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"
34 Differences with QCOW:
36 - Support for multiple incremental snapshots.
37 - Memory management by reference counts.
38 - Clusters which have a reference count of one have the bit
39 QCOW_OFLAG_COPIED to optimize write performance.
40 - Size of compressed clusters is stored in sectors to reduce bit usage
41 in the cluster offsets.
42 - Support for storing additional data (such as the VM state) in the
44 - If a backing store is used, the cluster size is not constrained
45 (could be backported to QCOW).
46 - L2 tables have always a size of one cluster.
54 #define QCOW2_EXT_MAGIC_END 0
55 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
57 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
59 const QCowHeader
*cow_header
= (const void *)buf
;
61 if (buf_size
>= sizeof(QCowHeader
) &&
62 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
63 be32_to_cpu(cow_header
->version
) >= QCOW_VERSION
)
71 * read qcow2 extension and fill bs
72 * start reading from start_offset
73 * finish reading upon magic of value 0 or when end_offset reached
74 * unknown magic is skipped (future extension this version knows nothing about)
75 * return 0 upon success, non-0 otherwise
77 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
84 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
86 offset
= start_offset
;
87 while (offset
< end_offset
) {
90 BDRVQcowState
*s
= bs
->opaque
;
92 if (offset
> s
->cluster_size
)
93 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
95 printf("attemting to read extended header in offset %lu\n", offset
);
98 if (bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
)) != sizeof(ext
)) {
99 fprintf(stderr
, "qcow2_read_extension: ERROR: "
100 "pread fail from offset %" PRIu64
"\n",
104 be32_to_cpus(&ext
.magic
);
105 be32_to_cpus(&ext
.len
);
106 offset
+= sizeof(ext
);
108 printf("ext.magic = 0x%x\n", ext
.magic
);
111 case QCOW2_EXT_MAGIC_END
:
114 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
115 if (ext
.len
>= sizeof(bs
->backing_format
)) {
116 fprintf(stderr
, "ERROR: ext_backing_format: len=%u too large"
118 ext
.len
, sizeof(bs
->backing_format
));
121 if (bdrv_pread(bs
->file
, offset
, bs
->backing_format
,
124 bs
->backing_format
[ext
.len
] = '\0';
126 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
128 offset
= ((offset
+ ext
.len
+ 7) & ~7);
132 /* unknown magic -- just skip it */
133 offset
= ((offset
+ ext
.len
+ 7) & ~7);
142 static int qcow2_open(BlockDriverState
*bs
, int flags
)
144 BDRVQcowState
*s
= bs
->opaque
;
150 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
154 be32_to_cpus(&header
.magic
);
155 be32_to_cpus(&header
.version
);
156 be64_to_cpus(&header
.backing_file_offset
);
157 be32_to_cpus(&header
.backing_file_size
);
158 be64_to_cpus(&header
.size
);
159 be32_to_cpus(&header
.cluster_bits
);
160 be32_to_cpus(&header
.crypt_method
);
161 be64_to_cpus(&header
.l1_table_offset
);
162 be32_to_cpus(&header
.l1_size
);
163 be64_to_cpus(&header
.refcount_table_offset
);
164 be32_to_cpus(&header
.refcount_table_clusters
);
165 be64_to_cpus(&header
.snapshots_offset
);
166 be32_to_cpus(&header
.nb_snapshots
);
168 if (header
.magic
!= QCOW_MAGIC
) {
172 if (header
.version
!= QCOW_VERSION
) {
174 snprintf(version
, sizeof(version
), "QCOW version %d", header
.version
);
175 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
176 bs
->device_name
, "qcow2", version
);
180 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
181 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
185 if (header
.crypt_method
> QCOW_CRYPT_AES
) {
189 s
->crypt_method_header
= header
.crypt_method
;
190 if (s
->crypt_method_header
) {
193 s
->cluster_bits
= header
.cluster_bits
;
194 s
->cluster_size
= 1 << s
->cluster_bits
;
195 s
->cluster_sectors
= 1 << (s
->cluster_bits
- 9);
196 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
197 s
->l2_size
= 1 << s
->l2_bits
;
198 bs
->total_sectors
= header
.size
/ 512;
199 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
