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 "qemu/memalign.h"
42 #include "qapi/qobject-input-visitor.h"
43 #include "qapi/qapi-visit-block-core.h"
45 #include "block/aio_task.h"
46 #include "block/dirty-bitmap.h"
49 Differences with QCOW:
51 - Support for multiple incremental snapshots.
52 - Memory management by reference counts.
53 - Clusters which have a reference count of one have the bit
54 QCOW_OFLAG_COPIED to optimize write performance.
55 - Size of compressed clusters is stored in sectors to reduce bit usage
56 in the cluster offsets.
57 - Support for storing additional data (such as the VM state) in the
59 - If a backing store is used, the cluster size is not constrained
60 (could be backported to QCOW).
61 - L2 tables have always a size of one cluster.
68 } QEMU_PACKED QCowExtension
;
70 #define QCOW2_EXT_MAGIC_END 0
71 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca
72 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
73 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
74 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
75 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
77 static int coroutine_fn
78 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
85 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
87 const QCowHeader
*cow_header
= (const void *)buf
;
89 if (buf_size
>= sizeof(QCowHeader
) &&
90 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
91 be32_to_cpu(cow_header
->version
) >= 2)
98 static int qcow2_crypto_hdr_read_func(QCryptoBlock
*block
, size_t offset
,
99 uint8_t *buf
, size_t buflen
,
100 void *opaque
, Error
**errp
)
102 BlockDriverState
*bs
= opaque
;
103 BDRVQcow2State
*s
= bs
->opaque
;
106 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
107 error_setg(errp
, "Request for data outside of extension header");
111 ret
= bdrv_pread(bs
->file
, s
->crypto_header
.offset
+ offset
, buflen
, buf
,
114 error_setg_errno(errp
, -ret
, "Could not read encryption header");
121 static int qcow2_crypto_hdr_init_func(QCryptoBlock
*block
, size_t headerlen
,
122 void *opaque
, Error
**errp
)
124 BlockDriverState
*bs
= opaque
;
125 BDRVQcow2State
*s
= bs
->opaque
;
129 ret
= qcow2_alloc_clusters(bs
, headerlen
);
131 error_setg_errno(errp
, -ret
,
132 "Cannot allocate cluster for LUKS header size %zu",
137 s
->crypto_header
.length
= headerlen
;
138 s
->crypto_header
.offset
= ret
;
141 * Zero fill all space in cluster so it has predictable
142 * content, as we may not initialize some regions of the
143 * header (eg only 1 out of 8 key slots will be initialized)
145 clusterlen
= size_to_clusters(s
, headerlen
) * s
->cluster_size
;
146 assert(qcow2_pre_write_overlap_check(bs
, 0, ret
, clusterlen
, false) == 0);
147 ret
= bdrv_pwrite_zeroes(bs
->file
,
151 error_setg_errno(errp
, -ret
, "Could not zero fill encryption header");
159 static int qcow2_crypto_hdr_write_func(QCryptoBlock
*block
, size_t offset
,
160 const uint8_t *buf
, size_t buflen
,
161 void *opaque
, Error
**errp
)
163 BlockDriverState
*bs
= opaque
;
164 BDRVQcow2State
*s
= bs
->opaque
;
167 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
168 error_setg(errp
, "Request for data outside of extension header");
172 ret
= bdrv_pwrite(bs
->file
, s
->crypto_header
.offset
+ offset
, buflen
, buf
,
175 error_setg_errno(errp
, -ret
, "Could not read encryption header");
182 qcow2_extract_crypto_opts(QemuOpts
*opts
, const char *fmt
, Error
**errp
)
184 QDict
*cryptoopts_qdict
;
187 /* Extract "encrypt." options into a qdict */
188 opts_qdict
= qemu_opts_to_qdict(opts
, NULL
);
189 qdict_extract_subqdict(opts_qdict
, &cryptoopts_qdict
, "encrypt.");
190 qobject_unref(opts_qdict
);
191 qdict_put_str(cryptoopts_qdict
, "format", fmt
);
192 return cryptoopts_qdict
;
196 * read qcow2 extension and fill bs
197 * start reading from start_offset
198 * finish reading upon magic of value 0 or when end_offset reached
199 * unknown magic is skipped (future extension this version knows nothing about)
200 * return 0 upon success, non-0 otherwise
202 static int coroutine_fn GRAPH_RDLOCK
203 qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
204 uint64_t end_offset
, void **p_feature_table
,
205 int flags
, bool *need_update_header
, Error
**errp
)
207 BDRVQcow2State
*s
= bs
->opaque
;
211 Qcow2BitmapHeaderExt bitmaps_ext
;
213 if (need_update_header
!= NULL
) {
214 *need_update_header
= false;
218 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
220 offset
= start_offset
;
221 while (offset
< end_offset
) {
225 if (offset
> s
->cluster_size
)
226 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
228 printf("attempting to read extended header in offset %lu\n", offset
);
231 ret
= bdrv_co_pread(bs
->file
, offset
, sizeof(ext
), &ext
, 0);
233 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
234 "pread fail from offset %" PRIu64
, offset
);
237 ext
.magic
= be32_to_cpu(ext
.magic
);
238 ext
.len
= be32_to_cpu(ext
.len
);
239 offset
+= sizeof(ext
);
241 printf("ext.magic = 0x%x\n", ext
.magic
);
243 if (offset
> end_offset
|| ext
.len
> end_offset
- offset
) {
244 error_setg(errp
, "Header extension too large");
249 case QCOW2_EXT_MAGIC_END
:
252 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
253 if (ext
.len
>= sizeof(bs
->backing_format
)) {
254 error_setg(errp
, "ERROR: ext_backing_format: len=%" PRIu32
255 " too large (>=%zu)", ext
.len
,
256 sizeof(bs
->backing_format
));
259 ret
= bdrv_co_pread(bs
->file
, offset
, ext
.len
, bs
->backing_format
, 0);
261 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
262 "Could not read format name");
265 bs
->backing_format
[ext
.len
] = '\0';
266 s
->image_backing_format
= g_strdup(bs
->backing_format
);
268 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
272 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
273 if (p_feature_table
!= NULL
) {
274 void *feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
275 ret
= bdrv_co_pread(bs
->file
, offset
, ext
.len
, feature_table
, 0);
277 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
278 "Could not read table");
279 g_free(feature_table
);
283 *p_feature_table
= feature_table
;
287 case QCOW2_EXT_MAGIC_CRYPTO_HEADER
: {
288 unsigned int cflags
= 0;
289 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
290 error_setg(errp
, "CRYPTO header extension only "
291 "expected with LUKS encryption method");
294 if (ext
.len
!= sizeof(Qcow2CryptoHeaderExtension
)) {
295 error_setg(errp
, "CRYPTO header extension size %u, "
296 "but expected size %zu", ext
.len
,
297 sizeof(Qcow2CryptoHeaderExtension
));
301 ret
= bdrv_co_pread(bs
->file
, offset
, ext
.len
, &s
->crypto_header
, 0);
303 error_setg_errno(errp
, -ret
,
304 "Unable to read CRYPTO header extension");
307 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
308 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
310 if ((s
->crypto_header
.offset
% s
->cluster_size
) != 0) {
311 error_setg(errp
, "Encryption header offset '%" PRIu64
"' is "
312 "not a multiple of cluster size '%u'",
313 s
->crypto_header
.offset
, s
->cluster_size
);
317 if (flags
& BDRV_O_NO_IO
) {
318 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
320 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
321 qcow2_crypto_hdr_read_func
,
322 bs
, cflags
, QCOW2_MAX_THREADS
, errp
);
328 case QCOW2_EXT_MAGIC_BITMAPS
:
329 if (ext
.len
!= sizeof(bitmaps_ext
)) {
330 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
331 "Invalid extension length");
335 if (!(s
->autoclear_features
& QCOW2_AUTOCLEAR_BITMAPS
)) {
336 if (s
->qcow_version
< 3) {
337 /* Let's be a bit more specific */
338 warn_report("This qcow2 v2 image contains bitmaps, but "
339 "they may have been modified by a program "
340 "without persistent bitmap support; so now "
341 "they must all be considered inconsistent");
343 warn_report("a program lacking bitmap support "
344 "modified this file, so all bitmaps are now "
345 "considered inconsistent");
347 error_printf("Some clusters may be leaked, "
348 "run 'qemu-img check -r' on the image "
350 if (need_update_header
!= NULL
) {
351 /* Updating is needed to drop invalid bitmap extension. */
352 *need_update_header
= true;
357 ret
= bdrv_co_pread(bs
->file
, offset
, ext
.len
, &bitmaps_ext
, 0);
359 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
360 "Could not read ext header");
364 if (bitmaps_ext
.reserved32
!= 0) {
365 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
366 "Reserved field is not zero");
370 bitmaps_ext
.nb_bitmaps
= be32_to_cpu(bitmaps_ext
.nb_bitmaps
);
371 bitmaps_ext
.bitmap_directory_size
=
372 be64_to_cpu(bitmaps_ext
.bitmap_directory_size
);
373 bitmaps_ext
.bitmap_directory_offset
=
374 be64_to_cpu(bitmaps_ext
.bitmap_directory_offset
);
376 if (bitmaps_ext
.nb_bitmaps
> QCOW2_MAX_BITMAPS
) {
378 "bitmaps_ext: Image has %" PRIu32
" bitmaps, "
379 "exceeding the QEMU supported maximum of %d",
380 bitmaps_ext
.nb_bitmaps
, QCOW2_MAX_BITMAPS
);
384 if (bitmaps_ext
.nb_bitmaps
== 0) {
385 error_setg(errp
, "found bitmaps extension with zero bitmaps");
389 if (offset_into_cluster(s
, bitmaps_ext
.bitmap_directory_offset
)) {
390 error_setg(errp
, "bitmaps_ext: "
391 "invalid bitmap directory offset");
395 if (bitmaps_ext
.bitmap_directory_size
>
396 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
) {
397 error_setg(errp
, "bitmaps_ext: "
398 "bitmap directory size (%" PRIu64
") exceeds "
399 "the maximum supported size (%d)",
400 bitmaps_ext
.bitmap_directory_size
,
401 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
);
405 s
->nb_bitmaps
= bitmaps_ext
.nb_bitmaps
;
406 s
->bitmap_directory_offset
=
407 bitmaps_ext
.bitmap_directory_offset
;
408 s
->bitmap_directory_size
=
409 bitmaps_ext
.bitmap_directory_size
;
412 printf("Qcow2: Got bitmaps extension: "
413 "offset=%" PRIu64
" nb_bitmaps=%" PRIu32
"\n",
414 s
->bitmap_directory_offset
, s
->nb_bitmaps
);
418 case QCOW2_EXT_MAGIC_DATA_FILE
:
420 s
->image_data_file
= g_malloc0(ext
.len
+ 1);
421 ret
= bdrv_co_pread(bs
->file
, offset
, ext
.len
, s
->image_data_file
, 0);
423 error_setg_errno(errp
, -ret
,
424 "ERROR: Could not read data file name");
428 printf("Qcow2: Got external data file %s\n", s
->image_data_file
);
434 /* unknown magic - save it in case we need to rewrite the header */
435 /* If you add a new feature, make sure to also update the fast
436 * path of qcow2_make_empty() to deal with it. */
438 Qcow2UnknownHeaderExtension
*uext
;
440 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
441 uext
->magic
= ext
.magic
;
443 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
445 ret
= bdrv_co_pread(bs
->file
, offset
, uext
->len
, uext
->data
, 0);
447 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
448 "Could not read data");
455 offset
+= ((ext
.len
+ 7) & ~7);
461 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
463 BDRVQcow2State
*s
= bs
->opaque
;
464 Qcow2UnknownHeaderExtension
*uext
, *next
;
466 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
467 QLIST_REMOVE(uext
, next
);
472 static void report_unsupported_feature(Error
**errp
, Qcow2Feature
*table
,
475 g_autoptr(GString
) features
= g_string_sized_new(60);
477 while (table
&& table
->name
[0] != '\0') {
478 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
479 if (mask
& (1ULL << table
->bit
)) {
480 if (features
->len
> 0) {
481 g_string_append(features
, ", ");
483 g_string_append_printf(features
, "%.46s", table
->name
);
484 mask
&= ~(1ULL << table
->bit
);
491 if (features
->len
> 0) {
492 g_string_append(features
, ", ");
494 g_string_append_printf(features
,
495 "Unknown incompatible feature: %" PRIx64
, mask
);
498 error_setg(errp
, "Unsupported qcow2 feature(s): %s", features
->str
);
502 * Sets the dirty bit and flushes afterwards if necessary.
504 * The incompatible_features bit is only set if the image file header was
505 * updated successfully. Therefore it is not required to check the return
506 * value of this function.
508 int qcow2_mark_dirty(BlockDriverState
*bs
)
510 BDRVQcow2State
*s
= bs
->opaque
;
514 assert(s
->qcow_version
>= 3);
516 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
517 return 0; /* already dirty */
520 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
521 ret
= bdrv_pwrite_sync(bs
->file
,
522 offsetof(QCowHeader
, incompatible_features
),
523 sizeof(val
), &val
, 0);
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 GRAPH_RDLOCK
605 qcow2_co_check_locked(BlockDriverState
*bs
, 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 GRAPH_RDLOCK
644 qcow2_co_check(BlockDriverState
*bs
, 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
=
844 aio_timer_new_with_attrs(context
, QEMU_CLOCK_VIRTUAL
,
845 SCALE_MS
, QEMU_TIMER_ATTR_EXTERNAL
,
846 cache_clean_timer_cb
, bs
);
847 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
848 (int64_t) s
->cache_clean_interval
* 1000);
852 static void cache_clean_timer_del(BlockDriverState
*bs
)
854 BDRVQcow2State
*s
= bs
->opaque
;
855 if (s
->cache_clean_timer
) {
856 timer_free(s
->cache_clean_timer
);
857 s
->cache_clean_timer
= NULL
;
861 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
863 cache_clean_timer_del(bs
);
866 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
867 AioContext
*new_context
)
869 cache_clean_timer_init(bs
, new_context
);
872 static bool read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
873 uint64_t *l2_cache_size
,
874 uint64_t *l2_cache_entry_size
,
875 uint64_t *refcount_cache_size
, Error
**errp
)
877 BDRVQcow2State
*s
= bs
->opaque
;
878 uint64_t combined_cache_size
, l2_cache_max_setting
;
879 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
880 bool l2_cache_entry_size_set
;
881 int min_refcount_cache
= MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
882 uint64_t virtual_disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
883 uint64_t max_l2_entries
= DIV_ROUND_UP(virtual_disk_size
, s
->cluster_size
);
884 /* An L2 table is always one cluster in size so the max cache size
885 * should be a multiple of the cluster size. */
886 uint64_t max_l2_cache
= ROUND_UP(max_l2_entries
* l2_entry_size(s
),
889 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
890 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
891 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
892 l2_cache_entry_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
);
894 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
895 l2_cache_max_setting
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
,
896 DEFAULT_L2_CACHE_MAX_SIZE
);
897 *refcount_cache_size
= qemu_opt_get_size(opts
,
898 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
900 *l2_cache_entry_size
= qemu_opt_get_size(
901 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
903 *l2_cache_size
= MIN(max_l2_cache
, l2_cache_max_setting
);
905 if (combined_cache_size_set
) {
906 if (l2_cache_size_set
&& refcount_cache_size_set
) {
907 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
908 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
911 } else if (l2_cache_size_set
&&
912 (l2_cache_max_setting
> combined_cache_size
)) {
913 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
914 QCOW2_OPT_CACHE_SIZE
);
916 } else if (*refcount_cache_size
> combined_cache_size
) {
917 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
918 QCOW2_OPT_CACHE_SIZE
);
922 if (l2_cache_size_set
) {
923 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
924 } else if (refcount_cache_size_set
) {
925 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
927 /* Assign as much memory as possible to the L2 cache, and
928 * use the remainder for the refcount cache */
929 if (combined_cache_size
>= max_l2_cache
+ min_refcount_cache
) {
930 *l2_cache_size
= max_l2_cache
;
931 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
933 *refcount_cache_size
=
934 MIN(combined_cache_size
, min_refcount_cache
);
935 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
941 * If the L2 cache is not enough to cover the whole disk then
942 * default to 4KB entries. Smaller entries reduce the cost of
943 * loads and evictions and increase I/O performance.
