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 coroutine_fn GRAPH_RDLOCK
122 qcow2_crypto_hdr_init_func(QCryptoBlock
*block
, size_t headerlen
, void *opaque
,
125 BlockDriverState
*bs
= opaque
;
126 BDRVQcow2State
*s
= bs
->opaque
;
130 ret
= qcow2_alloc_clusters(bs
, headerlen
);
132 error_setg_errno(errp
, -ret
,
133 "Cannot allocate cluster for LUKS header size %zu",
138 s
->crypto_header
.length
= headerlen
;
139 s
->crypto_header
.offset
= ret
;
142 * Zero fill all space in cluster so it has predictable
143 * content, as we may not initialize some regions of the
144 * header (eg only 1 out of 8 key slots will be initialized)
146 clusterlen
= size_to_clusters(s
, headerlen
) * s
->cluster_size
;
147 assert(qcow2_pre_write_overlap_check(bs
, 0, ret
, clusterlen
, false) == 0);
148 ret
= bdrv_co_pwrite_zeroes(bs
->file
, ret
, clusterlen
, 0);
150 error_setg_errno(errp
, -ret
, "Could not zero fill encryption header");
158 /* The graph lock must be held when called in coroutine context */
159 static int coroutine_mixed_fn
160 qcow2_crypto_hdr_write_func(QCryptoBlock
*block
, size_t offset
,
161 const uint8_t *buf
, size_t buflen
,
162 void *opaque
, Error
**errp
)
164 BlockDriverState
*bs
= opaque
;
165 BDRVQcow2State
*s
= bs
->opaque
;
168 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
169 error_setg(errp
, "Request for data outside of extension header");
173 ret
= bdrv_pwrite(bs
->file
, s
->crypto_header
.offset
+ offset
, buflen
, buf
,
176 error_setg_errno(errp
, -ret
, "Could not read encryption header");
183 qcow2_extract_crypto_opts(QemuOpts
*opts
, const char *fmt
, Error
**errp
)
185 QDict
*cryptoopts_qdict
;
188 /* Extract "encrypt." options into a qdict */
189 opts_qdict
= qemu_opts_to_qdict(opts
, NULL
);
190 qdict_extract_subqdict(opts_qdict
, &cryptoopts_qdict
, "encrypt.");
191 qobject_unref(opts_qdict
);
192 qdict_put_str(cryptoopts_qdict
, "format", fmt
);
193 return cryptoopts_qdict
;
197 * read qcow2 extension and fill bs
198 * start reading from start_offset
199 * finish reading upon magic of value 0 or when end_offset reached
200 * unknown magic is skipped (future extension this version knows nothing about)
201 * return 0 upon success, non-0 otherwise
203 static int coroutine_fn GRAPH_RDLOCK
204 qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
205 uint64_t end_offset
, void **p_feature_table
,
206 int flags
, bool *need_update_header
, Error
**errp
)
208 BDRVQcow2State
*s
= bs
->opaque
;
212 Qcow2BitmapHeaderExt bitmaps_ext
;
214 if (need_update_header
!= NULL
) {
215 *need_update_header
= false;
219 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
221 offset
= start_offset
;
222 while (offset
< end_offset
) {
226 if (offset
> s
->cluster_size
)
227 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
229 printf("attempting to read extended header in offset %lu\n", offset
);
232 ret
= bdrv_co_pread(bs
->file
, offset
, sizeof(ext
), &ext
, 0);
234 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
235 "pread fail from offset %" PRIu64
, offset
);
238 ext
.magic
= be32_to_cpu(ext
.magic
);
239 ext
.len
= be32_to_cpu(ext
.len
);
240 offset
+= sizeof(ext
);
242 printf("ext.magic = 0x%x\n", ext
.magic
);
244 if (offset
> end_offset
|| ext
.len
> end_offset
- offset
) {
245 error_setg(errp
, "Header extension too large");
250 case QCOW2_EXT_MAGIC_END
:
253 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
254 if (ext
.len
>= sizeof(bs
->backing_format
)) {
255 error_setg(errp
, "ERROR: ext_backing_format: len=%" PRIu32
256 " too large (>=%zu)", ext
.len
,
257 sizeof(bs
->backing_format
));
260 ret
= bdrv_co_pread(bs
->file
, offset
, ext
.len
, bs
->backing_format
, 0);
262 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
263 "Could not read format name");
266 bs
->backing_format
[ext
.len
] = '\0';
267 s
->image_backing_format
= g_strdup(bs
->backing_format
);
269 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
273 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
274 if (p_feature_table
!= NULL
) {
275 void *feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
276 ret
= bdrv_co_pread(bs
->file
, offset
, ext
.len
, feature_table
, 0);
278 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
279 "Could not read table");
280 g_free(feature_table
);
284 *p_feature_table
= feature_table
;
288 case QCOW2_EXT_MAGIC_CRYPTO_HEADER
: {
289 unsigned int cflags
= 0;
290 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
291 error_setg(errp
, "CRYPTO header extension only "
292 "expected with LUKS encryption method");
295 if (ext
.len
!= sizeof(Qcow2CryptoHeaderExtension
)) {
296 error_setg(errp
, "CRYPTO header extension size %u, "
297 "but expected size %zu", ext
.len
,
298 sizeof(Qcow2CryptoHeaderExtension
));
302 ret
= bdrv_co_pread(bs
->file
, offset
, ext
.len
, &s
->crypto_header
, 0);
304 error_setg_errno(errp
, -ret
,
305 "Unable to read CRYPTO header extension");
308 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
309 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
311 if ((s
->crypto_header
.offset
% s
->cluster_size
) != 0) {
312 error_setg(errp
, "Encryption header offset '%" PRIu64
"' is "
313 "not a multiple of cluster size '%u'",
314 s
->crypto_header
.offset
, s
->cluster_size
);
318 if (flags
& BDRV_O_NO_IO
) {
319 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
321 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
322 qcow2_crypto_hdr_read_func
,
323 bs
, cflags
, QCOW2_MAX_THREADS
, errp
);
329 case QCOW2_EXT_MAGIC_BITMAPS
:
330 if (ext
.len
!= sizeof(bitmaps_ext
)) {
331 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
332 "Invalid extension length");
336 if (!(s
->autoclear_features
& QCOW2_AUTOCLEAR_BITMAPS
)) {
337 if (s
->qcow_version
< 3) {
338 /* Let's be a bit more specific */
339 warn_report("This qcow2 v2 image contains bitmaps, but "
340 "they may have been modified by a program "
341 "without persistent bitmap support; so now "
342 "they must all be considered inconsistent");
344 warn_report("a program lacking bitmap support "
345 "modified this file, so all bitmaps are now "
346 "considered inconsistent");
348 error_printf("Some clusters may be leaked, "
349 "run 'qemu-img check -r' on the image "
351 if (need_update_header
!= NULL
) {
352 /* Updating is needed to drop invalid bitmap extension. */
353 *need_update_header
= true;
358 ret
= bdrv_co_pread(bs
->file
, offset
, ext
.len
, &bitmaps_ext
, 0);
360 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
361 "Could not read ext header");
365 if (bitmaps_ext
.reserved32
!= 0) {
366 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
367 "Reserved field is not zero");
371 bitmaps_ext
.nb_bitmaps
= be32_to_cpu(bitmaps_ext
.nb_bitmaps
);
372 bitmaps_ext
.bitmap_directory_size
=
373 be64_to_cpu(bitmaps_ext
.bitmap_directory_size
);
374 bitmaps_ext
.bitmap_directory_offset
=
375 be64_to_cpu(bitmaps_ext
.bitmap_directory_offset
);
377 if (bitmaps_ext
.nb_bitmaps
> QCOW2_MAX_BITMAPS
) {
379 "bitmaps_ext: Image has %" PRIu32
" bitmaps, "
380 "exceeding the QEMU supported maximum of %d",
381 bitmaps_ext
.nb_bitmaps
, QCOW2_MAX_BITMAPS
);
385 if (bitmaps_ext
.nb_bitmaps
== 0) {
386 error_setg(errp
, "found bitmaps extension with zero bitmaps");
390 if (offset_into_cluster(s
, bitmaps_ext
.bitmap_directory_offset
)) {
391 error_setg(errp
, "bitmaps_ext: "
392 "invalid bitmap directory offset");
396 if (bitmaps_ext
.bitmap_directory_size
>
397 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
) {
398 error_setg(errp
, "bitmaps_ext: "
399 "bitmap directory size (%" PRIu64
") exceeds "
400 "the maximum supported size (%d)",
401 bitmaps_ext
.bitmap_directory_size
,
402 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
);
406 s
->nb_bitmaps
= bitmaps_ext
.nb_bitmaps
;
407 s
->bitmap_directory_offset
=
408 bitmaps_ext
.bitmap_directory_offset
;
409 s
->bitmap_directory_size
=
410 bitmaps_ext
.bitmap_directory_size
;
413 printf("Qcow2: Got bitmaps extension: "
414 "offset=%" PRIu64
" nb_bitmaps=%" PRIu32
"\n",
415 s
->bitmap_directory_offset
, s
->nb_bitmaps
);
419 case QCOW2_EXT_MAGIC_DATA_FILE
:
421 s
->image_data_file
= g_malloc0(ext
.len
+ 1);
422 ret
= bdrv_co_pread(bs
->file
, offset
, ext
.len
, s
->image_data_file
, 0);
424 error_setg_errno(errp
, -ret
,
425 "ERROR: Could not read data file name");
429 printf("Qcow2: Got external data file %s\n", s
->image_data_file
);
435 /* unknown magic - save it in case we need to rewrite the header */
436 /* If you add a new feature, make sure to also update the fast
437 * path of qcow2_make_empty() to deal with it. */
439 Qcow2UnknownHeaderExtension
*uext
;
441 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
442 uext
->magic
= ext
.magic
;
444 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
446 ret
= bdrv_co_pread(bs
->file
, offset
, uext
->len
, uext
->data
, 0);
448 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
449 "Could not read data");
456 offset
+= ((ext
.len
+ 7) & ~7);
462 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
464 BDRVQcow2State
*s
= bs
->opaque
;
465 Qcow2UnknownHeaderExtension
*uext
, *next
;
467 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
468 QLIST_REMOVE(uext
, next
);
473 static void report_unsupported_feature(Error
**errp
, Qcow2Feature
*table
,
476 g_autoptr(GString
) features
= g_string_sized_new(60);
478 while (table
&& table
->name
[0] != '\0') {
479 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
480 if (mask
& (1ULL << table
->bit
)) {
481 if (features
->len
> 0) {
482 g_string_append(features
, ", ");
484 g_string_append_printf(features
, "%.46s", table
->name
);
485 mask
&= ~(1ULL << table
->bit
);
492 if (features
->len
> 0) {
493 g_string_append(features
, ", ");
495 g_string_append_printf(features
,
496 "Unknown incompatible feature: %" PRIx64
, mask
);
499 error_setg(errp
, "Unsupported qcow2 feature(s): %s", features
->str
);
503 * Sets the dirty bit and flushes afterwards if necessary.
505 * The incompatible_features bit is only set if the image file header was
506 * updated successfully. Therefore it is not required to check the return
507 * value of this function.
509 int qcow2_mark_dirty(BlockDriverState
*bs
)
511 BDRVQcow2State
*s
= bs
->opaque
;
515 assert(s
->qcow_version
>= 3);
517 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
518 return 0; /* already dirty */
521 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
522 ret
= bdrv_pwrite_sync(bs
->file
,
523 offsetof(QCowHeader
, incompatible_features
),
524 sizeof(val
), &val
, 0);
529 /* Only treat image as dirty if the header was updated successfully */
530 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
535 * Clears the dirty bit and flushes before if necessary. Only call this
536 * function when there are no pending requests, it does not guard against
537 * concurrent requests dirtying the image.
539 static int qcow2_mark_clean(BlockDriverState
*bs
)
541 BDRVQcow2State
*s
= bs
->opaque
;
543 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
546 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
548 ret
= qcow2_flush_caches(bs
);
553 return qcow2_update_header(bs
);
559 * Marks the image as corrupt.
561 int qcow2_mark_corrupt(BlockDriverState
*bs
)
563 BDRVQcow2State
*s
= bs
->opaque
;
565 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
566 return qcow2_update_header(bs
);
570 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
571 * before if necessary.
573 int qcow2_mark_consistent(BlockDriverState
*bs
)
575 BDRVQcow2State
*s
= bs
->opaque
;
577 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
578 int ret
= qcow2_flush_caches(bs
);
583 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
584 return qcow2_update_header(bs
);
589 static void qcow2_add_check_result(BdrvCheckResult
*out
,
590 const BdrvCheckResult
*src
,
591 bool set_allocation_info
)
593 out
->corruptions
+= src
->corruptions
;
594 out
->leaks
+= src
->leaks
;
595 out
->check_errors
+= src
->check_errors
;
596 out
->corruptions_fixed
+= src
->corruptions_fixed
;
597 out
->leaks_fixed
+= src
->leaks_fixed
;
599 if (set_allocation_info
) {
600 out
->image_end_offset
= src
->image_end_offset
;
605 static int coroutine_fn GRAPH_RDLOCK
606 qcow2_co_check_locked(BlockDriverState
*bs
, BdrvCheckResult
*result
,
609 BdrvCheckResult snapshot_res
= {};
610 BdrvCheckResult refcount_res
= {};
613 memset(result
, 0, sizeof(*result
));
615 ret
= qcow2_check_read_snapshot_table(bs
, &snapshot_res
, fix
);
617 qcow2_add_check_result(result
, &snapshot_res
, false);
621 ret
= qcow2_check_refcounts(bs
, &refcount_res
, fix
);
622 qcow2_add_check_result(result
, &refcount_res
, true);
624 qcow2_add_check_result(result
, &snapshot_res
, false);
628 ret
= qcow2_check_fix_snapshot_table(bs
, &snapshot_res
, fix
);
629 qcow2_add_check_result(result
, &snapshot_res
, false);
634 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
635 ret
= qcow2_mark_clean(bs
);
639 return qcow2_mark_consistent(bs
);
644 static int coroutine_fn GRAPH_RDLOCK
645 qcow2_co_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
648 BDRVQcow2State
*s
= bs
->opaque
;
651 qemu_co_mutex_lock(&s
->lock
);
652 ret
= qcow2_co_check_locked(bs
, result
, fix
);
653 qemu_co_mutex_unlock(&s
->lock
);
657 int qcow2_validate_table(BlockDriverState
*bs
, uint64_t offset
,
658 uint64_t entries
, size_t entry_len
,
659 int64_t max_size_bytes
, const char *table_name
,
662 BDRVQcow2State
*s
= bs
->opaque
;
664 if (entries
> max_size_bytes
/ entry_len
) {
665 error_setg(errp
, "%s too large", table_name
);
669 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
670 * because values will be passed to qemu functions taking int64_t. */
671 if ((INT64_MAX
- entries
* entry_len
< offset
) ||
672 (offset_into_cluster(s
, offset
) != 0)) {
673 error_setg(errp
, "%s offset invalid", table_name
);
680 static const char *const mutable_opts
[] = {
681 QCOW2_OPT_LAZY_REFCOUNTS
,
682 QCOW2_OPT_DISCARD_REQUEST
,
683 QCOW2_OPT_DISCARD_SNAPSHOT
,
684 QCOW2_OPT_DISCARD_OTHER
,
686 QCOW2_OPT_OVERLAP_TEMPLATE
,
687 QCOW2_OPT_OVERLAP_MAIN_HEADER
,
688 QCOW2_OPT_OVERLAP_ACTIVE_L1
,
689 QCOW2_OPT_OVERLAP_ACTIVE_L2
,
690 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
691 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
692 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
693 QCOW2_OPT_OVERLAP_INACTIVE_L1
,
694 QCOW2_OPT_OVERLAP_INACTIVE_L2
,
695 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
696 QCOW2_OPT_CACHE_SIZE
,
697 QCOW2_OPT_L2_CACHE_SIZE
,
698 QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
699 QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
700 QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
704 static QemuOptsList qcow2_runtime_opts
= {
706 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
709 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
710 .type
= QEMU_OPT_BOOL
,
711 .help
= "Postpone refcount updates",
714 .name
= QCOW2_OPT_DISCARD_REQUEST
,
715 .type
= QEMU_OPT_BOOL
,
716 .help
= "Pass guest discard requests to the layer below",
719 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
720 .type
= QEMU_OPT_BOOL
,
721 .help
= "Generate discard requests when snapshot related space "
725 .name
= QCOW2_OPT_DISCARD_OTHER
,
726 .type
= QEMU_OPT_BOOL
,
727 .help
= "Generate discard requests when other clusters are freed",
730 .name
= QCOW2_OPT_OVERLAP
,
731 .type
= QEMU_OPT_STRING
,
732 .help
= "Selects which overlap checks to perform from a range of "
733 "templates (none, constant, cached, all)",
736 .name
= QCOW2_OPT_OVERLAP_TEMPLATE
,
737 .type
= QEMU_OPT_STRING
,
738 .help
= "Selects which overlap checks to perform from a range of "
739 "templates (none, constant, cached, all)",
742 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
743 .type
= QEMU_OPT_BOOL
,
744 .help
= "Check for unintended writes into the main qcow2 header",
747 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
748 .type
= QEMU_OPT_BOOL
,
749 .help
= "Check for unintended writes into the active L1 table",
752 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
753 .type
= QEMU_OPT_BOOL
,
754 .help
= "Check for unintended writes into an active L2 table",
757 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
758 .type
= QEMU_OPT_BOOL
,
759 .help
= "Check for unintended writes into the refcount table",
762 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
763 .type
= QEMU_OPT_BOOL
,
764 .help
= "Check for unintended writes into a refcount block",
767 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
768 .type
= QEMU_OPT_BOOL
,
769 .help
= "Check for unintended writes into the snapshot table",
772 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
773 .type
= QEMU_OPT_BOOL
,
774 .help
= "Check for unintended writes into an inactive L1 table",
777 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
778 .type
= QEMU_OPT_BOOL
,
779 .help
= "Check for unintended writes into an inactive L2 table",
782 .name
= QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
783 .type
= QEMU_OPT_BOOL
,
784 .help
= "Check for unintended writes into the bitmap directory",
787 .name
= QCOW2_OPT_CACHE_SIZE
,
788 .type
= QEMU_OPT_SIZE
,
789 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
793 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
794 .type
= QEMU_OPT_SIZE
,
795 .help
= "Maximum L2 table cache size",
798 .name
= QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
799 .type
= QEMU_OPT_SIZE
,
800 .help
= "Size of each entry in the L2 cache",
803 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
804 .type
= QEMU_OPT_SIZE
,
805 .help
= "Maximum refcount block cache size",
808 .name
= QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
809 .type
= QEMU_OPT_NUMBER
,
810 .help
= "Clean unused cache entries after this time (in seconds)",
812 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
813 "ID of secret providing qcow2 AES key or LUKS passphrase"),
814 { /* end of list */ }
818 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
819 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
820 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
821 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
822 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
823 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
824 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
825 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
826 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
827 [QCOW2_OL_BITMAP_DIRECTORY_BITNR
] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
830 static void cache_clean_timer_cb(void *opaque
)
832 BlockDriverState
*bs
= opaque
;
833 BDRVQcow2State
*s
= bs
->opaque
;
834 qcow2_cache_clean_unused(s
->l2_table_cache
);
835 qcow2_cache_clean_unused(s
->refcount_block_cache
);
836 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
837 (int64_t) s
->cache_clean_interval
* 1000);
840 static void cache_clean_timer_init(BlockDriverState
*bs
, AioContext
*context
)
842 BDRVQcow2State
*s
= bs
->opaque
;
843 if (s
->cache_clean_interval
> 0) {
844 s
->cache_clean_timer
=
845 aio_timer_new_with_attrs(context
, QEMU_CLOCK_VIRTUAL
,
846 SCALE_MS
, QEMU_TIMER_ATTR_EXTERNAL
,
847 cache_clean_timer_cb
, bs
);
848 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
849 (int64_t) s
->cache_clean_interval
* 1000);
853 static void cache_clean_timer_del(BlockDriverState
*bs
)
855 BDRVQcow2State
*s
= bs
->opaque
;
856 if (s
->cache_clean_timer
) {
857 timer_free(s
->cache_clean_timer
);
858 s
->cache_clean_timer
= NULL
;
862 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
864 cache_clean_timer_del(bs
);
867 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
868 AioContext
*new_context
)
870 cache_clean_timer_init(bs
, new_context
);
873 static bool read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
874 uint64_t *l2_cache_size
,
875 uint64_t *l2_cache_entry_size
,
876 uint64_t *refcount_cache_size
, Error
**errp
)
878 BDRVQcow2State
*s
= bs
->opaque
;
879 uint64_t combined_cache_size
, l2_cache_max_setting
;
880 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
881 bool l2_cache_entry_size_set
;
882 int min_refcount_cache
= MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
883 uint64_t virtual_disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
884 uint64_t max_l2_entries
= DIV_ROUND_UP(virtual_disk_size
, s
->cluster_size
);
885 /* An L2 table is always one cluster in size so the max cache size
886 * should be a multiple of the cluster size. */
887 uint64_t max_l2_cache
= ROUND_UP(max_l2_entries
* l2_entry_size(s
),
890 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
891 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
892 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
893 l2_cache_entry_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
);
895 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
896 l2_cache_max_setting
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
,
897 DEFAULT_L2_CACHE_MAX_SIZE
);
898 *refcount_cache_size
= qemu_opt_get_size(opts
,
899 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
901 *l2_cache_entry_size
= qemu_opt_get_size(
902 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
904 *l2_cache_size
= MIN(max_l2_cache
, l2_cache_max_setting
);
906 if (combined_cache_size_set
) {
907 if (l2_cache_size_set
&& refcount_cache_size_set
) {
908 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
909 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
912 } else if (l2_cache_size_set
&&
913 (l2_cache_max_setting
> combined_cache_size
)) {
914 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
915 QCOW2_OPT_CACHE_SIZE
);
917 } else if (*refcount_cache_size
> combined_cache_size
) {
918 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
919 QCOW2_OPT_CACHE_SIZE
);
923 if (l2_cache_size_set
) {
924 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
925 } else if (refcount_cache_size_set
) {
926 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
928 /* Assign as much memory as possible to the L2 cache, and
929 * use the remainder for the refcount cache */
930 if (combined_cache_size
>= max_l2_cache
+ min_refcount_cache
) {
931 *l2_cache_size
= max_l2_cache
;
932 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
934 *refcount_cache_size
=
935 MIN(combined_cache_size
, min_refcount_cache
);
936 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
942 * If the L2 cache is not enough to cover the whole disk then
943 * default to 4KB entries. Smaller entries reduce the cost of
944 * loads and evictions and increase I/O performance.
