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
) {
274 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
276 ret
= bdrv_flush(bs
);
281 return qcow2_update_header(bs
);
287 * Marks the image as corrupt.
289 int qcow2_mark_corrupt(BlockDriverState
*bs
)
291 BDRVQcowState
*s
= bs
->opaque
;
293 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
294 return qcow2_update_header(bs
);
298 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
299 * before if necessary.
301 int qcow2_mark_consistent(BlockDriverState
*bs
)
303 BDRVQcowState
*s
= bs
->opaque
;
305 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
306 int ret
= bdrv_flush(bs
);
311 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
312 return qcow2_update_header(bs
);
317 static int qcow2_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
320 int ret
= qcow2_check_refcounts(bs
, result
, fix
);
325 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
326 ret
= qcow2_mark_clean(bs
);
330 return qcow2_mark_consistent(bs
);
335 static int validate_table_offset(BlockDriverState
*bs
, uint64_t offset
,
336 uint64_t entries
, size_t entry_len
)
338 BDRVQcowState
*s
= bs
->opaque
;
341 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
342 * because values will be passed to qemu functions taking int64_t. */
343 if (entries
> INT64_MAX
/ entry_len
) {
347 size
= entries
* entry_len
;
349 if (INT64_MAX
- size
< offset
) {
353 /* Tables must be cluster aligned */
354 if (offset
& (s
->cluster_size
- 1)) {
361 static QemuOptsList qcow2_runtime_opts
= {
363 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
366 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
367 .type
= QEMU_OPT_BOOL
,
368 .help
= "Postpone refcount updates",
371 .name
= QCOW2_OPT_DISCARD_REQUEST
,
372 .type
= QEMU_OPT_BOOL
,
373 .help
= "Pass guest discard requests to the layer below",
376 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
377 .type
= QEMU_OPT_BOOL
,
378 .help
= "Generate discard requests when snapshot related space "
382 .name
= QCOW2_OPT_DISCARD_OTHER
,
383 .type
= QEMU_OPT_BOOL
,
384 .help
= "Generate discard requests when other clusters are freed",
387 .name
= QCOW2_OPT_OVERLAP
,
388 .type
= QEMU_OPT_STRING
,
389 .help
= "Selects which overlap checks to perform from a range of "
390 "templates (none, constant, cached, all)",
393 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
394 .type
= QEMU_OPT_BOOL
,
395 .help
= "Check for unintended writes into the main qcow2 header",
398 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
399 .type
= QEMU_OPT_BOOL
,
400 .help
= "Check for unintended writes into the active L1 table",
403 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
404 .type
= QEMU_OPT_BOOL
,
405 .help
= "Check for unintended writes into an active L2 table",
408 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
409 .type
= QEMU_OPT_BOOL
,
410 .help
= "Check for unintended writes into the refcount table",
413 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
414 .type
= QEMU_OPT_BOOL
,
415 .help
= "Check for unintended writes into a refcount block",
418 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
419 .type
= QEMU_OPT_BOOL
,
420 .help
= "Check for unintended writes into the snapshot table",
423 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
424 .type
= QEMU_OPT_BOOL
,
425 .help
= "Check for unintended writes into an inactive L1 table",
428 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
429 .type
= QEMU_OPT_BOOL
,
430 .help
= "Check for unintended writes into an inactive L2 table",
432 { /* end of list */ }
436 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
437 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
438 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
439 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
440 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
441 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
442 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
443 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
444 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
447 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
450 BDRVQcowState
*s
= bs
->opaque
;
455 Error
*local_err
= NULL
;
457 uint64_t l1_vm_state_index
;
458 const char *opt_overlap_check
;
459 int overlap_check_template
= 0;
461 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
463 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
466 be32_to_cpus(&header
.magic
);
467 be32_to_cpus(&header
.version
);
468 be64_to_cpus(&header
.backing_file_offset
);
469 be32_to_cpus(&header
.backing_file_size
);
470 be64_to_cpus(&header
.size
);
471 be32_to_cpus(&header
.cluster_bits
);
472 be32_to_cpus(&header
.crypt_method
);
473 be64_to_cpus(&header
.l1_table_offset
);
474 be32_to_cpus(&header
.l1_size
);
475 be64_to_cpus(&header
.refcount_table_offset
);
476 be32_to_cpus(&header
.refcount_table_clusters
);
477 be64_to_cpus(&header
.snapshots_offset
);
478 be32_to_cpus(&header
.nb_snapshots
);
480 if (header
.magic
!= QCOW_MAGIC
) {
481 error_setg(errp
, "Image is not in qcow2 format");
485 if (header
.version
< 2 || header
.version
> 3) {
486 report_unsupported(bs
, errp
, "QCOW version %d", header
.version
);
491 s
->qcow_version
= header
.version
;
493 /* Initialise cluster size */
494 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
495 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
496 error_setg(errp
, "Unsupported cluster size: 2^%i", header
.cluster_bits
);
501 s
->cluster_bits
= header
.cluster_bits
;
502 s
->cluster_size
= 1 << s
->cluster_bits
;
503 s
->cluster_sectors
= 1 << (s
->cluster_bits
- 9);
505 /* Initialise version 3 header fields */
506 if (header
.version
== 2) {
507 header
.incompatible_features
= 0;
508 header
.compatible_features
= 0;
509 header
.autoclear_features
= 0;
510 header
.refcount_order
= 4;
511 header
.header_length
= 72;
513 be64_to_cpus(&header
.incompatible_features
);
514 be64_to_cpus(&header
.compatible_features
);
515 be64_to_cpus(&header
.autoclear_features
);
516 be32_to_cpus(&header
.refcount_order
);
517 be32_to_cpus(&header
.header_length
);
519 if (header
.header_length
< 104) {
520 error_setg(errp
, "qcow2 header too short");
526 if (header
.header_length
> s
->cluster_size
) {
527 error_setg(errp
, "qcow2 header exceeds cluster size");
532 if (header
.header_length
> sizeof(header
)) {
