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"
36 Differences with QCOW:
38 - Support for multiple incremental snapshots.
39 - Memory management by reference counts.
40 - Clusters which have a reference count of one have the bit
41 QCOW_OFLAG_COPIED to optimize write performance.
42 - Size of compressed clusters is stored in sectors to reduce bit usage
43 in the cluster offsets.
44 - Support for storing additional data (such as the VM state) in the
46 - If a backing store is used, the cluster size is not constrained
47 (could be backported to QCOW).
48 - L2 tables have always a size of one cluster.
55 } QEMU_PACKED QCowExtension
;
57 #define QCOW2_EXT_MAGIC_END 0
58 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
59 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
61 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
63 const QCowHeader
*cow_header
= (const void *)buf
;
65 if (buf_size
>= sizeof(QCowHeader
) &&
66 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
67 be32_to_cpu(cow_header
->version
) >= 2)
75 * read qcow2 extension and fill bs
76 * start reading from start_offset
77 * finish reading upon magic of value 0 or when end_offset reached
78 * unknown magic is skipped (future extension this version knows nothing about)
79 * return 0 upon success, non-0 otherwise
81 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
82 uint64_t end_offset
, void **p_feature_table
,
85 BDRVQcowState
*s
= bs
->opaque
;
91 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
93 offset
= start_offset
;
94 while (offset
< end_offset
) {
98 if (offset
> s
->cluster_size
)
99 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
101 printf("attempting to read extended header in offset %lu\n", offset
);
104 ret
= bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
));
106 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
107 "pread fail from offset %" PRIu64
, offset
);
110 be32_to_cpus(&ext
.magic
);
111 be32_to_cpus(&ext
.len
);
112 offset
+= sizeof(ext
);
114 printf("ext.magic = 0x%x\n", ext
.magic
);
116 if (ext
.len
> end_offset
- offset
) {
117 error_setg(errp
, "Header extension too large");
122 case QCOW2_EXT_MAGIC_END
:
125 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
126 if (ext
.len
>= sizeof(bs
->backing_format
)) {
127 error_setg(errp
, "ERROR: ext_backing_format: len=%u too large"
128 " (>=%zu)", ext
.len
, sizeof(bs
->backing_format
));
131 ret
= bdrv_pread(bs
->file
, offset
, bs
->backing_format
, ext
.len
);
133 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
134 "Could not read format name");
137 bs
->backing_format
[ext
.len
] = '\0';
139 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
143 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
144 if (p_feature_table
!= NULL
) {
145 void* feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
146 ret
= bdrv_pread(bs
->file
, offset
, feature_table
, ext
.len
);
148 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
149 "Could not read table");
153 *p_feature_table
= feature_table
;
158 /* unknown magic - save it in case we need to rewrite the header */
160 Qcow2UnknownHeaderExtension
*uext
;
162 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
163 uext
->magic
= ext
.magic
;
165 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
167 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
169 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
170 "Could not read data");
177 offset
+= ((ext
.len
+ 7) & ~7);
183 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
185 BDRVQcowState
*s
= bs
->opaque
;
186 Qcow2UnknownHeaderExtension
*uext
, *next
;
188 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
189 QLIST_REMOVE(uext
, next
);
194 static void GCC_FMT_ATTR(3, 4) report_unsupported(BlockDriverState
*bs
,
195 Error
**errp
, const char *fmt
, ...)
201 vsnprintf(msg
, sizeof(msg
), fmt
, ap
);
204 error_set(errp
, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
, bs
->device_name
, "qcow2",
208 static void report_unsupported_feature(BlockDriverState
*bs
,
209 Error
**errp
, Qcow2Feature
*table
, uint64_t mask
)
211 while (table
&& table
->name
[0] != '\0') {
212 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
213 if (mask
& (1 << table
->bit
)) {
214 report_unsupported(bs
, errp
, "%.46s", table
->name
);
215 mask
&= ~(1 << table
->bit
);
222 report_unsupported(bs
, errp
, "Unknown incompatible feature: %" PRIx64
,
228 * Sets the dirty bit and flushes afterwards if necessary.
230 * The incompatible_features bit is only set if the image file header was
231 * updated successfully. Therefore it is not required to check the return
232 * value of this function.
234 int qcow2_mark_dirty(BlockDriverState
*bs
)
236 BDRVQcowState
*s
= bs
->opaque
;
240 assert(s
->qcow_version
>= 3);
242 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
243 return 0; /* already dirty */
246 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
247 ret
= bdrv_pwrite(bs
->file
, offsetof(QCowHeader
, incompatible_features
),
252 ret
= bdrv_flush(bs
->file
);
257 /* Only treat image as dirty if the header was updated successfully */
258 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
263 * Clears the dirty bit and flushes before if necessary. Only call this
264 * function when there are no pending requests, it does not guard against
265 * concurrent requests dirtying the image.
267 static int qcow2_mark_clean(BlockDriverState
*bs
)
269 BDRVQcowState
*s
= bs
->opaque
;
271 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
272 int ret
= bdrv_flush(bs
);
277 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
278 return qcow2_update_header(bs
);
284 * Marks the image as corrupt.
286 int qcow2_mark_corrupt(BlockDriverState
*bs
)
288 BDRVQcowState
*s
= bs
->opaque
;
290 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
291 return qcow2_update_header(bs
);
295 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
296 * before if necessary.