946 if (*l2_cache_size
< max_l2_cache
&& !l2_cache_entry_size_set
) {
947 *l2_cache_entry_size
= MIN(s
->cluster_size
, 4096);
950 /* l2_cache_size and refcount_cache_size are ensured to have at least
951 * their minimum values in qcow2_update_options_prepare() */
953 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
954 *l2_cache_entry_size
> s
->cluster_size
||
955 !is_power_of_2(*l2_cache_entry_size
)) {
956 error_setg(errp
, "L2 cache entry size must be a power of two "
957 "between %d and the cluster size (%d)",
958 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
965 typedef struct Qcow2ReopenState
{
966 Qcow2Cache
*l2_table_cache
;
967 Qcow2Cache
*refcount_block_cache
;
968 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
969 bool use_lazy_refcounts
;
971 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
972 uint64_t cache_clean_interval
;
973 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
976 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
978 QDict
*options
, int flags
,
981 BDRVQcow2State
*s
= bs
->opaque
;
982 QemuOpts
*opts
= NULL
;
983 const char *opt_overlap_check
, *opt_overlap_check_template
;
984 int overlap_check_template
= 0;
985 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
987 const char *encryptfmt
;
988 QDict
*encryptopts
= NULL
;
991 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
992 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
994 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
995 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1000 /* get L2 table/refcount block cache size from command line options */
1001 if (!read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
1002 &refcount_cache_size
, errp
)) {
1007 l2_cache_size
/= l2_cache_entry_size
;
1008 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
1009 l2_cache_size
= MIN_L2_CACHE_SIZE
;
1011 if (l2_cache_size
> INT_MAX
) {
1012 error_setg(errp
, "L2 cache size too big");
1017 refcount_cache_size
/= s
->cluster_size
;
1018 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
1019 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
1021 if (refcount_cache_size
> INT_MAX
) {
1022 error_setg(errp
, "Refcount cache size too big");
1027 /* alloc new L2 table/refcount block cache, flush old one */
1028 if (s
->l2_table_cache
) {
1029 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1031 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
1036 if (s
->refcount_block_cache
) {
1037 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1039 error_setg_errno(errp
, -ret
,
1040 "Failed to flush the refcount block cache");
1045 r
->l2_slice_size
= l2_cache_entry_size
/ l2_entry_size(s
);
1046 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
1047 l2_cache_entry_size
);
1048 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
1050 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
1051 error_setg(errp
, "Could not allocate metadata caches");
1056 /* New interval for cache cleanup timer */
1057 r
->cache_clean_interval
=
1058 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
1059 DEFAULT_CACHE_CLEAN_INTERVAL
);
1060 #ifndef CONFIG_LINUX
1061 if (r
->cache_clean_interval
!= 0) {
1062 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1063 " not supported on this host");
1068 if (r
->cache_clean_interval
> UINT_MAX
) {
1069 error_setg(errp
, "Cache clean interval too big");
1074 /* lazy-refcounts; flush if going from enabled to disabled */
1075 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
1076 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
1077 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
1078 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
1079 "qemu 1.1 compatibility level");
1084 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
1085 ret
= qcow2_mark_clean(bs
);
1087 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
1092 /* Overlap check options */
1093 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
1094 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1095 if (opt_overlap_check_template
&& opt_overlap_check
&&
1096 strcmp(opt_overlap_check_template
, opt_overlap_check
))
1098 error_setg(errp
, "Conflicting values for qcow2 options '"
1099 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1100 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
1104 if (!opt_overlap_check
) {
1105 opt_overlap_check
= opt_overlap_check_template
?: "cached";
1108 if (!strcmp(opt_overlap_check
, "none")) {
1109 overlap_check_template
= 0;
1110 } else if (!strcmp(opt_overlap_check
, "constant")) {
1111 overlap_check_template
= QCOW2_OL_CONSTANT
;
1112 } else if (!strcmp(opt_overlap_check
, "cached")) {
1113 overlap_check_template
= QCOW2_OL_CACHED
;
1114 } else if (!strcmp(opt_overlap_check
, "all")) {
1115 overlap_check_template
= QCOW2_OL_ALL
;
1117 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1118 "'overlap-check'. Allowed are any of the following: "
1119 "none, constant, cached, all", opt_overlap_check
);
1124 r
->overlap_check
= 0;
1125 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1126 /* overlap-check defines a template bitmask, but every flag may be
1127 * overwritten through the associated boolean option */
1129 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1130 overlap_check_template
& (1 << i
)) << i
;
1133 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1134 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1135 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1136 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1137 flags
& BDRV_O_UNMAP
);
1138 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1139 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1140 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1141 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1143 switch (s
->crypt_method_header
) {
1144 case QCOW_CRYPT_NONE
:
1146 error_setg(errp
, "No encryption in image header, but options "
1147 "specified format '%s'", encryptfmt
);
1153 case QCOW_CRYPT_AES
:
1154 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1156 "Header reported 'aes' encryption format but "
1157 "options specify '%s'", encryptfmt
);
1161 qdict_put_str(encryptopts
, "format", "qcow");
1162 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1163 if (!r
->crypto_opts
) {
1169 case QCOW_CRYPT_LUKS
:
1170 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1172 "Header reported 'luks' encryption format but "
1173 "options specify '%s'", encryptfmt
);
1177 qdict_put_str(encryptopts
, "format", "luks");
1178 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1179 if (!r
->crypto_opts
) {
1186 error_setg(errp
, "Unsupported encryption method %d",
1187 s
->crypt_method_header
);
1194 qobject_unref(encryptopts
);
1195 qemu_opts_del(opts
);
1200 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1201 Qcow2ReopenState
*r
)
1203 BDRVQcow2State
*s
= bs
->opaque
;
1206 if (s
->l2_table_cache
) {
1207 qcow2_cache_destroy(s
->l2_table_cache
);
1209 if (s
->refcount_block_cache
) {
1210 qcow2_cache_destroy(s
->refcount_block_cache
);
1212 s
->l2_table_cache
= r
->l2_table_cache
;
1213 s
->refcount_block_cache
= r
->refcount_block_cache
;
1214 s
->l2_slice_size
= r
->l2_slice_size
;
1216 s
->overlap_check
= r
->overlap_check
;
1217 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1219 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1220 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1223 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1224 cache_clean_timer_del(bs
);
1225 s
->cache_clean_interval
= r
->cache_clean_interval
;
1226 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1229 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1230 s
->crypto_opts
= r
->crypto_opts
;
1233 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1234 Qcow2ReopenState
*r
)
1236 if (r
->l2_table_cache
) {
1237 qcow2_cache_destroy(r
->l2_table_cache
);
1239 if (r
->refcount_block_cache
) {
1240 qcow2_cache_destroy(r
->refcount_block_cache
);
1242 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1245 static int coroutine_fn
1246 qcow2_update_options(BlockDriverState
*bs
, QDict
*options
, int flags
,
1249 Qcow2ReopenState r
= {};
1252 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1254 qcow2_update_options_commit(bs
, &r
);
1256 qcow2_update_options_abort(bs
, &r
);
1262 static int validate_compression_type(BDRVQcow2State
*s
, Error
**errp
)
1264 switch (s
->compression_type
) {
1265 case QCOW2_COMPRESSION_TYPE_ZLIB
:
1267 case QCOW2_COMPRESSION_TYPE_ZSTD
:
1272 error_setg(errp
, "qcow2: unknown compression type: %u",
1273 s
->compression_type
);
1278 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1279 * the incompatible feature flag must be set
1281 if (s
->compression_type
== QCOW2_COMPRESSION_TYPE_ZLIB
) {
1282 if (s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
) {
1283 error_setg(errp
, "qcow2: Compression type incompatible feature "
1284 "bit must not be set");
1288 if (!(s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
)) {
1289 error_setg(errp
, "qcow2: Compression type incompatible feature "
1298 /* Called with s->lock held. */
1299 static int coroutine_fn GRAPH_RDLOCK
1300 qcow2_do_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1301 bool open_data_file
, Error
**errp
)
1304 BDRVQcow2State
*s
= bs
->opaque
;
1305 unsigned int len
, i
;
1309 uint64_t l1_vm_state_index
;
1310 bool update_header
= false;
1312 ret
= bdrv_co_pread(bs
->file
, 0, sizeof(header
), &header
, 0);
1314 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1317 header
.magic
= be32_to_cpu(header
.magic
);
1318 header
.version
= be32_to_cpu(header
.version
);
1319 header
.backing_file_offset
= be64_to_cpu(header
.backing_file_offset
);
1320 header
.backing_file_size
= be32_to_cpu(header
.backing_file_size
);
1321 header
.size
= be64_to_cpu(header
.size
);
1322 header
.cluster_bits
= be32_to_cpu(header
.cluster_bits
);
1323 header
.crypt_method
= be32_to_cpu(header
.crypt_method
);
1324 header
.l1_table_offset
= be64_to_cpu(header
.l1_table_offset
);
1325 header
.l1_size
= be32_to_cpu(header
.l1_size
);
1326 header
.refcount_table_offset
= be64_to_cpu(header
.refcount_table_offset
);
1327 header
.refcount_table_clusters
=
1328 be32_to_cpu(header
.refcount_table_clusters
);
1329 header
.snapshots_offset
= be64_to_cpu(header
.snapshots_offset
);
1330 header
.nb_snapshots
= be32_to_cpu(header
.nb_snapshots
);
1332 if (header
.magic
!= QCOW_MAGIC
) {
1333 error_setg(errp
, "Image is not in qcow2 format");
1337 if (header
.version
< 2 || header
.version
> 3) {
1338 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1343 s
->qcow_version
= header
.version
;
1345 /* Initialise cluster size */
1346 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1347 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1348 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1349 header
.cluster_bits
);
1354 s
->cluster_bits
= header
.cluster_bits
;
1355 s
->cluster_size
= 1 << s
->cluster_bits
;
1357 /* Initialise version 3 header fields */
1358 if (header
.version
== 2) {
1359 header
.incompatible_features
= 0;
1360 header
.compatible_features
= 0;
1361 header
.autoclear_features
= 0;
1362 header
.refcount_order
= 4;
1363 header
.header_length
= 72;
1365 header
.incompatible_features
=
1366 be64_to_cpu(header
.incompatible_features
);
1367 header
.compatible_features
= be64_to_cpu(header
.compatible_features
);
1368 header
.autoclear_features
= be64_to_cpu(header
.autoclear_features
);
1369 header
.refcount_order
= be32_to_cpu(header
.refcount_order
);
1370 header
.header_length
= be32_to_cpu(header
.header_length
);
1372 if (header
.header_length
< 104) {
1373 error_setg(errp
, "qcow2 header too short");
1379 if (header
.header_length
> s
->cluster_size
) {
1380 error_setg(errp
, "qcow2 header exceeds cluster size");
1385 if (header
.header_length
> sizeof(header
)) {
1386 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1387 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1388 ret
= bdrv_co_pread(bs
->file
, sizeof(header
),
1389 s
->unknown_header_fields_size
,
1390 s
->unknown_header_fields
, 0);
1392 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1398 if (header
.backing_file_offset
> s
->cluster_size
) {
1399 error_setg(errp
, "Invalid backing file offset");
1404 if (header
.backing_file_offset
) {
1405 ext_end
= header
.backing_file_offset
;
1407 ext_end
= 1 << header
.cluster_bits
;
1410 /* Handle feature bits */
1411 s
->incompatible_features
= header
.incompatible_features
;
1412 s
->compatible_features
= header
.compatible_features
;
1413 s
->autoclear_features
= header
.autoclear_features
;
1416 * Handle compression type
1417 * Older qcow2 images don't contain the compression type header.
1418 * Distinguish them by the header length and use
1419 * the only valid (default) compression type in that case
1421 if (header
.header_length
> offsetof(QCowHeader
, compression_type
)) {
1422 s
->compression_type
= header
.compression_type
;
1424 s
->compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
1427 ret
= validate_compression_type(s
, errp
);
1432 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1433 void *feature_table
= NULL
;
1434 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1435 &feature_table
, flags
, NULL
, NULL
);
1436 report_unsupported_feature(errp
, feature_table
,
1437 s
->incompatible_features
&
1438 ~QCOW2_INCOMPAT_MASK
);
1440 g_free(feature_table
);
1444 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1445 /* Corrupt images may not be written to unless they are being repaired
1447 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1448 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1455 s
->subclusters_per_cluster
=
1456 has_subclusters(s
) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
: 1;
1457 s
->subcluster_size
= s
->cluster_size
/ s
->subclusters_per_cluster
;
1458 s
->subcluster_bits
= ctz32(s
->subcluster_size
);
1460 if (s
->subcluster_size
< (1 << MIN_CLUSTER_BITS
)) {
1461 error_setg(errp
, "Unsupported subcluster size: %d", s
->subcluster_size
);
1466 /* Check support for various header values */
1467 if (header
.refcount_order
> 6) {
1468 error_setg(errp
, "Reference count entry width too large; may not "
1473 s
->refcount_order
= header
.refcount_order
;
1474 s
->refcount_bits
= 1 << s
->refcount_order
;
1475 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1476 s
->refcount_max
+= s
->refcount_max
- 1;
1478 s
->crypt_method_header
= header
.crypt_method
;
1479 if (s
->crypt_method_header
) {
1480 if (bdrv_uses_whitelist() &&
1481 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1483 "Use of AES-CBC encrypted qcow2 images is no longer "
1484 "supported in system emulators");
1485 error_append_hint(errp
,
1486 "You can use 'qemu-img convert' to convert your "
1487 "image to an alternative supported format, such "
1488 "as unencrypted qcow2, or raw with the LUKS "
1489 "format instead.\n");
1494 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1495 s
->crypt_physical_offset
= false;
1497 /* Assuming LUKS and any future crypt methods we
1498 * add will all use physical offsets, due to the
1499 * fact that the alternative is insecure... */
1500 s
->crypt_physical_offset
= true;
1503 bs
->encrypted
= true;
1506 s
->l2_bits
= s
->cluster_bits
- ctz32(l2_entry_size(s
));
1507 s
->l2_size
= 1 << s
->l2_bits
;
1508 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1509 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1510 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1511 bs
->total_sectors
= header
.size
/ BDRV_SECTOR_SIZE
;
1512 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1513 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1514 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1516 s
->refcount_table_offset
= header
.refcount_table_offset
;
1517 s
->refcount_table_size
=
1518 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1520 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1521 error_setg(errp
, "Image does not contain a reference count table");
1526 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1527 header
.refcount_table_clusters
,
1528 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1529 "Reference count table", errp
);
1534 if (!(flags
& BDRV_O_CHECK
)) {
1536 * The total size in bytes of the snapshot table is checked in
1537 * qcow2_read_snapshots() because the size of each snapshot is
1538 * variable and we don't know it yet.
1539 * Here we only check the offset and number of snapshots.