945 if (*l2_cache_size
< max_l2_cache
&& !l2_cache_entry_size_set
) {
946 *l2_cache_entry_size
= MIN(s
->cluster_size
, 4096);
949 /* l2_cache_size and refcount_cache_size are ensured to have at least
950 * their minimum values in qcow2_update_options_prepare() */
952 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
953 *l2_cache_entry_size
> s
->cluster_size
||
954 !is_power_of_2(*l2_cache_entry_size
)) {
955 error_setg(errp
, "L2 cache entry size must be a power of two "
956 "between %d and the cluster size (%d)",
957 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
964 typedef struct Qcow2ReopenState
{
965 Qcow2Cache
*l2_table_cache
;
966 Qcow2Cache
*refcount_block_cache
;
967 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
968 bool use_lazy_refcounts
;
970 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
971 uint64_t cache_clean_interval
;
972 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
975 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
977 QDict
*options
, int flags
,
980 BDRVQcow2State
*s
= bs
->opaque
;
981 QemuOpts
*opts
= NULL
;
982 const char *opt_overlap_check
, *opt_overlap_check_template
;
983 int overlap_check_template
= 0;
984 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
986 const char *encryptfmt
;
987 QDict
*encryptopts
= NULL
;
990 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
991 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
993 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
994 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
999 /* get L2 table/refcount block cache size from command line options */
1000 if (!read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
1001 &refcount_cache_size
, errp
)) {
1006 l2_cache_size
/= l2_cache_entry_size
;
1007 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
1008 l2_cache_size
= MIN_L2_CACHE_SIZE
;
1010 if (l2_cache_size
> INT_MAX
) {
1011 error_setg(errp
, "L2 cache size too big");
1016 refcount_cache_size
/= s
->cluster_size
;
1017 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
1018 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
1020 if (refcount_cache_size
> INT_MAX
) {
1021 error_setg(errp
, "Refcount cache size too big");
1026 /* alloc new L2 table/refcount block cache, flush old one */
1027 if (s
->l2_table_cache
) {
1028 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1030 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
1035 if (s
->refcount_block_cache
) {
1036 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1038 error_setg_errno(errp
, -ret
,
1039 "Failed to flush the refcount block cache");
1044 r
->l2_slice_size
= l2_cache_entry_size
/ l2_entry_size(s
);
1045 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
1046 l2_cache_entry_size
);
1047 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
1049 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
1050 error_setg(errp
, "Could not allocate metadata caches");
1055 /* New interval for cache cleanup timer */
1056 r
->cache_clean_interval
=
1057 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
1058 DEFAULT_CACHE_CLEAN_INTERVAL
);
1059 #ifndef CONFIG_LINUX
1060 if (r
->cache_clean_interval
!= 0) {
1061 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1062 " not supported on this host");
1067 if (r
->cache_clean_interval
> UINT_MAX
) {
1068 error_setg(errp
, "Cache clean interval too big");
1073 /* lazy-refcounts; flush if going from enabled to disabled */
1074 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
1075 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
1076 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
1077 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
1078 "qemu 1.1 compatibility level");
1083 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
1084 ret
= qcow2_mark_clean(bs
);
1086 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
1091 /* Overlap check options */
1092 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
1093 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1094 if (opt_overlap_check_template
&& opt_overlap_check
&&
1095 strcmp(opt_overlap_check_template
, opt_overlap_check
))
1097 error_setg(errp
, "Conflicting values for qcow2 options '"
1098 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1099 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
1103 if (!opt_overlap_check
) {
1104 opt_overlap_check
= opt_overlap_check_template
?: "cached";
1107 if (!strcmp(opt_overlap_check
, "none")) {
1108 overlap_check_template
= 0;
1109 } else if (!strcmp(opt_overlap_check
, "constant")) {
1110 overlap_check_template
= QCOW2_OL_CONSTANT
;
1111 } else if (!strcmp(opt_overlap_check
, "cached")) {
1112 overlap_check_template
= QCOW2_OL_CACHED
;
1113 } else if (!strcmp(opt_overlap_check
, "all")) {
1114 overlap_check_template
= QCOW2_OL_ALL
;
1116 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1117 "'overlap-check'. Allowed are any of the following: "
1118 "none, constant, cached, all", opt_overlap_check
);
1123 r
->overlap_check
= 0;
1124 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1125 /* overlap-check defines a template bitmask, but every flag may be
1126 * overwritten through the associated boolean option */
1128 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1129 overlap_check_template
& (1 << i
)) << i
;
1132 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1133 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1134 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1135 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1136 flags
& BDRV_O_UNMAP
);
1137 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1138 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1139 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1140 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1142 switch (s
->crypt_method_header
) {
1143 case QCOW_CRYPT_NONE
:
1145 error_setg(errp
, "No encryption in image header, but options "
1146 "specified format '%s'", encryptfmt
);
1152 case QCOW_CRYPT_AES
:
1153 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1155 "Header reported 'aes' encryption format but "
1156 "options specify '%s'", encryptfmt
);
1160 qdict_put_str(encryptopts
, "format", "qcow");
1161 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1162 if (!r
->crypto_opts
) {
1168 case QCOW_CRYPT_LUKS
:
1169 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1171 "Header reported 'luks' encryption format but "
1172 "options specify '%s'", encryptfmt
);
1176 qdict_put_str(encryptopts
, "format", "luks");
1177 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1178 if (!r
->crypto_opts
) {
1185 error_setg(errp
, "Unsupported encryption method %d",
1186 s
->crypt_method_header
);
1193 qobject_unref(encryptopts
);
1194 qemu_opts_del(opts
);
1199 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1200 Qcow2ReopenState
*r
)
1202 BDRVQcow2State
*s
= bs
->opaque
;
1205 if (s
->l2_table_cache
) {
1206 qcow2_cache_destroy(s
->l2_table_cache
);
1208 if (s
->refcount_block_cache
) {
1209 qcow2_cache_destroy(s
->refcount_block_cache
);
1211 s
->l2_table_cache
= r
->l2_table_cache
;
1212 s
->refcount_block_cache
= r
->refcount_block_cache
;
1213 s
->l2_slice_size
= r
->l2_slice_size
;
1215 s
->overlap_check
= r
->overlap_check
;
1216 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1218 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1219 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1222 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1223 cache_clean_timer_del(bs
);
1224 s
->cache_clean_interval
= r
->cache_clean_interval
;
1225 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1228 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1229 s
->crypto_opts
= r
->crypto_opts
;
1232 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1233 Qcow2ReopenState
*r
)
1235 if (r
->l2_table_cache
) {
1236 qcow2_cache_destroy(r
->l2_table_cache
);
1238 if (r
->refcount_block_cache
) {
1239 qcow2_cache_destroy(r
->refcount_block_cache
);
1241 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1244 static int coroutine_fn
1245 qcow2_update_options(BlockDriverState
*bs
, QDict
*options
, int flags
,
1248 Qcow2ReopenState r
= {};
1251 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1253 qcow2_update_options_commit(bs
, &r
);
1255 qcow2_update_options_abort(bs
, &r
);
1261 static int validate_compression_type(BDRVQcow2State
*s
, Error
**errp
)
1263 switch (s
->compression_type
) {
1264 case QCOW2_COMPRESSION_TYPE_ZLIB
:
1266 case QCOW2_COMPRESSION_TYPE_ZSTD
:
1271 error_setg(errp
, "qcow2: unknown compression type: %u",
1272 s
->compression_type
);
1277 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1278 * the incompatible feature flag must be set
1280 if (s
->compression_type
== QCOW2_COMPRESSION_TYPE_ZLIB
) {
1281 if (s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
) {
1282 error_setg(errp
, "qcow2: Compression type incompatible feature "
1283 "bit must not be set");
1287 if (!(s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
)) {
1288 error_setg(errp
, "qcow2: Compression type incompatible feature "
1297 /* Called with s->lock held. */
1298 static int coroutine_fn GRAPH_RDLOCK
1299 qcow2_do_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1300 bool open_data_file
, Error
**errp
)
1303 BDRVQcow2State
*s
= bs
->opaque
;
1304 unsigned int len
, i
;
1308 uint64_t l1_vm_state_index
;
1309 bool update_header
= false;
1311 ret
= bdrv_co_pread(bs
->file
, 0, sizeof(header
), &header
, 0);
1313 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1316 header
.magic
= be32_to_cpu(header
.magic
);
1317 header
.version
= be32_to_cpu(header
.version
);
1318 header
.backing_file_offset
= be64_to_cpu(header
.backing_file_offset
);
1319 header
.backing_file_size
= be32_to_cpu(header
.backing_file_size
);
1320 header
.size
= be64_to_cpu(header
.size
);
1321 header
.cluster_bits
= be32_to_cpu(header
.cluster_bits
);
1322 header
.crypt_method
= be32_to_cpu(header
.crypt_method
);
1323 header
.l1_table_offset
= be64_to_cpu(header
.l1_table_offset
);
1324 header
.l1_size
= be32_to_cpu(header
.l1_size
);
1325 header
.refcount_table_offset
= be64_to_cpu(header
.refcount_table_offset
);
1326 header
.refcount_table_clusters
=
1327 be32_to_cpu(header
.refcount_table_clusters
);
1328 header
.snapshots_offset
= be64_to_cpu(header
.snapshots_offset
);
1329 header
.nb_snapshots
= be32_to_cpu(header
.nb_snapshots
);
1331 if (header
.magic
!= QCOW_MAGIC
) {
1332 error_setg(errp
, "Image is not in qcow2 format");
1336 if (header
.version
< 2 || header
.version
> 3) {
1337 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1342 s
->qcow_version
= header
.version
;
1344 /* Initialise cluster size */
1345 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1346 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1347 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1348 header
.cluster_bits
);
1353 s
->cluster_bits
= header
.cluster_bits
;
1354 s
->cluster_size
= 1 << s
->cluster_bits
;
1356 /* Initialise version 3 header fields */
1357 if (header
.version
== 2) {
1358 header
.incompatible_features
= 0;
1359 header
.compatible_features
= 0;
1360 header
.autoclear_features
= 0;
1361 header
.refcount_order
= 4;
1362 header
.header_length
= 72;
1364 header
.incompatible_features
=
1365 be64_to_cpu(header
.incompatible_features
);
1366 header
.compatible_features
= be64_to_cpu(header
.compatible_features
);
1367 header
.autoclear_features
= be64_to_cpu(header
.autoclear_features
);
1368 header
.refcount_order
= be32_to_cpu(header
.refcount_order
);
1369 header
.header_length
= be32_to_cpu(header
.header_length
);
1371 if (header
.header_length
< 104) {
1372 error_setg(errp
, "qcow2 header too short");
1378 if (header
.header_length
> s
->cluster_size
) {
1379 error_setg(errp
, "qcow2 header exceeds cluster size");
1384 if (header
.header_length
> sizeof(header
)) {
1385 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1386 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1387 ret
= bdrv_co_pread(bs
->file
, sizeof(header
),
1388 s
->unknown_header_fields_size
,
1389 s
->unknown_header_fields
, 0);
1391 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1397 if (header
.backing_file_offset
> s
->cluster_size
) {
1398 error_setg(errp
, "Invalid backing file offset");
1403 if (header
.backing_file_offset
) {
1404 ext_end
= header
.backing_file_offset
;
1406 ext_end
= 1 << header
.cluster_bits
;
1409 /* Handle feature bits */
1410 s
->incompatible_features
= header
.incompatible_features
;
1411 s
->compatible_features
= header
.compatible_features
;
1412 s
->autoclear_features
= header
.autoclear_features
;
1415 * Handle compression type
1416 * Older qcow2 images don't contain the compression type header.
1417 * Distinguish them by the header length and use
1418 * the only valid (default) compression type in that case
1420 if (header
.header_length
> offsetof(QCowHeader
, compression_type
)) {
1421 s
->compression_type
= header
.compression_type
;
1423 s
->compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
1426 ret
= validate_compression_type(s
, errp
);
1431 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1432 void *feature_table
= NULL
;
1433 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1434 &feature_table
, flags
, NULL
, NULL
);
1435 report_unsupported_feature(errp
, feature_table
,
1436 s
->incompatible_features
&
1437 ~QCOW2_INCOMPAT_MASK
);
1439 g_free(feature_table
);
1443 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1444 /* Corrupt images may not be written to unless they are being repaired
1446 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1447 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1454 s
->subclusters_per_cluster
=
1455 has_subclusters(s
) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
: 1;
1456 s
->subcluster_size
= s
->cluster_size
/ s
->subclusters_per_cluster
;
1457 s
->subcluster_bits
= ctz32(s
->subcluster_size
);
1459 if (s
->subcluster_size
< (1 << MIN_CLUSTER_BITS
)) {
1460 error_setg(errp
, "Unsupported subcluster size: %d", s
->subcluster_size
);
1465 /* Check support for various header values */
1466 if (header
.refcount_order
> 6) {
1467 error_setg(errp
, "Reference count entry width too large; may not "
1472 s
->refcount_order
= header
.refcount_order
;
1473 s
->refcount_bits
= 1 << s
->refcount_order
;
1474 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1475 s
->refcount_max
+= s
->refcount_max
- 1;
1477 s
->crypt_method_header
= header
.crypt_method
;
1478 if (s
->crypt_method_header
) {
1479 if (bdrv_uses_whitelist() &&
1480 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1482 "Use of AES-CBC encrypted qcow2 images is no longer "
1483 "supported in system emulators");
1484 error_append_hint(errp
,
1485 "You can use 'qemu-img convert' to convert your "
1486 "image to an alternative supported format, such "
1487 "as unencrypted qcow2, or raw with the LUKS "
1488 "format instead.\n");
1493 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1494 s
->crypt_physical_offset
= false;
1496 /* Assuming LUKS and any future crypt methods we
1497 * add will all use physical offsets, due to the
1498 * fact that the alternative is insecure... */
1499 s
->crypt_physical_offset
= true;
1502 bs
->encrypted
= true;
1505 s
->l2_bits
= s
->cluster_bits
- ctz32(l2_entry_size(s
));
1506 s
->l2_size
= 1 << s
->l2_bits
;
1507 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1508 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1509 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1510 bs
->total_sectors
= header
.size
/ BDRV_SECTOR_SIZE
;
1511 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1512 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1513 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1515 s
->refcount_table_offset
= header
.refcount_table_offset
;
1516 s
->refcount_table_size
=
1517 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1519 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1520 error_setg(errp
, "Image does not contain a reference count table");
1525 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1526 header
.refcount_table_clusters
,
1527 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1528 "Reference count table", errp
);
1533 if (!(flags
& BDRV_O_CHECK
)) {
1535 * The total size in bytes of the snapshot table is checked in
1536 * qcow2_read_snapshots() because the size of each snapshot is
1537 * variable and we don't know it yet.
1538 * Here we only check the offset and number of snapshots.
1540 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1541 header
.nb_snapshots
,
1542 sizeof(QCowSnapshotHeader
),
1543 sizeof(QCowSnapshotHeader
) *
1545 "Snapshot table", errp
);
1551 /* read the level 1 table */
1552 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1553 header
.l1_size
, L1E_SIZE
,
1554 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1558 s
->l1_size
= header
.l1_size
;
1559 s
->l1_table_offset
= header
.l1_table_offset
;
1561 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1562 if (l1_vm_state_index
> INT_MAX
) {
1563 error_setg(errp
, "Image is too big");
1567 s
->l1_vm_state_index
= l1_vm_state_index
;
1569 /* the L1 table must contain at least enough entries to put
1570 header.size bytes */
1571 if (s
->l1_size
< s
->l1_vm_state_index
) {
1572 error_setg(errp
, "L1 table is too small");
1577 if (s
->l1_size
> 0) {
1578 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
, s
->l1_size
* L1E_SIZE
);
1579 if (s
->l1_table
== NULL
) {
1580 error_setg(errp
, "Could not allocate L1 table");
1584 ret
= bdrv_co_pread(bs
->file
, s
->l1_table_offset
, s
->l1_size
* L1E_SIZE
,
1587 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1590 for(i
= 0;i
< s
->l1_size
; i
++) {
1591 s
->l1_table
[i
] = be64_to_cpu(s
->l1_table
[i
]);
1595 /* Parse driver-specific options */
1596 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1603 ret
= qcow2_refcount_init(bs
);
1605 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1609 QLIST_INIT(&s
->cluster_allocs
);
1610 QTAILQ_INIT(&s
->discards
);
1612 /* read qcow2 extensions */
1613 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1614 flags
, &update_header
, errp
)) {
1619 if (open_data_file
) {
1620 /* Open external data file */
1621 s
->data_file
= bdrv_co_open_child(NULL
, options
, "data-file", bs
,
1622 &child_of_bds
, BDRV_CHILD_DATA
,
1629 if (s
->incompatible_features
& QCOW2_INCOMPAT_DATA_FILE
) {
1630 if (!s
->data_file
&& s
->image_data_file
) {
1631 s
->data_file
= bdrv_co_open_child(s
->image_data_file
, options
,
1634 BDRV_CHILD_DATA
, false, errp
);
1635 if (!s
->data_file
) {
1640 if (!s
->data_file
) {
1641 error_setg(errp
, "'data-file' is required for this image");
1647 bs
->file
->role
&= ~BDRV_CHILD_DATA
;
1649 /* Must succeed because we have given up permissions if anything */
1650 bdrv_child_refresh_perms(bs
, bs
->file
, &error_abort
);
1653 error_setg(errp
, "'data-file' can only be set for images with "
1654 "an external data file");
1659 s
->data_file
= bs
->file
;
1661 if (data_file_is_raw(bs
)) {
1662 error_setg(errp
, "data-file-raw requires a data file");
1669 /* qcow2_read_extension may have set up the crypto context
1670 * if the crypt method needs a header region, some methods
1671 * don't need header extensions, so must check here
1673 if (s
->crypt_method_header
&& !s
->crypto
) {
1674 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1675 unsigned int cflags
= 0;
1676 if (flags
& BDRV_O_NO_IO
) {
1677 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1679 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1681 QCOW2_MAX_THREADS
, errp
);
1686 } else if (!(flags
& BDRV_O_NO_IO
)) {
1687 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1688 s
->crypt_method_header
);
1694 /* read the backing file name */
1695 if (header
.backing_file_offset
!= 0) {
1696 len
= header
.backing_file_size
;
1697 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1698 len
>= sizeof(bs
->backing_file
)) {
1699 error_setg(errp
, "Backing file name too long");
1704 s
->image_backing_file
= g_malloc(len
+ 1);
1705 ret
= bdrv_co_pread(bs
->file
, header
.backing_file_offset
, len
,
1706 s
->image_backing_file
, 0);
1708 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1711 s
->image_backing_file
[len
] = '\0';
1714 * Update only when something has changed. This function is called by
1715 * qcow2_co_invalidate_cache(), and we do not want to reset
1716 * auto_backing_file unless necessary.
1718 if (!g_str_equal(s
->image_backing_file
, bs
->backing_file
)) {
1719 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
),
1720 s
->image_backing_file
);
1721 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
1722 s
->image_backing_file
);
1727 * Internal snapshots; skip reading them in check mode, because
1728 * we do not need them then, and we do not want to abort because
1729 * of a broken table.
1731 if (!(flags
& BDRV_O_CHECK
)) {
1732 s
->snapshots_offset
= header
.snapshots_offset
;
1733 s
->nb_snapshots
= header
.nb_snapshots
;
1735 ret
= qcow2_read_snapshots(bs
, errp
);
1741 /* Clear unknown autoclear feature bits */
1742 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1743 update_header
= update_header
&& bdrv_is_writable(bs
);
1744 if (update_header
) {
1745 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1748 /* == Handle persistent dirty bitmaps ==
1750 * We want load dirty bitmaps in three cases:
1752 * 1. Normal open of the disk in active mode, not related to invalidation
1755 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1756 * bitmaps are _not_ migrating through migration channel, i.e.
1757 * 'dirty-bitmaps' capability is disabled.
1759 * 3. Invalidation of source vm after failed or canceled migration.
1760 * This is a very interesting case. There are two possible types of
1763 * A. Stored on inactivation and removed. They should be loaded from the
1766 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1767 * the migration channel (with dirty-bitmaps capability).
1769 * On the other hand, there are two possible sub-cases:
1771 * 3.1 disk was changed by somebody else while were inactive. In this
1772 * case all in-RAM dirty bitmaps (both persistent and not) are
1773 * definitely invalid. And we don't have any method to determine
1776 * Simple and safe thing is to just drop all the bitmaps of type B on
1777 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1779 * On the other hand, resuming source vm, if disk was already changed
1780 * is a bad thing anyway: not only bitmaps, the whole vm state is
1781 * out of sync with disk.