200 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
201 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
202 s
->refcount_table_offset
= header
.refcount_table_offset
;
203 s
->refcount_table_size
=
204 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
206 s
->snapshots_offset
= header
.snapshots_offset
;
207 s
->nb_snapshots
= header
.nb_snapshots
;
209 /* read the level 1 table */
210 s
->l1_size
= header
.l1_size
;
211 s
->l1_vm_state_index
= size_to_l1(s
, header
.size
);
212 /* the L1 table must contain at least enough entries to put
214 if (s
->l1_size
< s
->l1_vm_state_index
) {
218 s
->l1_table_offset
= header
.l1_table_offset
;
219 if (s
->l1_size
> 0) {
220 s
->l1_table
= g_malloc0(
221 align_offset(s
->l1_size
* sizeof(uint64_t), 512));
222 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
223 s
->l1_size
* sizeof(uint64_t));
227 for(i
= 0;i
< s
->l1_size
; i
++) {
228 be64_to_cpus(&s
->l1_table
[i
]);
232 /* alloc L2 table/refcount block cache */
233 writethrough
= ((flags
& BDRV_O_CACHE_WB
) == 0);
234 s
->l2_table_cache
= qcow2_cache_create(bs
, L2_CACHE_SIZE
, writethrough
);
235 s
->refcount_block_cache
= qcow2_cache_create(bs
, REFCOUNT_CACHE_SIZE
,
238 s
->cluster_cache
= g_malloc(s
->cluster_size
);
239 /* one more sector for decompressed data alignment */
240 s
->cluster_data
= g_malloc(QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
242 s
->cluster_cache_offset
= -1;
244 ret
= qcow2_refcount_init(bs
);
249 QLIST_INIT(&s
->cluster_allocs
);
251 /* read qcow2 extensions */
252 if (header
.backing_file_offset
) {
253 ext_end
= header
.backing_file_offset
;
255 ext_end
= s
->cluster_size
;
257 if (qcow2_read_extensions(bs
, sizeof(header
), ext_end
)) {
262 /* read the backing file name */
263 if (header
.backing_file_offset
!= 0) {
264 len
= header
.backing_file_size
;
268 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
269 bs
->backing_file
, len
);
273 bs
->backing_file
[len
] = '\0';
275 if (qcow2_read_snapshots(bs
) < 0) {
280 /* Initialise locks */
281 qemu_co_mutex_init(&s
->lock
);
285 BdrvCheckResult result
= {0};
286 qcow2_check_refcounts(bs
, &result
);
292 qcow2_free_snapshots(bs
);
293 qcow2_refcount_close(bs
);
295 if (s
->l2_table_cache
) {
296 qcow2_cache_destroy(bs
, s
->l2_table_cache
);
298 g_free(s
->cluster_cache
);
299 g_free(s
->cluster_data
);
303 static int qcow2_set_key(BlockDriverState
*bs
, const char *key
)
305 BDRVQcowState
*s
= bs
->opaque
;
309 memset(keybuf
, 0, 16);
313 /* XXX: we could compress the chars to 7 bits to increase
315 for(i
= 0;i
< len
;i
++) {
318 s
->crypt_method
= s
->crypt_method_header
;
320 if (AES_set_encrypt_key(keybuf
, 128, &s
->aes_encrypt_key
) != 0)
322 if (AES_set_decrypt_key(keybuf
, 128, &s
->aes_decrypt_key
) != 0)
332 AES_encrypt(in
, tmp
, &s
->aes_encrypt_key
);
333 AES_decrypt(tmp
, out
, &s
->aes_decrypt_key
);
334 for(i
= 0; i
< 16; i
++)
335 printf(" %02x", tmp
[i
]);
337 for(i
= 0; i
< 16; i
++)
338 printf(" %02x", out
[i
]);
345 static int qcow2_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
346 int nb_sectors
, int *pnum
)
348 uint64_t cluster_offset
;
352 /* FIXME We can get errors here, but the bdrv_is_allocated interface can't
353 * pass them on today */
354 ret
= qcow2_get_cluster_offset(bs
, sector_num
<< 9, pnum
, &cluster_offset
);
359 return (cluster_offset
!= 0);
362 /* handle reading after the end of the backing file */
363 int qcow2_backing_read1(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
364 int64_t sector_num
, int nb_sectors
)
367 if ((sector_num
+ nb_sectors
) <= bs
->total_sectors
)
369 if (sector_num
>= bs
->total_sectors
)
372 n1
= bs
->total_sectors
- sector_num
;
374 qemu_iovec_memset_skip(qiov
, 0, 512 * (nb_sectors
- n1
), 512 * n1
);
379 typedef struct QCowAIOCB
{
380 BlockDriverAIOCB common
;
383 int remaining_sectors
;
385 uint8_t *cluster_data
;
386 QEMUIOVector hd_qiov
;
390 * Returns 0 when the request is completed successfully, 1 when there is still
391 * a part left to do and -errno in error cases.
