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_int.h"
29 #include "block/qcow2.h"
30 #include "qemu-error.h"
35 Differences with QCOW:
37 - Support for multiple incremental snapshots.
38 - Memory management by reference counts.
39 - Clusters which have a reference count of one have the bit
40 QCOW_OFLAG_COPIED to optimize write performance.
41 - Size of compressed clusters is stored in sectors to reduce bit usage
42 in the cluster offsets.
43 - Support for storing additional data (such as the VM state) in the
45 - If a backing store is used, the cluster size is not constrained
46 (could be backported to QCOW).
47 - L2 tables have always a size of one cluster.
55 #define QCOW2_EXT_MAGIC_END 0
56 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
58 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
60 const QCowHeader
*cow_header
= (const void *)buf
;
62 if (buf_size
>= sizeof(QCowHeader
) &&
63 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
64 be32_to_cpu(cow_header
->version
) >= QCOW_VERSION
)
72 * read qcow2 extension and fill bs
73 * start reading from start_offset
74 * finish reading upon magic of value 0 or when end_offset reached
75 * unknown magic is skipped (future extension this version knows nothing about)
76 * return 0 upon success, non-0 otherwise
78 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
81 BDRVQcowState
*s
= bs
->opaque
;
87 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
89 offset
= start_offset
;
90 while (offset
< end_offset
) {
94 if (offset
> s
->cluster_size
)
95 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
97 printf("attempting to read extended header in offset %lu\n", offset
);
100 if (bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
)) != sizeof(ext
)) {
101 fprintf(stderr
, "qcow2_read_extension: ERROR: "
102 "pread fail from offset %" PRIu64
"\n",
106 be32_to_cpus(&ext
.magic
);
107 be32_to_cpus(&ext
.len
);
108 offset
+= sizeof(ext
);
110 printf("ext.magic = 0x%x\n", ext
.magic
);
112 if (ext
.len
> end_offset
- offset
) {
113 error_report("Header extension too large");
118 case QCOW2_EXT_MAGIC_END
:
121 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
122 if (ext
.len
>= sizeof(bs
->backing_format
)) {
123 fprintf(stderr
, "ERROR: ext_backing_format: len=%u too large"
125 ext
.len
, sizeof(bs
->backing_format
));
128 if (bdrv_pread(bs
->file
, offset
, bs
->backing_format
,
131 bs
->backing_format
[ext
.len
] = '\0';
133 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
138 /* unknown magic - save it in case we need to rewrite the header */
140 Qcow2UnknownHeaderExtension
*uext
;
142 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
143 uext
->magic
= ext
.magic
;
145 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
147 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
155 offset
+= ((ext
.len
+ 7) & ~7);
161 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
163 BDRVQcowState
*s
= bs
->opaque
;
164 Qcow2UnknownHeaderExtension
*uext
, *next
;
166 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
167 QLIST_REMOVE(uext
, next
);
172 static int qcow2_open(BlockDriverState
*bs
, int flags
)
174 BDRVQcowState
*s
= bs
->opaque
;
180 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
184 be32_to_cpus(&header
.magic
);
185 be32_to_cpus(&header
.version
);
186 be64_to_cpus(&header
.backing_file_offset
);
187 be32_to_cpus(&header
.backing_file_size
);
188 be64_to_cpus(&header
.size
);
189 be32_to_cpus(&header
.cluster_bits
);
190 be32_to_cpus(&header
.crypt_method
);
191 be64_to_cpus(&header
.l1_table_offset
);
192 be32_to_cpus(&header
.l1_size
);
193 be64_to_cpus(&header
.refcount_table_offset
);
194 be32_to_cpus(&header
.refcount_table_clusters
);
195 be64_to_cpus(&header
.snapshots_offset
);
196 be32_to_cpus(&header
.nb_snapshots
);
198 if (header
.magic
!= QCOW_MAGIC
) {
202 if (header
.version
!= QCOW_VERSION
) {
