2 * Block driver for the QCOW 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/osdep.h"
25 #include "qemu-common.h"
26 #include "block/block_int.h"
27 #include "sysemu/block-backend.h"
28 #include "qemu/module.h"
30 #include "qapi/qmp/qerror.h"
31 #include "crypto/cipher.h"
32 #include "migration/migration.h"
34 /**************************************************************/
35 /* QEMU COW block driver with compression and encryption support */
37 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
38 #define QCOW_VERSION 1
40 #define QCOW_CRYPT_NONE 0
41 #define QCOW_CRYPT_AES 1
43 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
45 typedef struct QCowHeader
{
48 uint64_t backing_file_offset
;
49 uint32_t backing_file_size
;
51 uint64_t size
; /* in bytes */
55 uint32_t crypt_method
;
56 uint64_t l1_table_offset
;
57 } QEMU_PACKED QCowHeader
;
59 #define L2_CACHE_SIZE 16
61 typedef struct BDRVQcowState
{
68 uint64_t cluster_offset_mask
;
69 uint64_t l1_table_offset
;
72 uint64_t l2_cache_offsets
[L2_CACHE_SIZE
];
73 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
74 uint8_t *cluster_cache
;
75 uint8_t *cluster_data
;
76 uint64_t cluster_cache_offset
;
77 QCryptoCipher
*cipher
; /* NULL if no key yet */
78 uint32_t crypt_method_header
;
80 Error
*migration_blocker
;
83 static int decompress_cluster(BlockDriverState
*bs
, uint64_t cluster_offset
);
85 static int qcow_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
87 const QCowHeader
*cow_header
= (const void *)buf
;
89 if (buf_size
>= sizeof(QCowHeader
) &&
90 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
91 be32_to_cpu(cow_header
->version
) == QCOW_VERSION
)
97 static int qcow_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
100 BDRVQcowState
*s
= bs
->opaque
;
101 unsigned int len
, i
, shift
;
105 ret
= bdrv_pread(bs
->file
->bs
, 0, &header
, sizeof(header
));
109 be32_to_cpus(&header
.magic
);
110 be32_to_cpus(&header
.version
);
111 be64_to_cpus(&header
.backing_file_offset
);
112 be32_to_cpus(&header
.backing_file_size
);
113 be32_to_cpus(&header
.mtime
);
114 be64_to_cpus(&header
.size
);
115 be32_to_cpus(&header
.crypt_method
);
116 be64_to_cpus(&header
.l1_table_offset
);
118 if (header
.magic
!= QCOW_MAGIC
) {
119 error_setg(errp
, "Image not in qcow format");
123 if (header
.version
!= QCOW_VERSION
) {
125 snprintf(version
, sizeof(version
), "QCOW version %" PRIu32
,
127 error_setg(errp
, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
128 bdrv_get_device_or_node_name(bs
), "qcow", version
);
133 if (header
.size
<= 1) {
134 error_setg(errp
, "Image size is too small (must be at least 2 bytes)");
138 if (header
.cluster_bits
< 9 || header
.cluster_bits
> 16) {
139 error_setg(errp
, "Cluster size must be between 512 and 64k");
144 /* l2_bits specifies number of entries; storing a uint64_t in each entry,
145 * so bytes = num_entries << 3. */
146 if (header
.l2_bits
< 9 - 3 || header
.l2_bits
> 16 - 3) {
147 error_setg(errp
, "L2 table size must be between 512 and 64k");
152 if (header
.crypt_method
> QCOW_CRYPT_AES
) {
153 error_setg(errp
, "invalid encryption method in qcow header");
157 if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128
)) {
158 error_setg(errp
, "AES cipher not available");
162 s
->crypt_method_header
= header
.crypt_method
;
163 if (s
->crypt_method_header
) {
166 s
->cluster_bits
= header
.cluster_bits
;
167 s
->cluster_size
= 1 << s
->cluster_bits
;
168 s
->cluster_sectors
= 1 << (s
->cluster_bits
- 9);
169 s
->l2_bits
= header
.l2_bits
;
170 s
->l2_size