393 static int qcow2_aio_read_cb(QCowAIOCB
*acb
)
395 BlockDriverState
*bs
= acb
->common
.bs
;
396 BDRVQcowState
*s
= bs
->opaque
;
397 int index_in_cluster
, n1
;
399 int cur_nr_sectors
; /* number of sectors in current iteration */
400 uint64_t cluster_offset
= 0;
402 if (acb
->remaining_sectors
== 0) {
403 /* request completed */
407 /* prepare next request */
408 cur_nr_sectors
= acb
->remaining_sectors
;
409 if (s
->crypt_method
) {
410 cur_nr_sectors
= MIN(cur_nr_sectors
,
411 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
);
414 ret
= qcow2_get_cluster_offset(bs
, acb
->sector_num
<< 9,
415 &cur_nr_sectors
, &cluster_offset
);
420 index_in_cluster
= acb
->sector_num
& (s
->cluster_sectors
- 1);
422 qemu_iovec_reset(&acb
->hd_qiov
);
423 qemu_iovec_copy(&acb
->hd_qiov
, acb
->qiov
, acb
->bytes_done
,
424 cur_nr_sectors
* 512);
426 if (!cluster_offset
) {
428 if (bs
->backing_hd
) {
429 /* read from the base image */
430 n1
= qcow2_backing_read1(bs
->backing_hd
, &acb
->hd_qiov
,
431 acb
->sector_num
, cur_nr_sectors
);
433 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
434 qemu_co_mutex_unlock(&s
->lock
);
435 ret
= bdrv_co_readv(bs
->backing_hd
, acb
->sector_num
,
437 qemu_co_mutex_lock(&s
->lock
);
443 /* Note: in this case, no need to wait */
444 qemu_iovec_memset(&acb
->hd_qiov
, 0, 512 * cur_nr_sectors
);
446 } else if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
447 /* add AIO support for compressed blocks ? */
448 ret
= qcow2_decompress_cluster(bs
, cluster_offset
);
453 qemu_iovec_from_buffer(&acb
->hd_qiov
,
454 s
->cluster_cache
+ index_in_cluster
* 512,
455 512 * cur_nr_sectors
);
457 if ((cluster_offset
& 511) != 0) {
461 if (s
->crypt_method
) {
463 * For encrypted images, read everything into a temporary
464 * contiguous buffer on which the AES functions can work.
466 if (!acb
->cluster_data
) {
468 g_malloc0(QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
471 assert(cur_nr_sectors
<=
472 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
);
473 qemu_iovec_reset(&acb
->hd_qiov
);
474 qemu_iovec_add(&acb
->hd_qiov
, acb
->cluster_data
,
475 512 * cur_nr_sectors
);
478 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
479 qemu_co_mutex_unlock(&s
->lock
);
480 ret
= bdrv_co_readv(bs
->file
,
481 (cluster_offset
>> 9) + index_in_cluster
,
482 cur_nr_sectors
, &acb
->hd_qiov
);
483 qemu_co_mutex_lock(&s
->lock
);
487 if (s
->crypt_method
) {
488 qcow2_encrypt_sectors(s
, acb
->sector_num
, acb
->cluster_data
,
489 acb
->cluster_data
, cur_nr_sectors
, 0, &s
->aes_decrypt_key
);
490 qemu_iovec_reset(&acb
->hd_qiov
);
491 qemu_iovec_copy(&acb
->hd_qiov
, acb
->qiov
, acb
->bytes_done
,
492 cur_nr_sectors
* 512);
493 qemu_iovec_from_buffer(&acb
->hd_qiov
, acb
->cluster_data
,
494 512 * cur_nr_sectors
);
498 acb
->remaining_sectors
-= cur_nr_sectors
;
499 acb
->sector_num
+= cur_nr_sectors
;
500 acb
->bytes_done
+= cur_nr_sectors
* 512;
505 static QCowAIOCB
*qcow2_aio_setup(BlockDriverState
*bs
, int64_t sector_num
,
506 QEMUIOVector
*qiov
, int nb_sectors
,
507 BlockDriverCompletionFunc
*cb
,
508 void *opaque
, QCowAIOCB
*acb
)
510 memset(acb
, 0, sizeof(*acb
));
512 acb
->sector_num
= sector_num
;
515 qemu_iovec_init(&acb
->hd_qiov
, qiov
->niov
);
518 acb
->remaining_sectors
= nb_sectors
;
522 static int qcow2_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
523 int nb_sectors
, QEMUIOVector
*qiov
)
525 BDRVQcowState
*s
= bs
->opaque
;
529 qcow2_aio_setup(bs
, sector_num
, qiov
, nb_sectors
, NULL
, NULL
, &acb
);
531 qemu_co_mutex_lock(&s
->lock
);
533 ret
= qcow2_aio_read_cb(&acb
);
535 qemu_co_mutex_unlock(&s
->lock
);
537 qemu_iovec_destroy(&acb
.hd_qiov
);
542 static void run_dependent_requests(BDRVQcowState
*s
, QCowL2Meta
*m
)
544 /* Take the request off the list of running requests */
545 if (m
->nb_clusters
!= 0) {
546 QLIST_REMOVE(m
, next_in_flight
);
549 /* Restart all dependent requests */
550 if (!qemu_co_queue_empty(&m
->dependent_requests
)) {
551 qemu_co_mutex_unlock(&s
->lock
);
552 while(qemu_co_queue_next(&m
->dependent_requests
));
553 qemu_co_mutex_lock(&s
->lock
);
558 * Returns 0 when the request is completed successfully, 1 when there is still
559 * a part left to do and -errno in error cases.