204 snprintf(version
, sizeof(version
), "QCOW version %d", header
.version
);
205 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
206 bs
->device_name
, "qcow2", version
);
210 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
211 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
215 if (header
.crypt_method
> QCOW_CRYPT_AES
) {
219 s
->crypt_method_header
= header
.crypt_method
;
220 if (s
->crypt_method_header
) {
223 s
->cluster_bits
= header
.cluster_bits
;
224 s
->cluster_size
= 1 << s
->cluster_bits
;
225 s
->cluster_sectors
= 1 << (s
->cluster_bits
- 9);
226 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
227 s
->l2_size
= 1 << s
->l2_bits
;
228 bs
->total_sectors
= header
.size
/ 512;
229 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
230 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
231 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
232 s
->refcount_table_offset
= header
.refcount_table_offset
;
233 s
->refcount_table_size
=
234 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
236 s
->snapshots_offset
= header
.snapshots_offset
;
237 s
->nb_snapshots
= header
.nb_snapshots
;
239 /* read the level 1 table */
240 s
->l1_size
= header
.l1_size
;
241 s
->l1_vm_state_index
= size_to_l1(s
, header
.size
);
242 /* the L1 table must contain at least enough entries to put
244 if (s
->l1_size
< s
->l1_vm_state_index
) {
248 s
->l1_table_offset
= header
.l1_table_offset
;
249 if (s
->l1_size
> 0) {
250 s
->l1_table
= g_malloc0(
251 align_offset(s
->l1_size
* sizeof(uint64_t), 512));
252 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
253 s
->l1_size
* sizeof(uint64_t));
257 for(i
= 0;i
< s
->l1_size
; i
++) {
258 be64_to_cpus(&s
->l1_table
[i
]);
262 /* alloc L2 table/refcount block cache */
263 writethrough
= ((flags
& BDRV_O_CACHE_WB
) == 0);
264 s
->l2_table_cache
= qcow2_cache_create(bs
, L2_CACHE_SIZE
, writethrough
);
265 s
->refcount_block_cache
= qcow2_cache_create(bs
, REFCOUNT_CACHE_SIZE
,
268 s
->cluster_cache
= g_malloc(s
->cluster_size
);
269 /* one more sector for decompressed data alignment */
270 s
->cluster_data
= qemu_blockalign(bs
, QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
272 s
->cluster_cache_offset
= -1;
275 ret
= qcow2_refcount_init(bs
);
280 QLIST_INIT(&s
->cluster_allocs
);
282 /* read qcow2 extensions */
283 if (header
.backing_file_offset
) {
284 ext_end
= header
.backing_file_offset
;
286 ext_end
= s
->cluster_size
;
288 if (qcow2_read_extensions(bs
, sizeof(header
), ext_end
)) {
293 /* read the backing file name */
294 if (header
.backing_file_offset
!= 0) {
295 len
= header
.backing_file_size
;
299 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
300 bs
->backing_file
, len
);
304 bs
->backing_file
[len
] = '\0';
307 ret
= qcow2_read_snapshots(bs
);
312 /* Initialise locks */
313 qemu_co_mutex_init(&s
->lock
);
317 BdrvCheckResult result
= {0};
318 qcow2_check_refcounts(bs
, &result
);
324 cleanup_unknown_header_ext(bs
);
325 qcow2_free_snapshots(bs
);
326 qcow2_refcount_close(bs
);
328 if (s
->l2_table_cache
) {
329 qcow2_cache_destroy(bs
, s
->l2_table_cache
);
331 g_free(s
->cluster_cache
);
332 qemu_vfree(s
->cluster_data
);
336 static int qcow2_set_key(BlockDriverState
*bs
, const char *key
)
338 BDRVQcowState
*s
= bs
->opaque
;
342 memset(keybuf
, 0, 16);
346 /* XXX: we could compress the chars to 7 bits to increase
348 for(i
= 0;i
< len
;i
++) {
351 s
->crypt_method
= s
->crypt_method_header
;
353 if (AES_set_encrypt_key(keybuf
, 128, &s
->aes_encrypt_key
) != 0)
355 if (AES_set_decrypt_key(keybuf
, 128, &s
->aes_decrypt_key
) != 0)
365 AES_encrypt(in
, tmp
, &s
->aes_encrypt_key
);
366 AES_decrypt(tmp
, out
, &s
->aes_decrypt_key
);
367 for(i
= 0; i
< 16; i
++)
368 printf(" %02x", tmp
[i
]);
370 for(i
= 0; i
< 16; i
++)
371 printf(" %02x", out
[i
]);
378 static int coroutine_fn
qcow2_co_is_allocated(BlockDriverState
*bs
,
