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 static int coroutine_fn
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
,
685 QCOW2_OPT_DISCARD_NO_UNREF
,
687 QCOW2_OPT_OVERLAP_TEMPLATE
,
688 QCOW2_OPT_OVERLAP_MAIN_HEADER
,
689 QCOW2_OPT_OVERLAP_ACTIVE_L1
,
690 QCOW2_OPT_OVERLAP_ACTIVE_L2
,
691 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
692 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
693 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
694 QCOW2_OPT_OVERLAP_INACTIVE_L1
,
695 QCOW2_OPT_OVERLAP_INACTIVE_L2
,
696 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
697 QCOW2_OPT_CACHE_SIZE
,
698 QCOW2_OPT_L2_CACHE_SIZE
,
699 QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
700 QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
701 QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
705 static QemuOptsList qcow2_runtime_opts
= {
707 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
710 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
711 .type
= QEMU_OPT_BOOL
,
712 .help
= "Postpone refcount updates",
715 .name
= QCOW2_OPT_DISCARD_REQUEST
,
716 .type
= QEMU_OPT_BOOL
,
717 .help
= "Pass guest discard requests to the layer below",
720 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
721 .type
= QEMU_OPT_BOOL
,
722 .help
= "Generate discard requests when snapshot related space "
726 .name
= QCOW2_OPT_DISCARD_OTHER
,
727 .type
= QEMU_OPT_BOOL
,
728 .help
= "Generate discard requests when other clusters are freed",
731 .name
= QCOW2_OPT_DISCARD_NO_UNREF
,
732 .type
= QEMU_OPT_BOOL
,
733 .help
= "Do not unreference discarded clusters",
736 .name
= QCOW2_OPT_OVERLAP
,
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_TEMPLATE
,
743 .type
= QEMU_OPT_STRING
,
744 .help
= "Selects which overlap checks to perform from a range of "
745 "templates (none, constant, cached, all)",
748 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
749 .type
= QEMU_OPT_BOOL
,
750 .help
= "Check for unintended writes into the main qcow2 header",
753 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
754 .type
= QEMU_OPT_BOOL
,
755 .help
= "Check for unintended writes into the active L1 table",
758 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
759 .type
= QEMU_OPT_BOOL
,
760 .help
= "Check for unintended writes into an active L2 table",
763 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
764 .type
= QEMU_OPT_BOOL
,
765 .help
= "Check for unintended writes into the refcount table",
768 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
769 .type
= QEMU_OPT_BOOL
,
770 .help
= "Check for unintended writes into a refcount block",
773 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
774 .type
= QEMU_OPT_BOOL
,
775 .help
= "Check for unintended writes into the snapshot table",
778 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
779 .type
= QEMU_OPT_BOOL
,
780 .help
= "Check for unintended writes into an inactive L1 table",
783 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
784 .type
= QEMU_OPT_BOOL
,
785 .help
= "Check for unintended writes into an inactive L2 table",
788 .name
= QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
789 .type
= QEMU_OPT_BOOL
,
790 .help
= "Check for unintended writes into the bitmap directory",
793 .name
= QCOW2_OPT_CACHE_SIZE
,
794 .type
= QEMU_OPT_SIZE
,
795 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
799 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
800 .type
= QEMU_OPT_SIZE
,
801 .help
= "Maximum L2 table cache size",
804 .name
= QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
805 .type
= QEMU_OPT_SIZE
,
806 .help
= "Size of each entry in the L2 cache",
809 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
810 .type
= QEMU_OPT_SIZE
,
811 .help
= "Maximum refcount block cache size",
814 .name
= QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
815 .type
= QEMU_OPT_NUMBER
,
816 .help
= "Clean unused cache entries after this time (in seconds)",
818 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
819 "ID of secret providing qcow2 AES key or LUKS passphrase"),
820 { /* end of list */ }
824 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
825 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
826 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
827 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
828 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
829 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
830 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
831 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
832 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
833 [QCOW2_OL_BITMAP_DIRECTORY_BITNR
] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
836 static void cache_clean_timer_cb(void *opaque
)
838 BlockDriverState
*bs
= opaque
;
839 BDRVQcow2State
*s
= bs
->opaque
;
840 qcow2_cache_clean_unused(s
->l2_table_cache
);
841 qcow2_cache_clean_unused(s
->refcount_block_cache
);
842 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
843 (int64_t) s
->cache_clean_interval
* 1000);
846 static void cache_clean_timer_init(BlockDriverState
*bs
, AioContext
*context
)
848 BDRVQcow2State
*s
= bs
->opaque
;
849 if (s
->cache_clean_interval
> 0) {
850 s
->cache_clean_timer
=
851 aio_timer_new_with_attrs(context
, QEMU_CLOCK_VIRTUAL
,
852 SCALE_MS
, QEMU_TIMER_ATTR_EXTERNAL
,
853 cache_clean_timer_cb
, bs
);
854 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
855 (int64_t) s
->cache_clean_interval
* 1000);
859 static void cache_clean_timer_del(BlockDriverState
*bs
)
861 BDRVQcow2State
*s
= bs
->opaque
;
862 if (s
->cache_clean_timer
) {
863 timer_free(s
->cache_clean_timer
);
864 s
->cache_clean_timer
= NULL
;
868 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
870 cache_clean_timer_del(bs
);
873 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
874 AioContext
*new_context
)
876 cache_clean_timer_init(bs
, new_context
);
879 static bool read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
880 uint64_t *l2_cache_size
,
881 uint64_t *l2_cache_entry_size
,
882 uint64_t *refcount_cache_size
, Error
**errp
)
884 BDRVQcow2State
*s
= bs
->opaque
;
885 uint64_t combined_cache_size
, l2_cache_max_setting
;
886 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
887 bool l2_cache_entry_size_set
;
888 int min_refcount_cache
= MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
889 uint64_t virtual_disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
890 uint64_t max_l2_entries
= DIV_ROUND_UP(virtual_disk_size
, s
->cluster_size
);
891 /* An L2 table is always one cluster in size so the max cache size
892 * should be a multiple of the cluster size. */
893 uint64_t max_l2_cache
= ROUND_UP(max_l2_entries
* l2_entry_size(s
),
896 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
897 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
898 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
899 l2_cache_entry_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
);
901 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
902 l2_cache_max_setting
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
,
903 DEFAULT_L2_CACHE_MAX_SIZE
);
904 *refcount_cache_size
= qemu_opt_get_size(opts
,
905 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
907 *l2_cache_entry_size
= qemu_opt_get_size(
908 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
910 *l2_cache_size
= MIN(max_l2_cache
, l2_cache_max_setting
);
912 if (combined_cache_size_set
) {
913 if (l2_cache_size_set
&& refcount_cache_size_set
) {
914 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
915 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
918 } else if (l2_cache_size_set
&&
919 (l2_cache_max_setting
> combined_cache_size
)) {
920 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
921 QCOW2_OPT_CACHE_SIZE
);
923 } else if (*refcount_cache_size
> combined_cache_size
) {
924 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
925 QCOW2_OPT_CACHE_SIZE
);
929 if (l2_cache_size_set
) {
930 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
931 } else if (refcount_cache_size_set
) {
932 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
934 /* Assign as much memory as possible to the L2 cache, and
935 * use the remainder for the refcount cache */
936 if (combined_cache_size
>= max_l2_cache
+ min_refcount_cache
) {
937 *l2_cache_size
= max_l2_cache
;
938 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
940 *refcount_cache_size
=
941 MIN(combined_cache_size
, min_refcount_cache
);
942 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
948 * If the L2 cache is not enough to cover the whole disk then
949 * default to 4KB entries. Smaller entries reduce the cost of
950 * loads and evictions and increase I/O performance.
952 if (*l2_cache_size
< max_l2_cache
&& !l2_cache_entry_size_set
) {
953 *l2_cache_entry_size
= MIN(s
->cluster_size
, 4096);
956 /* l2_cache_size and refcount_cache_size are ensured to have at least
957 * their minimum values in qcow2_update_options_prepare() */
959 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
960 *l2_cache_entry_size
> s
->cluster_size
||
961 !is_power_of_2(*l2_cache_entry_size
)) {
962 error_setg(errp
, "L2 cache entry size must be a power of two "
963 "between %d and the cluster size (%d)",
964 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
971 typedef struct Qcow2ReopenState
{
972 Qcow2Cache
*l2_table_cache
;
973 Qcow2Cache
*refcount_block_cache
;
974 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
975 bool use_lazy_refcounts
;
977 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
978 bool discard_no_unref
;
979 uint64_t cache_clean_interval
;
980 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
983 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
985 QDict
*options
, int flags
,
988 BDRVQcow2State
*s
= bs
->opaque
;
989 QemuOpts
*opts
= NULL
;
990 const char *opt_overlap_check
, *opt_overlap_check_template
;
991 int overlap_check_template
= 0;
992 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
994 const char *encryptfmt
;
995 QDict
*encryptopts
= NULL
;
998 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
999 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
1001 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
1002 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1007 /* get L2 table/refcount block cache size from command line options */
1008 if (!read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
1009 &refcount_cache_size
, errp
)) {
1014 l2_cache_size
/= l2_cache_entry_size
;
1015 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
1016 l2_cache_size
= MIN_L2_CACHE_SIZE
;
1018 if (l2_cache_size
> INT_MAX
) {
1019 error_setg(errp
, "L2 cache size too big");
1024 refcount_cache_size
/= s
->cluster_size
;
1025 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
1026 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
1028 if (refcount_cache_size
> INT_MAX
) {
1029 error_setg(errp
, "Refcount cache size too big");
1034 /* alloc new L2 table/refcount block cache, flush old one */
1035 if (s
->l2_table_cache
) {
1036 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1038 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
1043 if (s
->refcount_block_cache
) {
1044 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1046 error_setg_errno(errp
, -ret
,
1047 "Failed to flush the refcount block cache");
1052 r
->l2_slice_size
= l2_cache_entry_size
/ l2_entry_size(s
);
1053 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
1054 l2_cache_entry_size
);
1055 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
1057 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
1058 error_setg(errp
, "Could not allocate metadata caches");
1063 /* New interval for cache cleanup timer */
1064 r
->cache_clean_interval
=
1065 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
1066 DEFAULT_CACHE_CLEAN_INTERVAL
);
1067 #ifndef CONFIG_LINUX
1068 if (r
->cache_clean_interval
!= 0) {
1069 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1070 " not supported on this host");
1075 if (r
->cache_clean_interval
> UINT_MAX
) {
1076 error_setg(errp
, "Cache clean interval too big");
1081 /* lazy-refcounts; flush if going from enabled to disabled */
1082 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
1083 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
1084 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
1085 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
1086 "qemu 1.1 compatibility level");
1091 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
1092 ret
= qcow2_mark_clean(bs
);
1094 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
1099 /* Overlap check options */
1100 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
1101 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1102 if (opt_overlap_check_template
&& opt_overlap_check
&&
1103 strcmp(opt_overlap_check_template
, opt_overlap_check
))
1105 error_setg(errp
, "Conflicting values for qcow2 options '"
1106 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1107 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
1111 if (!opt_overlap_check
) {
1112 opt_overlap_check
= opt_overlap_check_template
?: "cached";
1115 if (!strcmp(opt_overlap_check
, "none")) {
1116 overlap_check_template
= 0;
1117 } else if (!strcmp(opt_overlap_check
, "constant")) {
1118 overlap_check_template
= QCOW2_OL_CONSTANT
;
1119 } else if (!strcmp(opt_overlap_check
, "cached")) {
1120 overlap_check_template
= QCOW2_OL_CACHED
;
1121 } else if (!strcmp(opt_overlap_check
, "all")) {
1122 overlap_check_template
= QCOW2_OL_ALL
;
1124 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1125 "'overlap-check'. Allowed are any of the following: "
1126 "none, constant, cached, all", opt_overlap_check
);
1131 r
->overlap_check
= 0;
1132 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1133 /* overlap-check defines a template bitmask, but every flag may be
1134 * overwritten through the associated boolean option */
1136 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1137 overlap_check_template
& (1 << i
)) << i
;
1140 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1141 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1142 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1143 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1144 flags
& BDRV_O_UNMAP
);
1145 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1146 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1147 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1148 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1150 r
->discard_no_unref
= qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_NO_UNREF
,
1152 if (r
->discard_no_unref
&& s
->qcow_version
< 3) {
1154 "discard-no-unref is only supported since qcow2 version 3");
1159 switch (s
->crypt_method_header
) {
1160 case QCOW_CRYPT_NONE
:
1162 error_setg(errp
, "No encryption in image header, but options "
1163 "specified format '%s'", encryptfmt
);
1169 case QCOW_CRYPT_AES
:
1170 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1172 "Header reported 'aes' encryption format but "
1173 "options specify '%s'", encryptfmt
);
1177 qdict_put_str(encryptopts
, "format", "qcow");
1178 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1179 if (!r
->crypto_opts
) {
1185 case QCOW_CRYPT_LUKS
:
1186 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1188 "Header reported 'luks' encryption format but "
1189 "options specify '%s'", encryptfmt
);
1193 qdict_put_str(encryptopts
, "format", "luks");
1194 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1195 if (!r
->crypto_opts
) {
1202 error_setg(errp
, "Unsupported encryption method %d",
1203 s
->crypt_method_header
);
1210 qobject_unref(encryptopts
);
1211 qemu_opts_del(opts
);
1216 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1217 Qcow2ReopenState
*r
)
1219 BDRVQcow2State
*s
= bs
->opaque
;
1222 if (s
->l2_table_cache
) {
1223 qcow2_cache_destroy(s
->l2_table_cache
);
1225 if (s
->refcount_block_cache
) {
1226 qcow2_cache_destroy(s
->refcount_block_cache
);
1228 s
->l2_table_cache
= r
->l2_table_cache
;
1229 s
->refcount_block_cache
= r
->refcount_block_cache
;
1230 s
->l2_slice_size
= r
->l2_slice_size
;
1232 s
->overlap_check
= r
->overlap_check
;
1233 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1235 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1236 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1239 s
->discard_no_unref
= r
->discard_no_unref
;
1241 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1242 cache_clean_timer_del(bs
);
1243 s
->cache_clean_interval
= r
->cache_clean_interval
;
1244 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1247 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1248 s
->crypto_opts
= r
->crypto_opts
;
1251 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1252 Qcow2ReopenState
*r
)
1254 if (r
->l2_table_cache
) {
1255 qcow2_cache_destroy(r
->l2_table_cache
);
1257 if (r
->refcount_block_cache
) {
1258 qcow2_cache_destroy(r
->refcount_block_cache
);
1260 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1263 static int coroutine_fn
1264 qcow2_update_options(BlockDriverState
*bs
, QDict
*options
, int flags
,
1267 Qcow2ReopenState r
= {};
1270 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1272 qcow2_update_options_commit(bs
, &r
);
1274 qcow2_update_options_abort(bs
, &r
);
1280 static int validate_compression_type(BDRVQcow2State
*s
, Error
**errp
)
1282 switch (s
->compression_type
) {
1283 case QCOW2_COMPRESSION_TYPE_ZLIB
:
1285 case QCOW2_COMPRESSION_TYPE_ZSTD
:
1290 error_setg(errp
, "qcow2: unknown compression type: %u",
1291 s
->compression_type
);
1296 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1297 * the incompatible feature flag must be set
1299 if (s
->compression_type
== QCOW2_COMPRESSION_TYPE_ZLIB
) {
1300 if (s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
) {
1301 error_setg(errp
, "qcow2: Compression type incompatible feature "
1302 "bit must not be set");
1306 if (!(s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
)) {
1307 error_setg(errp
, "qcow2: Compression type incompatible feature "
1316 /* Called with s->lock held. */
1317 static int coroutine_fn GRAPH_RDLOCK
1318 qcow2_do_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1319 bool open_data_file
, Error
**errp
)
1322 BDRVQcow2State
*s
= bs
->opaque
;
1323 unsigned int len
, i
;
1327 uint64_t l1_vm_state_index
;
1328 bool update_header
= false;
1330 ret
= bdrv_co_pread(bs
->file
, 0, sizeof(header
), &header
, 0);
1332 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1335 header
.magic
= be32_to_cpu(header
.magic
);
1336 header
.version
= be32_to_cpu(header
.version
);
1337 header
.backing_file_offset
= be64_to_cpu(header
.backing_file_offset
);
1338 header
.backing_file_size
= be32_to_cpu(header
.backing_file_size
);
1339 header
.size
= be64_to_cpu(header
.size
);
1340 header
.cluster_bits
= be32_to_cpu(header
.cluster_bits
);
1341 header
.crypt_method
= be32_to_cpu(header
.crypt_method
);
1342 header
.l1_table_offset
= be64_to_cpu(header
.l1_table_offset
);
1343 header
.l1_size
= be32_to_cpu(header
.l1_size
);
1344 header
.refcount_table_offset
= be64_to_cpu(header
.refcount_table_offset
);
1345 header
.refcount_table_clusters
=
1346 be32_to_cpu(header
.refcount_table_clusters
);
1347 header
.snapshots_offset
= be64_to_cpu(header
.snapshots_offset
);
1348 header
.nb_snapshots
= be32_to_cpu(header
.nb_snapshots
);
1350 if (header
.magic
!= QCOW_MAGIC
) {
1351 error_setg(errp
, "Image is not in qcow2 format");
1355 if (header
.version
< 2 || header
.version
> 3) {
1356 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1361 s
->qcow_version
= header
.version
;
1363 /* Initialise cluster size */
1364 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1365 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1366 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1367 header
.cluster_bits
);
1372 s
->cluster_bits
= header
.cluster_bits
;
1373 s
->cluster_size
= 1 << s
->cluster_bits
;
1375 /* Initialise version 3 header fields */
1376 if (header
.version
== 2) {
1377 header
.incompatible_features
= 0;
1378 header
.compatible_features
= 0;
1379 header
.autoclear_features
= 0;
1380 header
.refcount_order
= 4;
1381 header
.header_length
= 72;
1383 header
.incompatible_features
=
1384 be64_to_cpu(header
.incompatible_features
);
1385 header
.compatible_features
= be64_to_cpu(header
.compatible_features
);
1386 header
.autoclear_features
= be64_to_cpu(header
.autoclear_features
);
1387 header
.refcount_order
= be32_to_cpu(header
.refcount_order
);
1388 header
.header_length
= be32_to_cpu(header
.header_length
);
1390 if (header
.header_length
< 104) {
1391 error_setg(errp
, "qcow2 header too short");
1397 if (header
.header_length
> s
->cluster_size
) {
1398 error_setg(errp
, "qcow2 header exceeds cluster size");
1403 if (header
.header_length
> sizeof(header
)) {
1404 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1405 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1406 ret
= bdrv_co_pread(bs
->file
, sizeof(header
),
1407 s
->unknown_header_fields_size
,
1408 s
->unknown_header_fields
, 0);
1410 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1416 if (header
.backing_file_offset
> s
->cluster_size
) {
1417 error_setg(errp
, "Invalid backing file offset");
1422 if (header
.backing_file_offset
) {
1423 ext_end
= header
.backing_file_offset
;
1425 ext_end
= 1 << header
.cluster_bits
;
1428 /* Handle feature bits */
1429 s
->incompatible_features
= header
.incompatible_features
;
1430 s
->compatible_features
= header
.compatible_features
;
1431 s
->autoclear_features
= header
.autoclear_features
;
1434 * Handle compression type
1435 * Older qcow2 images don't contain the compression type header.