533 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
534 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
535 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
536 s
->unknown_header_fields_size
);
538 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
544 if (header
.backing_file_offset
> s
->cluster_size
) {
545 error_setg(errp
, "Invalid backing file offset");
550 if (header
.backing_file_offset
) {
551 ext_end
= header
.backing_file_offset
;
553 ext_end
= 1 << header
.cluster_bits
;
556 /* Handle feature bits */
557 s
->incompatible_features
= header
.incompatible_features
;
558 s
->compatible_features
= header
.compatible_features
;
559 s
->autoclear_features
= header
.autoclear_features
;
561 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
562 void *feature_table
= NULL
;
563 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
564 &feature_table
, NULL
);
565 report_unsupported_feature(bs
, errp
, feature_table
,
566 s
->incompatible_features
&
567 ~QCOW2_INCOMPAT_MASK
);
569 g_free(feature_table
);
573 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
574 /* Corrupt images may not be written to unless they are being repaired
576 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
577 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
584 /* Check support for various header values */
585 if (header
.refcount_order
!= 4) {
586 report_unsupported(bs
, errp
, "%d bit reference counts",
587 1 << header
.refcount_order
);
591 s
->refcount_order
= header
.refcount_order
;
593 if (header
.crypt_method
> QCOW_CRYPT_AES
) {
594 error_setg(errp
, "Unsupported encryption method: %i",
595 header
.crypt_method
);
599 s
->crypt_method_header
= header
.crypt_method
;
600 if (s
->crypt_method_header
) {
604 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
605 s
->l2_size
= 1 << s
->l2_bits
;
606 bs
->total_sectors
= header
.size
/ 512;
607 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
608 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
609 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
611 s
->refcount_table_offset
= header
.refcount_table_offset
;
612 s
->refcount_table_size
=
613 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
615 if (header
.refcount_table_clusters
> qcow2_max_refcount_clusters(s
)) {
616 error_setg(errp
, "Reference count table too large");
621 ret
= validate_table_offset(bs
, s
->refcount_table_offset
,
622 s
->refcount_table_size
, sizeof(uint64_t));
624 error_setg(errp
, "Invalid reference count table offset");
628 /* Snapshot table offset/length */
629 if (header
.nb_snapshots
> QCOW_MAX_SNAPSHOTS
) {
630 error_setg(errp
, "Too many snapshots");
635 ret
= validate_table_offset(bs
, header
.snapshots_offset
,
637 sizeof(QCowSnapshotHeader
));
639 error_setg(errp
, "Invalid snapshot table offset");
643 /* read the level 1 table */
644 if (header
.l1_size
> QCOW_MAX_L1_SIZE
) {
645 error_setg(errp
, "Active L1 table too large");
649 s
->l1_size
= header
.l1_size
;
651 l1_vm_state_index
= size_to_l1(s
, header
.size
);
652 if (l1_vm_state_index
> INT_MAX
) {
653 error_setg(errp
, "Image is too big");
657 s
->l1_vm_state_index
= l1_vm_state_index
;
659 /* the L1 table must contain at least enough entries to put
661 if (s
->l1_size
< s
->l1_vm_state_index
) {
662 error_setg(errp
, "L1 table is too small");
667 ret
= validate_table_offset(bs
, header
.l1_table_offset
,
668 header
.l1_size
, sizeof(uint64_t));
670 error_setg(errp
, "Invalid L1 table offset");
673 s
->l1_table_offset
= header
.l1_table_offset
;
676 if (s
->l1_size
> 0) {
677 s
->l1_table
= g_malloc0(
678 align_offset(s
->l1_size
* sizeof(uint64_t), 512));
679 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
680 s
->l1_size
* sizeof(uint64_t));
682 error_setg_errno(errp
, -ret
, "Could not read L1 table");
685 for(i
= 0;i
< s
->l1_size
; i
++) {
686 be64_to_cpus(&s
->l1_table
[i
]);
690 /* alloc L2 table/refcount block cache */
691 s
->l2_table_cache
= qcow2_cache_create(bs
, L2_CACHE_SIZE
);
692 s
->refcount_block_cache
= qcow2_cache_create(bs
, REFCOUNT_CACHE_SIZE
);
694 s
->cluster_cache
= g_malloc(s
->cluster_size
);
695 /* one more sector for decompressed data alignment */
696 s
->cluster_data
= qemu_blockalign(bs
, QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
698 s
->cluster_cache_offset
= -1;
701 ret
= qcow2_refcount_init(bs
);
703 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
707 QLIST_INIT(&s
->cluster_allocs
);
708 QTAILQ_INIT(&s
->discards
);
710 /* read qcow2 extensions */
711 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
713 error_propagate(errp
, local_err
);
718 /* read the backing file name */
719 if (header
.backing_file_offset
!= 0) {
720 len
= header
.backing_file_size
;
721 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
)) {
722 error_setg(errp
, "Backing file name too long");
726 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
727 bs
->backing_file
, len
);
729 error_setg_errno(errp
, -ret
, "Could not read backing file name");
732 bs
->backing_file
[len
] = '\0';
735 /* Internal snapshots */
736 s
->snapshots_offset
= header
.snapshots_offset
;
737 s
->nb_snapshots
= header
.nb_snapshots
;
739 ret
= qcow2_read_snapshots(bs
);
741 error_setg_errno(errp
, -ret
, "Could not read snapshots");
745 /* Clear unknown autoclear feature bits */
746 if (!bs
->read_only
&& !(flags
& BDRV_O_INCOMING
) && s
->autoclear_features
) {
747 s
->autoclear_features
= 0;
748 ret
= qcow2_update_header(bs
);
750 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
755 /* Initialise locks */
756 qemu_co_mutex_init(&s
->lock
);
758 /* Repair image if dirty */
759 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INCOMING
)) && !bs
->read_only
&&
760 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
761 BdrvCheckResult result
= {0};
763 ret
= qcow2_check(bs
, &result
, BDRV_FIX_ERRORS
);
765 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
770 /* Enable lazy_refcounts according to image and command line options */
771 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
772 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
774 error_propagate(errp
, local_err
);
779 s
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
780 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
782 s
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
783 s
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
784 s
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
