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
24 #include "qemu-common.h"
25 #include "block/block_int.h"
26 #include "qemu/module.h"
29 #include "block/qcow2.h"
30 #include "qemu/error-report.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qapi/qmp/qbool.h"
33 #include "qapi/util.h"
34 #include "qapi/qmp/types.h"
35 #include "qapi-event.h"
37 #include "qemu/option_int.h"
40 Differences with QCOW:
42 - Support for multiple incremental snapshots.
43 - Memory management by reference counts.
44 - Clusters which have a reference count of one have the bit
45 QCOW_OFLAG_COPIED to optimize write performance.
46 - Size of compressed clusters is stored in sectors to reduce bit usage
47 in the cluster offsets.
48 - Support for storing additional data (such as the VM state) in the
50 - If a backing store is used, the cluster size is not constrained
51 (could be backported to QCOW).
52 - L2 tables have always a size of one cluster.
59 } QEMU_PACKED QCowExtension
;
61 #define QCOW2_EXT_MAGIC_END 0
62 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
63 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
65 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
67 const QCowHeader
*cow_header
= (const void *)buf
;
69 if (buf_size
>= sizeof(QCowHeader
) &&
70 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
71 be32_to_cpu(cow_header
->version
) >= 2)
79 * read qcow2 extension and fill bs
80 * start reading from start_offset
81 * finish reading upon magic of value 0 or when end_offset reached
82 * unknown magic is skipped (future extension this version knows nothing about)
83 * return 0 upon success, non-0 otherwise
85 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
86 uint64_t end_offset
, void **p_feature_table
,
89 BDRVQcowState
*s
= bs
->opaque
;
95 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
97 offset
= start_offset
;
98 while (offset
< end_offset
) {
102 if (offset
> s
->cluster_size
)
103 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
105 printf("attempting to read extended header in offset %lu\n", offset
);
108 ret
= bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
));
110 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
111 "pread fail from offset %" PRIu64
, offset
);
114 be32_to_cpus(&ext
.magic
);
115 be32_to_cpus(&ext
.len
);
116 offset
+= sizeof(ext
);
118 printf("ext.magic = 0x%x\n", ext
.magic
);
120 if (offset
> end_offset
|| ext
.len
> end_offset
- offset
) {
121 error_setg(errp
, "Header extension too large");
126 case QCOW2_EXT_MAGIC_END
:
129 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
130 if (ext
.len
>= sizeof(bs
->backing_format
)) {
131 error_setg(errp
, "ERROR: ext_backing_format: len=%" PRIu32
132 " too large (>=%zu)", ext
.len
,
133 sizeof(bs
->backing_format
));
136 ret
= bdrv_pread(bs
->file
, offset
, bs
->backing_format
, ext
.len
);
138 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
139 "Could not read format name");
142 bs
->backing_format
[ext
.len
] = '\0';
143 s
->image_backing_format
= g_strdup(bs
->backing_format
);
145 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
149 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
150 if (p_feature_table
!= NULL
) {
151 void* feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
152 ret
= bdrv_pread(bs
->file
, offset
, feature_table
, ext
.len
);
154 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
155 "Could not read table");
159 *p_feature_table
= feature_table
;
164 /* unknown magic - save it in case we need to rewrite the header */
166 Qcow2UnknownHeaderExtension
*uext
;
168 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
169 uext
->magic
= ext
.magic
;
171 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
173 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
175 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
176 "Could not read data");
183 offset
+= ((ext
.len
+ 7) & ~7);
189 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
191 BDRVQcowState
*s
= bs
->opaque
;
192 Qcow2UnknownHeaderExtension
*uext
, *next
;
194 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
195 QLIST_REMOVE(uext
, next
);
200 static void GCC_FMT_ATTR(3, 4) report_unsupported(BlockDriverState
*bs
,
201 Error
**errp
, const char *fmt
, ...)
207 vsnprintf(msg
, sizeof(msg
), fmt
, ap
);
210 error_set(errp
, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
211 bdrv_get_device_name(bs
), "qcow2", msg
);
214 static void report_unsupported_feature(BlockDriverState
*bs
,
215 Error
**errp
, Qcow2Feature
*table
, uint64_t mask
)
217 char *features
= g_strdup("");
220 while (table
&& table
->name
[0] != '\0') {
221 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
222 if (mask
& (1ULL << table
->bit
)) {
224 features
= g_strdup_printf("%s%s%.46s", old
, *old
? ", " : "",
227 mask
&= ~(1ULL << table
->bit
);
235 features
= g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64
,
236 old
, *old
? ", " : "", mask
);
240 report_unsupported(bs
, errp
, "%s", features
);
245 * Sets the dirty bit and flushes afterwards if necessary.
247 * The incompatible_features bit is only set if the image file header was
248 * updated successfully. Therefore it is not required to check the return
249 * value of this function.
251 int qcow2_mark_dirty(BlockDriverState
*bs
)
253 BDRVQcowState
*s
= bs
->opaque
;
257 assert(s
->qcow_version
>= 3);
259 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
260 return 0; /* already dirty */
263 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
264 ret
= bdrv_pwrite(bs
->file
, offsetof(QCowHeader
, incompatible_features
),
269 ret
= bdrv_flush(bs
->file
);
274 /* Only treat image as dirty if the header was updated successfully */
275 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
280 * Clears the dirty bit and flushes before if necessary. Only call this
281 * function when there are no pending requests, it does not guard against
282 * concurrent requests dirtying the image.
284 static int qcow2_mark_clean(BlockDriverState
*bs
)
286 BDRVQcowState
*s
= bs
->opaque
;
288 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
291 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
293 ret
= bdrv_flush(bs
);
298 return qcow2_update_header(bs
);
304 * Marks the image as corrupt.
306 int qcow2_mark_corrupt(BlockDriverState
*bs
)
308 BDRVQcowState
*s
= bs
->opaque
;
310 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
311 return qcow2_update_header(bs
);
315 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
316 * before if necessary.
318 int qcow2_mark_consistent(BlockDriverState
*bs
)
320 BDRVQcowState
*s
= bs
->opaque
;
322 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
323 int ret
= bdrv_flush(bs
);
328 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
329 return qcow2_update_header(bs
);
334 static int qcow2_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
337 int ret
= qcow2_check_refcounts(bs
, result
, fix
);
342 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
343 ret
= qcow2_mark_clean(bs
);
347 return qcow2_mark_consistent(bs
);
352 static int validate_table_offset(BlockDriverState
*bs
, uint64_t offset
,
353 uint64_t entries
, size_t entry_len
)
355 BDRVQcowState
*s
= bs
->opaque
;
358 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
359 * because values will be passed to qemu functions taking int64_t. */
360 if (entries
> INT64_MAX
/ entry_len
) {
364 size
= entries
* entry_len
;
366 if (INT64_MAX
- size
< offset
) {
370 /* Tables must be cluster aligned */
371 if (offset
& (s
->cluster_size
- 1)) {
378 static QemuOptsList qcow2_runtime_opts
= {
380 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
383 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
384 .type
= QEMU_OPT_BOOL
,
385 .help
= "Postpone refcount updates",
388 .name
= QCOW2_OPT_DISCARD_REQUEST
,
389 .type
= QEMU_OPT_BOOL
,
390 .help
= "Pass guest discard requests to the layer below",
393 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
394 .type
= QEMU_OPT_BOOL
,
395 .help
= "Generate discard requests when snapshot related space "
399 .name
= QCOW2_OPT_DISCARD_OTHER
,
400 .type
= QEMU_OPT_BOOL
,
401 .help
= "Generate discard requests when other clusters are freed",
404 .name
= QCOW2_OPT_OVERLAP
,
405 .type
= QEMU_OPT_STRING
,
406 .help
= "Selects which overlap checks to perform from a range of "
407 "templates (none, constant, cached, all)",
410 .name
= QCOW2_OPT_OVERLAP_TEMPLATE
,
411 .type
= QEMU_OPT_STRING
,
412 .help
= "Selects which overlap checks to perform from a range of "
413 "templates (none, constant, cached, all)",
416 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
417 .type
= QEMU_OPT_BOOL
,
418 .help
= "Check for unintended writes into the main qcow2 header",
421 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
422 .type
= QEMU_OPT_BOOL
,
423 .help
= "Check for unintended writes into the active L1 table",
426 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
427 .type
= QEMU_OPT_BOOL
,
428 .help
= "Check for unintended writes into an active L2 table",
431 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
432 .type
= QEMU_OPT_BOOL
,
433 .help
= "Check for unintended writes into the refcount table",
436 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
437 .type
= QEMU_OPT_BOOL
,
438 .help
= "Check for unintended writes into a refcount block",
441 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
442 .type
= QEMU_OPT_BOOL
,
443 .help
= "Check for unintended writes into the snapshot table",
446 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
447 .type
= QEMU_OPT_BOOL
,
448 .help
= "Check for unintended writes into an inactive L1 table",
451 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
452 .type
= QEMU_OPT_BOOL
,
453 .help
= "Check for unintended writes into an inactive L2 table",
456 .name
= QCOW2_OPT_CACHE_SIZE
,
457 .type
= QEMU_OPT_SIZE
,
458 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
462 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
463 .type
= QEMU_OPT_SIZE
,
464 .help
= "Maximum L2 table cache size",