= 1 << s
->l2_bits
;
171 bs
->total_sectors
= header
.size
/ 512;
172 s
->cluster_offset_mask
= (1LL << (63 - s
->cluster_bits
)) - 1;
174 /* read the level 1 table */
175 shift
= s
->cluster_bits
+ s
->l2_bits
;
176 if (header
.size
> UINT64_MAX
- (1LL << shift
)) {
177 error_setg(errp
, "Image too large");
181 uint64_t l1_size
= (header
.size
+ (1LL << shift
) - 1) >> shift
;
182 if (l1_size
> INT_MAX
/ sizeof(uint64_t)) {
183 error_setg(errp
, "Image too large");
187 s
->l1_size
= l1_size
;
190 s
->l1_table_offset
= header
.l1_table_offset
;
191 s
->l1_table
= g_try_new(uint64_t, s
->l1_size
);
192 if (s
->l1_table
== NULL
) {
193 error_setg(errp
, "Could not allocate memory for L1 table");
198 ret
= bdrv_pread(bs
->file
->bs
, s
->l1_table_offset
, s
->l1_table
,
199 s
->l1_size
* sizeof(uint64_t));
204 for(i
= 0;i
< s
->l1_size
; i
++) {
205 be64_to_cpus(&s
->l1_table
[i
]);
208 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
210 qemu_try_blockalign(bs
->file
->bs
,
211 s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
212 if (s
->l2_cache
== NULL
) {
213 error_setg(errp
, "Could not allocate L2 table cache");
217 s
->cluster_cache
= g_malloc(s
->cluster_size
);
218 s
->cluster_data
= g_malloc(s
->cluster_size
);
219 s
->cluster_cache_offset
= -1;
221 /* read the backing file name */
222 if (header
.backing_file_offset
!= 0) {
223 len
= header
.backing_file_size
;
224 if (len
> 1023 || len
>= sizeof(bs
->backing_file
)) {
225 error_setg(errp
, "Backing file name too long");
229 ret
= bdrv_pread(bs
->file
->bs
, header
.backing_file_offset
,
230 bs
->backing_file
, len
);
234 bs
->backing_file
[len
] = '\0';
237 /* Disable migration when qcow images are used */
238 error_setg(&s
->migration_blocker
, "The qcow format used by node '%s' "
239 "does not support live migration",
240 bdrv_get_device_or_node_name(bs
));
241 migrate_add_blocker(s
->migration_blocker
);
243 qemu_co_mutex_init(&s
->lock
);
248 qemu_vfree(s
->l2_cache
);
249 g_free(s
->cluster_cache
);
250 g_free(s
->cluster_data
);
255 /* We have nothing to do for QCOW reopen, stubs just return
257 static int qcow_reopen_prepare(BDRVReopenState
*state
,
258 BlockReopenQueue
*queue
, Error
**errp
)
263 static int qcow_set_key(BlockDriverState
*bs
, const char *key
)
265 BDRVQcowState
*s
= bs
->opaque
;
270 memset(keybuf
, 0, 16);
274 /* XXX: we could compress the chars to 7 bits to increase
276 for(i
= 0;i
< len
;i
++) {
279 assert(bs
->encrypted
);
281 qcrypto_cipher_free(s
->cipher
);
282 s
->cipher
= qcrypto_cipher_new(
283 QCRYPTO_CIPHER_ALG_AES_128
,
284 QCRYPTO_CIPHER_MODE_CBC
,
285 keybuf
, G_N_ELEMENTS(keybuf
),
289 /* XXX would be nice if errors in this method could
290 * be properly propagate to the caller. Would need
291 * the bdrv_set_key() API signature to be fixed. */
298 /* The crypt function is compatible with the linux cryptoloop
299 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
301 static int encrypt_sectors(BDRVQcowState
*s
, int64_t sector_num
,
302 uint8_t *out_buf
, const uint8_t *in_buf
,
303 int nb_sectors
, bool enc
, Error
**errp
)
312 for(i
= 0; i
< nb_sectors
; i
++) {
313 ivec
.ll
[0] = cpu_to_le64(sector_num
);
315 if (qcrypto_cipher_setiv(s
->cipher
,
316 ivec
.b
, G_N_ELEMENTS(ivec
.b
),
321 ret
= qcrypto_cipher_encrypt(s
->cipher
,
327 ret
= qcrypto_cipher_decrypt(s
->cipher
,
347 * 1 to allocate a normal cluster (for sector indexes 'n_start' to
350 * 2 to allocate a compressed cluster of size
351 * 'compressed_size'. 'compressed_size' must be > 0 and <
354 * return 0 if not allocated.