785 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
786 flags
& BDRV_O_UNMAP
);
787 s
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
788 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
789 s
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
790 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
792 opt_overlap_check
= qemu_opt_get(opts
, "overlap-check") ?: "cached";
793 if (!strcmp(opt_overlap_check
, "none")) {
794 overlap_check_template
= 0;
795 } else if (!strcmp(opt_overlap_check
, "constant")) {
796 overlap_check_template
= QCOW2_OL_CONSTANT
;
797 } else if (!strcmp(opt_overlap_check
, "cached")) {
798 overlap_check_template
= QCOW2_OL_CACHED
;
799 } else if (!strcmp(opt_overlap_check
, "all")) {
800 overlap_check_template
= QCOW2_OL_ALL
;
802 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
803 "'overlap-check'. Allowed are either of the following: "
804 "none, constant, cached, all", opt_overlap_check
);
810 s
->overlap_check
= 0;
811 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
812 /* overlap-check defines a template bitmask, but every flag may be
813 * overwritten through the associated boolean option */
815 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
816 overlap_check_template
& (1 << i
)) << i
;
821 if (s
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
822 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
823 "qemu 1.1 compatibility level");
830 BdrvCheckResult result
= {0};
831 qcow2_check_refcounts(bs
, &result
, 0);
837 g_free(s
->unknown_header_fields
);
838 cleanup_unknown_header_ext(bs
);
839 qcow2_free_snapshots(bs
);
840 qcow2_refcount_close(bs
);
842 /* else pre-write overlap checks in cache_destroy may crash */
844 if (s
->l2_table_cache
) {
845 qcow2_cache_destroy(bs
, s
->l2_table_cache
);
847 if (s
->refcount_block_cache
) {
848 qcow2_cache_destroy(bs
, s
->refcount_block_cache
);
850 g_free(s
->cluster_cache
);
851 qemu_vfree(s
->cluster_data
);
855 static int qcow2_refresh_limits(BlockDriverState
*bs
)
857 BDRVQcowState
*s
= bs
->opaque
;
859 bs
->bl
.write_zeroes_alignment
= s
->cluster_sectors
;
864 static int qcow2_set_key(BlockDriverState
*bs
, const char *key
)
866 BDRVQcowState
*s
= bs
->opaque
;
870 memset(keybuf
, 0, 16);
874 /* XXX: we could compress the chars to 7 bits to increase
876 for(i
= 0;i
< len
;i
++) {
879 s
->crypt_method
= s
->crypt_method_header
;
881 if (AES_set_encrypt_key(keybuf
, 128, &s
->aes_encrypt_key
) != 0)
883 if (AES_set_decrypt_key(keybuf
, 128, &s
->aes_decrypt_key
) != 0)
893 AES_encrypt(in
, tmp
, &s
->aes_encrypt_key
);
894 AES_decrypt(tmp
, out
, &s
->aes_decrypt_key
);
895 for(i
= 0; i
< 16; i
++)
896 printf(" %02x", tmp
[i
]);
898 for(i
= 0; i
< 16; i
++)
899 printf(" %02x", out
[i
]);
906 /* We have no actual commit/abort logic for qcow2, but we need to write out any
907 * unwritten data if we reopen read-only. */
908 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
909 BlockReopenQueue
*queue
, Error
**errp
)
913 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
914 ret
= bdrv_flush(state
->bs
);
919 ret
= qcow2_mark_clean(state
->bs
);
928 static int64_t coroutine_fn
qcow2_co_get_block_status(BlockDriverState
*bs
,
929 int64_t sector_num
, int nb_sectors
, int *pnum
)
931 BDRVQcowState
*s
= bs
->opaque
;
932 uint64_t cluster_offset
;
933 int index_in_cluster
, ret
;
937 qemu_co_mutex_lock(&s
->lock
);
938 ret
= qcow2_get_cluster_offset(bs
, sector_num
<< 9, pnum
, &cluster_offset
);
939 qemu_co_mutex_unlock(&s
->lock
);
944 if (cluster_offset
!= 0 && ret
!= QCOW2_CLUSTER_COMPRESSED
&&
946 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
947 cluster_offset
|= (index_in_cluster
<< BDRV_SECTOR_BITS
);
948 status
|= BDRV_BLOCK_OFFSET_VALID
| cluster_offset
;
950 if (ret
== QCOW2_CLUSTER_ZERO
) {
951 status
|= BDRV_BLOCK_ZERO
;
952 } else if (ret
!= QCOW2_CLUSTER_UNALLOCATED
) {
953 status
|= BDRV_BLOCK_DATA
;
958 /* handle reading after the end of the backing file */
959 int qcow2_backing_read1(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
960 int64_t sector_num
, int nb_sectors
)
963 if ((sector_num
+ nb_sectors
) <= bs
->total_sectors
)
965 if (sector_num
>= bs
->total_sectors
)
968 n1
= bs
->total_sectors
- sector_num
;
970 qemu_iovec_memset(qiov
, 512 * n1
, 0, 512 * (nb_sectors
- n1
));
975 static coroutine_fn
int qcow2_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
976 int remaining_sectors
, QEMUIOVector
*qiov
)
978 BDRVQcowState
*s
= bs
->opaque
;
979 int index_in_cluster
, n1
;
981 int cur_nr_sectors
; /* number of sectors in current iteration */
982 uint64_t cluster_offset
= 0;
983 uint64_t bytes_done
= 0;
984 QEMUIOVector hd_qiov
;
985 uint8_t *cluster_data
= NULL
;
987 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
989 qemu_co_mutex_lock(&s
->lock
);
991 while (remaining_sectors
!= 0) {
993 /* prepare next request */
994 cur_nr_sectors
= remaining_sectors
;
995 if (s
->crypt_method
) {
996 cur_nr_sectors
= MIN(cur_nr_sectors
,
997 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
);
1000 ret
= qcow2_get_cluster_offset(bs
, sector_num
<< 9,
1001 &cur_nr_sectors
, &cluster_offset
);
1006 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
1008 qemu_iovec_reset(&hd_qiov
);
1009 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
,
1010 cur_nr_sectors
* 512);
1013 case QCOW2_CLUSTER_UNALLOCATED
:
1015 if (bs
->backing_hd
) {
1016 /* read from the base image */
1017 n1
= qcow2_backing_read1(bs
->backing_hd
, &hd_qiov
,
1018 sector_num
, cur_nr_sectors
);
1020 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
1021 qemu_co_mutex_unlock(&s
->lock
);
1022 ret
= bdrv_co_readv(bs
->backing_hd
, sector_num
,
1024 qemu_co_mutex_lock(&s
->lock
);
1030 /* Note: in this case, no need to wait */
1031 qemu_iovec_memset(&hd_qiov
, 0, 0, 512 * cur_nr_sectors
);
1035 case QCOW2_CLUSTER_ZERO
:
1036 qemu_iovec_memset(&hd_qiov
, 0, 0, 512 * cur_nr_sectors
);
1039 case QCOW2_CLUSTER_COMPRESSED
:
1040 /* add AIO support for compressed blocks ? */
1041 ret
= qcow2_decompress_cluster(bs
, cluster_offset
);
1046 qemu_iovec_from_buf(&hd_qiov
, 0,
1047 s
->cluster_cache
+ index_in_cluster
* 512,
1048 512 * cur_nr_sectors
);
1051 case QCOW2_CLUSTER_NORMAL
:
1052 if ((cluster_offset
& 511) != 0) {
1057 if (s
->crypt_method
) {
1059 * For encrypted images, read everything into a temporary
1060 * contiguous buffer on which the AES functions can work.