1783 * This means, that user or management tool, who for some reason
1784 * decided to resume source vm, after disk was already changed by
1785 * target vm, should at least drop all dirty bitmaps by hand.
1787 * So, we can ignore this case for now, but TODO: "generation"
1788 * extension for qcow2, to determine, that image was changed after
1789 * last inactivation. And if it is changed, we will drop (or at least
1790 * mark as 'invalid' all the bitmaps of type B, both persistent
1793 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1794 * to disk ('dirty-bitmaps' capability disabled), or not saved
1795 * ('dirty-bitmaps' capability enabled), but we don't need to care
1796 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1797 * and not stored has flag IN_USE=1 in the image and will be skipped
1800 * One remaining possible case when we don't want load bitmaps:
1802 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1803 * will be loaded on invalidation, no needs try loading them before)
1806 if (!(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
)) {
1807 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1808 bool header_updated
;
1809 if (!qcow2_load_dirty_bitmaps(bs
, &header_updated
, errp
)) {
1814 update_header
= update_header
&& !header_updated
;
1817 if (update_header
) {
1818 ret
= qcow2_update_header(bs
);
1820 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1825 bs
->supported_zero_flags
= header
.version
>= 3 ?
1826 BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
: 0;
1827 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
1829 /* Repair image if dirty */
1830 if (!(flags
& BDRV_O_CHECK
) && bdrv_is_writable(bs
) &&
1831 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1832 BdrvCheckResult result
= {0};
1834 ret
= qcow2_co_check_locked(bs
, &result
,
1835 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1836 if (ret
< 0 || result
.check_errors
) {
1840 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1847 BdrvCheckResult result
= {0};
1848 qcow2_check_refcounts(bs
, &result
, 0);
1852 qemu_co_queue_init(&s
->thread_task_queue
);
1857 g_free(s
->image_data_file
);
1858 if (open_data_file
&& has_data_file(bs
)) {
1859 bdrv_unref_child(bs
, s
->data_file
);
1860 s
->data_file
= NULL
;
1862 g_free(s
->unknown_header_fields
);
1863 cleanup_unknown_header_ext(bs
);
1864 qcow2_free_snapshots(bs
);
1865 qcow2_refcount_close(bs
);
1866 qemu_vfree(s
->l1_table
);
1867 /* else pre-write overlap checks in cache_destroy may crash */
1869 cache_clean_timer_del(bs
);
1870 if (s
->l2_table_cache
) {
1871 qcow2_cache_destroy(s
->l2_table_cache
);
1873 if (s
->refcount_block_cache
) {
1874 qcow2_cache_destroy(s
->refcount_block_cache
);
1876 qcrypto_block_free(s
->crypto
);
1877 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1881 typedef struct QCow2OpenCo
{
1882 BlockDriverState
*bs
;
1889 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1891 QCow2OpenCo
*qoc
= opaque
;
1892 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1894 assume_graph_lock(); /* FIXME */
1896 qemu_co_mutex_lock(&s
->lock
);
1897 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, true,
1899 qemu_co_mutex_unlock(&s
->lock
);
1902 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1905 BDRVQcow2State
*s
= bs
->opaque
;
1915 ret
= bdrv_open_file_child(NULL
, options
, "file", bs
, errp
);
1920 /* Initialise locks */
1921 qemu_co_mutex_init(&s
->lock
);
1923 if (qemu_in_coroutine()) {
1924 /* From bdrv_co_create. */
1925 qcow2_open_entry(&qoc
);
1927 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1928 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1929 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1934 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1936 BDRVQcow2State
*s
= bs
->opaque
;
1938 if (bs
->encrypted
) {
1939 /* Encryption works on a sector granularity */
1940 bs
->bl
.request_alignment
= qcrypto_block_get_sector_size(s
->crypto
);
1942 bs
->bl
.pwrite_zeroes_alignment
= s
->subcluster_size
;
1943 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1946 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1947 BlockReopenQueue
*queue
, Error
**errp
)
1949 BDRVQcow2State
*s
= state
->bs
->opaque
;
1950 Qcow2ReopenState
*r
;
1953 r
= g_new0(Qcow2ReopenState
, 1);
1956 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1957 state
->flags
, errp
);
1962 /* We need to write out any unwritten data if we reopen read-only. */
1963 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1964 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1969 ret
= bdrv_flush(state
->bs
);
1974 ret
= qcow2_mark_clean(state
->bs
);
1981 * Without an external data file, s->data_file points to the same BdrvChild
1982 * as bs->file. It needs to be resynced after reopen because bs->file may
1983 * be changed. We can't use it in the meantime.
1985 if (!has_data_file(state
->bs
)) {
1986 assert(s
->data_file
== state
->bs
->file
);
1987 s
->data_file
= NULL
;
1993 qcow2_update_options_abort(state
->bs
, r
);
1998 static void qcow2_reopen_commit(BDRVReopenState
*state
)
2000 BDRVQcow2State
*s
= state
->bs
->opaque
;
2002 qcow2_update_options_commit(state
->bs
, state
->opaque
);
2003 if (!s
->data_file
) {
2005 * If we don't have an external data file, s->data_file was cleared by
2006 * qcow2_reopen_prepare() and needs to be updated.
2008 s
->data_file
= state
->bs
->file
;
2010 g_free(state
->opaque
);
2013 static void qcow2_reopen_commit_post(BDRVReopenState
*state
)
2015 if (state
->flags
& BDRV_O_RDWR
) {
2016 Error
*local_err
= NULL
;
2018 if (qcow2_reopen_bitmaps_rw(state
->bs
, &local_err
) < 0) {
2020 * This is not fatal, bitmaps just left read-only, so all following
2021 * writes will fail. User can remove read-only bitmaps to unblock
2022 * writes or retry reopen.
2024 error_reportf_err(local_err
,
2025 "%s: Failed to make dirty bitmaps writable: ",
2026 bdrv_get_node_name(state
->bs
));
2031 static void qcow2_reopen_abort(BDRVReopenState
*state
)
2033 BDRVQcow2State
*s
= state
->bs
->opaque
;
2035 if (!s
->data_file
) {
2037 * If we don't have an external data file, s->data_file was cleared by
2038 * qcow2_reopen_prepare() and needs to be restored.
2040 s
->data_file
= state
->bs
->file
;
2042 qcow2_update_options_abort(state
->bs
, state
->opaque
);
2043 g_free(state
->opaque
);
2046 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
2048 bool has_new_overlap_template
=
2049 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
2050 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2051 bool has_new_total_cache_size
=
2052 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
2053 bool has_all_cache_options
;
2055 /* New overlap template overrides all old overlap options */
2056 if (has_new_overlap_template
) {
2057 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
2058 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2059 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
2060 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
2061 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
2062 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
2063 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
2064 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
2065 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
2066 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
2069 /* New total cache size overrides all old options */
2070 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
2071 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
2072 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2075 qdict_join(options
, old_options
, false);
2078 * If after merging all cache size options are set, an old total size is
2079 * overwritten. Do keep all options, however, if all three are new. The
2080 * resulting error message is what we want to happen.
2082 has_all_cache_options
=
2083 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
2084 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
2085 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2087 if (has_all_cache_options
&& !has_new_total_cache_size
) {
2088 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
2092 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
2094 int64_t offset
, int64_t count
,
2095 int64_t *pnum
, int64_t *map
,
2096 BlockDriverState
**file
)
2098 BDRVQcow2State
*s
= bs
->opaque
;
2099 uint64_t host_offset
;
2101 QCow2SubclusterType type
;
2102 int ret
, status
= 0;
2104 qemu_co_mutex_lock(&s
->lock
);
2106 if (!s
->metadata_preallocation_checked
) {
2107 ret
= qcow2_detect_metadata_preallocation(bs
);
2108 s
->metadata_preallocation
= (ret
== 1);
2109 s
->metadata_preallocation_checked
= true;
2112 bytes
= MIN(INT_MAX
, count
);
2113 ret
= qcow2_get_host_offset(bs
, offset
, &bytes
, &host_offset
, &type
);
2114 qemu_co_mutex_unlock(&s
->lock
);
2121 if ((type
== QCOW2_SUBCLUSTER_NORMAL
||
2122 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2123 type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) && !s
->crypto
) {
2125 *file
= s
->data_file
->bs
;
2126 status
|= BDRV_BLOCK_OFFSET_VALID
;
2128 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2129 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
) {
2130 status
|= BDRV_BLOCK_ZERO
;
2131 } else if (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
2132 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) {
2133 status
|= BDRV_BLOCK_DATA
;
2135 if (s
->metadata_preallocation
&& (status
& BDRV_BLOCK_DATA
) &&
2136 (status
& BDRV_BLOCK_OFFSET_VALID
))
2138 status
|= BDRV_BLOCK_RECURSE
;
2143 static int coroutine_fn GRAPH_RDLOCK
2144 qcow2_handle_l2meta(BlockDriverState
*bs
, QCowL2Meta
**pl2meta
, bool link_l2
)
2147 QCowL2Meta
*l2meta
= *pl2meta
;
2149 while (l2meta
!= NULL
) {
2153 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
2158 qcow2_alloc_cluster_abort(bs
, l2meta
);
2161 /* Take the request off the list of running requests */
2162 QLIST_REMOVE(l2meta
, next_in_flight
);
2164 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2166 next
= l2meta
->next
;
2175 static int coroutine_fn GRAPH_RDLOCK
2176 qcow2_co_preadv_encrypted(BlockDriverState
*bs
,
2177 uint64_t host_offset
,
2181 uint64_t qiov_offset
)
2184 BDRVQcow2State
*s
= bs
->opaque
;
2187 assert(bs
->encrypted
&& s
->crypto
);
2188 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2191 * For encrypted images, read everything into a temporary
2192 * contiguous buffer on which the AES functions can work.
2193 * Also, decryption in a separate buffer is better as it
2194 * prevents the guest from learning information about the
2195 * encrypted nature of the virtual disk.
2198 buf
= qemu_try_blockalign(s
->data_file
->bs
, bytes
);
2203 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2204 ret
= bdrv_co_pread(s
->data_file
, host_offset
, bytes
, buf
, 0);
2209 if (qcow2_co_decrypt(bs
, host_offset
, offset
, buf
, bytes
) < 0)
2214 qemu_iovec_from_buf(qiov
, qiov_offset
, buf
, bytes
);
2222 typedef struct Qcow2AioTask
{
2225 BlockDriverState
*bs
;
2226 QCow2SubclusterType subcluster_type
; /* only for read */
2227 uint64_t host_offset
; /* or l2_entry for compressed read */
2231 uint64_t qiov_offset
;
2232 QCowL2Meta
*l2meta
; /* only for write */
2235 static coroutine_fn
int qcow2_co_preadv_task_entry(AioTask
*task
);
2236 static coroutine_fn
int qcow2_add_task(BlockDriverState
*bs
,
2239 QCow2SubclusterType subcluster_type
,
2240 uint64_t host_offset
,
2247 Qcow2AioTask local_task
;
2248 Qcow2AioTask
*task
= pool
? g_new(Qcow2AioTask
, 1) : &local_task
;
2250 *task
= (Qcow2AioTask
) {
2253 .subcluster_type
= subcluster_type
,
2255 .host_offset
= host_offset
,
2258 .qiov_offset
= qiov_offset
,
2262 trace_qcow2_add_task(qemu_coroutine_self(), bs
, pool
,
2263 func
== qcow2_co_preadv_task_entry
? "read" : "write",
2264 subcluster_type
, host_offset
, offset
, bytes
,
2268 return func(&task
->task
);
2271 aio_task_pool_start_task(pool
, &task
->task
);
2276 static int coroutine_fn GRAPH_RDLOCK
2277 qcow2_co_preadv_task(BlockDriverState
*bs
, QCow2SubclusterType subc_type
,
2278 uint64_t host_offset
, uint64_t offset
, uint64_t bytes
,
2279 QEMUIOVector
*qiov
, size_t qiov_offset
)
2281 BDRVQcow2State
*s
= bs
->opaque
;
2283 switch (subc_type
) {
2284 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
2285 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
2286 /* Both zero types are handled in qcow2_co_preadv_part */
2287 g_assert_not_reached();
2289 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
2290 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
2291 assert(bs
->backing
); /* otherwise handled in qcow2_co_preadv_part */
2293 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
2294 return bdrv_co_preadv_part(bs
->backing
, offset
, bytes
,
2295 qiov
, qiov_offset
, 0);
2297 case QCOW2_SUBCLUSTER_COMPRESSED
:
2298 return qcow2_co_preadv_compressed(bs
, host_offset
,
2299 offset
, bytes
, qiov
, qiov_offset
);
2301 case QCOW2_SUBCLUSTER_NORMAL
:
2302 if (bs
->encrypted
) {
2303 return qcow2_co_preadv_encrypted(bs
, host_offset
,
2304 offset
, bytes
, qiov
, qiov_offset
);
2307 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2308 return bdrv_co_preadv_part(s
->data_file
, host_offset
,
2309 bytes
, qiov
, qiov_offset
, 0);
2312 g_assert_not_reached();
2315 g_assert_not_reached();
2319 * This function can count as GRAPH_RDLOCK because qcow2_co_preadv_part() holds
2320 * the graph lock and keeps it until this coroutine has terminated.
2322 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_preadv_task_entry(AioTask
*task
)
2324 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2328 return qcow2_co_preadv_task(t
->bs
, t
->subcluster_type
,
2329 t
->host_offset
, t
->offset
, t
->bytes
,
2330 t
->qiov
, t
->qiov_offset
);
2333 static int coroutine_fn GRAPH_RDLOCK
2334 qcow2_co_preadv_part(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
2335 QEMUIOVector
*qiov
, size_t qiov_offset
,
2336 BdrvRequestFlags flags
)
2338 BDRVQcow2State
*s
= bs
->opaque
;
2340 unsigned int cur_bytes
; /* number of bytes in current iteration */
2341 uint64_t host_offset
= 0;
2342 QCow2SubclusterType type
;
2343 AioTaskPool
*aio
= NULL
;
2345 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2346 /* prepare next request */
2347 cur_bytes
= MIN(bytes
, INT_MAX
);
2349 cur_bytes
= MIN(cur_bytes
,
2350 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2353 qemu_co_mutex_lock(&s
->lock
);
2354 ret
= qcow2_get_host_offset(bs
, offset
, &cur_bytes
,
2355 &host_offset
, &type
);
2356 qemu_co_mutex_unlock(&s
->lock
);
2361 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2362 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2363 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&& !bs
->backing
) ||
2364 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&& !bs
->backing
))
2366 qemu_iovec_memset(qiov
, qiov_offset
, 0, cur_bytes
);
2368 if (!aio
&& cur_bytes
!= bytes
) {
2369 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2371 ret
= qcow2_add_task(bs
, aio
, qcow2_co_preadv_task_entry
, type
,
2372 host_offset
, offset
, cur_bytes
,
2373 qiov
, qiov_offset
, NULL
);
2380 offset
+= cur_bytes
;
2381 qiov_offset
+= cur_bytes
;
2386 aio_task_pool_wait_all(aio
);
2388 ret
= aio_task_pool_status(aio
);
2396 /* Check if it's possible to merge a write request with the writing of
2397 * the data from the COW regions */
2398 static bool merge_cow(uint64_t offset
, unsigned bytes
,
2399 QEMUIOVector
*qiov
, size_t qiov_offset
,
2404 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2405 /* If both COW regions are empty then there's nothing to merge */
2406 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
2410 /* If COW regions are handled already, skip this too */
2416 * The write request should start immediately after the first
2417 * COW region. This does not always happen because the area
2418 * touched by the request can be larger than the one defined
2419 * by @m (a single request can span an area consisting of a
2420 * mix of previously unallocated and allocated clusters, that
2421 * is why @l2meta is a list).
2423 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
2424 /* In this case the request starts before this region */
2425 assert(offset
< l2meta_cow_start(m
));
2426 assert(m
->cow_start
.nb_bytes
== 0);
2430 /* The write request should end immediately before the second
2431 * COW region (see above for why it does not always happen) */
2432 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
2433 assert(offset
+ bytes
> m
->offset
+ m
->cow_end
.offset
);
2434 assert(m
->cow_end
.nb_bytes
== 0);
2438 /* Make sure that adding both COW regions to the QEMUIOVector
2439 * does not exceed IOV_MAX */
2440 if (qemu_iovec_subvec_niov(qiov
, qiov_offset
, bytes
) > IOV_MAX
- 2) {
2444 m
->data_qiov
= qiov
;
2445 m
->data_qiov_offset
= qiov_offset
;
2453 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error.
2454 * Note that returning 0 does not guarantee non-zero data.
2456 static int coroutine_fn GRAPH_RDLOCK
2457 is_zero_cow(BlockDriverState
*bs
, QCowL2Meta
*m
)
2460 * This check is designed for optimization shortcut so it must be
2462 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is
2463 * faster (but not as accurate and can result in false negatives).