298 int qcow2_mark_consistent(BlockDriverState
*bs
)
300 BDRVQcowState
*s
= bs
->opaque
;
302 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
303 int ret
= bdrv_flush(bs
);
308 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
309 return qcow2_update_header(bs
);
314 static int qcow2_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
317 int ret
= qcow2_check_refcounts(bs
, result
, fix
);
322 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
323 ret
= qcow2_mark_clean(bs
);
327 return qcow2_mark_consistent(bs
);
332 static int validate_table_offset(BlockDriverState
*bs
, uint64_t offset
,
333 uint64_t entries
, size_t entry_len
)
335 BDRVQcowState
*s
= bs
->opaque
;
338 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
339 * because values will be passed to qemu functions taking int64_t. */
340 if (entries
> INT64_MAX
/ entry_len
) {
344 size
= entries
* entry_len
;
346 if (INT64_MAX
- size
< offset
) {
350 /* Tables must be cluster aligned */
351 if (offset
& (s
->cluster_size
- 1)) {
358 static QemuOptsList qcow2_runtime_opts
= {
360 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
363 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
364 .type
= QEMU_OPT_BOOL
,
365 .help
= "Postpone refcount updates",
368 .name
= QCOW2_OPT_DISCARD_REQUEST
,
369 .type
= QEMU_OPT_BOOL
,
370 .help
= "Pass guest discard requests to the layer below",
373 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
374 .type
= QEMU_OPT_BOOL
,
375 .help
= "Generate discard requests when snapshot related space "
379 .name
= QCOW2_OPT_DISCARD_OTHER
,
380 .type
= QEMU_OPT_BOOL
,
381 .help
= "Generate discard requests when other clusters are freed",
384 .name
= QCOW2_OPT_OVERLAP
,
385 .type
= QEMU_OPT_STRING
,
386 .help
= "Selects which overlap checks to perform from a range of "
387 "templates (none, constant, cached, all)",
390 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
391 .type
= QEMU_OPT_BOOL
,
392 .help
= "Check for unintended writes into the main qcow2 header",
395 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
396 .type
= QEMU_OPT_BOOL
,
397 .help
= "Check for unintended writes into the active L1 table",
400 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
401 .type
= QEMU_OPT_BOOL
,
402 .help
= "Check for unintended writes into an active L2 table",
405 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
406 .type
= QEMU_OPT_BOOL
,
407 .help
= "Check for unintended writes into the refcount table",
410 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
411 .type
= QEMU_OPT_BOOL
,
412 .help
= "Check for unintended writes into a refcount block",
415 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
416 .type
= QEMU_OPT_BOOL
,
417 .help
= "Check for unintended writes into the snapshot table",
420 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
421 .type
= QEMU_OPT_BOOL
,
422 .help
= "Check for unintended writes into an inactive L1 table",
425 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
426 .type
= QEMU_OPT_BOOL
,
427 .help
= "Check for unintended writes into an inactive L2 table",
429 { /* end of list */ }
433 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
434 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
435 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
436 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
437 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
438 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
439 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
440 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
441 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
444 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
447 BDRVQcowState
*s
= bs
->opaque
;
452 Error
*local_err
= NULL
;
454 uint64_t l1_vm_state_index
;
455 const char *opt_overlap_check
;
456 int overlap_check_template
= 0;
458 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
460 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
463 be32_to_cpus(&header
.magic
);
464 be32_to_cpus(&header
.version
);
465 be64_to_cpus(&header
.backing_file_offset
);
466 be32_to_cpus(&header
.backing_file_size
);
467 be64_to_cpus(&header
.size
);
468 be32_to_cpus(&header
.cluster_bits
);
469 be32_to_cpus(&header
.crypt_method
);
470 be64_to_cpus(&header
.l1_table_offset
);
471 be32_to_cpus(&header
.l1_size
);
472 be64_to_cpus(&header
.refcount_table_offset
);
473 be32_to_cpus(&header
.refcount_table_clusters
);
474 be64_to_cpus(&header
.snapshots_offset
);
475 be32_to_cpus(&header
.nb_snapshots
);
477 if (header
.magic
!= QCOW_MAGIC
) {
478 error_setg(errp
, "Image is not in qcow2 format");
482 if (header
.version
< 2 || header
.version
> 3) {
483 report_unsupported(bs
, errp
, "QCOW version %d", header
.version
);
488 s
->qcow_version
= header
.version
;
490 /* Initialise cluster size */
491 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
492 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
493 error_setg(errp
, "Unsupported cluster size: 2^%i", header
.cluster_bits
);
498 s
->cluster_bits
= header
.cluster_bits
;
499 s
->cluster_size
= 1 << s
->cluster_bits
;
500 s
->cluster_sectors
= 1 << (s
->cluster_bits
- 9);
502 /* Initialise version 3 header fields */
503 if (header
.version
== 2) {
504 header
.incompatible_features
= 0;
505 header
.compatible_features
= 0;
506 header
.autoclear_features
= 0;
507 header
.refcount_order
= 4;
508 header
.header_length
= 72;
510 be64_to_cpus(&header
.incompatible_features
);
511 be64_to_cpus(&header
.compatible_features
);
512 be64_to_cpus(&header
.autoclear_features
);
513 be32_to_cpus(&header
.refcount_order
);
514 be32_to_cpus(&header
.header_length
);
516 if (header
.header_length
< 104) {
517 error_setg(errp
, "qcow2 header too short");
523 if (header
.header_length
> s
->cluster_size
) {
524 error_setg(errp
, "qcow2 header exceeds cluster size");
529 if (header
.header_length
> sizeof(header
)) {
530 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
531 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
532 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
533 s
->unknown_header_fields_size
);
535 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
541 if (header
.backing_file_offset
> s
->cluster_size
) {
542 error_setg(errp
, "Invalid backing file offset");
547 if (header
.backing_file_offset
) {
548 ext_end
= header
.backing_file_offset
;
550 ext_end
= 1 << header
.cluster_bits
;
553 /* Handle feature bits */
554 s
->incompatible_features
= header
.incompatible_features
;
555 s
->compatible_features
= header
.compatible_features
;
556 s
->autoclear_features
= header
.autoclear_features
;
558 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
559 void *feature_table
= NULL
;
560 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
561 &feature_table
, NULL
);
562 report_unsupported_feature(bs
, errp
, feature_table
,
563 s
->incompatible_features
&
564 ~QCOW2_INCOMPAT_MASK
);
566 g_free(feature_table
);
570 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
571 /* Corrupt images may not be written to unless they are being repaired
573 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
574 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
581 /* Check support for various header values */
582 if (header
.refcount_order
!= 4) {
583 report_unsupported(bs
, errp
, "%d bit reference counts",
584 1 << header
.refcount_order
);
588 s
->refcount_order
= header
.refcount_order
;
590 if (header
.crypt_method
> QCOW_CRYPT_AES
) {
591 error_setg(errp
, "Unsupported encryption method: %i",
592 header
.crypt_method
);
596 s
->crypt_method_header
= header
.crypt_method
;
597 if (s
->crypt_method_header
) {
601 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
602 s
->l2_size
= 1 << s
->l2_bits
;
603 bs
->total_sectors
= header
.size
/ 512;
604 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
605 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
606 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
608 s
->refcount_table_offset
= header
.refcount_table_offset
;
609 s
->refcount_table_size
=
610 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
612 if (header
.refcount_table_clusters
> qcow2_max_refcount_clusters(s
)) {
613 error_setg(errp
, "Reference count table too large");
618 ret
= validate_table_offset(bs
, s
->refcount_table_offset
,
619 s
->refcount_table_size
, sizeof(uint64_t));
621 error_setg(errp
, "Invalid reference count table offset");
625 /* Snapshot table offset/length */
626 if (header
.nb_snapshots
> QCOW_MAX_SNAPSHOTS
) {
627 error_setg(errp
, "Too many snapshots");
632 ret
= validate_table_offset(bs
, header
.snapshots_offset
,
634 sizeof(QCowSnapshotHeader
));
636 error_setg(errp
, "Invalid snapshot table offset");
640 /* read the level 1 table */
641 if (header
.l1_size
> 0x2000000) {
642 /* 32 MB L1 table is enough for 2 PB images at 64k cluster size
643 * (128 GB for 512 byte clusters, 2 EB for 2 MB clusters) */
644 error_setg(errp
, "Active L1 table too large");
648 s
->l1_size
= header
.l1_size
;
650 l1_vm_state_index
= size_to_l1(s
, header
.size
);
651 if (l1_vm_state_index
> INT_MAX
) {
652 error_setg(errp
, "Image is too big");
656 s
->l1_vm_state_index
= l1_vm_state_index
;
658 /* the L1 table must contain at least enough entries to put
660 if (s
->l1_size
< s
->l1_vm_state_index
) {
661 error_setg(errp