1062 if (!cluster_data
) {
1064 qemu_blockalign(bs
, QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1067 assert(cur_nr_sectors
<=
1068 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
);
1069 qemu_iovec_reset(&hd_qiov
);
1070 qemu_iovec_add(&hd_qiov
, cluster_data
,
1071 512 * cur_nr_sectors
);
1074 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1075 qemu_co_mutex_unlock(&s
->lock
);
1076 ret
= bdrv_co_readv(bs
->file
,
1077 (cluster_offset
>> 9) + index_in_cluster
,
1078 cur_nr_sectors
, &hd_qiov
);
1079 qemu_co_mutex_lock(&s
->lock
);
1083 if (s
->crypt_method
) {
1084 qcow2_encrypt_sectors(s
, sector_num
, cluster_data
,
1085 cluster_data
, cur_nr_sectors
, 0, &s
->aes_decrypt_key
);
1086 qemu_iovec_from_buf(qiov
, bytes_done
,
1087 cluster_data
, 512 * cur_nr_sectors
);
1092 g_assert_not_reached();
1097 remaining_sectors
-= cur_nr_sectors
;
1098 sector_num
+= cur_nr_sectors
;
1099 bytes_done
+= cur_nr_sectors
* 512;
1104 qemu_co_mutex_unlock(&s
->lock
);
1106 qemu_iovec_destroy(&hd_qiov
);
1107 qemu_vfree(cluster_data
);
1112 static coroutine_fn
int qcow2_co_writev(BlockDriverState
*bs
,
1114 int remaining_sectors
,
1117 BDRVQcowState
*s
= bs
->opaque
;
1118 int index_in_cluster
;
1120 int cur_nr_sectors
; /* number of sectors in current iteration */
1121 uint64_t cluster_offset
;
1122 QEMUIOVector hd_qiov
;
1123 uint64_t bytes_done
= 0;
1124 uint8_t *cluster_data
= NULL
;
1125 QCowL2Meta
*l2meta
= NULL
;
1127 trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num
,
1130 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1132 s
->cluster_cache_offset
= -1; /* disable compressed cache */
1134 qemu_co_mutex_lock(&s
->lock
);
1136 while (remaining_sectors
!= 0) {
1140 trace_qcow2_writev_start_part(qemu_coroutine_self());
1141 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
1142 cur_nr_sectors
= remaining_sectors
;
1143 if (s
->crypt_method
&&
1145 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
- index_in_cluster
) {
1147 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
- index_in_cluster
;
1150 ret
= qcow2_alloc_cluster_offset(bs
, sector_num
<< 9,
1151 &cur_nr_sectors
, &cluster_offset
, &l2meta
);
1156 assert((cluster_offset
& 511) == 0);
1158 qemu_iovec_reset(&hd_qiov
);
1159 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
,
1160 cur_nr_sectors
* 512);
1162 if (s
->crypt_method
) {
1163 if (!cluster_data
) {
1164 cluster_data
= qemu_blockalign(bs
, QCOW_MAX_CRYPT_CLUSTERS
*
1168 assert(hd_qiov
.size
<=
1169 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1170 qemu_iovec_to_buf(&hd_qiov
, 0, cluster_data
, hd_qiov
.size
);
1172 qcow2_encrypt_sectors(s
, sector_num
, cluster_data
,
1173 cluster_data
, cur_nr_sectors
, 1, &s
->aes_encrypt_key
);
1175 qemu_iovec_reset(&hd_qiov
);
1176 qemu_iovec_add(&hd_qiov
, cluster_data
,
1177 cur_nr_sectors
* 512);
1180 ret
= qcow2_pre_write_overlap_check(bs
, 0,
1181 cluster_offset
+ index_in_cluster
* BDRV_SECTOR_SIZE
,
1182 cur_nr_sectors
* BDRV_SECTOR_SIZE
);
1187 qemu_co_mutex_unlock(&s
->lock
);
1188 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
1189 trace_qcow2_writev_data(qemu_coroutine_self(),
1190 (cluster_offset
>> 9) + index_in_cluster
);
1191 ret
= bdrv_co_writev(bs
->file
,
1192 (cluster_offset
>> 9) + index_in_cluster
,
1193 cur_nr_sectors
, &hd_qiov
);
1194 qemu_co_mutex_lock(&s
->lock
);
1199 while (l2meta
!= NULL
) {
1202 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
1207 /* Take the request off the list of running requests */
1208 if (l2meta
->nb_clusters
!= 0) {
1209 QLIST_REMOVE(l2meta
, next_in_flight
);
1212 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
1214 next
= l2meta
->next
;
1219 remaining_sectors
-= cur_nr_sectors
;
1220 sector_num
+= cur_nr_sectors
;
1221 bytes_done
+= cur_nr_sectors
* 512;
1222 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors
);
1227 qemu_co_mutex_unlock(&s
->lock
);
1229 while (l2meta
!= NULL
) {
1232 if (l2meta
->nb_clusters
!= 0) {
1233 QLIST_REMOVE(l2meta
, next_in_flight
);
1235 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
1237 next
= l2meta
->next
;
1242 qemu_iovec_destroy(&hd_qiov
);
1243 qemu_vfree(cluster_data
);
1244 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
1249 static void qcow2_close(BlockDriverState
*bs
)
1251 BDRVQcowState
*s
= bs
->opaque
;
1252 g_free(s
->l1_table
);
1253 /* else pre-write overlap checks in cache_destroy may crash */
1256 if (!(bs
->open_flags
& BDRV_O_INCOMING
)) {
1257 qcow2_cache_flush(bs
, s
->l2_table_cache
);
1258 qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1260 qcow2_mark_clean(bs
);
1263 qcow2_cache_destroy(bs
, s
->l2_table_cache
);
1264 qcow2_cache_destroy(bs
, s
->refcount_block_cache
);
1266 g_free(s
->unknown_header_fields
);
1267 cleanup_unknown_header_ext(bs
);
1269 g_free(s
->cluster_cache
);
1270 qemu_vfree(s
->cluster_data
);
1271 qcow2_refcount_close(bs
);
1272 qcow2_free_snapshots(bs
);
1275 static void qcow2_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
1277 BDRVQcowState
*s
= bs
->opaque
;
1278 int flags
= s
->flags
;
1279 AES_KEY aes_encrypt_key
;
1280 AES_KEY aes_decrypt_key
;
1281 uint32_t crypt_method
= 0;
1283 Error
*local_err
= NULL
;
1287 * Backing files are read-only which makes all of their metadata immutable,
1288 * that means we don't have to worry about reopening them here.
1291 if (s
->crypt_method
) {
1292 crypt_method
= s
->crypt_method
;
1293 memcpy(&aes_encrypt_key
, &s
->aes_encrypt_key
, sizeof(aes_encrypt_key
));
1294 memcpy(&aes_decrypt_key
, &s
->aes_decrypt_key
, sizeof(aes_decrypt_key
));
1299 bdrv_invalidate_cache(bs
->file
, &local_err
);
1301 error_propagate(errp
, local_err
);
1305 memset(s
, 0, sizeof(BDRVQcowState
));
1306 options
= qdict_clone_shallow(bs
->options
);
1308 ret
= qcow2_open(bs
, options
, flags
, &local_err
);
1310 error_setg(errp
, "Could not reopen qcow2 layer: %s",
1311 error_get_pretty(local_err
));
1312 error_free(local_err
);
1314 } else if (ret
< 0) {
1315 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
1322 s
->crypt_method
= crypt_method
;
1323 memcpy(&s
->aes_encrypt_key
, &aes_encrypt_key
, sizeof(aes_encrypt_key
));
1324 memcpy(&s
->aes_decrypt_key
, &aes_decrypt_key
, sizeof(aes_decrypt_key
));
1328 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
1329 size_t len
, size_t buflen
)
1331 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
1332 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
1334 if (buflen
< ext_len
) {
1338 *ext_backing_fmt
= (QCowExtension
) {
1339 .magic
= cpu_to_be32(magic
),
1340 .len
= cpu_to_be32(len
),
1342 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
1348 * Updates the qcow2 header, including the variable length parts of it, i.e.