379 int64_t sector_num
, int nb_sectors
, int *pnum
)
381 BDRVQcowState
*s
= bs
->opaque
;
382 uint64_t cluster_offset
;
386 /* FIXME We can get errors here, but the bdrv_co_is_allocated interface
387 * can't pass them on today */
388 qemu_co_mutex_lock(&s
->lock
);
389 ret
= qcow2_get_cluster_offset(bs
, sector_num
<< 9, pnum
, &cluster_offset
);
390 qemu_co_mutex_unlock(&s
->lock
);
395 return (cluster_offset
!= 0);
398 /* handle reading after the end of the backing file */
399 int qcow2_backing_read1(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
400 int64_t sector_num
, int nb_sectors
)
403 if ((sector_num
+ nb_sectors
) <= bs
->total_sectors
)
405 if (sector_num
>= bs
->total_sectors
)
408 n1
= bs
->total_sectors
- sector_num
;
410 qemu_iovec_memset_skip(qiov
, 0, 512 * (nb_sectors
- n1
), 512 * n1
);
415 static coroutine_fn
int qcow2_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
416 int remaining_sectors
, QEMUIOVector
*qiov
)
418 BDRVQcowState
*s
= bs
->opaque
;
419 int index_in_cluster
, n1
;
421 int cur_nr_sectors
; /* number of sectors in current iteration */
422 uint64_t cluster_offset
= 0;
423 uint64_t bytes_done
= 0;
424 QEMUIOVector hd_qiov
;
425 uint8_t *cluster_data
= NULL
;
427 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
429 qemu_co_mutex_lock(&s
->lock
);
431 while (remaining_sectors
!= 0) {
433 /* prepare next request */
434 cur_nr_sectors
= remaining_sectors
;
435 if (s
->crypt_method
) {
436 cur_nr_sectors
= MIN(cur_nr_sectors
,
437 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
);
440 ret
= qcow2_get_cluster_offset(bs
, sector_num
<< 9,
441 &cur_nr_sectors
, &cluster_offset
);
446 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
448 qemu_iovec_reset(&hd_qiov
);
449 qemu_iovec_copy(&hd_qiov
, qiov
, bytes_done
,
450 cur_nr_sectors
* 512);
453 case QCOW2_CLUSTER_UNALLOCATED
:
455 if (bs
->backing_hd
) {
456 /* read from the base image */
457 n1
= qcow2_backing_read1(bs
->backing_hd
, &hd_qiov
,
458 sector_num
, cur_nr_sectors
);
460 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
461 qemu_co_mutex_unlock(&s
->lock
);
462 ret
= bdrv_co_readv(bs
->backing_hd
, sector_num
,
464 qemu_co_mutex_lock(&s
->lock
);
470 /* Note: in this case, no need to wait */
471 qemu_iovec_memset(&hd_qiov
, 0, 512 * cur_nr_sectors
);
475 case QCOW2_CLUSTER_COMPRESSED
:
476 /* add AIO support for compressed blocks ? */
477 ret
= qcow2_decompress_cluster(bs
, cluster_offset
);
482 qemu_iovec_from_buffer(&hd_qiov
,
483 s
->cluster_cache
+ index_in_cluster
* 512,
484 512 * cur_nr_sectors
);
487 case QCOW2_CLUSTER_NORMAL
:
488 if ((cluster_offset
& 511) != 0) {
493 if (s
->crypt_method
) {
495 * For encrypted images, read everything into a temporary
496 * contiguous buffer on which the AES functions can work.
500 qemu_blockalign(bs
, QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
503 assert(cur_nr_sectors
<=
504 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
);
505 qemu_iovec_reset(&hd_qiov
);
506 qemu_iovec_add(&hd_qiov
, cluster_data
,
507 512 * cur_nr_sectors
);
510 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
511 qemu_co_mutex_unlock(&s
->lock
);
512 ret
= bdrv_co_readv(bs
->file
,
513 (cluster_offset
>> 9) + index_in_cluster
,
514 cur_nr_sectors
, &hd_qiov
);
515 qemu_co_mutex_lock(&s
->lock
);
519 if (s
->crypt_method
) {
520 qcow2_encrypt_sectors(s
, sector_num
, cluster_data
,
521 cluster_data
, cur_nr_sectors
, 0, &s
->aes_decrypt_key
);
522 qemu_iovec_reset(&hd_qiov
);
523 qemu_iovec_copy(&hd_qiov
, qiov
, bytes_done
,
524 cur_nr_sectors
* 512);
525 qemu_iovec_from_buffer(&hd_qiov
, cluster_data
,
526 512 * cur_nr_sectors
);
531 g_assert_not_reached();
536 remaining_sectors
-= cur_nr_sectors
;
537 sector_num
+= cur_nr_sectors
;
538 bytes_done
+= cur_nr_sectors
* 512;
543 qemu_co_mutex_unlock(&s
->lock
);
545 qemu_iovec_destroy(&hd_qiov
);
546 qemu_vfree(cluster_data
