refcount block. It
4848 * only resets the image file, i.e. does not work with an
4849 * external data file. */
4850 return make_completely_empty(bs
);
4853 /* This fallback code simply discards every active cluster; this is slow,
4854 * but works in all cases */
4855 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4856 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
4857 /* As this function is generally used after committing an external
4858 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4859 * default action for this kind of discard is to pass the discard,
4860 * which will ideally result in an actually smaller image file, as
4861 * is probably desired. */
4862 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
4863 QCOW2_DISCARD_SNAPSHOT
, true);
4872 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
4874 BDRVQcow2State
*s
= bs
->opaque
;
4877 qemu_co_mutex_lock(&s
->lock
);
4878 ret
= qcow2_write_caches(bs
);
4879 qemu_co_mutex_unlock(&s
->lock
);
4884 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
4887 Error
*local_err
= NULL
;
4888 BlockMeasureInfo
*info
;
4889 uint64_t required
= 0; /* bytes that contribute to required size */
4890 uint64_t virtual_size
; /* disk size as seen by guest */
4891 uint64_t refcount_bits
;
4893 uint64_t luks_payload_size
= 0;
4894 size_t cluster_size
;
4897 PreallocMode prealloc
;
4898 bool has_backing_file
;
4903 /* Parse image creation options */
4904 extended_l2
= qemu_opt_get_bool_del(opts
, BLOCK_OPT_EXTL2
, false);
4906 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, extended_l2
,
4912 version
= qcow2_opt_get_version_del(opts
, &local_err
);
4917 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
4922 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
4923 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
4924 PREALLOC_MODE_OFF
, &local_err
);
4930 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
4931 has_backing_file
= !!optstr
;
4934 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
4935 has_luks
= optstr
&& strcmp(optstr
, "luks") == 0;
4939 g_autoptr(QCryptoBlockCreateOptions
) create_opts
= NULL
;
4940 QDict
*cryptoopts
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
4943 create_opts
= block_crypto_create_opts_init(cryptoopts
, errp
);
4944 qobject_unref(cryptoopts
);
4949 if (!qcrypto_block_calculate_payload_offset(create_opts
,
4956 luks_payload_size
= ROUND_UP(headerlen
, cluster_size
);
4959 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
4960 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
4962 /* Check that virtual disk size is valid */
4963 l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
4964 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
4965 cluster_size
/ l2e_size
);
4966 if (l2_tables
* L1E_SIZE
> QCOW_MAX_L1_SIZE
) {
4967 error_setg(&local_err
, "The image size is too large "
4968 "(try using a larger cluster size)");
4972 /* Account for input image */
4974 int64_t ssize
= bdrv_getlength(in_bs
);
4976 error_setg_errno(&local_err
, -ssize
,
4977 "Unable to get image virtual_size");
4981 virtual_size
= ROUND_UP(ssize
, cluster_size
);
4983 if (has_backing_file
) {
4984 /* We don't how much of the backing chain is shared by the input
4985 * image and the new image file. In the worst case the new image's
4986 * backing file has nothing in common with the input image. Be
4987 * conservative and assume all clusters need to be written.
4989 required
= virtual_size
;
4994 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
4997 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
4998 ssize
- offset
, &pnum
, NULL
,
5001 error_setg_errno(&local_err
, -ret
,
5002 "Unable to get block status");
5006 if (ret
& BDRV_BLOCK_ZERO
) {
5007 /* Skip zero regions (safe with no backing file) */
5008 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
5009 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
5010 /* Extend pnum to end of cluster for next iteration */
5011 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
5013 /* Count clusters we've seen */
5014 required
+= offset
% cluster_size
+ pnum
;
5020 /* Take into account preallocation. Nothing special is needed for
5021 * PREALLOC_MODE_METADATA since metadata is always counted.
5023 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
5024 required
= virtual_size
;
5027 info
= g_new0(BlockMeasureInfo
, 1);
5028 info
->fully_allocated
= luks_payload_size
+
5029 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
5030 ctz32(refcount_bits
), extended_l2
);
5033 * Remove data clusters that are not required. This overestimates the
5034 * required size because metadata needed for the fully allocated file is
5035 * still counted. Show bitmaps only if both source and destination
5036 * would support them.