, "L1 table is too small");
666 ret
= validate_table_offset(bs
, header
.l1_table_offset
,
667 header
.l1_size
, sizeof(uint64_t));
669 error_setg(errp
, "Invalid L1 table offset");
672 s
->l1_table_offset
= header
.l1_table_offset
;
675 if (s
->l1_size
> 0) {
676 s
->l1_table
= g_malloc0(
677 align_offset(s
->l1_size
* sizeof(uint64_t), 512));
678 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
679 s
->l1_size
* sizeof(uint64_t));
681 error_setg_errno(errp
, -ret
, "Could not read L1 table");
684 for(i
= 0;i
< s
->l1_size
; i
++) {
685 be64_to_cpus(&s
->l1_table
[i
]);
689 /* alloc L2 table/refcount block cache */
690 s
->l2_table_cache
= qcow2_cache_create(bs
, L2_CACHE_SIZE
);
691 s
->refcount_block_cache
= qcow2_cache_create(bs
, REFCOUNT_CACHE_SIZE
);
693 s
->cluster_cache
= g_malloc(s
->cluster_size
);
694 /* one more sector for decompressed data alignment */
695 s
->cluster_data
= qemu_blockalign(bs
, QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
697 s
->cluster_cache_offset
= -1;
700 ret
= qcow2_refcount_init(bs
);
702 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
706 QLIST_INIT(&s
->cluster_allocs
);
707 QTAILQ_INIT(&s
->discards
);
709 /* read qcow2 extensions */
710 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
712 error_propagate(errp
, local_err
);
717 /* read the backing file name */
718 if (header
.backing_file_offset
!= 0) {
719 len
= header
.backing_file_size
;
720 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
)) {
721 error_setg(errp
, "Backing file name too long");
725 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
726 bs
->backing_file
, len
);
728 error_setg_errno(errp
, -ret
, "Could not read backing file name");
731 bs
->backing_file
[len
] = '\0';
734 /* Internal snapshots */
735 s
->snapshots_offset
= header
.snapshots_offset
;
736 s
->nb_snapshots
= header
.nb_snapshots
;
738 ret
= qcow2_read_snapshots(bs
);
740 error_setg_errno(errp
, -ret
, "Could not read snapshots");
744 /* Clear unknown autoclear feature bits */
745 if (!bs
->read_only
&& !(flags
& BDRV_O_INCOMING
) && s
->autoclear_features
) {
746 s
->autoclear_features
= 0;
747 ret
= qcow2_update_header(bs
);
749 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
754 /* Initialise locks */
755 qemu_co_mutex_init(&s
->lock
);
757 /* Repair image if dirty */
758 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INCOMING
)) && !bs
->read_only
&&
759 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
760 BdrvCheckResult result
= {0};
762 ret
= qcow2_check(bs
, &result
, BDRV_FIX_ERRORS
);
764 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
769 /* Enable lazy_refcounts according to image and command line options */
770 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
771 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
773 error_propagate(errp
, local_err
);
778 s
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
779 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
781 s
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
782 s
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
783 s
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
784 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
785 flags
& BDRV_O_UNMAP
);
786 s
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
787 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
788 s
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
789 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
791 opt_overlap_check
= qemu_opt_get(opts
, "overlap-check") ?: "cached";
792 if (!strcmp(opt_overlap_check
, "none")) {
793 overlap_check_template
= 0;
794 } else if (!strcmp(opt_overlap_check
, "constant")) {
795 overlap_check_template
= QCOW2_OL_CONSTANT
;
796 } else if (!strcmp(opt_overlap_check
, "cached")) {
797 overlap_check_template
= QCOW2_OL_CACHED
;
798 } else if (!strcmp(opt_overlap_check
, "all")) {
799 overlap_check_template
= QCOW2_OL_ALL
;
801 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
802 "'overlap-check'. Allowed are either of the following: "
803 "none, constant, cached, all", opt_overlap_check
);
809 s
->overlap_check
= 0;
810 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
811 /* overlap-check defines a template bitmask, but every flag may be
812 * overwritten through the associated boolean option */
814 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
815 overlap_check_template
& (1 << i
)) << i
;
820 if (s
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
821 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
822 "qemu 1.1 compatibility level");
829 BdrvCheckResult result
= {0};
830 qcow2_check_refcounts(bs
, &result
, 0);
836 g_free(s
->unknown_header_fields
);
837 cleanup_unknown_header_ext(bs
);
838 qcow2_free_snapshots(bs
);
839 qcow2_refcount_close(bs
);
841 /* else pre-write overlap checks in cache_destroy may crash */
843 if (s
->l2_table_cache
) {
844 qcow2_cache_destroy(bs
, s
->l2_table_cache
);
846 if (s
->refcount_block_cache
) {
847 qcow2_cache_destroy(bs
, s
->refcount_block_cache
);
849 g_free(s
->cluster_cache
);
850 qemu_vfree(s
->cluster_data
);
854 static int qcow2_refresh_limits(BlockDriverState
*bs
)
856 BDRVQcowState
*s
= bs
->opaque
;
858 bs
->bl
.write_zeroes_alignment
= s
->cluster_sectors
;
863 static int qcow2_set_key(BlockDriverState
*bs
, const char *key
)
865 BDRVQcowState
*s
= bs
->opaque
;
869 memset(keybuf
, 0, 16);
873 /* XXX: we could compress the chars to 7 bits to increase
875 for(i
= 0;i
< len
;i
++) {
878 s
->crypt_method
= s
->crypt_method_header
;
880 if (AES_set_encrypt_key(keybuf
, 128, &s
->aes_encrypt_key
) != 0)
882 if (AES_set_decrypt_key(keybuf
, 128, &s
->aes_decrypt_key
) != 0)
892 AES_encrypt(in
, tmp
, &s
->aes_encrypt_key
);
893 AES_decrypt(tmp
, out
, &s
->aes_decrypt_key
);
894 for(i
= 0; i
< 16; i
++)
895 printf(" %02x", tmp
[i
]);
897 for(i
= 0; i
< 16; i
++)
898 printf(" %02x", out
[i
]);
905 /* We have nothing to do for QCOW2 reopen, stubs just return
907 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
908 BlockReopenQueue
*queue
, Error
**errp
)
913 static int64_t coroutine_fn
qcow2_co_get_block_status(BlockDriverState
*bs
,
914 int64_t sector_num
, int nb_sectors
, int *pnum
)
916 BDRVQcowState
*s
= bs
->opaque
;
917 uint64_t cluster_offset
;
918 int index_in_cluster
, ret
;
922 qemu_co_mutex_lock(&s
->lock
);
923 ret
= qcow2_get_cluster_offset(bs
, sector_num
<< 9, pnum
, &cluster_offset
);
924 qemu_co_mutex_unlock(&s
->lock
);
929 if (cluster_offset
!= 0 && ret
!= QCOW2_CLUSTER_COMPRESSED
&&
931 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
932 cluster_offset
|= (index_in_cluster
<< BDRV_SECTOR_BITS
);
933 status
|= BDRV_BLOCK_OFFSET_VALID
| cluster_offset
;
935 if (ret
== QCOW2_CLUSTER_ZERO
) {
936 status
|= BDRV_BLOCK_ZERO
;
937 } else if (ret
!= QCOW2_CLUSTER_UNALLOCATED
) {
938 status
|= BDRV_BLOCK_DATA
;
943 /* handle reading after the end of the backing file */
944 int qcow2_backing_read1(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
945 int64_t sector_num
, int nb_sectors
)
948 if ((sector_num
+ nb_sectors
) <= bs
->total_sectors
)
950 if (sector_num
>= bs
->total_sectors
)
953 n1
= bs
->total_sectors
- sector_num
;
955 qemu_iovec_memset(qiov
, 512 * n1
, 0, 512 * (nb_sectors
- n1
));
960 static coroutine_fn
int qcow2_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
961 int remaining_sectors
, QEMUIOVector
*qiov
)
963 BDRVQcowState
*s
= bs
->opaque
;
964 int index_in_cluster
, n1
;
966 int cur_nr_sectors
; /* number of sectors in current iteration */
967 uint64_t cluster_offset
= 0;
968 uint64_t bytes_done
= 0;
969 QEMUIOVector hd_qiov
;
970 uint8_t *cluster_data
= NULL
;
972 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
974 qemu_co_mutex_lock(&s
->lock
);
976 while (remaining_sectors
!= 0) {
978 /* prepare next request */
979 cur_nr_sectors
= remaining_sectors
;
980 if (s
->crypt_method
) {
981 cur_nr_sectors
= MIN(cur_nr_sectors
,
982 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
);
985 ret
= qcow2_get_cluster_offset(bs
, sector_num
<< 9,
986 &cur_nr_sectors
, &cluster_offset
);
991 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
993 qemu_iovec_reset(&hd_qiov
);
994 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
,
995 cur_nr_sectors
* 512);
998 case QCOW2_CLUSTER_UNALLOCATED
:
1000 if (bs
->backing_hd
) {
1001 /* read from the base image */
1002 n1
= qcow2_backing_read1(bs
->backing_hd
, &hd_qiov
,
1003 sector_num
, cur_nr_sectors
);
1005 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
1006 qemu_co_mutex_unlock(&s
->lock
);
1007 ret
= bdrv_co_readv(bs
->backing_hd
, sector_num
,
1009 qemu_co_mutex_lock(&s
->lock
);
1015 /* Note: in this case, no need to wait */
1016 qemu_iovec_memset(&hd_qiov
, 0, 0, 512 * cur_nr_sectors
);
1020 case QCOW2_CLUSTER_ZERO
:
1021 qemu_iovec_memset(&hd_qiov
, 0, 0, 512 * cur_nr_sectors
);
1024 case QCOW2_CLUSTER_COMPRESSED
:
1025 /* add AIO support for compressed blocks ? */
1026 ret
= qcow2_decompress_cluster(bs
, cluster_offset
);
1031 qemu_iovec_from_buf(&hd_qiov
, 0,
1032 s
->cluster_cache
+ index_in_cluster
* 512,
1033 512 * cur_nr_sectors
);
1036 case QCOW2_CLUSTER_NORMAL
:
1037 if ((cluster_offset
& 511) != 0) {
1042 if (s
->crypt_method
) {
1044 * For encrypted images, read everything into a temporary
1045 * contiguous buffer on which the AES functions can work.