);
551 static void run_dependent_requests(BDRVQcowState
*s
, QCowL2Meta
*m
)
553 /* Take the request off the list of running requests */
554 if (m
->nb_clusters
!= 0) {
555 QLIST_REMOVE(m
, next_in_flight
);
558 /* Restart all dependent requests */
559 if (!qemu_co_queue_empty(&m
->dependent_requests
)) {
560 qemu_co_mutex_unlock(&s
->lock
);
561 qemu_co_queue_restart_all(&m
->dependent_requests
);
562 qemu_co_mutex_lock(&s
->lock
);
566 static coroutine_fn
int qcow2_co_writev(BlockDriverState
*bs
,
568 int remaining_sectors
,
571 BDRVQcowState
*s
= bs
->opaque
;
572 int index_in_cluster
;
575 int cur_nr_sectors
; /* number of sectors in current iteration */
576 uint64_t cluster_offset
;
577 QEMUIOVector hd_qiov
;
578 uint64_t bytes_done
= 0;
579 uint8_t *cluster_data
= NULL
;
580 QCowL2Meta l2meta
= {
584 trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num
,
587 qemu_co_queue_init(&l2meta
.dependent_requests
);
589 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
591 s
->cluster_cache_offset
= -1; /* disable compressed cache */
593 qemu_co_mutex_lock(&s
->lock
);
595 while (remaining_sectors
!= 0) {
597 trace_qcow2_writev_start_part(qemu_coroutine_self());
598 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
599 n_end
= index_in_cluster
+ remaining_sectors
;
600 if (s
->crypt_method
&&
601 n_end
> QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
) {
602 n_end
= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
;
605 ret
= qcow2_alloc_cluster_offset(bs
, sector_num
<< 9,
606 index_in_cluster
, n_end
, &cur_nr_sectors
, &l2meta
);
611 cluster_offset
= l2meta
.cluster_offset
;
612 assert((cluster_offset
& 511) == 0);
614 qemu_iovec_reset(&hd_qiov
);
615 qemu_iovec_copy(&hd_qiov
, qiov
, bytes_done
,
616 cur_nr_sectors
* 512);
618 if (s
->crypt_method
) {
620 cluster_data
= qemu_blockalign(bs
, QCOW_MAX_CRYPT_CLUSTERS
*
624 assert(hd_qiov
.size
<=
625 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
626 qemu_iovec_to_buffer(&hd_qiov
, cluster_data
);
628 qcow2_encrypt_sectors(s
, sector_num
, cluster_data
,
629 cluster_data
, cur_nr_sectors
, 1, &s
->aes_encrypt_key
);
631 qemu_iovec_reset(&hd_qiov
);
632 qemu_iovec_add(&hd_qiov
, cluster_data
,
633 cur_nr_sectors
* 512);
636 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
637 qemu_co_mutex_unlock(&s
->lock
);
638 trace_qcow2_writev_data(qemu_coroutine_self(),
639 (cluster_offset
>> 9) + index_in_cluster
);
640 ret
= bdrv_co_writev(bs
->file
,
641 (cluster_offset
>> 9) + index_in_cluster
,
642 cur_nr_sectors
, &hd_qiov
);
643 qemu_co_mutex_lock(&s
->lock
);
648 ret
= qcow2_alloc_cluster_link_l2(bs
, &l2meta
);
653 run_dependent_requests(s
, &l2meta
);
655 remaining_sectors
-= cur_nr_sectors
;
656 sector_num
+= cur_nr_sectors
;
657 bytes_done
+= cur_nr_sectors
* 512;
658 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors
);
663 run_dependent_requests(s
, &l2meta
);
665 qemu_co_mutex_unlock(&s
->lock
);
667 qemu_iovec_destroy(&hd_qiov
);
668 qemu_vfree(cluster_data
);
669 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
674 static void qcow2_close(BlockDriverState
*bs
)
676 BDRVQcowState
*s
= bs
->opaque
;
679 qcow2_cache_flush(bs
, s
->l2_table_cache
);
680 qcow2_cache_flush(bs
, s
->refcount_block_cache
);
682 qcow2_cache_destroy(bs
, s
->l2_table_cache
);
683 qcow2_cache_destroy(bs
, s
->refcount_block_cache
);
685 cleanup_unknown_header_ext(bs
);
686 g_free(s
->cluster_cache
);
687 qemu_vfree(s
->cluster_data
);
688 qcow2_refcount_close(bs
);
689 qcow2_free_snapshots(bs
);
692 static void qcow2_invalidate_cache(BlockDriverState
*bs
)
694 BDRVQcowState
*s
= bs
->opaque
;
695 int flags
= s
->flags
;
696 AES_KEY aes_encrypt_key
;
697 AES_KEY aes_decrypt_key
;
698 uint32_t crypt_method
= 0;
701 * Backing files are read-only which makes all of their metadata immutable,
702 * that means we don't have to worry about reopening them here.