467 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
468 .type
= QEMU_OPT_SIZE
,
469 .help
= "Maximum refcount block cache size",
471 { /* end of list */ }
475 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
476 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
477 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
478 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
479 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
480 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
481 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
482 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
483 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
486 static void read_cache_sizes(QemuOpts
*opts
, uint64_t *l2_cache_size
,
487 uint64_t *refcount_cache_size
, Error
**errp
)
489 uint64_t combined_cache_size
;
490 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
492 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
493 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
494 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
496 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
497 *l2_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
, 0);
498 *refcount_cache_size
= qemu_opt_get_size(opts
,
499 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
501 if (combined_cache_size_set
) {
502 if (l2_cache_size_set
&& refcount_cache_size_set
) {
503 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
504 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
507 } else if (*l2_cache_size
> combined_cache_size
) {
508 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
509 QCOW2_OPT_CACHE_SIZE
);
511 } else if (*refcount_cache_size
> combined_cache_size
) {
512 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
513 QCOW2_OPT_CACHE_SIZE
);
517 if (l2_cache_size_set
) {
518 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
519 } else if (refcount_cache_size_set
) {
520 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
522 *refcount_cache_size
= combined_cache_size
523 / (DEFAULT_L2_REFCOUNT_SIZE_RATIO
+ 1);
524 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
527 if (!l2_cache_size_set
&& !refcount_cache_size_set
) {
528 *l2_cache_size
= DEFAULT_L2_CACHE_BYTE_SIZE
;
529 *refcount_cache_size
= *l2_cache_size
530 / DEFAULT_L2_REFCOUNT_SIZE_RATIO
;
531 } else if (!l2_cache_size_set
) {
532 *l2_cache_size
= *refcount_cache_size
533 * DEFAULT_L2_REFCOUNT_SIZE_RATIO
;
534 } else if (!refcount_cache_size_set
) {
535 *refcount_cache_size
= *l2_cache_size
536 / DEFAULT_L2_REFCOUNT_SIZE_RATIO
;
541 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
544 BDRVQcowState
*s
= bs
->opaque
;
548 QemuOpts
*opts
= NULL
;
549 Error
*local_err
= NULL
;
551 uint64_t l1_vm_state_index
;
552 const char *opt_overlap_check
, *opt_overlap_check_template
;
553 int overlap_check_template
= 0;
554 uint64_t l2_cache_size
, refcount_cache_size
;
556 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
558 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
561 be32_to_cpus(&header
.magic
);
562 be32_to_cpus(&header
.version
);
563 be64_to_cpus(&header
.backing_file_offset
);
564 be32_to_cpus(&header
.backing_file_size
);
565 be64_to_cpus(&header
.size
);
566 be32_to_cpus(&header
.cluster_bits
);
567 be32_to_cpus(&header
.crypt_method
);
568 be64_to_cpus(&header
.l1_table_offset
);
569 be32_to_cpus(&header
.l1_size
);
570 be64_to_cpus(&header
.refcount_table_offset
);
571 be32_to_cpus(&header
.refcount_table_clusters
);
572 be64_to_cpus(&header
.snapshots_offset
);
573 be32_to_cpus(&header
.nb_snapshots
);
575 if (header
.magic
!= QCOW_MAGIC
) {
576 error_setg(errp
, "Image is not in qcow2 format");
580 if (header
.version
< 2 || header
.version
> 3) {
581 report_unsupported(bs
, errp
, "QCOW version %" PRIu32
, header
.version
);
586 s
->qcow_version
= header
.version
;
588 /* Initialise cluster size */
589 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
590 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
591 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
592 header
.cluster_bits
);
597 s
->cluster_bits
= header
.cluster_bits
;
598 s
->cluster_size
= 1 << s
->cluster_bits
;
599 s
->cluster_sectors
= 1 << (s
->cluster_bits
- 9);
601 /* Initialise version 3 header fields */
602 if (header
.version
== 2) {
603 header
.incompatible_features
= 0;
604 header
.compatible_features
= 0;
605 header
.autoclear_features
= 0;
606 header
.refcount_order
= 4;
607 header
.header_length
= 72;
609 be64_to_cpus(&header
.incompatible_features
);
610 be64_to_cpus(&header
.compatible_features
);
611 be64_to_cpus(&header
.autoclear_features
);
612 be32_to_cpus(&header
.refcount_order
);
613 be32_to_cpus(&header
.header_length
);
615 if (header
.header_length
< 104) {
616 error_setg(errp
, "qcow2 header too short");
622 if (header
.header_length
> s
->cluster_size
) {
623 error_setg(errp
, "qcow2 header exceeds cluster size");
628 if (header
.header_length
> sizeof(header
)) {
629 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
630 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
631 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
632 s
->unknown_header_fields_size
);
634 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
640 if (header
.backing_file_offset
> s
->cluster_size
) {
641 error_setg(errp
, "Invalid backing file offset");
646 if (header
.backing_file_offset
) {
647 ext_end
= header
.backing_file_offset
;
649 ext_end
= 1 << header
.cluster_bits
;
652 /* Handle feature bits */
653 s
->incompatible_features
= header
.incompatible_features
;
654 s
->compatible_features
= header
.compatible_features
;
655 s
->autoclear_features
= header
.autoclear_features
;
657 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
658 void *feature_table
= NULL
;
659 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
660 &feature_table
, NULL
);
661 report_unsupported_feature(bs
, errp
, feature_table
,
662 s
->incompatible_features
&
663 ~QCOW2_INCOMPAT_MASK
);
665 g_free(feature_table
);
669 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
670 /* Corrupt images may not be written to unless they are being repaired
672 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
673 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
680 /* Check support for various header values */
681 if (header
.refcount_order
> 6) {
682 error_setg(errp
, "Reference count entry width too large; may not "
687 s
->refcount_order
= header
.refcount_order
;
688 s
->refcount_bits
= 1 << s
->refcount_order
;
689 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
690 s
->refcount_max
+= s
->refcount_max
- 1;
692 if (header
.crypt_method
> QCOW_CRYPT_AES
) {
693 error_setg(errp
, "Unsupported encryption method: %" PRIu32
,
694 header
.crypt_method
);
698 s
->crypt_method_header
= header
.crypt_method
;
699 if (s
->crypt_method_header
) {
703 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
704 s
->l2_size
= 1 << s
->l2_bits
;
705 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
706 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
707 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
708 bs
->total_sectors
= header
.size
/ 512;
709 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
710 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
711 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
713 s
->refcount_table_offset
= header
.refcount_table_offset
;
714 s
->refcount_table_size
=
715 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
717 if (header
.refcount_table_clusters
> qcow2_max_refcount_clusters(s
)) {
718 error_setg(errp
, "Reference count table too large");
723 ret
= validate_table_offset(bs
, s
->refcount_table_offset
,
724 s
->refcount_table_size
, sizeof(uint64_t));
726 error_setg(errp
, "Invalid reference count table offset");
730 /* Snapshot table offset/length */
731 if (header
.nb_snapshots
> QCOW_MAX_SNAPSHOTS
) {
732 error_setg(errp
, "Too many snapshots");
737 ret
= validate_table_offset(bs
, header
.snapshots_offset
,
739 sizeof(QCowSnapshotHeader
));
741 error_setg(errp
, "Invalid snapshot table offset");
745 /* read the level 1 table */
746 if (header
.l1_size
> QCOW_MAX_L1_SIZE
/ sizeof(uint64_t)) {
747 error_setg(errp
, "Active L1 table too large");
751 s
->l1_size
= header
.l1_size
;
753 l1_vm_state_index
= size_to_l1(s
, header
.size
);
754 if (l1_vm_state_index
> INT_MAX
) {
755 error_setg(errp
, "Image is too big");
759 s
->l1_vm_state_index
= l1_vm_state_index
;
761 /* the L1 table must contain at least enough entries to put
763 if (s
->l1_size
< s
->l1_vm_state_index
) {
764 error_setg(errp
, "L1 table is too small");
769 ret
= validate_table_offset(bs
, header
.l1_table_offset
,
770 header
.l1_size
, sizeof(uint64_t));
772 error_setg(errp
, "Invalid L1 table offset");
775 s
->l1_table_offset
= header
.l1_table_offset
;
778 if (s
->l1_size
> 0) {
779 s
->l1_table
= qemu_try_blockalign(bs
->file
,
780 align_offset(s
->l1_size
* sizeof(uint64_t), 512));
781 if (s
->l1_table
== NULL
) {
782 error_setg(errp
, "Could not allocate L1 table");
786 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
787 s
->l1_size
* sizeof(uint64_t));
789 error_setg_errno(errp
, -ret
, "Could not read L1 table");
792 for(i
= 0;i
< s
->l1_size
; i
++) {
793 be64_to_cpus(&s
->l1_table
[i
]);
797 /* get L2 table/refcount block cache size from command line options */
798 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
799 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
801 error_propagate(errp
, local_err
);
806 read_cache_sizes(opts
, &l2_cache_size
, &refcount_cache_size
, &local_err
);
808 error_propagate(errp
, local_err
);
813 l2_cache_size
/= s
->cluster_size
;
814 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
815 l2_cache_size
= MIN_L2_CACHE_SIZE
;
817 if (l2_cache_size
> INT_MAX
) {
818 error_setg(errp
, "L2 cache size too big");
823 refcount_cache_size
/= s
->cluster_size
;
824 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
825 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