1047 if (!cluster_data
) {
1049 qemu_blockalign(bs
, QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1052 assert(cur_nr_sectors
<=
1053 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
);
1054 qemu_iovec_reset(&hd_qiov
);
1055 qemu_iovec_add(&hd_qiov
, cluster_data
,
1056 512 * cur_nr_sectors
);
1059 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1060 qemu_co_mutex_unlock(&s
->lock
);
1061 ret
= bdrv_co_readv(bs
->file
,
1062 (cluster_offset
>> 9) + index_in_cluster
,
1063 cur_nr_sectors
, &hd_qiov
);
1064 qemu_co_mutex_lock(&s
->lock
);
1068 if (s
->crypt_method
) {
1069 qcow2_encrypt_sectors(s
, sector_num
, cluster_data
,
1070 cluster_data
, cur_nr_sectors
, 0, &s
->aes_decrypt_key
);
1071 qemu_iovec_from_buf(qiov
, bytes_done
,
1072 cluster_data
, 512 * cur_nr_sectors
);
1077 g_assert_not_reached();
1082 remaining_sectors
-= cur_nr_sectors
;
1083 sector_num
+= cur_nr_sectors
;
1084 bytes_done
+= cur_nr_sectors
* 512;
1089 qemu_co_mutex_unlock(&s
->lock
);
1091 qemu_iovec_destroy(&hd_qiov
);
1092 qemu_vfree(cluster_data
);
1097 static coroutine_fn
int qcow2_co_writev(BlockDriverState
*bs
,
1099 int remaining_sectors
,
1102 BDRVQcowState
*s
= bs
->opaque
;
1103 int index_in_cluster
;
1105 int cur_nr_sectors
; /* number of sectors in current iteration */
1106 uint64_t cluster_offset
;
1107 QEMUIOVector hd_qiov
;
1108 uint64_t bytes_done
= 0;
1109 uint8_t *cluster_data
= NULL
;
1110 QCowL2Meta
*l2meta
= NULL
;
1112 trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num
,
1115 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1117 s
->cluster_cache_offset
= -1; /* disable compressed cache */
1119 qemu_co_mutex_lock(&s
->lock
);
1121 while (remaining_sectors
!= 0) {
1125 trace_qcow2_writev_start_part(qemu_coroutine_self());
1126 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
1127 cur_nr_sectors
= remaining_sectors
;
1128 if (s
->crypt_method
&&
1130 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
- index_in_cluster
) {
1132 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
- index_in_cluster
;
1135 ret
= qcow2_alloc_cluster_offset(bs
, sector_num
<< 9,
1136 &cur_nr_sectors
, &cluster_offset
, &l2meta
);
1141 assert((cluster_offset
& 511) == 0);
1143 qemu_iovec_reset(&hd_qiov
);
1144 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
,
1145 cur_nr_sectors
* 512);
1147 if (s
->crypt_method
) {
1148 if (!cluster_data
) {
1149 cluster_data
= qemu_blockalign(bs
, QCOW_MAX_CRYPT_CLUSTERS
*
1153 assert(hd_qiov
.size
<=
1154 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1155 qemu_iovec_to_buf(&hd_qiov
, 0, cluster_data
, hd_qiov
.size
);
1157 qcow2_encrypt_sectors(s
, sector_num
, cluster_data
,
1158 cluster_data
, cur_nr_sectors
, 1, &s
->aes_encrypt_key
);
1160 qemu_iovec_reset(&hd_qiov
);
1161 qemu_iovec_add(&hd_qiov
, cluster_data
,
1162 cur_nr_sectors
* 512);
1165 ret
= qcow2_pre_write_overlap_check(bs
, 0,
1166 cluster_offset
+ index_in_cluster
* BDRV_SECTOR_SIZE
,
1167 cur_nr_sectors
* BDRV_SECTOR_SIZE
);
1172 qemu_co_mutex_unlock(&s
->lock
);
1173 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
1174 trace_qcow2_writev_data(qemu_coroutine_self(),
1175 (cluster_offset
>> 9) + index_in_cluster
);
1176 ret
= bdrv_co_writev(bs
->file
,
1177 (cluster_offset
>> 9) + index_in_cluster
,
1178 cur_nr_sectors
, &hd_qiov
);
1179 qemu_co_mutex_lock(&s
->lock
);
1184 while (l2meta
!= NULL
) {
1187 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
1192 /* Take the request off the list of running requests */
1193 if (l2meta
->nb_clusters
!= 0) {
1194 QLIST_REMOVE(l2meta
, next_in_flight
);
1197 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
1199 next
= l2meta
->next
;
1204 remaining_sectors
-= cur_nr_sectors
;
1205 sector_num
+= cur_nr_sectors
;
1206 bytes_done
+= cur_nr_sectors
* 512;
1207 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors
);
1212 qemu_co_mutex_unlock(&s
->lock
);
1214 while (l2meta
!= NULL
) {
1217 if (l2meta
->nb_clusters
!= 0) {
1218 QLIST_REMOVE(l2meta
, next_in_flight
);
1220 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
1222 next
= l2meta
->next
;
1227 qemu_iovec_destroy(&hd_qiov
);
1228 qemu_vfree(cluster_data
);
1229 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
1234 static void qcow2_close(BlockDriverState
*bs
)
1236 BDRVQcowState
*s
= bs
->opaque
;
1237 g_free(s
->l1_table
);
1238 /* else pre-write overlap checks in cache_destroy may crash */
1241 if (!(bs
->open_flags
& BDRV_O_INCOMING
)) {
1242 qcow2_cache_flush(bs
, s
->l2_table_cache
);
1243 qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1245 qcow2_mark_clean(bs
);
1248 qcow2_cache_destroy(bs
, s
->l2_table_cache
);
1249 qcow2_cache_destroy(bs
, s
->refcount_block_cache
);
1251 g_free(s
->unknown_header_fields
);
1252 cleanup_unknown_header_ext(bs
);
1254 g_free(s
->cluster_cache
);
1255 qemu_vfree(s
->cluster_data
);
1256 qcow2_refcount_close(bs
);
1257 qcow2_free_snapshots(bs
);
1260 static void qcow2_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
1262 BDRVQcowState
*s
= bs
->opaque
;
1263 int flags
= s
->flags
;
1264 AES_KEY aes_encrypt_key
;
1265 AES_KEY aes_decrypt_key
;
1266 uint32_t crypt_method
= 0;
1268 Error
*local_err
= NULL
;
1272 * Backing files are read-only which makes all of their metadata immutable,
1273 * that means we don't have to worry about reopening them here.