561 static int qcow2_aio_write_cb(QCowAIOCB
*acb
)
563 BlockDriverState
*bs
= acb
->common
.bs
;
564 BDRVQcowState
*s
= bs
->opaque
;
565 int index_in_cluster
;
568 int cur_nr_sectors
; /* number of sectors in current iteration */
570 uint64_t cluster_offset
;
572 l2meta
.nb_clusters
= 0;
573 qemu_co_queue_init(&l2meta
.dependent_requests
);
575 if (acb
->remaining_sectors
== 0) {
576 /* request completed */
580 index_in_cluster
= acb
->sector_num
& (s
->cluster_sectors
- 1);
581 n_end
= index_in_cluster
+ acb
->remaining_sectors
;
582 if (s
->crypt_method
&&
583 n_end
> QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
)
584 n_end
= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
;
586 ret
= qcow2_alloc_cluster_offset(bs
, acb
->sector_num
<< 9,
587 index_in_cluster
, n_end
, &cur_nr_sectors
, &l2meta
);
592 cluster_offset
= l2meta
.cluster_offset
;
593 assert((cluster_offset
& 511) == 0);
595 qemu_iovec_reset(&acb
->hd_qiov
);
596 qemu_iovec_copy(&acb
->hd_qiov
, acb
->qiov
, acb
->bytes_done
,
597 cur_nr_sectors
* 512);
599 if (s
->crypt_method
) {
600 if (!acb
->cluster_data
) {
601 acb
->cluster_data
= g_malloc0(QCOW_MAX_CRYPT_CLUSTERS
*
605 assert(acb
->hd_qiov
.size
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
606 qemu_iovec_to_buffer(&acb
->hd_qiov
, acb
->cluster_data
);
608 qcow2_encrypt_sectors(s
, acb
->sector_num
, acb
->cluster_data
,
609 acb
->cluster_data
, cur_nr_sectors
, 1, &s
->aes_encrypt_key
);
611 qemu_iovec_reset(&acb
->hd_qiov
);
612 qemu_iovec_add(&acb
->hd_qiov
, acb
->cluster_data
,
613 cur_nr_sectors
* 512);
616 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
617 qemu_co_mutex_unlock(&s
->lock
);
618 ret
= bdrv_co_writev(bs
->file
,
619 (cluster_offset
>> 9) + index_in_cluster
,
620 cur_nr_sectors
, &acb
->hd_qiov
);
621 qemu_co_mutex_lock(&s
->lock
);
626 ret
= qcow2_alloc_cluster_link_l2(bs
, &l2meta
);
628 run_dependent_requests(s
, &l2meta
);
634 acb
->remaining_sectors
-= cur_nr_sectors
;
635 acb
->sector_num
+= cur_nr_sectors
;
636 acb
->bytes_done
+= cur_nr_sectors
* 512;
641 static int qcow2_co_writev(BlockDriverState
*bs
,
646 BDRVQcowState
*s
= bs
->opaque
;
650 qcow2_aio_setup(bs
, sector_num
, qiov
, nb_sectors
, NULL
, NULL
, &acb
);
651 s
->cluster_cache_offset
= -1; /* disable compressed cache */
653 qemu_co_mutex_lock(&s
->lock
);
655 ret
= qcow2_aio_write_cb(&acb
);
657 qemu_co_mutex_unlock(&s
->lock
);
659 qemu_iovec_destroy(&acb
.hd_qiov
);
664 static void qcow2_close(BlockDriverState
*bs
)
666 BDRVQcowState
*s
= bs
->opaque
;
669 qcow2_cache_flush(bs
, s
->l2_table_cache
);
670 qcow2_cache_flush(bs
, s
->refcount_block_cache
);
672 qcow2_cache_destroy(bs
, s
->l2_table_cache
);
673 qcow2_cache_destroy(bs
, s
->refcount_block_cache
);
675 g_free(s
->cluster_cache
);
676 g_free(s
->cluster_data
);
677 qcow2_refcount_close(bs
);
681 * Updates the variable length parts of the qcow2 header, i.e. the backing file
682 * name and all extensions. qcow2 was not designed to allow such changes, so if
683 * we run out of space (we can only use the first cluster) this function may
686 * Returns 0 on success, -errno in error cases.