1349 * the backing file name and all extensions. qcow2 was not designed to allow
1350 * such changes, so if we run out of space (we can only use the first cluster)
1351 * this function may fail.
1353 * Returns 0 on success, -errno in error cases.
1355 int qcow2_update_header(BlockDriverState
*bs
)
1357 BDRVQcowState
*s
= bs
->opaque
;
1360 size_t buflen
= s
->cluster_size
;
1362 uint64_t total_size
;
1363 uint32_t refcount_table_clusters
;
1364 size_t header_length
;
1365 Qcow2UnknownHeaderExtension
*uext
;
1367 buf
= qemu_blockalign(bs
, buflen
);
1369 /* Header structure */
1370 header
= (QCowHeader
*) buf
;
1372 if (buflen
< sizeof(*header
)) {
1377 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
1378 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
1379 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
1381 *header
= (QCowHeader
) {
1382 /* Version 2 fields */
1383 .magic
= cpu_to_be32(QCOW_MAGIC
),
1384 .version
= cpu_to_be32(s
->qcow_version
),
1385 .backing_file_offset
= 0,
1386 .backing_file_size
= 0,
1387 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
1388 .size
= cpu_to_be64(total_size
),
1389 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
1390 .l1_size
= cpu_to_be32(s
->l1_size
),
1391 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
1392 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
1393 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
1394 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
1395 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
1397 /* Version 3 fields */
1398 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
1399 .compatible_features
= cpu_to_be64(s
->compatible_features
),
1400 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
1401 .refcount_order
= cpu_to_be32(s
->refcount_order
),
1402 .header_length
= cpu_to_be32(header_length
),
1405 /* For older versions, write a shorter header */
1406 switch (s
->qcow_version
) {
1408 ret
= offsetof(QCowHeader
, incompatible_features
);
1411 ret
= sizeof(*header
);
1420 memset(buf
, 0, buflen
);
1422 /* Preserve any unknown field in the header */
1423 if (s
->unknown_header_fields_size
) {
1424 if (buflen
< s
->unknown_header_fields_size
) {
1429 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
1430 buf
+= s
->unknown_header_fields_size
;
1431 buflen
-= s
->unknown_header_fields_size
;
1434 /* Backing file format header extension */
1435 if (*bs
->backing_format
) {
1436 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
1437 bs
->backing_format
, strlen(bs
->backing_format
),
1448 Qcow2Feature features
[] = {
1450 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
1451 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
1452 .name
= "dirty bit",
1455 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
1456 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
1457 .name
= "corrupt bit",
1460 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
1461 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
1462 .name
= "lazy refcounts",
1466 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
1467 features
, sizeof(features
), buflen
);
1474 /* Keep unknown header extensions */
1475 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
1476 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
1485 /* End of header extensions */
1486 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
1494 /* Backing file name */
1495 if (*bs
->backing_file
) {
1496 size_t backing_file_len
= strlen(bs
->backing_file
);
1498 if (buflen
< backing_file_len
) {
1503 /* Using strncpy is ok here, since buf is not NUL-terminated. */
1504 strncpy(buf
, bs
->backing_file
, buflen
);
1506 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
1507 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
1510 /* Write the new header */
1511 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
1522 static int qcow2_change_backing_file(BlockDriverState
*bs
,
1523 const char *backing_file
, const char *backing_fmt
)
1525 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
1526 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
1528 return qcow2_update_header(bs
);
1531 static int preallocate(BlockDriverState
*bs
)
1533 uint64_t nb_sectors
;
1535 uint64_t host_offset
= 0;
1540 nb_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
1543 while (nb_sectors
) {
1544 num
= MIN(nb_sectors
, INT_MAX
>> BDRV_SECTOR_BITS
);
1545 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &num
,
1546 &host_offset
, &meta
);
1552 QCowL2Meta
*next
= meta
->next
;
1554 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
1556 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
1557 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
1561 /* There are no dependent requests, but we need to remove our
1562 * request from the list of in-flight requests */
1563 QLIST_REMOVE(meta
, next_in_flight
);
1569 /* TODO Preallocate data if requested */
1572 offset
+= num
<< BDRV_SECTOR_BITS
;
1576 * It is expected that the image file is large enough to actually contain
1577 * all of the allocated clusters (otherwise we get failing reads after
1578 * EOF). Extend the image to the last allocated sector.
1580 if (host_offset
!= 0) {
1581 uint8_t buf
[BDRV_SECTOR_SIZE
];
1582 memset(buf
, 0, BDRV_SECTOR_SIZE
);
1583 ret
= bdrv_write(bs
->file
, (host_offset
>> BDRV_SECTOR_BITS
) + num
- 1,
1593 static int qcow2_create2(const char *filename
, int64_t total_size
,
1594 const char *backing_file
, const char *backing_format
,
1595 int flags
, size_t cluster_size
, int prealloc
,
1596 QEMUOptionParameter
*options
, int version
,
1599 /* Calculate cluster_bits */
1601 cluster_bits
= ffs(cluster_size
) - 1;
1602 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
1603 (1 << cluster_bits
) != cluster_size
)
1605 error_setg(errp
, "Cluster size must be a power of two between %d and "
1606 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
1611 * Open the image file and write a minimal qcow2 header.
1613 * We keep things simple and start with a zero-sized image. We also
1614 * do without refcount blocks or a L1 table for now. We'll fix the
1615 * inconsistency later.
1617 * We do need a refcount table because growing the refcount table means
1618 * allocating two new refcount blocks - the seconds of which would be at
1619 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1620 * size for any qcow2 image.