356 static uint64_t get_cluster_offset(BlockDriverState
*bs
,
357 uint64_t offset
, int allocate
,
359 int n_start
, int n_end
)
361 BDRVQcowState
*s
= bs
->opaque
;
362 int min_index
, i
, j
, l1_index
, l2_index
;
363 uint64_t l2_offset
, *l2_table
, cluster_offset
, tmp
;
367 l1_index
= offset
>> (s
->l2_bits
+ s
->cluster_bits
);
368 l2_offset
= s
->l1_table
[l1_index
];
373 /* allocate a new l2 entry */
374 l2_offset
= bdrv_getlength(bs
->file
->bs
);
375 /* round to cluster size */
376 l2_offset
= (l2_offset
+ s
->cluster_size
- 1) & ~(s
->cluster_size
- 1);
377 /* update the L1 entry */
378 s
->l1_table
[l1_index
] = l2_offset
;
379 tmp
= cpu_to_be64(l2_offset
);
380 if (bdrv_pwrite_sync(bs
->file
->bs
,
381 s
->l1_table_offset
+ l1_index
* sizeof(tmp
),
382 &tmp
, sizeof(tmp
)) < 0)
386 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
387 if (l2_offset
== s
->l2_cache_offsets
[i
]) {
388 /* increment the hit count */
389 if (++s
->l2_cache_counts
[i
] == 0xffffffff) {
390 for(j
= 0; j
< L2_CACHE_SIZE
; j
++) {
391 s
->l2_cache_counts
[j
] >>= 1;
394 l2_table
= s
->l2_cache
+ (i
<< s
->l2_bits
);
398 /* not found: load a new entry in the least used one */
400 min_count
= 0xffffffff;
401 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
402 if (s
->l2_cache_counts
[i
] < min_count
) {
403 min_count
= s
->l2_cache_counts
[i
];
407 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
409 memset(l2_table
, 0, s
->l2_size
* sizeof(uint64_t));
410 if (bdrv_pwrite_sync(bs
->file
->bs
, l2_offset
, l2_table
,
411 s
->l2_size
* sizeof(uint64_t)) < 0)
414 if (bdrv_pread(bs
->file
->bs
, l2_offset
, l2_table
,
415 s
->l2_size
* sizeof(uint64_t)) !=
416 s
->l2_size
* sizeof(uint64_t))
419 s
->l2_cache_offsets
[min_index
] = l2_offset
;
420 s
->l2_cache_counts
[min_index
] = 1;
422 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
423 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
424 if (!cluster_offset
||
425 ((cluster_offset
& QCOW_OFLAG_COMPRESSED
) && allocate
== 1)) {
428 /* allocate a new cluster */
429 if ((cluster_offset
& QCOW_OFLAG_COMPRESSED
) &&
430 (n_end
- n_start
) < s
->cluster_sectors
) {
431 /* if the cluster is already compressed, we must
432 decompress it in the case it is not completely
434 if (decompress_cluster(bs
, cluster_offset
) < 0)
436 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
437 cluster_offset
= (cluster_offset
+ s
->cluster_size
- 1) &
438 ~(s
->cluster_size
- 1);
439 /* write the cluster content */
440 if (bdrv_pwrite(bs
->file
->bs
, cluster_offset
, s
->cluster_cache
,
445 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
447 /* round to cluster size */
448 cluster_offset
= (cluster_offset
+ s
->cluster_size
- 1) &
449 ~(s
->cluster_size
- 1);
450 bdrv_truncate(bs
->file
->bs
, cluster_offset
+ s
->cluster_size
);
451 /* if encrypted, we must initialize the cluster
452 content which won't be written */
454 (n_end
- n_start
) < s
->cluster_sectors
) {
457 start_sect
= (offset
& ~(s
->cluster_size
- 1)) >> 9;
458 memset(s
->cluster_data
+ 512, 0x00, 512);
459 for(i
= 0; i
< s
->cluster_sectors
; i
++) {
460 if (i
< n_start
|| i
>= n_end
) {
462 if (encrypt_sectors(s
, start_sect
+ i
,
464 s
->cluster_data
+ 512, 1,
470 if (bdrv_pwrite(bs
->file
->bs
,
471 cluster_offset
+ i
* 512,
472 s
->cluster_data
, 512) != 512)
477 } else if (allocate
== 2) {
478 cluster_offset
|= QCOW_OFLAG_COMPRESSED