688 static int qcow2_update_ext_header(BlockDriverState
*bs
,
689 const char *backing_file
, const char *backing_fmt
)
691 size_t backing_file_len
= 0;
692 size_t backing_fmt_len
= 0;
693 BDRVQcowState
*s
= bs
->opaque
;
694 QCowExtension ext_backing_fmt
= {0, 0};
697 /* Backing file format doesn't make sense without a backing file */
698 if (backing_fmt
&& !backing_file
) {
702 /* Prepare the backing file format extension if needed */
704 ext_backing_fmt
.len
= cpu_to_be32(strlen(backing_fmt
));
705 ext_backing_fmt
.magic
= cpu_to_be32(QCOW2_EXT_MAGIC_BACKING_FORMAT
);
706 backing_fmt_len
= ((sizeof(ext_backing_fmt
)
707 + strlen(backing_fmt
) + 7) & ~7);
710 /* Check if we can fit the new header into the first cluster */
712 backing_file_len
= strlen(backing_file
);
715 size_t header_size
= sizeof(QCowHeader
) + backing_file_len
718 if (header_size
> s
->cluster_size
) {
722 /* Rewrite backing file name and qcow2 extensions */
723 size_t ext_size
= header_size
- sizeof(QCowHeader
);
724 uint8_t buf
[ext_size
];
726 size_t backing_file_offset
= 0;
730 int padding
= backing_fmt_len
-
731 (sizeof(ext_backing_fmt
) + strlen(backing_fmt
));
733 memcpy(buf
+ offset
, &ext_backing_fmt
, sizeof(ext_backing_fmt
));
734 offset
+= sizeof(ext_backing_fmt
);
736 memcpy(buf
+ offset
, backing_fmt
, strlen(backing_fmt
));
737 offset
+= strlen(backing_fmt
);
739 memset(buf
+ offset
, 0, padding
);
743 memcpy(buf
+ offset
, backing_file
, backing_file_len
);
744 backing_file_offset
= sizeof(QCowHeader
) + offset
;
747 ret
= bdrv_pwrite_sync(bs
->file
, sizeof(QCowHeader
), buf
, ext_size
);
752 /* Update header fields */
753 uint64_t be_backing_file_offset
= cpu_to_be64(backing_file_offset
);
754 uint32_t be_backing_file_size
= cpu_to_be32(backing_file_len
);
756 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, backing_file_offset
),
757 &be_backing_file_offset
, sizeof(uint64_t));
762 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, backing_file_size
),
763 &be_backing_file_size
, sizeof(uint32_t));
773 static int qcow2_change_backing_file(BlockDriverState
*bs
,
774 const char *backing_file
, const char *backing_fmt
)
776 return qcow2_update_ext_header(bs
, backing_file
, backing_fmt
);
779 static int preallocate(BlockDriverState
*bs
)
787 nb_sectors
= bdrv_getlength(bs
) >> 9;
789 qemu_co_queue_init(&meta
.dependent_requests
);
790 meta
.cluster_offset
= 0;
793 num
= MIN(nb_sectors
, INT_MAX
>> 9);
794 ret
= qcow2_alloc_cluster_offset(bs
, offset
, 0, num
, &num
, &meta
);
799 ret
= qcow2_alloc_cluster_link_l2(bs
, &meta
);
801 qcow2_free_any_clusters(bs
, meta
.cluster_offset
, meta
.nb_clusters
);
805 /* There are no dependent requests, but we need to remove our request
806 * from the list of in-flight requests */
807 run_dependent_requests(bs
->opaque
, &meta
);
809 /* TODO Preallocate data if requested */
816 * It is expected that the image file is large enough to actually contain
817 * all of the allocated clusters (otherwise we get failing reads after
818 * EOF). Extend the image to the last allocated sector.