1436 * Distinguish them by the header length and use
1437 * the only valid (default) compression type in that case
1439 if (header
.header_length
> offsetof(QCowHeader
, compression_type
)) {
1440 s
->compression_type
= header
.compression_type
;
1442 s
->compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
1445 ret
= validate_compression_type(s
, errp
);
1450 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1451 void *feature_table
= NULL
;
1452 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1453 &feature_table
, flags
, NULL
, NULL
);
1454 report_unsupported_feature(errp
, feature_table
,
1455 s
->incompatible_features
&
1456 ~QCOW2_INCOMPAT_MASK
);
1458 g_free(feature_table
);
1462 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1463 /* Corrupt images may not be written to unless they are being repaired
1465 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1466 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1473 s
->subclusters_per_cluster
=
1474 has_subclusters(s
) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
: 1;
1475 s
->subcluster_size
= s
->cluster_size
/ s
->subclusters_per_cluster
;
1476 s
->subcluster_bits
= ctz32(s
->subcluster_size
);
1478 if (s
->subcluster_size
< (1 << MIN_CLUSTER_BITS
)) {
1479 error_setg(errp
, "Unsupported subcluster size: %d", s
->subcluster_size
);
1484 /* Check support for various header values */
1485 if (header
.refcount_order
> 6) {
1486 error_setg(errp
, "Reference count entry width too large; may not "
1491 s
->refcount_order
= header
.refcount_order
;
1492 s
->refcount_bits
= 1 << s
->refcount_order
;
1493 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1494 s
->refcount_max
+= s
->refcount_max
- 1;
1496 s
->crypt_method_header
= header
.crypt_method
;
1497 if (s
->crypt_method_header
) {
1498 if (bdrv_uses_whitelist() &&
1499 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1501 "Use of AES-CBC encrypted qcow2 images is no longer "
1502 "supported in system emulators");
1503 error_append_hint(errp
,
1504 "You can use 'qemu-img convert' to convert your "
1505 "image to an alternative supported format, such "
1506 "as unencrypted qcow2, or raw with the LUKS "
1507 "format instead.\n");
1512 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1513 s
->crypt_physical_offset
= false;
1515 /* Assuming LUKS and any future crypt methods we
1516 * add will all use physical offsets, due to the
1517 * fact that the alternative is insecure... */
1518 s
->crypt_physical_offset
= true;
1521 bs
->encrypted
= true;
1524 s
->l2_bits
= s
->cluster_bits
- ctz32(l2_entry_size(s
));
1525 s
->l2_size
= 1 << s
->l2_bits
;
1526 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1527 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1528 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1529 bs
->total_sectors
= header
.size
/ BDRV_SECTOR_SIZE
;
1530 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1531 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1532 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1534 s
->refcount_table_offset
= header
.refcount_table_offset
;
1535 s
->refcount_table_size
=
1536 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1538 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1539 error_setg(errp
, "Image does not contain a reference count table");
1544 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1545 header
.refcount_table_clusters
,
1546 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1547 "Reference count table", errp
);
1552 if (!(flags
& BDRV_O_CHECK
)) {
1554 * The total size in bytes of the snapshot table is checked in
1555 * qcow2_read_snapshots() because the size of each snapshot is
1556 * variable and we don't know it yet.
1557 * Here we only check the offset and number of snapshots.
1559 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1560 header
.nb_snapshots
,
1561 sizeof(QCowSnapshotHeader
),
1562 sizeof(QCowSnapshotHeader
) *
1564 "Snapshot table", errp
);
1570 /* read the level 1 table */
1571 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1572 header
.l1_size
, L1E_SIZE
,
1573 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1577 s
->l1_size
= header
.l1_size
;
1578 s
->l1_table_offset
= header
.l1_table_offset
;
1580 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1581 if (l1_vm_state_index
> INT_MAX
) {
1582 error_setg(errp
, "Image is too big");
1586 s
->l1_vm_state_index
= l1_vm_state_index
;
1588 /* the L1 table must contain at least enough entries to put
1589 header.size bytes */
1590 if (s
->l1_size
< s
->l1_vm_state_index
) {
1591 error_setg(errp
, "L1 table is too small");
1596 if (s
->l1_size
> 0) {
1597 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
, s
->l1_size
* L1E_SIZE
);
1598 if (s
->l1_table
== NULL
) {
1599 error_setg(errp
, "Could not allocate L1 table");
1603 ret
= bdrv_co_pread(bs
->file
, s
->l1_table_offset
, s
->l1_size
* L1E_SIZE
,
1606 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1609 for(i
= 0;i
< s
->l1_size
; i
++) {
1610 s
->l1_table
[i
] = be64_to_cpu(s
->l1_table
[i
]);
1614 /* Parse driver-specific options */
1615 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1622 ret
= qcow2_refcount_init(bs
);
1624 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1628 QLIST_INIT(&s
->cluster_allocs
);
1629 QTAILQ_INIT(&s
->discards
);
1631 /* read qcow2 extensions */
1632 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1633 flags
, &update_header
, errp
)) {
1638 if (open_data_file
) {
1639 /* Open external data file */
1640 bdrv_graph_co_rdunlock();
1641 s
->data_file
= bdrv_co_open_child(NULL
, options
, "data-file", bs
,
1642 &child_of_bds
, BDRV_CHILD_DATA
,
1644 bdrv_graph_co_rdlock();
1650 if (s
->incompatible_features
& QCOW2_INCOMPAT_DATA_FILE
) {
1651 if (!s
->data_file
&& s
->image_data_file
) {
1652 bdrv_graph_co_rdunlock();
1653 s
->data_file
= bdrv_co_open_child(s
->image_data_file
, options
,
1656 BDRV_CHILD_DATA
, false, errp
);
1657 bdrv_graph_co_rdlock();
1658 if (!s
->data_file
) {
1663 if (!s
->data_file
) {
1664 error_setg(errp
, "'data-file' is required for this image");
1670 bs
->file
->role
&= ~BDRV_CHILD_DATA
;
1672 /* Must succeed because we have given up permissions if anything */
1673 bdrv_child_refresh_perms(bs
, bs
->file
, &error_abort
);
1676 error_setg(errp
, "'data-file' can only be set for images with "
1677 "an external data file");
1682 s
->data_file
= bs
->file
;
1684 if (data_file_is_raw(bs
)) {
1685 error_setg(errp
, "data-file-raw requires a data file");
1692 /* qcow2_read_extension may have set up the crypto context
1693 * if the crypt method needs a header region, some methods
1694 * don't need header extensions, so must check here
1696 if (s
->crypt_method_header
&& !s
->crypto
) {
1697 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1698 unsigned int cflags
= 0;
1699 if (flags
& BDRV_O_NO_IO
) {
1700 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1702 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1704 QCOW2_MAX_THREADS
, errp
);
1709 } else if (!(flags
& BDRV_O_NO_IO
)) {
1710 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1711 s
->crypt_method_header
);
1717 /* read the backing file name */
1718 if (header
.backing_file_offset
!= 0) {
1719 len
= header
.backing_file_size
;
1720 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1721 len
>= sizeof(bs
->backing_file
)) {
1722 error_setg(errp
, "Backing file name too long");
1727 s
->image_backing_file
= g_malloc(len
+ 1);
1728 ret
= bdrv_co_pread(bs
->file
, header
.backing_file_offset
, len
,
1729 s
->image_backing_file
, 0);
1731 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1734 s
->image_backing_file
[len
] = '\0';
1737 * Update only when something has changed. This function is called by
1738 * qcow2_co_invalidate_cache(), and we do not want to reset
1739 * auto_backing_file unless necessary.
1741 if (!g_str_equal(s
->image_backing_file
, bs
->backing_file
)) {
1742 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
),
1743 s
->image_backing_file
);
1744 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
1745 s
->image_backing_file
);
1750 * Internal snapshots; skip reading them in check mode, because
1751 * we do not need them then, and we do not want to abort because
1752 * of a broken table.
1754 if (!(flags
& BDRV_O_CHECK
)) {
1755 s
->snapshots_offset
= header
.snapshots_offset
;
1756 s
->nb_snapshots
= header
.nb_snapshots
;
1758 ret
= qcow2_read_snapshots(bs
, errp
);
1764 /* Clear unknown autoclear feature bits */
1765 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1766 update_header
= update_header
&& bdrv_is_writable(bs
);
1767 if (update_header
) {
1768 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1771 /* == Handle persistent dirty bitmaps ==
1773 * We want load dirty bitmaps in three cases:
1775 * 1. Normal open of the disk in active mode, not related to invalidation
1778 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1779 * bitmaps are _not_ migrating through migration channel, i.e.
1780 * 'dirty-bitmaps' capability is disabled.
1782 * 3. Invalidation of source vm after failed or canceled migration.
1783 * This is a very interesting case. There are two possible types of
1786 * A. Stored on inactivation and removed. They should be loaded from the
1789 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1790 * the migration channel (with dirty-bitmaps capability).
1792 * On the other hand, there are two possible sub-cases:
1794 * 3.1 disk was changed by somebody else while were inactive. In this
1795 * case all in-RAM dirty bitmaps (both persistent and not) are
1796 * definitely invalid. And we don't have any method to determine
1799 * Simple and safe thing is to just drop all the bitmaps of type B on
1800 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1802 * On the other hand, resuming source vm, if disk was already changed
1803 * is a bad thing anyway: not only bitmaps, the whole vm state is
1804 * out of sync with disk.
1806 * This means, that user or management tool, who for some reason
1807 * decided to resume source vm, after disk was already changed by
1808 * target vm, should at least drop all dirty bitmaps by hand.
1810 * So, we can ignore this case for now, but TODO: "generation"
1811 * extension for qcow2, to determine, that image was changed after
1812 * last inactivation. And if it is changed, we will drop (or at least
1813 * mark as 'invalid' all the bitmaps of type B, both persistent
1816 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1817 * to disk ('dirty-bitmaps' capability disabled), or not saved
1818 * ('dirty-bitmaps' capability enabled), but we don't need to care
1819 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1820 * and not stored has flag IN_USE=1 in the image and will be skipped
1823 * One remaining possible case when we don't want load bitmaps:
1825 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1826 * will be loaded on invalidation, no needs try loading them before)
1829 if (!(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
)) {
1830 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1831 bool header_updated
;
1832 if (!qcow2_load_dirty_bitmaps(bs
, &header_updated
, errp
)) {
1837 update_header
= update_header
&& !header_updated
;
1840 if (update_header
) {
1841 ret
= qcow2_update_header(bs
);
1843 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1848 bs
->supported_zero_flags
= header
.version
>= 3 ?
1849 BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
: 0;
1850 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
1852 /* Repair image if dirty */
1853 if (!(flags
& BDRV_O_CHECK
) && bdrv_is_writable(bs
) &&
1854 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1855 BdrvCheckResult result
= {0};
1857 ret
= qcow2_co_check_locked(bs
, &result
,
1858 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1859 if (ret
< 0 || result
.check_errors
) {
1863 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1870 BdrvCheckResult result
= {0};
1871 qcow2_check_refcounts(bs
, &result
, 0);
1875 qemu_co_queue_init(&s
->thread_task_queue
);
1880 g_free(s
->image_data_file
);
1881 if (open_data_file
&& has_data_file(bs
)) {
1882 bdrv_graph_co_rdunlock();
1883 bdrv_co_unref_child(bs
, s
->data_file
);
1884 bdrv_graph_co_rdlock();
1885 s
->data_file
= NULL
;
1887 g_free(s
->unknown_header_fields
);
1888 cleanup_unknown_header_ext(bs
);
1889 qcow2_free_snapshots(bs
);
1890 qcow2_refcount_close(bs
);
1891 qemu_vfree(s
->l1_table
);
1892 /* else pre-write overlap checks in cache_destroy may crash */
1894 cache_clean_timer_del(bs
);
1895 if (s
->l2_table_cache
) {
1896 qcow2_cache_destroy(s
->l2_table_cache
);
1898 if (s
->refcount_block_cache
) {
1899 qcow2_cache_destroy(s
->refcount_block_cache
);
1901 qcrypto_block_free(s
->crypto
);
1902 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1906 typedef struct QCow2OpenCo
{
1907 BlockDriverState
*bs
;
1914 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1916 QCow2OpenCo
*qoc
= opaque
;
1917 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1919 GRAPH_RDLOCK_GUARD();
1921 qemu_co_mutex_lock(&s
->lock
);
1922 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, true,
1924 qemu_co_mutex_unlock(&s
->lock
);
1929 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1932 BDRVQcow2State
*s
= bs
->opaque
;
1942 ret
= bdrv_open_file_child(NULL
, options
, "file", bs
, errp
);
1947 /* Initialise locks */
1948 qemu_co_mutex_init(&s
->lock
);
1950 assert(!qemu_in_coroutine());
1951 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1953 aio_co_enter(bdrv_get_aio_context(bs
),
1954 qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1955 AIO_WAIT_WHILE_UNLOCKED(NULL
, qoc
.ret
== -EINPROGRESS
);
1960 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1962 BDRVQcow2State
*s
= bs
->opaque
;
1964 if (bs
->encrypted
) {
1965 /* Encryption works on a sector granularity */
1966 bs
->bl
.request_alignment
= qcrypto_block_get_sector_size(s
->crypto
);
1968 bs
->bl
.pwrite_zeroes_alignment
= s
->subcluster_size
;
1969 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1972 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1973 BlockReopenQueue
*queue
, Error
**errp
)
1975 BDRVQcow2State
*s
= state
->bs
->opaque
;
1976 Qcow2ReopenState
*r
;
1979 r
= g_new0(Qcow2ReopenState
, 1);
1982 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1983 state
->flags
, errp
);
1988 /* We need to write out any unwritten data if we reopen read-only. */
1989 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1990 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1995 ret
= bdrv_flush(state
->bs
);
2000 ret
= qcow2_mark_clean(state
->bs
);
2007 * Without an external data file, s->data_file points to the same BdrvChild
2008 * as bs->file. It needs to be resynced after reopen because bs->file may
2009 * be changed. We can't use it in the meantime.
2011 if (!has_data_file(state
->bs
)) {
2012 assert(s
->data_file
== state
->bs
->file
);
2013 s
->data_file
= NULL
;
2019 qcow2_update_options_abort(state
->bs
, r
);
2024 static void qcow2_reopen_commit(BDRVReopenState
*state
)
2026 BDRVQcow2State
*s
= state
->bs
->opaque
;
2028 qcow2_update_options_commit(state
->bs
, state
->opaque
);
2029 if (!s
->data_file
) {
2031 * If we don't have an external data file, s->data_file was cleared by
2032 * qcow2_reopen_prepare() and needs to be updated.
2034 s
->data_file
= state
->bs
->file
;
2036 g_free(state
->opaque
);
2039 static void qcow2_reopen_commit_post(BDRVReopenState
*state
)
2041 if (state
->flags
& BDRV_O_RDWR
) {
2042 Error
*local_err
= NULL
;
2044 if (qcow2_reopen_bitmaps_rw(state
->bs
, &local_err
) < 0) {
2046 * This is not fatal, bitmaps just left read-only, so all following
2047 * writes will fail. User can remove read-only bitmaps to unblock
2048 * writes or retry reopen.
2050 error_reportf_err(local_err
,
2051 "%s: Failed to make dirty bitmaps writable: ",
2052 bdrv_get_node_name(state
->bs
));
2057 static void qcow2_reopen_abort(BDRVReopenState
*state
)
2059 BDRVQcow2State
*s
= state
->bs
->opaque
;
2061 if (!s
->data_file
) {
2063 * If we don't have an external data file, s->data_file was cleared by
2064 * qcow2_reopen_prepare() and needs to be restored.