1541 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1542 header
.nb_snapshots
,
1543 sizeof(QCowSnapshotHeader
),
1544 sizeof(QCowSnapshotHeader
) *
1546 "Snapshot table", errp
);
1552 /* read the level 1 table */
1553 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1554 header
.l1_size
, L1E_SIZE
,
1555 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1559 s
->l1_size
= header
.l1_size
;
1560 s
->l1_table_offset
= header
.l1_table_offset
;
1562 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1563 if (l1_vm_state_index
> INT_MAX
) {
1564 error_setg(errp
, "Image is too big");
1568 s
->l1_vm_state_index
= l1_vm_state_index
;
1570 /* the L1 table must contain at least enough entries to put
1571 header.size bytes */
1572 if (s
->l1_size
< s
->l1_vm_state_index
) {
1573 error_setg(errp
, "L1 table is too small");
1578 if (s
->l1_size
> 0) {
1579 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
, s
->l1_size
* L1E_SIZE
);
1580 if (s
->l1_table
== NULL
) {
1581 error_setg(errp
, "Could not allocate L1 table");
1585 ret
= bdrv_co_pread(bs
->file
, s
->l1_table_offset
, s
->l1_size
* L1E_SIZE
,
1588 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1591 for(i
= 0;i
< s
->l1_size
; i
++) {
1592 s
->l1_table
[i
] = be64_to_cpu(s
->l1_table
[i
]);
1596 /* Parse driver-specific options */
1597 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1604 ret
= qcow2_refcount_init(bs
);
1606 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1610 QLIST_INIT(&s
->cluster_allocs
);
1611 QTAILQ_INIT(&s
->discards
);
1613 /* read qcow2 extensions */
1614 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1615 flags
, &update_header
, errp
)) {
1620 if (open_data_file
) {
1621 /* Open external data file */
1622 s
->data_file
= bdrv_co_open_child(NULL
, options
, "data-file", bs
,
1623 &child_of_bds
, BDRV_CHILD_DATA
,
1630 if (s
->incompatible_features
& QCOW2_INCOMPAT_DATA_FILE
) {
1631 if (!s
->data_file
&& s
->image_data_file
) {
1632 s
->data_file
= bdrv_co_open_child(s
->image_data_file
, options
,
1635 BDRV_CHILD_DATA
, false, errp
);
1636 if (!s
->data_file
) {
1641 if (!s
->data_file
) {
1642 error_setg(errp
, "'data-file' is required for this image");
1648 bs
->file
->role
&= ~BDRV_CHILD_DATA
;
1650 /* Must succeed because we have given up permissions if anything */
1651 bdrv_child_refresh_perms(bs
, bs
->file
, &error_abort
);
1654 error_setg(errp
, "'data-file' can only be set for images with "
1655 "an external data file");
1660 s
->data_file
= bs
->file
;
1662 if (data_file_is_raw(bs
)) {
1663 error_setg(errp
, "data-file-raw requires a data file");
1670 /* qcow2_read_extension may have set up the crypto context
1671 * if the crypt method needs a header region, some methods
1672 * don't need header extensions, so must check here
1674 if (s
->crypt_method_header
&& !s
->crypto
) {
1675 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1676 unsigned int cflags
= 0;
1677 if (flags
& BDRV_O_NO_IO
) {
1678 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1680 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1682 QCOW2_MAX_THREADS
, errp
);
1687 } else if (!(flags
& BDRV_O_NO_IO
)) {
1688 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1689 s
->crypt_method_header
);
1695 /* read the backing file name */
1696 if (header
.backing_file_offset
!= 0) {
1697 len
= header
.backing_file_size
;
1698 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1699 len
>= sizeof(bs
->backing_file
)) {
1700 error_setg(errp
, "Backing file name too long");
1705 s
->image_backing_file
= g_malloc(len
+ 1);
1706 ret
= bdrv_co_pread(bs
->file
, header
.backing_file_offset
, len
,
1707 s
->image_backing_file
, 0);
1709 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1712 s
->image_backing_file
[len
] = '\0';
1715 * Update only when something has changed. This function is called by
1716 * qcow2_co_invalidate_cache(), and we do not want to reset
1717 * auto_backing_file unless necessary.
1719 if (!g_str_equal(s
->image_backing_file
, bs
->backing_file
)) {
1720 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
),
1721 s
->image_backing_file
);
1722 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
1723 s
->image_backing_file
);
1728 * Internal snapshots; skip reading them in check mode, because
1729 * we do not need them then, and we do not want to abort because
1730 * of a broken table.
1732 if (!(flags
& BDRV_O_CHECK
)) {
1733 s
->snapshots_offset
= header
.snapshots_offset
;
1734 s
->nb_snapshots
= header
.nb_snapshots
;
1736 ret
= qcow2_read_snapshots(bs
, errp
);
1742 /* Clear unknown autoclear feature bits */
1743 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1744 update_header
= update_header
&& bdrv_is_writable(bs
);
1745 if (update_header
) {
1746 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1749 /* == Handle persistent dirty bitmaps ==
1751 * We want load dirty bitmaps in three cases:
1753 * 1. Normal open of the disk in active mode, not related to invalidation
1756 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1757 * bitmaps are _not_ migrating through migration channel, i.e.
1758 * 'dirty-bitmaps' capability is disabled.
1760 * 3. Invalidation of source vm after failed or canceled migration.
1761 * This is a very interesting case. There are two possible types of
1764 * A. Stored on inactivation and removed. They should be loaded from the
1767 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1768 * the migration channel (with dirty-bitmaps capability).
1770 * On the other hand, there are two possible sub-cases:
1772 * 3.1 disk was changed by somebody else while were inactive. In this
1773 * case all in-RAM dirty bitmaps (both persistent and not) are
1774 * definitely invalid. And we don't have any method to determine
1777 * Simple and safe thing is to just drop all the bitmaps of type B on
1778 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1780 * On the other hand, resuming source vm, if disk was already changed
1781 * is a bad thing anyway: not only bitmaps, the whole vm state is
1782 * out of sync with disk.
1784 * This means, that user or management tool, who for some reason
1785 * decided to resume source vm, after disk was already changed by
1786 * target vm, should at least drop all dirty bitmaps by hand.
1788 * So, we can ignore this case for now, but TODO: "generation"
1789 * extension for qcow2, to determine, that image was changed after
1790 * last inactivation. And if it is changed, we will drop (or at least
1791 * mark as 'invalid' all the bitmaps of type B, both persistent
1794 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1795 * to disk ('dirty-bitmaps' capability disabled), or not saved
1796 * ('dirty-bitmaps' capability enabled), but we don't need to care
1797 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1798 * and not stored has flag IN_USE=1 in the image and will be skipped
1801 * One remaining possible case when we don't want load bitmaps:
1803 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1804 * will be loaded on invalidation, no needs try loading them before)
1807 if (!(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
)) {
1808 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1809 bool header_updated
;
1810 if (!qcow2_load_dirty_bitmaps(bs
, &header_updated
, errp
)) {
1815 update_header
= update_header
&& !header_updated
;
1818 if (update_header
) {
1819 ret
= qcow2_update_header(bs
);
1821 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1826 bs
->supported_zero_flags
= header
.version
>= 3 ?
1827 BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
: 0;
1828 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
1830 /* Repair image if dirty */
1831 if (!(flags
& BDRV_O_CHECK
) && bdrv_is_writable(bs
) &&
1832 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1833 BdrvCheckResult result
= {0};
1835 ret
= qcow2_co_check_locked(bs
, &result
,
1836 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1837 if (ret
< 0 || result
.check_errors
) {
1841 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1848 BdrvCheckResult result
= {0};
1849 qcow2_check_refcounts(bs
, &result
, 0);
1853 qemu_co_queue_init(&s
->thread_task_queue
);
1858 g_free(s
->image_data_file
);
1859 if (open_data_file
&& has_data_file(bs
)) {
1860 bdrv_unref_child(bs
, s
->data_file
);
1861 s
->data_file
= NULL
;
1863 g_free(s
->unknown_header_fields
);
1864 cleanup_unknown_header_ext(bs
);
1865 qcow2_free_snapshots(bs
);
1866 qcow2_refcount_close(bs
);
1867 qemu_vfree(s
->l1_table
);
1868 /* else pre-write overlap checks in cache_destroy may crash */
1870 cache_clean_timer_del(bs
);
1871 if (s
->l2_table_cache
) {
1872 qcow2_cache_destroy(s
->l2_table_cache
);
1874 if (s
->refcount_block_cache
) {
1875 qcow2_cache_destroy(s
->refcount_block_cache
);
1877 qcrypto_block_free(s
->crypto
);
1878 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1882 typedef struct QCow2OpenCo
{
1883 BlockDriverState
*bs
;
1890 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1892 QCow2OpenCo
*qoc
= opaque
;
1893 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1895 GRAPH_RDLOCK_GUARD();
1897 qemu_co_mutex_lock(&s
->lock
);
1898 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, true,
1900 qemu_co_mutex_unlock(&s
->lock
);
1903 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1906 BDRVQcow2State
*s
= bs
->opaque
;
1916 ret
= bdrv_open_file_child(NULL
, options
, "file", bs
, errp
);
1921 /* Initialise locks */
1922 qemu_co_mutex_init(&s
->lock
);
1924 assert(!qemu_in_coroutine());
1925 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1926 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1927 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1932 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1934 BDRVQcow2State
*s
= bs
->opaque
;
1936 if (bs
->encrypted
) {
1937 /* Encryption works on a sector granularity */
1938 bs
->bl
.request_alignment
= qcrypto_block_get_sector_size(s
->crypto
);
1940 bs
->bl
.pwrite_zeroes_alignment
= s
->subcluster_size
;
1941 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1944 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1945 BlockReopenQueue
*queue
, Error
**errp
)
1947 BDRVQcow2State
*s
= state
->bs
->opaque
;
1948 Qcow2ReopenState
*r
;
1951 r
= g_new0(Qcow2ReopenState
, 1);
1954 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1955 state
->flags
, errp
);
1960 /* We need to write out any unwritten data if we reopen read-only. */
1961 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1962 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1967 ret
= bdrv_flush(state
->bs
);
1972 ret
= qcow2_mark_clean(state
->bs
);
1979 * Without an external data file, s->data_file points to the same BdrvChild
1980 * as bs->file. It needs to be resynced after reopen because bs->file may
1981 * be changed. We can't use it in the meantime.
1983 if (!has_data_file(state
->bs
)) {
1984 assert(s
->data_file
== state
->bs
->file
);
1985 s
->data_file
= NULL
;
1991 qcow2_update_options_abort(state
->bs
, r
);
1996 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1998 BDRVQcow2State
*s
= state
->bs
->opaque
;
2000 qcow2_update_options_commit(state
->bs
, state
->opaque
);
2001 if (!s
->data_file
) {
2003 * If we don't have an external data file, s->data_file was cleared by
2004 * qcow2_reopen_prepare() and needs to be updated.
2006 s
->data_file
= state
->bs
->file
;
2008 g_free(state
->opaque
);
2011 static void qcow2_reopen_commit_post(BDRVReopenState
*state
)
2013 if (state
->flags
& BDRV_O_RDWR
) {
2014 Error
*local_err
= NULL
;
2016 if (qcow2_reopen_bitmaps_rw(state
->bs
, &local_err
) < 0) {
2018 * This is not fatal, bitmaps just left read-only, so all following
2019 * writes will fail. User can remove read-only bitmaps to unblock
2020 * writes or retry reopen.
2022 error_reportf_err(local_err
,
2023 "%s: Failed to make dirty bitmaps writable: ",
2024 bdrv_get_node_name(state
->bs
));
2029 static void qcow2_reopen_abort(BDRVReopenState
*state
)
2031 BDRVQcow2State
*s
= state
->bs
->opaque
;
2033 if (!s
->data_file
) {
2035 * If we don't have an external data file, s->data_file was cleared by
2036 * qcow2_reopen_prepare() and needs to be restored.
2038 s
->data_file
= state
->bs
->file
;
2040 qcow2_update_options_abort(state
->bs
, state
->opaque
);
2041 g_free(state
->opaque
);
2044 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
2046 bool has_new_overlap_template
=
2047 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
2048 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2049 bool has_new_total_cache_size
=
2050 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
2051 bool has_all_cache_options
;
2053 /* New overlap template overrides all old overlap options */
2054 if (has_new_overlap_template
) {
2055 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
2056 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2057 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
2058 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
2059 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
2060 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
2061 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
2062 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
2063 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
2064 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
2067 /* New total cache size overrides all old options */
2068 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
2069 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
2070 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2073 qdict_join(options
, old_options
, false);
2076 * If after merging all cache size options are set, an old total size is
2077 * overwritten. Do keep all options, however, if all three are new. The
2078 * resulting error message is what we want to happen.
2080 has_all_cache_options
=
2081 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
2082 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
2083 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2085 if (has_all_cache_options
&& !has_new_total_cache_size
) {
2086 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
2090 static int coroutine_fn GRAPH_RDLOCK
2091 qcow2_co_block_status(BlockDriverState
*bs
, bool want_zero
, int64_t offset
,
2092 int64_t count
, int64_t *pnum
, int64_t *map
,
2093 BlockDriverState
**file
)
2095 BDRVQcow2State
*s
= bs
->opaque
;
2096 uint64_t host_offset
;
2098 QCow2SubclusterType type
;
2099 int ret
, status
= 0;
2101 qemu_co_mutex_lock(&s
->lock
);
2103 if (!s
->metadata_preallocation_checked
) {
2104 ret
= qcow2_detect_metadata_preallocation(bs
);
2105 s
->metadata_preallocation
= (ret
== 1);
2106 s
->metadata_preallocation_checked
= true;
2109 bytes
= MIN(INT_MAX
, count
);
2110 ret
= qcow2_get_host_offset(bs
, offset
, &bytes
, &host_offset
, &type
);
2111 qemu_co_mutex_unlock(&s
->lock
);
2118 if ((type
== QCOW2_SUBCLUSTER_NORMAL
||
2119 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2120 type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) && !s
->crypto
) {
2122 *file
= s
->data_file
->bs
;
2123 status
|= BDRV_BLOCK_OFFSET_VALID
;
2125 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2126 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
) {
2127 status
|= BDRV_BLOCK_ZERO
;
2128 } else if (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
2129 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) {
2130 status
|= BDRV_BLOCK_DATA
;
2132 if (s
->metadata_preallocation
&& (status
& BDRV_BLOCK_DATA
) &&
2133 (status
& BDRV_BLOCK_OFFSET_VALID
))
2135 status
|= BDRV_BLOCK_RECURSE
;
2140 static int coroutine_fn GRAPH_RDLOCK
2141 qcow2_handle_l2meta(BlockDriverState
*bs
, QCowL2Meta
**pl2meta
, bool link_l2
)
2144 QCowL2Meta
*l2meta
= *pl2meta
;
2146 while (l2meta
!= NULL
) {
2150 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
2155 qcow2_alloc_cluster_abort(bs
, l2meta
);
2158 /* Take the request off the list of running requests */
2159 QLIST_REMOVE(l2meta
, next_in_flight
);
2161 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2163 next
= l2meta
->next
;
2172 static int coroutine_fn GRAPH_RDLOCK
2173 qcow2_co_preadv_encrypted(BlockDriverState
*bs
,
2174 uint64_t host_offset
,
2178 uint64_t qiov_offset
)
2181 BDRVQcow2State
*s
= bs
->opaque
;
2184 assert(bs
->encrypted
&& s
->crypto
);
2185 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2188 * For encrypted images, read everything into a temporary
2189 * contiguous buffer on which the AES functions can work.
2190 * Also, decryption in a separate buffer is better as it
2191 * prevents the guest from learning information about the
2192 * encrypted nature of the virtual disk.
2195 buf
= qemu_try_blockalign(s
->data_file
->bs
, bytes
);
2200 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2201 ret
= bdrv_co_pread(s
->data_file
, host_offset
, bytes
, buf
, 0);
2206 if (qcow2_co_decrypt(bs
, host_offset
, offset
, buf
, bytes
) < 0)
2211 qemu_iovec_from_buf(qiov
, qiov_offset
, buf
, bytes
);
2219 typedef struct Qcow2AioTask
{
2222 BlockDriverState
*bs
;
2223 QCow2SubclusterType subcluster_type
; /* only for read */
2224 uint64_t host_offset
; /* or l2_entry for compressed read */
2228 uint64_t qiov_offset
;
2229 QCowL2Meta
*l2meta
; /* only for write */
2232 static coroutine_fn
int qcow2_co_preadv_task_entry(AioTask
*task
);
2233 static coroutine_fn
int qcow2_add_task(BlockDriverState
*bs
,
2236 QCow2SubclusterType subcluster_type
,
2237 uint64_t host_offset
,
2244 Qcow2AioTask local_task
;
2245 Qcow2AioTask
*task
= pool
? g_new(Qcow2AioTask
, 1) : &local_task
;
2247 *task
= (Qcow2AioTask
) {
2250 .subcluster_type
= subcluster_type
,
2252 .host_offset
= host_offset
,
2255 .qiov_offset
= qiov_offset
,
2259 trace_qcow2_add_task(qemu_coroutine_self(), bs
, pool
,
2260 func
== qcow2_co_preadv_task_entry
? "read" : "write",
2261 subcluster_type
, host_offset
, offset
, bytes
,
2265 return func(&task
->task
);
2268 aio_task_pool_start_task(pool
, &task
->task
);
2273 static int coroutine_fn GRAPH_RDLOCK
2274 qcow2_co_preadv_task(BlockDriverState
*bs
, QCow2SubclusterType subc_type
,
2275 uint64_t host_offset
, uint64_t offset
, uint64_t bytes
,
2276 QEMUIOVector
*qiov
, size_t qiov_offset
)
2278 BDRVQcow2State
*s
= bs
->opaque
;
2280 switch (subc_type
) {
2281 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
2282 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
2283 /* Both zero types are handled in qcow2_co_preadv_part */
2284 g_assert_not_reached();
2286 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
2287 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
2288 assert(bs
->backing
); /* otherwise handled in qcow2_co_preadv_part */
2290 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
2291 return bdrv_co_preadv_part(bs
->backing
, offset
, bytes
,
2292 qiov
, qiov_offset
, 0);
2294 case QCOW2_SUBCLUSTER_COMPRESSED
:
2295 return qcow2_co_preadv_compressed(bs
, host_offset
,
2296 offset
, bytes
, qiov
, qiov_offset
);
2298 case QCOW2_SUBCLUSTER_NORMAL
:
2299 if (bs
->encrypted
) {
2300 return qcow2_co_preadv_encrypted(bs
, host_offset
,
2301 offset
, bytes
, qiov
, qiov_offset
);
2304 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2305 return bdrv_co_preadv_part(s
->data_file
, host_offset
,
2306 bytes
, qiov
, qiov_offset
, 0);
2309 g_assert_not_reached();
2312 g_assert_not_reached();
2316 * This function can count as GRAPH_RDLOCK because qcow2_co_preadv_part() holds
2317 * the graph lock and keeps it until this coroutine has terminated.