2465 int ret
= bdrv_co_is_zero_fast(bs
, m
->offset
+ m
->cow_start
.offset
,
2466 m
->cow_start
.nb_bytes
);
2471 return bdrv_co_is_zero_fast(bs
, m
->offset
+ m
->cow_end
.offset
,
2472 m
->cow_end
.nb_bytes
);
2475 static int coroutine_fn GRAPH_RDLOCK
2476 handle_alloc_space(BlockDriverState
*bs
, QCowL2Meta
*l2meta
)
2478 BDRVQcow2State
*s
= bs
->opaque
;
2481 if (!(s
->data_file
->bs
->supported_zero_flags
& BDRV_REQ_NO_FALLBACK
)) {
2485 if (bs
->encrypted
) {
2489 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2491 uint64_t start_offset
= m
->alloc_offset
+ m
->cow_start
.offset
;
2492 unsigned nb_bytes
= m
->cow_end
.offset
+ m
->cow_end
.nb_bytes
-
2493 m
->cow_start
.offset
;
2495 if (!m
->cow_start
.nb_bytes
&& !m
->cow_end
.nb_bytes
) {
2499 ret
= is_zero_cow(bs
, m
);
2502 } else if (ret
== 0) {
2507 * instead of writing zero COW buffers,
2508 * efficiently zero out the whole clusters
2511 ret
= qcow2_pre_write_overlap_check(bs
, 0, start_offset
, nb_bytes
,
2517 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC_SPACE
);
2518 ret
= bdrv_co_pwrite_zeroes(s
->data_file
, start_offset
, nb_bytes
,
2519 BDRV_REQ_NO_FALLBACK
);
2521 if (ret
!= -ENOTSUP
&& ret
!= -EAGAIN
) {
2527 trace_qcow2_skip_cow(qemu_coroutine_self(), m
->offset
, m
->nb_clusters
);
2534 * qcow2_co_pwritev_task
2535 * Called with s->lock unlocked
2536 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2537 * not use it somehow after qcow2_co_pwritev_task() call
2539 static coroutine_fn GRAPH_RDLOCK
2540 int qcow2_co_pwritev_task(BlockDriverState
*bs
, uint64_t host_offset
,
2541 uint64_t offset
, uint64_t bytes
, QEMUIOVector
*qiov
,
2542 uint64_t qiov_offset
, QCowL2Meta
*l2meta
)
2545 BDRVQcow2State
*s
= bs
->opaque
;
2546 void *crypt_buf
= NULL
;
2547 QEMUIOVector encrypted_qiov
;
2549 if (bs
->encrypted
) {
2551 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2552 crypt_buf
= qemu_try_blockalign(bs
->file
->bs
, bytes
);
2553 if (crypt_buf
== NULL
) {
2557 qemu_iovec_to_buf(qiov
, qiov_offset
, crypt_buf
, bytes
);
2559 if (qcow2_co_encrypt(bs
, host_offset
, offset
, crypt_buf
, bytes
) < 0) {
2564 qemu_iovec_init_buf(&encrypted_qiov
, crypt_buf
, bytes
);
2565 qiov
= &encrypted_qiov
;
2569 /* Try to efficiently initialize the physical space with zeroes */
2570 ret
= handle_alloc_space(bs
, l2meta
);
2576 * If we need to do COW, check if it's possible to merge the
2577 * writing of the guest data together with that of the COW regions.
2578 * If it's not possible (or not necessary) then write the
2581 if (!merge_cow(offset
, bytes
, qiov
, qiov_offset
, l2meta
)) {
2582 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2583 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset
);
2584 ret
= bdrv_co_pwritev_part(s
->data_file
, host_offset
,
2585 bytes
, qiov
, qiov_offset
, 0);
2591 qemu_co_mutex_lock(&s
->lock
);
2593 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
2597 qemu_co_mutex_lock(&s
->lock
);
2600 qcow2_handle_l2meta(bs
, &l2meta
, false);
2601 qemu_co_mutex_unlock(&s
->lock
);
2603 qemu_vfree(crypt_buf
);
2609 * This function can count as GRAPH_RDLOCK because qcow2_co_pwritev_part() holds
2610 * the graph lock and keeps it until this coroutine has terminated.
2612 static coroutine_fn GRAPH_RDLOCK
int qcow2_co_pwritev_task_entry(AioTask
*task
)
2614 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2616 assert(!t
->subcluster_type
);
2618 return qcow2_co_pwritev_task(t
->bs
, t
->host_offset
,
2619 t
->offset
, t
->bytes
, t
->qiov
, t
->qiov_offset
,
2623 static int coroutine_fn GRAPH_RDLOCK
2624 qcow2_co_pwritev_part(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
2625 QEMUIOVector
*qiov
, size_t qiov_offset
,
2626 BdrvRequestFlags flags
)
2628 BDRVQcow2State
*s
= bs
->opaque
;
2629 int offset_in_cluster
;
2631 unsigned int cur_bytes
; /* number of sectors in current iteration */
2632 uint64_t host_offset
;
2633 QCowL2Meta
*l2meta
= NULL
;
2634 AioTaskPool
*aio
= NULL
;
2636 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
2638 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2642 trace_qcow2_writev_start_part(qemu_coroutine_self());
2643 offset_in_cluster
= offset_into_cluster(s
, offset
);
2644 cur_bytes
= MIN(bytes
, INT_MAX
);
2645 if (bs
->encrypted
) {
2646 cur_bytes
= MIN(cur_bytes
,
2647 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
2648 - offset_in_cluster
);
2651 qemu_co_mutex_lock(&s
->lock
);
2653 ret
= qcow2_alloc_host_offset(bs
, offset
, &cur_bytes
,
2654 &host_offset
, &l2meta
);
2659 ret
= qcow2_pre_write_overlap_check(bs
, 0, host_offset
,
2665 qemu_co_mutex_unlock(&s
->lock
);
2667 if (!aio
&& cur_bytes
!= bytes
) {
2668 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2670 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_task_entry
, 0,
2671 host_offset
, offset
,
2672 cur_bytes
, qiov
, qiov_offset
, l2meta
);
2673 l2meta
= NULL
; /* l2meta is consumed by qcow2_co_pwritev_task() */
2679 offset
+= cur_bytes
;
2680 qiov_offset
+= cur_bytes
;
2681 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2685 qemu_co_mutex_lock(&s
->lock
);
2688 qcow2_handle_l2meta(bs
, &l2meta
, false);
2690 qemu_co_mutex_unlock(&s
->lock
);
2694 aio_task_pool_wait_all(aio
);
2696 ret
= aio_task_pool_status(aio
);
2701 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2706 static int qcow2_inactivate(BlockDriverState
*bs
)
2708 BDRVQcow2State
*s
= bs
->opaque
;
2709 int ret
, result
= 0;
2710 Error
*local_err
= NULL
;
2712 qcow2_store_persistent_dirty_bitmaps(bs
, true, &local_err
);
2713 if (local_err
!= NULL
) {
2715 error_reportf_err(local_err
, "Lost persistent bitmaps during "
2716 "inactivation of node '%s': ",
2717 bdrv_get_device_or_node_name(bs
));
2720 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2723 error_report("Failed to flush the L2 table cache: %s",
2727 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2730 error_report("Failed to flush the refcount block cache: %s",
2735 qcow2_mark_clean(bs
);
2741 static void qcow2_do_close(BlockDriverState
*bs
, bool close_data_file
)
2743 BDRVQcow2State
*s
= bs
->opaque
;
2744 qemu_vfree(s
->l1_table
);
2745 /* else pre-write overlap checks in cache_destroy may crash */
2748 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2749 qcow2_inactivate(bs
);
2752 cache_clean_timer_del(bs
);
2753 qcow2_cache_destroy(s
->l2_table_cache
);
2754 qcow2_cache_destroy(s
->refcount_block_cache
);
2756 qcrypto_block_free(s
->crypto
);
2758 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
2760 g_free(s
->unknown_header_fields
);
2761 cleanup_unknown_header_ext(bs
);
2763 g_free(s
->image_data_file
);
2764 g_free(s
->image_backing_file
);
2765 g_free(s
->image_backing_format
);
2767 if (close_data_file
&& has_data_file(bs
)) {
2768 bdrv_unref_child(bs
, s
->data_file
);
2769 s
->data_file
= NULL
;
2772 qcow2_refcount_close(bs
);
2773 qcow2_free_snapshots(bs
);
2776 static void qcow2_close(BlockDriverState
*bs
)
2778 qcow2_do_close(bs
, true);
2781 static void coroutine_fn GRAPH_RDLOCK
2782 qcow2_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
2785 BDRVQcow2State
*s
= bs
->opaque
;
2786 BdrvChild
*data_file
;
2787 int flags
= s
->flags
;
2788 QCryptoBlock
*crypto
= NULL
;
2793 * Backing files are read-only which makes all of their metadata immutable,
2794 * that means we don't have to worry about reopening them here.
2801 * Do not reopen s->data_file (i.e., have qcow2_do_close() not close it,
2802 * and then prevent qcow2_do_open() from opening it), because this function
2803 * runs in the I/O path and as such we must not invoke global-state
2804 * functions like bdrv_unref_child() and bdrv_open_child().
2807 qcow2_do_close(bs
, false);
2809 data_file
= s
->data_file
;
2810 memset(s
, 0, sizeof(BDRVQcow2State
));
2811 s
->data_file
= data_file
;
2813 options
= qdict_clone_shallow(bs
->options
);
2815 flags
&= ~BDRV_O_INACTIVE
;
2816 qemu_co_mutex_lock(&s
->lock
);
2817 ret
= qcow2_do_open(bs
, options
, flags
, false, errp
);
2818 qemu_co_mutex_unlock(&s
->lock
);
2819 qobject_unref(options
);
2821 error_prepend(errp
, "Could not reopen qcow2 layer: ");
2829 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2830 size_t len
, size_t buflen
)
2832 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2833 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2835 if (buflen
< ext_len
) {
2839 *ext_backing_fmt
= (QCowExtension
) {
2840 .magic
= cpu_to_be32(magic
),
2841 .len
= cpu_to_be32(len
),
2845 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2852 * Updates the qcow2 header, including the variable length parts of it, i.e.
2853 * the backing file name and all extensions. qcow2 was not designed to allow
2854 * such changes, so if we run out of space (we can only use the first cluster)
2855 * this function may fail.
2857 * Returns 0 on success, -errno in error cases.
2859 int qcow2_update_header(BlockDriverState
*bs
)
2861 BDRVQcow2State
*s
= bs
->opaque
;
2864 size_t buflen
= s
->cluster_size
;
2866 uint64_t total_size
;
2867 uint32_t refcount_table_clusters
;
2868 size_t header_length
;
2869 Qcow2UnknownHeaderExtension
*uext
;
2871 buf
= qemu_blockalign(bs
, buflen
);
2873 /* Header structure */
2874 header
= (QCowHeader
*) buf
;
2876 if (buflen
< sizeof(*header
)) {
2881 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2882 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2883 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2885 ret
= validate_compression_type(s
, NULL
);
2890 *header
= (QCowHeader
) {
2891 /* Version 2 fields */
2892 .magic
= cpu_to_be32(QCOW_MAGIC
),
2893 .version
= cpu_to_be32(s
->qcow_version
),
2894 .backing_file_offset
= 0,
2895 .backing_file_size
= 0,
2896 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2897 .size
= cpu_to_be64(total_size
),
2898 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2899 .l1_size
= cpu_to_be32(s
->l1_size
),
2900 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2901 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2902 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2903 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2904 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2906 /* Version 3 fields */
2907 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2908 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2909 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2910 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2911 .header_length
= cpu_to_be32(header_length
),
2912 .compression_type
= s
->compression_type
,
2915 /* For older versions, write a shorter header */
2916 switch (s
->qcow_version
) {
2918 ret
= offsetof(QCowHeader
, incompatible_features
);
2921 ret
= sizeof(*header
);
2930 memset(buf
, 0, buflen
);
2932 /* Preserve any unknown field in the header */
2933 if (s
->unknown_header_fields_size
) {
2934 if (buflen
< s
->unknown_header_fields_size
) {
2939 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2940 buf
+= s
->unknown_header_fields_size
;
2941 buflen
-= s
->unknown_header_fields_size
;
2944 /* Backing file format header extension */
2945 if (s
->image_backing_format
) {
2946 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2947 s
->image_backing_format
,
2948 strlen(s
->image_backing_format
),
2958 /* External data file header extension */
2959 if (has_data_file(bs
) && s
->image_data_file
) {
2960 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_DATA_FILE
,
2961 s
->image_data_file
, strlen(s
->image_data_file
),
2971 /* Full disk encryption header pointer extension */
2972 if (s
->crypto_header
.offset
!= 0) {
2973 s
->crypto_header
.offset
= cpu_to_be64(s
->crypto_header
.offset
);
2974 s
->crypto_header
.length
= cpu_to_be64(s
->crypto_header
.length
);
2975 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2976 &s
->crypto_header
, sizeof(s
->crypto_header
),
2978 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
2979 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
2988 * Feature table. A mere 8 feature names occupies 392 bytes, and
2989 * when coupled with the v3 minimum header of 104 bytes plus the
2990 * 8-byte end-of-extension marker, that would leave only 8 bytes
2991 * for a backing file name in an image with 512-byte clusters.
2992 * Thus, we choose to omit this header for cluster sizes 4k and
2995 if (s
->qcow_version
>= 3 && s
->cluster_size
> 4096) {
2996 static const Qcow2Feature features
[] = {
2998 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2999 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
3000 .name
= "dirty bit",
3003 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3004 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
3005 .name
= "corrupt bit",
3008 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3009 .bit
= QCOW2_INCOMPAT_DATA_FILE_BITNR
,
3010 .name
= "external data file",
3013 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3014 .bit
= QCOW2_INCOMPAT_COMPRESSION_BITNR
,
3015 .name
= "compression type",
3018 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3019 .bit
= QCOW2_INCOMPAT_EXTL2_BITNR
,
3020 .name
= "extended L2 entries",
3023 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
3024 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
3025 .name
= "lazy refcounts",
3028 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
3029 .bit
= QCOW2_AUTOCLEAR_BITMAPS_BITNR
,
3033 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
3034 .bit
= QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR
,
3035 .name
= "raw external data",
3039 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
3040 features
, sizeof(features
), buflen
);
3048 /* Bitmap extension */
3049 if (s
->nb_bitmaps
> 0) {
3050 Qcow2BitmapHeaderExt bitmaps_header
= {
3051 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
3052 .bitmap_directory_size
=
3053 cpu_to_be64(s
->bitmap_directory_size
),
3054 .bitmap_directory_offset
=
3055 cpu_to_be64(s
->bitmap_directory_offset
)
3057 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
3058 &bitmaps_header
, sizeof(bitmaps_header
),
3067 /* Keep unknown header extensions */
3068 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
3069 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
3078 /* End of header extensions */
3079 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
3087 /* Backing file name */
3088 if (s
->image_backing_file
) {
3089 size_t backing_file_len
= strlen(s
->image_backing_file
);
3091 if (buflen
< backing_file_len
) {
3096 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3097 strncpy(buf
, s
->image_backing_file
, buflen
);
3099 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
3100 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
3103 /* Write the new header */
3104 ret
= bdrv_pwrite(bs
->file
, 0, s
->cluster_size
, header
, 0);
3115 static int qcow2_change_backing_file(BlockDriverState
*bs
,
3116 const char *backing_file
, const char *backing_fmt
)
3118 BDRVQcow2State
*s
= bs
->opaque
;
3120 /* Adding a backing file means that the external data file alone won't be
3121 * enough to make sense of the content */
3122 if (backing_file
&& data_file_is_raw(bs
)) {
3126 if (backing_file
&& strlen(backing_file
) > 1023) {
3130 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3131 backing_file
?: "");
3132 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3133 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3135 g_free(s
->image_backing_file
);
3136 g_free(s
->image_backing_format
);
3138 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
3139 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
3141 return qcow2_update_header(bs
);
3144 static int qcow2_set_up_encryption(BlockDriverState
*bs
,
3145 QCryptoBlockCreateOptions
*cryptoopts
,
3148 BDRVQcow2State
*s
= bs
->opaque
;
3149 QCryptoBlock
*crypto
= NULL
;
3152 switch (cryptoopts
->format
) {
3153 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
3154 fmt
= QCOW_CRYPT_LUKS
;
3156 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
3157 fmt
= QCOW_CRYPT_AES
;
3160 error_setg(errp
, "Crypto format not supported in qcow2");
3164 s
->crypt_method_header
= fmt
;
3166 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
3167 qcow2_crypto_hdr_init_func
,
3168 qcow2_crypto_hdr_write_func
,
3174 ret
= qcow2_update_header(bs
);
3176 error_setg_errno(errp
, -ret
, "Could not write encryption header");
3182 qcrypto_block_free(crypto
);
3187 * Preallocates metadata structures for data clusters between @offset (in the
3188 * guest disk) and @new_length (which is thus generally the new guest disk
3191 * Returns: 0 on success, -errno on failure.
3193 static int coroutine_fn GRAPH_RDLOCK
3194 preallocate_co(BlockDriverState
*bs
, uint64_t offset
, uint64_t new_length
,
3195 PreallocMode mode
, Error
**errp
)
3197 BDRVQcow2State
*s
= bs
->opaque
;
3199 uint64_t host_offset
= 0;
3200 int64_t file_length
;
3201 unsigned int cur_bytes
;
3203 QCowL2Meta
*meta
= NULL
, *m
;
3205 assert(offset
<= new_length
);
3206 bytes
= new_length
- offset
;
3209 cur_bytes
= MIN(bytes
, QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
));
3210 ret
= qcow2_alloc_host_offset(bs
, offset
, &cur_bytes
,
3211 &host_offset
, &meta
);
3213 error_setg_errno(errp
, -ret
, "Allocating clusters failed");
3217 for (m
= meta
; m
!= NULL
; m
= m
->next
) {
3221 ret
= qcow2_handle_l2meta(bs
, &meta
, true);
3223 error_setg_errno(errp
, -ret
, "Mapping clusters failed");
3227 /* TODO Preallocate data if requested */
3230 offset
+= cur_bytes
;
3234 * It is expected that the image file is large enough to actually contain
3235 * all of the allocated clusters (otherwise we get failing reads after
3236 * EOF). Extend the image to the last allocated sector.
3238 file_length
= bdrv_getlength(s
->data_file
->bs
);
3239 if (file_length
< 0) {
3240 error_setg_errno(errp
, -file_length
, "Could not get file size");
3245 if (host_offset
+ cur_bytes
> file_length
) {
3246 if (mode
== PREALLOC_MODE_METADATA
) {
3247 mode
= PREALLOC_MODE_OFF
;
3249 ret
= bdrv_co_truncate(s
->data_file
, host_offset
+ cur_bytes
, false,
3259 qcow2_handle_l2meta(bs
, &meta
, false);
3263 /* qcow2_refcount_metadata_size:
3264 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3265 * @cluster_size: size of a cluster, in bytes
3266 * @refcount_order: refcount bits power-of-2 exponent
3267 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3270 * Returns: Number of bytes required for refcount blocks and table metadata.
3272 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
3273 int refcount_order
, bool generous_increase
,
3274 uint64_t *refblock_count
)
3277 * Every host cluster is reference-counted, including metadata (even
3278 * refcount metadata is recursively included).
3280 * An accurate formula for the size of refcount metadata size is difficult
3281 * to derive. An easier method of calculation is finding the fixed point
3282 * where no further refcount blocks or table clusters are required to
3283 * reference count every cluster.