|
479 (uint64_t)compressed_size
<< (63 - s
->cluster_bits
);
482 /* update L2 table */
483 tmp
= cpu_to_be64(cluster_offset
);
484 l2_table
[l2_index
] = tmp
;
485 if (bdrv_pwrite_sync(bs
->file
->bs
, l2_offset
+ l2_index
* sizeof(tmp
),
486 &tmp
, sizeof(tmp
)) < 0)
489 return cluster_offset
;
492 static int64_t coroutine_fn
qcow_co_get_block_status(BlockDriverState
*bs
,
493 int64_t sector_num
, int nb_sectors
, int *pnum
, BlockDriverState
**file
)
495 BDRVQcowState
*s
= bs
->opaque
;
496 int index_in_cluster
, n
;
497 uint64_t cluster_offset
;
499 qemu_co_mutex_lock(&s
->lock
);
500 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 0, 0, 0, 0);
501 qemu_co_mutex_unlock(&s
->lock
);
502 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
503 n
= s
->cluster_sectors
- index_in_cluster
;
507 if (!cluster_offset
) {
510 if ((cluster_offset
& QCOW_OFLAG_COMPRESSED
) || s
->cipher
) {
511 return BDRV_BLOCK_DATA
;
513 cluster_offset
|= (index_in_cluster
<< BDRV_SECTOR_BITS
);
514 *file
= bs
->file
->bs
;
515 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| cluster_offset
;
518 static int decompress_buffer(uint8_t *out_buf
, int out_buf_size
,
519 const uint8_t *buf
, int buf_size
)
521 z_stream strm1
, *strm
= &strm1
;
524 memset(strm
, 0, sizeof(*strm
));
526 strm
->next_in
= (uint8_t *)buf
;
527 strm
->avail_in
= buf_size
;
528 strm
->next_out
= out_buf
;
529 strm
->avail_out
= out_buf_size
;
531 ret
= inflateInit2(strm
, -12);
534 ret
= inflate(strm
, Z_FINISH
);
535 out_len
= strm
->next_out
- out_buf
;
536 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) ||
537 out_len
!= out_buf_size
) {
545 static int decompress_cluster(BlockDriverState
*bs
, uint64_t cluster_offset
)
547 BDRVQcowState
*s
= bs
->opaque
;
551 coffset
= cluster_offset
& s
->cluster_offset_mask
;
552 if (s
->cluster_cache_offset
!= coffset
) {
553 csize
= cluster_offset
>> (63 - s
->cluster_bits
);
554 csize
&= (s
->cluster_size
- 1);
555 ret
= bdrv_pread(bs
->file
->bs
, coffset
, s
->cluster_data
, csize
);
558 if (decompress_buffer(s
->cluster_cache
, s
->cluster_size
,
559 s
->cluster_data
, csize
) < 0) {
562 s
->cluster_cache_offset
= coffset
;
567 static coroutine_fn
int qcow_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
568 int nb_sectors
, QEMUIOVector
*qiov
)
570 BDRVQcowState
*s
= bs
->opaque
;
571 int index_in_cluster
;
573 uint64_t cluster_offset
;
575 QEMUIOVector hd_qiov
;
580 if (qiov
->niov
> 1) {
581 buf
= orig_buf
= qemu_try_blockalign(bs
, qiov
->size
);
587 buf
= (uint8_t *)qiov
->iov
->iov_base
;
590 qemu_co_mutex_lock(&s
->lock
);
592 while (nb_sectors
!= 0) {
593 /* prepare next request */
594 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9,
596 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
597 n
= s
->cluster_sectors
- index_in_cluster
;
598 if (n
> nb_sectors
) {
602 if (!cluster_offset
) {
604 /* read from the base image */
605 hd_iov
.iov_base
= (void *)buf
;
606 hd_iov
.iov_len
= n
* 512;
607 qemu_iovec_init_external(&hd_qiov
, &hd_iov
, 1);
608 qemu_co_mutex_unlock(&s
->lock
);
609 ret
= bdrv_co_readv(bs
->backing
->bs
, sector_num
,
611 qemu_co_mutex_lock(&s
->lock
);
616 /* Note: in this case, no need to wait */
617 memset(buf
, 0, 512 * n
);
619 } else if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
620 /* add AIO support for compressed blocks ? */
621 if (decompress_cluster(bs
, cluster_offset