705 if (s
->crypt_method
) {
706 crypt_method
= s
->crypt_method
;
707 memcpy(&aes_encrypt_key
, &s
->aes_encrypt_key
, sizeof(aes_encrypt_key
));
708 memcpy(&aes_decrypt_key
, &s
->aes_decrypt_key
, sizeof(aes_decrypt_key
));
713 memset(s
, 0, sizeof(BDRVQcowState
));
714 qcow2_open(bs
, flags
);
717 s
->crypt_method
= crypt_method
;
718 memcpy(&s
->aes_encrypt_key
, &aes_encrypt_key
, sizeof(aes_encrypt_key
));
719 memcpy(&s
->aes_decrypt_key
, &aes_decrypt_key
, sizeof(aes_decrypt_key
));
723 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
724 size_t len
, size_t buflen
)
726 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
727 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
729 if (buflen
< ext_len
) {
733 *ext_backing_fmt
= (QCowExtension
) {
734 .magic
= cpu_to_be32(magic
),
735 .len
= cpu_to_be32(len
),
737 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
743 * Updates the qcow2 header, including the variable length parts of it, i.e.
744 * the backing file name and all extensions. qcow2 was not designed to allow
745 * such changes, so if we run out of space (we can only use the first cluster)
746 * this function may fail.
748 * Returns 0 on success, -errno in error cases.
750 int qcow2_update_header(BlockDriverState
*bs
)
752 BDRVQcowState
*s
= bs
->opaque
;
755 size_t buflen
= s
->cluster_size
;
758 uint32_t refcount_table_clusters
;
759 Qcow2UnknownHeaderExtension
*uext
;
761 buf
= qemu_blockalign(bs
, buflen
);
762 memset(buf
, 0, s
->cluster_size
);
764 /* Header structure */
765 header
= (QCowHeader
*) buf
;
767 if (buflen
< sizeof(*header
)) {
772 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
773 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
775 *header
= (QCowHeader
) {
776 .magic
= cpu_to_be32(QCOW_MAGIC
),
777 .version
= cpu_to_be32(QCOW_VERSION
),
778 .backing_file_offset
= 0,
779 .backing_file_size
= 0,
780 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
781 .size
= cpu_to_be64(total_size
),
782 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
783 .l1_size
= cpu_to_be32(s
->l1_size
),
784 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
785 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
786 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
787 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
788 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
791 buf
+= sizeof(*header
);
792 buflen
-= sizeof(*header
);
794 /* Backing file format header extension */
795 if (*bs
->backing_format
) {
796 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
797 bs
->backing_format
, strlen(bs
->backing_format
),
807 /* Keep unknown header extensions */
808 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
809 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
818 /* End of header extensions */
819 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
827 /* Backing file name */
828 if (*bs
->backing_file
) {
829 size_t backing_file_len
= strlen(bs
->backing_file
);
831 if (buflen
< backing_file_len
) {
836 strncpy(buf
, bs
->backing_file
, buflen
);
838 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
839 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
842 /* Write the new header */
843 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
854 static int qcow2_change_backing_file(BlockDriverState
*bs
,
855 const char *backing_file
, const char *backing_fmt
)
857 /* Backing file format doesn't make sense without a backing file */
858 if (backing_fmt
&& !backing_file
) {
862 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
863 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
865 return qcow2_update_header(bs
);
868 static int preallocate(BlockDriverState
*bs
)
876 nb_sectors
= bdrv_getlength(bs
) >> 9;
878 qemu_co_queue_init(&meta
.dependent_requests
);
879 meta
.cluster_offset
= 0;
882 num
= MIN(nb_sectors
, INT_MAX
>> 9);
883 ret
= qcow2_alloc_cluster_offset(bs
, offset
, 0, num
, &num
, &meta
);
888 ret
= qcow2_alloc_cluster_link_l2(bs
, &meta
);
890 qcow2_free_any_clusters(bs
, meta
.cluster_offset
, meta
.nb_clusters
);
894 /* There are no dependent requests, but we need to remove our request
895 * from the list of in-flight requests */
896 run_dependent_requests(bs
->opaque
, &meta
);
898 /* TODO Preallocate data if requested */
905 * It is expected that the image file is large enough to actually contain
906 * all of the allocated clusters (otherwise we get failing reads after
907 * EOF). Extend the image to the last allocated sector.