3285 int64_t blocks_per_table_cluster
= cluster_size
/ REFTABLE_ENTRY_SIZE
;
3286 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
3287 int64_t table
= 0; /* number of refcount table clusters */
3288 int64_t blocks
= 0; /* number of refcount block clusters */
3294 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
3295 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
3296 n
= clusters
+ blocks
+ table
;
3298 if (n
== last
&& generous_increase
) {
3299 clusters
+= DIV_ROUND_UP(table
, 2);
3300 n
= 0; /* force another loop */
3301 generous_increase
= false;
3303 } while (n
!= last
);
3305 if (refblock_count
) {
3306 *refblock_count
= blocks
;
3309 return (blocks
+ table
) * cluster_size
;
3313 * qcow2_calc_prealloc_size:
3314 * @total_size: virtual disk size in bytes
3315 * @cluster_size: cluster size in bytes
3316 * @refcount_order: refcount bits power-of-2 exponent
3317 * @extended_l2: true if the image has extended L2 entries
3319 * Returns: Total number of bytes required for the fully allocated image
3320 * (including metadata).
3322 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
3323 size_t cluster_size
,
3327 int64_t meta_size
= 0;
3328 uint64_t nl1e
, nl2e
;
3329 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
3330 size_t l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
3332 /* header: 1 cluster */
3333 meta_size
+= cluster_size
;
3335 /* total size of L2 tables */
3336 nl2e
= aligned_total_size
/ cluster_size
;
3337 nl2e
= ROUND_UP(nl2e
, cluster_size
/ l2e_size
);
3338 meta_size
+= nl2e
* l2e_size
;
3340 /* total size of L1 tables */
3341 nl1e
= nl2e
* l2e_size
/ cluster_size
;
3342 nl1e
= ROUND_UP(nl1e
, cluster_size
/ L1E_SIZE
);
3343 meta_size
+= nl1e
* L1E_SIZE
;
3345 /* total size of refcount table and blocks */
3346 meta_size
+= qcow2_refcount_metadata_size(
3347 (meta_size
+ aligned_total_size
) / cluster_size
,
3348 cluster_size
, refcount_order
, false, NULL
);
3350 return meta_size
+ aligned_total_size
;
3353 static bool validate_cluster_size(size_t cluster_size
, bool extended_l2
,
3356 int cluster_bits
= ctz32(cluster_size
);
3357 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
3358 (1 << cluster_bits
) != cluster_size
)
3360 error_setg(errp
, "Cluster size must be a power of two between %d and "
3361 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
3366 unsigned min_cluster_size
=
3367 (1 << MIN_CLUSTER_BITS
) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
;
3368 if (cluster_size
< min_cluster_size
) {
3369 error_setg(errp
, "Extended L2 entries are only supported with "
3370 "cluster sizes of at least %u bytes", min_cluster_size
);
3378 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, bool extended_l2
,
3381 size_t cluster_size
;
3383 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
3384 DEFAULT_CLUSTER_SIZE
);
3385 if (!validate_cluster_size(cluster_size
, extended_l2
, errp
)) {
3388 return cluster_size
;
3391 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
3396 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
3398 ret
= 3; /* default */
3399 } else if (!strcmp(buf
, "0.10")) {
3401 } else if (!strcmp(buf
, "1.1")) {
3404 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
3411 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
3414 uint64_t refcount_bits
;
3416 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
3417 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
3418 error_setg(errp
, "Refcount width must be a power of two and may not "
3423 if (version
< 3 && refcount_bits
!= 16) {
3424 error_setg(errp
, "Different refcount widths than 16 bits require "
3425 "compatibility level 1.1 or above (use compat=1.1 or "
3430 return refcount_bits
;
3433 static int coroutine_fn
3434 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
3436 BlockdevCreateOptionsQcow2
*qcow2_opts
;
3440 * Open the image file and write a minimal qcow2 header.
3442 * We keep things simple and start with a zero-sized image. We also
3443 * do without refcount blocks or a L1 table for now. We'll fix the
3444 * inconsistency later.
3446 * We do need a refcount table because growing the refcount table means
3447 * allocating two new refcount blocks - the second of which would be at
3448 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3449 * size for any qcow2 image.
3451 BlockBackend
*blk
= NULL
;
3452 BlockDriverState
*bs
= NULL
;
3453 BlockDriverState
*data_bs
= NULL
;
3455 size_t cluster_size
;
3458 uint64_t *refcount_table
;
3460 uint8_t compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
3462 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
3463 qcow2_opts
= &create_options
->u
.qcow2
;
3465 bs
= bdrv_co_open_blockdev_ref(qcow2_opts
->file
, errp
);
3470 /* Validate options and set default values */
3471 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
3472 error_setg(errp
, "Image size must be a multiple of %u bytes",
3473 (unsigned) BDRV_SECTOR_SIZE
);
3478 if (qcow2_opts
->has_version
) {
3479 switch (qcow2_opts
->version
) {
3480 case BLOCKDEV_QCOW2_VERSION_V2
:
3483 case BLOCKDEV_QCOW2_VERSION_V3
:
3487 g_assert_not_reached();
3493 if (qcow2_opts
->has_cluster_size
) {
3494 cluster_size
= qcow2_opts
->cluster_size
;
3496 cluster_size
= DEFAULT_CLUSTER_SIZE
;
3499 if (!qcow2_opts
->has_extended_l2
) {
3500 qcow2_opts
->extended_l2
= false;
3502 if (qcow2_opts
->extended_l2
) {
3504 error_setg(errp
, "Extended L2 entries are only supported with "
3505 "compatibility level 1.1 and above (use version=v3 or "
3512 if (!validate_cluster_size(cluster_size
, qcow2_opts
->extended_l2
, errp
)) {
3517 if (!qcow2_opts
->has_preallocation
) {
3518 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
3520 if (qcow2_opts
->backing_file
&&
3521 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
&&
3522 !qcow2_opts
->extended_l2
)
3524 error_setg(errp
, "Backing file and preallocation can only be used at "
3525 "the same time if extended_l2 is on");
3529 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->backing_file
) {
3530 error_setg(errp
, "Backing format cannot be used without backing file");
3535 if (!qcow2_opts
->has_lazy_refcounts
) {
3536 qcow2_opts
->lazy_refcounts
= false;
3538 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
3539 error_setg(errp
, "Lazy refcounts only supported with compatibility "
3540 "level 1.1 and above (use version=v3 or greater)");
3545 if (!qcow2_opts
->has_refcount_bits
) {
3546 qcow2_opts
->refcount_bits
= 16;
3548 if (qcow2_opts
->refcount_bits
> 64 ||
3549 !is_power_of_2(qcow2_opts
->refcount_bits
))
3551 error_setg(errp
, "Refcount width must be a power of two and may not "
3556 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
3557 error_setg(errp
, "Different refcount widths than 16 bits require "
3558 "compatibility level 1.1 or above (use version=v3 or "
3563 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
3565 if (qcow2_opts
->data_file_raw
&& !qcow2_opts
->data_file
) {
3566 error_setg(errp
, "data-file-raw requires data-file");
3570 if (qcow2_opts
->data_file_raw
&& qcow2_opts
->backing_file
) {
3571 error_setg(errp
, "Backing file and data-file-raw cannot be used at "
3576 if (qcow2_opts
->data_file_raw
&&
3577 qcow2_opts
->preallocation
== PREALLOC_MODE_OFF
)
3580 * data-file-raw means that "the external data file can be
3581 * read as a consistent standalone raw image without looking
3582 * at the qcow2 metadata." It does not say that the metadata
3583 * must be ignored, though (and the qcow2 driver in fact does
3584 * not ignore it), so the L1/L2 tables must be present and
3585 * give a 1:1 mapping, so you get the same result regardless
3586 * of whether you look at the metadata or whether you ignore
3589 qcow2_opts
->preallocation
= PREALLOC_MODE_METADATA
;
3592 * Cannot use preallocation with backing files, but giving a
3593 * backing file when specifying data_file_raw is an error
3596 assert(!qcow2_opts
->backing_file
);
3599 if (qcow2_opts
->data_file
) {
3601 error_setg(errp
, "External data files are only supported with "
3602 "compatibility level 1.1 and above (use version=v3 or "
3607 data_bs
= bdrv_co_open_blockdev_ref(qcow2_opts
->data_file
, errp
);
3608 if (data_bs
== NULL
) {
3614 if (qcow2_opts
->has_compression_type
&&
3615 qcow2_opts
->compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3620 error_setg(errp
, "Non-zlib compression type is only supported with "
3621 "compatibility level 1.1 and above (use version=v3 or "
3626 switch (qcow2_opts
->compression_type
) {
3628 case QCOW2_COMPRESSION_TYPE_ZSTD
:
3632 error_setg(errp
, "Unknown compression type");
3636 compression_type
= qcow2_opts
->compression_type
;
3639 /* Create BlockBackend to write to the image */
3640 blk
= blk_co_new_with_bs(bs
, BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
,
3646 blk_set_allow_write_beyond_eof(blk
, true);
3648 /* Write the header */
3649 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
3650 header
= g_malloc0(cluster_size
);
3651 *header
= (QCowHeader
) {
3652 .magic
= cpu_to_be32(QCOW_MAGIC
),
3653 .version
= cpu_to_be32(version
),
3654 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
3655 .size
= cpu_to_be64(0),
3656 .l1_table_offset
= cpu_to_be64(0),
3657 .l1_size
= cpu_to_be32(0),
3658 .refcount_table_offset
= cpu_to_be64(cluster_size
),
3659 .refcount_table_clusters
= cpu_to_be32(1),
3660 .refcount_order
= cpu_to_be32(refcount_order
),
3661 /* don't deal with endianness since compression_type is 1 byte long */
3662 .compression_type
= compression_type
,
3663 .header_length
= cpu_to_be32(sizeof(*header
)),
3666 /* We'll update this to correct value later */
3667 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
3669 if (qcow2_opts
->lazy_refcounts
) {
3670 header
->compatible_features
|=
3671 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
3674 header
->incompatible_features
|=
3675 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE
);
3677 if (qcow2_opts
->data_file_raw
) {
3678 header
->autoclear_features
|=
3679 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW
);
3681 if (compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3682 header
->incompatible_features
|=
3683 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION
);
3686 if (qcow2_opts
->extended_l2
) {
3687 header
->incompatible_features
|=
3688 cpu_to_be64(QCOW2_INCOMPAT_EXTL2
);
3691 ret
= blk_co_pwrite(blk
, 0, cluster_size
, header
, 0);
3694 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
3698 /* Write a refcount table with one refcount block */
3699 refcount_table
= g_malloc0(2 * cluster_size
);
3700 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
3701 ret
= blk_co_pwrite(blk
, cluster_size
, 2 * cluster_size
, refcount_table
, 0);
3702 g_free(refcount_table
);
3705 error_setg_errno(errp
, -ret
, "Could not write refcount table");
3713 * And now open the image and make it consistent first (i.e. increase the
3714 * refcount of the cluster that is occupied by the header and the refcount
3717 options
= qdict_new();
3718 qdict_put_str(options
, "driver", "qcow2");
3719 qdict_put_str(options
, "file", bs
->node_name
);
3721 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3723 blk
= blk_co_new_open(NULL
, NULL
, options
,
3724 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
3731 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
3733 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
3734 "header and refcount table");
3737 } else if (ret
!= 0) {
3738 error_report("Huh, first cluster in empty image is already in use?");
3742 /* Set the external data file if necessary */
3744 BDRVQcow2State
*s
= blk_bs(blk
)->opaque
;
3745 s
->image_data_file
= g_strdup(data_bs
->filename
);
3748 /* Create a full header (including things like feature table) */
3749 ret
= qcow2_update_header(blk_bs(blk
));
3751 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
3755 /* Okay, now that we have a valid image, let's give it the right size */
3756 ret
= blk_co_truncate(blk
, qcow2_opts
->size
, false,
3757 qcow2_opts
->preallocation
, 0, errp
);
3759 error_prepend(errp
, "Could not resize image: ");
3763 /* Want a backing file? There you go. */
3764 if (qcow2_opts
->backing_file
) {
3765 const char *backing_format
= NULL
;
3767 if (qcow2_opts
->has_backing_fmt
) {
3768 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
3771 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
3772 backing_format
, false);
3774 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
3775 "with format '%s'", qcow2_opts
->backing_file
,
3781 /* Want encryption? There you go. */
3782 if (qcow2_opts
->encrypt
) {
3783 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3792 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3793 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3794 * have to setup decryption context. We're not doing any I/O on the top
3795 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3798 options
= qdict_new();
3799 qdict_put_str(options
, "driver", "qcow2");
3800 qdict_put_str(options
, "file", bs
->node_name
);
3802 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3804 blk
= blk_co_new_open(NULL
, NULL
, options
,
3805 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3816 bdrv_unref(data_bs
);
3820 static int coroutine_fn GRAPH_RDLOCK
3821 qcow2_co_create_opts(BlockDriver
*drv
, const char *filename
, QemuOpts
*opts
,
3824 BlockdevCreateOptions
*create_options
= NULL
;
3827 BlockDriverState
*bs
= NULL
;
3828 BlockDriverState
*data_bs
= NULL
;
3832 /* Only the keyval visitor supports the dotted syntax needed for
3833 * encryption, so go through a QDict before getting a QAPI type. Ignore
3834 * options meant for the protocol layer so that the visitor doesn't
3836 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3839 /* Handle encryption options */
3840 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3841 if (val
&& !strcmp(val
, "on")) {
3842 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3843 } else if (val
&& !strcmp(val
, "off")) {
3844 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3847 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3848 if (val
&& !strcmp(val
, "aes")) {
3849 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3852 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3853 * version=v2/v3 below. */
3854 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3855 if (val
&& !strcmp(val
, "0.10")) {
3856 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3857 } else if (val
&& !strcmp(val
, "1.1")) {
3858 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3861 /* Change legacy command line options into QMP ones */
3862 static const QDictRenames opt_renames
[] = {
3863 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3864 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3865 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3866 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3867 { BLOCK_OPT_EXTL2
, "extended-l2" },
3868 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3869 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3870 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3871 { BLOCK_OPT_DATA_FILE_RAW
, "data-file-raw" },
3872 { BLOCK_OPT_COMPRESSION_TYPE
, "compression-type" },
3876 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3881 /* Create and open the file (protocol layer) */
3882 ret
= bdrv_co_create_file(filename
, opts
, errp
);
3887 bs
= bdrv_co_open(filename
, NULL
, NULL
,
3888 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3894 /* Create and open an external data file (protocol layer) */
3895 val
= qdict_get_try_str(qdict
, BLOCK_OPT_DATA_FILE
);
3897 ret
= bdrv_co_create_file(val
, opts
, errp
);
3902 data_bs
= bdrv_co_open(val
, NULL
, NULL
,
3903 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
,
3905 if (data_bs
== NULL
) {
3910 qdict_del(qdict
, BLOCK_OPT_DATA_FILE
);
3911 qdict_put_str(qdict
, "data-file", data_bs
->node_name
);
3914 /* Set 'driver' and 'node' options */
3915 qdict_put_str(qdict
, "driver", "qcow2");
3916 qdict_put_str(qdict
, "file", bs
->node_name
);
3918 /* Now get the QAPI type BlockdevCreateOptions */
3919 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
3925 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, errp
);
3927 if (!create_options
) {
3932 /* Silently round up size */
3933 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3936 /* Create the qcow2 image (format layer) */
3937 ret
= qcow2_co_create(create_options
, errp
);
3940 bdrv_co_delete_file_noerr(bs
);
3941 bdrv_co_delete_file_noerr(data_bs
);
3946 qobject_unref(qdict
);
3948 bdrv_unref(data_bs
);
3949 qapi_free_BlockdevCreateOptions(create_options
);
3954 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3959 /* Clamp to image length, before checking status of underlying sectors */
3960 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3961 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3969 * bdrv_block_status_above doesn't merge different types of zeros, for
3970 * example, zeros which come from the region which is unallocated in
3971 * the whole backing chain, and zeros which come because of a short
3972 * backing file. So, we need a loop.