) < 0) {
625 s
->cluster_cache
+ index_in_cluster
* 512, 512 * n
);
627 if ((cluster_offset
& 511) != 0) {
630 hd_iov
.iov_base
= (void *)buf
;
631 hd_iov
.iov_len
= n
* 512;
632 qemu_iovec_init_external(&hd_qiov
, &hd_iov
, 1);
633 qemu_co_mutex_unlock(&s
->lock
);
634 ret
= bdrv_co_readv(bs
->file
->bs
,
635 (cluster_offset
>> 9) + index_in_cluster
,
637 qemu_co_mutex_lock(&s
->lock
);
643 if (encrypt_sectors(s
, sector_num
, buf
, buf
,
644 n
, false, &err
) < 0) {
657 qemu_co_mutex_unlock(&s
->lock
);
659 if (qiov
->niov
> 1) {
660 qemu_iovec_from_buf(qiov
, 0, orig_buf
, qiov
->size
);
661 qemu_vfree(orig_buf
);
672 static coroutine_fn
int qcow_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
673 int nb_sectors
, QEMUIOVector
*qiov
)
675 BDRVQcowState
*s
= bs
->opaque
;
676 int index_in_cluster
;
677 uint64_t cluster_offset
;
678 const uint8_t *src_buf
;
680 uint8_t *cluster_data
= NULL
;
682 QEMUIOVector hd_qiov
;
686 s
->cluster_cache_offset
= -1; /* disable compressed cache */
688 if (qiov
->niov
> 1) {
689 buf
= orig_buf
= qemu_try_blockalign(bs
, qiov
->size
);
693 qemu_iovec_to_buf(qiov
, 0, buf
, qiov
->size
);
696 buf
= (uint8_t *)qiov
->iov
->iov_base
;
699 qemu_co_mutex_lock(&s
->lock
);
701 while (nb_sectors
!= 0) {
703 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
704 n
= s
->cluster_sectors
- index_in_cluster
;
705 if (n
> nb_sectors
) {
708 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 1, 0,
710 index_in_cluster
+ n
);
711 if (!cluster_offset
|| (cluster_offset
& 511) != 0) {
719 cluster_data
= g_malloc0(s
->cluster_size
);
721 if (encrypt_sectors(s
, sector_num
, cluster_data
, buf
,
722 n
, true, &err
) < 0) {
727 src_buf
= cluster_data
;
732 hd_iov
.iov_base
= (void *)src_buf
;
733 hd_iov
.iov_len
= n
* 512;
734 qemu_iovec_init_external(&hd_qiov
, &hd_iov
, 1);
735 qemu_co_mutex_unlock(&s
->lock
);
736 ret
= bdrv_co_writev(bs
->file
->bs
,
737 (cluster_offset
>> 9) + index_in_cluster
,
739 qemu_co_mutex_lock(&s
->lock
);
749 qemu_co_mutex_unlock(&s
->lock
);
751 if (qiov
->niov
> 1) {
752 qemu_vfree(orig_buf
);
754 g_free(cluster_data
);
759 static void qcow_close(BlockDriverState
*bs
)
761 BDRVQcowState
*s
= bs
->opaque
;
763 qcrypto_cipher_free(s
->cipher
);
766 qemu_vfree(s
->l2_cache
);
767 g_free(s
->cluster_cache
);
768 g_free(s
->cluster_data
);
770 migrate_del_blocker(s
->migration_blocker
);
771 error_free(s
->migration_blocker
);
774 static int qcow_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
776 int header_size
, backing_filename_len
, l1_size
, shift
, i
;
779 int64_t total_size
= 0;
780 char *backing_file
= NULL
;
782 Error
*local_err
= NULL
;
784 BlockBackend
*qcow_blk
;
786 /* Read out options */
787 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
789 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
790 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_ENCRYPT
, false)) {
791 flags
|= BLOCK_FLAG_ENCRYPT
;
794 ret
= bdrv_create_file(filename
, opts
, &local_err
);
796 error_propagate(errp
, local_err
);
800 qcow_blk
= blk_new_open("image", filename
, NULL
, NULL
,
801 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_PROTOCOL
,
803 if (qcow_blk
== NULL
) {
804 error_propagate(errp
, local_err
);
809 blk_set_allow_write_beyond_eof(qcow_blk
, true);
811 ret
= blk_truncate(qcow_blk
, 0);
816 memset(&header
, 0, sizeof(header
));
817 header
.magic
= cpu_to_be32(QCOW_MAGIC
);