5038 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
5039 info
->has_bitmaps
= version
>= 3 && in_bs
&&
5040 bdrv_supports_persistent_dirty_bitmap(in_bs
);
5041 if (info
->has_bitmaps
) {
5042 info
->bitmaps
= qcow2_get_persistent_dirty_bitmap_size(in_bs
,
5048 error_propagate(errp
, local_err
);
5052 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5054 BDRVQcow2State
*s
= bs
->opaque
;
5055 bdi
->cluster_size
= s
->cluster_size
;
5056 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
5060 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
,
5063 BDRVQcow2State
*s
= bs
->opaque
;
5064 ImageInfoSpecific
*spec_info
;
5065 QCryptoBlockInfo
*encrypt_info
= NULL
;
5067 if (s
->crypto
!= NULL
) {
5068 encrypt_info
= qcrypto_block_get_info(s
->crypto
, errp
);
5069 if (!encrypt_info
) {
5074 spec_info
= g_new(ImageInfoSpecific
, 1);
5075 *spec_info
= (ImageInfoSpecific
){
5076 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
5077 .u
.qcow2
.data
= g_new0(ImageInfoSpecificQCow2
, 1),
5079 if (s
->qcow_version
== 2) {
5080 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5081 .compat
= g_strdup("0.10"),
5082 .refcount_bits
= s
->refcount_bits
,
5084 } else if (s
->qcow_version
== 3) {
5085 Qcow2BitmapInfoList
*bitmaps
;
5086 if (!qcow2_get_bitmap_info_list(bs
, &bitmaps
, errp
)) {
5087 qapi_free_ImageInfoSpecific(spec_info
);
5088 qapi_free_QCryptoBlockInfo(encrypt_info
);
5091 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5092 .compat
= g_strdup("1.1"),
5093 .lazy_refcounts
= s
->compatible_features
&
5094 QCOW2_COMPAT_LAZY_REFCOUNTS
,
5095 .has_lazy_refcounts
= true,
5096 .corrupt
= s
->incompatible_features
&
5097 QCOW2_INCOMPAT_CORRUPT
,
5098 .has_corrupt
= true,
5099 .has_extended_l2
= true,
5100 .extended_l2
= has_subclusters(s
),
5101 .refcount_bits
= s
->refcount_bits
,
5102 .has_bitmaps
= !!bitmaps
,
5104 .has_data_file
= !!s
->image_data_file
,
5105 .data_file
= g_strdup(s
->image_data_file
),
5106 .has_data_file_raw
= has_data_file(bs
),
5107 .data_file_raw
= data_file_is_raw(bs
),
5108 .compression_type
= s
->compression_type
,
5111 /* if this assertion fails, this probably means a new version was
5112 * added without having it covered here */
5117 ImageInfoSpecificQCow2Encryption
*qencrypt
=
5118 g_new(ImageInfoSpecificQCow2Encryption
, 1);
5119 switch (encrypt_info
->format
) {
5120 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
5121 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
5123 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
5124 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
5125 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
5130 /* Since we did shallow copy above, erase any pointers
5131 * in the original info */
5132 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
5133 qapi_free_QCryptoBlockInfo(encrypt_info
);
5135 spec_info
->u
.qcow2
.data
->has_encrypt
= true;
5136 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
5142 static int qcow2_has_zero_init(BlockDriverState
*bs
)
5144 BDRVQcow2State
*s
= bs
->opaque
;
5147 if (qemu_in_coroutine()) {
5148 qemu_co_mutex_lock(&s
->lock
);
5151 * Check preallocation status: Preallocated images have all L2
5152 * tables allocated, nonpreallocated images have none. It is
5153 * therefore enough to check the first one.
5155 preallocated
= s
->l1_size
> 0 && s
->l1_table
[0] != 0;
5156 if (qemu_in_coroutine()) {
5157 qemu_co_mutex_unlock(&s
->lock
);
5160 if (!preallocated
) {
5162 } else if (bs
->encrypted
) {
5165 return bdrv_has_zero_init(s
->data_file
->bs
);
5169 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
5172 BDRVQcow2State
*s
= bs
->opaque
;
5174 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
5175 return bs
->drv
->bdrv_co_pwritev_part(bs
, qcow2_vm_state_offset(s
) + pos
,
5176 qiov
->size
, qiov
, 0, 0);
5179 static int qcow2_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
5182 BDRVQcow2State
*s
= bs
->opaque
;
5184 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
5185 return bs
->drv
->bdrv_co_preadv_part(bs
, qcow2_vm_state_offset(s
) + pos
,
5186 qiov
->size
, qiov
, 0, 0);
5190 * Downgrades an image's version. To achieve this, any incompatible features
5191 * have to be removed.