1276 if (s
->crypt_method
) {
1277 crypt_method
= s
->crypt_method
;
1278 memcpy(&aes_encrypt_key
, &s
->aes_encrypt_key
, sizeof(aes_encrypt_key
));
1279 memcpy(&aes_decrypt_key
, &s
->aes_decrypt_key
, sizeof(aes_decrypt_key
));
1284 bdrv_invalidate_cache(bs
->file
, &local_err
);
1286 error_propagate(errp
, local_err
);
1290 memset(s
, 0, sizeof(BDRVQcowState
));
1291 options
= qdict_clone_shallow(bs
->options
);
1293 ret
= qcow2_open(bs
, options
, flags
, &local_err
);
1295 error_setg(errp
, "Could not reopen qcow2 layer: %s",
1296 error_get_pretty(local_err
));
1297 error_free(local_err
);
1299 } else if (ret
< 0) {
1300 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
1307 s
->crypt_method
= crypt_method
;
1308 memcpy(&s
->aes_encrypt_key
, &aes_encrypt_key
, sizeof(aes_encrypt_key
));
1309 memcpy(&s
->aes_decrypt_key
, &aes_decrypt_key
, sizeof(aes_decrypt_key
));
1313 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
1314 size_t len
, size_t buflen
)
1316 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
1317 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
1319 if (buflen
< ext_len
) {
1323 *ext_backing_fmt
= (QCowExtension
) {
1324 .magic
= cpu_to_be32(magic
),
1325 .len
= cpu_to_be32(len
),
1327 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
1333 * Updates the qcow2 header, including the variable length parts of it, i.e.
1334 * the backing file name and all extensions. qcow2 was not designed to allow
1335 * such changes, so if we run out of space (we can only use the first cluster)
1336 * this function may fail.
1338 * Returns 0 on success, -errno in error cases.
1340 int qcow2_update_header(BlockDriverState
*bs
)
1342 BDRVQcowState
*s
= bs
->opaque
;
1345 size_t buflen
= s
->cluster_size
;
1347 uint64_t total_size
;
1348 uint32_t refcount_table_clusters
;
1349 size_t header_length
;
1350 Qcow2UnknownHeaderExtension
*uext
;
1352 buf
= qemu_blockalign(bs
, buflen
);
1354 /* Header structure */
1355 header
= (QCowHeader
*) buf
;
1357 if (buflen
< sizeof(*header
)) {
1362 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
1363 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
1364 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
1366 *header
= (QCowHeader
) {
1367 /* Version 2 fields */
1368 .magic
= cpu_to_be32(QCOW_MAGIC
),
1369 .version
= cpu_to_be32(s
->qcow_version
),
1370 .backing_file_offset
= 0,
1371 .backing_file_size
= 0,
1372 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
1373 .size
= cpu_to_be64(total_size
),
1374 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
1375 .l1_size
= cpu_to_be32(s
->l1_size
),
1376 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
1377 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
1378 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
1379 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
1380 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
1382 /* Version 3 fields */
1383 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
1384 .compatible_features
= cpu_to_be64(s
->compatible_features
),
1385 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
1386 .refcount_order
= cpu_to_be32(s
->refcount_order
),
1387 .header_length
= cpu_to_be32(header_length
),
1390 /* For older versions, write a shorter header */
1391 switch (s
->qcow_version
) {
1393 ret
= offsetof(QCowHeader
, incompatible_features
);
1396 ret
= sizeof(*header
);
1405 memset(buf
, 0, buflen
);
1407 /* Preserve any unknown field in the header */
1408 if (s
->unknown_header_fields_size
) {
1409 if (buflen
< s
->unknown_header_fields_size
) {
1414 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
1415 buf
+= s
->unknown_header_fields_size
;
1416 buflen
-= s
->unknown_header_fields_size
;
1419 /* Backing file format header extension */
1420 if (*bs
->backing_format
) {
1421 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
1422 bs
->backing_format
, strlen(bs
->backing_format
),
1433 Qcow2Feature features
[] = {
1435 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
1436 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
1437 .name
= "dirty bit",
1440 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
1441 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
1442 .name
= "corrupt bit",
1445 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
1446 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
1447 .name
= "lazy refcounts",
1451 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
1452 features
, sizeof(features
), buflen
);
1459 /* Keep unknown header extensions */
1460 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
1461 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
1470 /* End of header extensions */
1471 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
1479 /* Backing file name */
1480 if (*bs
->backing_file
) {
1481 size_t backing_file_len
= strlen(bs
->backing_file
);
1483 if (buflen
< backing_file_len
) {
1488 /* Using strncpy is ok here, since buf is not NUL-terminated. */
1489 strncpy(buf
, bs
->backing_file
, buflen
);
1491 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
1492 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
1495 /* Write the new header */
1496 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
1507 static int qcow2_change_backing_file(BlockDriverState
*bs
,
1508 const char *backing_file
, const char *backing_fmt
)
1510 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
1511 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
1513 return qcow2_update_header(bs
);
1516 static int preallocate(BlockDriverState
*bs
)
1518 uint64_t nb_sectors
;
1520 uint64_t host_offset
= 0;
1525 nb_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
1528 while (nb_sectors
) {
1529 num
= MIN(nb_sectors
, INT_MAX
>> BDRV_SECTOR_BITS
);
1530 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &num
,
1531 &host_offset
, &meta
);
1537 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
1539 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
1540 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
1544 /* There are no dependent requests, but we need to remove our
1545 * request from the list of in-flight requests */
1546 QLIST_REMOVE(meta
, next_in_flight
);
1549 /* TODO Preallocate data if requested */
1552 offset
+= num
<< BDRV_SECTOR_BITS
;
1556 * It is expected that the image file is large enough to actually contain
1557 * all of the allocated clusters (otherwise we get failing reads after
1558 * EOF). Extend the image to the last allocated sector.
1560 if (host_offset
!= 0) {
1561 uint8_t buf
[BDRV_SECTOR_SIZE
];
1562 memset(buf
, 0, BDRV_SECTOR_SIZE
);
1563 ret
= bdrv_write(bs
->file
, (host_offset
>> BDRV_SECTOR_BITS
) + num
- 1,
1573 static int qcow2_create2(const char *filename
, int64_t total_size
,
1574 const char *backing_file
, const char *backing_format
,
1575 int flags
, size_t cluster_size
, int prealloc
,
1576 QEMUOptionParameter
*options
, int version
,
1579 /* Calculate cluster_bits */
1581 cluster_bits
= ffs(cluster_size
) - 1;
1582 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
1583 (1 << cluster_bits
) != cluster_size
)
1585 error_setg(errp
, "Cluster size must be a power of two between %d and "
1586 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
1591 * Open the image file and write a minimal qcow2 header.