909 if (meta
.cluster_offset
!= 0) {
912 ret
= bdrv_write(bs
->file
, (meta
.cluster_offset
>> 9) + num
- 1, buf
, 1);
921 static int qcow2_create2(const char *filename
, int64_t total_size
,
922 const char *backing_file
, const char *backing_format
,
923 int flags
, size_t cluster_size
, int prealloc
,
924 QEMUOptionParameter
*options
)
926 /* Calculate cluster_bits */
928 cluster_bits
= ffs(cluster_size
) - 1;
929 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
930 (1 << cluster_bits
) != cluster_size
)
933 "Cluster size must be a power of two between %d and %dk",
934 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
939 * Open the image file and write a minimal qcow2 header.
941 * We keep things simple and start with a zero-sized image. We also
942 * do without refcount blocks or a L1 table for now. We'll fix the
943 * inconsistency later.
945 * We do need a refcount table because growing the refcount table means
946 * allocating two new refcount blocks - the seconds of which would be at
947 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
948 * size for any qcow2 image.
950 BlockDriverState
* bs
;
952 uint8_t* refcount_table
;
955 ret
= bdrv_create_file(filename
, options
);
960 ret
= bdrv_file_open(&bs
, filename
, BDRV_O_RDWR
);
965 /* Write the header */
966 memset(&header
, 0, sizeof(header
));
967 header
.magic
= cpu_to_be32(QCOW_MAGIC
);
968 header
.version
= cpu_to_be32(QCOW_VERSION
);
969 header
.cluster_bits
= cpu_to_be32(cluster_bits
);
970 header
.size
= cpu_to_be64(0);
971 header
.l1_table_offset
= cpu_to_be64(0);
972 header
.l1_size
= cpu_to_be32(0);
973 header
.refcount_table_offset
= cpu_to_be64(cluster_size
);
974 header
.refcount_table_clusters
= cpu_to_be32(1);
976 if (flags
& BLOCK_FLAG_ENCRYPT
) {
977 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
979 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
982 ret
= bdrv_pwrite(bs
, 0, &header
, sizeof(header
));
987 /* Write an empty refcount table */
988 refcount_table
= g_malloc0(cluster_size
);
989 ret
= bdrv_pwrite(bs
, cluster_size
, refcount_table
, cluster_size
);
990 g_free(refcount_table
);
999 * And now open the image and make it consistent first (i.e. increase the
1000 * refcount of the cluster that is occupied by the header and the refcount
1003 BlockDriver
* drv
= bdrv_find_format("qcow2");
1004 assert(drv
!= NULL
);
1005 ret
= bdrv_open(bs
, filename
,
1006 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_NO_FLUSH
, drv
);
1011 ret
= qcow2_alloc_clusters(bs
, 2 * cluster_size
);
1015 } else if (ret
!= 0) {
1016 error_report("Huh, first cluster in empty image is already in use?");
1020 /* Okay, now that we have a valid image, let's give it the right size */
1021 ret
= bdrv_truncate(bs
, total_size
* BDRV_SECTOR_SIZE
);
1026 /* Want a backing file? There you go.*/
1028 ret
= bdrv_change_backing_file(bs
, backing_file
, backing_format
);
1034 /* And if we're supposed to preallocate metadata, do that now */
1036 ret
= preallocate(bs
);
1048 static int qcow2_create(const char *filename
, QEMUOptionParameter
*options
)
1050 const char *backing_file
= NULL
;
1051 const char *backing_fmt
= NULL
;
1052 uint64_t sectors
= 0;
1054 size_t cluster_size
= DEFAULT_CLUSTER_SIZE
;
1057 /* Read out options */
1058 while (options
&& options
->name
) {
1059 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
1060 sectors
= options
->value
.n
/ 512;
1061 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
1062 backing_file
= options
->value
.s
;
1063 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FMT
)) {
1064 backing_fmt
= options
->value
.s
;
1065 } else if (!strcmp(options
->name
, BLOCK_OPT_ENCRYPT
)) {
1066 flags
|= options
->value
.n
? BLOCK_FLAG_ENCRYPT
: 0;
1067 } else if (!strcmp(options
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