1622 BlockDriverState
* bs
;
1624 uint64_t* refcount_table
;
1625 Error
*local_err
= NULL
;
1628 ret
= bdrv_create_file(filename
, options
, &local_err
);
1630 error_propagate(errp
, local_err
);
1635 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
, BDRV_O_RDWR
| BDRV_O_PROTOCOL
,
1638 error_propagate(errp
, local_err
);
1642 /* Write the header */
1643 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
1644 header
= g_malloc0(cluster_size
);
1645 *header
= (QCowHeader
) {
1646 .magic
= cpu_to_be32(QCOW_MAGIC
),
1647 .version
= cpu_to_be32(version
),
1648 .cluster_bits
= cpu_to_be32(cluster_bits
),
1649 .size
= cpu_to_be64(0),
1650 .l1_table_offset
= cpu_to_be64(0),
1651 .l1_size
= cpu_to_be32(0),
1652 .refcount_table_offset
= cpu_to_be64(cluster_size
),
1653 .refcount_table_clusters
= cpu_to_be32(1),
1654 .refcount_order
= cpu_to_be32(3 + REFCOUNT_SHIFT
),
1655 .header_length
= cpu_to_be32(sizeof(*header
)),
1658 if (flags
& BLOCK_FLAG_ENCRYPT
) {
1659 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
1661 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
1664 if (flags
& BLOCK_FLAG_LAZY_REFCOUNTS
) {
1665 header
->compatible_features
|=
1666 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
1669 ret
= bdrv_pwrite(bs
, 0, header
, cluster_size
);
1672 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
1676 /* Write a refcount table with one refcount block */
1677 refcount_table
= g_malloc0(2 * cluster_size
);
1678 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
1679 ret
= bdrv_pwrite(bs
, cluster_size
, refcount_table
, 2 * cluster_size
);
1680 g_free(refcount_table
);
1683 error_setg_errno(errp
, -ret
, "Could not write refcount table");
1691 * And now open the image and make it consistent first (i.e. increase the
1692 * refcount of the cluster that is occupied by the header and the refcount
1695 BlockDriver
* drv
= bdrv_find_format("qcow2");
1696 assert(drv
!= NULL
);
1697 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
,
1698 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_NO_FLUSH
, drv
, &local_err
);
1700 error_propagate(errp
, local_err
);
1704 ret
= qcow2_alloc_clusters(bs
, 3 * cluster_size
);
1706 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
1707 "header and refcount table");
1710 } else if (ret
!= 0) {
1711 error_report("Huh, first cluster in empty image is already in use?");
1715 /* Okay, now that we have a valid image, let's give it the right size */
1716 ret
= bdrv_truncate(bs
, total_size
* BDRV_SECTOR_SIZE
);
1718 error_setg_errno(errp
, -ret
, "Could not resize image");
1722 /* Want a backing file? There you go.*/
1724 ret
= bdrv_change_backing_file(bs
, backing_file
, backing_format
);
1726 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
1727 "with format '%s'", backing_file
, backing_format
);
1732 /* And if we're supposed to preallocate metadata, do that now */
1734 BDRVQcowState
*s
= bs
->opaque
;
1735 qemu_co_mutex_lock(&s
->lock
);
1736 ret
= preallocate(bs
);
1737 qemu_co_mutex_unlock(&s
->lock
);
1739 error_setg_errno(errp
, -ret
, "Could not preallocate metadata");
1747 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
1748 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
,
1749 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_NO_BACKING
,
1752 error_propagate(errp
, local_err
);
1764 static int qcow2_create(const char *filename
, QEMUOptionParameter
*options
,
1767 const char *backing_file
= NULL
;
1768 const char *backing_fmt
= NULL
;
1769 uint64_t sectors
= 0;
1771 size_t cluster_size
= DEFAULT_CLUSTER_SIZE
;
1774 Error
*local_err
= NULL
;
1777 /* Read out options */
1778 while (options
&& options
->name
) {
1779 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
1780 sectors
= options
->value
.n
/ 512;
1781 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
1782 backing_file
= options
->value
.s
;
1783 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FMT
)) {
1784 backing_fmt
= options
->value
.s
;
1785 } else if (!strcmp(options
->name
, BLOCK_OPT_ENCRYPT
)) {
1786 flags
|= options
->value
.n
? BLOCK_FLAG_ENCRYPT
: 0;
1787 } else if (!strcmp(options
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
1788 if (options
->value
.n
) {
1789 cluster_size
= options
->value
.n
;
1791 } else if (!strcmp(options
->name
, BLOCK_OPT_PREALLOC
)) {
1792 if (!options
->value
.s
|| !strcmp(options
->value
.s
, "off")) {
1794 } else if (!strcmp(options
->value
.s
, "metadata")) {
1797 error_setg(errp
, "Invalid preallocation mode: '%s'",
1801 } else if (!strcmp(options
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
1802 if (!options
->value
.s
) {
1803 /* keep the default */
1804 } else if (!strcmp(options
->value
.s
, "0.10")) {
1806 } else if (!strcmp(options
->value
.s
, "1.1")) {
1809 error_setg(errp
, "Invalid compatibility level: '%s'",
1813 } else if (!strcmp(options
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
1814 flags
|= options
->value
.n
? BLOCK_FLAG_LAZY_REFCOUNTS
: 0;
1819 if (backing_file
&& prealloc
) {
1820 error_setg(errp
, "Backing file and preallocation cannot be used at "
1825 if (version
< 3 && (flags
& BLOCK_FLAG_LAZY_REFCOUNTS
)) {
1826 error_setg(errp
, "Lazy refcounts only supported with compatibility "
1827 "level 1.1 and above (use compat=1.1 or greater)");
1831 ret
= qcow2_create2(filename
, sectors
, backing_file
, backing_fmt
, flags
,
1832 cluster_size
, prealloc
, options
, version
, &local_err
);
1834 error_propagate(errp
, local_err
);
1839 static coroutine_fn
int qcow2_co_write_zeroes(BlockDriverState
*bs
,
1840 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
1843 BDRVQcowState
*s
= bs
->opaque
;
1845 /* Emulate misaligned zero writes */
1846 if (sector_num
% s
->cluster_sectors
|| nb_sectors
% s
->cluster_sectors
) {
1850 /* Whatever is left can use real zero clusters */
1851 qemu_co_mutex_lock(&s
->lock
);
1852 ret
= qcow2_zero_clusters(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1854 qemu_co_mutex_unlock(&s
->lock
);
1859 static coroutine_fn
int qcow2_co_discard(BlockDriverState
*bs
,
1860 int64_t sector_num
, int nb_sectors
)
1863 BDRVQcowState
*s
= bs
->opaque
;
1865 qemu_co_mutex_lock(&s
->lock
);
1866 ret
= qcow2_discard_clusters(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1867 nb_sectors