2319 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_preadv_task_entry(AioTask
*task
)
2321 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2325 return qcow2_co_preadv_task(t
->bs
, t
->subcluster_type
,
2326 t
->host_offset
, t
->offset
, t
->bytes
,
2327 t
->qiov
, t
->qiov_offset
);
2330 static int coroutine_fn GRAPH_RDLOCK
2331 qcow2_co_preadv_part(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
2332 QEMUIOVector
*qiov
, size_t qiov_offset
,
2333 BdrvRequestFlags flags
)
2335 BDRVQcow2State
*s
= bs
->opaque
;
2337 unsigned int cur_bytes
; /* number of bytes in current iteration */
2338 uint64_t host_offset
= 0;
2339 QCow2SubclusterType type
;
2340 AioTaskPool
*aio
= NULL
;
2342 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2343 /* prepare next request */
2344 cur_bytes
= MIN(bytes
, INT_MAX
);
2346 cur_bytes
= MIN(cur_bytes
,
2347 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2350 qemu_co_mutex_lock(&s
->lock
);
2351 ret
= qcow2_get_host_offset(bs
, offset
, &cur_bytes
,
2352 &host_offset
, &type
);
2353 qemu_co_mutex_unlock(&s
->lock
);
2358 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2359 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2360 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&& !bs
->backing
) ||
2361 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&& !bs
->backing
))
2363 qemu_iovec_memset(qiov
, qiov_offset
, 0, cur_bytes
);
2365 if (!aio
&& cur_bytes
!= bytes
) {
2366 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2368 ret
= qcow2_add_task(bs
, aio
, qcow2_co_preadv_task_entry
, type
,
2369 host_offset
, offset
, cur_bytes
,
2370 qiov
, qiov_offset
, NULL
);
2377 offset
+= cur_bytes
;
2378 qiov_offset
+= cur_bytes
;
2383 aio_task_pool_wait_all(aio
);
2385 ret
= aio_task_pool_status(aio
);
2393 /* Check if it's possible to merge a write request with the writing of
2394 * the data from the COW regions */
2395 static bool merge_cow(uint64_t offset
, unsigned bytes
,
2396 QEMUIOVector
*qiov
, size_t qiov_offset
,
2401 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2402 /* If both COW regions are empty then there's nothing to merge */
2403 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
2407 /* If COW regions are handled already, skip this too */
2413 * The write request should start immediately after the first
2414 * COW region. This does not always happen because the area
2415 * touched by the request can be larger than the one defined
2416 * by @m (a single request can span an area consisting of a
2417 * mix of previously unallocated and allocated clusters, that
2418 * is why @l2meta is a list).
2420 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
2421 /* In this case the request starts before this region */
2422 assert(offset
< l2meta_cow_start(m
));
2423 assert(m
->cow_start
.nb_bytes
== 0);
2427 /* The write request should end immediately before the second
2428 * COW region (see above for why it does not always happen) */
2429 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
2430 assert(offset
+ bytes
> m
->offset
+ m
->cow_end
.offset
);
2431 assert(m
->cow_end
.nb_bytes
== 0);
2435 /* Make sure that adding both COW regions to the QEMUIOVector
2436 * does not exceed IOV_MAX */
2437 if (qemu_iovec_subvec_niov(qiov
, qiov_offset
, bytes
) > IOV_MAX
- 2) {
2441 m
->data_qiov
= qiov
;
2442 m
->data_qiov_offset
= qiov_offset
;
2450 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error.
2451 * Note that returning 0 does not guarantee non-zero data.
2453 static int coroutine_fn GRAPH_RDLOCK
2454 is_zero_cow(BlockDriverState
*bs
, QCowL2Meta
*m
)
2457 * This check is designed for optimization shortcut so it must be
2459 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is
2460 * faster (but not as accurate and can result in false negatives).
2462 int ret
= bdrv_co_is_zero_fast(bs
, m
->offset
+ m
->cow_start
.offset
,
2463 m
->cow_start
.nb_bytes
);
2468 return bdrv_co_is_zero_fast(bs
, m
->offset
+ m
->cow_end
.offset
,
2469 m
->cow_end
.nb_bytes
);
2472 static int coroutine_fn GRAPH_RDLOCK
2473 handle_alloc_space(BlockDriverState
*bs
, QCowL2Meta
*l2meta
)
2475 BDRVQcow2State
*s
= bs
->opaque
;
2478 if (!(s
->data_file
->bs
->supported_zero_flags
& BDRV_REQ_NO_FALLBACK
)) {
2482 if (bs
->encrypted
) {
2486 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2488 uint64_t start_offset
= m
->alloc_offset
+ m
->cow_start
.offset
;
2489 unsigned nb_bytes
= m
->cow_end
.offset
+ m
->cow_end
.nb_bytes
-
2490 m
->cow_start
.offset
;
2492 if (!m
->cow_start
.nb_bytes
&& !m
->cow_end
.nb_bytes
) {
2496 ret
= is_zero_cow(bs
, m
);
2499 } else if (ret
== 0) {
2504 * instead of writing zero COW buffers,
2505 * efficiently zero out the whole clusters
2508 ret
= qcow2_pre_write_overlap_check(bs
, 0, start_offset
, nb_bytes
,
2514 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC_SPACE
);
2515 ret
= bdrv_co_pwrite_zeroes(s
->data_file
, start_offset
, nb_bytes
,
2516 BDRV_REQ_NO_FALLBACK
);
2518 if (ret
!= -ENOTSUP
&& ret
!= -EAGAIN
) {
2524 trace_qcow2_skip_cow(qemu_coroutine_self(), m
->offset
, m
->nb_clusters
);
2531 * qcow2_co_pwritev_task
2532 * Called with s->lock unlocked
2533 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2534 * not use it somehow after qcow2_co_pwritev_task() call
2536 static coroutine_fn GRAPH_RDLOCK
2537 int qcow2_co_pwritev_task(BlockDriverState
*bs
, uint64_t host_offset
,
2538 uint64_t offset
, uint64_t bytes
, QEMUIOVector
*qiov
,
2539 uint64_t qiov_offset
, QCowL2Meta
*l2meta
)
2542 BDRVQcow2State
*s
= bs
->opaque
;
2543 void *crypt_buf
= NULL
;
2544 QEMUIOVector encrypted_qiov
;
2546 if (bs
->encrypted
) {
2548 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2549 crypt_buf
= qemu_try_blockalign(bs
->file
->bs
, bytes
);
2550 if (crypt_buf
== NULL
) {
2554 qemu_iovec_to_buf(qiov
, qiov_offset
, crypt_buf
, bytes
);
2556 if (qcow2_co_encrypt(bs
, host_offset
, offset
, crypt_buf
, bytes
) < 0) {
2561 qemu_iovec_init_buf(&encrypted_qiov
, crypt_buf
, bytes
);
2562 qiov
= &encrypted_qiov
;
2566 /* Try to efficiently initialize the physical space with zeroes */
2567 ret
= handle_alloc_space(bs
, l2meta
);
2573 * If we need to do COW, check if it's possible to merge the
2574 * writing of the guest data together with that of the COW regions.
2575 * If it's not possible (or not necessary) then write the
2578 if (!merge_cow(offset
, bytes
, qiov
, qiov_offset
, l2meta
)) {
2579 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2580 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset
);
2581 ret
= bdrv_co_pwritev_part(s
->data_file
, host_offset
,
2582 bytes
, qiov
, qiov_offset
, 0);
2588 qemu_co_mutex_lock(&s
->lock
);
2590 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
2594 qemu_co_mutex_lock(&s
->lock
);
2597 qcow2_handle_l2meta(bs
, &l2meta
, false);
2598 qemu_co_mutex_unlock(&s
->lock
);
2600 qemu_vfree(crypt_buf
);
2606 * This function can count as GRAPH_RDLOCK because qcow2_co_pwritev_part() holds
2607 * the graph lock and keeps it until this coroutine has terminated.
2609 static coroutine_fn GRAPH_RDLOCK
int qcow2_co_pwritev_task_entry(AioTask
*task
)
2611 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2613 assert(!t
->subcluster_type
);
2615 return qcow2_co_pwritev_task(t
->bs
, t
->host_offset
,
2616 t
->offset
, t
->bytes
, t
->qiov
, t
->qiov_offset
,
2620 static int coroutine_fn GRAPH_RDLOCK
2621 qcow2_co_pwritev_part(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
2622 QEMUIOVector
*qiov
, size_t qiov_offset
,
2623 BdrvRequestFlags flags
)
2625 BDRVQcow2State
*s
= bs
->opaque
;
2626 int offset_in_cluster
;
2628 unsigned int cur_bytes
; /* number of sectors in current iteration */
2629 uint64_t host_offset
;
2630 QCowL2Meta
*l2meta
= NULL
;
2631 AioTaskPool
*aio
= NULL
;
2633 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
2635 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2639 trace_qcow2_writev_start_part(qemu_coroutine_self());
2640 offset_in_cluster
= offset_into_cluster(s
, offset
);
2641 cur_bytes
= MIN(bytes
, INT_MAX
);
2642 if (bs
->encrypted
) {
2643 cur_bytes
= MIN(cur_bytes
,
2644 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
2645 - offset_in_cluster
);
2648 qemu_co_mutex_lock(&s
->lock
);
2650 ret
= qcow2_alloc_host_offset(bs
, offset
, &cur_bytes
,
2651 &host_offset
, &l2meta
);
2656 ret
= qcow2_pre_write_overlap_check(bs
, 0, host_offset
,
2662 qemu_co_mutex_unlock(&s
->lock
);
2664 if (!aio
&& cur_bytes
!= bytes
) {
2665 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2667 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_task_entry
, 0,
2668 host_offset
, offset
,
2669 cur_bytes
, qiov
, qiov_offset
, l2meta
);
2670 l2meta
= NULL
; /* l2meta is consumed by qcow2_co_pwritev_task() */
2676 offset
+= cur_bytes
;
2677 qiov_offset
+= cur_bytes
;
2678 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2682 qemu_co_mutex_lock(&s
->lock
);
2685 qcow2_handle_l2meta(bs
, &l2meta
, false);
2687 qemu_co_mutex_unlock(&s
->lock
);
2691 aio_task_pool_wait_all(aio
);
2693 ret
= aio_task_pool_status(aio
);
2698 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2703 static int qcow2_inactivate(BlockDriverState
*bs
)
2705 BDRVQcow2State
*s
= bs
->opaque
;
2706 int ret
, result
= 0;
2707 Error
*local_err
= NULL
;
2709 qcow2_store_persistent_dirty_bitmaps(bs
, true, &local_err
);
2710 if (local_err
!= NULL
) {
2712 error_reportf_err(local_err
, "Lost persistent bitmaps during "
2713 "inactivation of node '%s': ",
2714 bdrv_get_device_or_node_name(bs
));
2717 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2720 error_report("Failed to flush the L2 table cache: %s",
2724 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2727 error_report("Failed to flush the refcount block cache: %s",
2732 qcow2_mark_clean(bs
);
2738 static void qcow2_do_close(BlockDriverState
*bs
, bool close_data_file
)
2740 BDRVQcow2State
*s
= bs
->opaque
;
2741 qemu_vfree(s
->l1_table
);
2742 /* else pre-write overlap checks in cache_destroy may crash */
2745 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2746 qcow2_inactivate(bs
);
2749 cache_clean_timer_del(bs
);
2750 qcow2_cache_destroy(s
->l2_table_cache
);
2751 qcow2_cache_destroy(s
->refcount_block_cache
);
2753 qcrypto_block_free(s
->crypto
);
2755 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
2757 g_free(s
->unknown_header_fields
);
2758 cleanup_unknown_header_ext(bs
);
2760 g_free(s
->image_data_file
);
2761 g_free(s
->image_backing_file
);
2762 g_free(s
->image_backing_format
);
2764 if (close_data_file
&& has_data_file(bs
)) {
2765 bdrv_unref_child(bs
, s
->data_file
);
2766 s
->data_file
= NULL
;
2769 qcow2_refcount_close(bs
);
2770 qcow2_free_snapshots(bs
);
2773 static void qcow2_close(BlockDriverState
*bs
)
2775 qcow2_do_close(bs
, true);
2778 static void coroutine_fn GRAPH_RDLOCK
2779 qcow2_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
2782 BDRVQcow2State
*s
= bs
->opaque
;
2783 BdrvChild
*data_file
;
2784 int flags
= s
->flags
;
2785 QCryptoBlock
*crypto
= NULL
;
2790 * Backing files are read-only which makes all of their metadata immutable,
2791 * that means we don't have to worry about reopening them here.
2798 * Do not reopen s->data_file (i.e., have qcow2_do_close() not close it,
2799 * and then prevent qcow2_do_open() from opening it), because this function
2800 * runs in the I/O path and as such we must not invoke global-state
2801 * functions like bdrv_unref_child() and bdrv_open_child().
2804 qcow2_do_close(bs
, false);
2806 data_file
= s
->data_file
;
2807 memset(s
, 0, sizeof(BDRVQcow2State
));
2808 s
->data_file
= data_file
;
2810 options
= qdict_clone_shallow(bs
->options
);
2812 flags
&= ~BDRV_O_INACTIVE
;
2813 qemu_co_mutex_lock(&s
->lock
);
2814 ret
= qcow2_do_open(bs
, options
, flags
, false, errp
);
2815 qemu_co_mutex_unlock(&s
->lock
);
2816 qobject_unref(options
);
2818 error_prepend(errp
, "Could not reopen qcow2 layer: ");
2826 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2827 size_t len
, size_t buflen
)
2829 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2830 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2832 if (buflen
< ext_len
) {
2836 *ext_backing_fmt
= (QCowExtension
) {
2837 .magic
= cpu_to_be32(magic
),
2838 .len
= cpu_to_be32(len
),
2842 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2849 * Updates the qcow2 header, including the variable length parts of it, i.e.
2850 * the backing file name and all extensions. qcow2 was not designed to allow
2851 * such changes, so if we run out of space (we can only use the first cluster)
2852 * this function may fail.
2854 * Returns 0 on success, -errno in error cases.
2856 int qcow2_update_header(BlockDriverState
*bs
)
2858 BDRVQcow2State
*s
= bs
->opaque
;
2861 size_t buflen
= s
->cluster_size
;
2863 uint64_t total_size
;
2864 uint32_t refcount_table_clusters
;
2865 size_t header_length
;
2866 Qcow2UnknownHeaderExtension
*uext
;
2868 buf
= qemu_blockalign(bs
, buflen
);
2870 /* Header structure */
2871 header
= (QCowHeader
*) buf
;
2873 if (buflen
< sizeof(*header
)) {
2878 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2879 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2880 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2882 ret
= validate_compression_type(s
, NULL
);
2887 *header
= (QCowHeader
) {
2888 /* Version 2 fields */
2889 .magic
= cpu_to_be32(QCOW_MAGIC
),
2890 .version
= cpu_to_be32(s
->qcow_version
),
2891 .backing_file_offset
= 0,
2892 .backing_file_size
= 0,
2893 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2894 .size
= cpu_to_be64(total_size
),
2895 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2896 .l1_size
= cpu_to_be32(s
->l1_size
),
2897 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2898 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2899 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2900 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2901 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2903 /* Version 3 fields */
2904 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2905 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2906 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2907 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2908 .header_length
= cpu_to_be32(header_length
),
2909 .compression_type
= s
->compression_type
,
2912 /* For older versions, write a shorter header */
2913 switch (s
->qcow_version
) {
2915 ret
= offsetof(QCowHeader
, incompatible_features
);
2918 ret
= sizeof(*header
);
2927 memset(buf
, 0, buflen
);
2929 /* Preserve any unknown field in the header */
2930 if (s
->unknown_header_fields_size
) {
2931 if (buflen
< s
->unknown_header_fields_size
) {
2936 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2937 buf
+= s
->unknown_header_fields_size
;
2938 buflen
-= s
->unknown_header_fields_size
;
2941 /* Backing file format header extension */
2942 if (s
->image_backing_format
) {
2943 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2944 s
->image_backing_format
,
2945 strlen(s
->image_backing_format
),
2955 /* External data file header extension */
2956 if (has_data_file(bs
) && s
->image_data_file
) {
2957 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_DATA_FILE
,
2958 s
->image_data_file
, strlen(s
->image_data_file
),
2968 /* Full disk encryption header pointer extension */
2969 if (s
->crypto_header
.offset
!= 0) {
2970 s
->crypto_header
.offset
= cpu_to_be64(s
->crypto_header
.offset
);
2971 s
->crypto_header
.length
= cpu_to_be64(s
->crypto_header
.length
);
2972 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2973 &s
->crypto_header
, sizeof(s
->crypto_header
),
2975 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
2976 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
2985 * Feature table. A mere 8 feature names occupies 392 bytes, and
2986 * when coupled with the v3 minimum header of 104 bytes plus the
2987 * 8-byte end-of-extension marker, that would leave only 8 bytes
2988 * for a backing file name in an image with 512-byte clusters.
2989 * Thus, we choose to omit this header for cluster sizes 4k and
2992 if (s
->qcow_version
>= 3 && s
->cluster_size
> 4096) {
2993 static const Qcow2Feature features
[] = {
2995 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2996 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2997 .name
= "dirty bit",
3000 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3001 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
3002 .name
= "corrupt bit",
3005 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3006 .bit
= QCOW2_INCOMPAT_DATA_FILE_BITNR
,
3007 .name
= "external data file",
3010 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3011 .bit
= QCOW2_INCOMPAT_COMPRESSION_BITNR
,
3012 .name
= "compression type",
3015 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3016 .bit
= QCOW2_INCOMPAT_EXTL2_BITNR
,
3017 .name
= "extended L2 entries",
3020 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
3021 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
3022 .name
= "lazy refcounts",
3025 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
3026 .bit
= QCOW2_AUTOCLEAR_BITMAPS_BITNR
,
3030 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
3031 .bit
= QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR
,
3032 .name
= "raw external data",
3036 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
3037 features
, sizeof(features
), buflen
);
3045 /* Bitmap extension */
3046 if (s
->nb_bitmaps
> 0) {
3047 Qcow2BitmapHeaderExt bitmaps_header
= {
3048 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
3049 .bitmap_directory_size
=
3050 cpu_to_be64(s
->bitmap_directory_size
),
3051 .bitmap_directory_offset
=
3052 cpu_to_be64(s
->bitmap_directory_offset
)
3054 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
3055 &bitmaps_header
, sizeof(bitmaps_header
),
3064 /* Keep unknown header extensions */
3065 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
3066 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
3075 /* End of header extensions */
3076 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
3084 /* Backing file name */
3085 if (s
->image_backing_file
) {
3086 size_t backing_file_len
= strlen(s
->image_backing_file
);
3088 if (buflen
< backing_file_len
) {
3093 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3094 strncpy(buf
, s
->image_backing_file
, buflen
);
3096 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
3097 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
3100 /* Write the new header */
3101 ret
= bdrv_pwrite(bs
->file
, 0, s
->cluster_size
, header
, 0);
3112 static int qcow2_change_backing_file(BlockDriverState
*bs
,
3113 const char *backing_file
, const char *backing_fmt
)
3115 BDRVQcow2State
*s
= bs
->opaque
;
3117 /* Adding a backing file means that the external data file alone won't be
3118 * enough to make sense of the content */
3119 if (backing_file
&& data_file_is_raw(bs
)) {
3123 if (backing_file
&& strlen(backing_file
) > 1023) {
3127 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3128 backing_file
?: "");
3129 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3130 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3132 g_free(s
->image_backing_file
);
3133 g_free(s
->image_backing_format
);
3135 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
3136 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
3138 return qcow2_update_header(bs
);
3141 static int coroutine_fn GRAPH_RDLOCK
3142 qcow2_set_up_encryption(BlockDriverState
*bs
,
3143 QCryptoBlockCreateOptions
*cryptoopts
,
3146 BDRVQcow2State
*s
= bs
->opaque
;
3147 QCryptoBlock
*crypto
= NULL
;
3150 switch (cryptoopts
->format
) {
3151 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
3152 fmt
= QCOW_CRYPT_LUKS
;
3154 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
3155 fmt
= QCOW_CRYPT_AES
;
3158 error_setg(errp
, "Crypto format not supported in qcow2");
3162 s
->crypt_method_header
= fmt
;
3164 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
3165 qcow2_crypto_hdr_init_func
,
3166 qcow2_crypto_hdr_write_func
,
3172 ret
= qcow2_update_header(bs
);
3174 error_setg_errno(errp
, -ret
, "Could not write encryption header");
3180 qcrypto_block_free(crypto
);
3185 * Preallocates metadata structures for data clusters between @offset (in the
3186 * guest disk) and @new_length (which is thus generally the new guest disk
3189 * Returns: 0 on success, -errno on failure.