820 if (meta
.cluster_offset
!= 0) {
823 ret
= bdrv_write(bs
->file
, (meta
.cluster_offset
>> 9) + num
- 1, buf
, 1);
832 static int qcow2_create2(const char *filename
, int64_t total_size
,
833 const char *backing_file
, const char *backing_format
,
834 int flags
, size_t cluster_size
, int prealloc
,
835 QEMUOptionParameter
*options
)
837 /* Calulate cluster_bits */
839 cluster_bits
= ffs(cluster_size
) - 1;
840 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
841 (1 << cluster_bits
) != cluster_size
)
844 "Cluster size must be a power of two between %d and %dk",
845 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
850 * Open the image file and write a minimal qcow2 header.
852 * We keep things simple and start with a zero-sized image. We also
853 * do without refcount blocks or a L1 table for now. We'll fix the
854 * inconsistency later.
856 * We do need a refcount table because growing the refcount table means
857 * allocating two new refcount blocks - the seconds of which would be at
858 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
859 * size for any qcow2 image.
861 BlockDriverState
* bs
;
863 uint8_t* refcount_table
;
866 ret
= bdrv_create_file(filename
, options
);
871 ret
= bdrv_file_open(&bs
, filename
, BDRV_O_RDWR
);
876 /* Write the header */
877 memset(&header
, 0, sizeof(header
));
878 header
.magic
= cpu_to_be32(QCOW_MAGIC
);
879 header
.version
= cpu_to_be32(QCOW_VERSION
);
880 header
.cluster_bits
= cpu_to_be32(cluster_bits
);
881 header
.size
= cpu_to_be64(0);
882 header
.l1_table_offset
= cpu_to_be64(0);
883 header
.l1_size
= cpu_to_be32(0);
884 header
.refcount_table_offset
= cpu_to_be64(cluster_size
);
885 header
.refcount_table_clusters
= cpu_to_be32(1);
887 if (flags
& BLOCK_FLAG_ENCRYPT
) {
888 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
890 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
893 ret
= bdrv_pwrite(bs
, 0, &header
, sizeof(header
));
898 /* Write an empty refcount table */
899 refcount_table
= g_malloc0(cluster_size
);
900 ret
= bdrv_pwrite(bs
, cluster_size
, refcount_table
, cluster_size
);
901 g_free(refcount_table
);
910 * And now open the image and make it consistent first (i.e. increase the
911 * refcount of the cluster that is occupied by the header and the refcount
914 BlockDriver
* drv
= bdrv_find_format("qcow2");
916 ret
= bdrv_open(bs
, filename
,
917 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_NO_FLUSH
, drv
);
922 ret
= qcow2_alloc_clusters(bs
, 2 * cluster_size
);
926 } else if (ret
!= 0) {
927 error_report("Huh, first cluster in empty image is already in use?");
931 /* Okay, now that we have a valid image, let's give it the right size */
932 ret
= bdrv_truncate(bs
, total_size
* BDRV_SECTOR_SIZE
);
937 /* Want a backing file? There you go.*/
939 ret
= bdrv_change_backing_file(bs
, backing_file
, backing_format
);
945 /* And if we're supposed to preallocate metadata, do that now */
947 ret
= preallocate(bs
);
959 static int qcow2_create(const char *filename
, QEMUOptionParameter
*options
)
961 const char *backing_file
= NULL
;
962 const char *backing_fmt
= NULL
;
963 uint64_t sectors
= 0;
965 size_t cluster_size
= DEFAULT_CLUSTER_SIZE
;
968 /* Read out options */
969 while (options
&& options
->name
) {
970 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
971 sectors
= options
->value
.n
/ 512;
972 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
973 backing_file
= options
->value
.s
;
974 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FMT
)) {
975 backing_fmt
= options
->value
.s
;
976 } else if (!strcmp(options
->name
, BLOCK_OPT_ENCRYPT
)) {
977 flags
|= options
->value
.n
? BLOCK_FLAG_ENCRYPT
: 0;
978 } else if (!strcmp(options