, QCOW2_DISCARD_REQUEST
);
1868 qemu_co_mutex_unlock(&s
->lock
);
1872 static int qcow2_truncate(BlockDriverState
*bs
, int64_t offset
)
1874 BDRVQcowState
*s
= bs
->opaque
;
1875 int64_t new_l1_size
;
1879 error_report("The new size must be a multiple of 512");
1883 /* cannot proceed if image has snapshots */
1884 if (s
->nb_snapshots
) {
1885 error_report("Can't resize an image which has snapshots");
1889 /* shrinking is currently not supported */
1890 if (offset
< bs
->total_sectors
* 512) {
1891 error_report("qcow2 doesn't support shrinking images yet");
1895 new_l1_size
= size_to_l1(s
, offset
);
1896 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
1901 /* write updated header.size */
1902 offset
= cpu_to_be64(offset
);
1903 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
1904 &offset
, sizeof(uint64_t));
1909 s
->l1_vm_state_index
= new_l1_size
;
1913 /* XXX: put compressed sectors first, then all the cluster aligned
1914 tables to avoid losing bytes in alignment */
1915 static int qcow2_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1916 const uint8_t *buf
, int nb_sectors
)
1918 BDRVQcowState
*s
= bs
->opaque
;
1922 uint64_t cluster_offset
;
1924 if (nb_sectors
== 0) {
1925 /* align end of file to a sector boundary to ease reading with
1926 sector based I/Os */
1927 cluster_offset
= bdrv_getlength(bs
->file
);
1928 cluster_offset
= (cluster_offset
+ 511) & ~511;
1929 bdrv_truncate(bs
->file
, cluster_offset
);
1933 if (nb_sectors
!= s
->cluster_sectors
) {
1936 /* Zero-pad last write if image size is not cluster aligned */
1937 if (sector_num
+ nb_sectors
== bs
->total_sectors
&&
1938 nb_sectors
< s
->cluster_sectors
) {
1939 uint8_t *pad_buf
= qemu_blockalign(bs
, s
->cluster_size
);
1940 memset(pad_buf
, 0, s
->cluster_size
);
1941 memcpy(pad_buf
, buf
, nb_sectors
* BDRV_SECTOR_SIZE
);
1942 ret
= qcow2_write_compressed(bs
, sector_num
,
1943 pad_buf
, s
->cluster_sectors
);
1944 qemu_vfree(pad_buf
);
1949 out_buf
= g_malloc(s
->cluster_size
+ (s
->cluster_size
/ 1000) + 128);
1951 /* best compression, small window, no zlib header */
1952 memset(&strm
, 0, sizeof(strm
));
1953 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
1955 9, Z_DEFAULT_STRATEGY
);
1961 strm
.avail_in
= s
->cluster_size
;
1962 strm
.next_in
= (uint8_t *)buf
;
1963 strm
.avail_out
= s
->cluster_size
;
1964 strm
.next_out
= out_buf
;
1966 ret
= deflate(&strm
, Z_FINISH
);
1967 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
1972 out_len
= strm
.next_out
- out_buf
;
1976 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
1977 /* could not compress: write normal cluster */
1978 ret
= bdrv_write(bs
, sector_num
, buf
, s
->cluster_sectors
);
1983 cluster_offset
= qcow2_alloc_compressed_cluster_offset(bs
,
1984 sector_num
<< 9, out_len
);
1985 if (!cluster_offset
) {
1989 cluster_offset
&= s
->cluster_offset_mask
;
1991 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
);
1996 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
1997 ret
= bdrv_pwrite(bs
->file
, cluster_offset
, out_buf
, out_len
);
2009 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
2011 BDRVQcowState
*s
= bs
->opaque
;
2014 qemu_co_mutex_lock(&s
->lock
);
2015 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2017 qemu_co_mutex_unlock(&s
->lock
);
2021 if (qcow2_need_accurate_refcounts(s
)) {
2022 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2024 qemu_co_mutex_unlock(&s
->lock
);
2028 qemu_co_mutex_unlock(&s
->lock
);
2033 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2035 BDRVQcowState
*s
= bs
->opaque
;
2036 bdi
->unallocated_blocks_are_zero
= true;
2037 bdi
->can_write_zeroes_with_unmap
= (s
->qcow_version
>= 3);
2038 bdi
->cluster_size
= s
->cluster_size
;
2039 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
2043 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
)
2045 BDRVQcowState
*s
= bs
->opaque
;
2046 ImageInfoSpecific
*spec_info
= g_new(ImageInfoSpecific
, 1);
2048 *spec_info
= (ImageInfoSpecific
){
2049 .kind
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
2051 .qcow2
= g_new(ImageInfoSpecificQCow2
, 1),
2054 if (s
->qcow_version
== 2) {
2055 *spec_info
->qcow2
= (ImageInfoSpecificQCow2
){
2056 .compat
= g_strdup("0.10"),
2058 } else if (s
->qcow_version
== 3) {
2059 *spec_info
->qcow2
= (ImageInfoSpecificQCow2
){
2060 .compat
= g_strdup("1.1"),
2061 .lazy_refcounts
= s
->compatible_features
&
2062 QCOW2_COMPAT_LAZY_REFCOUNTS
,
2063 .has_lazy_refcounts
= true,
2071 static void dump_refcounts(BlockDriverState
*bs
)
2073 BDRVQcowState
*s
= bs
->opaque
;
2074 int64_t nb_clusters
, k
, k1
, size
;
2077 size
= bdrv_getlength(bs
->file
);
2078 nb_clusters
= size_to_clusters(s
, size
);
2079 for(k
= 0; k
< nb_clusters
;) {
2081 refcount
= get_refcount(bs
, k
);
2083 while (k
< nb_clusters
&& get_refcount(bs
, k
) == refcount
)
2085 printf("%" PRId64
": refcount=%d nb=%" PRId64
"\n", k
, refcount
,
2091 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
2094 BDRVQcowState
*s
= bs
->opaque
;
2095 int64_t total_sectors
= bs
->total_sectors
;
2096 int growable
= bs
->growable
;
2097 bool zero_beyond_eof
= bs
->zero_beyond_eof
;
2100 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
2102 bs
->zero_beyond_eof
= false;
2103 ret
= bdrv_pwritev(bs
, qcow2_vm_state_offset(s
) + pos
, qiov
);
2104 bs
->growable
= growable
;
2105 bs
->zero_beyond_eof
= zero_beyond_eof
;
2107 /* bdrv_co_do_writev will have increased the total_sectors value to include
2108 * the VM state - the VM state is however not an actual part of the block
2109 * device, therefore, we need to restore the old value. */
2110 bs
->total_sectors
= total_sectors
;
2115 static int qcow2_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2116 int64_t pos
, int size
)
2118 BDRVQcowState
*s
= bs
->opaque
;
2119 int growable
= bs
->growable
;
2120 bool zero_beyond_eof
= bs
->zero_beyond_eof
;
2123 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
2125 bs
->zero_beyond_eof
= false;
2126 ret
= bdrv_pread(bs
, qcow2_vm_state_offset(s
) + pos
, buf
, size
);
2127 bs
->growable
= growable
;
2128 bs
->zero_beyond_eof
= zero_beyond_eof
;
2134 * Downgrades an image's version. To achieve this, any incompatible features
2135 * have to be removed.