3191 static int coroutine_fn GRAPH_RDLOCK
3192 preallocate_co(BlockDriverState
*bs
, uint64_t offset
, uint64_t new_length
,
3193 PreallocMode mode
, Error
**errp
)
3195 BDRVQcow2State
*s
= bs
->opaque
;
3197 uint64_t host_offset
= 0;
3198 int64_t file_length
;
3199 unsigned int cur_bytes
;
3201 QCowL2Meta
*meta
= NULL
, *m
;
3203 assert(offset
<= new_length
);
3204 bytes
= new_length
- offset
;
3207 cur_bytes
= MIN(bytes
, QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
));
3208 ret
= qcow2_alloc_host_offset(bs
, offset
, &cur_bytes
,
3209 &host_offset
, &meta
);
3211 error_setg_errno(errp
, -ret
, "Allocating clusters failed");
3215 for (m
= meta
; m
!= NULL
; m
= m
->next
) {
3219 ret
= qcow2_handle_l2meta(bs
, &meta
, true);
3221 error_setg_errno(errp
, -ret
, "Mapping clusters failed");
3225 /* TODO Preallocate data if requested */
3228 offset
+= cur_bytes
;
3232 * It is expected that the image file is large enough to actually contain
3233 * all of the allocated clusters (otherwise we get failing reads after
3234 * EOF). Extend the image to the last allocated sector.
3236 file_length
= bdrv_co_getlength(s
->data_file
->bs
);
3237 if (file_length
< 0) {
3238 error_setg_errno(errp
, -file_length
, "Could not get file size");
3243 if (host_offset
+ cur_bytes
> file_length
) {
3244 if (mode
== PREALLOC_MODE_METADATA
) {
3245 mode
= PREALLOC_MODE_OFF
;
3247 ret
= bdrv_co_truncate(s
->data_file
, host_offset
+ cur_bytes
, false,
3257 qcow2_handle_l2meta(bs
, &meta
, false);
3261 /* qcow2_refcount_metadata_size:
3262 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3263 * @cluster_size: size of a cluster, in bytes
3264 * @refcount_order: refcount bits power-of-2 exponent
3265 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3268 * Returns: Number of bytes required for refcount blocks and table metadata.
3270 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
3271 int refcount_order
, bool generous_increase
,
3272 uint64_t *refblock_count
)
3275 * Every host cluster is reference-counted, including metadata (even
3276 * refcount metadata is recursively included).
3278 * An accurate formula for the size of refcount metadata size is difficult
3279 * to derive. An easier method of calculation is finding the fixed point
3280 * where no further refcount blocks or table clusters are required to
3281 * reference count every cluster.
3283 int64_t blocks_per_table_cluster
= cluster_size
/ REFTABLE_ENTRY_SIZE
;
3284 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
3285 int64_t table
= 0; /* number of refcount table clusters */
3286 int64_t blocks
= 0; /* number of refcount block clusters */
3292 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
3293 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
3294 n
= clusters
+ blocks
+ table
;
3296 if (n
== last
&& generous_increase
) {
3297 clusters
+= DIV_ROUND_UP(table
, 2);
3298 n
= 0; /* force another loop */
3299 generous_increase
= false;
3301 } while (n
!= last
);
3303 if (refblock_count
) {
3304 *refblock_count
= blocks
;
3307 return (blocks
+ table
) * cluster_size
;
3311 * qcow2_calc_prealloc_size:
3312 * @total_size: virtual disk size in bytes
3313 * @cluster_size: cluster size in bytes
3314 * @refcount_order: refcount bits power-of-2 exponent
3315 * @extended_l2: true if the image has extended L2 entries
3317 * Returns: Total number of bytes required for the fully allocated image
3318 * (including metadata).
3320 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
3321 size_t cluster_size
,
3325 int64_t meta_size
= 0;
3326 uint64_t nl1e
, nl2e
;
3327 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
3328 size_t l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
3330 /* header: 1 cluster */
3331 meta_size
+= cluster_size
;
3333 /* total size of L2 tables */
3334 nl2e
= aligned_total_size
/ cluster_size
;
3335 nl2e
= ROUND_UP(nl2e
, cluster_size
/ l2e_size
);
3336 meta_size
+= nl2e
* l2e_size
;
3338 /* total size of L1 tables */
3339 nl1e
= nl2e
* l2e_size
/ cluster_size
;
3340 nl1e
= ROUND_UP(nl1e
, cluster_size
/ L1E_SIZE
);
3341 meta_size
+= nl1e
* L1E_SIZE
;
3343 /* total size of refcount table and blocks */
3344 meta_size
+= qcow2_refcount_metadata_size(
3345 (meta_size
+ aligned_total_size
) / cluster_size
,
3346 cluster_size
, refcount_order
, false, NULL
);
3348 return meta_size
+ aligned_total_size
;
3351 static bool validate_cluster_size(size_t cluster_size
, bool extended_l2
,
3354 int cluster_bits
= ctz32(cluster_size
);
3355 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
3356 (1 << cluster_bits
) != cluster_size
)
3358 error_setg(errp
, "Cluster size must be a power of two between %d and "
3359 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
3364 unsigned min_cluster_size
=
3365 (1 << MIN_CLUSTER_BITS
) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
;
3366 if (cluster_size
< min_cluster_size
) {
3367 error_setg(errp
, "Extended L2 entries are only supported with "
3368 "cluster sizes of at least %u bytes", min_cluster_size
);
3376 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, bool extended_l2
,
3379 size_t cluster_size
;
3381 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
3382 DEFAULT_CLUSTER_SIZE
);
3383 if (!validate_cluster_size(cluster_size
, extended_l2
, errp
)) {
3386 return cluster_size
;
3389 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
3394 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
3396 ret
= 3; /* default */
3397 } else if (!strcmp(buf
, "0.10")) {
3399 } else if (!strcmp(buf
, "1.1")) {
3402 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
3409 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
3412 uint64_t refcount_bits
;
3414 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
3415 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
3416 error_setg(errp
, "Refcount width must be a power of two and may not "
3421 if (version
< 3 && refcount_bits
!= 16) {
3422 error_setg(errp
, "Different refcount widths than 16 bits require "
3423 "compatibility level 1.1 or above (use compat=1.1 or "
3428 return refcount_bits
;
3431 static int coroutine_fn GRAPH_UNLOCKED
3432 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
3434 BlockdevCreateOptionsQcow2
*qcow2_opts
;
3438 * Open the image file and write a minimal qcow2 header.
3440 * We keep things simple and start with a zero-sized image. We also
3441 * do without refcount blocks or a L1 table for now. We'll fix the
3442 * inconsistency later.
3444 * We do need a refcount table because growing the refcount table means
3445 * allocating two new refcount blocks - the second of which would be at
3446 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3447 * size for any qcow2 image.
3449 BlockBackend
*blk
= NULL
;
3450 BlockDriverState
*bs
= NULL
;
3451 BlockDriverState
*data_bs
= NULL
;
3453 size_t cluster_size
;
3456 uint64_t *refcount_table
;
3458 uint8_t compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
3460 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
3461 qcow2_opts
= &create_options
->u
.qcow2
;
3463 bs
= bdrv_co_open_blockdev_ref(qcow2_opts
->file
, errp
);
3468 /* Validate options and set default values */
3469 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
3470 error_setg(errp
, "Image size must be a multiple of %u bytes",
3471 (unsigned) BDRV_SECTOR_SIZE
);
3476 if (qcow2_opts
->has_version
) {
3477 switch (qcow2_opts
->version
) {
3478 case BLOCKDEV_QCOW2_VERSION_V2
:
3481 case BLOCKDEV_QCOW2_VERSION_V3
:
3485 g_assert_not_reached();
3491 if (qcow2_opts
->has_cluster_size
) {
3492 cluster_size
= qcow2_opts
->cluster_size
;
3494 cluster_size
= DEFAULT_CLUSTER_SIZE
;
3497 if (!qcow2_opts
->has_extended_l2
) {
3498 qcow2_opts
->extended_l2
= false;
3500 if (qcow2_opts
->extended_l2
) {
3502 error_setg(errp
, "Extended L2 entries are only supported with "
3503 "compatibility level 1.1 and above (use version=v3 or "
3510 if (!validate_cluster_size(cluster_size
, qcow2_opts
->extended_l2
, errp
)) {
3515 if (!qcow2_opts
->has_preallocation
) {
3516 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
3518 if (qcow2_opts
->backing_file
&&
3519 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
&&
3520 !qcow2_opts
->extended_l2
)
3522 error_setg(errp
, "Backing file and preallocation can only be used at "
3523 "the same time if extended_l2 is on");
3527 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->backing_file
) {
3528 error_setg(errp
, "Backing format cannot be used without backing file");
3533 if (!qcow2_opts
->has_lazy_refcounts
) {
3534 qcow2_opts
->lazy_refcounts
= false;
3536 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
3537 error_setg(errp
, "Lazy refcounts only supported with compatibility "
3538 "level 1.1 and above (use version=v3 or greater)");
3543 if (!qcow2_opts
->has_refcount_bits
) {
3544 qcow2_opts
->refcount_bits
= 16;
3546 if (qcow2_opts
->refcount_bits
> 64 ||
3547 !is_power_of_2(qcow2_opts
->refcount_bits
))
3549 error_setg(errp
, "Refcount width must be a power of two and may not "
3554 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
3555 error_setg(errp
, "Different refcount widths than 16 bits require "
3556 "compatibility level 1.1 or above (use version=v3 or "
3561 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
3563 if (qcow2_opts
->data_file_raw
&& !qcow2_opts
->data_file
) {
3564 error_setg(errp
, "data-file-raw requires data-file");
3568 if (qcow2_opts
->data_file_raw
&& qcow2_opts
->backing_file
) {
3569 error_setg(errp
, "Backing file and data-file-raw cannot be used at "
3574 if (qcow2_opts
->data_file_raw
&&
3575 qcow2_opts
->preallocation
== PREALLOC_MODE_OFF
)
3578 * data-file-raw means that "the external data file can be
3579 * read as a consistent standalone raw image without looking
3580 * at the qcow2 metadata." It does not say that the metadata
3581 * must be ignored, though (and the qcow2 driver in fact does
3582 * not ignore it), so the L1/L2 tables must be present and
3583 * give a 1:1 mapping, so you get the same result regardless
3584 * of whether you look at the metadata or whether you ignore
3587 qcow2_opts
->preallocation
= PREALLOC_MODE_METADATA
;
3590 * Cannot use preallocation with backing files, but giving a
3591 * backing file when specifying data_file_raw is an error
3594 assert(!qcow2_opts
->backing_file
);
3597 if (qcow2_opts
->data_file
) {
3599 error_setg(errp
, "External data files are only supported with "
3600 "compatibility level 1.1 and above (use version=v3 or "
3605 data_bs
= bdrv_co_open_blockdev_ref(qcow2_opts
->data_file
, errp
);
3606 if (data_bs
== NULL
) {
3612 if (qcow2_opts
->has_compression_type
&&
3613 qcow2_opts
->compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3618 error_setg(errp
, "Non-zlib compression type is only supported with "
3619 "compatibility level 1.1 and above (use version=v3 or "
3624 switch (qcow2_opts
->compression_type
) {
3626 case QCOW2_COMPRESSION_TYPE_ZSTD
:
3630 error_setg(errp
, "Unknown compression type");
3634 compression_type
= qcow2_opts
->compression_type
;
3637 /* Create BlockBackend to write to the image */
3638 blk
= blk_co_new_with_bs(bs
, BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
,
3644 blk_set_allow_write_beyond_eof(blk
, true);
3646 /* Write the header */
3647 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
3648 header
= g_malloc0(cluster_size
);
3649 *header
= (QCowHeader
) {
3650 .magic
= cpu_to_be32(QCOW_MAGIC
),
3651 .version
= cpu_to_be32(version
),
3652 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
3653 .size
= cpu_to_be64(0),
3654 .l1_table_offset
= cpu_to_be64(0),
3655 .l1_size
= cpu_to_be32(0),
3656 .refcount_table_offset
= cpu_to_be64(cluster_size
),
3657 .refcount_table_clusters
= cpu_to_be32(1),
3658 .refcount_order
= cpu_to_be32(refcount_order
),
3659 /* don't deal with endianness since compression_type is 1 byte long */
3660 .compression_type
= compression_type
,
3661 .header_length
= cpu_to_be32(sizeof(*header
)),
3664 /* We'll update this to correct value later */
3665 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
3667 if (qcow2_opts
->lazy_refcounts
) {
3668 header
->compatible_features
|=
3669 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
3672 header
->incompatible_features
|=
3673 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE
);
3675 if (qcow2_opts
->data_file_raw
) {
3676 header
->autoclear_features
|=
3677 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW
);
3679 if (compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3680 header
->incompatible_features
|=
3681 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION
);
3684 if (qcow2_opts
->extended_l2
) {
3685 header
->incompatible_features
|=
3686 cpu_to_be64(QCOW2_INCOMPAT_EXTL2
);
3689 ret
= blk_co_pwrite(blk
, 0, cluster_size
, header
, 0);
3692 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
3696 /* Write a refcount table with one refcount block */
3697 refcount_table
= g_malloc0(2 * cluster_size
);
3698 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
3699 ret
= blk_co_pwrite(blk
, cluster_size
, 2 * cluster_size
, refcount_table
, 0);
3700 g_free(refcount_table
);
3703 error_setg_errno(errp
, -ret
, "Could not write refcount table");
3711 * And now open the image and make it consistent first (i.e. increase the
3712 * refcount of the cluster that is occupied by the header and the refcount
3715 options
= qdict_new();
3716 qdict_put_str(options
, "driver", "qcow2");
3717 qdict_put_str(options
, "file", bs
->node_name
);
3719 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3721 blk
= blk_co_new_open(NULL
, NULL
, options
,
3722 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
3729 bdrv_graph_co_rdlock();
3730 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
3732 bdrv_graph_co_rdunlock();
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
));
3750 bdrv_graph_co_rdunlock();
3753 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
3757 /* Okay, now that we have a valid image, let's give it the right size */
3758 ret
= blk_co_truncate(blk
, qcow2_opts
->size
, false,
3759 qcow2_opts
->preallocation
, 0, errp
);
3761 error_prepend(errp
, "Could not resize image: ");
3765 /* Want a backing file? There you go. */
3766 if (qcow2_opts
->backing_file
) {
3767 const char *backing_format
= NULL
;
3769 if (qcow2_opts
->has_backing_fmt
) {
3770 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
3773 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
3774 backing_format
, false);
3776 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
3777 "with format '%s'", qcow2_opts
->backing_file
,
3783 /* Want encryption? There you go. */
3784 if (qcow2_opts
->encrypt
) {
3785 bdrv_graph_co_rdlock();
3786 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3787 bdrv_graph_co_rdunlock();
3797 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3798 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3799 * have to setup decryption context. We're not doing any I/O on the top
3800 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3803 options
= qdict_new();
3804 qdict_put_str(options
, "driver", "qcow2");
3805 qdict_put_str(options
, "file", bs
->node_name
);
3807 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3809 blk
= blk_co_new_open(NULL
, NULL
, options
,
3810 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3821 bdrv_co_unref(data_bs
);
3825 static int coroutine_fn GRAPH_UNLOCKED
3826 qcow2_co_create_opts(BlockDriver
*drv
, const char *filename
, QemuOpts
*opts
,
3829 BlockdevCreateOptions
*create_options
= NULL
;
3832 BlockDriverState
*bs
= NULL
;
3833 BlockDriverState
*data_bs
= NULL
;
3837 /* Only the keyval visitor supports the dotted syntax needed for
3838 * encryption, so go through a QDict before getting a QAPI type. Ignore
3839 * options meant for the protocol layer so that the visitor doesn't
3841 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3844 /* Handle encryption options */
3845 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3846 if (val
&& !strcmp(val
, "on")) {
3847 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3848 } else if (val
&& !strcmp(val
, "off")) {
3849 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3852 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3853 if (val
&& !strcmp(val
, "aes")) {
3854 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3857 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3858 * version=v2/v3 below. */
3859 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3860 if (val
&& !strcmp(val
, "0.10")) {
3861 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3862 } else if (val
&& !strcmp(val
, "1.1")) {
3863 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3866 /* Change legacy command line options into QMP ones */
3867 static const QDictRenames opt_renames
[] = {
3868 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3869 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3870 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3871 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3872 { BLOCK_OPT_EXTL2
, "extended-l2" },
3873 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3874 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3875 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3876 { BLOCK_OPT_DATA_FILE_RAW
, "data-file-raw" },
3877 { BLOCK_OPT_COMPRESSION_TYPE
, "compression-type" },
3881 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3886 /* Create and open the file (protocol layer) */
3887 ret
= bdrv_co_create_file(filename
, opts
, errp
);
3892 bs
= bdrv_co_open(filename
, NULL
, NULL
,
3893 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3899 /* Create and open an external data file (protocol layer) */
3900 val
= qdict_get_try_str(qdict
, BLOCK_OPT_DATA_FILE
);
3902 ret
= bdrv_co_create_file(val
, opts
, errp
);
3907 data_bs
= bdrv_co_open(val
, NULL
, NULL
,
3908 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
,
3910 if (data_bs
== NULL
) {
3915 qdict_del(qdict
, BLOCK_OPT_DATA_FILE
);
3916 qdict_put_str(qdict
, "data-file", data_bs
->node_name
);
3919 /* Set 'driver' and 'node' options */
3920 qdict_put_str(qdict
, "driver", "qcow2");
3921 qdict_put_str(qdict
, "file", bs
->node_name
);
3923 /* Now get the QAPI type BlockdevCreateOptions */
3924 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
3930 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, errp
);
3932 if (!create_options
) {
3937 /* Silently round up size */
3938 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3941 /* Create the qcow2 image (format layer) */
3942 ret
= qcow2_co_create(create_options
, errp
);
3945 bdrv_graph_co_rdlock();
3946 bdrv_co_delete_file_noerr(bs
);
3947 bdrv_co_delete_file_noerr(data_bs
);
3948 bdrv_graph_co_rdunlock();
3953 qobject_unref(qdict
);
3955 bdrv_co_unref(data_bs
);
3956 qapi_free_BlockdevCreateOptions(create_options
);
3961 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3966 /* Clamp to image length, before checking status of underlying sectors */
3967 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3968 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3976 * bdrv_block_status_above doesn't merge different types of zeros, for
3977 * example, zeros which come from the region which is unallocated in
3978 * the whole backing chain, and zeros which come because of a short
3979 * backing file. So, we need a loop.