2066 s
->data_file
= state
->bs
->file
;
2068 qcow2_update_options_abort(state
->bs
, state
->opaque
);
2069 g_free(state
->opaque
);
2072 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
2074 bool has_new_overlap_template
=
2075 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
2076 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2077 bool has_new_total_cache_size
=
2078 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
2079 bool has_all_cache_options
;
2081 /* New overlap template overrides all old overlap options */
2082 if (has_new_overlap_template
) {
2083 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
2084 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
2085 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
2086 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
2087 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
2088 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
2089 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
2090 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
2091 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
2092 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
2095 /* New total cache size overrides all old options */
2096 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
2097 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
2098 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2101 qdict_join(options
, old_options
, false);
2104 * If after merging all cache size options are set, an old total size is
2105 * overwritten. Do keep all options, however, if all three are new. The
2106 * resulting error message is what we want to happen.
2108 has_all_cache_options
=
2109 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
2110 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
2111 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
2113 if (has_all_cache_options
&& !has_new_total_cache_size
) {
2114 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
2118 static int coroutine_fn GRAPH_RDLOCK
2119 qcow2_co_block_status(BlockDriverState
*bs
, bool want_zero
, int64_t offset
,
2120 int64_t count
, int64_t *pnum
, int64_t *map
,
2121 BlockDriverState
**file
)
2123 BDRVQcow2State
*s
= bs
->opaque
;
2124 uint64_t host_offset
;
2126 QCow2SubclusterType type
;
2127 int ret
, status
= 0;
2129 qemu_co_mutex_lock(&s
->lock
);
2131 if (!s
->metadata_preallocation_checked
) {
2132 ret
= qcow2_detect_metadata_preallocation(bs
);
2133 s
->metadata_preallocation
= (ret
== 1);
2134 s
->metadata_preallocation_checked
= true;
2137 bytes
= MIN(INT_MAX
, count
);
2138 ret
= qcow2_get_host_offset(bs
, offset
, &bytes
, &host_offset
, &type
);
2139 qemu_co_mutex_unlock(&s
->lock
);
2146 if ((type
== QCOW2_SUBCLUSTER_NORMAL
||
2147 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2148 type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) && !s
->crypto
) {
2150 *file
= s
->data_file
->bs
;
2151 status
|= BDRV_BLOCK_OFFSET_VALID
;
2153 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2154 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
) {
2155 status
|= BDRV_BLOCK_ZERO
;
2156 } else if (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
2157 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
) {
2158 status
|= BDRV_BLOCK_DATA
;
2160 if (s
->metadata_preallocation
&& (status
& BDRV_BLOCK_DATA
) &&
2161 (status
& BDRV_BLOCK_OFFSET_VALID
))
2163 status
|= BDRV_BLOCK_RECURSE
;
2168 static int coroutine_fn GRAPH_RDLOCK
2169 qcow2_handle_l2meta(BlockDriverState
*bs
, QCowL2Meta
**pl2meta
, bool link_l2
)
2172 QCowL2Meta
*l2meta
= *pl2meta
;
2174 while (l2meta
!= NULL
) {
2178 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
2183 qcow2_alloc_cluster_abort(bs
, l2meta
);
2186 /* Take the request off the list of running requests */
2187 QLIST_REMOVE(l2meta
, next_in_flight
);
2189 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
2191 next
= l2meta
->next
;
2200 static int coroutine_fn GRAPH_RDLOCK
2201 qcow2_co_preadv_encrypted(BlockDriverState
*bs
,
2202 uint64_t host_offset
,
2206 uint64_t qiov_offset
)
2209 BDRVQcow2State
*s
= bs
->opaque
;
2212 assert(bs
->encrypted
&& s
->crypto
);
2213 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2216 * For encrypted images, read everything into a temporary
2217 * contiguous buffer on which the AES functions can work.
2218 * Also, decryption in a separate buffer is better as it
2219 * prevents the guest from learning information about the
2220 * encrypted nature of the virtual disk.
2223 buf
= qemu_try_blockalign(s
->data_file
->bs
, bytes
);
2228 BLKDBG_CO_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2229 ret
= bdrv_co_pread(s
->data_file
, host_offset
, bytes
, buf
, 0);
2234 if (qcow2_co_decrypt(bs
, host_offset
, offset
, buf
, bytes
) < 0)
2239 qemu_iovec_from_buf(qiov
, qiov_offset
, buf
, bytes
);
2247 typedef struct Qcow2AioTask
{
2250 BlockDriverState
*bs
;
2251 QCow2SubclusterType subcluster_type
; /* only for read */
2252 uint64_t host_offset
; /* or l2_entry for compressed read */
2256 uint64_t qiov_offset
;
2257 QCowL2Meta
*l2meta
; /* only for write */
2260 static coroutine_fn
int qcow2_co_preadv_task_entry(AioTask
*task
);
2261 static coroutine_fn
int qcow2_add_task(BlockDriverState
*bs
,
2264 QCow2SubclusterType subcluster_type
,
2265 uint64_t host_offset
,
2272 Qcow2AioTask local_task
;
2273 Qcow2AioTask
*task
= pool
? g_new(Qcow2AioTask
, 1) : &local_task
;
2275 *task
= (Qcow2AioTask
) {
2278 .subcluster_type
= subcluster_type
,
2280 .host_offset
= host_offset
,
2283 .qiov_offset
= qiov_offset
,
2287 trace_qcow2_add_task(qemu_coroutine_self(), bs
, pool
,
2288 func
== qcow2_co_preadv_task_entry
? "read" : "write",
2289 subcluster_type
, host_offset
, offset
, bytes
,
2293 return func(&task
->task
);
2296 aio_task_pool_start_task(pool
, &task
->task
);
2301 static int coroutine_fn GRAPH_RDLOCK
2302 qcow2_co_preadv_task(BlockDriverState
*bs
, QCow2SubclusterType subc_type
,
2303 uint64_t host_offset
, uint64_t offset
, uint64_t bytes
,
2304 QEMUIOVector
*qiov
, size_t qiov_offset
)
2306 BDRVQcow2State
*s
= bs
->opaque
;
2308 switch (subc_type
) {
2309 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
2310 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
2311 /* Both zero types are handled in qcow2_co_preadv_part */
2312 g_assert_not_reached();
2314 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
2315 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
2316 assert(bs
->backing
); /* otherwise handled in qcow2_co_preadv_part */
2318 BLKDBG_CO_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
2319 return bdrv_co_preadv_part(bs
->backing
, offset
, bytes
,
2320 qiov
, qiov_offset
, 0);
2322 case QCOW2_SUBCLUSTER_COMPRESSED
:
2323 return qcow2_co_preadv_compressed(bs
, host_offset
,
2324 offset
, bytes
, qiov
, qiov_offset
);
2326 case QCOW2_SUBCLUSTER_NORMAL
:
2327 if (bs
->encrypted
) {
2328 return qcow2_co_preadv_encrypted(bs
, host_offset
,
2329 offset
, bytes
, qiov
, qiov_offset
);
2332 BLKDBG_CO_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2333 return bdrv_co_preadv_part(s
->data_file
, host_offset
,
2334 bytes
, qiov
, qiov_offset
, 0);
2337 g_assert_not_reached();
2340 g_assert_not_reached();
2344 * This function can count as GRAPH_RDLOCK because qcow2_co_preadv_part() holds
2345 * the graph lock and keeps it until this coroutine has terminated.
2347 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_preadv_task_entry(AioTask
*task
)
2349 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2353 return qcow2_co_preadv_task(t
->bs
, t
->subcluster_type
,
2354 t
->host_offset
, t
->offset
, t
->bytes
,
2355 t
->qiov
, t
->qiov_offset
);
2358 static int coroutine_fn GRAPH_RDLOCK
2359 qcow2_co_preadv_part(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
2360 QEMUIOVector
*qiov
, size_t qiov_offset
,
2361 BdrvRequestFlags flags
)
2363 BDRVQcow2State
*s
= bs
->opaque
;
2365 unsigned int cur_bytes
; /* number of bytes in current iteration */
2366 uint64_t host_offset
= 0;
2367 QCow2SubclusterType type
;
2368 AioTaskPool
*aio
= NULL
;
2370 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2371 /* prepare next request */
2372 cur_bytes
= MIN(bytes
, INT_MAX
);
2374 cur_bytes
= MIN(cur_bytes
,
2375 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2378 qemu_co_mutex_lock(&s
->lock
);
2379 ret
= qcow2_get_host_offset(bs
, offset
, &cur_bytes
,
2380 &host_offset
, &type
);
2381 qemu_co_mutex_unlock(&s
->lock
);
2386 if (type
== QCOW2_SUBCLUSTER_ZERO_PLAIN
||
2387 type
== QCOW2_SUBCLUSTER_ZERO_ALLOC
||
2388 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&& !bs
->backing
) ||
2389 (type
== QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&& !bs
->backing
))
2391 qemu_iovec_memset(qiov
, qiov_offset
, 0, cur_bytes
);
2393 if (!aio
&& cur_bytes
!= bytes
) {
2394 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2396 ret
= qcow2_add_task(bs
, aio
, qcow2_co_preadv_task_entry
, type
,
2397 host_offset
, offset
, cur_bytes
,
2398 qiov
, qiov_offset
, NULL
);
2405 offset
+= cur_bytes
;
2406 qiov_offset
+= cur_bytes
;
2411 aio_task_pool_wait_all(aio
);
2413 ret
= aio_task_pool_status(aio
);
2421 /* Check if it's possible to merge a write request with the writing of
2422 * the data from the COW regions */
2423 static bool merge_cow(uint64_t offset
, unsigned bytes
,
2424 QEMUIOVector
*qiov
, size_t qiov_offset
,
2429 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2430 /* If both COW regions are empty then there's nothing to merge */
2431 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
2435 /* If COW regions are handled already, skip this too */
2441 * The write request should start immediately after the first
2442 * COW region. This does not always happen because the area
2443 * touched by the request can be larger than the one defined
2444 * by @m (a single request can span an area consisting of a
2445 * mix of previously unallocated and allocated clusters, that
2446 * is why @l2meta is a list).
2448 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
2449 /* In this case the request starts before this region */
2450 assert(offset
< l2meta_cow_start(m
));
2451 assert(m
->cow_start
.nb_bytes
== 0);
2455 /* The write request should end immediately before the second
2456 * COW region (see above for why it does not always happen) */
2457 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
2458 assert(offset
+ bytes
> m
->offset
+ m
->cow_end
.offset
);
2459 assert(m
->cow_end
.nb_bytes
== 0);
2463 /* Make sure that adding both COW regions to the QEMUIOVector
2464 * does not exceed IOV_MAX */
2465 if (qemu_iovec_subvec_niov(qiov
, qiov_offset
, bytes
) > IOV_MAX
- 2) {
2469 m
->data_qiov
= qiov
;
2470 m
->data_qiov_offset
= qiov_offset
;
2478 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error.
2479 * Note that returning 0 does not guarantee non-zero data.
2481 static int coroutine_fn GRAPH_RDLOCK
2482 is_zero_cow(BlockDriverState
*bs
, QCowL2Meta
*m
)
2485 * This check is designed for optimization shortcut so it must be
2487 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is
2488 * faster (but not as accurate and can result in false negatives).
2490 int ret
= bdrv_co_is_zero_fast(bs
, m
->offset
+ m
->cow_start
.offset
,
2491 m
->cow_start
.nb_bytes
);
2496 return bdrv_co_is_zero_fast(bs
, m
->offset
+ m
->cow_end
.offset
,
2497 m
->cow_end
.nb_bytes
);
2500 static int coroutine_fn GRAPH_RDLOCK
2501 handle_alloc_space(BlockDriverState
*bs
, QCowL2Meta
*l2meta
)
2503 BDRVQcow2State
*s
= bs
->opaque
;
2506 if (!(s
->data_file
->bs
->supported_zero_flags
& BDRV_REQ_NO_FALLBACK
)) {
2510 if (bs
->encrypted
) {
2514 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2516 uint64_t start_offset
= m
->alloc_offset
+ m
->cow_start
.offset
;
2517 unsigned nb_bytes
= m
->cow_end
.offset
+ m
->cow_end
.nb_bytes
-
2518 m
->cow_start
.offset
;
2520 if (!m
->cow_start
.nb_bytes
&& !m
->cow_end
.nb_bytes
) {
2524 ret
= is_zero_cow(bs
, m
);
2527 } else if (ret
== 0) {
2532 * instead of writing zero COW buffers,
2533 * efficiently zero out the whole clusters
2536 ret
= qcow2_pre_write_overlap_check(bs
, 0, start_offset
, nb_bytes
,
2542 BLKDBG_CO_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC_SPACE
);
2543 ret
= bdrv_co_pwrite_zeroes(s
->data_file
, start_offset
, nb_bytes
,
2544 BDRV_REQ_NO_FALLBACK
);
2546 if (ret
!= -ENOTSUP
&& ret
!= -EAGAIN
) {
2552 trace_qcow2_skip_cow(qemu_coroutine_self(), m
->offset
, m
->nb_clusters
);
2559 * qcow2_co_pwritev_task
2560 * Called with s->lock unlocked
2561 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2562 * not use it somehow after qcow2_co_pwritev_task() call
2564 static coroutine_fn GRAPH_RDLOCK
2565 int qcow2_co_pwritev_task(BlockDriverState
*bs
, uint64_t host_offset
,
2566 uint64_t offset
, uint64_t bytes
, QEMUIOVector
*qiov
,
2567 uint64_t qiov_offset
, QCowL2Meta
*l2meta
)
2570 BDRVQcow2State
*s
= bs
->opaque
;
2571 void *crypt_buf
= NULL
;
2572 QEMUIOVector encrypted_qiov
;
2574 if (bs
->encrypted
) {
2576 assert(bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2577 crypt_buf
= qemu_try_blockalign(bs
->file
->bs
, bytes
);
2578 if (crypt_buf
== NULL
) {
2582 qemu_iovec_to_buf(qiov
, qiov_offset
, crypt_buf
, bytes
);
2584 if (qcow2_co_encrypt(bs
, host_offset
, offset
, crypt_buf
, bytes
) < 0) {
2589 qemu_iovec_init_buf(&encrypted_qiov
, crypt_buf
, bytes
);
2590 qiov
= &encrypted_qiov
;
2594 /* Try to efficiently initialize the physical space with zeroes */
2595 ret
= handle_alloc_space(bs
, l2meta
);
2601 * If we need to do COW, check if it's possible to merge the
2602 * writing of the guest data together with that of the COW regions.
2603 * If it's not possible (or not necessary) then write the
2606 if (!merge_cow(offset
, bytes
, qiov
, qiov_offset
, l2meta
)) {
2607 BLKDBG_CO_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2608 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset
);
2609 ret
= bdrv_co_pwritev_part(s
->data_file
, host_offset
,
2610 bytes
, qiov
, qiov_offset
, 0);
2616 qemu_co_mutex_lock(&s
->lock
);
2618 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
2622 qemu_co_mutex_lock(&s
->lock
);
2625 qcow2_handle_l2meta(bs
, &l2meta
, false);
2626 qemu_co_mutex_unlock(&s
->lock
);
2628 qemu_vfree(crypt_buf
);
2634 * This function can count as GRAPH_RDLOCK because qcow2_co_pwritev_part() holds
2635 * the graph lock and keeps it until this coroutine has terminated.
2637 static coroutine_fn GRAPH_RDLOCK
int qcow2_co_pwritev_task_entry(AioTask
*task
)
2639 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
2641 assert(!t
->subcluster_type
);
2643 return qcow2_co_pwritev_task(t
->bs
, t
->host_offset
,
2644 t
->offset
, t
->bytes
, t
->qiov
, t
->qiov_offset
,
2648 static int coroutine_fn GRAPH_RDLOCK
2649 qcow2_co_pwritev_part(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
2650 QEMUIOVector
*qiov
, size_t qiov_offset
,
2651 BdrvRequestFlags flags
)
2653 BDRVQcow2State
*s
= bs
->opaque
;
2654 int offset_in_cluster
;
2656 unsigned int cur_bytes
; /* number of sectors in current iteration */
2657 uint64_t host_offset
;
2658 QCowL2Meta
*l2meta
= NULL
;
2659 AioTaskPool
*aio
= NULL
;
2661 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
2663 while (bytes
!= 0 && aio_task_pool_status(aio
) == 0) {
2667 trace_qcow2_writev_start_part(qemu_coroutine_self());
2668 offset_in_cluster
= offset_into_cluster(s
, offset
);
2669 cur_bytes
= MIN(bytes
, INT_MAX
);
2670 if (bs
->encrypted
) {
2671 cur_bytes
= MIN(cur_bytes
,
2672 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
2673 - offset_in_cluster
);
2676 qemu_co_mutex_lock(&s
->lock
);
2678 ret
= qcow2_alloc_host_offset(bs
, offset
, &cur_bytes
,
2679 &host_offset
, &l2meta
);
2684 ret
= qcow2_pre_write_overlap_check(bs
, 0, host_offset
,
2690 qemu_co_mutex_unlock(&s
->lock
);
2692 if (!aio
&& cur_bytes
!= bytes
) {
2693 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
2695 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_task_entry
, 0,
2696 host_offset
, offset
,
2697 cur_bytes
, qiov
, qiov_offset
, l2meta
);
2698 l2meta
= NULL
; /* l2meta is consumed by qcow2_co_pwritev_task() */
2704 offset
+= cur_bytes
;
2705 qiov_offset
+= cur_bytes
;
2706 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2710 qemu_co_mutex_lock(&s
->lock
);
2713 qcow2_handle_l2meta(bs
, &l2meta
, false);
2715 qemu_co_mutex_unlock(&s
->lock
);
2719 aio_task_pool_wait_all(aio
);
2721 ret
= aio_task_pool_status(aio
);
2726 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2731 static int qcow2_inactivate(BlockDriverState
*bs
)
2733 BDRVQcow2State
*s
= bs
->opaque
;
2734 int ret
, result
= 0;
2735 Error
*local_err
= NULL
;
2737 qcow2_store_persistent_dirty_bitmaps(bs
, true, &local_err
);
2738 if (local_err
!= NULL
) {
2740 error_reportf_err(local_err
, "Lost persistent bitmaps during "
2741 "inactivation of node '%s': ",
2742 bdrv_get_device_or_node_name(bs
));
2745 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2748 error_report("Failed to flush the L2 table cache: %s",
2752 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2755 error_report("Failed to flush the refcount block cache: %s",
2760 qcow2_mark_clean(bs
);
2766 static void qcow2_do_close(BlockDriverState
*bs
, bool close_data_file
)
2768 BDRVQcow2State
*s
= bs
->opaque
;
2769 qemu_vfree(s
->l1_table
);
2770 /* else pre-write overlap checks in cache_destroy may crash */
2773 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2774 qcow2_inactivate(bs
);
2777 cache_clean_timer_del(bs
);
2778 qcow2_cache_destroy(s
->l2_table_cache
);
2779 qcow2_cache_destroy(s
->refcount_block_cache
);
2781 qcrypto_block_free(s
->crypto
);
2783 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
2785 g_free(s
->unknown_header_fields
);
2786 cleanup_unknown_header_ext(bs
);
2788 g_free(s
->image_data_file
);
2789 g_free(s
->image_backing_file
);
2790 g_free(s
->image_backing_format
);
2792 if (close_data_file
&& has_data_file(bs
)) {
2793 bdrv_graph_wrlock(NULL
);
2794 bdrv_unref_child(bs
, s
->data_file
);
2795 bdrv_graph_wrunlock();
2796 s
->data_file
= NULL
;
2799 qcow2_refcount_close(bs
);
2800 qcow2_free_snapshots(bs
);
2803 static void qcow2_close(BlockDriverState
*bs
)
2805 qcow2_do_close(bs
, true);
2808 static void coroutine_fn GRAPH_RDLOCK
2809 qcow2_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
2812 BDRVQcow2State
*s
= bs
->opaque
;
2813 BdrvChild
*data_file
;
2814 int flags
= s
->flags
;
2815 QCryptoBlock
*crypto
= NULL
;
2820 * Backing files are read-only which makes all of their metadata immutable,
2821 * that means we don't have to worry about reopening them here.