827 if (refcount_cache_size
> INT_MAX
) {
828 error_setg(errp
, "Refcount cache size too big");
833 /* alloc L2 table/refcount block cache */
834 s
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
);
835 s
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
);
836 if (s
->l2_table_cache
== NULL
|| s
->refcount_block_cache
== NULL
) {
837 error_setg(errp
, "Could not allocate metadata caches");
842 s
->cluster_cache
= g_malloc(s
->cluster_size
);
843 /* one more sector for decompressed data alignment */
844 s
->cluster_data
= qemu_try_blockalign(bs
->file
, QCOW_MAX_CRYPT_CLUSTERS
845 * s
->cluster_size
+ 512);
846 if (s
->cluster_data
== NULL
) {
847 error_setg(errp
, "Could not allocate temporary cluster buffer");
852 s
->cluster_cache_offset
= -1;
855 ret
= qcow2_refcount_init(bs
);
857 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
861 QLIST_INIT(&s
->cluster_allocs
);
862 QTAILQ_INIT(&s
->discards
);
864 /* read qcow2 extensions */
865 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
867 error_propagate(errp
, local_err
);
872 /* read the backing file name */
873 if (header
.backing_file_offset
!= 0) {
874 len
= header
.backing_file_size
;
875 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
876 len
>= sizeof(bs
->backing_file
)) {
877 error_setg(errp
, "Backing file name too long");
881 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
882 bs
->backing_file
, len
);
884 error_setg_errno(errp
, -ret
, "Could not read backing file name");
887 bs
->backing_file
[len
] = '\0';
888 s
->image_backing_file
= g_strdup(bs
->backing_file
);
891 /* Internal snapshots */
892 s
->snapshots_offset
= header
.snapshots_offset
;
893 s
->nb_snapshots
= header
.nb_snapshots
;
895 ret
= qcow2_read_snapshots(bs
);
897 error_setg_errno(errp
, -ret
, "Could not read snapshots");
901 /* Clear unknown autoclear feature bits */
902 if (!bs
->read_only
&& !(flags
& BDRV_O_INCOMING
) && s
->autoclear_features
) {
903 s
->autoclear_features
= 0;
904 ret
= qcow2_update_header(bs
);
906 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
911 /* Initialise locks */
912 qemu_co_mutex_init(&s
->lock
);
914 /* Repair image if dirty */
915 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INCOMING
)) && !bs
->read_only
&&
916 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
917 BdrvCheckResult result
= {0};
919 ret
= qcow2_check(bs
, &result
, BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
921 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
926 /* Enable lazy_refcounts according to image and command line options */
927 s
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
928 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
930 s
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
931 s
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
932 s
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
933 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
934 flags
& BDRV_O_UNMAP
);
935 s
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
936 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
937 s
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
938 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
940 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
941 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
942 if (opt_overlap_check_template
&& opt_overlap_check
&&
943 strcmp(opt_overlap_check_template
, opt_overlap_check
))
945 error_setg(errp
, "Conflicting values for qcow2 options '"
946 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
947 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
951 if (!opt_overlap_check
) {
952 opt_overlap_check
= opt_overlap_check_template
?: "cached";
955 if (!strcmp(opt_overlap_check
, "none")) {
956 overlap_check_template
= 0;
957 } else if (!strcmp(opt_overlap_check
, "constant")) {
958 overlap_check_template
= QCOW2_OL_CONSTANT
;
959 } else if (!strcmp(opt_overlap_check
, "cached")) {
960 overlap_check_template
= QCOW2_OL_CACHED
;
961 } else if (!strcmp(opt_overlap_check
, "all")) {
962 overlap_check_template
= QCOW2_OL_ALL
;
964 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
965 "'overlap-check'. Allowed are either of the following: "
966 "none, constant, cached, all", opt_overlap_check
);
971 s
->overlap_check
= 0;
972 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
973 /* overlap-check defines a template bitmask, but every flag may be
974 * overwritten through the associated boolean option */
976 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
977 overlap_check_template
& (1 << i
)) << i
;
983 if (s
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
984 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
985 "qemu 1.1 compatibility level");
992 BdrvCheckResult result
= {0};
993 qcow2_check_refcounts(bs
, &result
, 0);
1000 g_free(s
->unknown_header_fields
);
1001 cleanup_unknown_header_ext(bs
);
1002 qcow2_free_snapshots(bs
);
1003 qcow2_refcount_close(bs
);
1004 qemu_vfree(s
->l1_table
);
1005 /* else pre-write overlap checks in cache_destroy may crash */
1007 if (s
->l2_table_cache
) {
1008 qcow2_cache_destroy(bs
, s
->l2_table_cache
);
1010 if (s
->refcount_block_cache
) {
1011 qcow2_cache_destroy(bs
, s
->refcount_block_cache
);
1013 g_free(s
->cluster_cache
);
1014 qemu_vfree(s
->cluster_data
);
1018 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1020 BDRVQcowState
*s
= bs
->opaque
;
1022 bs
->bl
.write_zeroes_alignment
= s
->cluster_sectors
;
1025 static int qcow2_set_key(BlockDriverState
*bs
, const char *key
)
1027 BDRVQcowState
*s
= bs
->opaque
;
1031 memset(keybuf
, 0, 16);
1035 /* XXX: we could compress the chars to 7 bits to increase
1037 for(i
= 0;i
< len
;i
++) {
1040 s
->crypt_method
= s
->crypt_method_header
;
1042 if (AES_set_encrypt_key(keybuf
, 128, &s
->aes_encrypt_key
) != 0)
1044 if (AES_set_decrypt_key(keybuf
, 128, &s
->aes_decrypt_key
) != 0)
1054 AES_encrypt(in
, tmp
, &s
->aes_encrypt_key
);
1055 AES_decrypt(tmp
, out
, &s
->aes_decrypt_key
);
1056 for(i
= 0; i
< 16; i
++)
1057 printf(" %02x", tmp
[i
]);
1059 for(i
= 0; i
< 16; i
++)
1060 printf(" %02x", out
[i
]);
1067 /* We have no actual commit/abort logic for qcow2, but we need to write out any
1068 * unwritten data if we reopen read-only. */
1069 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1070 BlockReopenQueue
*queue
, Error
**errp
)
1074 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1075 ret
= bdrv_flush(state
->bs
);
1080 ret
= qcow2_mark_clean(state
->bs
);
1089 static int64_t coroutine_fn
qcow2_co_get_block_status(BlockDriverState
*bs
,
1090 int64_t sector_num
, int nb_sectors
, int *pnum
)
1092 BDRVQcowState
*s
= bs
->opaque
;
1093 uint64_t cluster_offset
;
1094 int index_in_cluster
, ret
;
1098 qemu_co_mutex_lock(&s
->lock
);
1099 ret
= qcow2_get_cluster_offset(bs
, sector_num
<< 9, pnum
, &cluster_offset
);
1100 qemu_co_mutex_unlock(&s
->lock
);
1105 if (cluster_offset
!= 0 && ret
!= QCOW2_CLUSTER_COMPRESSED
&&
1107 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
1108 cluster_offset
|= (index_in_cluster
<< BDRV_SECTOR_BITS
);
1109 status
|= BDRV_BLOCK_OFFSET_VALID
| cluster_offset
;
1111 if (ret
== QCOW2_CLUSTER_ZERO
) {
1112 status
|= BDRV_BLOCK_ZERO
;
1113 } else if (ret
!= QCOW2_CLUSTER_UNALLOCATED
) {
1114 status
|= BDRV_BLOCK_DATA
;
1119 /* handle reading after the end of the backing file */
1120 int qcow2_backing_read1(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
1121 int64_t sector_num
, int nb_sectors
)
1124 if ((sector_num
+ nb_sectors
) <= bs
->total_sectors
)
1126 if (sector_num
>= bs
->total_sectors
)
1129 n1
= bs
->total_sectors
- sector_num
;
1131 qemu_iovec_memset(qiov
, 512 * n1
, 0, 512 * (nb_sectors
- n1
));
1136 static coroutine_fn
int qcow2_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1137 int remaining_sectors
, QEMUIOVector
*qiov
)
1139 BDRVQcowState
*s
= bs
->opaque
;
1140 int index_in_cluster
, n1
;
1142 int cur_nr_sectors
; /* number of sectors in current iteration */
1143 uint64_t cluster_offset
= 0;
1144 uint64_t bytes_done
= 0;
1145 QEMUIOVector hd_qiov
;
1146 uint8_t *cluster_data
= NULL
;
1148 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1150 qemu_co_mutex_lock(&s
->lock
);
1152 while (remaining_sectors
!= 0) {
1154 /* prepare next request */
1155 cur_nr_sectors
= remaining_sectors
;
1156 if (s
->crypt_method
) {
1157 cur_nr_sectors
= MIN(cur_nr_sectors
,
1158 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
);
1161 ret
= qcow2_get_cluster_offset(bs
, sector_num
<< 9,
1162 &cur_nr_sectors
, &cluster_offset
);
1167 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
1169 qemu_iovec_reset(&hd_qiov
);
1170 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
,
1171 cur_nr_sectors
* 512);
1174 case QCOW2_CLUSTER_UNALLOCATED
:
1176 if (bs
->backing_hd
) {
1177 /* read from the base image */
1178 n1
= qcow2_backing_read1(bs
->backing_hd
, &hd_qiov
,
1179 sector_num
, cur_nr_sectors
);
1181 QEMUIOVector local_qiov
;
1183 qemu_iovec_init(&local_qiov
, hd_qiov
.niov
);
1184 qemu_iovec_concat(&local_qiov
, &hd_qiov
, 0,
1185 n1
* BDRV_SECTOR_SIZE
);
1187 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
1188 qemu_co_mutex_unlock(&s
->lock
);
1189 ret
= bdrv_co_readv(bs
->backing_hd
, sector_num
,
1191 qemu_co_mutex_lock(&s
->lock
);
1193 qemu_iovec_destroy(&local_qiov
);
1200 /* Note: in this case, no need to wait */
1201 qemu_iovec_memset(&hd_qiov
, 0, 0, 512 * cur_nr_sectors
);
1205 case QCOW2_CLUSTER_ZERO
:
1206 qemu_iovec_memset(&hd_qiov
, 0, 0, 512 * cur_nr_sectors
);
1209 case QCOW2_CLUSTER_COMPRESSED
:
1210 /* add AIO support for compressed blocks ? */
1211 ret
= qcow2_decompress_cluster(bs
, cluster_offset
);
1216 qemu_iovec_from_buf(&hd_qiov
, 0,
1217 s
->cluster_cache
+ index_in_cluster
* 512,
1218 512 * cur_nr_sectors
);
1221 case QCOW2_CLUSTER_NORMAL
:
1222 if ((cluster_offset
& 511) != 0) {
1227 if (s
->crypt_method
) {
1229 * For encrypted images, read everything into a temporary
1230 * contiguous buffer on which the AES functions can work.