3982 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3985 } while (res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
&& bytes
);
3987 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && bytes
== 0;
3990 static int coroutine_fn GRAPH_RDLOCK
3991 qcow2_co_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
3992 BdrvRequestFlags flags
)
3995 BDRVQcow2State
*s
= bs
->opaque
;
3997 uint32_t head
= offset_into_subcluster(s
, offset
);
3998 uint32_t tail
= ROUND_UP(offset
+ bytes
, s
->subcluster_size
) -
4001 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
4002 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
4009 QCow2SubclusterType type
;
4011 assert(head
+ bytes
+ tail
<= s
->subcluster_size
);
4013 /* check whether remainder of cluster already reads as zero */
4014 if (!(is_zero(bs
, offset
- head
, head
) &&
4015 is_zero(bs
, offset
+ bytes
, tail
))) {
4019 qemu_co_mutex_lock(&s
->lock
);
4020 /* We can have new write after previous check */
4022 bytes
= s
->subcluster_size
;
4023 nr
= s
->subcluster_size
;
4024 ret
= qcow2_get_host_offset(bs
, offset
, &nr
, &off
, &type
);
4026 (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
4027 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&&
4028 type
!= QCOW2_SUBCLUSTER_ZERO_PLAIN
&&
4029 type
!= QCOW2_SUBCLUSTER_ZERO_ALLOC
)) {
4030 qemu_co_mutex_unlock(&s
->lock
);
4031 return ret
< 0 ? ret
: -ENOTSUP
;
4034 qemu_co_mutex_lock(&s
->lock
);
4037 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
4039 /* Whatever is left can use real zero subclusters */
4040 ret
= qcow2_subcluster_zeroize(bs
, offset
, bytes
, flags
);
4041 qemu_co_mutex_unlock(&s
->lock
);
4046 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
4047 int64_t offset
, int64_t bytes
)
4050 BDRVQcow2State
*s
= bs
->opaque
;
4052 /* If the image does not support QCOW_OFLAG_ZERO then discarding
4053 * clusters could expose stale data from the backing file. */
4054 if (s
->qcow_version
< 3 && bs
->backing
) {
4058 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
4059 assert(bytes
< s
->cluster_size
);
4060 /* Ignore partial clusters, except for the special case of the
4061 * complete partial cluster at the end of an unaligned file */
4062 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
4063 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
4068 qemu_co_mutex_lock(&s
->lock
);
4069 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
4071 qemu_co_mutex_unlock(&s
->lock
);
4075 static int coroutine_fn GRAPH_RDLOCK
4076 qcow2_co_copy_range_from(BlockDriverState
*bs
,
4077 BdrvChild
*src
, int64_t src_offset
,
4078 BdrvChild
*dst
, int64_t dst_offset
,
4079 int64_t bytes
, BdrvRequestFlags read_flags
,
4080 BdrvRequestFlags write_flags
)
4082 BDRVQcow2State
*s
= bs
->opaque
;
4084 unsigned int cur_bytes
; /* number of bytes in current iteration */
4085 BdrvChild
*child
= NULL
;
4086 BdrvRequestFlags cur_write_flags
;
4088 assert(!bs
->encrypted
);
4089 qemu_co_mutex_lock(&s
->lock
);
4091 while (bytes
!= 0) {
4092 uint64_t copy_offset
= 0;
4093 QCow2SubclusterType type
;
4094 /* prepare next request */
4095 cur_bytes
= MIN(bytes
, INT_MAX
);
4096 cur_write_flags
= write_flags
;
4098 ret
= qcow2_get_host_offset(bs
, src_offset
, &cur_bytes
,
4099 ©_offset
, &type
);
4105 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
4106 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
4107 if (bs
->backing
&& bs
->backing
->bs
) {
4108 int64_t backing_length
= bdrv_co_getlength(bs
->backing
->bs
);
4109 if (src_offset
>= backing_length
) {
4110 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4112 child
= bs
->backing
;
4113 cur_bytes
= MIN(cur_bytes
, backing_length
- src_offset
);
4114 copy_offset
= src_offset
;
4117 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4121 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
4122 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
4123 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4126 case QCOW2_SUBCLUSTER_COMPRESSED
:
4130 case QCOW2_SUBCLUSTER_NORMAL
:
4131 child
= s
->data_file
;
4137 qemu_co_mutex_unlock(&s
->lock
);
4138 ret
= bdrv_co_copy_range_from(child
,
4141 cur_bytes
, read_flags
, cur_write_flags
);
4142 qemu_co_mutex_lock(&s
->lock
);
4148 src_offset
+= cur_bytes
;
4149 dst_offset
+= cur_bytes
;
4154 qemu_co_mutex_unlock(&s
->lock
);
4158 static int coroutine_fn GRAPH_RDLOCK
4159 qcow2_co_copy_range_to(BlockDriverState
*bs
,
4160 BdrvChild
*src
, int64_t src_offset
,
4161 BdrvChild
*dst
, int64_t dst_offset
,
4162 int64_t bytes
, BdrvRequestFlags read_flags
,
4163 BdrvRequestFlags write_flags
)
4165 BDRVQcow2State
*s
= bs
->opaque
;
4167 unsigned int cur_bytes
; /* number of sectors in current iteration */
4168 uint64_t host_offset
;
4169 QCowL2Meta
*l2meta
= NULL
;
4171 assert(!bs
->encrypted
);
4173 qemu_co_mutex_lock(&s
->lock
);
4175 while (bytes
!= 0) {
4179 cur_bytes
= MIN(bytes
, INT_MAX
);
4182 * If src->bs == dst->bs, we could simply copy by incrementing
4183 * the refcnt, without copying user data.
4184 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4185 ret
= qcow2_alloc_host_offset(bs
, dst_offset
, &cur_bytes
,
4186 &host_offset
, &l2meta
);
4191 ret
= qcow2_pre_write_overlap_check(bs
, 0, host_offset
, cur_bytes
,
4197 qemu_co_mutex_unlock(&s
->lock
);
4198 ret
= bdrv_co_copy_range_to(src
, src_offset
, s
->data_file
, host_offset
,
4199 cur_bytes
, read_flags
, write_flags
);
4200 qemu_co_mutex_lock(&s
->lock
);
4205 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
4211 src_offset
+= cur_bytes
;
4212 dst_offset
+= cur_bytes
;
4217 qcow2_handle_l2meta(bs
, &l2meta
, false);
4219 qemu_co_mutex_unlock(&s
->lock
);
4221 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
4226 static int coroutine_fn GRAPH_RDLOCK
4227 qcow2_co_truncate(BlockDriverState
*bs
, int64_t offset
, bool exact
,
4228 PreallocMode prealloc
, BdrvRequestFlags flags
, Error
**errp
)
4230 BDRVQcow2State
*s
= bs
->opaque
;
4231 uint64_t old_length
;
4232 int64_t new_l1_size
;
4236 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
4237 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
4239 error_setg(errp
, "Unsupported preallocation mode '%s'",
4240 PreallocMode_str(prealloc
));
4244 if (!QEMU_IS_ALIGNED(offset
, BDRV_SECTOR_SIZE
)) {
4245 error_setg(errp
, "The new size must be a multiple of %u",
4246 (unsigned) BDRV_SECTOR_SIZE
);
4250 qemu_co_mutex_lock(&s
->lock
);
4253 * Even though we store snapshot size for all images, it was not
4254 * required until v3, so it is not safe to proceed for v2.
4256 if (s
->nb_snapshots
&& s
->qcow_version
< 3) {
4257 error_setg(errp
, "Can't resize a v2 image which has snapshots");
4262 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4263 if (qcow2_truncate_bitmaps_check(bs
, errp
)) {
4268 old_length
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4269 new_l1_size
= size_to_l1(s
, offset
);
4271 if (offset
< old_length
) {
4272 int64_t last_cluster
, old_file_size
;
4273 if (prealloc
!= PREALLOC_MODE_OFF
) {
4275 "Preallocation can't be used for shrinking an image");
4280 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
4281 old_length
- ROUND_UP(offset
,
4283 QCOW2_DISCARD_ALWAYS
, true);
4285 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
4289 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
4291 error_setg_errno(errp
, -ret
,
4292 "Failed to reduce the number of L2 tables");
4296 ret
= qcow2_shrink_reftable(bs
);
4298 error_setg_errno(errp
, -ret
,
4299 "Failed to discard unused refblocks");
4303 old_file_size
= bdrv_co_getlength(bs
->file
->bs
);
4304 if (old_file_size
< 0) {
4305 error_setg_errno(errp
, -old_file_size
,
4306 "Failed to inquire current file length");
4307 ret
= old_file_size
;
4310 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4311 if (last_cluster
< 0) {
4312 error_setg_errno(errp
, -last_cluster
,
4313 "Failed to find the last cluster");
4317 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
4318 Error
*local_err
= NULL
;
4321 * Do not pass @exact here: It will not help the user if
4322 * we get an error here just because they wanted to shrink
4323 * their qcow2 image (on a block device) with qemu-img.
4324 * (And on the qcow2 layer, the @exact requirement is
4325 * always fulfilled, so there is no need to pass it on.)
4327 bdrv_co_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
4328 false, PREALLOC_MODE_OFF
, 0, &local_err
);
4330 warn_reportf_err(local_err
,
4331 "Failed to truncate the tail of the image: ");
4335 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
4337 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
4341 if (data_file_is_raw(bs
) && prealloc
== PREALLOC_MODE_OFF
) {
4343 * When creating a qcow2 image with data-file-raw, we enforce
4344 * at least prealloc=metadata, so that the L1/L2 tables are
4345 * fully allocated and reading from the data file will return
4346 * the same data as reading from the qcow2 image. When the
4347 * image is grown, we must consequently preallocate the
4348 * metadata structures to cover the added area.
4350 prealloc
= PREALLOC_MODE_METADATA
;
4355 case PREALLOC_MODE_OFF
:
4356 if (has_data_file(bs
)) {
4358 * If the caller wants an exact resize, the external data
4359 * file should be resized to the exact target size, too,
4360 * so we pass @exact here.
4362 ret
= bdrv_co_truncate(s
->data_file
, offset
, exact
, prealloc
, 0,
4370 case PREALLOC_MODE_METADATA
:
4371 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4377 case PREALLOC_MODE_FALLOC
:
4378 case PREALLOC_MODE_FULL
:
4380 int64_t allocation_start
, host_offset
, guest_offset
;
4381 int64_t clusters_allocated
;
4382 int64_t old_file_size
, last_cluster
, new_file_size
;
4383 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
4384 bool subclusters_need_allocation
= false;
4386 /* With a data file, preallocation means just allocating the metadata
4387 * and forwarding the truncate request to the data file */
4388 if (has_data_file(bs
)) {
4389 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4396 old_file_size
= bdrv_co_getlength(bs
->file
->bs
);
4397 if (old_file_size
< 0) {
4398 error_setg_errno(errp
, -old_file_size
,
4399 "Failed to inquire current file length");
4400 ret
= old_file_size
;
4404 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4405 if (last_cluster
>= 0) {
4406 old_file_size
= (last_cluster
+ 1) * s
->cluster_size
;
4408 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
4411 nb_new_data_clusters
= (ROUND_UP(offset
, s
->cluster_size
) -
4412 start_of_cluster(s
, old_length
)) >> s
->cluster_bits
;
4414 /* This is an overestimation; we will not actually allocate space for
4415 * these in the file but just make sure the new refcount structures are
4416 * able to cover them so we will not have to allocate new refblocks
4417 * while entering the data blocks in the potentially new L2 tables.
4418 * (We do not actually care where the L2 tables are placed. Maybe they
4419 * are already allocated or they can be placed somewhere before
4420 * @old_file_size. It does not matter because they will be fully
4421 * allocated automatically, so they do not need to be covered by the
4422 * preallocation. All that matters is that we will not have to allocate
4423 * new refcount structures for them.) */
4424 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
4425 s
->cluster_size
/ l2_entry_size(s
));
4426 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4427 * table for a potential head/tail */
4430 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
4431 nb_new_data_clusters
+
4434 if (allocation_start
< 0) {
4435 error_setg_errno(errp
, -allocation_start
,
4436 "Failed to resize refcount structures");
4437 ret
= allocation_start
;
4441 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
4442 nb_new_data_clusters
);
4443 if (clusters_allocated
< 0) {
4444 error_setg_errno(errp
, -clusters_allocated
,
4445 "Failed to allocate data clusters");
4446 ret
= clusters_allocated
;
4450 assert(clusters_allocated
== nb_new_data_clusters
);
4452 /* Allocate the data area */
4453 new_file_size
= allocation_start
+
4454 nb_new_data_clusters
* s
->cluster_size
;
4456 * Image file grows, so @exact does not matter.
4458 * If we need to zero out the new area, try first whether the protocol
4459 * driver can already take care of this.
4461 if (flags
& BDRV_REQ_ZERO_WRITE
) {
4462 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
,
4463 BDRV_REQ_ZERO_WRITE
, NULL
);
4465 flags
&= ~BDRV_REQ_ZERO_WRITE
;
4466 /* Ensure that we read zeroes and not backing file data */
4467 subclusters_need_allocation
= true;
4473 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
, 0,
4477 error_prepend(errp
, "Failed to resize underlying file: ");
4478 qcow2_free_clusters(bs
, allocation_start
,
4479 nb_new_data_clusters
* s
->cluster_size
,
4480 QCOW2_DISCARD_OTHER
);
4484 /* Create the necessary L2 entries */
4485 host_offset
= allocation_start
;
4486 guest_offset
= old_length
;
4487 while (nb_new_data_clusters
) {
4488 int64_t nb_clusters
= MIN(
4489 nb_new_data_clusters
,
4490 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
4491 unsigned cow_start_length
= offset_into_cluster(s
, guest_offset
);
4492 QCowL2Meta allocation
;
4493 guest_offset
= start_of_cluster(s
, guest_offset
);
4494 allocation
= (QCowL2Meta
) {
4495 .offset
= guest_offset
,
4496 .alloc_offset
= host_offset
,
4497 .nb_clusters
= nb_clusters
,
4500 .nb_bytes
= cow_start_length
,
4503 .offset
= nb_clusters
<< s
->cluster_bits
,
4506 .prealloc
= !subclusters_need_allocation
,
4508 qemu_co_queue_init(&allocation
.dependent_requests
);
4510 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
4512 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
4513 qcow2_free_clusters(bs
, host_offset
,
4514 nb_new_data_clusters
* s
->cluster_size
,
4515 QCOW2_DISCARD_OTHER
);
4519 guest_offset
+= nb_clusters
* s
->cluster_size
;
4520 host_offset
+= nb_clusters
* s
->cluster_size
;
4521 nb_new_data_clusters
-= nb_clusters
;
4527 g_assert_not_reached();
4530 if ((flags
& BDRV_REQ_ZERO_WRITE
) && offset
> old_length
) {
4531 uint64_t zero_start
= QEMU_ALIGN_UP(old_length
, s
->subcluster_size
);
4534 * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4535 * requires a subcluster-aligned start. The end may be unaligned if
4536 * it is at the end of the image (which it is here).
4538 if (offset
> zero_start
) {
4539 ret
= qcow2_subcluster_zeroize(bs
, zero_start
, offset
- zero_start
,
4542 error_setg_errno(errp
, -ret
, "Failed to zero out new clusters");
4547 /* Write explicit zeros for the unaligned head */
4548 if (zero_start
> old_length
) {
4549 uint64_t len
= MIN(zero_start
, offset
) - old_length
;
4550 uint8_t *buf
= qemu_blockalign0(bs
, len
);
4552 qemu_iovec_init_buf(&qiov
, buf
, len
);
4554 qemu_co_mutex_unlock(&s
->lock
);
4555 ret
= qcow2_co_pwritev_part(bs
, old_length
, len
, &qiov
, 0, 0);
4556 qemu_co_mutex_lock(&s
->lock
);
4560 error_setg_errno(errp
, -ret
, "Failed to zero out the new area");
4566 if (prealloc
!= PREALLOC_MODE_OFF
) {
4567 /* Flush metadata before actually changing the image size */
4568 ret
= qcow2_write_caches(bs
);
4570 error_setg_errno(errp
, -ret
,
4571 "Failed to flush the preallocated area to disk");
4576 bs
->total_sectors
= offset
/ BDRV_SECTOR_SIZE
;
4578 /* write updated header.size */
4579 offset
= cpu_to_be64(offset
);
4580 ret
= bdrv_co_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
4581 sizeof(offset
), &offset
, 0);
4583 error_setg_errno(errp
, -ret
, "Failed to update the image size");
4587 s
->l1_vm_state_index
= new_l1_size
;
4589 /* Update cache sizes */
4590 options
= qdict_clone_shallow(bs
->options
);
4591 ret
= qcow2_update_options(bs
, options
, s
->flags
, errp
);
4592 qobject_unref(options
);
4598 qemu_co_mutex_unlock(&s
->lock
);
4602 static int coroutine_fn GRAPH_RDLOCK
4603 qcow2_co_pwritev_compressed_task(BlockDriverState
*bs
,
4604 uint64_t offset
, uint64_t bytes
,
4605 QEMUIOVector
*qiov
, size_t qiov_offset
)
4607 BDRVQcow2State
*s
= bs
->opaque
;
4610 uint8_t *buf
, *out_buf
;
4611 uint64_t cluster_offset
;
4613 assert(bytes
== s
->cluster_size
|| (bytes
< s
->cluster_size
&&
4614 (offset
+ bytes
== bs
->total_sectors
<< BDRV_SECTOR_BITS
)));
4616 buf
= qemu_blockalign(bs
, s
->cluster_size
);
4617 if (bytes
< s
->cluster_size
) {
4618 /* Zero-pad last write if image size is not cluster aligned */
4619 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
4621 qemu_iovec_to_buf(qiov
, qiov_offset
, buf
, bytes
);
4623 out_buf
= g_malloc(s
->cluster_size
);
4625 out_len
= qcow2_co_compress(bs
, out_buf
, s
->cluster_size
- 1,
4626 buf
, s
->cluster_size
);
4627 if (out_len
== -ENOMEM
) {
4628 /* could not compress: write normal cluster */
4629 ret
= qcow2_co_pwritev_part(bs
, offset
, bytes
, qiov
, qiov_offset
, 0);
4634 } else if (out_len
< 0) {
4639 qemu_co_mutex_lock(&s
->lock
);
4640 ret
= qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
,
4643 qemu_co_mutex_unlock(&s
->lock
);
4647 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
, true);
4648 qemu_co_mutex_unlock(&s
->lock
);
4653 BLKDBG_EVENT(s
->data_file
, BLKDBG_WRITE_COMPRESSED
);
4654 ret
= bdrv_co_pwrite(s
->data_file
, cluster_offset
, out_len
, out_buf
, 0);
4667 * This function can count as GRAPH_RDLOCK because
4668 * qcow2_co_pwritev_compressed_part() holds the graph lock and keeps it until
4669 * this coroutine has terminated.