2828 * Do not reopen s->data_file (i.e., have qcow2_do_close() not close it,
2829 * and then prevent qcow2_do_open() from opening it), because this function
2830 * runs in the I/O path and as such we must not invoke global-state
2831 * functions like bdrv_unref_child() and bdrv_open_child().
2834 qcow2_do_close(bs
, false);
2836 data_file
= s
->data_file
;
2837 memset(s
, 0, sizeof(BDRVQcow2State
));
2838 s
->data_file
= data_file
;
2840 options
= qdict_clone_shallow(bs
->options
);
2842 flags
&= ~BDRV_O_INACTIVE
;
2843 qemu_co_mutex_lock(&s
->lock
);
2844 ret
= qcow2_do_open(bs
, options
, flags
, false, errp
);
2845 qemu_co_mutex_unlock(&s
->lock
);
2846 qobject_unref(options
);
2848 error_prepend(errp
, "Could not reopen qcow2 layer: ");
2856 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2857 size_t len
, size_t buflen
)
2859 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2860 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2862 if (buflen
< ext_len
) {
2866 *ext_backing_fmt
= (QCowExtension
) {
2867 .magic
= cpu_to_be32(magic
),
2868 .len
= cpu_to_be32(len
),
2872 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2879 * Updates the qcow2 header, including the variable length parts of it, i.e.
2880 * the backing file name and all extensions. qcow2 was not designed to allow
2881 * such changes, so if we run out of space (we can only use the first cluster)
2882 * this function may fail.
2884 * Returns 0 on success, -errno in error cases.
2886 int qcow2_update_header(BlockDriverState
*bs
)
2888 BDRVQcow2State
*s
= bs
->opaque
;
2891 size_t buflen
= s
->cluster_size
;
2893 uint64_t total_size
;
2894 uint32_t refcount_table_clusters
;
2895 size_t header_length
;
2896 Qcow2UnknownHeaderExtension
*uext
;
2898 buf
= qemu_blockalign(bs
, buflen
);
2900 /* Header structure */
2901 header
= (QCowHeader
*) buf
;
2903 if (buflen
< sizeof(*header
)) {
2908 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2909 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2910 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2912 ret
= validate_compression_type(s
, NULL
);
2917 *header
= (QCowHeader
) {
2918 /* Version 2 fields */
2919 .magic
= cpu_to_be32(QCOW_MAGIC
),
2920 .version
= cpu_to_be32(s
->qcow_version
),
2921 .backing_file_offset
= 0,
2922 .backing_file_size
= 0,
2923 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2924 .size
= cpu_to_be64(total_size
),
2925 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2926 .l1_size
= cpu_to_be32(s
->l1_size
),
2927 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2928 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2929 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2930 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2931 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2933 /* Version 3 fields */
2934 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2935 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2936 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2937 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2938 .header_length
= cpu_to_be32(header_length
),
2939 .compression_type
= s
->compression_type
,
2942 /* For older versions, write a shorter header */
2943 switch (s
->qcow_version
) {
2945 ret
= offsetof(QCowHeader
, incompatible_features
);
2948 ret
= sizeof(*header
);
2957 memset(buf
, 0, buflen
);
2959 /* Preserve any unknown field in the header */
2960 if (s
->unknown_header_fields_size
) {
2961 if (buflen
< s
->unknown_header_fields_size
) {
2966 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2967 buf
+= s
->unknown_header_fields_size
;
2968 buflen
-= s
->unknown_header_fields_size
;
2971 /* Backing file format header extension */
2972 if (s
->image_backing_format
) {
2973 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2974 s
->image_backing_format
,
2975 strlen(s
->image_backing_format
),
2985 /* External data file header extension */
2986 if (has_data_file(bs
) && s
->image_data_file
) {
2987 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_DATA_FILE
,
2988 s
->image_data_file
, strlen(s
->image_data_file
),
2998 /* Full disk encryption header pointer extension */
2999 if (s
->crypto_header
.offset
!= 0) {
3000 s
->crypto_header
.offset
= cpu_to_be64(s
->crypto_header
.offset
);
3001 s
->crypto_header
.length
= cpu_to_be64(s
->crypto_header
.length
);
3002 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
3003 &s
->crypto_header
, sizeof(s
->crypto_header
),
3005 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
3006 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
3015 * Feature table. A mere 8 feature names occupies 392 bytes, and
3016 * when coupled with the v3 minimum header of 104 bytes plus the
3017 * 8-byte end-of-extension marker, that would leave only 8 bytes
3018 * for a backing file name in an image with 512-byte clusters.
3019 * Thus, we choose to omit this header for cluster sizes 4k and
3022 if (s
->qcow_version
>= 3 && s
->cluster_size
> 4096) {
3023 static const Qcow2Feature features
[] = {
3025 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3026 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
3027 .name
= "dirty bit",
3030 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3031 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
3032 .name
= "corrupt bit",
3035 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3036 .bit
= QCOW2_INCOMPAT_DATA_FILE_BITNR
,
3037 .name
= "external data file",
3040 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3041 .bit
= QCOW2_INCOMPAT_COMPRESSION_BITNR
,
3042 .name
= "compression type",
3045 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
3046 .bit
= QCOW2_INCOMPAT_EXTL2_BITNR
,
3047 .name
= "extended L2 entries",
3050 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
3051 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
3052 .name
= "lazy refcounts",
3055 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
3056 .bit
= QCOW2_AUTOCLEAR_BITMAPS_BITNR
,
3060 .type
= QCOW2_FEAT_TYPE_AUTOCLEAR
,
3061 .bit
= QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR
,
3062 .name
= "raw external data",
3066 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
3067 features
, sizeof(features
), buflen
);
3075 /* Bitmap extension */
3076 if (s
->nb_bitmaps
> 0) {
3077 Qcow2BitmapHeaderExt bitmaps_header
= {
3078 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
3079 .bitmap_directory_size
=
3080 cpu_to_be64(s
->bitmap_directory_size
),
3081 .bitmap_directory_offset
=
3082 cpu_to_be64(s
->bitmap_directory_offset
)
3084 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
3085 &bitmaps_header
, sizeof(bitmaps_header
),
3094 /* Keep unknown header extensions */
3095 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
3096 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
3105 /* End of header extensions */
3106 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
3114 /* Backing file name */
3115 if (s
->image_backing_file
) {
3116 size_t backing_file_len
= strlen(s
->image_backing_file
);
3118 if (buflen
< backing_file_len
) {
3123 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3124 strncpy(buf
, s
->image_backing_file
, buflen
);
3126 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
3127 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
3130 /* Write the new header */
3131 ret
= bdrv_pwrite(bs
->file
, 0, s
->cluster_size
, header
, 0);
3142 static int qcow2_change_backing_file(BlockDriverState
*bs
,
3143 const char *backing_file
, const char *backing_fmt
)
3145 BDRVQcow2State
*s
= bs
->opaque
;
3147 /* Adding a backing file means that the external data file alone won't be
3148 * enough to make sense of the content */
3149 if (backing_file
&& data_file_is_raw(bs
)) {
3153 if (backing_file
&& strlen(backing_file
) > 1023) {
3157 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3158 backing_file
?: "");
3159 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3160 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3162 g_free(s
->image_backing_file
);
3163 g_free(s
->image_backing_format
);
3165 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
3166 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
3168 return qcow2_update_header(bs
);
3171 static int coroutine_fn GRAPH_RDLOCK
3172 qcow2_set_up_encryption(BlockDriverState
*bs
,
3173 QCryptoBlockCreateOptions
*cryptoopts
,
3176 BDRVQcow2State
*s
= bs
->opaque
;
3177 QCryptoBlock
*crypto
= NULL
;
3180 switch (cryptoopts
->format
) {
3181 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
3182 fmt
= QCOW_CRYPT_LUKS
;
3184 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
3185 fmt
= QCOW_CRYPT_AES
;
3188 error_setg(errp
, "Crypto format not supported in qcow2");
3192 s
->crypt_method_header
= fmt
;
3194 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
3195 qcow2_crypto_hdr_init_func
,
3196 qcow2_crypto_hdr_write_func
,
3202 ret
= qcow2_update_header(bs
);
3204 error_setg_errno(errp
, -ret
, "Could not write encryption header");
3210 qcrypto_block_free(crypto
);
3215 * Preallocates metadata structures for data clusters between @offset (in the
3216 * guest disk) and @new_length (which is thus generally the new guest disk
3219 * Returns: 0 on success, -errno on failure.
3221 static int coroutine_fn GRAPH_RDLOCK
3222 preallocate_co(BlockDriverState
*bs
, uint64_t offset
, uint64_t new_length
,
3223 PreallocMode mode
, Error
**errp
)
3225 BDRVQcow2State
*s
= bs
->opaque
;
3227 uint64_t host_offset
= 0;
3228 int64_t file_length
;
3229 unsigned int cur_bytes
;
3231 QCowL2Meta
*meta
= NULL
, *m
;
3233 assert(offset
<= new_length
);
3234 bytes
= new_length
- offset
;
3237 cur_bytes
= MIN(bytes
, QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
));
3238 ret
= qcow2_alloc_host_offset(bs
, offset
, &cur_bytes
,
3239 &host_offset
, &meta
);
3241 error_setg_errno(errp
, -ret
, "Allocating clusters failed");
3245 for (m
= meta
; m
!= NULL
; m
= m
->next
) {
3249 ret
= qcow2_handle_l2meta(bs
, &meta
, true);
3251 error_setg_errno(errp
, -ret
, "Mapping clusters failed");
3255 /* TODO Preallocate data if requested */
3258 offset
+= cur_bytes
;
3262 * It is expected that the image file is large enough to actually contain
3263 * all of the allocated clusters (otherwise we get failing reads after
3264 * EOF). Extend the image to the last allocated sector.
3266 file_length
= bdrv_co_getlength(s
->data_file
->bs
);
3267 if (file_length
< 0) {
3268 error_setg_errno(errp
, -file_length
, "Could not get file size");
3273 if (host_offset
+ cur_bytes
> file_length
) {
3274 if (mode
== PREALLOC_MODE_METADATA
) {
3275 mode
= PREALLOC_MODE_OFF
;
3277 ret
= bdrv_co_truncate(s
->data_file
, host_offset
+ cur_bytes
, false,
3287 qcow2_handle_l2meta(bs
, &meta
, false);
3291 /* qcow2_refcount_metadata_size:
3292 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3293 * @cluster_size: size of a cluster, in bytes
3294 * @refcount_order: refcount bits power-of-2 exponent
3295 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3298 * Returns: Number of bytes required for refcount blocks and table metadata.
3300 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
3301 int refcount_order
, bool generous_increase
,
3302 uint64_t *refblock_count
)
3305 * Every host cluster is reference-counted, including metadata (even
3306 * refcount metadata is recursively included).
3308 * An accurate formula for the size of refcount metadata size is difficult
3309 * to derive. An easier method of calculation is finding the fixed point
3310 * where no further refcount blocks or table clusters are required to
3311 * reference count every cluster.
3313 int64_t blocks_per_table_cluster
= cluster_size
/ REFTABLE_ENTRY_SIZE
;
3314 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
3315 int64_t table
= 0; /* number of refcount table clusters */
3316 int64_t blocks
= 0; /* number of refcount block clusters */
3322 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
3323 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
3324 n
= clusters
+ blocks
+ table
;
3326 if (n
== last
&& generous_increase
) {
3327 clusters
+= DIV_ROUND_UP(table
, 2);
3328 n
= 0; /* force another loop */
3329 generous_increase
= false;
3331 } while (n
!= last
);
3333 if (refblock_count
) {
3334 *refblock_count
= blocks
;
3337 return (blocks
+ table
) * cluster_size
;
3341 * qcow2_calc_prealloc_size:
3342 * @total_size: virtual disk size in bytes
3343 * @cluster_size: cluster size in bytes
3344 * @refcount_order: refcount bits power-of-2 exponent
3345 * @extended_l2: true if the image has extended L2 entries
3347 * Returns: Total number of bytes required for the fully allocated image
3348 * (including metadata).
3350 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
3351 size_t cluster_size
,
3355 int64_t meta_size
= 0;
3356 uint64_t nl1e
, nl2e
;
3357 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
3358 size_t l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
3360 /* header: 1 cluster */
3361 meta_size
+= cluster_size
;
3363 /* total size of L2 tables */
3364 nl2e
= aligned_total_size
/ cluster_size
;
3365 nl2e
= ROUND_UP(nl2e
, cluster_size
/ l2e_size
);
3366 meta_size
+= nl2e
* l2e_size
;
3368 /* total size of L1 tables */
3369 nl1e
= nl2e
* l2e_size
/ cluster_size
;
3370 nl1e
= ROUND_UP(nl1e
, cluster_size
/ L1E_SIZE
);
3371 meta_size
+= nl1e
* L1E_SIZE
;
3373 /* total size of refcount table and blocks */
3374 meta_size
+= qcow2_refcount_metadata_size(
3375 (meta_size
+ aligned_total_size
) / cluster_size
,
3376 cluster_size
, refcount_order
, false, NULL
);
3378 return meta_size
+ aligned_total_size
;
3381 static bool validate_cluster_size(size_t cluster_size
, bool extended_l2
,
3384 int cluster_bits
= ctz32(cluster_size
);
3385 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
3386 (1 << cluster_bits
) != cluster_size
)
3388 error_setg(errp
, "Cluster size must be a power of two between %d and "
3389 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
3394 unsigned min_cluster_size
=
3395 (1 << MIN_CLUSTER_BITS
) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER
;
3396 if (cluster_size
< min_cluster_size
) {
3397 error_setg(errp
, "Extended L2 entries are only supported with "
3398 "cluster sizes of at least %u bytes", min_cluster_size
);
3406 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, bool extended_l2
,
3409 size_t cluster_size
;
3411 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
3412 DEFAULT_CLUSTER_SIZE
);
3413 if (!validate_cluster_size(cluster_size
, extended_l2
, errp
)) {
3416 return cluster_size
;
3419 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
3424 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
3426 ret
= 3; /* default */
3427 } else if (!strcmp(buf
, "0.10")) {
3429 } else if (!strcmp(buf
, "1.1")) {
3432 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
3439 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
3442 uint64_t refcount_bits
;
3444 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
3445 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
3446 error_setg(errp
, "Refcount width must be a power of two and may not "
3451 if (version
< 3 && refcount_bits
!= 16) {
3452 error_setg(errp
, "Different refcount widths than 16 bits require "
3453 "compatibility level 1.1 or above (use compat=1.1 or "
3458 return refcount_bits
;
3461 static int coroutine_fn GRAPH_UNLOCKED
3462 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
3464 BlockdevCreateOptionsQcow2
*qcow2_opts
;
3468 * Open the image file and write a minimal qcow2 header.
3470 * We keep things simple and start with a zero-sized image. We also
3471 * do without refcount blocks or a L1 table for now. We'll fix the
3472 * inconsistency later.
3474 * We do need a refcount table because growing the refcount table means
3475 * allocating two new refcount blocks - the second of which would be at
3476 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3477 * size for any qcow2 image.