3975 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3978 } while (res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
&& bytes
);
3980 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && bytes
== 0;
3983 static int coroutine_fn GRAPH_RDLOCK
3984 qcow2_co_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
3985 BdrvRequestFlags flags
)
3988 BDRVQcow2State
*s
= bs
->opaque
;
3990 uint32_t head
= offset_into_subcluster(s
, offset
);
3991 uint32_t tail
= ROUND_UP(offset
+ bytes
, s
->subcluster_size
) -
3994 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3995 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
4002 QCow2SubclusterType type
;
4004 assert(head
+ bytes
+ tail
<= s
->subcluster_size
);
4006 /* check whether remainder of cluster already reads as zero */
4007 if (!(is_zero(bs
, offset
- head
, head
) &&
4008 is_zero(bs
, offset
+ bytes
, tail
))) {
4012 qemu_co_mutex_lock(&s
->lock
);
4013 /* We can have new write after previous check */
4015 bytes
= s
->subcluster_size
;
4016 nr
= s
->subcluster_size
;
4017 ret
= qcow2_get_host_offset(bs
, offset
, &nr
, &off
, &type
);
4019 (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
4020 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&&
4021 type
!= QCOW2_SUBCLUSTER_ZERO_PLAIN
&&
4022 type
!= QCOW2_SUBCLUSTER_ZERO_ALLOC
)) {
4023 qemu_co_mutex_unlock(&s
->lock
);
4024 return ret
< 0 ? ret
: -ENOTSUP
;
4027 qemu_co_mutex_lock(&s
->lock
);
4030 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
4032 /* Whatever is left can use real zero subclusters */
4033 ret
= qcow2_subcluster_zeroize(bs
, offset
, bytes
, flags
);
4034 qemu_co_mutex_unlock(&s
->lock
);
4039 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
4040 int64_t offset
, int64_t bytes
)
4043 BDRVQcow2State
*s
= bs
->opaque
;
4045 /* If the image does not support QCOW_OFLAG_ZERO then discarding
4046 * clusters could expose stale data from the backing file. */
4047 if (s
->qcow_version
< 3 && bs
->backing
) {
4051 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
4052 assert(bytes
< s
->cluster_size
);
4053 /* Ignore partial clusters, except for the special case of the
4054 * complete partial cluster at the end of an unaligned file */
4055 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
4056 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
4061 qemu_co_mutex_lock(&s
->lock
);
4062 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
4064 qemu_co_mutex_unlock(&s
->lock
);
4068 static int coroutine_fn GRAPH_RDLOCK
4069 qcow2_co_copy_range_from(BlockDriverState
*bs
,
4070 BdrvChild
*src
, int64_t src_offset
,
4071 BdrvChild
*dst
, int64_t dst_offset
,
4072 int64_t bytes
, BdrvRequestFlags read_flags
,
4073 BdrvRequestFlags write_flags
)
4075 BDRVQcow2State
*s
= bs
->opaque
;
4077 unsigned int cur_bytes
; /* number of bytes in current iteration */
4078 BdrvChild
*child
= NULL
;
4079 BdrvRequestFlags cur_write_flags
;
4081 assert(!bs
->encrypted
);
4082 qemu_co_mutex_lock(&s
->lock
);
4084 while (bytes
!= 0) {
4085 uint64_t copy_offset
= 0;
4086 QCow2SubclusterType type
;
4087 /* prepare next request */
4088 cur_bytes
= MIN(bytes
, INT_MAX
);
4089 cur_write_flags
= write_flags
;
4091 ret
= qcow2_get_host_offset(bs
, src_offset
, &cur_bytes
,
4092 ©_offset
, &type
);
4098 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
4099 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
4100 if (bs
->backing
&& bs
->backing
->bs
) {
4101 int64_t backing_length
= bdrv_getlength(bs
->backing
->bs
);
4102 if (src_offset
>= backing_length
) {
4103 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4105 child
= bs
->backing
;
4106 cur_bytes
= MIN(cur_bytes
, backing_length
- src_offset
);
4107 copy_offset
= src_offset
;
4110 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4114 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
4115 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
4116 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4119 case QCOW2_SUBCLUSTER_COMPRESSED
:
4123 case QCOW2_SUBCLUSTER_NORMAL
:
4124 child
= s
->data_file
;
4130 qemu_co_mutex_unlock(&s
->lock
);
4131 ret
= bdrv_co_copy_range_from(child
,
4134 cur_bytes
, read_flags
, cur_write_flags
);
4135 qemu_co_mutex_lock(&s
->lock
);
4141 src_offset
+= cur_bytes
;
4142 dst_offset
+= cur_bytes
;
4147 qemu_co_mutex_unlock(&s
->lock
);
4151 static int coroutine_fn GRAPH_RDLOCK
4152 qcow2_co_copy_range_to(BlockDriverState
*bs
,
4153 BdrvChild
*src
, int64_t src_offset
,
4154 BdrvChild
*dst
, int64_t dst_offset
,
4155 int64_t bytes
, BdrvRequestFlags read_flags
,
4156 BdrvRequestFlags write_flags
)
4158 BDRVQcow2State
*s
= bs
->opaque
;
4160 unsigned int cur_bytes
; /* number of sectors in current iteration */
4161 uint64_t host_offset
;
4162 QCowL2Meta
*l2meta
= NULL
;
4164 assert(!bs
->encrypted
);
4166 qemu_co_mutex_lock(&s
->lock
);
4168 while (bytes
!= 0) {
4172 cur_bytes
= MIN(bytes
, INT_MAX
);
4175 * If src->bs == dst->bs, we could simply copy by incrementing
4176 * the refcnt, without copying user data.
4177 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4178 ret
= qcow2_alloc_host_offset(bs
, dst_offset
, &cur_bytes
,
4179 &host_offset
, &l2meta
);
4184 ret
= qcow2_pre_write_overlap_check(bs
, 0, host_offset
, cur_bytes
,
4190 qemu_co_mutex_unlock(&s
->lock
);
4191 ret
= bdrv_co_copy_range_to(src
, src_offset
, s
->data_file
, host_offset
,
4192 cur_bytes
, read_flags
, write_flags
);
4193 qemu_co_mutex_lock(&s
->lock
);
4198 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
4204 src_offset
+= cur_bytes
;
4205 dst_offset
+= cur_bytes
;
4210 qcow2_handle_l2meta(bs
, &l2meta
, false);
4212 qemu_co_mutex_unlock(&s
->lock
);
4214 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
4219 static int coroutine_fn GRAPH_RDLOCK
4220 qcow2_co_truncate(BlockDriverState
*bs
, int64_t offset
, bool exact
,
4221 PreallocMode prealloc
, BdrvRequestFlags flags
, Error
**errp
)
4223 BDRVQcow2State
*s
= bs
->opaque
;
4224 uint64_t old_length
;
4225 int64_t new_l1_size
;
4229 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
4230 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
4232 error_setg(errp
, "Unsupported preallocation mode '%s'",
4233 PreallocMode_str(prealloc
));
4237 if (!QEMU_IS_ALIGNED(offset
, BDRV_SECTOR_SIZE
)) {
4238 error_setg(errp
, "The new size must be a multiple of %u",
4239 (unsigned) BDRV_SECTOR_SIZE
);
4243 qemu_co_mutex_lock(&s
->lock
);
4246 * Even though we store snapshot size for all images, it was not
4247 * required until v3, so it is not safe to proceed for v2.
4249 if (s
->nb_snapshots
&& s
->qcow_version
< 3) {
4250 error_setg(errp
, "Can't resize a v2 image which has snapshots");
4255 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4256 if (qcow2_truncate_bitmaps_check(bs
, errp
)) {
4261 old_length
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4262 new_l1_size
= size_to_l1(s
, offset
);
4264 if (offset
< old_length
) {
4265 int64_t last_cluster
, old_file_size
;
4266 if (prealloc
!= PREALLOC_MODE_OFF
) {
4268 "Preallocation can't be used for shrinking an image");
4273 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
4274 old_length
- ROUND_UP(offset
,
4276 QCOW2_DISCARD_ALWAYS
, true);
4278 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
4282 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
4284 error_setg_errno(errp
, -ret
,
4285 "Failed to reduce the number of L2 tables");
4289 ret
= qcow2_shrink_reftable(bs
);
4291 error_setg_errno(errp
, -ret
,
4292 "Failed to discard unused refblocks");
4296 old_file_size
= bdrv_getlength(bs
->file
->bs
);
4297 if (old_file_size
< 0) {
4298 error_setg_errno(errp
, -old_file_size
,
4299 "Failed to inquire current file length");
4300 ret
= old_file_size
;
4303 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4304 if (last_cluster
< 0) {
4305 error_setg_errno(errp
, -last_cluster
,
4306 "Failed to find the last cluster");
4310 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
4311 Error
*local_err
= NULL
;
4314 * Do not pass @exact here: It will not help the user if
4315 * we get an error here just because they wanted to shrink
4316 * their qcow2 image (on a block device) with qemu-img.
4317 * (And on the qcow2 layer, the @exact requirement is
4318 * always fulfilled, so there is no need to pass it on.)
4320 bdrv_co_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
4321 false, PREALLOC_MODE_OFF
, 0, &local_err
);
4323 warn_reportf_err(local_err
,
4324 "Failed to truncate the tail of the image: ");
4328 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
4330 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
4334 if (data_file_is_raw(bs
) && prealloc
== PREALLOC_MODE_OFF
) {
4336 * When creating a qcow2 image with data-file-raw, we enforce
4337 * at least prealloc=metadata, so that the L1/L2 tables are
4338 * fully allocated and reading from the data file will return
4339 * the same data as reading from the qcow2 image. When the
4340 * image is grown, we must consequently preallocate the
4341 * metadata structures to cover the added area.
4343 prealloc
= PREALLOC_MODE_METADATA
;
4348 case PREALLOC_MODE_OFF
:
4349 if (has_data_file(bs
)) {
4351 * If the caller wants an exact resize, the external data
4352 * file should be resized to the exact target size, too,
4353 * so we pass @exact here.
4355 ret
= bdrv_co_truncate(s
->data_file
, offset
, exact
, prealloc
, 0,
4363 case PREALLOC_MODE_METADATA
:
4364 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4370 case PREALLOC_MODE_FALLOC
:
4371 case PREALLOC_MODE_FULL
:
4373 int64_t allocation_start
, host_offset
, guest_offset
;
4374 int64_t clusters_allocated
;
4375 int64_t old_file_size
, last_cluster
, new_file_size
;
4376 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
4377 bool subclusters_need_allocation
= false;
4379 /* With a data file, preallocation means just allocating the metadata
4380 * and forwarding the truncate request to the data file */
4381 if (has_data_file(bs
)) {
4382 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4389 old_file_size
= bdrv_getlength(bs
->file
->bs
);
4390 if (old_file_size
< 0) {
4391 error_setg_errno(errp
, -old_file_size
,
4392 "Failed to inquire current file length");
4393 ret
= old_file_size
;
4397 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4398 if (last_cluster
>= 0) {
4399 old_file_size
= (last_cluster
+ 1) * s
->cluster_size
;
4401 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
4404 nb_new_data_clusters
= (ROUND_UP(offset
, s
->cluster_size
) -
4405 start_of_cluster(s
, old_length
)) >> s
->cluster_bits
;
4407 /* This is an overestimation; we will not actually allocate space for
4408 * these in the file but just make sure the new refcount structures are
4409 * able to cover them so we will not have to allocate new refblocks
4410 * while entering the data blocks in the potentially new L2 tables.
4411 * (We do not actually care where the L2 tables are placed. Maybe they
4412 * are already allocated or they can be placed somewhere before
4413 * @old_file_size. It does not matter because they will be fully
4414 * allocated automatically, so they do not need to be covered by the
4415 * preallocation. All that matters is that we will not have to allocate
4416 * new refcount structures for them.) */
4417 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
4418 s
->cluster_size
/ l2_entry_size(s
));
4419 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4420 * table for a potential head/tail */
4423 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
4424 nb_new_data_clusters
+
4427 if (allocation_start
< 0) {
4428 error_setg_errno(errp
, -allocation_start
,
4429 "Failed to resize refcount structures");
4430 ret
= allocation_start
;
4434 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
4435 nb_new_data_clusters
);
4436 if (clusters_allocated
< 0) {
4437 error_setg_errno(errp
, -clusters_allocated
,
4438 "Failed to allocate data clusters");
4439 ret
= clusters_allocated
;
4443 assert(clusters_allocated
== nb_new_data_clusters
);
4445 /* Allocate the data area */
4446 new_file_size
= allocation_start
+
4447 nb_new_data_clusters
* s
->cluster_size
;
4449 * Image file grows, so @exact does not matter.
4451 * If we need to zero out the new area, try first whether the protocol
4452 * driver can already take care of this.
4454 if (flags
& BDRV_REQ_ZERO_WRITE
) {
4455 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
,
4456 BDRV_REQ_ZERO_WRITE
, NULL
);
4458 flags
&= ~BDRV_REQ_ZERO_WRITE
;
4459 /* Ensure that we read zeroes and not backing file data */
4460 subclusters_need_allocation
= true;
4466 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
, 0,
4470 error_prepend(errp
, "Failed to resize underlying file: ");
4471 qcow2_free_clusters(bs
, allocation_start
,
4472 nb_new_data_clusters
* s
->cluster_size
,
4473 QCOW2_DISCARD_OTHER
);
4477 /* Create the necessary L2 entries */
4478 host_offset
= allocation_start
;
4479 guest_offset
= old_length
;
4480 while (nb_new_data_clusters
) {
4481 int64_t nb_clusters
= MIN(
4482 nb_new_data_clusters
,
4483 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
4484 unsigned cow_start_length
= offset_into_cluster(s
, guest_offset
);
4485 QCowL2Meta allocation
;
4486 guest_offset
= start_of_cluster(s
, guest_offset
);
4487 allocation
= (QCowL2Meta
) {
4488 .offset
= guest_offset
,
4489 .alloc_offset
= host_offset
,
4490 .nb_clusters
= nb_clusters
,
4493 .nb_bytes
= cow_start_length
,
4496 .offset
= nb_clusters
<< s
->cluster_bits
,
4499 .prealloc
= !subclusters_need_allocation
,
4501 qemu_co_queue_init(&allocation
.dependent_requests
);
4503 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
4505 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
4506 qcow2_free_clusters(bs
, host_offset
,
4507 nb_new_data_clusters
* s
->cluster_size
,
4508 QCOW2_DISCARD_OTHER
);
4512 guest_offset
+= nb_clusters
* s
->cluster_size
;
4513 host_offset
+= nb_clusters
* s
->cluster_size
;
4514 nb_new_data_clusters
-= nb_clusters
;
4520 g_assert_not_reached();
4523 if ((flags
& BDRV_REQ_ZERO_WRITE
) && offset
> old_length
) {
4524 uint64_t zero_start
= QEMU_ALIGN_UP(old_length
, s
->subcluster_size
);
4527 * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4528 * requires a subcluster-aligned start. The end may be unaligned if
4529 * it is at the end of the image (which it is here).
4531 if (offset
> zero_start
) {
4532 ret
= qcow2_subcluster_zeroize(bs
, zero_start
, offset
- zero_start
,
4535 error_setg_errno(errp
, -ret
, "Failed to zero out new clusters");
4540 /* Write explicit zeros for the unaligned head */
4541 if (zero_start
> old_length
) {
4542 uint64_t len
= MIN(zero_start
, offset
) - old_length
;
4543 uint8_t *buf
= qemu_blockalign0(bs
, len
);
4545 qemu_iovec_init_buf(&qiov
, buf
, len
);
4547 qemu_co_mutex_unlock(&s
->lock
);
4548 ret
= qcow2_co_pwritev_part(bs
, old_length
, len
, &qiov
, 0, 0);
4549 qemu_co_mutex_lock(&s
->lock
);
4553 error_setg_errno(errp
, -ret
, "Failed to zero out the new area");
4559 if (prealloc
!= PREALLOC_MODE_OFF
) {
4560 /* Flush metadata before actually changing the image size */
4561 ret
= qcow2_write_caches(bs
);
4563 error_setg_errno(errp
, -ret
,
4564 "Failed to flush the preallocated area to disk");
4569 bs
->total_sectors
= offset
/ BDRV_SECTOR_SIZE
;
4571 /* write updated header.size */
4572 offset
= cpu_to_be64(offset
);
4573 ret
= bdrv_co_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
4574 sizeof(offset
), &offset
, 0);
4576 error_setg_errno(errp
, -ret
, "Failed to update the image size");
4580 s
->l1_vm_state_index
= new_l1_size
;
4582 /* Update cache sizes */
4583 options
= qdict_clone_shallow(bs
->options
);
4584 ret
= qcow2_update_options(bs
, options
, s
->flags
, errp
);
4585 qobject_unref(options
);
4591 qemu_co_mutex_unlock(&s
->lock
);
4595 static int coroutine_fn GRAPH_RDLOCK
4596 qcow2_co_pwritev_compressed_task(BlockDriverState
*bs
,
4597 uint64_t offset
, uint64_t bytes
,
4598 QEMUIOVector
*qiov
, size_t qiov_offset
)
4600 BDRVQcow2State
*s
= bs
->opaque
;
4603 uint8_t *buf
, *out_buf
;
4604 uint64_t cluster_offset
;
4606 assert(bytes
== s
->cluster_size
|| (bytes
< s
->cluster_size
&&
4607 (offset
+ bytes
== bs
->total_sectors
<< BDRV_SECTOR_BITS
)));
4609 buf
= qemu_blockalign(bs
, s
->cluster_size
);
4610 if (bytes
< s
->cluster_size
) {
4611 /* Zero-pad last write if image size is not cluster aligned */
4612 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
4614 qemu_iovec_to_buf(qiov
, qiov_offset
, buf
, bytes
);
4616 out_buf
= g_malloc(s
->cluster_size
);
4618 out_len
= qcow2_co_compress(bs
, out_buf
, s
->cluster_size
- 1,
4619 buf
, s
->cluster_size
);
4620 if (out_len
== -ENOMEM
) {
4621 /* could not compress: write normal cluster */
4622 ret
= qcow2_co_pwritev_part(bs
, offset
, bytes
, qiov
, qiov_offset
, 0);
4627 } else if (out_len
< 0) {
4632 qemu_co_mutex_lock(&s
->lock
);
4633 ret
= qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
,
4636 qemu_co_mutex_unlock(&s
->lock
);
4640 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
, true);
4641 qemu_co_mutex_unlock(&s
->lock
);
4646 BLKDBG_EVENT(s
->data_file
, BLKDBG_WRITE_COMPRESSED
);
4647 ret
= bdrv_co_pwrite(s
->data_file
, cluster_offset
, out_len
, out_buf
, 0);
4660 * This function can count as GRAPH_RDLOCK because
4661 * qcow2_co_pwritev_compressed_part() holds the graph lock and keeps it until
4662 * this coroutine has terminated.