1232 if (!cluster_data
) {
1234 qemu_try_blockalign(bs
->file
, QCOW_MAX_CRYPT_CLUSTERS
1236 if (cluster_data
== NULL
) {
1242 assert(cur_nr_sectors
<=
1243 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
);
1244 qemu_iovec_reset(&hd_qiov
);
1245 qemu_iovec_add(&hd_qiov
, cluster_data
,
1246 512 * cur_nr_sectors
);
1249 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1250 qemu_co_mutex_unlock(&s
->lock
);
1251 ret
= bdrv_co_readv(bs
->file
,
1252 (cluster_offset
>> 9) + index_in_cluster
,
1253 cur_nr_sectors
, &hd_qiov
);
1254 qemu_co_mutex_lock(&s
->lock
);
1258 if (s
->crypt_method
) {
1259 qcow2_encrypt_sectors(s
, sector_num
, cluster_data
,
1260 cluster_data
, cur_nr_sectors
, 0, &s
->aes_decrypt_key
);
1261 qemu_iovec_from_buf(qiov
, bytes_done
,
1262 cluster_data
, 512 * cur_nr_sectors
);
1267 g_assert_not_reached();
1272 remaining_sectors
-= cur_nr_sectors
;
1273 sector_num
+= cur_nr_sectors
;
1274 bytes_done
+= cur_nr_sectors
* 512;
1279 qemu_co_mutex_unlock(&s
->lock
);
1281 qemu_iovec_destroy(&hd_qiov
);
1282 qemu_vfree(cluster_data
);
1287 static coroutine_fn
int qcow2_co_writev(BlockDriverState
*bs
,
1289 int remaining_sectors
,
1292 BDRVQcowState
*s
= bs
->opaque
;
1293 int index_in_cluster
;
1295 int cur_nr_sectors
; /* number of sectors in current iteration */
1296 uint64_t cluster_offset
;
1297 QEMUIOVector hd_qiov
;
1298 uint64_t bytes_done
= 0;
1299 uint8_t *cluster_data
= NULL
;
1300 QCowL2Meta
*l2meta
= NULL
;
1302 trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num
,
1305 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1307 s
->cluster_cache_offset
= -1; /* disable compressed cache */
1309 qemu_co_mutex_lock(&s
->lock
);
1311 while (remaining_sectors
!= 0) {
1315 trace_qcow2_writev_start_part(qemu_coroutine_self());
1316 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
1317 cur_nr_sectors
= remaining_sectors
;
1318 if (s
->crypt_method
&&
1320 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
- index_in_cluster
) {
1322 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
- index_in_cluster
;
1325 ret
= qcow2_alloc_cluster_offset(bs
, sector_num
<< 9,
1326 &cur_nr_sectors
, &cluster_offset
, &l2meta
);
1331 assert((cluster_offset
& 511) == 0);
1333 qemu_iovec_reset(&hd_qiov
);
1334 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
,
1335 cur_nr_sectors
* 512);
1337 if (s
->crypt_method
) {
1338 if (!cluster_data
) {
1339 cluster_data
= qemu_try_blockalign(bs
->file
,
1340 QCOW_MAX_CRYPT_CLUSTERS
1342 if (cluster_data
== NULL
) {
1348 assert(hd_qiov
.size
<=
1349 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1350 qemu_iovec_to_buf(&hd_qiov
, 0, cluster_data
, hd_qiov
.size
);
1352 qcow2_encrypt_sectors(s
, sector_num
, cluster_data
,
1353 cluster_data
, cur_nr_sectors
, 1, &s
->aes_encrypt_key
);
1355 qemu_iovec_reset(&hd_qiov
);
1356 qemu_iovec_add(&hd_qiov
, cluster_data
,
1357 cur_nr_sectors
* 512);
1360 ret
= qcow2_pre_write_overlap_check(bs
, 0,
1361 cluster_offset
+ index_in_cluster
* BDRV_SECTOR_SIZE
,
1362 cur_nr_sectors
* BDRV_SECTOR_SIZE
);
1367 qemu_co_mutex_unlock(&s
->lock
);
1368 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
1369 trace_qcow2_writev_data(qemu_coroutine_self(),
1370 (cluster_offset
>> 9) + index_in_cluster
);
1371 ret
= bdrv_co_writev(bs
->file
,
1372 (cluster_offset
>> 9) + index_in_cluster
,
1373 cur_nr_sectors
, &hd_qiov
);
1374 qemu_co_mutex_lock(&s
->lock
);
1379 while (l2meta
!= NULL
) {
1382 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
1387 /* Take the request off the list of running requests */
1388 if (l2meta
->nb_clusters
!= 0) {
1389 QLIST_REMOVE(l2meta
, next_in_flight
);
1392 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
1394 next
= l2meta
->next
;
1399 remaining_sectors
-= cur_nr_sectors
;
1400 sector_num
+= cur_nr_sectors
;
1401 bytes_done
+= cur_nr_sectors
* 512;
1402 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors
);
1407 qemu_co_mutex_unlock(&s
->lock
);
1409 while (l2meta
!= NULL
) {
1412 if (l2meta
->nb_clusters
!= 0) {
1413 QLIST_REMOVE(l2meta
, next_in_flight
);
1415 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
1417 next
= l2meta
->next
;
1422 qemu_iovec_destroy(&hd_qiov
);
1423 qemu_vfree(cluster_data
);
1424 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
1429 static void qcow2_close(BlockDriverState
*bs
)
1431 BDRVQcowState
*s
= bs
->opaque
;
1432 qemu_vfree(s
->l1_table
);
1433 /* else pre-write overlap checks in cache_destroy may crash */
1436 if (!(bs
->open_flags
& BDRV_O_INCOMING
)) {
1439 ret1
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1440 ret2
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1443 error_report("Failed to flush the L2 table cache: %s",
1447 error_report("Failed to flush the refcount block cache: %s",
1451 if (!ret1
&& !ret2
) {
1452 qcow2_mark_clean(bs
);
1456 qcow2_cache_destroy(bs
, s
->l2_table_cache
);
1457 qcow2_cache_destroy(bs
, s
->refcount_block_cache
);
1459 g_free(s
->unknown_header_fields
);
1460 cleanup_unknown_header_ext(bs
);
1462 g_free(s
->image_backing_file
);
1463 g_free(s
->image_backing_format
);
1465 g_free(s
->cluster_cache
);
1466 qemu_vfree(s
->cluster_data
);
1467 qcow2_refcount_close(bs
);
1468 qcow2_free_snapshots(bs
);
1471 static void qcow2_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
1473 BDRVQcowState
*s
= bs
->opaque
;
1474 int flags
= s
->flags
;
1475 AES_KEY aes_encrypt_key
;
1476 AES_KEY aes_decrypt_key
;
1477 uint32_t crypt_method
= 0;
1479 Error
*local_err
= NULL
;
1483 * Backing files are read-only which makes all of their metadata immutable,
1484 * that means we don't have to worry about reopening them here.
1487 if (s
->crypt_method
) {
1488 crypt_method
= s
->crypt_method
;
1489 memcpy(&aes_encrypt_key
, &s
->aes_encrypt_key
, sizeof(aes_encrypt_key
));
1490 memcpy(&aes_decrypt_key
, &s
->aes_decrypt_key
, sizeof(aes_decrypt_key
));
1495 bdrv_invalidate_cache(bs
->file
, &local_err
);
1497 error_propagate(errp
, local_err
);
1501 memset(s
, 0, sizeof(BDRVQcowState
));
1502 options
= qdict_clone_shallow(bs
->options
);
1504 ret
= qcow2_open(bs
, options
, flags
, &local_err
);
1507 error_setg(errp
, "Could not reopen qcow2 layer: %s",
1508 error_get_pretty(local_err
));
1509 error_free(local_err
);
1511 } else if (ret
< 0) {
1512 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
1517 s
->crypt_method
= crypt_method
;
1518 memcpy(&s
->aes_encrypt_key
, &aes_encrypt_key
, sizeof(aes_encrypt_key
));
1519 memcpy(&s
->aes_decrypt_key
, &aes_decrypt_key
, sizeof(aes_decrypt_key
));
1523 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
1524 size_t len
, size_t buflen
)
1526 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
1527 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
1529 if (buflen
< ext_len
) {
1533 *ext_backing_fmt
= (QCowExtension
) {
1534 .magic
= cpu_to_be32(magic
),
1535 .len
= cpu_to_be32(len
),
1537 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
1543 * Updates the qcow2 header, including the variable length parts of it, i.e.