3479 BlockBackend
*blk
= NULL
;
3480 BlockDriverState
*bs
= NULL
;
3481 BlockDriverState
*data_bs
= NULL
;
3483 size_t cluster_size
;
3486 uint64_t *refcount_table
;
3488 uint8_t compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
3490 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
3491 qcow2_opts
= &create_options
->u
.qcow2
;
3493 bs
= bdrv_co_open_blockdev_ref(qcow2_opts
->file
, errp
);
3498 /* Validate options and set default values */
3499 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
3500 error_setg(errp
, "Image size must be a multiple of %u bytes",
3501 (unsigned) BDRV_SECTOR_SIZE
);
3506 if (qcow2_opts
->has_version
) {
3507 switch (qcow2_opts
->version
) {
3508 case BLOCKDEV_QCOW2_VERSION_V2
:
3511 case BLOCKDEV_QCOW2_VERSION_V3
:
3515 g_assert_not_reached();
3521 if (qcow2_opts
->has_cluster_size
) {
3522 cluster_size
= qcow2_opts
->cluster_size
;
3524 cluster_size
= DEFAULT_CLUSTER_SIZE
;
3527 if (!qcow2_opts
->has_extended_l2
) {
3528 qcow2_opts
->extended_l2
= false;
3530 if (qcow2_opts
->extended_l2
) {
3532 error_setg(errp
, "Extended L2 entries are only supported with "
3533 "compatibility level 1.1 and above (use version=v3 or "
3540 if (!validate_cluster_size(cluster_size
, qcow2_opts
->extended_l2
, errp
)) {
3545 if (!qcow2_opts
->has_preallocation
) {
3546 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
3548 if (qcow2_opts
->backing_file
&&
3549 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
&&
3550 !qcow2_opts
->extended_l2
)
3552 error_setg(errp
, "Backing file and preallocation can only be used at "
3553 "the same time if extended_l2 is on");
3557 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->backing_file
) {
3558 error_setg(errp
, "Backing format cannot be used without backing file");
3563 if (!qcow2_opts
->has_lazy_refcounts
) {
3564 qcow2_opts
->lazy_refcounts
= false;
3566 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
3567 error_setg(errp
, "Lazy refcounts only supported with compatibility "
3568 "level 1.1 and above (use version=v3 or greater)");
3573 if (!qcow2_opts
->has_refcount_bits
) {
3574 qcow2_opts
->refcount_bits
= 16;
3576 if (qcow2_opts
->refcount_bits
> 64 ||
3577 !is_power_of_2(qcow2_opts
->refcount_bits
))
3579 error_setg(errp
, "Refcount width must be a power of two and may not "
3584 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
3585 error_setg(errp
, "Different refcount widths than 16 bits require "
3586 "compatibility level 1.1 or above (use version=v3 or "
3591 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
3593 if (qcow2_opts
->data_file_raw
&& !qcow2_opts
->data_file
) {
3594 error_setg(errp
, "data-file-raw requires data-file");
3598 if (qcow2_opts
->data_file_raw
&& qcow2_opts
->backing_file
) {
3599 error_setg(errp
, "Backing file and data-file-raw cannot be used at "
3604 if (qcow2_opts
->data_file_raw
&&
3605 qcow2_opts
->preallocation
== PREALLOC_MODE_OFF
)
3608 * data-file-raw means that "the external data file can be
3609 * read as a consistent standalone raw image without looking
3610 * at the qcow2 metadata." It does not say that the metadata
3611 * must be ignored, though (and the qcow2 driver in fact does
3612 * not ignore it), so the L1/L2 tables must be present and
3613 * give a 1:1 mapping, so you get the same result regardless
3614 * of whether you look at the metadata or whether you ignore
3617 qcow2_opts
->preallocation
= PREALLOC_MODE_METADATA
;
3620 * Cannot use preallocation with backing files, but giving a
3621 * backing file when specifying data_file_raw is an error
3624 assert(!qcow2_opts
->backing_file
);
3627 if (qcow2_opts
->data_file
) {
3629 error_setg(errp
, "External data files are only supported with "
3630 "compatibility level 1.1 and above (use version=v3 or "
3635 data_bs
= bdrv_co_open_blockdev_ref(qcow2_opts
->data_file
, errp
);
3636 if (data_bs
== NULL
) {
3642 if (qcow2_opts
->has_compression_type
&&
3643 qcow2_opts
->compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3648 error_setg(errp
, "Non-zlib compression type is only supported with "
3649 "compatibility level 1.1 and above (use version=v3 or "
3654 switch (qcow2_opts
->compression_type
) {
3656 case QCOW2_COMPRESSION_TYPE_ZSTD
:
3660 error_setg(errp
, "Unknown compression type");
3664 compression_type
= qcow2_opts
->compression_type
;
3667 /* Create BlockBackend to write to the image */
3668 blk
= blk_co_new_with_bs(bs
, BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
,
3674 blk_set_allow_write_beyond_eof(blk
, true);
3676 /* Write the header */
3677 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
3678 header
= g_malloc0(cluster_size
);
3679 *header
= (QCowHeader
) {
3680 .magic
= cpu_to_be32(QCOW_MAGIC
),
3681 .version
= cpu_to_be32(version
),
3682 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
3683 .size
= cpu_to_be64(0),
3684 .l1_table_offset
= cpu_to_be64(0),
3685 .l1_size
= cpu_to_be32(0),
3686 .refcount_table_offset
= cpu_to_be64(cluster_size
),
3687 .refcount_table_clusters
= cpu_to_be32(1),
3688 .refcount_order
= cpu_to_be32(refcount_order
),
3689 /* don't deal with endianness since compression_type is 1 byte long */
3690 .compression_type
= compression_type
,
3691 .header_length
= cpu_to_be32(sizeof(*header
)),
3694 /* We'll update this to correct value later */
3695 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
3697 if (qcow2_opts
->lazy_refcounts
) {
3698 header
->compatible_features
|=
3699 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
3702 header
->incompatible_features
|=
3703 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE
);
3705 if (qcow2_opts
->data_file_raw
) {
3706 header
->autoclear_features
|=
3707 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW
);
3709 if (compression_type
!= QCOW2_COMPRESSION_TYPE_ZLIB
) {
3710 header
->incompatible_features
|=
3711 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION
);
3714 if (qcow2_opts
->extended_l2
) {
3715 header
->incompatible_features
|=
3716 cpu_to_be64(QCOW2_INCOMPAT_EXTL2
);
3719 ret
= blk_co_pwrite(blk
, 0, cluster_size
, header
, 0);
3722 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
3726 /* Write a refcount table with one refcount block */
3727 refcount_table
= g_malloc0(2 * cluster_size
);
3728 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
3729 ret
= blk_co_pwrite(blk
, cluster_size
, 2 * cluster_size
, refcount_table
, 0);
3730 g_free(refcount_table
);
3733 error_setg_errno(errp
, -ret
, "Could not write refcount table");
3741 * And now open the image and make it consistent first (i.e. increase the
3742 * refcount of the cluster that is occupied by the header and the refcount
3745 options
= qdict_new();
3746 qdict_put_str(options
, "driver", "qcow2");
3747 qdict_put_str(options
, "file", bs
->node_name
);
3749 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3751 blk
= blk_co_new_open(NULL
, NULL
, options
,
3752 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
3759 bdrv_graph_co_rdlock();
3760 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
3762 bdrv_graph_co_rdunlock();
3763 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
3764 "header and refcount table");
3767 } else if (ret
!= 0) {
3768 error_report("Huh, first cluster in empty image is already in use?");
3772 /* Set the external data file if necessary */
3774 BDRVQcow2State
*s
= blk_bs(blk
)->opaque
;
3775 s
->image_data_file
= g_strdup(data_bs
->filename
);
3778 /* Create a full header (including things like feature table) */
3779 ret
= qcow2_update_header(blk_bs(blk
));
3780 bdrv_graph_co_rdunlock();
3783 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
3787 /* Okay, now that we have a valid image, let's give it the right size */
3788 ret
= blk_co_truncate(blk
, qcow2_opts
->size
, false,
3789 qcow2_opts
->preallocation
, 0, errp
);
3791 error_prepend(errp
, "Could not resize image: ");
3795 /* Want a backing file? There you go. */
3796 if (qcow2_opts
->backing_file
) {
3797 const char *backing_format
= NULL
;
3799 if (qcow2_opts
->has_backing_fmt
) {
3800 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
3803 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
3804 backing_format
, false);
3806 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
3807 "with format '%s'", qcow2_opts
->backing_file
,
3813 /* Want encryption? There you go. */
3814 if (qcow2_opts
->encrypt
) {
3815 bdrv_graph_co_rdlock();
3816 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3817 bdrv_graph_co_rdunlock();
3827 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3828 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3829 * have to setup decryption context. We're not doing any I/O on the top
3830 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3833 options
= qdict_new();
3834 qdict_put_str(options
, "driver", "qcow2");
3835 qdict_put_str(options
, "file", bs
->node_name
);
3837 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3839 blk
= blk_co_new_open(NULL
, NULL
, options
,
3840 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3851 bdrv_co_unref(data_bs
);
3855 static int coroutine_fn GRAPH_UNLOCKED
3856 qcow2_co_create_opts(BlockDriver
*drv
, const char *filename
, QemuOpts
*opts
,
3859 BlockdevCreateOptions
*create_options
= NULL
;
3862 BlockDriverState
*bs
= NULL
;
3863 BlockDriverState
*data_bs
= NULL
;
3867 /* Only the keyval visitor supports the dotted syntax needed for
3868 * encryption, so go through a QDict before getting a QAPI type. Ignore
3869 * options meant for the protocol layer so that the visitor doesn't
3871 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3874 /* Handle encryption options */
3875 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3876 if (val
&& !strcmp(val
, "on")) {
3877 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3878 } else if (val
&& !strcmp(val
, "off")) {
3879 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3882 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3883 if (val
&& !strcmp(val
, "aes")) {
3884 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3887 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3888 * version=v2/v3 below. */
3889 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3890 if (val
&& !strcmp(val
, "0.10")) {
3891 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3892 } else if (val
&& !strcmp(val
, "1.1")) {
3893 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3896 /* Change legacy command line options into QMP ones */
3897 static const QDictRenames opt_renames
[] = {
3898 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3899 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3900 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3901 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3902 { BLOCK_OPT_EXTL2
, "extended-l2" },
3903 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3904 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3905 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3906 { BLOCK_OPT_DATA_FILE_RAW
, "data-file-raw" },
3907 { BLOCK_OPT_COMPRESSION_TYPE
, "compression-type" },
3911 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3916 /* Create and open the file (protocol layer) */
3917 ret
= bdrv_co_create_file(filename
, opts
, errp
);
3922 bs
= bdrv_co_open(filename
, NULL
, NULL
,
3923 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3929 /* Create and open an external data file (protocol layer) */
3930 val
= qdict_get_try_str(qdict
, BLOCK_OPT_DATA_FILE
);
3932 ret
= bdrv_co_create_file(val
, opts
, errp
);
3937 data_bs
= bdrv_co_open(val
, NULL
, NULL
,
3938 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
,
3940 if (data_bs
== NULL
) {
3945 qdict_del(qdict
, BLOCK_OPT_DATA_FILE
);
3946 qdict_put_str(qdict
, "data-file", data_bs
->node_name
);
3949 /* Set 'driver' and 'node' options */
3950 qdict_put_str(qdict
, "driver", "qcow2");
3951 qdict_put_str(qdict
, "file", bs
->node_name
);
3953 /* Now get the QAPI type BlockdevCreateOptions */
3954 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
3960 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, errp
);
3962 if (!create_options
) {
3967 /* Silently round up size */
3968 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3971 /* Create the qcow2 image (format layer) */
3972 ret
= qcow2_co_create(create_options
, errp
);
3975 bdrv_graph_co_rdlock();
3976 bdrv_co_delete_file_noerr(bs
);
3977 bdrv_co_delete_file_noerr(data_bs
);
3978 bdrv_graph_co_rdunlock();
3983 qobject_unref(qdict
);
3985 bdrv_co_unref(data_bs
);
3986 qapi_free_BlockdevCreateOptions(create_options
);
3991 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3996 /* Clamp to image length, before checking status of underlying sectors */
3997 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3998 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
4006 * bdrv_block_status_above doesn't merge different types of zeros, for
4007 * example, zeros which come from the region which is unallocated in
4008 * the whole backing chain, and zeros which come because of a short
4009 * backing file. So, we need a loop.
4012 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
4015 } while (res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
&& bytes
);
4017 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && bytes
== 0;
4020 static int coroutine_fn GRAPH_RDLOCK
4021 qcow2_co_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
4022 BdrvRequestFlags flags
)
4025 BDRVQcow2State
*s
= bs
->opaque
;
4027 uint32_t head
= offset_into_subcluster(s
, offset
);
4028 uint32_t tail
= ROUND_UP(offset
+ bytes
, s
->subcluster_size
) -
4031 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
4032 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
4039 QCow2SubclusterType type
;
4041 assert(head
+ bytes
+ tail
<= s
->subcluster_size
);
4043 /* check whether remainder of cluster already reads as zero */
4044 if (!(is_zero(bs
, offset
- head
, head
) &&
4045 is_zero(bs
, offset
+ bytes
, tail
))) {
4049 qemu_co_mutex_lock(&s
->lock
);
4050 /* We can have new write after previous check */
4052 bytes
= s
->subcluster_size
;
4053 nr
= s
->subcluster_size
;
4054 ret
= qcow2_get_host_offset(bs
, offset
, &nr
, &off
, &type
);
4056 (type
!= QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
&&
4057 type
!= QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
&&
4058 type
!= QCOW2_SUBCLUSTER_ZERO_PLAIN
&&
4059 type
!= QCOW2_SUBCLUSTER_ZERO_ALLOC
)) {
4060 qemu_co_mutex_unlock(&s
->lock
);
4061 return ret
< 0 ? ret
: -ENOTSUP
;
4064 qemu_co_mutex_lock(&s
->lock
);
4067 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
4069 /* Whatever is left can use real zero subclusters */
4070 ret
= qcow2_subcluster_zeroize(bs
, offset
, bytes
, flags
);
4071 qemu_co_mutex_unlock(&s
->lock
);
4076 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
4077 int64_t offset
, int64_t bytes
)
4080 BDRVQcow2State
*s
= bs
->opaque
;
4082 /* If the image does not support QCOW_OFLAG_ZERO then discarding
4083 * clusters could expose stale data from the backing file. */
4084 if (s
->qcow_version
< 3 && bs
->backing
) {
4088 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
4089 assert(bytes
< s
->cluster_size
);
4090 /* Ignore partial clusters, except for the special case of the
4091 * complete partial cluster at the end of an unaligned file */
4092 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
4093 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
4098 qemu_co_mutex_lock(&s
->lock
);
4099 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
4101 qemu_co_mutex_unlock(&s
->lock
);
4105 static int coroutine_fn GRAPH_RDLOCK
4106 qcow2_co_copy_range_from(BlockDriverState
*bs
,
4107 BdrvChild
*src
, int64_t src_offset
,
4108 BdrvChild
*dst
, int64_t dst_offset
,
4109 int64_t bytes
, BdrvRequestFlags read_flags
,
4110 BdrvRequestFlags write_flags
)
4112 BDRVQcow2State
*s
= bs
->opaque
;
4114 unsigned int cur_bytes
; /* number of bytes in current iteration */
4115 BdrvChild
*child
= NULL
;
4116 BdrvRequestFlags cur_write_flags
;
4118 assert(!bs
->encrypted
);
4119 qemu_co_mutex_lock(&s
->lock
);
4121 while (bytes
!= 0) {
4122 uint64_t copy_offset
= 0;
4123 QCow2SubclusterType type
;
4124 /* prepare next request */
4125 cur_bytes
= MIN(bytes
, INT_MAX
);
4126 cur_write_flags
= write_flags
;
4128 ret
= qcow2_get_host_offset(bs
, src_offset
, &cur_bytes
,
4129 ©_offset
, &type
);
4135 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN
:
4136 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC
:
4137 if (bs
->backing
&& bs
->backing
->bs
) {
4138 int64_t backing_length
= bdrv_co_getlength(bs
->backing
->bs
);
4139 if (src_offset
>= backing_length
) {
4140 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4142 child
= bs
->backing
;
4143 cur_bytes
= MIN(cur_bytes
, backing_length
- src_offset
);
4144 copy_offset
= src_offset
;
4147 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4151 case QCOW2_SUBCLUSTER_ZERO_PLAIN
:
4152 case QCOW2_SUBCLUSTER_ZERO_ALLOC
:
4153 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
4156 case QCOW2_SUBCLUSTER_COMPRESSED
:
4160 case QCOW2_SUBCLUSTER_NORMAL
:
4161 child
= s
->data_file
;
4167 qemu_co_mutex_unlock(&s
->lock
);
4168 ret
= bdrv_co_copy_range_from(child
,
4171 cur_bytes
, read_flags
, cur_write_flags
);
4172 qemu_co_mutex_lock(&s
->lock
);
4178 src_offset
+= cur_bytes
;
4179 dst_offset
+= cur_bytes
;
4184 qemu_co_mutex_unlock(&s
->lock
);
4188 static int coroutine_fn GRAPH_RDLOCK
4189 qcow2_co_copy_range_to(BlockDriverState
*bs
,
4190 BdrvChild
*src
, int64_t src_offset
,
4191 BdrvChild
*dst
, int64_t dst_offset
,
4192 int64_t bytes
, BdrvRequestFlags read_flags
,
4193 BdrvRequestFlags write_flags
)
4195 BDRVQcow2State
*s
= bs
->opaque
;
4197 unsigned int cur_bytes
; /* number of sectors in current iteration */
4198 uint64_t host_offset
;
4199 QCowL2Meta
*l2meta
= NULL
;
4201 assert(!bs
->encrypted
);
4203 qemu_co_mutex_lock(&s
->lock
);
4205 while (bytes
!= 0) {
4209 cur_bytes
= MIN(bytes
, INT_MAX
);
4212 * If src->bs == dst->bs, we could simply copy by incrementing
4213 * the refcnt, without copying user data.