4664 static int coroutine_fn GRAPH_RDLOCK
4665 qcow2_co_pwritev_compressed_task_entry(AioTask
*task
)
4667 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
4669 assert(!t
->subcluster_type
&& !t
->l2meta
);
4671 return qcow2_co_pwritev_compressed_task(t
->bs
, t
->offset
, t
->bytes
, t
->qiov
,
4676 * XXX: put compressed sectors first, then all the cluster aligned
4677 * tables to avoid losing bytes in alignment
4679 static int coroutine_fn GRAPH_RDLOCK
4680 qcow2_co_pwritev_compressed_part(BlockDriverState
*bs
,
4681 int64_t offset
, int64_t bytes
,
4682 QEMUIOVector
*qiov
, size_t qiov_offset
)
4684 BDRVQcow2State
*s
= bs
->opaque
;
4685 AioTaskPool
*aio
= NULL
;
4688 if (has_data_file(bs
)) {
4694 * align end of file to a sector boundary to ease reading with
4697 int64_t len
= bdrv_getlength(bs
->file
->bs
);
4701 return bdrv_co_truncate(bs
->file
, len
, false, PREALLOC_MODE_OFF
, 0,
4705 if (offset_into_cluster(s
, offset
)) {
4709 if (offset_into_cluster(s
, bytes
) &&
4710 (offset
+ bytes
) != (bs
->total_sectors
<< BDRV_SECTOR_BITS
)) {
4714 while (bytes
&& aio_task_pool_status(aio
) == 0) {
4715 uint64_t chunk_size
= MIN(bytes
, s
->cluster_size
);
4717 if (!aio
&& chunk_size
!= bytes
) {
4718 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
4721 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_compressed_task_entry
,
4722 0, 0, offset
, chunk_size
, qiov
, qiov_offset
, NULL
);
4726 qiov_offset
+= chunk_size
;
4727 offset
+= chunk_size
;
4728 bytes
-= chunk_size
;
4732 aio_task_pool_wait_all(aio
);
4734 ret
= aio_task_pool_status(aio
);
4742 static int coroutine_fn GRAPH_RDLOCK
4743 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
4750 BDRVQcow2State
*s
= bs
->opaque
;
4753 uint8_t *buf
, *out_buf
;
4754 int offset_in_cluster
= offset_into_cluster(s
, offset
);
4756 qcow2_parse_compressed_l2_entry(bs
, l2_entry
, &coffset
, &csize
);
4758 buf
= g_try_malloc(csize
);
4763 out_buf
= qemu_blockalign(bs
, s
->cluster_size
);
4765 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
4766 ret
= bdrv_co_pread(bs
->file
, coffset
, csize
, buf
, 0);
4771 if (qcow2_co_decompress(bs
, out_buf
, s
->cluster_size
, buf
, csize
) < 0) {
4776 qemu_iovec_from_buf(qiov
, qiov_offset
, out_buf
+ offset_in_cluster
, bytes
);
4779 qemu_vfree(out_buf
);
4785 static int make_completely_empty(BlockDriverState
*bs
)
4787 BDRVQcow2State
*s
= bs
->opaque
;
4788 Error
*local_err
= NULL
;
4789 int ret
, l1_clusters
;
4791 uint64_t *new_reftable
= NULL
;
4792 uint64_t rt_entry
, l1_size2
;
4795 uint64_t reftable_offset
;
4796 uint32_t reftable_clusters
;
4797 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
4799 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
4804 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
4809 /* Refcounts will be broken utterly */
4810 ret
= qcow2_mark_dirty(bs
);
4815 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4817 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ L1E_SIZE
);
4818 l1_size2
= (uint64_t)s
->l1_size
* L1E_SIZE
;
4820 /* After this call, neither the in-memory nor the on-disk refcount
4821 * information accurately describe the actual references */
4823 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
4824 l1_clusters
* s
->cluster_size
, 0);
4826 goto fail_broken_refcounts
;
4828 memset(s
->l1_table
, 0, l1_size2
);
4830 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
4832 /* Overwrite enough clusters at the beginning of the sectors to place
4833 * the refcount table, a refcount block and the L1 table in; this may
4834 * overwrite parts of the existing refcount and L1 table, which is not
4835 * an issue because the dirty flag is set, complete data loss is in fact
4836 * desired and partial data loss is consequently fine as well */
4837 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
4838 (2 + l1_clusters
) * s
->cluster_size
, 0);
4839 /* This call (even if it failed overall) may have overwritten on-disk
4840 * refcount structures; in that case, the in-memory refcount information
4841 * will probably differ from the on-disk information which makes the BDS
4844 goto fail_broken_refcounts
;
4847 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4848 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
4850 /* "Create" an empty reftable (one cluster) directly after the image
4851 * header and an empty L1 table three clusters after the image header;
4852 * the cluster between those two will be used as the first refblock */
4853 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
4854 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
4855 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
4856 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
4857 sizeof(l1_ofs_rt_ofs_cls
), &l1_ofs_rt_ofs_cls
, 0);
4859 goto fail_broken_refcounts
;
4862 s
->l1_table_offset
= 3 * s
->cluster_size
;
4864 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ REFTABLE_ENTRY_SIZE
);
4865 if (!new_reftable
) {
4867 goto fail_broken_refcounts
;
4870 s
->refcount_table_offset
= s
->cluster_size
;
4871 s
->refcount_table_size
= s
->cluster_size
/ REFTABLE_ENTRY_SIZE
;
4872 s
->max_refcount_table_index
= 0;
4874 g_free(s
->refcount_table
);
4875 s
->refcount_table
= new_reftable
;
4876 new_reftable
= NULL
;
4878 /* Now the in-memory refcount information again corresponds to the on-disk
4879 * information (reftable is empty and no refblocks (the refblock cache is
4880 * empty)); however, this means some clusters (e.g. the image header) are
4881 * referenced, but not refcounted, but the normal qcow2 code assumes that
4882 * the in-memory information is always correct */
4884 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
4886 /* Enter the first refblock into the reftable */
4887 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
4888 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
, sizeof(rt_entry
),
4891 goto fail_broken_refcounts
;
4893 s
->refcount_table
[0] = 2 * s
->cluster_size
;
4895 s
->free_cluster_index
= 0;
4896 assert(3 + l1_clusters
<= s
->refcount_block_size
);
4897 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
4900 goto fail_broken_refcounts
;
4901 } else if (offset
> 0) {
4902 error_report("First cluster in emptied image is in use");
4906 /* Now finally the in-memory information corresponds to the on-disk
4907 * structures and is correct */
4908 ret
= qcow2_mark_clean(bs
);
4913 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
, false,
4914 PREALLOC_MODE_OFF
, 0, &local_err
);
4916 error_report_err(local_err
);
4922 fail_broken_refcounts
:
4923 /* The BDS is unusable at this point. If we wanted to make it usable, we
4924 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4925 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4926 * again. However, because the functions which could have caused this error
4927 * path to be taken are used by those functions as well, it's very likely
4928 * that that sequence will fail as well. Therefore, just eject the BDS. */
4932 g_free(new_reftable
);
4936 static int qcow2_make_empty(BlockDriverState
*bs
)
4938 BDRVQcow2State
*s
= bs
->opaque
;
4939 uint64_t offset
, end_offset
;
4940 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
4941 int l1_clusters
, ret
= 0;
4943 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ L1E_SIZE
);
4945 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
4946 3 + l1_clusters
<= s
->refcount_block_size
&&
4947 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
&&
4948 !has_data_file(bs
)) {
4949 /* The following function only works for qcow2 v3 images (it
4950 * requires the dirty flag) and only as long as there are no
4951 * features that reserve extra clusters (such as snapshots,
4952 * LUKS header, or persistent bitmaps), because it completely
4953 * empties the image. Furthermore, the L1 table and three
4954 * additional clusters (image header, refcount table, one
4955 * refcount block) have to fit inside one refcount block. It
4956 * only resets the image file, i.e. does not work with an
4957 * external data file. */
4958 return make_completely_empty(bs
);
4961 /* This fallback code simply discards every active cluster; this is slow,
4962 * but works in all cases */
4963 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4964 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
4965 /* As this function is generally used after committing an external
4966 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4967 * default action for this kind of discard is to pass the discard,
4968 * which will ideally result in an actually smaller image file, as
4969 * is probably desired. */
4970 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
4971 QCOW2_DISCARD_SNAPSHOT
, true);
4980 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
4982 BDRVQcow2State
*s
= bs
->opaque
;
4985 qemu_co_mutex_lock(&s
->lock
);
4986 ret
= qcow2_write_caches(bs
);
4987 qemu_co_mutex_unlock(&s
->lock
);
4992 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
4995 Error
*local_err
= NULL
;
4996 BlockMeasureInfo
*info
;
4997 uint64_t required
= 0; /* bytes that contribute to required size */
4998 uint64_t virtual_size
; /* disk size as seen by guest */
4999 uint64_t refcount_bits
;
5001 uint64_t luks_payload_size
= 0;
5002 size_t cluster_size
;
5005 PreallocMode prealloc
;
5006 bool has_backing_file
;
5011 /* Parse image creation options */
5012 extended_l2
= qemu_opt_get_bool_del(opts
, BLOCK_OPT_EXTL2
, false);
5014 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, extended_l2
,
5020 version
= qcow2_opt_get_version_del(opts
, &local_err
);
5025 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
5030 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
5031 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
5032 PREALLOC_MODE_OFF
, &local_err
);
5038 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
5039 has_backing_file
= !!optstr
;
5042 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
5043 has_luks
= optstr
&& strcmp(optstr
, "luks") == 0;
5047 g_autoptr(QCryptoBlockCreateOptions
) create_opts
= NULL
;
5048 QDict
*cryptoopts
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
5051 create_opts
= block_crypto_create_opts_init(cryptoopts
, errp
);
5052 qobject_unref(cryptoopts
);
5057 if (!qcrypto_block_calculate_payload_offset(create_opts
,
5064 luks_payload_size
= ROUND_UP(headerlen
, cluster_size
);
5067 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
5068 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
5070 /* Check that virtual disk size is valid */
5071 l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
5072 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
5073 cluster_size
/ l2e_size
);
5074 if (l2_tables
* L1E_SIZE
> QCOW_MAX_L1_SIZE
) {
5075 error_setg(&local_err
, "The image size is too large "
5076 "(try using a larger cluster size)");
5080 /* Account for input image */
5082 int64_t ssize
= bdrv_getlength(in_bs
);
5084 error_setg_errno(&local_err
, -ssize
,
5085 "Unable to get image virtual_size");
5089 virtual_size
= ROUND_UP(ssize
, cluster_size
);
5091 if (has_backing_file
) {
5092 /* We don't how much of the backing chain is shared by the input
5093 * image and the new image file. In the worst case the new image's
5094 * backing file has nothing in common with the input image. Be
5095 * conservative and assume all clusters need to be written.
5097 required
= virtual_size
;
5102 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
5105 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
5106 ssize
- offset
, &pnum
, NULL
,
5109 error_setg_errno(&local_err
, -ret
,
5110 "Unable to get block status");
5114 if (ret
& BDRV_BLOCK_ZERO
) {
5115 /* Skip zero regions (safe with no backing file) */
5116 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
5117 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
5118 /* Extend pnum to end of cluster for next iteration */
5119 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
5121 /* Count clusters we've seen */
5122 required
+= offset
% cluster_size
+ pnum
;
5128 /* Take into account preallocation. Nothing special is needed for
5129 * PREALLOC_MODE_METADATA since metadata is always counted.
5131 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
5132 required
= virtual_size
;
5135 info
= g_new0(BlockMeasureInfo
, 1);
5136 info
->fully_allocated
= luks_payload_size
+
5137 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
5138 ctz32(refcount_bits
), extended_l2
);
5141 * Remove data clusters that are not required. This overestimates the
5142 * required size because metadata needed for the fully allocated file is
5143 * still counted. Show bitmaps only if both source and destination
5144 * would support them.
5146 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
5147 info
->has_bitmaps
= version
>= 3 && in_bs
&&
5148 bdrv_supports_persistent_dirty_bitmap(in_bs
);
5149 if (info
->has_bitmaps
) {
5150 info
->bitmaps
= qcow2_get_persistent_dirty_bitmap_size(in_bs
,
5156 error_propagate(errp
, local_err
);
5160 static int coroutine_fn
5161 qcow2_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5163 BDRVQcow2State
*s
= bs
->opaque
;
5164 bdi
->cluster_size
= s
->cluster_size
;
5165 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
5166 bdi
->is_dirty
= s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
;
5170 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
,
5173 BDRVQcow2State
*s
= bs
->opaque
;
5174 ImageInfoSpecific
*spec_info
;
5175 QCryptoBlockInfo
*encrypt_info
= NULL
;
5177 if (s
->crypto
!= NULL
) {
5178 encrypt_info
= qcrypto_block_get_info(s
->crypto
, errp
);
5179 if (!encrypt_info
) {
5184 spec_info
= g_new(ImageInfoSpecific
, 1);
5185 *spec_info
= (ImageInfoSpecific
){
5186 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
5187 .u
.qcow2
.data
= g_new0(ImageInfoSpecificQCow2
, 1),
5189 if (s
->qcow_version
== 2) {
5190 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5191 .compat
= g_strdup("0.10"),
5192 .refcount_bits
= s
->refcount_bits
,
5194 } else if (s
->qcow_version
== 3) {
5195 Qcow2BitmapInfoList
*bitmaps
;
5196 if (!qcow2_get_bitmap_info_list(bs
, &bitmaps
, errp
)) {
5197 qapi_free_ImageInfoSpecific(spec_info
);
5198 qapi_free_QCryptoBlockInfo(encrypt_info
);
5201 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5202 .compat
= g_strdup("1.1"),
5203 .lazy_refcounts
= s
->compatible_features
&
5204 QCOW2_COMPAT_LAZY_REFCOUNTS
,
5205 .has_lazy_refcounts
= true,
5206 .corrupt
= s
->incompatible_features
&
5207 QCOW2_INCOMPAT_CORRUPT
,
5208 .has_corrupt
= true,
5209 .has_extended_l2
= true,
5210 .extended_l2
= has_subclusters(s
),
5211 .refcount_bits
= s
->refcount_bits
,
5212 .has_bitmaps
= !!bitmaps
,
5214 .data_file
= g_strdup(s
->image_data_file
),
5215 .has_data_file_raw
= has_data_file(bs
),
5216 .data_file_raw
= data_file_is_raw(bs
),
5217 .compression_type
= s
->compression_type
,
5220 /* if this assertion fails, this probably means a new version was
5221 * added without having it covered here */
5226 ImageInfoSpecificQCow2Encryption
*qencrypt
=
5227 g_new(ImageInfoSpecificQCow2Encryption
, 1);
5228 switch (encrypt_info
->format
) {
5229 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
5230 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
5232 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
5233 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
5234 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
5239 /* Since we did shallow copy above, erase any pointers
5240 * in the original info */
5241 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
5242 qapi_free_QCryptoBlockInfo(encrypt_info
);
5244 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
5250 static int qcow2_has_zero_init(BlockDriverState
*bs
)
5252 BDRVQcow2State
*s
= bs
->opaque
;
5255 if (qemu_in_coroutine()) {
5256 qemu_co_mutex_lock(&s
->lock
);
5259 * Check preallocation status: Preallocated images have all L2
5260 * tables allocated, nonpreallocated images have none. It is
5261 * therefore enough to check the first one.
5263 preallocated
= s
->l1_size
> 0 && s
->l1_table
[0] != 0;
5264 if (qemu_in_coroutine()) {
5265 qemu_co_mutex_unlock(&s
->lock
);
5268 if (!preallocated
) {
5270 } else if (bs
->encrypted
) {
5273 return bdrv_has_zero_init(s
->data_file
->bs
);
5278 * Check the request to vmstate. On success return
5279 * qcow2_vm_state_offset(bs) + @pos
5281 static int64_t qcow2_check_vmstate_request(BlockDriverState
*bs
,
5282 QEMUIOVector
*qiov
, int64_t pos
)
5284 BDRVQcow2State
*s
= bs
->opaque
;
5285 int64_t vmstate_offset
= qcow2_vm_state_offset(s
);
5288 /* Incoming requests must be OK */
5289 bdrv_check_qiov_request(pos
, qiov
->size
, qiov
, 0, &error_abort
);
5291 if (INT64_MAX
- pos
< vmstate_offset
) {
5295 pos
+= vmstate_offset
;
5296 ret
= bdrv_check_qiov_request(pos
, qiov
->size
, qiov
, 0, NULL
);
5304 static int coroutine_fn GRAPH_RDLOCK
5305 qcow2_co_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
5307 int64_t offset
= qcow2_check_vmstate_request(bs
, qiov
, pos
);
5312 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
5313 return bs
->drv
->bdrv_co_pwritev_part(bs
, offset
, qiov
->size
, qiov
, 0, 0);
5316 static int coroutine_fn GRAPH_RDLOCK
5317 qcow2_co_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
5319 int64_t offset
= qcow2_check_vmstate_request(bs
, qiov
, pos
);
5324 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
5325 return bs
->drv
->bdrv_co_preadv_part(bs
, offset
, qiov
->size
, qiov
, 0, 0);
5328 static int qcow2_has_compressed_clusters(BlockDriverState
*bs
)
5331 int64_t bytes
= bdrv_getlength(bs
);
5337 while (bytes
!= 0) {
5339 QCow2SubclusterType type
;
5340 unsigned int cur_bytes
= MIN(INT_MAX
, bytes
);
5341 uint64_t host_offset
;
5343 ret
= qcow2_get_host_offset(bs
, offset
, &cur_bytes
, &host_offset
,
5349 if (type
== QCOW2_SUBCLUSTER_COMPRESSED
) {
5353 offset
+= cur_bytes
;
5361 * Downgrades an image's version. To achieve this, any incompatible features
5362 * have to be removed.
5364 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
5365 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5368 BDRVQcow2State
*s
= bs
->opaque
;
5369 int current_version
= s
->qcow_version
;
5373 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5374 assert(target_version
< current_version
);
5376 /* There are no other versions (now) that you can downgrade to */
5377 assert(target_version
== 2);
5379 if (s
->refcount_order
!= 4) {
5380 error_setg(errp
, "compat=0.10 requires refcount_bits=16");
5384 if (has_data_file(bs
)) {
5385 error_setg(errp
, "Cannot downgrade an image with a data file");
5390 * If any internal snapshot has a different size than the current
5391 * image size, or VM state size that exceeds 32 bits, downgrading
5392 * is unsafe. Even though we would still use v3-compliant output
5393 * to preserve that data, other v2 programs might not realize
5394 * those optional fields are important.
5396 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5397 if (s
->snapshots
[i
].vm_state_size
> UINT32_MAX
||
5398 s
->snapshots
[i
].disk_size
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
5399 error_setg(errp
, "Internal snapshots prevent downgrade of image");
5404 /* clear incompatible features */
5405 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
5406 ret
= qcow2_mark_clean(bs
);
5408 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5413 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5414 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5415 * best thing to do anyway */
5417 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_COMPRESSION
) {
5418 error_setg(errp
, "Cannot downgrade an image with incompatible features "
5419 "0x%" PRIx64
" set",
5420 s
->incompatible_features
& ~QCOW2_INCOMPAT_COMPRESSION
);
5424 /* since we can ignore compatible features, we can set them to 0 as well */
5425 s
->compatible_features
= 0;
5426 /* if lazy refcounts have been used, they have already been fixed through
5427 * clearing the dirty flag */
5429 /* clearing autoclear features is trivial */
5430 s
->autoclear_features
= 0;
5432 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
5434 error_setg_errno(errp
, -ret
, "Failed to turn zero into data clusters");
5438 if (s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
) {
5439 ret
= qcow2_has_compressed_clusters(bs
);
5441 error_setg(errp
, "Failed to check block status");
5445 error_setg(errp
, "Cannot downgrade an image with zstd compression "
5446 "type and existing compressed clusters");
5450 * No compressed clusters for now, so just chose default zlib
5453 s
->incompatible_features
&= ~QCOW2_INCOMPAT_COMPRESSION
;
5454 s
->compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
5457 assert(s
->incompatible_features
== 0);
5459 s
->qcow_version
= target_version
;
5460 ret
= qcow2_update_header(bs
);
5462 s
->qcow_version
= current_version
;
5463 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5470 * Upgrades an image's version. While newer versions encompass all
5471 * features of older versions, some things may have to be presented
5474 static int qcow2_upgrade(BlockDriverState
*bs
, int target_version
,
5475 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5478 BDRVQcow2State
*s
= bs
->opaque
;
5479 bool need_snapshot_update
;
5480 int current_version
= s
->qcow_version
;
5484 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5485 assert(target_version
> current_version
);
5487 /* There are no other versions (yet) that you can upgrade to */
5488 assert(target_version
== 3);
5490 status_cb(bs
, 0, 2, cb_opaque
);
5493 * In v2, snapshots do not need to have extra data. v3 requires
5494 * the 64-bit VM state size and the virtual disk size to be
5496 * qcow2_write_snapshots() will always write the list in the
5497 * v3-compliant format.