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
979 if (options
->value
.n
) {
980 cluster_size
= options
->value
.n
;
982 } else if (!strcmp(options
->name
, BLOCK_OPT_PREALLOC
)) {
983 if (!options
->value
.s
|| !strcmp(options
->value
.s
, "off")) {
985 } else if (!strcmp(options
->value
.s
, "metadata")) {
988 fprintf(stderr
, "Invalid preallocation mode: '%s'\n",
996 if (backing_file
&& prealloc
) {
997 fprintf(stderr
, "Backing file and preallocation cannot be used at "
1002 return qcow2_create2(filename
, sectors
, backing_file
, backing_fmt
, flags
,
1003 cluster_size
, prealloc
, options
);
1006 static int qcow2_make_empty(BlockDriverState
*bs
)
1009 /* XXX: not correct */
1010 BDRVQcowState
*s
= bs
->opaque
;
1011 uint32_t l1_length
= s
->l1_size
* sizeof(uint64_t);
1014 memset(s
->l1_table
, 0, l1_length
);
1015 if (bdrv_pwrite(bs
->file
, s
->l1_table_offset
, s
->l1_table
, l1_length
) < 0)
1017 ret
= bdrv_truncate(bs
->file
, s
->l1_table_offset
+ l1_length
);
1026 static int qcow2_discard(BlockDriverState
*bs
, int64_t sector_num
,
1029 return qcow2_discard_clusters(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1033 static int qcow2_truncate(BlockDriverState
*bs
, int64_t offset
)
1035 BDRVQcowState
*s
= bs
->opaque
;
1036 int ret
, new_l1_size
;
1042 /* cannot proceed if image has snapshots */
1043 if (s
->nb_snapshots
) {
1047 /* shrinking is currently not supported */
1048 if (offset
< bs
->total_sectors
* 512) {
1052 new_l1_size
= size_to_l1(s
, offset
);
1053 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
1058 /* write updated header.size */
1059 offset
= cpu_to_be64(offset
);
1060 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
1061 &offset
, sizeof(uint64_t));
1066 s
->l1_vm_state_index
= new_l1_size
;
1070 /* XXX: put compressed sectors first, then all the cluster aligned
1071 tables to avoid losing bytes in alignment */
1072 static int qcow2_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1073 const uint8_t *buf
, int nb_sectors
)
1075 BDRVQcowState
*s
= bs
->opaque
;
1079 uint64_t cluster_offset
;
1081 if (nb_sectors
== 0) {
1082 /* align end of file to a sector boundary to ease reading with
1083 sector based I/Os */
1084 cluster_offset
= bdrv_getlength(bs
->file
);
1085 cluster_offset
= (cluster_offset
+ 511) & ~511;
1086 bdrv_truncate(bs
->file
, cluster_offset
);
1090 if (nb_sectors
!= s
->cluster_sectors
)
1093 out_buf
= g_malloc(s
->cluster_size
+ (s
->cluster_size
/ 1000) + 128);
1095 /* best compression, small window, no zlib header */
1096 memset(&strm
, 0, sizeof(strm
));
1097 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
1099 9, Z_DEFAULT_STRATEGY
);
1105 strm
.avail_in
= s
->cluster_size
;
1106 strm
.next_in
= (uint8_t *)buf
;
1107 strm
.avail_out
= s
->cluster_size
;
1108 strm
.next_out
= out_buf
;
1110 ret
= deflate(&strm
, Z_FINISH
);
1111 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
1116 out_len
= strm
.next_out
- out_buf
;
1120 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
1121 /* could not compress: write normal cluster */
1122 bdrv_write(bs
, sector_num
, buf
, s
->cluster_sectors
);
1124 cluster_offset
= qcow2_alloc_compressed_cluster_offset(bs
,
1125 sector_num
<< 9, out_len
);
1126 if (!cluster_offset
)
1128 cluster_offset
&= s
->cluster_offset_mask
;
1129 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
1130 if (bdrv_pwrite(bs
->file
, cluster_offset
, out_buf
, out_len
) != out_len
) {
1140 static int qcow2_flush(BlockDriverState
*bs
)
1142 BDRVQcowState
*s
= bs
->opaque
;
1145 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1150 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1155 return bdrv_flush(bs
->file
);
1158 static BlockDriverAIOCB
*qcow2_aio_flush(BlockDriverState