2137 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
)
2139 BDRVQcowState
*s
= bs
->opaque
;
2140 int current_version
= s
->qcow_version
;
2143 if (target_version
== current_version
) {
2145 } else if (target_version
> current_version
) {
2147 } else if (target_version
!= 2) {
2151 if (s
->refcount_order
!= 4) {
2152 /* we would have to convert the image to a refcount_order == 4 image
2153 * here; however, since qemu (at the time of writing this) does not
2154 * support anything different than 4 anyway, there is no point in doing
2155 * so right now; however, we should error out (if qemu supports this in
2156 * the future and this code has not been adapted) */
2157 error_report("qcow2_downgrade: Image refcount orders other than 4 are "
2158 "currently not supported.");
2162 /* clear incompatible features */
2163 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
2164 ret
= qcow2_mark_clean(bs
);
2170 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2171 * the first place; if that happens nonetheless, returning -ENOTSUP is the
2172 * best thing to do anyway */
2174 if (s
->incompatible_features
) {
2178 /* since we can ignore compatible features, we can set them to 0 as well */
2179 s
->compatible_features
= 0;
2180 /* if lazy refcounts have been used, they have already been fixed through
2181 * clearing the dirty flag */
2183 /* clearing autoclear features is trivial */
2184 s
->autoclear_features
= 0;
2186 ret
= qcow2_expand_zero_clusters(bs
);
2191 s
->qcow_version
= target_version
;
2192 ret
= qcow2_update_header(bs
);
2194 s
->qcow_version
= current_version
;
2200 static int qcow2_amend_options(BlockDriverState
*bs
,
2201 QEMUOptionParameter
*options
)
2203 BDRVQcowState
*s
= bs
->opaque
;
2204 int old_version
= s
->qcow_version
, new_version
= old_version
;
2205 uint64_t new_size
= 0;
2206 const char *backing_file
= NULL
, *backing_format
= NULL
;
2207 bool lazy_refcounts
= s
->use_lazy_refcounts
;
2211 for (i
= 0; options
[i
].name
; i
++)
2213 if (!options
[i
].assigned
) {
2214 /* only change explicitly defined options */
2218 if (!strcmp(options
[i
].name
, "compat")) {
2219 if (!options
[i
].value
.s
) {
2220 /* preserve default */
2221 } else if (!strcmp(options
[i
].value
.s
, "0.10")) {
2223 } else if (!strcmp(options
[i
].value
.s
, "1.1")) {
2226 fprintf(stderr
, "Unknown compatibility level %s.\n",
2227 options
[i
].value
.s
);
2230 } else if (!strcmp(options
[i
].name
, "preallocation")) {
2231 fprintf(stderr
, "Cannot change preallocation mode.\n");
2233 } else if (!strcmp(options
[i
].name
, "size")) {
2234 new_size
= options
[i
].value
.n
;
2235 } else if (!strcmp(options
[i
].name
, "backing_file")) {
2236 backing_file
= options
[i
].value
.s
;
2237 } else if (!strcmp(options
[i
].name
, "backing_fmt")) {
2238 backing_format
= options
[i
].value
.s
;
2239 } else if (!strcmp(options
[i
].name
, "encryption")) {
2240 if ((options
[i
].value
.n
!= !!s
->crypt_method
)) {
2241 fprintf(stderr
, "Changing the encryption flag is not "
2245 } else if (!strcmp(options
[i
].name
, "cluster_size")) {
2246 if (options
[i
].value
.n
!= s
->cluster_size
) {
2247 fprintf(stderr
, "Changing the cluster size is not "
2251 } else if (!strcmp(options
[i
].name
, "lazy_refcounts")) {
2252 lazy_refcounts
= options
[i
].value
.n
;
2254 /* if this assertion fails, this probably means a new option was
2255 * added without having it covered here */
2260 if (new_version
!= old_version
) {
2261 if (new_version
> old_version
) {
2263 s
->qcow_version
= new_version
;
2264 ret
= qcow2_update_header(bs
);
2266 s
->qcow_version
= old_version
;
2270 ret
= qcow2_downgrade(bs
, new_version
);
2277 if (backing_file
|| backing_format
) {
2278 ret
= qcow2_change_backing_file(bs
, backing_file
?: bs
->backing_file
,
2279 backing_format
?: bs
->backing_format
);
2285 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
2286 if (lazy_refcounts
) {
2287 if (s
->qcow_version
< 3) {
2288 fprintf(stderr
, "Lazy refcounts only supported with compatibility "
2289 "level 1.1 and above (use compat=1.1 or greater)\n");
2292 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
2293 ret
= qcow2_update_header(bs
);
2295 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
2298 s
->use_lazy_refcounts
= true;
2300 /* make image clean first */
2301 ret
= qcow2_mark_clean(bs
);
2305 /* now disallow lazy refcounts */
2306 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
2307 ret
= qcow2_update_header(bs
);
2309 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
2312 s
->use_lazy_refcounts
= false;
2317 ret
= bdrv_truncate(bs
, new_size
);
2326 static QEMUOptionParameter qcow2_create_options
[] = {
2328 .name
= BLOCK_OPT_SIZE
,
2330 .help
= "Virtual disk size"
2333 .name
= BLOCK_OPT_COMPAT_LEVEL
,
2335 .help
= "Compatibility level (0.10 or 1.1)"
2338 .name
= BLOCK_OPT_BACKING_FILE
,
2340 .help
= "File name of a base image"
2343 .name
= BLOCK_OPT_BACKING_FMT
,
2345 .help
= "Image format of the base image"
2348 .name
= BLOCK_OPT_ENCRYPT
,
2350 .help
= "Encrypt the image"
2353 .name
= BLOCK_OPT_CLUSTER_SIZE
,
2355 .help
= "qcow2 cluster size",
2356 .value
= { .n
= DEFAULT_CLUSTER_SIZE
},
2359 .name
= BLOCK_OPT_PREALLOC
,
2361 .help
= "Preallocation mode (allowed values: off, metadata)"
2364 .name
= BLOCK_OPT_LAZY_REFCOUNTS
,
2366 .help
= "Postpone refcount updates",
2371 static BlockDriver bdrv_qcow2
= {
2372 .format_name
= "qcow2",
2373 .instance_size
= sizeof(BDRVQcowState
),
2374 .bdrv_probe
= qcow2_probe
,
2375 .bdrv_open
= qcow2_open
,
2376 .bdrv_close
= qcow2_close
,
2377 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
2378 .bdrv_create
= qcow2_create
,
2379 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2380 .bdrv_co_get_block_status
= qcow2_co_get_block_status
,
2381 .bdrv_set_key
= qcow2_set_key
,
2383 .bdrv_co_readv
= qcow2_co_readv
,
2384 .bdrv_co_writev
= qcow2_co_writev
,
2385 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
2387 .bdrv_co_write_zeroes
= qcow2_co_write_zeroes
,
2388 .bdrv_co_discard
= qcow2_co_discard
,
2389 .bdrv_truncate
= qcow2_truncate
,
2390 .bdrv_write_compressed
= qcow2_write_compressed
,
2392 .bdrv_snapshot_create
= qcow2_snapshot_create
,
2393 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
2394 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
2395 .bdrv_snapshot_list
= qcow2_snapshot_list
,
2396 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
2397 .bdrv_get_info
= qcow2_get_info
,
2398 .bdrv_get_specific_info
= qcow2_get_specific_info
,
2400 .bdrv_save_vmstate
= qcow2_save_vmstate
,
2401 .bdrv_load_vmstate
= qcow2_load_vmstate
,
2403 .bdrv_change_backing_file
= qcow2_change_backing_file
,
2405 .bdrv_refresh_limits
= qcow2_refresh_limits
,
2406 .bdrv_invalidate_cache
= qcow2_invalidate_cache
,
2408 .create_options
= qcow2_create_options
,
2409 .bdrv_check
= qcow2_check
,
2410 .bdrv_amend_options
= qcow2_amend_options
,
2413 static void bdrv_qcow2_init(void)
2415 bdrv_register(&bdrv_qcow2
);
2418 block_init(bdrv_qcow2_init
);