1544 * the backing file name and all extensions. qcow2 was not designed to allow
1545 * such changes, so if we run out of space (we can only use the first cluster)
1546 * this function may fail.
1548 * Returns 0 on success, -errno in error cases.
1550 int qcow2_update_header(BlockDriverState
*bs
)
1552 BDRVQcowState
*s
= bs
->opaque
;
1555 size_t buflen
= s
->cluster_size
;
1557 uint64_t total_size
;
1558 uint32_t refcount_table_clusters
;
1559 size_t header_length
;
1560 Qcow2UnknownHeaderExtension
*uext
;
1562 buf
= qemu_blockalign(bs
, buflen
);
1564 /* Header structure */
1565 header
= (QCowHeader
*) buf
;
1567 if (buflen
< sizeof(*header
)) {
1572 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
1573 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
1574 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
1576 *header
= (QCowHeader
) {
1577 /* Version 2 fields */
1578 .magic
= cpu_to_be32(QCOW_MAGIC
),
1579 .version
= cpu_to_be32(s
->qcow_version
),
1580 .backing_file_offset
= 0,
1581 .backing_file_size
= 0,
1582 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
1583 .size
= cpu_to_be64(total_size
),
1584 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
1585 .l1_size
= cpu_to_be32(s
->l1_size
),
1586 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
1587 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
1588 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
1589 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
1590 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
1592 /* Version 3 fields */
1593 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
1594 .compatible_features
= cpu_to_be64(s
->compatible_features
),
1595 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
1596 .refcount_order
= cpu_to_be32(s
->refcount_order
),
1597 .header_length
= cpu_to_be32(header_length
),
1600 /* For older versions, write a shorter header */
1601 switch (s
->qcow_version
) {
1603 ret
= offsetof(QCowHeader
, incompatible_features
);
1606 ret
= sizeof(*header
);
1615 memset(buf
, 0, buflen
);
1617 /* Preserve any unknown field in the header */
1618 if (s
->unknown_header_fields_size
) {
1619 if (buflen
< s
->unknown_header_fields_size
) {
1624 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
1625 buf
+= s
->unknown_header_fields_size
;
1626 buflen
-= s
->unknown_header_fields_size
;
1629 /* Backing file format header extension */
1630 if (s
->image_backing_format
) {
1631 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
1632 s
->image_backing_format
,
1633 strlen(s
->image_backing_format
),
1644 Qcow2Feature features
[] = {
1646 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
1647 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
1648 .name
= "dirty bit",
1651 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
1652 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
1653 .name
= "corrupt bit",
1656 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
1657 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
1658 .name
= "lazy refcounts",
1662 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
1663 features
, sizeof(features
), buflen
);
1670 /* Keep unknown header extensions */
1671 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
1672 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
1681 /* End of header extensions */
1682 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
1690 /* Backing file name */
1691 if (s
->image_backing_file
) {
1692 size_t backing_file_len
= strlen(s
->image_backing_file
);
1694 if (buflen
< backing_file_len
) {
1699 /* Using strncpy is ok here, since buf is not NUL-terminated. */
1700 strncpy(buf
, s
->image_backing_file
, buflen
);
1702 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
1703 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
1706 /* Write the new header */
1707 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
1718 static int qcow2_change_backing_file(BlockDriverState
*bs
,
1719 const char *backing_file
, const char *backing_fmt
)
1721 BDRVQcowState
*s
= bs
->opaque
;
1723 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
1724 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
1726 g_free(s
->image_backing_file
);
1727 g_free(s
->image_backing_format
);
1729 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
1730 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
1732 return qcow2_update_header(bs
);
1735 static int preallocate(BlockDriverState
*bs
)
1737 uint64_t nb_sectors
;
1739 uint64_t host_offset
= 0;
1744 nb_sectors
= bdrv_nb_sectors(bs
);
1747 while (nb_sectors
) {
1748 num
= MIN(nb_sectors
, INT_MAX
>> BDRV_SECTOR_BITS
);
1749 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &num
,
1750 &host_offset
, &meta
);
1756 QCowL2Meta
*next
= meta
->next
;
1758 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
1760 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
1761 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
1765 /* There are no dependent requests, but we need to remove our
1766 * request from the list of in-flight requests */
1767 QLIST_REMOVE(meta
, next_in_flight
);
1773 /* TODO Preallocate data if requested */
1776 offset
+= num
<< BDRV_SECTOR_BITS
;
1780 * It is expected that the image file is large enough to actually contain
1781 * all of the allocated clusters (otherwise we get failing reads after
1782 * EOF). Extend the image to the last allocated sector.
1784 if (host_offset
!= 0) {
1785 uint8_t buf
[BDRV_SECTOR_SIZE
];
1786 memset(buf
, 0, BDRV_SECTOR_SIZE
);
1787 ret
= bdrv_write(bs
->file
, (host_offset
>> BDRV_SECTOR_BITS
) + num
- 1,
1797 static int qcow2_create2(const char *filename
, int64_t total_size
,
1798 const char *backing_file
, const char *backing_format
,
1799 int flags
, size_t cluster_size
, PreallocMode prealloc
,
1800 QemuOpts
*opts
, int version
, int refcount_order
,
1803 /* Calculate cluster_bits */
1805 cluster_bits
= ffs(cluster_size
) - 1;
1806 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
1807 (1 << cluster_bits
) != cluster_size
)
1809 error_setg(errp
, "Cluster size must be a power of two between %d and "
1810 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
1815 * Open the image file and write a minimal qcow2 header.
1817 * We keep things simple and start with a zero-sized image. We also
1818 * do without refcount blocks or a L1 table for now. We'll fix the
1819 * inconsistency later.
1821 * We do need a refcount table because growing the refcount table means
1822 * allocating two new refcount blocks - the seconds of which would be at
1823 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1824 * size for any qcow2 image.
1826 BlockDriverState
* bs
;
1828 uint64_t* refcount_table
;
1829 Error
*local_err
= NULL
;
1832 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
1833 /* Note: The following calculation does not need to be exact; if it is a
1834 * bit off, either some bytes will be "leaked" (which is fine) or we
1835 * will need to increase the file size by some bytes (which is fine,
1836 * too, as long as the bulk is allocated here). Therefore, using
1837 * floating point arithmetic is fine. */
1838 int64_t meta_size
= 0;
1839 uint64_t nreftablee
, nrefblocke
, nl1e
, nl2e
;
1840 int64_t aligned_total_size
= align_offset(total_size
, cluster_size
);
1841 int refblock_bits
, refblock_size
;
1842 /* refcount entry size in bytes */
1843 double rces
= (1 << refcount_order
) / 8.;
1845 /* see qcow2_open() */
1846 refblock_bits
= cluster_bits
- (refcount_order
- 3);
1847 refblock_size
= 1 << refblock_bits
;
1849 /* header: 1 cluster */
1850 meta_size
+= cluster_size
;
1852 /* total size of L2 tables */
1853 nl2e
= aligned_total_size
/ cluster_size
;
1854 nl2e
= align_offset(nl2e
, cluster_size
/ sizeof(uint64_t));
1855 meta_size
+= nl2e
* sizeof(uint64_t);
1857 /* total size of L1 tables */
1858 nl1e
= nl2e
* sizeof(uint64_t) / cluster_size
;
1859 nl1e
= align_offset(nl1e
, cluster_size
/ sizeof(uint64_t));
1860 meta_size
+= nl1e
* sizeof(uint64_t);
1862 /* total size of refcount blocks
1864 * note: every host cluster is reference-counted, including metadata
1865 * (even refcount blocks are recursively included).