1593 * We keep things simple and start with a zero-sized image. We also
1594 * do without refcount blocks or a L1 table for now. We'll fix the
1595 * inconsistency later.
1597 * We do need a refcount table because growing the refcount table means
1598 * allocating two new refcount blocks - the seconds of which would be at
1599 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1600 * size for any qcow2 image.
1602 BlockDriverState
* bs
;
1604 uint64_t* refcount_table
;
1605 Error
*local_err
= NULL
;
1608 ret
= bdrv_create_file(filename
, options
, &local_err
);
1610 error_propagate(errp
, local_err
);
1615 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
, BDRV_O_RDWR
| BDRV_O_PROTOCOL
,
1618 error_propagate(errp
, local_err
);
1622 /* Write the header */
1623 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
1624 header
= g_malloc0(cluster_size
);
1625 *header
= (QCowHeader
) {
1626 .magic
= cpu_to_be32(QCOW_MAGIC
),
1627 .version
= cpu_to_be32(version
),
1628 .cluster_bits
= cpu_to_be32(cluster_bits
),
1629 .size
= cpu_to_be64(0),
1630 .l1_table_offset
= cpu_to_be64(0),
1631 .l1_size
= cpu_to_be32(0),
1632 .refcount_table_offset
= cpu_to_be64(cluster_size
),
1633 .refcount_table_clusters
= cpu_to_be32(1),
1634 .refcount_order
= cpu_to_be32(3 + REFCOUNT_SHIFT
),
1635 .header_length
= cpu_to_be32(sizeof(*header
)),
1638 if (flags
& BLOCK_FLAG_ENCRYPT
) {
1639 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
1641 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
1644 if (flags
& BLOCK_FLAG_LAZY_REFCOUNTS
) {
1645 header
->compatible_features
|=
1646 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
1649 ret
= bdrv_pwrite(bs
, 0, header
, cluster_size
);
1652 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
1656 /* Write a refcount table with one refcount block */
1657 refcount_table
= g_malloc0(2 * cluster_size
);
1658 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
1659 ret
= bdrv_pwrite(bs
, cluster_size
, refcount_table
, 2 * cluster_size
);
1660 g_free(refcount_table
);
1663 error_setg_errno(errp
, -ret
, "Could not write refcount table");
1671 * And now open the image and make it consistent first (i.e. increase the
1672 * refcount of the cluster that is occupied by the header and the refcount
1675 BlockDriver
* drv
= bdrv_find_format("qcow2");
1676 assert(drv
!= NULL
);
1677 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
,
1678 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_NO_FLUSH
, drv
, &local_err
);
1680 error_propagate(errp
, local_err
);
1684 ret
= qcow2_alloc_clusters(bs
, 3 * cluster_size
);
1686 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
1687 "header and refcount table");
1690 } else if (ret
!= 0) {
1691 error_report("Huh, first cluster in empty image is already in use?");
1695 /* Okay, now that we have a valid image, let's give it the right size */
1696 ret
= bdrv_truncate(bs
, total_size
* BDRV_SECTOR_SIZE
);
1698 error_setg_errno(errp
, -ret
, "Could not resize image");
1702 /* Want a backing file? There you go.*/
1704 ret
= bdrv_change_backing_file(bs
, backing_file
, backing_format
);
1706 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
1707 "with format '%s'", backing_file
, backing_format
);
1712 /* And if we're supposed to preallocate metadata, do that now */
1714 BDRVQcowState
*s
= bs
->opaque
;
1715 qemu_co_mutex_lock(&s
->lock
);
1716 ret
= preallocate(bs
);
1717 qemu_co_mutex_unlock(&s
->lock
);
1719 error_setg_errno(errp
, -ret
, "Could not preallocate metadata");
1727 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
1728 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
,
1729 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_NO_BACKING
,
1732 error_propagate(errp
, local_err
);
1744 static int qcow2_create(const char *filename
, QEMUOptionParameter
*options
,
1747 const char *backing_file
= NULL
;
1748 const char *backing_fmt
= NULL
;
1749 uint64_t sectors
= 0;
1751 size_t cluster_size
= DEFAULT_CLUSTER_SIZE
;
1754 Error
*local_err
= NULL
;
1757 /* Read out options */
1758 while (options
&& options
->name
) {
1759 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
1760 sectors
= options
->value
.n
/ 512;
1761 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
1762 backing_file
= options
->value
.s
;
1763 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FMT
)) {
1764 backing_fmt
= options
->value
.s
;
1765 } else if (!strcmp(options
->name
, BLOCK_OPT_ENCRYPT
)) {
1766 flags
|= options
->value
.n
? BLOCK_FLAG_ENCRYPT
: 0;
1767 } else if (!strcmp(options
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
1768 if (options
->value
.n
) {
1769 cluster_size
= options
->value
.n
;
1771 } else if (!strcmp(options
->name
, BLOCK_OPT_PREALLOC
)) {
1772 if (!options
->value
.s
|| !strcmp(options
->value
.s
, "off")) {
1774 } else if (!strcmp(options
->value
.s
, "metadata")) {
1777 error_setg(errp
, "Invalid preallocation mode: '%s'",
1781 } else if (!strcmp(options
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
1782 if (!options
->value
.s
) {
1783 /* keep the default */
1784 } else if (!strcmp(options
->value
.s
, "0.10")) {
1786 } else if (!strcmp(options
->value
.s
, "1.1")) {
1789 error_setg(errp
, "Invalid compatibility level: '%s'",
1793 } else if (!strcmp(options
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
1794 flags
|= options
->value
.n
? BLOCK_FLAG_LAZY_REFCOUNTS
: 0;
1799 if (backing_file
&& prealloc
) {
1800 error_setg(errp
, "Backing file and preallocation cannot be used at "
1805 if (version
< 3 && (flags
& BLOCK_FLAG_LAZY_REFCOUNTS
)) {
1806 error_setg(errp
, "Lazy refcounts only supported with compatibility "
1807 "level 1.1 and above (use compat=1.1 or greater)");
1811 ret
= qcow2_create2(filename
, sectors
, backing_file
, backing_fmt
, flags
,
1812 cluster_size
, prealloc
, options
, version
, &local_err
);
1814 error_propagate(errp
, local_err
);
1819 static coroutine_fn
int qcow2_co_write_zeroes(BlockDriverState
*bs
,
1820 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
1823 BDRVQcowState
*s
= bs
->opaque
;
1825 /* Emulate misaligned zero writes */
1826 if (sector_num
% s
->cluster_sectors
|| nb_sectors
% s
->cluster_sectors
) {
1830 /* Whatever is left can use real zero clusters */
1831 qemu_co_mutex_lock(&s
->lock
);
1832 ret
= qcow2_zero_clusters(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1834 qemu_co_mutex_unlock(&s
->lock
);
1839 static coroutine_fn
int qcow2_co_discard(BlockDriverState