5193 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
5194 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5197 BDRVQcow2State
*s
= bs
->opaque
;
5198 int current_version
= s
->qcow_version
;
5202 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5203 assert(target_version
< current_version
);
5205 /* There are no other versions (now) that you can downgrade to */
5206 assert(target_version
== 2);
5208 if (s
->refcount_order
!= 4) {
5209 error_setg(errp
, "compat=0.10 requires refcount_bits=16");
5213 if (has_data_file(bs
)) {
5214 error_setg(errp
, "Cannot downgrade an image with a data file");
5219 * If any internal snapshot has a different size than the current
5220 * image size, or VM state size that exceeds 32 bits, downgrading
5221 * is unsafe. Even though we would still use v3-compliant output
5222 * to preserve that data, other v2 programs might not realize
5223 * those optional fields are important.
5225 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5226 if (s
->snapshots
[i
].vm_state_size
> UINT32_MAX
||
5227 s
->snapshots
[i
].disk_size
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
5228 error_setg(errp
, "Internal snapshots prevent downgrade of image");
5233 /* clear incompatible features */
5234 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
5235 ret
= qcow2_mark_clean(bs
);
5237 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5242 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5243 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5244 * best thing to do anyway */
5246 if (s
->incompatible_features
) {
5247 error_setg(errp
, "Cannot downgrade an image with incompatible features "
5248 "%#" PRIx64
" set", s
->incompatible_features
);
5252 /* since we can ignore compatible features, we can set them to 0 as well */
5253 s
->compatible_features
= 0;
5254 /* if lazy refcounts have been used, they have already been fixed through
5255 * clearing the dirty flag */
5257 /* clearing autoclear features is trivial */
5258 s
->autoclear_features
= 0;
5260 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
5262 error_setg_errno(errp
, -ret
, "Failed to turn zero into data clusters");
5266 s
->qcow_version
= target_version
;
5267 ret
= qcow2_update_header(bs
);
5269 s
->qcow_version
= current_version
;
5270 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5277 * Upgrades an image's version. While newer versions encompass all
5278 * features of older versions, some things may have to be presented
5281 static int qcow2_upgrade(BlockDriverState
*bs
, int target_version
,
5282 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5285 BDRVQcow2State
*s
= bs
->opaque
;
5286 bool need_snapshot_update
;
5287 int current_version
= s
->qcow_version
;
5291 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5292 assert(target_version
> current_version
);
5294 /* There are no other versions (yet) that you can upgrade to */
5295 assert(target_version
== 3);
5297 status_cb(bs
, 0, 2, cb_opaque
);
5300 * In v2, snapshots do not need to have extra data. v3 requires
5301 * the 64-bit VM state size and the virtual disk size to be
5303 * qcow2_write_snapshots() will always write the list in the
5304 * v3-compliant format.