4214 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4215 ret
= qcow2_alloc_host_offset(bs
, dst_offset
, &cur_bytes
,
4216 &host_offset
, &l2meta
);
4221 ret
= qcow2_pre_write_overlap_check(bs
, 0, host_offset
, cur_bytes
,
4227 qemu_co_mutex_unlock(&s
->lock
);
4228 ret
= bdrv_co_copy_range_to(src
, src_offset
, s
->data_file
, host_offset
,
4229 cur_bytes
, read_flags
, write_flags
);
4230 qemu_co_mutex_lock(&s
->lock
);
4235 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
4241 src_offset
+= cur_bytes
;
4242 dst_offset
+= cur_bytes
;
4247 qcow2_handle_l2meta(bs
, &l2meta
, false);
4249 qemu_co_mutex_unlock(&s
->lock
);
4251 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
4256 static int coroutine_fn GRAPH_RDLOCK
4257 qcow2_co_truncate(BlockDriverState
*bs
, int64_t offset
, bool exact
,
4258 PreallocMode prealloc
, BdrvRequestFlags flags
, Error
**errp
)
4260 BDRVQcow2State
*s
= bs
->opaque
;
4261 uint64_t old_length
;
4262 int64_t new_l1_size
;
4266 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
4267 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
4269 error_setg(errp
, "Unsupported preallocation mode '%s'",
4270 PreallocMode_str(prealloc
));
4274 if (!QEMU_IS_ALIGNED(offset
, BDRV_SECTOR_SIZE
)) {
4275 error_setg(errp
, "The new size must be a multiple of %u",
4276 (unsigned) BDRV_SECTOR_SIZE
);
4280 qemu_co_mutex_lock(&s
->lock
);
4283 * Even though we store snapshot size for all images, it was not
4284 * required until v3, so it is not safe to proceed for v2.
4286 if (s
->nb_snapshots
&& s
->qcow_version
< 3) {
4287 error_setg(errp
, "Can't resize a v2 image which has snapshots");
4292 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4293 if (qcow2_truncate_bitmaps_check(bs
, errp
)) {
4298 old_length
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4299 new_l1_size
= size_to_l1(s
, offset
);
4301 if (offset
< old_length
) {
4302 int64_t last_cluster
, old_file_size
;
4303 if (prealloc
!= PREALLOC_MODE_OFF
) {
4305 "Preallocation can't be used for shrinking an image");
4310 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
4311 old_length
- ROUND_UP(offset
,
4313 QCOW2_DISCARD_ALWAYS
, true);
4315 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
4319 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
4321 error_setg_errno(errp
, -ret
,
4322 "Failed to reduce the number of L2 tables");
4326 ret
= qcow2_shrink_reftable(bs
);
4328 error_setg_errno(errp
, -ret
,
4329 "Failed to discard unused refblocks");
4333 old_file_size
= bdrv_co_getlength(bs
->file
->bs
);
4334 if (old_file_size
< 0) {
4335 error_setg_errno(errp
, -old_file_size
,
4336 "Failed to inquire current file length");
4337 ret
= old_file_size
;
4340 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4341 if (last_cluster
< 0) {
4342 error_setg_errno(errp
, -last_cluster
,
4343 "Failed to find the last cluster");
4347 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
4348 Error
*local_err
= NULL
;
4351 * Do not pass @exact here: It will not help the user if
4352 * we get an error here just because they wanted to shrink
4353 * their qcow2 image (on a block device) with qemu-img.
4354 * (And on the qcow2 layer, the @exact requirement is
4355 * always fulfilled, so there is no need to pass it on.)
4357 bdrv_co_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
4358 false, PREALLOC_MODE_OFF
, 0, &local_err
);
4360 warn_reportf_err(local_err
,
4361 "Failed to truncate the tail of the image: ");
4365 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
4367 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
4371 if (data_file_is_raw(bs
) && prealloc
== PREALLOC_MODE_OFF
) {
4373 * When creating a qcow2 image with data-file-raw, we enforce
4374 * at least prealloc=metadata, so that the L1/L2 tables are
4375 * fully allocated and reading from the data file will return
4376 * the same data as reading from the qcow2 image. When the
4377 * image is grown, we must consequently preallocate the
4378 * metadata structures to cover the added area.
4380 prealloc
= PREALLOC_MODE_METADATA
;
4385 case PREALLOC_MODE_OFF
:
4386 if (has_data_file(bs
)) {
4388 * If the caller wants an exact resize, the external data
4389 * file should be resized to the exact target size, too,
4390 * so we pass @exact here.
4392 ret
= bdrv_co_truncate(s
->data_file
, offset
, exact
, prealloc
, 0,
4400 case PREALLOC_MODE_METADATA
:
4401 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4407 case PREALLOC_MODE_FALLOC
:
4408 case PREALLOC_MODE_FULL
:
4410 int64_t allocation_start
, host_offset
, guest_offset
;
4411 int64_t clusters_allocated
;
4412 int64_t old_file_size
, last_cluster
, new_file_size
;
4413 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
4414 bool subclusters_need_allocation
= false;
4416 /* With a data file, preallocation means just allocating the metadata
4417 * and forwarding the truncate request to the data file */
4418 if (has_data_file(bs
)) {
4419 ret
= preallocate_co(bs
, old_length
, offset
, prealloc
, errp
);
4426 old_file_size
= bdrv_co_getlength(bs
->file
->bs
);
4427 if (old_file_size
< 0) {
4428 error_setg_errno(errp
, -old_file_size
,
4429 "Failed to inquire current file length");
4430 ret
= old_file_size
;
4434 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
4435 if (last_cluster
>= 0) {
4436 old_file_size
= (last_cluster
+ 1) * s
->cluster_size
;
4438 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
4441 nb_new_data_clusters
= (ROUND_UP(offset
, s
->cluster_size
) -
4442 start_of_cluster(s
, old_length
)) >> s
->cluster_bits
;
4444 /* This is an overestimation; we will not actually allocate space for
4445 * these in the file but just make sure the new refcount structures are
4446 * able to cover them so we will not have to allocate new refblocks
4447 * while entering the data blocks in the potentially new L2 tables.
4448 * (We do not actually care where the L2 tables are placed. Maybe they
4449 * are already allocated or they can be placed somewhere before
4450 * @old_file_size. It does not matter because they will be fully
4451 * allocated automatically, so they do not need to be covered by the
4452 * preallocation. All that matters is that we will not have to allocate
4453 * new refcount structures for them.) */
4454 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
4455 s
->cluster_size
/ l2_entry_size(s
));
4456 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4457 * table for a potential head/tail */
4460 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
4461 nb_new_data_clusters
+
4464 if (allocation_start
< 0) {
4465 error_setg_errno(errp
, -allocation_start
,
4466 "Failed to resize refcount structures");
4467 ret
= allocation_start
;
4471 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
4472 nb_new_data_clusters
);
4473 if (clusters_allocated
< 0) {
4474 error_setg_errno(errp
, -clusters_allocated
,
4475 "Failed to allocate data clusters");
4476 ret
= clusters_allocated
;
4480 assert(clusters_allocated
== nb_new_data_clusters
);
4482 /* Allocate the data area */
4483 new_file_size
= allocation_start
+
4484 nb_new_data_clusters
* s
->cluster_size
;
4486 * Image file grows, so @exact does not matter.
4488 * If we need to zero out the new area, try first whether the protocol
4489 * driver can already take care of this.
4491 if (flags
& BDRV_REQ_ZERO_WRITE
) {
4492 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
,
4493 BDRV_REQ_ZERO_WRITE
, NULL
);
4495 flags
&= ~BDRV_REQ_ZERO_WRITE
;
4496 /* Ensure that we read zeroes and not backing file data */
4497 subclusters_need_allocation
= true;
4503 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, false, prealloc
, 0,
4507 error_prepend(errp
, "Failed to resize underlying file: ");
4508 qcow2_free_clusters(bs
, allocation_start
,
4509 nb_new_data_clusters
* s
->cluster_size
,
4510 QCOW2_DISCARD_OTHER
);
4514 /* Create the necessary L2 entries */
4515 host_offset
= allocation_start
;
4516 guest_offset
= old_length
;
4517 while (nb_new_data_clusters
) {
4518 int64_t nb_clusters
= MIN(
4519 nb_new_data_clusters
,
4520 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
4521 unsigned cow_start_length
= offset_into_cluster(s
, guest_offset
);
4522 QCowL2Meta allocation
;
4523 guest_offset
= start_of_cluster(s
, guest_offset
);
4524 allocation
= (QCowL2Meta
) {
4525 .offset
= guest_offset
,
4526 .alloc_offset
= host_offset
,
4527 .nb_clusters
= nb_clusters
,
4530 .nb_bytes
= cow_start_length
,
4533 .offset
= nb_clusters
<< s
->cluster_bits
,
4536 .prealloc
= !subclusters_need_allocation
,
4538 qemu_co_queue_init(&allocation
.dependent_requests
);
4540 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
4542 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
4543 qcow2_free_clusters(bs
, host_offset
,
4544 nb_new_data_clusters
* s
->cluster_size
,
4545 QCOW2_DISCARD_OTHER
);
4549 guest_offset
+= nb_clusters
* s
->cluster_size
;
4550 host_offset
+= nb_clusters
* s
->cluster_size
;
4551 nb_new_data_clusters
-= nb_clusters
;
4557 g_assert_not_reached();
4560 if ((flags
& BDRV_REQ_ZERO_WRITE
) && offset
> old_length
) {
4561 uint64_t zero_start
= QEMU_ALIGN_UP(old_length
, s
->subcluster_size
);
4564 * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4565 * requires a subcluster-aligned start. The end may be unaligned if
4566 * it is at the end of the image (which it is here).
4568 if (offset
> zero_start
) {
4569 ret
= qcow2_subcluster_zeroize(bs
, zero_start
, offset
- zero_start
,
4572 error_setg_errno(errp
, -ret
, "Failed to zero out new clusters");
4577 /* Write explicit zeros for the unaligned head */
4578 if (zero_start
> old_length
) {
4579 uint64_t len
= MIN(zero_start
, offset
) - old_length
;
4580 uint8_t *buf
= qemu_blockalign0(bs
, len
);
4582 qemu_iovec_init_buf(&qiov
, buf
, len
);
4584 qemu_co_mutex_unlock(&s
->lock
);
4585 ret
= qcow2_co_pwritev_part(bs
, old_length
, len
, &qiov
, 0, 0);
4586 qemu_co_mutex_lock(&s
->lock
);
4590 error_setg_errno(errp
, -ret
, "Failed to zero out the new area");
4596 if (prealloc
!= PREALLOC_MODE_OFF
) {
4597 /* Flush metadata before actually changing the image size */
4598 ret
= qcow2_write_caches(bs
);
4600 error_setg_errno(errp
, -ret
,
4601 "Failed to flush the preallocated area to disk");
4606 bs
->total_sectors
= offset
/ BDRV_SECTOR_SIZE
;
4608 /* write updated header.size */
4609 offset
= cpu_to_be64(offset
);
4610 ret
= bdrv_co_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
4611 sizeof(offset
), &offset
, 0);
4613 error_setg_errno(errp
, -ret
, "Failed to update the image size");
4617 s
->l1_vm_state_index
= new_l1_size
;
4619 /* Update cache sizes */
4620 options
= qdict_clone_shallow(bs
->options
);
4621 ret
= qcow2_update_options(bs
, options
, s
->flags
, errp
);
4622 qobject_unref(options
);
4628 qemu_co_mutex_unlock(&s
->lock
);
4632 static int coroutine_fn GRAPH_RDLOCK
4633 qcow2_co_pwritev_compressed_task(BlockDriverState
*bs
,
4634 uint64_t offset
, uint64_t bytes
,
4635 QEMUIOVector
*qiov
, size_t qiov_offset
)
4637 BDRVQcow2State
*s
= bs
->opaque
;
4640 uint8_t *buf
, *out_buf
;
4641 uint64_t cluster_offset
;
4643 assert(bytes
== s
->cluster_size
|| (bytes
< s
->cluster_size
&&
4644 (offset
+ bytes
== bs
->total_sectors
<< BDRV_SECTOR_BITS
)));
4646 buf
= qemu_blockalign(bs
, s
->cluster_size
);
4647 if (bytes
< s
->cluster_size
) {
4648 /* Zero-pad last write if image size is not cluster aligned */
4649 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
4651 qemu_iovec_to_buf(qiov
, qiov_offset
, buf
, bytes
);
4653 out_buf
= g_malloc(s
->cluster_size
);
4655 out_len
= qcow2_co_compress(bs
, out_buf
, s
->cluster_size
- 1,
4656 buf
, s
->cluster_size
);
4657 if (out_len
== -ENOMEM
) {
4658 /* could not compress: write normal cluster */
4659 ret
= qcow2_co_pwritev_part(bs
, offset
, bytes
, qiov
, qiov_offset
, 0);
4664 } else if (out_len
< 0) {
4669 qemu_co_mutex_lock(&s
->lock
);
4670 ret
= qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
,
4673 qemu_co_mutex_unlock(&s
->lock
);
4677 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
, true);
4678 qemu_co_mutex_unlock(&s
->lock
);
4683 BLKDBG_CO_EVENT(s
->data_file
, BLKDBG_WRITE_COMPRESSED
);
4684 ret
= bdrv_co_pwrite(s
->data_file
, cluster_offset
, out_len
, out_buf
, 0);
4697 * This function can count as GRAPH_RDLOCK because
4698 * qcow2_co_pwritev_compressed_part() holds the graph lock and keeps it until
4699 * this coroutine has terminated.