*bs
,
1840 int64_t sector_num
, int nb_sectors
)
1843 BDRVQcowState
*s
= bs
->opaque
;
1845 qemu_co_mutex_lock(&s
->lock
);
1846 ret
= qcow2_discard_clusters(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1847 nb_sectors
, QCOW2_DISCARD_REQUEST
);
1848 qemu_co_mutex_unlock(&s
->lock
);
1852 static int qcow2_truncate(BlockDriverState
*bs
, int64_t offset
)
1854 BDRVQcowState
*s
= bs
->opaque
;
1855 int64_t new_l1_size
;
1859 error_report("The new size must be a multiple of 512");
1863 /* cannot proceed if image has snapshots */
1864 if (s
->nb_snapshots
) {
1865 error_report("Can't resize an image which has snapshots");
1869 /* shrinking is currently not supported */
1870 if (offset
< bs
->total_sectors
* 512) {
1871 error_report("qcow2 doesn't support shrinking images yet");
1875 new_l1_size
= size_to_l1(s
, offset
);
1876 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
1881 /* write updated header.size */
1882 offset
= cpu_to_be64(offset
);
1883 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
1884 &offset
, sizeof(uint64_t));
1889 s
->l1_vm_state_index
= new_l1_size
;
1893 /* XXX: put compressed sectors first, then all the cluster aligned
1894 tables to avoid losing bytes in alignment */
1895 static int qcow2_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1896 const uint8_t *buf
, int nb_sectors
)
1898 BDRVQcowState
*s
= bs
->opaque
;
1902 uint64_t cluster_offset
;
1904 if (nb_sectors
== 0) {
1905 /* align end of file to a sector boundary to ease reading with
1906 sector based I/Os */
1907 cluster_offset
= bdrv_getlength(bs
->file
);
1908 cluster_offset
= (cluster_offset
+ 511) & ~511;
1909 bdrv_truncate(bs
->file
, cluster_offset
);
1913 if (nb_sectors
!= s
->cluster_sectors
) {
1916 /* Zero-pad last write if image size is not cluster aligned */
1917 if (sector_num
+ nb_sectors
== bs
->total_sectors
&&
1918 nb_sectors
< s
->cluster_sectors
) {
1919 uint8_t *pad_buf
= qemu_blockalign(bs
, s
->cluster_size
);
1920 memset(pad_buf
, 0, s
->cluster_size
);
1921 memcpy(pad_buf
, buf
, nb_sectors
* BDRV_SECTOR_SIZE
);
1922 ret
= qcow2_write_compressed(bs
, sector_num
,
1923 pad_buf
, s
->cluster_sectors
);
1924 qemu_vfree(pad_buf
);
1929 out_buf
= g_malloc(s
->cluster_size
+ (s
->cluster_size
/ 1000) + 128);
1931 /* best compression, small window, no zlib header */
1932 memset(&strm
, 0, sizeof(strm
));
1933 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
1935 9, Z_DEFAULT_STRATEGY
);
1941 strm
.avail_in
= s
->cluster_size
;
1942 strm
.next_in
= (uint8_t *)buf
;
1943 strm
.avail_out
= s
->cluster_size
;
1944 strm
.next_out
= out_buf
;
1946 ret
= deflate(&strm
, Z_FINISH
);
1947 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
1952 out_len
= strm
.next_out
- out_buf
;
1956 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
1957 /* could not compress: write normal cluster */
1958 ret
= bdrv_write(bs
, sector_num
, buf
, s
->cluster_sectors
);
1963 cluster_offset
= qcow2_alloc_compressed_cluster_offset(bs
,
1964 sector_num
<< 9, out_len
);
1965 if (!cluster_offset
) {
1969 cluster_offset
&= s
->cluster_offset_mask
;
1971 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
);
1976 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
1977 ret
= bdrv_pwrite(bs
->file
, cluster_offset
, out_buf
, out_len
);
1989 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
1991 BDRVQcowState
*s
= bs
->opaque
;
1994 qemu_co_mutex_lock(&s
->lock
);
1995 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1997 qemu_co_mutex_unlock(&s
->lock
);
2001 if (qcow2_need_accurate_refcounts(s
)) {
2002 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2004 qemu_co_mutex_unlock(&s
->lock
);
2008 qemu_co_mutex_unlock(&s
->lock
);
2013 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2015 BDRVQcowState
*s
= bs
->opaque
;
2016 bdi
->unallocated_blocks_are_zero
= true;
2017 bdi
->can_write_zeroes_with_unmap
= (s
->qcow_version
>= 3);
2018 bdi
->cluster_size
= s
->cluster_size
;
2019 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
2023 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
)
2025 BDRVQcowState
*s
= bs
->opaque
;
2026 ImageInfoSpecific
*spec_info
= g_new(ImageInfoSpecific
, 1);
2028 *spec_info
= (ImageInfoSpecific
){
2029 .kind
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
2031 .qcow2
= g_new(ImageInfoSpecificQCow2
, 1),
2034 if (s
->qcow_version
== 2) {
2035 *spec_info
->qcow2
= (ImageInfoSpecificQCow2
){
2036 .compat
= g_strdup("0.10"),
2038 } else if (s
->qcow_version
== 3) {
2039 *spec_info
->qcow2
= (ImageInfoSpecificQCow2
){
2040 .compat
= g_strdup("1.1"),
2041 .lazy_refcounts
= s
->compatible_features
&
2042 QCOW2_COMPAT_LAZY_REFCOUNTS
,
2043 .has_lazy_refcounts
= true,
2051 static void dump_refcounts(BlockDriverState
*bs
)
2053 BDRVQcowState
*s
= bs
->opaque
;
2054 int64_t nb_clusters
, k
, k1
, size
;
2057 size
= bdrv_getlength(bs
->file
);
2058 nb_clusters
= size_to_clusters(s
, size
);
2059 for(k
= 0; k
< nb_clusters
;) {
2061 refcount
= get_refcount(bs
, k
);
2063 while (k
< nb_clusters
&& get_refcount(bs
, k
) == refcount
)
2065 printf("%" PRId64
": refcount=%d nb=%" PRId64
"\n", k
, refcount
,
2071 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2074 BDRVQcowState
*s
= bs
->opaque
;
2075 int64_t total_sectors
= bs
->total_sectors
;
2076 int growable
= bs
->growable
;
2077 bool zero_beyond_eof
= bs
->zero_beyond_eof
;
2080 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
2082 bs
->zero_beyond_eof
= false;
2083 ret
= bdrv_pwritev(bs
, qcow2_vm_state_offset(s
) + pos
, qiov
);
2084 bs
->growable
= growable
;
2085 bs
->zero_beyond_eof
= zero_beyond_eof
;
2087 /* bdrv_co_do_writev will have increased the total_sectors value to include
2088 * the VM state - the VM state is however not an actual part of the block
2089 * device, therefore, we need to restore the old value. */
2090 bs
->total_sectors
= total_sectors
;
2095 static int qcow2_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2096 int64_t pos
, int size
)
2098 BDRVQcowState
*s
= bs
->opaque
;
2099 int growable
= bs
->growable
;
2100 bool zero_beyond_eof
= bs
->zero_beyond_eof
;
2103 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
2105 bs
->zero_beyond_eof
= false;
2106 ret
= bdrv_pread(bs
, qcow2_vm_state_offset(s
) + pos
, buf
, size
);
2107 bs
->growable
= growable
;
2108 bs
->zero_beyond_eof
= zero_beyond_eof
;
2114 * Downgrades an image's version. To achieve this, any incompatible features
2115 * have to be removed.