5306 need_snapshot_update
= false;
5307 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5308 if (s
->snapshots
[i
].extra_data_size
<
5309 sizeof_field(QCowSnapshotExtraData
, vm_state_size_large
) +
5310 sizeof_field(QCowSnapshotExtraData
, disk_size
))
5312 need_snapshot_update
= true;
5316 if (need_snapshot_update
) {
5317 ret
= qcow2_write_snapshots(bs
);
5319 error_setg_errno(errp
, -ret
, "Failed to update the snapshot table");
5323 status_cb(bs
, 1, 2, cb_opaque
);
5325 s
->qcow_version
= target_version
;
5326 ret
= qcow2_update_header(bs
);
5328 s
->qcow_version
= current_version
;
5329 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5332 status_cb(bs
, 2, 2, cb_opaque
);
5337 typedef enum Qcow2AmendOperation
{
5338 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5339 * statically initialized to so that the helper CB can discern the first
5340 * invocation from an operation change */
5341 QCOW2_NO_OPERATION
= 0,
5344 QCOW2_UPDATING_ENCRYPTION
,
5345 QCOW2_CHANGING_REFCOUNT_ORDER
,
5347 } Qcow2AmendOperation
;
5349 typedef struct Qcow2AmendHelperCBInfo
{
5350 /* The code coordinating the amend operations should only modify
5351 * these four fields; the rest will be managed by the CB */
5352 BlockDriverAmendStatusCB
*original_status_cb
;
5353 void *original_cb_opaque
;
5355 Qcow2AmendOperation current_operation
;
5357 /* Total number of operations to perform (only set once) */
5358 int total_operations
;
5360 /* The following fields are managed by the CB */
5362 /* Number of operations completed */
5363 int operations_completed
;
5365 /* Cumulative offset of all completed operations */
5366 int64_t offset_completed
;
5368 Qcow2AmendOperation last_operation
;
5369 int64_t last_work_size
;
5370 } Qcow2AmendHelperCBInfo
;
5372 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
5373 int64_t operation_offset
,
5374 int64_t operation_work_size
, void *opaque
)
5376 Qcow2AmendHelperCBInfo
*info
= opaque
;
5377 int64_t current_work_size
;
5378 int64_t projected_work_size
;
5380 if (info
->current_operation
!= info
->last_operation
) {
5381 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
5382 info
->offset_completed
+= info
->last_work_size
;
5383 info
->operations_completed
++;
5386 info
->last_operation
= info
->current_operation
;
5389 assert(info
->total_operations
> 0);
5390 assert(info
->operations_completed
< info
->total_operations
);
5392 info
->last_work_size
= operation_work_size
;
5394 current_work_size
= info
->offset_completed
+ operation_work_size
;
5396 /* current_work_size is the total work size for (operations_completed + 1)
5397 * operations (which includes this one), so multiply it by the number of
5398 * operations not covered and divide it by the number of operations
5399 * covered to get a projection for the operations not covered */
5400 projected_work_size
= current_work_size
* (info
->total_operations
-
5401 info
->operations_completed
- 1)
5402 / (info
->operations_completed
+ 1);
5404 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
5405 current_work_size
+ projected_work_size
,
5406 info
->original_cb_opaque
);
5409 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5410 BlockDriverAmendStatusCB
*status_cb
,
5415 BDRVQcow2State
*s
= bs
->opaque
;
5416 int old_version
= s
->qcow_version
, new_version
= old_version
;
5417 uint64_t new_size
= 0;
5418 const char *backing_file
= NULL
, *backing_format
= NULL
, *data_file
= NULL
;
5419 bool lazy_refcounts
= s
->use_lazy_refcounts
;
5420 bool data_file_raw
= data_file_is_raw(bs
);
5421 const char *compat
= NULL
;
5422 int refcount_bits
= s
->refcount_bits
;
5424 QemuOptDesc
*desc
= opts
->list
->desc
;
5425 Qcow2AmendHelperCBInfo helper_cb_info
;
5426 bool encryption_update
= false;
5428 while (desc
&& desc
->name
) {
5429 if (!qemu_opt_find(opts
, desc
->name
)) {
5430 /* only change explicitly defined options */
5435 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
5436 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
5438 /* preserve default */
5439 } else if (!strcmp(compat
, "0.10") || !strcmp(compat
, "v2")) {
5441 } else if (!strcmp(compat
, "1.1") || !strcmp(compat
, "v3")) {
5444 error_setg(errp
, "Unknown compatibility level %s", compat
);
5447 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
5448 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
5449 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
5450 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5451 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
5452 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5453 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
5456 "Can't amend encryption options - encryption not present");
5459 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5461 "Only LUKS encryption options can be amended");