4701 static int coroutine_fn GRAPH_RDLOCK
4702 qcow2_co_pwritev_compressed_task_entry(AioTask
*task
)
4704 Qcow2AioTask
*t
= container_of(task
, Qcow2AioTask
, task
);
4706 assert(!t
->subcluster_type
&& !t
->l2meta
);
4708 return qcow2_co_pwritev_compressed_task(t
->bs
, t
->offset
, t
->bytes
, t
->qiov
,
4713 * XXX: put compressed sectors first, then all the cluster aligned
4714 * tables to avoid losing bytes in alignment
4716 static int coroutine_fn GRAPH_RDLOCK
4717 qcow2_co_pwritev_compressed_part(BlockDriverState
*bs
,
4718 int64_t offset
, int64_t bytes
,
4719 QEMUIOVector
*qiov
, size_t qiov_offset
)
4721 BDRVQcow2State
*s
= bs
->opaque
;
4722 AioTaskPool
*aio
= NULL
;
4725 if (has_data_file(bs
)) {
4731 * align end of file to a sector boundary to ease reading with
4734 int64_t len
= bdrv_co_getlength(bs
->file
->bs
);
4738 return bdrv_co_truncate(bs
->file
, len
, false, PREALLOC_MODE_OFF
, 0,
4742 if (offset_into_cluster(s
, offset
)) {
4746 if (offset_into_cluster(s
, bytes
) &&
4747 (offset
+ bytes
) != (bs
->total_sectors
<< BDRV_SECTOR_BITS
)) {
4751 while (bytes
&& aio_task_pool_status(aio
) == 0) {
4752 uint64_t chunk_size
= MIN(bytes
, s
->cluster_size
);
4754 if (!aio
&& chunk_size
!= bytes
) {
4755 aio
= aio_task_pool_new(QCOW2_MAX_WORKERS
);
4758 ret
= qcow2_add_task(bs
, aio
, qcow2_co_pwritev_compressed_task_entry
,
4759 0, 0, offset
, chunk_size
, qiov
, qiov_offset
, NULL
);
4763 qiov_offset
+= chunk_size
;
4764 offset
+= chunk_size
;
4765 bytes
-= chunk_size
;
4769 aio_task_pool_wait_all(aio
);
4771 ret
= aio_task_pool_status(aio
);
4779 static int coroutine_fn GRAPH_RDLOCK
4780 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
4787 BDRVQcow2State
*s
= bs
->opaque
;
4790 uint8_t *buf
, *out_buf
;
4791 int offset_in_cluster
= offset_into_cluster(s
, offset
);
4793 qcow2_parse_compressed_l2_entry(bs
, l2_entry
, &coffset
, &csize
);
4795 buf
= g_try_malloc(csize
);
4800 out_buf
= qemu_blockalign(bs
, s
->cluster_size
);
4802 BLKDBG_CO_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
4803 ret
= bdrv_co_pread(bs
->file
, coffset
, csize
, buf
, 0);
4808 if (qcow2_co_decompress(bs
, out_buf
, s
->cluster_size
, buf
, csize
) < 0) {
4813 qemu_iovec_from_buf(qiov
, qiov_offset
, out_buf
+ offset_in_cluster
, bytes
);
4816 qemu_vfree(out_buf
);
4822 static int make_completely_empty(BlockDriverState
*bs
)
4824 BDRVQcow2State
*s
= bs
->opaque
;
4825 Error
*local_err
= NULL
;
4826 int ret
, l1_clusters
;
4828 uint64_t *new_reftable
= NULL
;
4829 uint64_t rt_entry
, l1_size2
;
4832 uint64_t reftable_offset
;
4833 uint32_t reftable_clusters
;
4834 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
4836 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
4841 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
4846 /* Refcounts will be broken utterly */
4847 ret
= qcow2_mark_dirty(bs
);
4852 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4854 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ L1E_SIZE
);
4855 l1_size2
= (uint64_t)s
->l1_size
* L1E_SIZE
;
4857 /* After this call, neither the in-memory nor the on-disk refcount
4858 * information accurately describe the actual references */
4860 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
4861 l1_clusters
* s
->cluster_size
, 0);
4863 goto fail_broken_refcounts
;
4865 memset(s
->l1_table
, 0, l1_size2
);
4867 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
4869 /* Overwrite enough clusters at the beginning of the sectors to place
4870 * the refcount table, a refcount block and the L1 table in; this may
4871 * overwrite parts of the existing refcount and L1 table, which is not
4872 * an issue because the dirty flag is set, complete data loss is in fact
4873 * desired and partial data loss is consequently fine as well */
4874 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
4875 (2 + l1_clusters
) * s
->cluster_size
, 0);
4876 /* This call (even if it failed overall) may have overwritten on-disk
4877 * refcount structures; in that case, the in-memory refcount information
4878 * will probably differ from the on-disk information which makes the BDS
4881 goto fail_broken_refcounts
;
4884 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4885 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
4887 /* "Create" an empty reftable (one cluster) directly after the image
4888 * header and an empty L1 table three clusters after the image header;
4889 * the cluster between those two will be used as the first refblock */
4890 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
4891 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
4892 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
4893 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
4894 sizeof(l1_ofs_rt_ofs_cls
), &l1_ofs_rt_ofs_cls
, 0);
4896 goto fail_broken_refcounts
;
4899 s
->l1_table_offset
= 3 * s
->cluster_size
;
4901 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ REFTABLE_ENTRY_SIZE
);
4902 if (!new_reftable
) {
4904 goto fail_broken_refcounts
;
4907 s
->refcount_table_offset
= s
->cluster_size
;
4908 s
->refcount_table_size
= s
->cluster_size
/ REFTABLE_ENTRY_SIZE
;
4909 s
->max_refcount_table_index
= 0;
4911 g_free(s
->refcount_table
);
4912 s
->refcount_table
= new_reftable
;
4913 new_reftable
= NULL
;
4915 /* Now the in-memory refcount information again corresponds to the on-disk
4916 * information (reftable is empty and no refblocks (the refblock cache is
4917 * empty)); however, this means some clusters (e.g. the image header) are
4918 * referenced, but not refcounted, but the normal qcow2 code assumes that
4919 * the in-memory information is always correct */
4921 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
4923 /* Enter the first refblock into the reftable */
4924 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
4925 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
, sizeof(rt_entry
),
4928 goto fail_broken_refcounts
;
4930 s
->refcount_table
[0] = 2 * s
->cluster_size
;
4932 s
->free_cluster_index
= 0;
4933 assert(3 + l1_clusters
<= s
->refcount_block_size
);
4934 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
4937 goto fail_broken_refcounts
;
4938 } else if (offset
> 0) {
4939 error_report("First cluster in emptied image is in use");
4943 /* Now finally the in-memory information corresponds to the on-disk
4944 * structures and is correct */
4945 ret
= qcow2_mark_clean(bs
);
4950 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
, false,
4951 PREALLOC_MODE_OFF
, 0, &local_err
);
4953 error_report_err(local_err
);
4959 fail_broken_refcounts
:
4960 /* The BDS is unusable at this point. If we wanted to make it usable, we
4961 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4962 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4963 * again. However, because the functions which could have caused this error
4964 * path to be taken are used by those functions as well, it's very likely
4965 * that that sequence will fail as well. Therefore, just eject the BDS. */
4969 g_free(new_reftable
);
4973 static int qcow2_make_empty(BlockDriverState
*bs
)
4975 BDRVQcow2State
*s
= bs
->opaque
;
4976 uint64_t offset
, end_offset
;
4977 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
4978 int l1_clusters
, ret
= 0;
4980 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ L1E_SIZE
);
4982 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
4983 3 + l1_clusters
<= s
->refcount_block_size
&&
4984 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
&&
4985 !has_data_file(bs
)) {
4986 /* The following function only works for qcow2 v3 images (it
4987 * requires the dirty flag) and only as long as there are no
4988 * features that reserve extra clusters (such as snapshots,
4989 * LUKS header, or persistent bitmaps), because it completely
4990 * empties the image. Furthermore, the L1 table and three
4991 * additional clusters (image header, refcount table, one
4992 * refcount block) have to fit inside one refcount block. It
4993 * only resets the image file, i.e. does not work with an
4994 * external data file. */
4995 return make_completely_empty(bs
);
4998 /* This fallback code simply discards every active cluster; this is slow,
4999 * but works in all cases */
5000 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
5001 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
5002 /* As this function is generally used after committing an external
5003 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
5004 * default action for this kind of discard is to pass the discard,
5005 * which will ideally result in an actually smaller image file, as
5006 * is probably desired. */
5007 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
5008 QCOW2_DISCARD_SNAPSHOT
, true);
5017 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
5019 BDRVQcow2State
*s
= bs
->opaque
;
5022 qemu_co_mutex_lock(&s
->lock
);
5023 ret
= qcow2_write_caches(bs
);
5024 qemu_co_mutex_unlock(&s
->lock
);
5029 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
5032 Error
*local_err
= NULL
;
5033 BlockMeasureInfo
*info
;
5034 uint64_t required
= 0; /* bytes that contribute to required size */
5035 uint64_t virtual_size
; /* disk size as seen by guest */
5036 uint64_t refcount_bits
;
5038 uint64_t luks_payload_size
= 0;
5039 size_t cluster_size
;
5042 PreallocMode prealloc
;
5043 bool has_backing_file
;
5048 /* Parse image creation options */
5049 extended_l2
= qemu_opt_get_bool_del(opts
, BLOCK_OPT_EXTL2
, false);
5051 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, extended_l2
,
5057 version
= qcow2_opt_get_version_del(opts
, &local_err
);
5062 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
5067 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
5068 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
5069 PREALLOC_MODE_OFF
, &local_err
);
5075 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
5076 has_backing_file
= !!optstr
;
5079 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
5080 has_luks
= optstr
&& strcmp(optstr
, "luks") == 0;
5084 g_autoptr(QCryptoBlockCreateOptions
) create_opts
= NULL
;
5085 QDict
*cryptoopts
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
5088 create_opts
= block_crypto_create_opts_init(cryptoopts
, errp
);
5089 qobject_unref(cryptoopts
);
5094 if (!qcrypto_block_calculate_payload_offset(create_opts
,
5101 luks_payload_size
= ROUND_UP(headerlen
, cluster_size
);
5104 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
5105 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
5107 /* Check that virtual disk size is valid */
5108 l2e_size
= extended_l2
? L2E_SIZE_EXTENDED
: L2E_SIZE_NORMAL
;
5109 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
5110 cluster_size
/ l2e_size
);
5111 if (l2_tables
* L1E_SIZE
> QCOW_MAX_L1_SIZE
) {
5112 error_setg(&local_err
, "The image size is too large "
5113 "(try using a larger cluster size)");
5117 /* Account for input image */
5119 int64_t ssize
= bdrv_getlength(in_bs
);
5121 error_setg_errno(&local_err
, -ssize
,
5122 "Unable to get image virtual_size");
5126 virtual_size
= ROUND_UP(ssize
, cluster_size
);
5128 if (has_backing_file
) {
5129 /* We don't how much of the backing chain is shared by the input
5130 * image and the new image file. In the worst case the new image's
5131 * backing file has nothing in common with the input image. Be
5132 * conservative and assume all clusters need to be written.
5134 required
= virtual_size
;
5139 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
5142 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
5143 ssize
- offset
, &pnum
, NULL
,
5146 error_setg_errno(&local_err
, -ret
,
5147 "Unable to get block status");
5151 if (ret
& BDRV_BLOCK_ZERO
) {
5152 /* Skip zero regions (safe with no backing file) */
5153 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
5154 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
5155 /* Extend pnum to end of cluster for next iteration */
5156 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
5158 /* Count clusters we've seen */
5159 required
+= offset
% cluster_size
+ pnum
;
5165 /* Take into account preallocation. Nothing special is needed for
5166 * PREALLOC_MODE_METADATA since metadata is always counted.
5168 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
5169 required
= virtual_size
;
5172 info
= g_new0(BlockMeasureInfo
, 1);
5173 info
->fully_allocated
= luks_payload_size
+
5174 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
5175 ctz32(refcount_bits
), extended_l2
);
5178 * Remove data clusters that are not required. This overestimates the
5179 * required size because metadata needed for the fully allocated file is
5180 * still counted. Show bitmaps only if both source and destination
5181 * would support them.
5183 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
5184 info
->has_bitmaps
= version
>= 3 && in_bs
&&
5185 bdrv_supports_persistent_dirty_bitmap(in_bs
);
5186 if (info
->has_bitmaps
) {
5187 info
->bitmaps
= qcow2_get_persistent_dirty_bitmap_size(in_bs
,
5193 error_propagate(errp
, local_err
);
5197 static int coroutine_fn
5198 qcow2_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5200 BDRVQcow2State
*s
= bs
->opaque
;
5201 bdi
->cluster_size
= s
->cluster_size
;
5202 bdi
->subcluster_size
= s
->subcluster_size
;
5203 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
5204 bdi
->is_dirty
= s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
;
5208 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
,
5211 BDRVQcow2State
*s
= bs
->opaque
;
5212 ImageInfoSpecific
*spec_info
;
5213 QCryptoBlockInfo
*encrypt_info
= NULL
;
5215 if (s
->crypto
!= NULL
) {
5216 encrypt_info
= qcrypto_block_get_info(s
->crypto
, errp
);
5217 if (!encrypt_info
) {
5222 spec_info
= g_new(ImageInfoSpecific
, 1);
5223 *spec_info
= (ImageInfoSpecific
){
5224 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
5225 .u
.qcow2
.data
= g_new0(ImageInfoSpecificQCow2
, 1),
5227 if (s
->qcow_version
== 2) {
5228 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5229 .compat
= g_strdup("0.10"),
5230 .refcount_bits
= s
->refcount_bits
,
5232 } else if (s
->qcow_version
== 3) {
5233 Qcow2BitmapInfoList
*bitmaps
;
5234 if (!qcow2_get_bitmap_info_list(bs
, &bitmaps
, errp
)) {
5235 qapi_free_ImageInfoSpecific(spec_info
);
5236 qapi_free_QCryptoBlockInfo(encrypt_info
);
5239 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
5240 .compat
= g_strdup("1.1"),
5241 .lazy_refcounts
= s
->compatible_features
&
5242 QCOW2_COMPAT_LAZY_REFCOUNTS
,
5243 .has_lazy_refcounts
= true,
5244 .corrupt
= s
->incompatible_features
&
5245 QCOW2_INCOMPAT_CORRUPT
,
5246 .has_corrupt
= true,
5247 .has_extended_l2
= true,
5248 .extended_l2
= has_subclusters(s
),
5249 .refcount_bits
= s
->refcount_bits
,
5250 .has_bitmaps
= !!bitmaps
,
5252 .data_file
= g_strdup(s
->image_data_file
),
5253 .has_data_file_raw
= has_data_file(bs
),
5254 .data_file_raw
= data_file_is_raw(bs
),
5255 .compression_type
= s
->compression_type
,
5258 /* if this assertion fails, this probably means a new version was
5259 * added without having it covered here */
5264 ImageInfoSpecificQCow2Encryption
*qencrypt
=
5265 g_new(ImageInfoSpecificQCow2Encryption
, 1);
5266 switch (encrypt_info
->format
) {
5267 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
5268 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
5270 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
5271 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
5272 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
5277 /* Since we did shallow copy above, erase any pointers
5278 * in the original info */
5279 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
5280 qapi_free_QCryptoBlockInfo(encrypt_info
);
5282 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
5288 static int qcow2_has_zero_init(BlockDriverState
*bs
)
5290 BDRVQcow2State
*s
= bs
->opaque
;
5293 if (qemu_in_coroutine()) {
5294 qemu_co_mutex_lock(&s
->lock
);
5297 * Check preallocation status: Preallocated images have all L2
5298 * tables allocated, nonpreallocated images have none. It is
5299 * therefore enough to check the first one.
5301 preallocated
= s
->l1_size
> 0 && s
->l1_table
[0] != 0;
5302 if (qemu_in_coroutine()) {
5303 qemu_co_mutex_unlock(&s
->lock
);
5306 if (!preallocated
) {
5308 } else if (bs
->encrypted
) {
5311 return bdrv_has_zero_init(s
->data_file
->bs
);
5316 * Check the request to vmstate. On success return
5317 * qcow2_vm_state_offset(bs) + @pos
5319 static int64_t qcow2_check_vmstate_request(BlockDriverState
*bs
,
5320 QEMUIOVector
*qiov
, int64_t pos
)
5322 BDRVQcow2State
*s
= bs
->opaque
;
5323 int64_t vmstate_offset
= qcow2_vm_state_offset(s
);
5326 /* Incoming requests must be OK */
5327 bdrv_check_qiov_request(pos
, qiov
->size
, qiov
, 0, &error_abort
);
5329 if (INT64_MAX
- pos
< vmstate_offset
) {
5333 pos
+= vmstate_offset
;
5334 ret
= bdrv_check_qiov_request(pos
, qiov
->size
, qiov
, 0, NULL
);
5342 static int coroutine_fn GRAPH_RDLOCK
5343 qcow2_co_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
5345 int64_t offset
= qcow2_check_vmstate_request(bs
, qiov
, pos
);
5350 BLKDBG_CO_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
5351 return bs
->drv
->bdrv_co_pwritev_part(bs
, offset
, qiov
->size
, qiov
, 0, 0);
5354 static int coroutine_fn GRAPH_RDLOCK
5355 qcow2_co_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
5357 int64_t offset
= qcow2_check_vmstate_request(bs
, qiov
, pos
);
5362 BLKDBG_CO_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
5363 return bs
->drv
->bdrv_co_preadv_part(bs
, offset
, qiov
->size
, qiov
, 0, 0);
5366 static int qcow2_has_compressed_clusters(BlockDriverState
*bs
)
5369 int64_t bytes
= bdrv_getlength(bs
);
5375 while (bytes
!= 0) {
5377 QCow2SubclusterType type
;
5378 unsigned int cur_bytes
= MIN(INT_MAX
, bytes
);
5379 uint64_t host_offset
;
5381 ret
= qcow2_get_host_offset(bs
, offset
, &cur_bytes
, &host_offset
,
5387 if (type
== QCOW2_SUBCLUSTER_COMPRESSED
) {
5391 offset
+= cur_bytes
;
5399 * Downgrades an image's version. To achieve this, any incompatible features
5400 * have to be removed.
5402 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
5403 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5406 BDRVQcow2State
*s
= bs
->opaque
;
5407 int current_version
= s
->qcow_version
;
5411 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5412 assert(target_version
< current_version
);
5414 /* There are no other versions (now) that you can downgrade to */
5415 assert(target_version
== 2);
5417 if (s
->refcount_order
!= 4) {
5418 error_setg(errp
, "compat=0.10 requires refcount_bits=16");
5422 if (has_data_file(bs
)) {
5423 error_setg(errp
, "Cannot downgrade an image with a data file");
5428 * If any internal snapshot has a different size than the current
5429 * image size, or VM state size that exceeds 32 bits, downgrading
5430 * is unsafe. Even though we would still use v3-compliant output
5431 * to preserve that data, other v2 programs might not realize
5432 * those optional fields are important.
5434 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5435 if (s
->snapshots
[i
].vm_state_size
> UINT32_MAX
||
5436 s
->snapshots
[i
].disk_size
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
5437 error_setg(errp
, "Internal snapshots prevent downgrade of image");
5442 /* clear incompatible features */
5443 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
5444 ret
= qcow2_mark_clean(bs
);
5446 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5451 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5452 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5453 * best thing to do anyway */
5455 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_COMPRESSION
) {
5456 error_setg(errp
, "Cannot downgrade an image with incompatible features "
5457 "0x%" PRIx64
" set",
5458 s
->incompatible_features
& ~QCOW2_INCOMPAT_COMPRESSION
);
5462 /* since we can ignore compatible features, we can set them to 0 as well */
5463 s
->compatible_features
= 0;
5464 /* if lazy refcounts have been used, they have already been fixed through
5465 * clearing the dirty flag */
5467 /* clearing autoclear features is trivial */
5468 s
->autoclear_features
= 0;
5470 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
5472 error_setg_errno(errp
, -ret
, "Failed to turn zero into data clusters");
5476 if (s
->incompatible_features
& QCOW2_INCOMPAT_COMPRESSION
) {
5477 ret
= qcow2_has_compressed_clusters(bs
);
5479 error_setg(errp
, "Failed to check block status");
5483 error_setg(errp
, "Cannot downgrade an image with zstd compression "
5484 "type and existing compressed clusters");
5488 * No compressed clusters for now, so just chose default zlib
5491 s
->incompatible_features
&= ~QCOW2_INCOMPAT_COMPRESSION
;
5492 s
->compression_type
= QCOW2_COMPRESSION_TYPE_ZLIB
;
5495 assert(s
->incompatible_features
== 0);
5497 s
->qcow_version
= target_version
;
5498 ret
= qcow2_update_header(bs
);
5500 s
->qcow_version
= current_version
;
5501 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5508 * Upgrades an image's version. While newer versions encompass all
5509 * features of older versions, some things may have to be presented
5512 static int qcow2_upgrade(BlockDriverState
*bs
, int target_version
,
5513 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5516 BDRVQcow2State
*s
= bs
->opaque
;
5517 bool need_snapshot_update
;
5518 int current_version
= s
->qcow_version
;
5522 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5523 assert(target_version
> current_version
);
5525 /* There are no other versions (yet) that you can upgrade to */
5526 assert(target_version
== 3);
5528 status_cb(bs
, 0, 2, cb_opaque
);
5531 * In v2, snapshots do not need to have extra data. v3 requires
5532 * the 64-bit VM state size and the virtual disk size to be
5534 * qcow2_write_snapshots() will always write the list in the
5535 * v3-compliant format.