1867 * a = total_size (this is the guest disk size)
1868 * m = meta size not including refcount blocks and refcount tables
1870 * y1 = number of refcount blocks entries
1871 * y2 = meta size including everything
1872 * rces = refcount entry size in bytes
1875 * y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
1877 * y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
1879 nrefblocke
= (aligned_total_size
+ meta_size
+ cluster_size
)
1880 / (cluster_size
- rces
- rces
* sizeof(uint64_t)
1882 meta_size
+= DIV_ROUND_UP(nrefblocke
, refblock_size
) * cluster_size
;
1884 /* total size of refcount tables */
1885 nreftablee
= nrefblocke
/ refblock_size
;
1886 nreftablee
= align_offset(nreftablee
, cluster_size
/ sizeof(uint64_t));
1887 meta_size
+= nreftablee
* sizeof(uint64_t);
1889 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
,
1890 aligned_total_size
+ meta_size
, &error_abort
);
1891 qemu_opt_set(opts
, BLOCK_OPT_PREALLOC
, PreallocMode_lookup
[prealloc
],
1895 ret
= bdrv_create_file(filename
, opts
, &local_err
);
1897 error_propagate(errp
, local_err
);
1902 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
, BDRV_O_RDWR
| BDRV_O_PROTOCOL
,
1905 error_propagate(errp
, local_err
);
1909 /* Write the header */
1910 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
1911 header
= g_malloc0(cluster_size
);
1912 *header
= (QCowHeader
) {
1913 .magic
= cpu_to_be32(QCOW_MAGIC
),
1914 .version
= cpu_to_be32(version
),
1915 .cluster_bits
= cpu_to_be32(cluster_bits
),
1916 .size
= cpu_to_be64(0),
1917 .l1_table_offset
= cpu_to_be64(0),
1918 .l1_size
= cpu_to_be32(0),
1919 .refcount_table_offset
= cpu_to_be64(cluster_size
),
1920 .refcount_table_clusters
= cpu_to_be32(1),
1921 .refcount_order
= cpu_to_be32(refcount_order
),
1922 .header_length
= cpu_to_be32(sizeof(*header
)),
1925 if (flags
& BLOCK_FLAG_ENCRYPT
) {
1926 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
1928 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
1931 if (flags
& BLOCK_FLAG_LAZY_REFCOUNTS
) {
1932 header
->compatible_features
|=
1933 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
1936 ret
= bdrv_pwrite(bs
, 0, header
, cluster_size
);
1939 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
1943 /* Write a refcount table with one refcount block */
1944 refcount_table
= g_malloc0(2 * cluster_size
);
1945 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
1946 ret
= bdrv_pwrite(bs
, cluster_size
, refcount_table
, 2 * cluster_size
);
1947 g_free(refcount_table
);
1950 error_setg_errno(errp
, -ret
, "Could not write refcount table");
1958 * And now open the image and make it consistent first (i.e. increase the
1959 * refcount of the cluster that is occupied by the header and the refcount
1962 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
,
1963 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_NO_FLUSH
,
1964 &bdrv_qcow2
, &local_err
);
1966 error_propagate(errp
, local_err
);
1970 ret
= qcow2_alloc_clusters(bs
, 3 * cluster_size
);
1972 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
1973 "header and refcount table");
1976 } else if (ret
!= 0) {
1977 error_report("Huh, first cluster in empty image is already in use?");
1981 /* Okay, now that we have a valid image, let's give it the right size */
1982 ret
= bdrv_truncate(bs
, total_size
);
1984 error_setg_errno(errp
, -ret
, "Could not resize image");
1988 /* Want a backing file? There you go.*/
1990 ret
= bdrv_change_backing_file(bs
, backing_file
, backing_format
);
1992 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
1993 "with format '%s'", backing_file
, backing_format
);
1998 /* And if we're supposed to preallocate metadata, do that now */
1999 if (prealloc
!= PREALLOC_MODE_OFF
) {
2000 BDRVQcowState
*s
= bs
->opaque
;
2001 qemu_co_mutex_lock(&s
->lock
);
2002 ret
= preallocate(bs
);
2003 qemu_co_mutex_unlock(&s
->lock
);
2005 error_setg_errno(errp
, -ret
, "Could not preallocate metadata");
2013 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2014 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
,
2015 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_NO_BACKING
,
2016 &bdrv_qcow2
, &local_err
);
2018 error_propagate(errp
, local_err
);
2030 static int qcow2_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
2032 char *backing_file
= NULL
;
2033 char *backing_fmt
= NULL
;
2037 size_t cluster_size
= DEFAULT_CLUSTER_SIZE
;
2038 PreallocMode prealloc
;
2040 uint64_t refcount_bits
= 16;
2042 Error
*local_err
= NULL
;
2045 /* Read out options */
2046 size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2048 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
2049 backing_fmt
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FMT
);
2050 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_ENCRYPT
, false)) {
2051 flags
|= BLOCK_FLAG_ENCRYPT
;
2053 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
2054 DEFAULT_CLUSTER_SIZE
);
2055 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2056 prealloc
= qapi_enum_parse(PreallocMode_lookup
, buf
,
2057 PREALLOC_MODE_MAX
, PREALLOC_MODE_OFF
,
2060 error_propagate(errp
, local_err
);
2065 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
2067 /* keep the default */
2068 } else if (!strcmp(buf
, "0.10")) {
2070 } else if (!strcmp(buf
, "1.1")) {
2073 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
2078 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_LAZY_REFCOUNTS
, false)) {
2079 flags
|= BLOCK_FLAG_LAZY_REFCOUNTS
;
2082 if (backing_file
&& prealloc
!= PREALLOC_MODE_OFF
) {
2083 error_setg(errp
, "Backing file and preallocation cannot be used at "
2089 if (version
< 3 && (flags
& BLOCK_FLAG_LAZY_REFCOUNTS
)) {
2090 error_setg(errp
, "Lazy refcounts only supported with compatibility "
2091 "level 1.1 and above (use compat=1.1 or greater)");
2096 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
,
2098 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
2099 error_setg(errp
, "Refcount width must be a power of two and may not "
2105 if (version
< 3 && refcount_bits
!= 16) {
2106 error_setg(errp
, "Different refcount widths than 16 bits require "
2107 "compatibility level 1.1 or above (use compat=1.1 or "
2113 refcount_order
= ffs(refcount_bits
) - 1;
2115 ret
= qcow2_create2(filename
, size
, backing_file
, backing_fmt
, flags
,
2116 cluster_size
, prealloc
, opts
, version
, refcount_order
,
2119 error_propagate(errp
, local_err
);
2123 g_free(backing_file
);
2124 g_free(backing_fmt
);
2129 static coroutine_fn
int qcow2_co_write_zeroes(BlockDriverState
*bs
,
2130 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
2133 BDRVQcowState
*s
= bs
->opaque
;
2135 /* Emulate misaligned zero writes */
2136 if (sector_num
% s
->cluster_sectors
|| nb_sectors
% s
->cluster_sectors
) {
2140 /* Whatever is left can use real zero clusters */
2141 qemu_co_mutex_lock(&s
->lock
);
2142 ret
= qcow2_zero_clusters(bs
, sector_num
<< BDRV_SECTOR_BITS
,
2144 qemu_co_mutex_unlock(&s
->lock
);
2149 static coroutine_fn
int qcow2_co_discard(BlockDriverState
*bs
,
2150 int64_t sector_num
, int nb_sectors
)
2153 BDRVQcowState
*s
= bs
->opaque
;
2155 qemu_co_mutex_lock(&s
->lock
);
2156 ret
= qcow2_discard_clusters(bs
, sector_num
<< BDRV_SECTOR_BITS
,
2157 nb_sectors
, QCOW2_DISCARD_REQUEST
, false);
2158 qemu_co_mutex_unlock(&s
->lock
);
2162 static int qcow2_truncate(BlockDriverState
*bs
, int64_t offset
)
2164 BDRVQcowState
*s
= bs
->opaque
;
2165 int64_t new_l1_size
;
2169 error_report("The new size must be a multiple of 512");
2173 /* cannot proceed if image has snapshots */
2174 if (s
->nb_snapshots
) {
2175 error_report("Can't resize an image which has snapshots");
2179 /* shrinking is currently not supported */
2180 if (offset
< bs
->total_sectors
* 512) {
2181 error_report("qcow2 doesn't support shrinking images yet");
2185 new_l1_size
= size_to_l1(s
, offset
);
2186 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
2191 /* write updated header.size */
2192 offset
= cpu_to_be64(offset
);
2193 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
2194 &offset
, sizeof(uint64_t));
2199 s
->l1_vm_state_index
= new_l1_size
;
2203 /* XXX: put compressed sectors first, then all the cluster aligned
2204 tables to avoid losing bytes in alignment */
2205 static int qcow2_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
2206 const uint8_t *buf
, int nb_sectors
)
2208 BDRVQcowState
*s
= bs
->opaque
;
2212 uint64_t cluster_offset
;
2214 if (nb_sectors
== 0) {
2215 /* align end of file to a sector boundary to ease reading with
2216 sector based I/Os */
2217 cluster_offset
= bdrv_getlength(bs
->file
);
2218 return bdrv_truncate(bs
->file
, cluster_offset
);
2221 if (nb_sectors
!= s
->cluster_sectors
) {
2224 /* Zero-pad last write if image size is not cluster aligned */
2225 if (sector_num
+ nb_sectors
== bs
->total_sectors
&&
2226 nb_sectors
< s
->cluster_sectors
) {
2227 uint8_t *pad_buf
= qemu_blockalign(bs
, s
->cluster_size
);
2228 memset(pad_buf
, 0, s
->cluster_size
);
2229 memcpy(pad_buf
, buf
, nb_sectors
* BDRV_SECTOR_SIZE
);
2230 ret
= qcow2_write_compressed(bs
, sector_num
,
2231 pad_buf
, s
->cluster_sectors
);
2232 qemu_vfree(pad_buf
);
2237 out_buf
= g_malloc(s
->cluster_size
+ (s
->cluster_size
/ 1000) + 128);
2239 /* best compression, small window, no zlib header */
2240 memset(&strm
, 0, sizeof(strm
));
2241 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
2243 9, Z_DEFAULT_STRATEGY
);
2249 strm
.avail_in
= s
->cluster_size
;
2250 strm
.next_in
= (uint8_t *)buf
;
2251 strm
.avail_out
= s
->cluster_size
;
2252 strm
.next_out
= out_buf
;
2254 ret
= deflate(&strm
, Z_FINISH
);
2255 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
2260 out_len
= strm
.next_out
- out_buf
;
2264 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