4671 static int coroutine_fn GRAPH_RDLOCK
4672 qcow2_co_pwritev_compressed_task_entry(AioTask
*task
)
4674 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
4676 assert(!t
->subcluster_type
&& !t
->l2meta
);
4678 return qcow2_co_pwritev_compressed_task(t
->bs
, t
->offset
, t
->bytes
, t
->qiov
,
4683 * XXX: put compressed sectors first, then all the cluster aligned
4684 * tables to avoid losing bytes in alignment
4686 static int coroutine_fn GRAPH_RDLOCK
4687 qcow2_co_pwritev_compressed_part(BlockDriverState
*bs
,
4688 int64_t offset
, int64_t bytes
,
4689 QEMUIOVector
*qiov
, size_t qiov_offset
)
4691 BDRVQcow2State
*s
= bs
->opaque
;
4692 AioTaskPool
*aio
= NULL
;
4695 if (has_data_file(bs
)) {
4701 * align end of file to a sector boundary to ease reading with
4704 int64_t len
= bdrv_co_getlength(bs
->file
->bs
);
4708 return bdrv_co_truncate(bs
->file
, len
, false, PREALLOC_MODE_OFF
, 0,
4712 if (offset_into_cluster(s
, offset
)) {
4716 if (offset_into_cluster(s
, bytes
) &&
4717 (offset
+ bytes
) != (bs
->total_sectors
<< BDRV_SECTOR_BITS
)) {
4721 while (bytes
&& aio_task_pool_status(aio
) == 0) {
4722 uint64_t chunk_size
= MIN(bytes
, s
->cluster_size
);
4724 if (!aio
&& chunk_size
!= bytes
) {
4725 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
4728 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_compressed_task_entry
,
4729 0, 0, offset
, chunk_size
, qiov
, qiov_offset
, NULL
);
4733 qiov_offset
+= chunk_size
;
4734 offset
+= chunk_size
;
4735 bytes
-= chunk_size
;
4739 aio_task_pool_wait_all(aio
);
4741 ret
= aio_task_pool_status(aio
);
4749 static int coroutine_fn GRAPH_RDLOCK
4750 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
4757 BDRVQcow2State
*s
= bs
->opaque
;
4760 uint8_t *buf
, *out_buf
;
4761 int offset_in_cluster
= offset_into_cluster(s
, offset
);
4763 qcow2_parse_compressed_l2_entry(bs
, l2_entry
, &coffset
, &csize
);
4765 buf
= g_try_malloc(csize
);
4770 out_buf
= qemu_blockalign(bs
, s
->cluster_size
);
4772 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
4773 ret
= bdrv_co_pread(bs
->file
, coffset
, csize
, buf
, 0);
4778 if (qcow2_co_decompress(bs
, out_buf
, s
->cluster_size
, buf
, csize
) < 0) {
4783 qemu_iovec_from_buf(qiov
, qiov_offset
, out_buf
+ offset_in_cluster
, bytes
);
4786 qemu_vfree(out_buf
);
4792 static int make_completely_empty(BlockDriverState
*bs
)
4794 BDRVQcow2State
*s
= bs
->opaque
;
4795 Error
*local_err
= NULL
;
4796 int ret
, l1_clusters
;
4798 uint64_t *new_reftable
= NULL
;
4799 uint64_t rt_entry
, l1_size2
;
4802 uint64_t reftable_offset
;
4803 uint32_t reftable_clusters
;
4804 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
4806 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
4811 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
4816 /* Refcounts will be broken utterly */
4817 ret
= qcow2_mark_dirty(bs
);
4822 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4824 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ L1E_SIZE
);
4825 l1_size2
= (uint64_t)s
->l1_size
* L1E_SIZE
;
4827 /* After this call, neither the in-memory nor the on-disk refcount
4828 * information accurately describe the actual references */
4830 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
4831 l1_clusters
* s
->cluster_size
, 0);
4833 goto fail_broken_refcounts
;
4835 memset(s
->l1_table
, 0, l1_size2
);
4837 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
4839 /* Overwrite enough clusters at the beginning of the sectors to place
4840 * the refcount table, a refcount block and the L1 table in; this may
4841 * overwrite parts of the existing refcount and L1 table, which is not
4842 * an issue because the dirty flag is set, complete data loss is in fact
4843 * desired and partial data loss is consequently fine as well */
4844 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
4845 (2 + l1_clusters
) * s
->cluster_size
, 0);
4846 /* This call (even if it failed overall) may have overwritten on-disk
4847 * refcount structures; in that case, the in-memory refcount information
4848 * will probably differ from the on-disk information which makes the BDS
4851 goto fail_broken_refcounts
;
4854 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4855 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
4857 /* "Create" an empty reftable (one cluster) directly after the image
4858 * header and an empty L1 table three clusters after the image header;
4859 * the cluster between those two will be used as the first refblock */
4860 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
4861 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
4862 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
4863 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
4864 sizeof(l1_ofs_rt_ofs_cls
), &l1_ofs_rt_ofs_cls
, 0);
4866 goto fail_broken_refcounts
;
4869 s
->l1_table_offset
= 3 * s
->cluster_size
;
4871 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ REFTABLE_ENTRY_SIZE
);
4872 if (!new_reftable
) {
4874 goto fail_broken_refcounts
;
4877 s
->refcount_table_offset
= s
->cluster_size
;
4878 s
->refcount_table_size
= s
->cluster_size
/ REFTABLE_ENTRY_SIZE
;
4879 s
->max_refcount_table_index
= 0;
4881 g_free(s
->refcount_table
);
4882 s
->refcount_table
= new_reftable
;
4883 new_reftable
= NULL
;
4885 /* Now the in-memory refcount information again corresponds to the on-disk
4886 * information (reftable is empty and no refblocks (the refblock cache is
4887 * empty)); however, this means some clusters (e.g. the image header) are
4888 * referenced, but not refcounted, but the normal qcow2 code assumes that
4889 * the in-memory information is always correct */
4891 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
4893 /* Enter the first refblock into the reftable */
4894 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
4895 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
, sizeof(rt_entry
),
4898 goto fail_broken_refcounts
;
4900 s
->refcount_table
[0] = 2 * s
->cluster_size
;
4902 s
->free_cluster_index
= 0;
4903 assert(3 + l1_clusters
<= s
->refcount_block_size
);
4904 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
4907 goto fail_broken_refcounts
;
4908 } else if (offset
> 0) {
4909 error_report("First cluster in emptied image is in use");
4913 /* Now finally the in-memory information corresponds to the on-disk
4914 * structures and is correct */
4915 ret
= qcow2_mark_clean(bs
);
4920 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
, false,
4921 PREALLOC_MODE_OFF
, 0, &local_err
);
4923 error_report_err(local_err
);
4929 fail_broken_refcounts
:
4930 /* The BDS is unusable at this point. If we wanted to make it usable, we
4931 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4932 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4933 * again. However, because the functions which could have caused this error
4934 * path to be taken are used by those functions as well, it's very likely
4935 * that that sequence will fail as well. Therefore, just eject the BDS. */
4939 g_free(new_reftable
);
4943 static int qcow2_make_empty(BlockDriverState
*bs
)
4945 BDRVQcow2State
*s
= bs
->opaque
;
4946 uint64_t offset
, end_offset
;
4947 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
4948 int l1_clusters
, ret
= 0;
4950 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ L1E_SIZE
);
4952 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
4953 3 + l1_clusters
<= s
->refcount_block_size
&&
4954 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
&&
4955 !has_data_file(bs
)) {
4956 /* The following function only works for qcow2 v3 images (it
4957 * requires the dirty flag) and only as long as there are no
4958 * features that reserve extra clusters (such as snapshots,
4959 * LUKS header, or persistent bitmaps), because it completely
4960 * empties the image. Furthermore, the L1 table and three
4961 * additional clusters (image header, refcount table, one
4962 * refcount block) have to fit inside one refcount block. It
4963 * only resets the image file, i.e. does not work with an
4964 * external data file. */
4965 return make_completely_empty(bs
);
4968 /* This fallback code simply discards every active cluster; this is slow,
4969 * but works in all cases */
4970 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4971 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
4972 /* As this function is generally used after committing an external
4973 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4974 * default action for this kind of discard is to pass the discard,
4975 * which will ideally result in an actually smaller image file, as
4976 * is probably desired. */
4977 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
4978 QCOW2_DISCARD_SNAPSHOT
, true);
4987 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
4989 BDRVQcow2State
*s
= bs
->opaque
;
4992 qemu_co_mutex_lock(&s
->lock
);
4993 ret
= qcow2_write_caches(bs
);
4994 qemu_co_mutex_unlock(&s
->lock
);
4999 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
5002 Error
*local_err
= NULL
;
5003 BlockMeasureInfo
*info
;
5004 uint64_t required
= 0; /* bytes that contribute to required size */
5005 uint64_t virtual_size
; /* disk size as seen by guest */
5006 uint64_t refcount_bits
;
5008 uint64_t luks_payload_size
= 0;
5009 size_t cluster_size
;
5012 PreallocMode prealloc
;
5013 bool has_backing_file
;
5018 /* Parse image creation options */
5019 extended_l2
= qemu_opt_get_bool_del(opts
, BLOCK_OPT_EXTL2
, false);
5021 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, extended_l2
,
5027 version
= qcow2_opt_get_version_del(opts
, &local_err
);
5032 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
5037 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
5038 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
5039 PREALLOC_MODE_OFF
, &local_err
);
5045 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
5046 has_backing_file
= !!optstr
;
5049 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
5050 has_luks
= optstr
&& strcmp(optstr
, "luks") == 0;
5054 g_autoptr(QCryptoBlockCreateOptions
) create_opts
= NULL
;
5055 QDict
*cryptoopts
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
5058 create_opts
= block_crypto_create_opts_init(cryptoopts
, errp
);
5059 qobject_unref(cryptoopts
);
5064 if (!qcrypto_block_calculate_payload_offset(create_opts
,
5071 luks_payload_size
= ROUND_UP(headerlen
, cluster_size
);
5074 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
5075 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
5077 /* Check that virtual disk size is valid */
5078 l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
5079 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
5080 cluster_size
/ l2e_size
);
5081 if (l2_tables
* L1E_SIZE
> QCOW_MAX_L1_SIZE
) {
5082 error_setg(&local_err
, "The image size is too large "
5083 "(try using a larger cluster size)");
5087 /* Account for input image */
5089 int64_t ssize
= bdrv_getlength(in_bs
);
5091 error_setg_errno(&local_err
, -ssize
,
5092 "Unable to get image virtual_size");
5096 virtual_size
= ROUND_UP(ssize
, cluster_size
);
5098 if (has_backing_file
) {
5099 /* We don't how much of the backing chain is shared by the input
5100 * image and the new image file. In the worst case the new image's
5101 * backing file has nothing in common with the input image. Be
5102 * conservative and assume all clusters need to be written.
5104 required
= virtual_size
;
5109 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
5112 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
5113 ssize
- offset
, &pnum
, NULL
,
5116 error_setg_errno(&local_err
, -ret
,
5117 "Unable to get block status");
5121 if (ret
& BDRV_BLOCK_ZERO
) {
5122 /* Skip zero regions (safe with no backing file) */
5123 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
5124 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
5125 /* Extend pnum to end of cluster for next iteration */
5126 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
5128 /* Count clusters we've seen */
5129 required
+= offset
% cluster_size
+ pnum
;
5135 /* Take into account preallocation. Nothing special is needed for
5136 * PREALLOC_MODE_METADATA since metadata is always counted.
5138 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
5139 required
= virtual_size
;
5142 info
= g_new0(BlockMeasureInfo
, 1);
5143 info
->fully_allocated
= luks_payload_size
+
5144 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
5145 ctz32(refcount_bits
), extended_l2
);
5148 * Remove data clusters that are not required. This overestimates the
5149 * required size because metadata needed for the fully allocated file is
5150 * still counted. Show bitmaps only if both source and destination
5151 * would support them.
5153 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
5154 info
->has_bitmaps
= version
>= 3 && in_bs
&&
5155 bdrv_supports_persistent_dirty_bitmap(in_bs
);
5156 if (info
->has_bitmaps
) {
5157 info
->bitmaps
= qcow2_get_persistent_dirty_bitmap_size(in_bs
,
5163 error_propagate(errp
, local_err
);
5167 static int coroutine_fn
5168 qcow2_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5170 BDRVQcow2State
*s
= bs
->opaque
;
5171 bdi
->cluster_size
= s
->cluster_size
;
5172 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
5173 bdi
->is_dirty
= s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
;
5177 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
,
5180 BDRVQcow2State
*s
= bs
->opaque
;
5181 ImageInfoSpecific
*spec_info
;
5182 QCryptoBlockInfo
*encrypt_info
= NULL
;
5184 if (s
->crypto
!= NULL
) {
5185 encrypt_info
= qcrypto_block_get_info(s
->crypto
, errp
);
5186 if (!encrypt_info
) {
5191 spec_info
= g_new(ImageInfoSpecific
, 1);
5192 *spec_info
= (ImageInfoSpecific
){
5193 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
5194 .u
.qcow2
.data
= g_new0(ImageInfoSpecificQCow2
, 1),
5196 if (s
->qcow_version
== 2) {
5197 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5198 .compat
= g_strdup("0.10"),
5199 .refcount_bits
= s
->refcount_bits
,
5201 } else if (s
->qcow_version
== 3) {
5202 Qcow2BitmapInfoList
*bitmaps
;
5203 if (!qcow2_get_bitmap_info_list(bs
, &bitmaps
, errp
)) {
5204 qapi_free_ImageInfoSpecific(spec_info
);
5205 qapi_free_QCryptoBlockInfo(encrypt_info
);
5208 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5209 .compat
= g_strdup("1.1"),
5210 .lazy_refcounts
= s
->compatible_features
&
5211 QCOW2_COMPAT_LAZY_REFCOUNTS
,
5212 .has_lazy_refcounts
= true,
5213 .corrupt
= s
->incompatible_features
&
5214 QCOW2_INCOMPAT_CORRUPT
,
5215 .has_corrupt
= true,
5216 .has_extended_l2
= true,
5217 .extended_l2
= has_subclusters(s
),
5218 .refcount_bits
= s
->refcount_bits
,
5219 .has_bitmaps
= !!bitmaps
,
5221 .data_file
= g_strdup(s
->image_data_file
),
5222 .has_data_file_raw
= has_data_file(bs
),
5223 .data_file_raw
= data_file_is_raw(bs
),
5224 .compression_type
= s
->compression_type
,
5227 /* if this assertion fails, this probably means a new version was
5228 * added without having it covered here */
5233 ImageInfoSpecificQCow2Encryption
*qencrypt
=
5234 g_new(ImageInfoSpecificQCow2Encryption
, 1);
5235 switch (encrypt_info
->format
) {
5236 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
5237 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
5239 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
5240 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
5241 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
5246 /* Since we did shallow copy above, erase any pointers
5247 * in the original info */
5248 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
5249 qapi_free_QCryptoBlockInfo(encrypt_info
);
5251 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
5257 static int qcow2_has_zero_init(BlockDriverState
*bs
)
5259 BDRVQcow2State
*s
= bs
->opaque
;
5262 if (qemu_in_coroutine()) {
5263 qemu_co_mutex_lock(&s
->lock
);
5266 * Check preallocation status: Preallocated images have all L2
5267 * tables allocated, nonpreallocated images have none. It is
5268 * therefore enough to check the first one.
5270 preallocated
= s
->l1_size
> 0 && s
->l1_table
[0] != 0;
5271 if (qemu_in_coroutine()) {
5272 qemu_co_mutex_unlock(&s
->lock
);
5275 if (!preallocated
) {
5277 } else if (bs
->encrypted
) {
5280 return bdrv_has_zero_init(s
->data_file
->bs
);
5285 * Check the request to vmstate. On success return
5286 * qcow2_vm_state_offset(bs) + @pos
5288 static int64_t qcow2_check_vmstate_request(BlockDriverState
*bs
,
5289 QEMUIOVector
*qiov
, int64_t pos
)
5291 BDRVQcow2State
*s
= bs
->opaque
;
5292 int64_t vmstate_offset
= qcow2_vm_state_offset(s
);
5295 /* Incoming requests must be OK */
5296 bdrv_check_qiov_request(pos
, qiov
->size
, qiov
, 0, &error_abort
);
5298 if (INT64_MAX
- pos
< vmstate_offset
) {
5302 pos
+= vmstate_offset
;
5303 ret
= bdrv_check_qiov_request(pos
, qiov
->size
, qiov
, 0, NULL
);
5311 static int coroutine_fn GRAPH_RDLOCK
5312 qcow2_co_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
5314 int64_t offset
= qcow2_check_vmstate_request(bs
, qiov
, pos
);
5319 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
5320 return bs
->drv
->bdrv_co_pwritev_part(bs
, offset
, qiov
->size
, qiov
, 0, 0);
5323 static int coroutine_fn GRAPH_RDLOCK
5324 qcow2_co_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
5326 int64_t offset
= qcow2_check_vmstate_request(bs
, qiov
, pos
);
5331 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
5332 return bs
->drv
->bdrv_co_preadv_part(bs
, offset
, qiov
->size
, qiov
, 0, 0);
5335 static int qcow2_has_compressed_clusters(BlockDriverState
*bs
)
5338 int64_t bytes
= bdrv_getlength(bs
);
5344 while (bytes
!= 0) {
5346 QCow2SubclusterType type
;
5347 unsigned int cur_bytes
= MIN(INT_MAX
, bytes
);
5348 uint64_t host_offset
;
5350 ret
= qcow2_get_host_offset(bs
, offset
, &cur_bytes
, &host_offset
,
5356 if (type
== QCOW2_SUBCLUSTER_COMPRESSED
) {
5360 offset
+= cur_bytes
;
5368 * Downgrades an image's version. To achieve this, any incompatible features
5369 * have to be removed.
5371 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
5372 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5375 BDRVQcow2State
*s
= bs
->opaque
;
5376 int current_version
= s
->qcow_version
;
5380 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5381 assert(target_version
< current_version
);
5383 /* There are no other versions (now) that you can downgrade to */
5384 assert(target_version
== 2);
5386 if (s
->refcount_order
!= 4) {
5387 error_setg(errp
, "compat=0.10 requires refcount_bits=16");
5391 if (has_data_file(bs
)) {
5392 error_setg(errp
, "Cannot downgrade an image with a data file");
5397 * If any internal snapshot has a different size than the current
5398 * image size, or VM state size that exceeds 32 bits, downgrading
5399 * is unsafe. Even though we would still use v3-compliant output
5400 * to preserve that data, other v2 programs might not realize
5401 * those optional fields are important.