5537 need_snapshot_update
= false;
5538 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
5539 if (s
->snapshots
[i
].extra_data_size
<
5540 sizeof_field(QCowSnapshotExtraData
, vm_state_size_large
) +
5541 sizeof_field(QCowSnapshotExtraData
, disk_size
))
5543 need_snapshot_update
= true;
5547 if (need_snapshot_update
) {
5548 ret
= qcow2_write_snapshots(bs
);
5550 error_setg_errno(errp
, -ret
, "Failed to update the snapshot table");
5554 status_cb(bs
, 1, 2, cb_opaque
);
5556 s
->qcow_version
= target_version
;
5557 ret
= qcow2_update_header(bs
);
5559 s
->qcow_version
= current_version
;
5560 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5563 status_cb(bs
, 2, 2, cb_opaque
);
5568 typedef enum Qcow2AmendOperation
{
5569 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5570 * statically initialized to so that the helper CB can discern the first
5571 * invocation from an operation change */
5572 QCOW2_NO_OPERATION
= 0,
5575 QCOW2_UPDATING_ENCRYPTION
,
5576 QCOW2_CHANGING_REFCOUNT_ORDER
,
5578 } Qcow2AmendOperation
;
5580 typedef struct Qcow2AmendHelperCBInfo
{
5581 /* The code coordinating the amend operations should only modify
5582 * these four fields; the rest will be managed by the CB */
5583 BlockDriverAmendStatusCB
*original_status_cb
;
5584 void *original_cb_opaque
;
5586 Qcow2AmendOperation current_operation
;
5588 /* Total number of operations to perform (only set once) */
5589 int total_operations
;
5591 /* The following fields are managed by the CB */
5593 /* Number of operations completed */
5594 int operations_completed
;
5596 /* Cumulative offset of all completed operations */
5597 int64_t offset_completed
;
5599 Qcow2AmendOperation last_operation
;
5600 int64_t last_work_size
;
5601 } Qcow2AmendHelperCBInfo
;
5603 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
5604 int64_t operation_offset
,
5605 int64_t operation_work_size
, void *opaque
)
5607 Qcow2AmendHelperCBInfo
*info
= opaque
;
5608 int64_t current_work_size
;
5609 int64_t projected_work_size
;
5611 if (info
->current_operation
!= info
->last_operation
) {
5612 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
5613 info
->offset_completed
+= info
->last_work_size
;
5614 info
->operations_completed
++;
5617 info
->last_operation
= info
->current_operation
;
5620 assert(info
->total_operations
> 0);
5621 assert(info
->operations_completed
< info
->total_operations
);
5623 info
->last_work_size
= operation_work_size
;
5625 current_work_size
= info
->offset_completed
+ operation_work_size
;
5627 /* current_work_size is the total work size for (operations_completed + 1)
5628 * operations (which includes this one), so multiply it by the number of
5629 * operations not covered and divide it by the number of operations
5630 * covered to get a projection for the operations not covered */
5631 projected_work_size
= current_work_size
* (info
->total_operations
-
5632 info
->operations_completed
- 1)
5633 / (info
->operations_completed
+ 1);
5635 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
5636 current_work_size
+ projected_work_size
,
5637 info
->original_cb_opaque
);
5640 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5641 BlockDriverAmendStatusCB
*status_cb
,
5646 BDRVQcow2State
*s
= bs
->opaque
;
5647 int old_version
= s
->qcow_version
, new_version
= old_version
;
5648 uint64_t new_size
= 0;
5649 const char *backing_file
= NULL
, *backing_format
= NULL
, *data_file
= NULL
;
5650 bool lazy_refcounts
= s
->use_lazy_refcounts
;
5651 bool data_file_raw
= data_file_is_raw(bs
);
5652 const char *compat
= NULL
;
5653 int refcount_bits
= s
->refcount_bits
;
5655 QemuOptDesc
*desc
= opts
->list
->desc
;
5656 Qcow2AmendHelperCBInfo helper_cb_info
;
5657 bool encryption_update
= false;
5659 while (desc
&& desc
->name
) {
5660 if (!qemu_opt_find(opts
, desc
->name
)) {
5661 /* only change explicitly defined options */
5666 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
5667 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
5669 /* preserve default */
5670 } else if (!strcmp(compat
, "0.10") || !strcmp(compat
, "v2")) {
5672 } else if (!strcmp(compat
, "1.1") || !strcmp(compat
, "v3")) {
5675 error_setg(errp
, "Unknown compatibility level %s", compat
);
5678 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
5679 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
5680 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
5681 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5682 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
5683 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5684 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
5687 "Can't amend encryption options - encryption not present");
5690 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5692 "Only LUKS encryption options can be amended");
5695 encryption_update
= true;
5696 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
5697 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
5699 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
5700 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
5703 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
5704 !is_power_of_2(refcount_bits
))
5706 error_setg(errp
, "Refcount width must be a power of two and "
5707 "may not exceed 64 bits");
5710 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE
)) {
5711 data_file
= qemu_opt_get(opts
, BLOCK_OPT_DATA_FILE
);
5712 if (data_file
&& !has_data_file(bs
)) {
5713 error_setg(errp
, "data-file can only be set for images that "
5714 "use an external data file");
5717 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE_RAW
)) {
5718 data_file_raw
= qemu_opt_get_bool(opts
, BLOCK_OPT_DATA_FILE_RAW
,
5720 if (data_file_raw
&& !data_file_is_raw(bs
)) {
5721 error_setg(errp
, "data-file-raw cannot be set on existing "
5726 /* if this point is reached, this probably means a new option was
5727 * added without having it covered here */
5734 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
5735 .original_status_cb
= status_cb
,
5736 .original_cb_opaque
= cb_opaque
,
5737 .total_operations
= (new_version
!= old_version
)
5738 + (s
->refcount_bits
!= refcount_bits
) +
5739 (encryption_update
== true)
5742 /* Upgrade first (some features may require compat=1.1) */
5743 if (new_version
> old_version
) {
5744 helper_cb_info
.current_operation
= QCOW2_UPGRADING
;
5745 ret
= qcow2_upgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5746 &helper_cb_info
, errp
);
5752 if (encryption_update
) {
5753 QDict
*amend_opts_dict
;
5754 QCryptoBlockAmendOptions
*amend_opts
;
5756 helper_cb_info
.current_operation
= QCOW2_UPDATING_ENCRYPTION
;
5757 amend_opts_dict
= qcow2_extract_crypto_opts(opts
, "luks", errp
);
5758 if (!amend_opts_dict
) {
5761 amend_opts
= block_crypto_amend_opts_init(amend_opts_dict
, errp
);
5762 qobject_unref(amend_opts_dict
);
5766 ret
= qcrypto_block_amend_options(s
->crypto
,
5767 qcow2_crypto_hdr_read_func
,
5768 qcow2_crypto_hdr_write_func
,
5773 qapi_free_QCryptoBlockAmendOptions(amend_opts
);
5779 if (s
->refcount_bits
!= refcount_bits
) {
5780 int refcount_order
= ctz32(refcount_bits
);
5782 if (new_version
< 3 && refcount_bits
!= 16) {
5783 error_setg(errp
, "Refcount widths other than 16 bits require "
5784 "compatibility level 1.1 or above (use compat=1.1 or "
5789 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
5790 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
5791 &qcow2_amend_helper_cb
,
5792 &helper_cb_info
, errp
);
5798 /* data-file-raw blocks backing files, so clear it first if requested */
5799 if (data_file_raw
) {
5800 s
->autoclear_features
|= QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5802 s
->autoclear_features
&= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5806 g_free(s
->image_data_file
);
5807 s
->image_data_file
= *data_file
? g_strdup(data_file
) : NULL
;
5810 ret
= qcow2_update_header(bs
);
5812 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5816 if (backing_file
|| backing_format
) {
5817 if (g_strcmp0(backing_file
, s
->image_backing_file
) ||
5818 g_strcmp0(backing_format
, s
->image_backing_format
)) {
5819 error_setg(errp
, "Cannot amend the backing file");
5820 error_append_hint(errp
,
5821 "You can use 'qemu-img rebase' instead.\n");
5826 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
5827 if (lazy_refcounts
) {
5828 if (new_version
< 3) {
5829 error_setg(errp
, "Lazy refcounts only supported with "
5830 "compatibility level 1.1 and above (use compat=1.1 "
5834 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5835 ret
= qcow2_update_header(bs
);
5837 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5838 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5841 s
->use_lazy_refcounts
= true;
5843 /* make image clean first */
5844 ret
= qcow2_mark_clean(bs
);
5846 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5849 /* now disallow lazy refcounts */
5850 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5851 ret
= qcow2_update_header(bs
);
5853 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5854 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5857 s
->use_lazy_refcounts
= false;
5862 BlockBackend
*blk
= blk_new_with_bs(bs
, BLK_PERM_RESIZE
, BLK_PERM_ALL
,
5869 * Amending image options should ensure that the image has
5870 * exactly the given new values, so pass exact=true here.
5872 ret
= blk_truncate(blk
, new_size
, true, PREALLOC_MODE_OFF
, 0, errp
);
5879 /* Downgrade last (so unsupported features can be removed before) */
5880 if (new_version
< old_version
) {
5881 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
5882 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5883 &helper_cb_info
, errp
);
5892 static int coroutine_fn
qcow2_co_amend(BlockDriverState
*bs
,
5893 BlockdevAmendOptions
*opts
,
5897 BlockdevAmendOptionsQcow2
*qopts
= &opts
->u
.qcow2
;
5898 BDRVQcow2State
*s
= bs
->opaque
;
5901 if (qopts
->encrypt
) {
5903 error_setg(errp
, "image is not encrypted, can't amend");
5907 if (qopts
->encrypt
->format
!= Q_CRYPTO_BLOCK_FORMAT_LUKS
) {
5909 "Amend can't be used to change the qcow2 encryption format");
5913 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
5915 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5919 ret
= qcrypto_block_amend_options(s
->crypto
,
5920 qcow2_crypto_hdr_read_func
,
5921 qcow2_crypto_hdr_write_func
,
5931 * If offset or size are negative, respectively, they will not be included in
5932 * the BLOCK_IMAGE_CORRUPTED event emitted.
5933 * fatal will be ignored for read-only BDS; corruptions found there will always
5934 * be considered non-fatal.
5936 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
5937 int64_t size
, const char *message_format
, ...)
5939 BDRVQcow2State
*s
= bs
->opaque
;
5940 const char *node_name
;
5944 fatal
= fatal
&& bdrv_is_writable(bs
);
5946 if (s
->signaled_corruption
&&
5947 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
5952 va_start(ap
, message_format
);
5953 message
= g_strdup_vprintf(message_format
, ap
);
5957 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
5958 "corruption events will be suppressed\n", message
);
5960 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
5961 "corruption events will be suppressed\n", message
);
5964 node_name
= bdrv_get_node_name(bs
);
5965 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
5966 *node_name
? node_name
: NULL
,
5967 message
, offset
>= 0, offset
,
5973 qcow2_mark_corrupt(bs
);
5974 bs
->drv
= NULL
; /* make BDS unusable */
5977 s
->signaled_corruption
= true;
5980 #define QCOW_COMMON_OPTIONS \
5982 .name = BLOCK_OPT_SIZE, \
5983 .type = QEMU_OPT_SIZE, \
5984 .help = "Virtual disk size" \
5987 .name = BLOCK_OPT_COMPAT_LEVEL, \
5988 .type = QEMU_OPT_STRING, \
5989 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
5992 .name = BLOCK_OPT_BACKING_FILE, \
5993 .type = QEMU_OPT_STRING, \
5994 .help = "File name of a base image" \
5997 .name = BLOCK_OPT_BACKING_FMT, \
5998 .type = QEMU_OPT_STRING, \
5999 .help = "Image format of the base image" \
6002 .name = BLOCK_OPT_DATA_FILE, \
6003 .type = QEMU_OPT_STRING, \
6004 .help = "File name of an external data file" \
6007 .name = BLOCK_OPT_DATA_FILE_RAW, \
6008 .type = QEMU_OPT_BOOL, \
6009 .help = "The external data file must stay valid " \
6013 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
6014 .type = QEMU_OPT_BOOL, \
6015 .help = "Postpone refcount updates", \
6016 .def_value_str = "off" \
6019 .name = BLOCK_OPT_REFCOUNT_BITS, \
6020 .type = QEMU_OPT_NUMBER, \
6021 .help = "Width of a reference count entry in bits", \
6022 .def_value_str = "16" \
6025 static QemuOptsList qcow2_create_opts
= {
6026 .name
= "qcow2-create-opts",
6027 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
6030 .name
= BLOCK_OPT_ENCRYPT
, \
6031 .type
= QEMU_OPT_BOOL
, \
6032 .help
= "Encrypt the image with format 'aes'. (Deprecated " \
6033 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)", \
6036 .name
= BLOCK_OPT_ENCRYPT_FORMAT
, \
6037 .type
= QEMU_OPT_STRING
, \
6038 .help
= "Encrypt the image, format choices: 'aes', 'luks'", \
6040 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
6041 "ID of secret providing qcow AES key or LUKS passphrase"), \
6042 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
6043 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
6044 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
6045 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
6046 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
6047 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
6049 .name
= BLOCK_OPT_CLUSTER_SIZE
, \
6050 .type
= QEMU_OPT_SIZE
, \
6051 .help
= "qcow2 cluster size", \
6052 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
) \
6055 .name
= BLOCK_OPT_EXTL2
, \
6056 .type
= QEMU_OPT_BOOL
, \
6057 .help
= "Extended L2 tables", \
6058 .def_value_str
= "off" \
6061 .name
= BLOCK_OPT_PREALLOC
, \
6062 .type
= QEMU_OPT_STRING
, \
6063 .help
= "Preallocation mode (allowed values: off, " \
6064 "metadata, falloc, full)" \
6067 .name
= BLOCK_OPT_COMPRESSION_TYPE
, \
6068 .type
= QEMU_OPT_STRING
, \
6069 .help
= "Compression method used for image cluster " \
6071 .def_value_str
= "zlib" \
6073 QCOW_COMMON_OPTIONS
,
6074 { /* end of list */ }
6078 static QemuOptsList qcow2_amend_opts
= {
6079 .name
= "qcow2-amend-opts",
6080 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts
.head
),
6082 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
6083 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
6084 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
6085 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
6086 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
6087 QCOW_COMMON_OPTIONS
,
6088 { /* end of list */ }
6092 static const char *const qcow2_strong_runtime_opts
[] = {
6093 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET
,
6098 BlockDriver bdrv_qcow2
= {
6099 .format_name
= "qcow2",
6100 .instance_size
= sizeof(BDRVQcow2State
),
6101 .bdrv_probe
= qcow2_probe
,
6102 .bdrv_open
= qcow2_open
,
6103 .bdrv_close
= qcow2_close
,
6104 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
6105 .bdrv_reopen_commit
= qcow2_reopen_commit
,
6106 .bdrv_reopen_commit_post
= qcow2_reopen_commit_post
,
6107 .bdrv_reopen_abort
= qcow2_reopen_abort
,
6108 .bdrv_join_options
= qcow2_join_options
,
6109 .bdrv_child_perm
= bdrv_default_perms
,
6110 .bdrv_co_create_opts
= qcow2_co_create_opts
,
6111 .bdrv_co_create
= qcow2_co_create
,
6112 .bdrv_has_zero_init
= qcow2_has_zero_init
,
6113 .bdrv_co_block_status
= qcow2_co_block_status
,
6115 .bdrv_co_preadv_part
= qcow2_co_preadv_part
,
6116 .bdrv_co_pwritev_part
= qcow2_co_pwritev_part
,
6117 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
6119 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
6120 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
6121 .bdrv_co_copy_range_from
= qcow2_co_copy_range_from
,
6122 .bdrv_co_copy_range_to
= qcow2_co_copy_range_to
,
6123 .bdrv_co_truncate
= qcow2_co_truncate
,
6124 .bdrv_co_pwritev_compressed_part
= qcow2_co_pwritev_compressed_part
,
6125 .bdrv_make_empty
= qcow2_make_empty
,
6127 .bdrv_snapshot_create
= qcow2_snapshot_create
,
6128 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
6129 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
6130 .bdrv_snapshot_list
= qcow2_snapshot_list
,
6131 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
6132 .bdrv_measure
= qcow2_measure
,
6133 .bdrv_co_get_info
= qcow2_co_get_info
,
6134 .bdrv_get_specific_info
= qcow2_get_specific_info
,
6136 .bdrv_co_save_vmstate
= qcow2_co_save_vmstate
,
6137 .bdrv_co_load_vmstate
= qcow2_co_load_vmstate
,
6140 .supports_backing
= true,
6141 .bdrv_change_backing_file
= qcow2_change_backing_file
,
6143 .bdrv_refresh_limits
= qcow2_refresh_limits
,
6144 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
6145 .bdrv_inactivate
= qcow2_inactivate
,
6147 .create_opts
= &qcow2_create_opts
,
6148 .amend_opts
= &qcow2_amend_opts
,
6149 .strong_runtime_opts
= qcow2_strong_runtime_opts
,
6150 .mutable_opts
= mutable_opts
,
6151 .bdrv_co_check
= qcow2_co_check
,
6152 .bdrv_amend_options
= qcow2_amend_options
,
6153 .bdrv_co_amend
= qcow2_co_amend
,
6155 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
6156 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
6158 .bdrv_supports_persistent_dirty_bitmap
=
6159 qcow2_supports_persistent_dirty_bitmap
,
6160 .bdrv_co_can_store_new_dirty_bitmap
= qcow2_co_can_store_new_dirty_bitmap
,
6161 .bdrv_co_remove_persistent_dirty_bitmap
=
6162 qcow2_co_remove_persistent_dirty_bitmap
,
6165 static void bdrv_qcow2_init(void)
6167 bdrv_register(&bdrv_qcow2
);
6170 block_init(bdrv_qcow2_init
);