1068 if (options
->value
.n
) {
1069 cluster_size
= options
->value
.n
;
1071 } else if (!strcmp(options
->name
, BLOCK_OPT_PREALLOC
)) {
1072 if (!options
->value
.s
|| !strcmp(options
->value
.s
, "off")) {
1074 } else if (!strcmp(options
->value
.s
, "metadata")) {
1077 fprintf(stderr
, "Invalid preallocation mode: '%s'\n",
1085 if (backing_file
&& prealloc
) {
1086 fprintf(stderr
, "Backing file and preallocation cannot be used at "
1091 return qcow2_create2(filename
, sectors
, backing_file
, backing_fmt
, flags
,
1092 cluster_size
, prealloc
, options
);
1095 static int qcow2_make_empty(BlockDriverState
*bs
)
1098 /* XXX: not correct */
1099 BDRVQcowState
*s
= bs
->opaque
;
1100 uint32_t l1_length
= s
->l1_size
* sizeof(uint64_t);
1103 memset(s
->l1_table
, 0, l1_length
);
1104 if (bdrv_pwrite(bs
->file
, s
->l1_table_offset
, s
->l1_table
, l1_length
) < 0)
1106 ret
= bdrv_truncate(bs
->file
, s
->l1_table_offset
+ l1_length
);
1115 static coroutine_fn
int qcow2_co_discard(BlockDriverState
*bs
,
1116 int64_t sector_num
, int nb_sectors
)
1119 BDRVQcowState
*s
= bs
->opaque
;
1121 qemu_co_mutex_lock(&s
->lock
);
1122 ret
= qcow2_discard_clusters(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1124 qemu_co_mutex_unlock(&s
->lock
);
1128 static int qcow2_truncate(BlockDriverState
*bs
, int64_t offset
)
1130 BDRVQcowState
*s
= bs
->opaque
;
1131 int ret
, new_l1_size
;
1134 error_report("The new size must be a multiple of 512");
1138 /* cannot proceed if image has snapshots */
1139 if (s
->nb_snapshots
) {
1140 error_report("Can't resize an image which has snapshots");
1144 /* shrinking is currently not supported */
1145 if (offset
< bs
->total_sectors
* 512) {
1146 error_report("qcow2 doesn't support shrinking images yet");
1150 new_l1_size
= size_to_l1(s
, offset
);
1151 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
1156 /* write updated header.size */
1157 offset
= cpu_to_be64(offset
);
1158 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
1159 &offset
, sizeof(uint64_t));
1164 s
->l1_vm_state_index
= new_l1_size
;
1168 /* XXX: put compressed sectors first, then all the cluster aligned
1169 tables to avoid losing bytes in alignment */
1170 static int qcow2_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1171 const uint8_t *buf
, int nb_sectors
)
1173 BDRVQcowState
*s
= bs
->opaque
;
1177 uint64_t cluster_offset
;
1179 if (nb_sectors
== 0) {
1180 /* align end of file to a sector boundary to ease reading with
1181 sector based I/Os */
1182 cluster_offset
= bdrv_getlength(bs
->file
);
1183 cluster_offset
= (cluster_offset
+ 511) & ~511;
1184 bdrv_truncate(bs
->file
, cluster_offset
);
1188 if (nb_sectors
!= s
->cluster_sectors
)
1191 out_buf
= g_malloc(s
->cluster_size
+ (s
->cluster_size
/ 1000) + 128);
1193 /* best compression, small window, no zlib header */
1194 memset(&strm
, 0, sizeof(strm
));
1195 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
1197 9, Z_DEFAULT_STRATEGY
);
1203 strm
.avail_in
= s
->cluster_size
;
1204 strm
.next_in
= (uint8_t *)buf
;
1205 strm
.avail_out
= s
->cluster_size
;
1206 strm
.next_out
= out_buf
;
1208 ret
= deflate(&strm
, Z_FINISH
);
1209 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
1214 out_len
= strm
.next_out
- out_buf
;
1218 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
1219 /* could not compress: write normal cluster */
1220 ret
= bdrv_write(bs
, sector_num
, buf
, s
->cluster_sectors
);
1225 cluster_offset
= qcow2_alloc_compressed_cluster_offset(bs
,
1226 sector_num
<< 9, out_len
);
1227 if (!cluster_offset
) {
1231 cluster_offset
&= s
->cluster_offset_mask
;
1232 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
1233 ret
= bdrv_pwrite(bs
->file
, cluster_offset
, out_buf
, out_len
);
1245 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
1247 BDRVQcowState