2265 /* could not compress: write normal cluster */
2266 ret
= bdrv_write(bs
, sector_num
, buf
, s
->cluster_sectors
);
2271 cluster_offset
= qcow2_alloc_compressed_cluster_offset(bs
,
2272 sector_num
<< 9, out_len
);
2273 if (!cluster_offset
) {
2277 cluster_offset
&= s
->cluster_offset_mask
;
2279 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
);
2284 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
2285 ret
= bdrv_pwrite(bs
->file
, cluster_offset
, out_buf
, out_len
);
2297 static int make_completely_empty(BlockDriverState
*bs
)
2299 BDRVQcowState
*s
= bs
->opaque
;
2300 int ret
, l1_clusters
;
2302 uint64_t *new_reftable
= NULL
;
2303 uint64_t rt_entry
, l1_size2
;
2306 uint64_t reftable_offset
;
2307 uint32_t reftable_clusters
;
2308 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
2310 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
2315 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
2320 /* Refcounts will be broken utterly */
2321 ret
= qcow2_mark_dirty(bs
);
2326 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
2328 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
2329 l1_size2
= (uint64_t)s
->l1_size
* sizeof(uint64_t);
2331 /* After this call, neither the in-memory nor the on-disk refcount
2332 * information accurately describe the actual references */
2334 ret
= bdrv_write_zeroes(bs
->file
, s
->l1_table_offset
/ BDRV_SECTOR_SIZE
,
2335 l1_clusters
* s
->cluster_sectors
, 0);
2337 goto fail_broken_refcounts
;
2339 memset(s
->l1_table
, 0, l1_size2
);
2341 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
2343 /* Overwrite enough clusters at the beginning of the sectors to place
2344 * the refcount table, a refcount block and the L1 table in; this may
2345 * overwrite parts of the existing refcount and L1 table, which is not
2346 * an issue because the dirty flag is set, complete data loss is in fact
2347 * desired and partial data loss is consequently fine as well */
2348 ret
= bdrv_write_zeroes(bs
->file
, s
->cluster_size
/ BDRV_SECTOR_SIZE
,
2349 (2 + l1_clusters
) * s
->cluster_size
/
2350 BDRV_SECTOR_SIZE
, 0);
2351 /* This call (even if it failed overall) may have overwritten on-disk
2352 * refcount structures; in that case, the in-memory refcount information
2353 * will probably differ from the on-disk information which makes the BDS
2356 goto fail_broken_refcounts
;
2359 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
2360 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
2362 /* "Create" an empty reftable (one cluster) directly after the image
2363 * header and an empty L1 table three clusters after the image header;
2364 * the cluster between those two will be used as the first refblock */
2365 cpu_to_be64w(&l1_ofs_rt_ofs_cls
.l1_offset
, 3 * s
->cluster_size
);
2366 cpu_to_be64w(&l1_ofs_rt_ofs_cls
.reftable_offset
, s
->cluster_size
);
2367 cpu_to_be32w(&l1_ofs_rt_ofs_cls
.reftable_clusters
, 1);
2368 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
2369 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
2371 goto fail_broken_refcounts
;
2374 s
->l1_table_offset
= 3 * s
->cluster_size
;
2376 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ sizeof(uint64_t));
2377 if (!new_reftable
) {
2379 goto fail_broken_refcounts
;
2382 s
->refcount_table_offset
= s
->cluster_size
;
2383 s
->refcount_table_size
= s
->cluster_size
/ sizeof(uint64_t);
2385 g_free(s
->refcount_table
);
2386 s
->refcount_table
= new_reftable
;
2387 new_reftable
= NULL
;
2389 /* Now the in-memory refcount information again corresponds to the on-disk
2390 * information (reftable is empty and no refblocks (the refblock cache is
2391 * empty)); however, this means some clusters (e.g. the image header) are
2392 * referenced, but not refcounted, but the normal qcow2 code assumes that
2393 * the in-memory information is always correct */
2395 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
2397 /* Enter the first refblock into the reftable */
2398 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
2399 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
2400 &rt_entry
, sizeof(rt_entry
));
2402 goto fail_broken_refcounts
;
2404 s
->refcount_table
[0] = 2 * s
->cluster_size
;
2406 s
->free_cluster_index
= 0;
2407 assert(3 + l1_clusters
<= s
->refcount_block_size
);
2408 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
2411 goto fail_broken_refcounts
;
2412 } else if (offset
> 0) {
2413 error_report("First cluster in emptied image is in use");
2417 /* Now finally the in-memory information corresponds to the on-disk
2418 * structures and is correct */
2419 ret
= qcow2_mark_clean(bs
);
2424 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
);
2431 fail_broken_refcounts
:
2432 /* The BDS is unusable at this point. If we wanted to make it usable, we
2433 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2434 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2435 * again. However, because the functions which could have caused this error
2436 * path to be taken are used by those functions as well, it's very likely
2437 * that that sequence will fail as well. Therefore, just eject the BDS. */
2441 g_free(new_reftable
);
2445 static int qcow2_make_empty(BlockDriverState
*bs
)
2447 BDRVQcowState
*s
= bs
->opaque
;
2448 uint64_t start_sector
;
2449 int sector_step
= INT_MAX
/ BDRV_SECTOR_SIZE
;
2450 int l1_clusters
, ret
= 0;
2452 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
2454 if (s
->qcow_version
>= 3 && !s
->snapshots
&&
2455 3 + l1_clusters
<= s
->refcount_block_size
) {
2456 /* The following function only works for qcow2 v3 images (it requires
2457 * the dirty flag) and only as long as there are no snapshots (because
2458 * it completely empties the image). Furthermore, the L1 table and three
2459 * additional clusters (image header, refcount table, one refcount
2460 * block) have to fit inside one refcount block. */
2461 return make_completely_empty(bs
);
2464 /* This fallback code simply discards every active cluster; this is slow,
2465 * but works in all cases */
2466 for (start_sector
= 0; start_sector
< bs
->total_sectors
;
2467 start_sector
+= sector_step
)
2469 /* As this function is generally used after committing an external
2470 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2471 * default action for this kind of discard is to pass the discard,
2472 * which will ideally result in an actually smaller image file, as
2473 * is probably desired. */
2474 ret
= qcow2_discard_clusters(bs
, start_sector
* BDRV_SECTOR_SIZE
,
2476 bs
->total_sectors
- start_sector
),
2477 QCOW2_DISCARD_SNAPSHOT
, true);
2486 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
2488 BDRVQcowState
*s
= bs
->opaque
;
2491 qemu_co_mutex_lock(&s
->lock
);
2492 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2494 qemu_co_mutex_unlock(&s
->lock
);
2498 if (qcow2_need_accurate_refcounts(s
)) {
2499 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2501 qemu_co_mutex_unlock(&s
->lock
);
2505 qemu_co_mutex_unlock(&s
->lock
);
2510 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2512 BDRVQcowState
*s
= bs
->opaque
;
2513 bdi
->unallocated_blocks_are_zero
= true;
2514 bdi
->can_write_zeroes_with_unmap
= (s
->qcow_version
>= 3);
2515 bdi
->cluster_size
= s
->cluster_size
;
2516 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
2520 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
)
2522 BDRVQcowState
*s
= bs
->opaque
;
2523 ImageInfoSpecific
*spec_info
= g_new(ImageInfoSpecific
, 1);
2525 *spec_info
= (ImageInfoSpecific
){
2526 .kind
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
2528 .qcow2
= g_new(ImageInfoSpecificQCow2
, 1),
2531 if (s
->qcow_version
== 2) {
2532 *spec_info
->qcow2
= (ImageInfoSpecificQCow2
){
2533 .compat
= g_strdup("0.10"),
2534 .refcount_bits
= s
->refcount_bits
,
2536 } else if (s
->qcow_version
== 3) {
2537 *spec_info
->qcow2
= (ImageInfoSpecificQCow2
){
2538 .compat
= g_strdup("1.1"),
2539 .lazy_refcounts
= s
->compatible_features
&
2540 QCOW2_COMPAT_LAZY_REFCOUNTS
,
2541 .has_lazy_refcounts
= true,
2542 .corrupt
= s
->incompatible_features
&
2543 QCOW2_INCOMPAT_CORRUPT
,
2544 .has_corrupt
= true,
2545 .refcount_bits
= s
->refcount_bits
,
2553 static void dump_refcounts(BlockDriverState
*bs
)
2555 BDRVQcowState
*s
= bs
->opaque
;
2556 int64_t nb_clusters
, k
, k1
, size
;
2559 size
= bdrv_getlength(bs
->file
);
2560 nb_clusters
= size_to_clusters(s
, size
);
2561 for(k
= 0; k
< nb_clusters
;) {
2563 refcount
= get_refcount(bs
, k
);
2565 while (k
< nb_clusters
&& get_refcount(bs
, k
) == refcount
)
2567 printf("%" PRId64
": refcount=%d nb=%" PRId64
"\n", k
, refcount
,
2573 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2576 BDRVQcowState
*s
= bs
->opaque
;
2577 int64_t total_sectors
= bs
->total_sectors
;
2578 bool zero_beyond_eof
= bs
->zero_beyond_eof
;
2581 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
2582 bs
->zero_beyond_eof
= false;
2583 ret
= bdrv_pwritev(bs
, qcow2_vm_state_offset(s
) + pos
, qiov
);
2584 bs
->zero_beyond_eof
= zero_beyond_eof
;
2586 /* bdrv_co_do_writev will have increased the total_sectors value to include
2587 * the VM state - the VM state is however not an actual part of the block
2588 * device, therefore, we need to restore the old value. */
2589 bs
->total_sectors
= total_sectors
;
2594 static int qcow2_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2595 int64_t pos
, int size
)
2597 BDRVQcowState
*s
= bs
->opaque
;
2598 bool zero_beyond_eof
= bs
->zero_beyond_eof
;
2601 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
2602 bs
->zero_beyond_eof
= false;
2603 ret
= bdrv_pread(bs
, qcow2_vm_state_offset(s
) + pos
, buf
, size
);
2604 bs
->zero_beyond_eof
= zero_beyond_eof
;
2610 * Downgrades an image's version. To achieve this, any incompatible features
2611 * have to be removed.