2117 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
)
2119 BDRVQcowState
*s
= bs
->opaque
;
2120 int current_version
= s
->qcow_version
;
2123 if (target_version
== current_version
) {
2125 } else if (target_version
> current_version
) {
2127 } else if (target_version
!= 2) {
2131 if (s
->refcount_order
!= 4) {
2132 /* we would have to convert the image to a refcount_order == 4 image
2133 * here; however, since qemu (at the time of writing this) does not
2134 * support anything different than 4 anyway, there is no point in doing
2135 * so right now; however, we should error out (if qemu supports this in
2136 * the future and this code has not been adapted) */
2137 error_report("qcow2_downgrade: Image refcount orders other than 4 are "
2138 "currently not supported.");
2142 /* clear incompatible features */
2143 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
2144 ret
= qcow2_mark_clean(bs
);
2150 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2151 * the first place; if that happens nonetheless, returning -ENOTSUP is the
2152 * best thing to do anyway */
2154 if (s
->incompatible_features
) {
2158 /* since we can ignore compatible features, we can set them to 0 as well */
2159 s
->compatible_features
= 0;
2160 /* if lazy refcounts have been used, they have already been fixed through
2161 * clearing the dirty flag */
2163 /* clearing autoclear features is trivial */
2164 s
->autoclear_features
= 0;
2166 ret
= qcow2_expand_zero_clusters(bs
);
2171 s
->qcow_version
= target_version
;
2172 ret
= qcow2_update_header(bs
);
2174 s
->qcow_version
= current_version
;
2180 static int qcow2_amend_options(BlockDriverState
*bs
,
2181 QEMUOptionParameter
*options
)
2183 BDRVQcowState
*s
= bs
->opaque
;
2184 int old_version
= s
->qcow_version
, new_version
= old_version
;
2185 uint64_t new_size
= 0;
2186 const char *backing_file
= NULL
, *backing_format
= NULL
;
2187 bool lazy_refcounts
= s
->use_lazy_refcounts
;
2191 for (i
= 0; options
[i
].name
; i
++)
2193 if (!options
[i
].assigned
) {
2194 /* only change explicitly defined options */
2198 if (!strcmp(options
[i
].name
, "compat")) {
2199 if (!options
[i
].value
.s
) {
2200 /* preserve default */
2201 } else if (!strcmp(options
[i
].value
.s
, "0.10")) {
2203 } else if (!strcmp(options
[i
].value
.s
, "1.1")) {
2206 fprintf(stderr
, "Unknown compatibility level %s.\n",
2207 options
[i
].value
.s
);
2210 } else if (!strcmp(options
[i
].name
, "preallocation")) {
2211 fprintf(stderr
, "Cannot change preallocation mode.\n");
2213 } else if (!strcmp(options
[i
].name
, "size")) {
2214 new_size
= options
[i
].value
.n
;
2215 } else if (!strcmp(options
[i
].name
, "backing_file")) {
2216 backing_file
= options
[i
].value
.s
;
2217 } else if (!strcmp(options
[i
].name
, "backing_fmt")) {
2218 backing_format
= options
[i
].value
.s
;
2219 } else if (!strcmp(options
[i
].name
, "encryption")) {
2220 if ((options
[i
].value
.n
!= !!s
->crypt_method
)) {
2221 fprintf(stderr
, "Changing the encryption flag is not "
2225 } else if (!strcmp(options
[i
].name
, "cluster_size")) {
2226 if (options
[i
].value
.n
!= s
->cluster_size
) {
2227 fprintf(stderr
, "Changing the cluster size is not "
2231 } else if (!strcmp(options
[i
].name
, "lazy_refcounts")) {
2232 lazy_refcounts
= options
[i
].value
.n
;
2234 /* if this assertion fails, this probably means a new option was
2235 * added without having it covered here */
2240 if (new_version
!= old_version
) {
2241 if (new_version
> old_version
) {
2243 s
->qcow_version
= new_version
;
2244 ret
= qcow2_update_header(bs
);
2246 s
->qcow_version
= old_version
;
2250 ret
= qcow2_downgrade(bs
, new_version
);
2257 if (backing_file
|| backing_format
) {
2258 ret
= qcow2_change_backing_file(bs
, backing_file
?: bs
->backing_file
,
2259 backing_format
?: bs
->backing_format
);
2265 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
2266 if (lazy_refcounts
) {
2267 if (s
->qcow_version
< 3) {
2268 fprintf(stderr
, "Lazy refcounts only supported with compatibility "
2269 "level 1.1 and above (use compat=1.1 or greater)\n");
2272 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
2273 ret
= qcow2_update_header(bs
);
2275 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
2278 s
->use_lazy_refcounts
= true;
2280 /* make image clean first */
2281 ret
= qcow2_mark_clean(bs
);
2285 /* now disallow lazy refcounts */
2286 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
2287 ret
= qcow2_update_header(bs
);
2289 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
2292 s
->use_lazy_refcounts
= false;
2297 ret
= bdrv_truncate(bs
, new_size
);
2306 static QEMUOptionParameter qcow2_create_options
[] = {
2308 .name
= BLOCK_OPT_SIZE
,
2310 .help
= "Virtual disk size"
2313 .name
= BLOCK_OPT_COMPAT_LEVEL
,
2315 .help
= "Compatibility level (0.10 or 1.1)"
2318 .name
= BLOCK_OPT_BACKING_FILE
,
2320 .help
= "File name of a base image"
2323 .name
= BLOCK_OPT_BACKING_FMT
,
2325 .help
= "Image format of the base image"
2328 .name
= BLOCK_OPT_ENCRYPT
,
2330 .help
= "Encrypt the image"
2333 .name
= BLOCK_OPT_CLUSTER_SIZE
,
2335 .help
= "qcow2 cluster size",
2336 .value
= { .n
= DEFAULT_CLUSTER_SIZE
},
2339 .name
= BLOCK_OPT_PREALLOC
,
2341 .help
= "Preallocation mode (allowed values: off, metadata)"
2344 .name
= BLOCK_OPT_LAZY_REFCOUNTS
,
2346 .help
= "Postpone refcount updates",
2351 static BlockDriver bdrv_qcow2
= {
2352 .format_name
= "qcow2",
2353 .instance_size
= sizeof(BDRVQcowState
),
2354 .bdrv_probe
= qcow2_probe
,
2355 .bdrv_open
= qcow2_open
,
2356 .bdrv_close
= qcow2_close
,
2357 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
2358 .bdrv_create
= qcow2_create
,
2359 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2360 .bdrv_co_get_block_status
= qcow2_co_get_block_status
,
2361 .bdrv_set_key
= qcow2_set_key
,
2363 .bdrv_co_readv
= qcow2_co_readv
,
2364 .bdrv_co_writev
= qcow2_co_writev
,
2365 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
2367 .bdrv_co_write_zeroes
= qcow2_co_write_zeroes
,
2368 .bdrv_co_discard
= qcow2_co_discard
,
2369 .bdrv_truncate
= qcow2_truncate
,
2370 .bdrv_write_compressed
= qcow2_write_compressed
,
2372 .bdrv_snapshot_create
= qcow2_snapshot_create
,
2373 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
2374 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
2375 .bdrv_snapshot_list
= qcow2_snapshot_list
,
2376 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
2377 .bdrv_get_info
= qcow2_get_info
,
2378 .bdrv_get_specific_info
= qcow2_get_specific_info
,
2380 .bdrv_save_vmstate
= qcow2_save_vmstate
,
2381 .bdrv_load_vmstate
= qcow2_load_vmstate
,
2383 .bdrv_change_backing_file
= qcow2_change_backing_file
,
2385 .bdrv_refresh_limits
= qcow2_refresh_limits
,
2386 .bdrv_invalidate_cache
= qcow2_invalidate_cache
,
2388 .create_options
= qcow2_create_options
,
2389 .bdrv_check
= qcow2_check
,
2390 .bdrv_amend_options
= qcow2_amend_options
,
2393 static void bdrv_qcow2_init(void)
2395 bdrv_register(&bdrv_qcow2
);
2398 block_init(bdrv_qcow2_init
);