5464 encryption_update
= true;
5465 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
5466 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
5468 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
5469 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
5472 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
5473 !is_power_of_2(refcount_bits
))
5475 error_setg(errp
, "Refcount width must be a power of two and "
5476 "may not exceed 64 bits");
5479 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE
)) {
5480 data_file
= qemu_opt_get(opts
, BLOCK_OPT_DATA_FILE
);
5481 if (data_file
&& !has_data_file(bs
)) {
5482 error_setg(errp
, "data-file can only be set for images that "
5483 "use an external data file");
5486 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE_RAW
)) {
5487 data_file_raw
= qemu_opt_get_bool(opts
, BLOCK_OPT_DATA_FILE_RAW
,
5489 if (data_file_raw
&& !data_file_is_raw(bs
)) {
5490 error_setg(errp
, "data-file-raw cannot be set on existing "
5495 /* if this point is reached, this probably means a new option was
5496 * added without having it covered here */
5503 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
5504 .original_status_cb
= status_cb
,
5505 .original_cb_opaque
= cb_opaque
,
5506 .total_operations
= (new_version
!= old_version
)
5507 + (s
->refcount_bits
!= refcount_bits
) +
5508 (encryption_update
== true)
5511 /* Upgrade first (some features may require compat=1.1) */
5512 if (new_version
> old_version
) {
5513 helper_cb_info
.current_operation
= QCOW2_UPGRADING
;
5514 ret
= qcow2_upgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5515 &helper_cb_info
, errp
);
5521 if (encryption_update
) {
5522 QDict
*amend_opts_dict
;
5523 QCryptoBlockAmendOptions
*amend_opts
;
5525 helper_cb_info
.current_operation
= QCOW2_UPDATING_ENCRYPTION
;
5526 amend_opts_dict
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
5527 if (!amend_opts_dict
) {
5530 amend_opts
= block_crypto_amend_opts_init(amend_opts_dict
, errp
);
5531 qobject_unref(amend_opts_dict
);
5535 ret
= qcrypto_block_amend_options(s
->crypto
,
5536 qcow2_crypto_hdr_read_func
,
5537 qcow2_crypto_hdr_write_func
,
5542 qapi_free_QCryptoBlockAmendOptions(amend_opts
);
5548 if (s
->refcount_bits
!= refcount_bits
) {
5549 int refcount_order
= ctz32(refcount_bits
);
5551 if (new_version
< 3 && refcount_bits
!= 16) {
5552 error_setg(errp
, "Refcount widths other than 16 bits require "
5553 "compatibility level 1.1 or above (use compat=1.1 or "
5558 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
5559 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
5560 &qcow2_amend_helper_cb
,
5561 &helper_cb_info
, errp
);
5567 /* data-file-raw blocks backing files, so clear it first if requested */
5568 if (data_file_raw
) {
5569 s
->autoclear_features
|= QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5571 s
->autoclear_features
&= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5575 g_free(s
->image_data_file
);
5576 s
->image_data_file
= *data_file
? g_strdup(data_file
) : NULL
;
5579 ret
= qcow2_update_header(bs
);
5581 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5585 if (backing_file
|| backing_format
) {
5586 if (g_strcmp0(backing_file
, s
->image_backing_file
) ||
5587 g_strcmp0(backing_format
, s
->image_backing_format
)) {
5588 warn_report("Deprecated use of amend to alter the backing file; "
5589 "use qemu-img rebase instead");
5591 ret
= qcow2_change_backing_file(bs
,
5592 backing_file
?: s
->image_backing_file
,
5593 backing_format
?: s
->image_backing_format
);
5595 error_setg_errno(errp
, -ret
, "Failed to change the backing file");
5600 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
5601 if (lazy_refcounts
) {
5602 if (new_version
< 3) {
5603 error_setg(errp
, "Lazy refcounts only supported with "
5604 "compatibility level 1.1 and above (use compat=1.1 "
5608 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5609 ret
= qcow2_update_header(bs
);
5611 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5612 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5615 s
->use_lazy_refcounts
= true;
5617 /* make image clean first */
5618 ret
= qcow2_mark_clean(bs
);
5620 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5623 /* now disallow lazy refcounts */
5624 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5625 ret
= qcow2_update_header(bs
);
5627 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5628 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5631 s
->use_lazy_refcounts
= false;
5636 BlockBackend
*blk
= blk_new_with_bs(bs
, BLK_PERM_RESIZE
, BLK_PERM_ALL
,
5643 * Amending image options should ensure that the image has
5644 * exactly the given new values, so pass exact=true here.