5499 need_snapshot_update
= false;
5500 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5501 if (s
->snapshots
[i
].extra_data_size
<
5502 sizeof_field(QCowSnapshotExtraData
, vm_state_size_large
) +
5503 sizeof_field(QCowSnapshotExtraData
, disk_size
))
5505 need_snapshot_update
= true;
5509 if (need_snapshot_update
) {
5510 ret
= qcow2_write_snapshots(bs
);
5512 error_setg_errno(errp
, -ret
, "Failed to update the snapshot table");
5516 status_cb(bs
, 1, 2, cb_opaque
);
5518 s
->qcow_version
= target_version
;
5519 ret
= qcow2_update_header(bs
);
5521 s
->qcow_version
= current_version
;
5522 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5525 status_cb(bs
, 2, 2, cb_opaque
);
5530 typedef enum Qcow2AmendOperation
{
5531 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5532 * statically initialized to so that the helper CB can discern the first
5533 * invocation from an operation change */
5534 QCOW2_NO_OPERATION
= 0,
5537 QCOW2_UPDATING_ENCRYPTION
,
5538 QCOW2_CHANGING_REFCOUNT_ORDER
,
5540 } Qcow2AmendOperation
;
5542 typedef struct Qcow2AmendHelperCBInfo
{
5543 /* The code coordinating the amend operations should only modify
5544 * these four fields; the rest will be managed by the CB */
5545 BlockDriverAmendStatusCB
*original_status_cb
;
5546 void *original_cb_opaque
;
5548 Qcow2AmendOperation current_operation
;
5550 /* Total number of operations to perform (only set once) */
5551 int total_operations
;
5553 /* The following fields are managed by the CB */
5555 /* Number of operations completed */
5556 int operations_completed
;
5558 /* Cumulative offset of all completed operations */
5559 int64_t offset_completed
;
5561 Qcow2AmendOperation last_operation
;
5562 int64_t last_work_size
;
5563 } Qcow2AmendHelperCBInfo
;
5565 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
5566 int64_t operation_offset
,
5567 int64_t operation_work_size
, void *opaque
)
5569 Qcow2AmendHelperCBInfo
*info
= opaque
;
5570 int64_t current_work_size
;
5571 int64_t projected_work_size
;
5573 if (info
->current_operation
!= info
->last_operation
) {
5574 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
5575 info
->offset_completed
+= info
->last_work_size
;
5576 info
->operations_completed
++;
5579 info
->last_operation
= info
->current_operation
;
5582 assert(info
->total_operations
> 0);
5583 assert(info
->operations_completed
< info
->total_operations
);
5585 info
->last_work_size
= operation_work_size
;
5587 current_work_size
= info
->offset_completed
+ operation_work_size
;
5589 /* current_work_size is the total work size for (operations_completed + 1)
5590 * operations (which includes this one), so multiply it by the number of
5591 * operations not covered and divide it by the number of operations
5592 * covered to get a projection for the operations not covered */
5593 projected_work_size
= current_work_size
* (info
->total_operations
-
5594 info
->operations_completed
- 1)
5595 / (info
->operations_completed
+ 1);
5597 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
5598 current_work_size
+ projected_work_size
,
5599 info
->original_cb_opaque
);
5602 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5603 BlockDriverAmendStatusCB
*status_cb
,
5608 BDRVQcow2State
*s
= bs
->opaque
;
5609 int old_version
= s
->qcow_version
, new_version
= old_version
;
5610 uint64_t new_size
= 0;
5611 const char *backing_file
= NULL
, *backing_format
= NULL
, *data_file
= NULL
;
5612 bool lazy_refcounts
= s
->use_lazy_refcounts
;
5613 bool data_file_raw
= data_file_is_raw(bs
);
5614 const char *compat
= NULL
;
5615 int refcount_bits
= s
->refcount_bits
;
5617 QemuOptDesc
*desc
= opts
->list
->desc
;
5618 Qcow2AmendHelperCBInfo helper_cb_info
;
5619 bool encryption_update
= false;
5621 while (desc
&& desc
->name
) {
5622 if (!qemu_opt_find(opts
, desc
->name
)) {
5623 /* only change explicitly defined options */
5628 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
5629 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
5631 /* preserve default */
5632 } else if (!strcmp(compat
, "0.10") || !strcmp(compat
, "v2")) {
5634 } else if (!strcmp(compat
, "1.1") || !strcmp(compat
, "v3")) {
5637 error_setg(errp
, "Unknown compatibility level %s", compat
);
5640 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
5641 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
5642 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
5643 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5644 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
5645 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5646 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
5649 "Can't amend encryption options - encryption not present");
5652 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5654 "Only LUKS encryption options can be amended");
5657 encryption_update
= true;
5658 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
5659 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
5661 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
5662 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
5665 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
5666 !is_power_of_2(refcount_bits
))
5668 error_setg(errp
, "Refcount width must be a power of two and "
5669 "may not exceed 64 bits");
5672 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE
)) {
5673 data_file
= qemu_opt_get(opts
, BLOCK_OPT_DATA_FILE
);
5674 if (data_file
&& !has_data_file(bs
)) {
5675 error_setg(errp
, "data-file can only be set for images that "
5676 "use an external data file");
5679 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE_RAW
)) {
5680 data_file_raw
= qemu_opt_get_bool(opts
, BLOCK_OPT_DATA_FILE_RAW
,
5682 if (data_file_raw
&& !data_file_is_raw(bs
)) {
5683 error_setg(errp
, "data-file-raw cannot be set on existing "
5688 /* if this point is reached, this probably means a new option was
5689 * added without having it covered here */
5696 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
5697 .original_status_cb
= status_cb
,
5698 .original_cb_opaque
= cb_opaque
,
5699 .total_operations
= (new_version
!= old_version
)
5700 + (s
->refcount_bits
!= refcount_bits
) +
5701 (encryption_update
== true)
5704 /* Upgrade first (some features may require compat=1.1) */
5705 if (new_version
> old_version
) {
5706 helper_cb_info
.current_operation
= QCOW2_UPGRADING
;
5707 ret
= qcow2_upgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5708 &helper_cb_info
, errp
);
5714 if (encryption_update
) {
5715 QDict
*amend_opts_dict
;
5716 QCryptoBlockAmendOptions
*amend_opts
;
5718 helper_cb_info
.current_operation
= QCOW2_UPDATING_ENCRYPTION
;
5719 amend_opts_dict
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
5720 if (!amend_opts_dict
) {
5723 amend_opts
= block_crypto_amend_opts_init(amend_opts_dict
, errp
);
5724 qobject_unref(amend_opts_dict
);
5728 ret
= qcrypto_block_amend_options(s
->crypto
,
5729 qcow2_crypto_hdr_read_func
,
5730 qcow2_crypto_hdr_write_func
,
5735 qapi_free_QCryptoBlockAmendOptions(amend_opts
);
5741 if (s
->refcount_bits
!= refcount_bits
) {
5742 int refcount_order
= ctz32(refcount_bits
);
5744 if (new_version
< 3 && refcount_bits
!= 16) {
5745 error_setg(errp
, "Refcount widths other than 16 bits require "
5746 "compatibility level 1.1 or above (use compat=1.1 or "
5751 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
5752 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
5753 &qcow2_amend_helper_cb
,
5754 &helper_cb_info
, errp
);
5760 /* data-file-raw blocks backing files, so clear it first if requested */
5761 if (data_file_raw
) {
5762 s
->autoclear_features
|= QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5764 s
->autoclear_features
&= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5768 g_free(s
->image_data_file
);
5769 s
->image_data_file
= *data_file
? g_strdup(data_file
) : NULL
;
5772 ret
= qcow2_update_header(bs
);
5774 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5778 if (backing_file
|| backing_format
) {
5779 if (g_strcmp0(backing_file
, s
->image_backing_file
) ||
5780 g_strcmp0(backing_format
, s
->image_backing_format
)) {
5781 error_setg(errp
, "Cannot amend the backing file");
5782 error_append_hint(errp
,
5783 "You can use 'qemu-img rebase' instead.\n");
5788 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
5789 if (lazy_refcounts
) {
5790 if (new_version
< 3) {
5791 error_setg(errp
, "Lazy refcounts only supported with "
5792 "compatibility level 1.1 and above (use compat=1.1 "
5796 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5797 ret
= qcow2_update_header(bs
);
5799 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5800 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5803 s
->use_lazy_refcounts
= true;
5805 /* make image clean first */
5806 ret
= qcow2_mark_clean(bs
);
5808 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5811 /* now disallow lazy refcounts */
5812 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5813 ret
= qcow2_update_header(bs
);
5815 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5816 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5819 s
->use_lazy_refcounts
= false;
5824 BlockBackend
*blk
= blk_new_with_bs(bs
, BLK_PERM_RESIZE
, BLK_PERM_ALL
,
5831 * Amending image options should ensure that the image has
5832 * exactly the given new values, so pass exact=true here.
5834 ret
= blk_truncate(blk
, new_size
, true, PREALLOC_MODE_OFF
, 0, errp
);
5841 /* Downgrade last (so unsupported features can be removed before) */
5842 if (new_version
< old_version
) {
5843 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
5844 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5845 &helper_cb_info
, errp
);
5854 static int coroutine_fn
qcow2_co_amend(BlockDriverState
*bs
,
5855 BlockdevAmendOptions
*opts
,
5859 BlockdevAmendOptionsQcow2
*qopts
= &opts
->u
.qcow2
;
5860 BDRVQcow2State
*s
= bs
->opaque
;
5863 if (qopts
->encrypt
) {
5865 error_setg(errp
, "image is not encrypted, can't amend");
5869 if (qopts
->encrypt
->format
!= Q_CRYPTO_BLOCK_FORMAT_LUKS
) {
5871 "Amend can't be used to change the qcow2 encryption format");
5875 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5877 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5881 ret
= qcrypto_block_amend_options(s
->crypto
,
5882 qcow2_crypto_hdr_read_func
,
5883 qcow2_crypto_hdr_write_func
,
5893 * If offset or size are negative, respectively, they will not be included in
5894 * the BLOCK_IMAGE_CORRUPTED event emitted.
5895 * fatal will be ignored for read-only BDS; corruptions found there will always
5896 * be considered non-fatal.
5898 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
5899 int64_t size
, const char *message_format
, ...)
5901 BDRVQcow2State
*s
= bs
->opaque
;
5902 const char *node_name
;
5906 fatal
= fatal
&& bdrv_is_writable(bs
);
5908 if (s
->signaled_corruption
&&
5909 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
5914 va_start(ap
, message_format
);
5915 message
= g_strdup_vprintf(message_format
, ap
);
5919 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
5920 "corruption events will be suppressed\n", message
);
5922 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
5923 "corruption events will be suppressed\n", message
);
5926 node_name
= bdrv_get_node_name(bs
);
5927 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
5928 *node_name
? node_name
: NULL
,
5929 message
, offset
>= 0, offset
,
5935 qcow2_mark_corrupt(bs
);
5936 bs
->drv
= NULL
; /* make BDS unusable */
5939 s
->signaled_corruption
= true;
5942 #define QCOW_COMMON_OPTIONS \
5944 .name = BLOCK_OPT_SIZE, \
5945 .type = QEMU_OPT_SIZE, \
5946 .help = "Virtual disk size" \
5949 .name = BLOCK_OPT_COMPAT_LEVEL, \
5950 .type = QEMU_OPT_STRING, \
5951 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
5954 .name = BLOCK_OPT_BACKING_FILE, \
5955 .type = QEMU_OPT_STRING, \
5956 .help = "File name of a base image" \
5959 .name = BLOCK_OPT_BACKING_FMT, \
5960 .type = QEMU_OPT_STRING, \
5961 .help = "Image format of the base image" \
5964 .name = BLOCK_OPT_DATA_FILE, \
5965 .type = QEMU_OPT_STRING, \
5966 .help = "File name of an external data file" \
5969 .name = BLOCK_OPT_DATA_FILE_RAW, \
5970 .type = QEMU_OPT_BOOL, \
5971 .help = "The external data file must stay valid " \
5975 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
5976 .type = QEMU_OPT_BOOL, \
5977 .help = "Postpone refcount updates", \
5978 .def_value_str = "off" \
5981 .name = BLOCK_OPT_REFCOUNT_BITS, \
5982 .type = QEMU_OPT_NUMBER, \
5983 .help = "Width of a reference count entry in bits", \
5984 .def_value_str = "16" \
5987 static QemuOptsList qcow2_create_opts
= {
5988 .name
= "qcow2-create-opts",
5989 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
5992 .name
= BLOCK_OPT_ENCRYPT
, \
5993 .type
= QEMU_OPT_BOOL
, \
5994 .help
= "Encrypt the image with format 'aes'. (Deprecated " \
5995 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)", \
5998 .name
= BLOCK_OPT_ENCRYPT_FORMAT
, \
5999 .type
= QEMU_OPT_STRING
, \
6000 .help
= "Encrypt the image, format choices: 'aes', 'luks'", \
6002 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
6003 "ID of secret providing qcow AES key or LUKS passphrase"), \
6004 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
6005 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
6006 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
6007 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
6008 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
6009 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
6011 .name
= BLOCK_OPT_CLUSTER_SIZE
, \
6012 .type
= QEMU_OPT_SIZE
, \
6013 .help
= "qcow2 cluster size", \
6014 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
) \
6017 .name
= BLOCK_OPT_EXTL2
, \
6018 .type
= QEMU_OPT_BOOL
, \
6019 .help
= "Extended L2 tables", \
6020 .def_value_str
= "off" \
6023 .name
= BLOCK_OPT_PREALLOC
, \
6024 .type
= QEMU_OPT_STRING
, \
6025 .help
= "Preallocation mode (allowed values: off, " \
6026 "metadata, falloc, full)" \
6029 .name
= BLOCK_OPT_COMPRESSION_TYPE
, \
6030 .type
= QEMU_OPT_STRING
, \
6031 .help
= "Compression method used for image cluster " \
6033 .def_value_str
= "zlib" \
6035 QCOW_COMMON_OPTIONS
,
6036 { /* end of list */ }
6040 static QemuOptsList qcow2_amend_opts
= {
6041 .name
= "qcow2-amend-opts",
6042 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts
.head
),
6044 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
6045 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
6046 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
6047 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
6048 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
6049 QCOW_COMMON_OPTIONS
,
6050 { /* end of list */ }
6054 static const char *const qcow2_strong_runtime_opts
[] = {
6055 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET
,
6060 BlockDriver bdrv_qcow2
= {
6061 .format_name
= "qcow2",
6062 .instance_size
= sizeof(BDRVQcow2State
),
6063 .bdrv_probe
= qcow2_probe
,
6064 .bdrv_open
= qcow2_open
,
6065 .bdrv_close
= qcow2_close
,
6066 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
6067 .bdrv_reopen_commit
= qcow2_reopen_commit
,
6068 .bdrv_reopen_commit_post
= qcow2_reopen_commit_post
,
6069 .bdrv_reopen_abort
= qcow2_reopen_abort
,
6070 .bdrv_join_options
= qcow2_join_options
,
6071 .bdrv_child_perm
= bdrv_default_perms
,
6072 .bdrv_co_create_opts
= qcow2_co_create_opts
,
6073 .bdrv_co_create
= qcow2_co_create
,
6074 .bdrv_has_zero_init
= qcow2_has_zero_init
,
6075 .bdrv_co_block_status
= qcow2_co_block_status
,
6077 .bdrv_co_preadv_part
= qcow2_co_preadv_part
,
6078 .bdrv_co_pwritev_part
= qcow2_co_pwritev_part
,
6079 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
6081 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
6082 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
6083 .bdrv_co_copy_range_from
= qcow2_co_copy_range_from
,
6084 .bdrv_co_copy_range_to
= qcow2_co_copy_range_to
,
6085 .bdrv_co_truncate
= qcow2_co_truncate
,
6086 .bdrv_co_pwritev_compressed_part
= qcow2_co_pwritev_compressed_part
,
6087 .bdrv_make_empty
= qcow2_make_empty
,
6089 .bdrv_snapshot_create
= qcow2_snapshot_create
,
6090 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
6091 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
6092 .bdrv_snapshot_list
= qcow2_snapshot_list
,
6093 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
6094 .bdrv_measure
= qcow2_measure
,
6095 .bdrv_co_get_info
= qcow2_co_get_info
,
6096 .bdrv_get_specific_info
= qcow2_get_specific_info
,
6098 .bdrv_co_save_vmstate
= qcow2_co_save_vmstate
,
6099 .bdrv_co_load_vmstate
= qcow2_co_load_vmstate
,
6102 .supports_backing
= true,
6103 .bdrv_change_backing_file
= qcow2_change_backing_file
,
6105 .bdrv_refresh_limits
= qcow2_refresh_limits
,
6106 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
6107 .bdrv_inactivate
= qcow2_inactivate
,
6109 .create_opts
= &qcow2_create_opts
,
6110 .amend_opts
= &qcow2_amend_opts
,
6111 .strong_runtime_opts
= qcow2_strong_runtime_opts
,
6112 .mutable_opts
= mutable_opts
,
6113 .bdrv_co_check
= qcow2_co_check
,
6114 .bdrv_amend_options
= qcow2_amend_options
,
6115 .bdrv_co_amend
= qcow2_co_amend
,
6117 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
6118 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
6120 .bdrv_supports_persistent_dirty_bitmap
=
6121 qcow2_supports_persistent_dirty_bitmap
,
6122 .bdrv_co_can_store_new_dirty_bitmap
= qcow2_co_can_store_new_dirty_bitmap
,
6123 .bdrv_co_remove_persistent_dirty_bitmap
=
6124 qcow2_co_remove_persistent_dirty_bitmap
,
6127 static void bdrv_qcow2_init(void)
6129 bdrv_register(&bdrv_qcow2
);
6132 block_init(bdrv_qcow2_init
);