*s
= bs
->opaque
;
1250 qemu_co_mutex_lock(&s
->lock
);
1251 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1253 qemu_co_mutex_unlock(&s
->lock
);
1257 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1259 qemu_co_mutex_unlock(&s
->lock
);
1262 qemu_co_mutex_unlock(&s
->lock
);
1267 static int64_t qcow2_vm_state_offset(BDRVQcowState
*s
)
1269 return (int64_t)s
->l1_vm_state_index
<< (s
->cluster_bits
+ s
->l2_bits
);
1272 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1274 BDRVQcowState
*s
= bs
->opaque
;
1275 bdi
->cluster_size
= s
->cluster_size
;
1276 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
1281 static int qcow2_check(BlockDriverState
*bs
, BdrvCheckResult
*result
)
1283 return qcow2_check_refcounts(bs
, result
);
1287 static void dump_refcounts(BlockDriverState
*bs
)
1289 BDRVQcowState
*s
= bs
->opaque
;
1290 int64_t nb_clusters
, k
, k1
, size
;
1293 size
= bdrv_getlength(bs
->file
);
1294 nb_clusters
= size_to_clusters(s
, size
);
1295 for(k
= 0; k
< nb_clusters
;) {
1297 refcount
= get_refcount(bs
, k
);
1299 while (k
< nb_clusters
&& get_refcount(bs
, k
) == refcount
)
1301 printf("%" PRId64
": refcount=%d nb=%" PRId64
"\n", k
, refcount
,
1307 static int qcow2_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1308 int64_t pos
, int size
)
1310 BDRVQcowState
*s
= bs
->opaque
;
1311 int growable
= bs
->growable
;
1314 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
1316 ret
= bdrv_pwrite(bs
, qcow2_vm_state_offset(s
) + pos
, buf
, size
);
1317 bs
->growable
= growable
;
1322 static int qcow2_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1323 int64_t pos
, int size
)
1325 BDRVQcowState
*s
= bs
->opaque
;
1326 int growable
= bs
->growable
;
1329 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
1331 ret
= bdrv_pread(bs
, qcow2_vm_state_offset(s
) + pos
, buf
, size
);
1332 bs
->growable
= growable
;
1337 static QEMUOptionParameter qcow2_create_options
[] = {
1339 .name
= BLOCK_OPT_SIZE
,
1341 .help
= "Virtual disk size"
1344 .name
= BLOCK_OPT_BACKING_FILE
,
1346 .help
= "File name of a base image"
1349 .name
= BLOCK_OPT_BACKING_FMT
,
1351 .help
= "Image format of the base image"
1354 .name
= BLOCK_OPT_ENCRYPT
,
1356 .help
= "Encrypt the image"
1359 .name
= BLOCK_OPT_CLUSTER_SIZE
,
1361 .help
= "qcow2 cluster size",
1362 .value
= { .n
= DEFAULT_CLUSTER_SIZE
},
1365 .name
= BLOCK_OPT_PREALLOC
,
1367 .help
= "Preallocation mode (allowed values: off, metadata)"
1372 static BlockDriver bdrv_qcow2
= {
1373 .format_name
= "qcow2",
1374 .instance_size
= sizeof(BDRVQcowState
),
1375 .bdrv_probe
= qcow2_probe
,
1376 .bdrv_open
= qcow2_open
,
1377 .bdrv_close
= qcow2_close
,
1378 .bdrv_create
= qcow2_create
,
1379 .bdrv_co_is_allocated
= qcow2_co_is_allocated
,
1380 .bdrv_set_key
= qcow2_set_key
,
1381 .bdrv_make_empty
= qcow2_make_empty
,
1383 .bdrv_co_readv
= qcow2_co_readv
,
1384 .bdrv_co_writev
= qcow2_co_writev
,
1385 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
1387 .bdrv_co_discard
= qcow2_co_discard
,
1388 .bdrv_truncate
= qcow2_truncate
,
1389 .bdrv_write_compressed
= qcow2_write_compressed
,
1391 .bdrv_snapshot_create
= qcow2_snapshot_create
,
1392 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
1393 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
1394 .bdrv_snapshot_list
= qcow2_snapshot_list
,
1395 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
1396 .bdrv_get_info
= qcow2_get_info
,
1398 .bdrv_save_vmstate
= qcow2_save_vmstate
,
1399 .bdrv_load_vmstate
= qcow2_load_vmstate
,
1401 .bdrv_change_backing_file
= qcow2_change_backing_file
,
1403 .bdrv_invalidate_cache
= qcow2_invalidate_cache
,
1405 .create_options
= qcow2_create_options
,
1406 .bdrv_check
= qcow2_check
,
1409 static void bdrv_qcow2_init(void)
1411 bdrv_register(&bdrv_qcow2
);
1414 block_init(bdrv_qcow2_init
);