5646 ret
= blk_truncate(blk
, new_size
, true, PREALLOC_MODE_OFF
, 0, errp
);
5653 /* Downgrade last (so unsupported features can be removed before) */
5654 if (new_version
< old_version
) {
5655 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
5656 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5657 &helper_cb_info
, errp
);
5666 static int coroutine_fn
qcow2_co_amend(BlockDriverState
*bs
,
5667 BlockdevAmendOptions
*opts
,
5671 BlockdevAmendOptionsQcow2
*qopts
= &opts
->u
.qcow2
;
5672 BDRVQcow2State
*s
= bs
->opaque
;
5675 if (qopts
->has_encrypt
) {
5677 error_setg(errp
, "image is not encrypted, can't amend");
5681 if (qopts
->encrypt
->format
!= Q_CRYPTO_BLOCK_FORMAT_LUKS
) {
5683 "Amend can't be used to change the qcow2 encryption format");
5687 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5689 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5693 ret
= qcrypto_block_amend_options(s
->crypto
,
5694 qcow2_crypto_hdr_read_func
,
5695 qcow2_crypto_hdr_write_func
,
5705 * If offset or size are negative, respectively, they will not be included in
5706 * the BLOCK_IMAGE_CORRUPTED event emitted.
5707 * fatal will be ignored for read-only BDS; corruptions found there will always
5708 * be considered non-fatal.
5710 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
5711 int64_t size
, const char *message_format
, ...)
5713 BDRVQcow2State
*s
= bs
->opaque
;
5714 const char *node_name
;
5718 fatal
= fatal
&& bdrv_is_writable(bs
);
5720 if (s
->signaled_corruption
&&
5721 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
5726 va_start(ap
, message_format
);
5727 message
= g_strdup_vprintf(message_format
, ap
);
5731 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
5732 "corruption events will be suppressed\n", message
);
5734 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
5735 "corruption events will be suppressed\n", message
);
5738 node_name
= bdrv_get_node_name(bs
);
5739 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
5740 *node_name
!= '\0', node_name
,
5741 message
, offset
>= 0, offset
,
5747 qcow2_mark_corrupt(bs
);
5748 bs
->drv
= NULL
; /* make BDS unusable */
5751 s
->signaled_corruption
= true;
5754 #define QCOW_COMMON_OPTIONS \
5756 .name = BLOCK_OPT_SIZE, \
5757 .type = QEMU_OPT_SIZE, \
5758 .help = "Virtual disk size" \
5761 .name = BLOCK_OPT_COMPAT_LEVEL, \
5762 .type = QEMU_OPT_STRING, \
5763 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
5766 .name = BLOCK_OPT_BACKING_FILE, \
5767 .type = QEMU_OPT_STRING, \
5768 .help = "File name of a base image" \
5771 .name = BLOCK_OPT_BACKING_FMT, \
5772 .type = QEMU_OPT_STRING, \
5773 .help = "Image format of the base image" \
5776 .name = BLOCK_OPT_DATA_FILE, \
5777 .type = QEMU_OPT_STRING, \
5778 .help = "File name of an external data file" \
5781 .name = BLOCK_OPT_DATA_FILE_RAW, \
5782 .type = QEMU_OPT_BOOL, \
5783 .help = "The external data file must stay valid " \
5787 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
5788 .type = QEMU_OPT_BOOL, \
5789 .help = "Postpone refcount updates", \
5790 .def_value_str = "off" \
5793 .name = BLOCK_OPT_REFCOUNT_BITS, \
5794 .type = QEMU_OPT_NUMBER, \
5795 .help = "Width of a reference count entry in bits", \
5796 .def_value_str = "16" \
5799 static QemuOptsList qcow2_create_opts
= {
5800 .name
= "qcow2-create-opts",
5801 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
5804 .name
= BLOCK_OPT_ENCRYPT
, \
5805 .type
= QEMU_OPT_BOOL
, \
5806 .help
= "Encrypt the image with format 'aes'. (Deprecated " \
5807 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)", \
5810 .name
= BLOCK_OPT_ENCRYPT_FORMAT
, \
5811 .type
= QEMU_OPT_STRING
, \
5812 .help
= "Encrypt the image, format choices: 'aes', 'luks'", \
5814 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
5815 "ID of secret providing qcow AES key or LUKS passphrase"), \
5816 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
5817 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
5818 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
5819 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
5820 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