*bs
,
1159 BlockDriverCompletionFunc
*cb
,
1162 BDRVQcowState
*s
= bs
->opaque
;
1165 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
1170 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
1175 return bdrv_aio_flush(bs
->file
, cb
, opaque
);
1178 static int64_t qcow2_vm_state_offset(BDRVQcowState
*s
)
1180 return (int64_t)s
->l1_vm_state_index
<< (s
->cluster_bits
+ s
->l2_bits
);
1183 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1185 BDRVQcowState
*s
= bs
->opaque
;
1186 bdi
->cluster_size
= s
->cluster_size
;
1187 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
1192 static int qcow2_check(BlockDriverState
*bs
, BdrvCheckResult
*result
)
1194 return qcow2_check_refcounts(bs
, result
);
1198 static void dump_refcounts(BlockDriverState
*bs
)
1200 BDRVQcowState
*s
= bs
->opaque
;
1201 int64_t nb_clusters
, k
, k1
, size
;
1204 size
= bdrv_getlength(bs
->file
);
1205 nb_clusters
= size_to_clusters(s
, size
);
1206 for(k
= 0; k
< nb_clusters
;) {
1208 refcount
= get_refcount(bs
, k
);
1210 while (k
< nb_clusters
&& get_refcount(bs
, k
) == refcount
)
1212 printf("%" PRId64
": refcount=%d nb=%" PRId64
"\n", k
, refcount
,
1218 static int qcow2_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1219 int64_t pos
, int size
)
1221 BDRVQcowState
*s
= bs
->opaque
;
1222 int growable
= bs
->growable
;
1225 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
1227 ret
= bdrv_pwrite(bs
, qcow2_vm_state_offset(s
) + pos
, buf
, size
);
1228 bs
->growable
= growable
;
1233 static int qcow2_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1234 int64_t pos
, int size
)
1236 BDRVQcowState
*s
= bs
->opaque
;
1237 int growable
= bs
->growable
;
1240 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
1242 ret
= bdrv_pread(bs
, qcow2_vm_state_offset(s
) + pos
, buf
, size
);
1243 bs
->growable
= growable
;
1248 static QEMUOptionParameter qcow2_create_options
[] = {
1250 .name
= BLOCK_OPT_SIZE
,
1252 .help
= "Virtual disk size"
1255 .name
= BLOCK_OPT_BACKING_FILE
,
1257 .help
= "File name of a base image"
1260 .name
= BLOCK_OPT_BACKING_FMT
,
1262 .help
= "Image format of the base image"
1265 .name
= BLOCK_OPT_ENCRYPT
,
1267 .help
= "Encrypt the image"
1270 .name
= BLOCK_OPT_CLUSTER_SIZE
,
1272 .help
= "qcow2 cluster size",
1273 .value
= { .n
= DEFAULT_CLUSTER_SIZE
},
1276 .name
= BLOCK_OPT_PREALLOC
,
1278 .help
= "Preallocation mode (allowed values: off, metadata)"
1283 static BlockDriver bdrv_qcow2
= {
1284 .format_name
= "qcow2",
1285 .instance_size
= sizeof(BDRVQcowState
),
1286 .bdrv_probe
= qcow2_probe
,
1287 .bdrv_open
= qcow2_open
,
1288 .bdrv_close
= qcow2_close
,
1289 .bdrv_create
= qcow2_create
,
1290 .bdrv_flush
= qcow2_flush
,
1291 .bdrv_is_allocated
= qcow2_is_allocated
,
1292 .bdrv_set_key
= qcow2_set_key
,
1293 .bdrv_make_empty
= qcow2_make_empty
,
1295 .bdrv_co_readv
= qcow2_co_readv
,
1296 .bdrv_co_writev
= qcow2_co_writev
,
1297 .bdrv_aio_flush
= qcow2_aio_flush
,
1299 .bdrv_discard
= qcow2_discard
,
1300 .bdrv_truncate
= qcow2_truncate
,
1301 .bdrv_write_compressed
= qcow2_write_compressed
,
1303 .bdrv_snapshot_create
= qcow2_snapshot_create
,
1304 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
1305 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
1306 .bdrv_snapshot_list
= qcow2_snapshot_list
,
1307 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
1308 .bdrv_get_info
= qcow2_get_info
,
1310 .bdrv_save_vmstate
= qcow2_save_vmstate
,
1311 .bdrv_load_vmstate
= qcow2_load_vmstate
,
1313 .bdrv_change_backing_file
= qcow2_change_backing_file
,
1315 .create_options
= qcow2_create_options
,
1316 .bdrv_check
= qcow2_check
,
1319 static void bdrv_qcow2_init(void)
1321 bdrv_register(&bdrv_qcow2
);
1324 block_init(bdrv_qcow2_init
);