5403 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5404 if (s
->snapshots
[i
].vm_state_size
> UINT32_MAX
||
5405 s
->snapshots
[i
].disk_size
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
5406 error_setg(errp
, "Internal snapshots prevent downgrade of image");
5411 /* clear incompatible features */
5412 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
5413 ret
= qcow2_mark_clean(bs
);
5415 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5420 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5421 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5422 * best thing to do anyway */
5424 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_COMPRESSION
) {
5425 error_setg(errp
, "Cannot downgrade an image with incompatible features "
5426 "0x%" PRIx64
" set",
5427 s
->incompatible_features
& ~QCOW2_INCOMPAT_COMPRESSION
);
5431 /* since we can ignore compatible features, we can set them to 0 as well */
5432 s
->compatible_features
= 0;
5433 /* if lazy refcounts have been used, they have already been fixed through
5434 * clearing the dirty flag */
5436 /* clearing autoclear features is trivial */
5437 s
->autoclear_features
= 0;
5439 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
5441 error_setg_errno(errp
, -ret
, "Failed to turn zero into data clusters");
5445 if (s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
) {
5446 ret
= qcow2_has_compressed_clusters(bs
);
5448 error_setg(errp
, "Failed to check block status");
5452 error_setg(errp
, "Cannot downgrade an image with zstd compression "
5453 "type and existing compressed clusters");
5457 * No compressed clusters for now, so just chose default zlib
5460 s
->incompatible_features
&= ~QCOW2_INCOMPAT_COMPRESSION
;
5461 s
->compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
5464 assert(s
->incompatible_features
== 0);
5466 s
->qcow_version
= target_version
;
5467 ret
= qcow2_update_header(bs
);
5469 s
->qcow_version
= current_version
;
5470 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5477 * Upgrades an image's version. While newer versions encompass all
5478 * features of older versions, some things may have to be presented
5481 static int qcow2_upgrade(BlockDriverState
*bs
, int target_version
,
5482 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5485 BDRVQcow2State
*s
= bs
->opaque
;
5486 bool need_snapshot_update
;
5487 int current_version
= s
->qcow_version
;
5491 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5492 assert(target_version
> current_version
);
5494 /* There are no other versions (yet) that you can upgrade to */
5495 assert(target_version
== 3);
5497 status_cb(bs
, 0, 2, cb_opaque
);
5500 * In v2, snapshots do not need to have extra data. v3 requires
5501 * the 64-bit VM state size and the virtual disk size to be
5503 * qcow2_write_snapshots() will always write the list in the
5504 * v3-compliant format.
5506 need_snapshot_update
= false;
5507 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5508 if (s
->snapshots
[i
].extra_data_size
<
5509 sizeof_field(QCowSnapshotExtraData
, vm_state_size_large
) +
5510 sizeof_field(QCowSnapshotExtraData
, disk_size
))
5512 need_snapshot_update
= true;
5516 if (need_snapshot_update
) {
5517 ret
= qcow2_write_snapshots(bs
);
5519 error_setg_errno(errp
, -ret
, "Failed to update the snapshot table");
5523 status_cb(bs
, 1, 2, cb_opaque
);
5525 s
->qcow_version
= target_version
;
5526 ret
= qcow2_update_header(bs
);
5528 s
->qcow_version
= current_version
;
5529 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5532 status_cb(bs
, 2, 2, cb_opaque
);
5537 typedef enum Qcow2AmendOperation
{
5538 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5539 * statically initialized to so that the helper CB can discern the first
5540 * invocation from an operation change */
5541 QCOW2_NO_OPERATION
= 0,
5544 QCOW2_UPDATING_ENCRYPTION
,
5545 QCOW2_CHANGING_REFCOUNT_ORDER
,
5547 } Qcow2AmendOperation
;
5549 typedef struct Qcow2AmendHelperCBInfo
{
5550 /* The code coordinating the amend operations should only modify
5551 * these four fields; the rest will be managed by the CB */
5552 BlockDriverAmendStatusCB
*original_status_cb
;
5553 void *original_cb_opaque
;
5555 Qcow2AmendOperation current_operation
;
5557 /* Total number of operations to perform (only set once) */
5558 int total_operations
;
5560 /* The following fields are managed by the CB */
5562 /* Number of operations completed */
5563 int operations_completed
;
5565 /* Cumulative offset of all completed operations */
5566 int64_t offset_completed
;
5568 Qcow2AmendOperation last_operation
;
5569 int64_t last_work_size
;
5570 } Qcow2AmendHelperCBInfo
;
5572 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
5573 int64_t operation_offset
,
5574 int64_t operation_work_size
, void *opaque
)
5576 Qcow2AmendHelperCBInfo
*info
= opaque
;
5577 int64_t current_work_size
;
5578 int64_t projected_work_size
;
5580 if (info
->current_operation
!= info
->last_operation
) {
5581 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
5582 info
->offset_completed
+= info
->last_work_size
;
5583 info
->operations_completed
++;
5586 info
->last_operation
= info
->current_operation
;
5589 assert(info
->total_operations
> 0);
5590 assert(info
->operations_completed
< info
->total_operations
);
5592 info
->last_work_size
= operation_work_size
;
5594 current_work_size
= info
->offset_completed
+ operation_work_size
;
5596 /* current_work_size is the total work size for (operations_completed + 1)
5597 * operations (which includes this one), so multiply it by the number of
5598 * operations not covered and divide it by the number of operations
5599 * covered to get a projection for the operations not covered */
5600 projected_work_size
= current_work_size
* (info
->total_operations
-
5601 info
->operations_completed
- 1)
5602 / (info
->operations_completed
+ 1);
5604 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
5605 current_work_size
+ projected_work_size
,
5606 info
->original_cb_opaque
);
5609 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5610 BlockDriverAmendStatusCB
*status_cb
,
5615 BDRVQcow2State
*s
= bs
->opaque
;
5616 int old_version
= s
->qcow_version
, new_version
= old_version
;
5617 uint64_t new_size
= 0;
5618 const char *backing_file
= NULL
, *backing_format
= NULL
, *data_file
= NULL
;
5619 bool lazy_refcounts
= s
->use_lazy_refcounts
;
5620 bool data_file_raw
= data_file_is_raw(bs
);
5621 const char *compat
= NULL
;
5622 int refcount_bits
= s
->refcount_bits
;
5624 QemuOptDesc
*desc
= opts
->list
->desc
;
5625 Qcow2AmendHelperCBInfo helper_cb_info
;
5626 bool encryption_update
= false;
5628 while (desc
&& desc
->name
) {
5629 if (!qemu_opt_find(opts
, desc
->name
)) {
5630 /* only change explicitly defined options */
5635 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
5636 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
5638 /* preserve default */
5639 } else if (!strcmp(compat
, "0.10") || !strcmp(compat
, "v2")) {
5641 } else if (!strcmp(compat
, "1.1") || !strcmp(compat
, "v3")) {
5644 error_setg(errp
, "Unknown compatibility level %s", compat
);
5647 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
5648 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
5649 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
5650 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5651 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
5652 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5653 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
5656 "Can't amend encryption options - encryption not present");
5659 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5661 "Only LUKS encryption options can be amended");
5664 encryption_update
= true;
5665 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
5666 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
5668 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
5669 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
5672 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
5673 !is_power_of_2(refcount_bits
))
5675 error_setg(errp
, "Refcount width must be a power of two and "
5676 "may not exceed 64 bits");
5679 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE
)) {
5680 data_file
= qemu_opt_get(opts
, BLOCK_OPT_DATA_FILE
);
5681 if (data_file
&& !has_data_file(bs
)) {
5682 error_setg(errp
, "data-file can only be set for images that "
5683 "use an external data file");
5686 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE_RAW
)) {
5687 data_file_raw
= qemu_opt_get_bool(opts
, BLOCK_OPT_DATA_FILE_RAW
,
5689 if (data_file_raw
&& !data_file_is_raw(bs
)) {
5690 error_setg(errp
, "data-file-raw cannot be set on existing "
5695 /* if this point is reached, this probably means a new option was
5696 * added without having it covered here */
5703 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
5704 .original_status_cb
= status_cb
,
5705 .original_cb_opaque
= cb_opaque
,
5706 .total_operations
= (new_version
!= old_version
)
5707 + (s
->refcount_bits
!= refcount_bits
) +
5708 (encryption_update
== true)
5711 /* Upgrade first (some features may require compat=1.1) */
5712 if (new_version
> old_version
) {
5713 helper_cb_info
.current_operation
= QCOW2_UPGRADING
;
5714 ret
= qcow2_upgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5715 &helper_cb_info
, errp
);
5721 if (encryption_update
) {
5722 QDict
*amend_opts_dict
;
5723 QCryptoBlockAmendOptions
*amend_opts
;
5725 helper_cb_info
.current_operation
= QCOW2_UPDATING_ENCRYPTION
;
5726 amend_opts_dict
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
5727 if (!amend_opts_dict
) {
5730 amend_opts
= block_crypto_amend_opts_init(amend_opts_dict
, errp
);
5731 qobject_unref(amend_opts_dict
);
5735 ret
= qcrypto_block_amend_options(s
->crypto
,
5736 qcow2_crypto_hdr_read_func
,
5737 qcow2_crypto_hdr_write_func
,
5742 qapi_free_QCryptoBlockAmendOptions(amend_opts
);
5748 if (s
->refcount_bits
!= refcount_bits
) {
5749 int refcount_order
= ctz32(refcount_bits
);
5751 if (new_version
< 3 && refcount_bits
!= 16) {
5752 error_setg(errp
, "Refcount widths other than 16 bits require "
5753 "compatibility level 1.1 or above (use compat=1.1 or "
5758 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
5759 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
5760 &qcow2_amend_helper_cb
,
5761 &helper_cb_info
, errp
);
5767 /* data-file-raw blocks backing files, so clear it first if requested */
5768 if (data_file_raw
) {
5769 s
->autoclear_features
|= QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5771 s
->autoclear_features
&= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5775 g_free(s
->image_data_file
);
5776 s
->image_data_file
= *data_file
? g_strdup(data_file
) : NULL
;
5779 ret
= qcow2_update_header(bs
);
5781 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5785 if (backing_file
|| backing_format
) {
5786 if (g_strcmp0(backing_file
, s
->image_backing_file
) ||
5787 g_strcmp0(backing_format
, s
->image_backing_format
)) {
5788 error_setg(errp
, "Cannot amend the backing file");
5789 error_append_hint(errp
,
5790 "You can use 'qemu-img rebase' instead.\n");
5795 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
5796 if (lazy_refcounts
) {
5797 if (new_version
< 3) {
5798 error_setg(errp
, "Lazy refcounts only supported with "
5799 "compatibility level 1.1 and above (use compat=1.1 "
5803 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5804 ret
= qcow2_update_header(bs
);
5806 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5807 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5810 s
->use_lazy_refcounts
= true;
5812 /* make image clean first */
5813 ret
= qcow2_mark_clean(bs
);
5815 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5818 /* now disallow lazy refcounts */
5819 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5820 ret
= qcow2_update_header(bs
);
5822 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5823 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5826 s
->use_lazy_refcounts
= false;
5831 BlockBackend
*blk
= blk_new_with_bs(bs
, BLK_PERM_RESIZE
, BLK_PERM_ALL
,
5838 * Amending image options should ensure that the image has
5839 * exactly the given new values, so pass exact=true here.
5841 ret
= blk_truncate(blk
, new_size
, true, PREALLOC_MODE_OFF
, 0, errp
);
5848 /* Downgrade last (so unsupported features can be removed before) */
5849 if (new_version
< old_version
) {
5850 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
5851 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5852 &helper_cb_info
, errp
);
5861 static int coroutine_fn
qcow2_co_amend(BlockDriverState
*bs
,
5862 BlockdevAmendOptions
*opts
,
5866 BlockdevAmendOptionsQcow2
*qopts
= &opts
->u
.qcow2
;
5867 BDRVQcow2State
*s
= bs
->opaque
;
5870 if (qopts
->encrypt
) {
5872 error_setg(errp
, "image is not encrypted, can't amend");
5876 if (qopts
->encrypt
->format
!= Q_CRYPTO_BLOCK_FORMAT_LUKS
) {
5878 "Amend can't be used to change the qcow2 encryption format");
5882 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5884 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5888 ret
= qcrypto_block_amend_options(s
->crypto
,
5889 qcow2_crypto_hdr_read_func
,
5890 qcow2_crypto_hdr_write_func
,
5900 * If offset or size are negative, respectively, they will not be included in
5901 * the BLOCK_IMAGE_CORRUPTED event emitted.
5902 * fatal will be ignored for read-only BDS; corruptions found there will always
5903 * be considered non-fatal.
5905 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
5906 int64_t size
, const char *message_format
, ...)
5908 BDRVQcow2State
*s
= bs
->opaque
;
5909 const char *node_name
;
5913 fatal
= fatal
&& bdrv_is_writable(bs
);
5915 if (s
->signaled_corruption
&&
5916 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
5921 va_start(ap
, message_format
);
5922 message
= g_strdup_vprintf(message_format
, ap
);
5926 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
5927 "corruption events will be suppressed\n", message
);
5929 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
5930 "corruption events will be suppressed\n", message
);
5933 node_name
= bdrv_get_node_name(bs
);
5934 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
5935 *node_name
? node_name
: NULL
,
5936 message
, offset
>= 0, offset
,
5942 qcow2_mark_corrupt(bs
);
5943 bs
->drv
= NULL
; /* make BDS unusable */
5946 s
->signaled_corruption
= true;
5949 #define QCOW_COMMON_OPTIONS \
5951 .name = BLOCK_OPT_SIZE, \
5952 .type = QEMU_OPT_SIZE, \
5953 .help = "Virtual disk size" \
5956 .name = BLOCK_OPT_COMPAT_LEVEL, \
5957 .type = QEMU_OPT_STRING, \
5958 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
5961 .name = BLOCK_OPT_BACKING_FILE, \
5962 .type = QEMU_OPT_STRING, \
5963 .help = "File name of a base image" \
5966 .name = BLOCK_OPT_BACKING_FMT, \
5967 .type = QEMU_OPT_STRING, \
5968 .help = "Image format of the base image" \
5971 .name = BLOCK_OPT_DATA_FILE, \
5972 .type = QEMU_OPT_STRING, \
5973 .help = "File name of an external data file" \
5976 .name = BLOCK_OPT_DATA_FILE_RAW, \
5977 .type = QEMU_OPT_BOOL, \
5978 .help = "The external data file must stay valid " \
5982 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
5983 .type = QEMU_OPT_BOOL, \
5984 .help = "Postpone refcount updates", \
5985 .def_value_str = "off" \
5988 .name = BLOCK_OPT_REFCOUNT_BITS, \
5989 .type = QEMU_OPT_NUMBER, \
5990 .help = "Width of a reference count entry in bits", \
5991 .def_value_str = "16" \
5994 static QemuOptsList qcow2_create_opts
= {
5995 .name
= "qcow2-create-opts",
5996 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
5999 .name
= BLOCK_OPT_ENCRYPT
, \
6000 .type
= QEMU_OPT_BOOL
, \
6001 .help
= "Encrypt the image with format 'aes'. (Deprecated " \
6002 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)", \
6005 .name
= BLOCK_OPT_ENCRYPT_FORMAT
, \
6006 .type
= QEMU_OPT_STRING
, \
6007 .help
= "Encrypt the image, format choices: 'aes', 'luks'", \
6009 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
6010 "ID of secret providing qcow AES key or LUKS passphrase"), \
6011 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
6012 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
6013 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
6014 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
6015 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
6016 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
6018 .name
= BLOCK_OPT_CLUSTER_SIZE
, \
6019 .type
= QEMU_OPT_SIZE
, \
6020 .help
= "qcow2 cluster size", \
6021 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
) \
6024 .name
= BLOCK_OPT_EXTL2
, \
6025 .type
= QEMU_OPT_BOOL
, \
6026 .help
= "Extended L2 tables", \
6027 .def_value_str
= "off" \
6030 .name
= BLOCK_OPT_PREALLOC
, \
6031 .type
= QEMU_OPT_STRING
, \
6032 .help
= "Preallocation mode (allowed values: off, " \
6033 "metadata, falloc, full)" \
6036 .name
= BLOCK_OPT_COMPRESSION_TYPE
, \
6037 .type
= QEMU_OPT_STRING
, \
6038 .help
= "Compression method used for image cluster " \
6040 .def_value_str
= "zlib" \
6042 QCOW_COMMON_OPTIONS
,
6043 { /* end of list */ }
6047 static QemuOptsList qcow2_amend_opts
= {
6048 .name
= "qcow2-amend-opts",
6049 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts
.head
),
6051 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
6052 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
6053 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
6054 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
6055 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
6056 QCOW_COMMON_OPTIONS
,
6057 { /* end of list */ }
6061 static const char *const qcow2_strong_runtime_opts
[] = {
6062 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET
,
6067 BlockDriver bdrv_qcow2
= {
6068 .format_name
= "qcow2",
6069 .instance_size
= sizeof(BDRVQcow2State
),
6070 .bdrv_probe
= qcow2_probe
,
6071 .bdrv_open
= qcow2_open
,
6072 .bdrv_close
= qcow2_close
,
6073 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
6074 .bdrv_reopen_commit
= qcow2_reopen_commit
,
6075 .bdrv_reopen_commit_post
= qcow2_reopen_commit_post
,
6076 .bdrv_reopen_abort
= qcow2_reopen_abort
,
6077 .bdrv_join_options
= qcow2_join_options
,
6078 .bdrv_child_perm
= bdrv_default_perms
,
6079 .bdrv_co_create_opts
= qcow2_co_create_opts
,
6080 .bdrv_co_create
= qcow2_co_create
,
6081 .bdrv_has_zero_init
= qcow2_has_zero_init
,
6082 .bdrv_co_block_status
= qcow2_co_block_status
,
6084 .bdrv_co_preadv_part
= qcow2_co_preadv_part
,
6085 .bdrv_co_pwritev_part
= qcow2_co_pwritev_part
,
6086 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
6088 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
6089 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
6090 .bdrv_co_copy_range_from
= qcow2_co_copy_range_from
,
6091 .bdrv_co_copy_range_to
= qcow2_co_copy_range_to
,
6092 .bdrv_co_truncate
= qcow2_co_truncate
,
6093 .bdrv_co_pwritev_compressed_part
= qcow2_co_pwritev_compressed_part
,
6094 .bdrv_make_empty
= qcow2_make_empty
,
6096 .bdrv_snapshot_create
= qcow2_snapshot_create
,
6097 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
6098 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
6099 .bdrv_snapshot_list
= qcow2_snapshot_list
,
6100 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
6101 .bdrv_measure
= qcow2_measure
,
6102 .bdrv_co_get_info
= qcow2_co_get_info
,
6103 .bdrv_get_specific_info
= qcow2_get_specific_info
,
6105 .bdrv_co_save_vmstate
= qcow2_co_save_vmstate
,
6106 .bdrv_co_load_vmstate
= qcow2_co_load_vmstate
,
6109 .supports_backing
= true,
6110 .bdrv_change_backing_file
= qcow2_change_backing_file
,
6112 .bdrv_refresh_limits
= qcow2_refresh_limits
,
6113 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
6114 .bdrv_inactivate
= qcow2_inactivate
,
6116 .create_opts
= &qcow2_create_opts
,
6117 .amend_opts
= &qcow2_amend_opts
,
6118 .strong_runtime_opts
= qcow2_strong_runtime_opts
,
6119 .mutable_opts
= mutable_opts
,
6120 .bdrv_co_check
= qcow2_co_check
,
6121 .bdrv_amend_options
= qcow2_amend_options
,
6122 .bdrv_co_amend
= qcow2_co_amend
,
6124 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
6125 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
6127 .bdrv_supports_persistent_dirty_bitmap
=
6128 qcow2_supports_persistent_dirty_bitmap
,
6129 .bdrv_co_can_store_new_dirty_bitmap
= qcow2_co_can_store_new_dirty_bitmap
,
6130 .bdrv_co_remove_persistent_dirty_bitmap
=
6131 qcow2_co_remove_persistent_dirty_bitmap
,
6134 static void bdrv_qcow2_init(void)
6136 bdrv_register(&bdrv_qcow2
);
6139 block_init(bdrv_qcow2_init
);