5821 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
5823 .name
= BLOCK_OPT_CLUSTER_SIZE
, \
5824 .type
= QEMU_OPT_SIZE
, \
5825 .help
= "qcow2 cluster size", \
5826 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
) \
5829 .name
= BLOCK_OPT_EXTL2
, \
5830 .type
= QEMU_OPT_BOOL
, \
5831 .help
= "Extended L2 tables", \
5832 .def_value_str
= "off" \
5835 .name
= BLOCK_OPT_PREALLOC
, \
5836 .type
= QEMU_OPT_STRING
, \
5837 .help
= "Preallocation mode (allowed values: off, " \
5838 "metadata, falloc, full)" \
5841 .name
= BLOCK_OPT_COMPRESSION_TYPE
, \
5842 .type
= QEMU_OPT_STRING
, \
5843 .help
= "Compression method used for image cluster " \
5845 .def_value_str
= "zlib" \
5847 QCOW_COMMON_OPTIONS
,
5848 { /* end of list */ }
5852 static QemuOptsList qcow2_amend_opts
= {
5853 .name
= "qcow2-amend-opts",
5854 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts
.head
),
5856 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
5857 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
5858 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
5859 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
5860 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
5861 QCOW_COMMON_OPTIONS
,
5862 { /* end of list */ }
5866 static const char *const qcow2_strong_runtime_opts
[] = {
5867 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET
,
5872 BlockDriver bdrv_qcow2
= {
5873 .format_name
= "qcow2",
5874 .instance_size
= sizeof(BDRVQcow2State
),
5875 .bdrv_probe
= qcow2_probe
,
5876 .bdrv_open
= qcow2_open
,
5877 .bdrv_close
= qcow2_close
,
5878 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
5879 .bdrv_reopen_commit
= qcow2_reopen_commit
,
5880 .bdrv_reopen_commit_post
= qcow2_reopen_commit_post
,
5881 .bdrv_reopen_abort
= qcow2_reopen_abort
,
5882 .bdrv_join_options
= qcow2_join_options
,
5883 .bdrv_child_perm
= bdrv_default_perms
,
5884 .bdrv_co_create_opts
= qcow2_co_create_opts
,
5885 .bdrv_co_create
= qcow2_co_create
,
5886 .bdrv_has_zero_init
= qcow2_has_zero_init
,
5887 .bdrv_co_block_status
= qcow2_co_block_status
,
5889 .bdrv_co_preadv_part
= qcow2_co_preadv_part
,
5890 .bdrv_co_pwritev_part
= qcow2_co_pwritev_part
,
5891 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
5893 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
5894 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
5895 .bdrv_co_copy_range_from
= qcow2_co_copy_range_from
,
5896 .bdrv_co_copy_range_to
= qcow2_co_copy_range_to
,
5897 .bdrv_co_truncate
= qcow2_co_truncate
,
5898 .bdrv_co_pwritev_compressed_part
= qcow2_co_pwritev_compressed_part
,
5899 .bdrv_make_empty
= qcow2_make_empty
,
5901 .bdrv_snapshot_create
= qcow2_snapshot_create
,
5902 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
5903 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
5904 .bdrv_snapshot_list
= qcow2_snapshot_list
,
5905 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
5906 .bdrv_measure
= qcow2_measure
,
5907 .bdrv_get_info
= qcow2_get_info
,
5908 .bdrv_get_specific_info
= qcow2_get_specific_info
,
5910 .bdrv_save_vmstate
= qcow2_save_vmstate
,
5911 .bdrv_load_vmstate
= qcow2_load_vmstate
,
5914 .supports_backing
= true,
5915 .bdrv_change_backing_file
= qcow2_change_backing_file
,
5917 .bdrv_refresh_limits
= qcow2_refresh_limits
,
5918 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
5919 .bdrv_inactivate
= qcow2_inactivate
,
5921 .create_opts
= &qcow2_create_opts
,
5922 .amend_opts
= &qcow2_amend_opts
,
5923 .strong_runtime_opts
= qcow2_strong_runtime_opts
,
5924 .mutable_opts
= mutable_opts
,
5925 .bdrv_co_check
= qcow2_co_check
,
5926 .bdrv_amend_options
= qcow2_amend_options
,
5927 .bdrv_co_amend
= qcow2_co_amend
,
5929 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
5930 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
5932 .bdrv_supports_persistent_dirty_bitmap
=
5933 qcow2_supports_persistent_dirty_bitmap
,
5934 .bdrv_co_can_store_new_dirty_bitmap
= qcow2_co_can_store_new_dirty_bitmap
,
5935 .bdrv_co_remove_persistent_dirty_bitmap
=
5936 qcow2_co_remove_persistent_dirty_bitmap
,
5939 static void bdrv_qcow2_init(void)
5941 bdrv_register(&bdrv_qcow2
);
5944 block_init(bdrv_qcow2_init
);