2613 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
2614 BlockDriverAmendStatusCB
*status_cb
)
2616 BDRVQcowState
*s
= bs
->opaque
;
2617 int current_version
= s
->qcow_version
;
2620 if (target_version
== current_version
) {
2622 } else if (target_version
> current_version
) {
2624 } else if (target_version
!= 2) {
2628 if (s
->refcount_order
!= 4) {
2629 /* we would have to convert the image to a refcount_order == 4 image
2630 * here; however, since qemu (at the time of writing this) does not
2631 * support anything different than 4 anyway, there is no point in doing
2632 * so right now; however, we should error out (if qemu supports this in
2633 * the future and this code has not been adapted) */
2634 error_report("qcow2_downgrade: Image refcount orders other than 4 are "
2635 "currently not supported.");
2639 /* clear incompatible features */
2640 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
2641 ret
= qcow2_mark_clean(bs
);
2647 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2648 * the first place; if that happens nonetheless, returning -ENOTSUP is the
2649 * best thing to do anyway */
2651 if (s
->incompatible_features
) {
2655 /* since we can ignore compatible features, we can set them to 0 as well */
2656 s
->compatible_features
= 0;
2657 /* if lazy refcounts have been used, they have already been fixed through
2658 * clearing the dirty flag */
2660 /* clearing autoclear features is trivial */
2661 s
->autoclear_features
= 0;
2663 ret
= qcow2_expand_zero_clusters(bs
, status_cb
);
2668 s
->qcow_version
= target_version
;
2669 ret
= qcow2_update_header(bs
);
2671 s
->qcow_version
= current_version
;
2677 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
2678 BlockDriverAmendStatusCB
*status_cb
)
2680 BDRVQcowState
*s
= bs
->opaque
;
2681 int old_version
= s
->qcow_version
, new_version
= old_version
;
2682 uint64_t new_size
= 0;
2683 const char *backing_file
= NULL
, *backing_format
= NULL
;
2684 bool lazy_refcounts
= s
->use_lazy_refcounts
;
2685 const char *compat
= NULL
;
2686 uint64_t cluster_size
= s
->cluster_size
;
2689 QemuOptDesc
*desc
= opts
->list
->desc
;
2691 while (desc
&& desc
->name
) {
2692 if (!qemu_opt_find(opts
, desc
->name
)) {
2693 /* only change explicitly defined options */
2698 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
2699 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
2701 /* preserve default */
2702 } else if (!strcmp(compat
, "0.10")) {
2704 } else if (!strcmp(compat
, "1.1")) {
2707 fprintf(stderr
, "Unknown compatibility level %s.\n", compat
);
2710 } else if (!strcmp(desc
->name
, BLOCK_OPT_PREALLOC
)) {
2711 fprintf(stderr
, "Cannot change preallocation mode.\n");
2713 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
2714 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
2715 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
2716 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
2717 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
2718 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
2719 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT
)) {
2720 encrypt
= qemu_opt_get_bool(opts
, BLOCK_OPT_ENCRYPT
,
2722 if (encrypt
!= !!s
->crypt_method
) {
2723 fprintf(stderr
, "Changing the encryption flag is not "
2727 } else if (!strcmp(desc
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
2728 cluster_size
= qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
,
2730 if (cluster_size
!= s
->cluster_size
) {
2731 fprintf(stderr
, "Changing the cluster size is not "
2735 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
2736 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
2738 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
2739 error_report("Cannot change refcount entry width");
2742 /* if this assertion fails, this probably means a new option was
2743 * added without having it covered here */
2750 if (new_version
!= old_version
) {
2751 if (new_version
> old_version
) {
2753 s
->qcow_version
= new_version
;
2754 ret
= qcow2_update_header(bs
);
2756 s
->qcow_version
= old_version
;
2760 ret
= qcow2_downgrade(bs
, new_version
, status_cb
);
2767 if (backing_file
|| backing_format
) {
2768 ret
= qcow2_change_backing_file(bs
,
2769 backing_file
?: s
->image_backing_file
,
2770 backing_format
?: s
->image_backing_format
);
2776 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
2777 if (lazy_refcounts
) {
2778 if (s
->qcow_version
< 3) {
2779 fprintf(stderr
, "Lazy refcounts only supported with compatibility "
2780 "level 1.1 and above (use compat=1.1 or greater)\n");
2783 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
2784 ret
= qcow2_update_header(bs
);
2786 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
2789 s
->use_lazy_refcounts
= true;
2791 /* make image clean first */
2792 ret
= qcow2_mark_clean(bs
);
2796 /* now disallow lazy refcounts */
2797 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
2798 ret
= qcow2_update_header(bs
);
2800 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
2803 s
->use_lazy_refcounts
= false;
2808 ret
= bdrv_truncate(bs
, new_size
);
2818 * If offset or size are negative, respectively, they will not be included in
2819 * the BLOCK_IMAGE_CORRUPTED event emitted.
2820 * fatal will be ignored for read-only BDS; corruptions found there will always
2821 * be considered non-fatal.
2823 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
2824 int64_t size
, const char *message_format
, ...)
2826 BDRVQcowState
*s
= bs
->opaque
;
2830 fatal
= fatal
&& !bs
->read_only
;
2832 if (s
->signaled_corruption
&&
2833 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
2838 va_start(ap
, message_format
);
2839 message
= g_strdup_vprintf(message_format
, ap
);
2843 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
2844 "corruption events will be suppressed\n", message
);
2846 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
2847 "corruption events will be suppressed\n", message
);
2850 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
), message
,
2851 offset
>= 0, offset
, size
>= 0, size
,
2852 fatal
, &error_abort
);
2856 qcow2_mark_corrupt(bs
);
2857 bs
->drv
= NULL
; /* make BDS unusable */
2860 s
->signaled_corruption
= true;
2863 static QemuOptsList qcow2_create_opts
= {
2864 .name
= "qcow2-create-opts",
2865 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
2868 .name
= BLOCK_OPT_SIZE
,
2869 .type
= QEMU_OPT_SIZE
,
2870 .help
= "Virtual disk size"
2873 .name
= BLOCK_OPT_COMPAT_LEVEL
,
2874 .type
= QEMU_OPT_STRING
,
2875 .help
= "Compatibility level (0.10 or 1.1)"
2878 .name
= BLOCK_OPT_BACKING_FILE
,
2879 .type
= QEMU_OPT_STRING
,
2880 .help
= "File name of a base image"
2883 .name
= BLOCK_OPT_BACKING_FMT
,
2884 .type
= QEMU_OPT_STRING
,
2885 .help
= "Image format of the base image"
2888 .name
= BLOCK_OPT_ENCRYPT
,
2889 .type
= QEMU_OPT_BOOL
,
2890 .help
= "Encrypt the image",
2891 .def_value_str
= "off"
2894 .name
= BLOCK_OPT_CLUSTER_SIZE
,
2895 .type
= QEMU_OPT_SIZE
,
2896 .help
= "qcow2 cluster size",
2897 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
)
2900 .name
= BLOCK_OPT_PREALLOC
,
2901 .type
= QEMU_OPT_STRING
,
2902 .help
= "Preallocation mode (allowed values: off, metadata, "
2906 .name
= BLOCK_OPT_LAZY_REFCOUNTS
,
2907 .type
= QEMU_OPT_BOOL
,
2908 .help
= "Postpone refcount updates",
2909 .def_value_str
= "off"
2912 .name
= BLOCK_OPT_REFCOUNT_BITS
,
2913 .type
= QEMU_OPT_NUMBER
,
2914 .help
= "Width of a reference count entry in bits",
2915 .def_value_str
= "16"
2917 { /* end of list */ }
2921 BlockDriver bdrv_qcow2
= {
2922 .format_name
= "qcow2",
2923 .instance_size
= sizeof(BDRVQcowState
),
2924 .bdrv_probe
= qcow2_probe
,
2925 .bdrv_open
= qcow2_open
,
2926 .bdrv_close
= qcow2_close
,
2927 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
2928 .bdrv_create
= qcow2_create
,
2929 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2930 .bdrv_co_get_block_status
= qcow2_co_get_block_status
,
2931 .bdrv_set_key
= qcow2_set_key
,
2933 .bdrv_co_readv
= qcow2_co_readv
,
2934 .bdrv_co_writev
= qcow2_co_writev
,
2935 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
2937 .bdrv_co_write_zeroes
= qcow2_co_write_zeroes
,
2938 .bdrv_co_discard
= qcow2_co_discard
,
2939 .bdrv_truncate
= qcow2_truncate
,
2940 .bdrv_write_compressed
= qcow2_write_compressed
,
2941 .bdrv_make_empty
= qcow2_make_empty
,
2943 .bdrv_snapshot_create
= qcow2_snapshot_create
,
2944 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
2945 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
2946 .bdrv_snapshot_list
= qcow2_snapshot_list
,
2947 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
2948 .bdrv_get_info
= qcow2_get_info
,
2949 .bdrv_get_specific_info
= qcow2_get_specific_info
,
2951 .bdrv_save_vmstate
= qcow2_save_vmstate
,
2952 .bdrv_load_vmstate
= qcow2_load_vmstate
,
2954 .supports_backing
= true,
2955 .bdrv_change_backing_file
= qcow2_change_backing_file
,
2957 .bdrv_refresh_limits
= qcow2_refresh_limits
,
2958 .bdrv_invalidate_cache
= qcow2_invalidate_cache
,
2960 .create_opts
= &qcow2_create_opts
,
2961 .bdrv_check
= qcow2_check
,
2962 .bdrv_amend_options
= qcow2_amend_options
,
2965 static void bdrv_qcow2_init(void)
2967 bdrv_register(&bdrv_qcow2
);
2970 block_init(bdrv_qcow2_init
);