818 header
.version
= cpu_to_be32(QCOW_VERSION
);
819 header
.size
= cpu_to_be64(total_size
);
820 header_size
= sizeof(header
);
821 backing_filename_len
= 0;
823 if (strcmp(backing_file
, "fat:")) {
824 header
.backing_file_offset
= cpu_to_be64(header_size
);
825 backing_filename_len
= strlen(backing_file
);
826 header
.backing_file_size
= cpu_to_be32(backing_filename_len
);
827 header_size
+= backing_filename_len
;
829 /* special backing file for vvfat */
832 header
.cluster_bits
= 9; /* 512 byte cluster to avoid copying
833 unmodified sectors */
834 header
.l2_bits
= 12; /* 32 KB L2 tables */
836 header
.cluster_bits
= 12; /* 4 KB clusters */
837 header
.l2_bits
= 9; /* 4 KB L2 tables */
839 header_size
= (header_size
+ 7) & ~7;
840 shift
= header
.cluster_bits
+ header
.l2_bits
;
841 l1_size
= (total_size
+ (1LL << shift
) - 1) >> shift
;
843 header
.l1_table_offset
= cpu_to_be64(header_size
);
844 if (flags
& BLOCK_FLAG_ENCRYPT
) {
845 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
847 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
850 /* write all the data */
851 ret
= blk_pwrite(qcow_blk
, 0, &header
, sizeof(header
));
852 if (ret
!= sizeof(header
)) {
857 ret
= blk_pwrite(qcow_blk
, sizeof(header
),
858 backing_file
, backing_filename_len
);
859 if (ret
!= backing_filename_len
) {
864 tmp
= g_malloc0(BDRV_SECTOR_SIZE
);
865 for (i
= 0; i
< ((sizeof(uint64_t)*l1_size
+ BDRV_SECTOR_SIZE
- 1)/
866 BDRV_SECTOR_SIZE
); i
++) {
867 ret
= blk_pwrite(qcow_blk
, header_size
+
868 BDRV_SECTOR_SIZE
*i
, tmp
, BDRV_SECTOR_SIZE
);
869 if (ret
!= BDRV_SECTOR_SIZE
) {
880 g_free(backing_file
);
884 static int qcow_make_empty(BlockDriverState
*bs
)
886 BDRVQcowState
*s
= bs
->opaque
;
887 uint32_t l1_length
= s
->l1_size
* sizeof(uint64_t);
890 memset(s
->l1_table
, 0, l1_length
);
891 if (bdrv_pwrite_sync(bs
->file
->bs
, s
->l1_table_offset
, s
->l1_table
,
894 ret
= bdrv_truncate(bs
->file
->bs
, s
->l1_table_offset
+ l1_length
);
898 memset(s
->l2_cache
, 0, s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
899 memset(s
->l2_cache_offsets
, 0, L2_CACHE_SIZE
* sizeof(uint64_t));
900 memset(s
->l2_cache_counts
, 0, L2_CACHE_SIZE
* sizeof(uint32_t));
905 /* XXX: put compressed sectors first, then all the cluster aligned
906 tables to avoid losing bytes in alignment */
907 static int qcow_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
908 const uint8_t *buf
, int nb_sectors
)
910 BDRVQcowState
*s
= bs
->opaque
;
914 uint64_t cluster_offset
;
916 if (nb_sectors
!= s
->cluster_sectors
) {
919 /* Zero-pad last write if image size is not cluster aligned */
920 if (sector_num
+ nb_sectors
== bs
->total_sectors
&&
921 nb_sectors
< s
->cluster_sectors
) {
922 uint8_t *pad_buf
= qemu_blockalign(bs
, s
->cluster_size
);
923 memset(pad_buf
, 0, s
->cluster_size
);
924 memcpy(pad_buf
, buf
, nb_sectors
* BDRV_SECTOR_SIZE
);
925 ret
= qcow_write_compressed(bs
, sector_num
,
926 pad_buf
, s
->cluster_sectors
);
932 out_buf
= g_malloc(s
->cluster_size
+ (s
->cluster_size
/ 1000) + 128);
934 /* best compression, small window, no zlib header */
935 memset(&strm
, 0, sizeof(strm
));
936 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
938 9, Z_DEFAULT_STRATEGY
);
944 strm
.avail_in
= s
->cluster_size
;
945 strm
.next_in
= (uint8_t *)buf
;
946 strm
.avail_out
= s
->cluster_size
;
947 strm
.next_out
= out_buf
;
949 ret
= deflate(&strm
, Z_FINISH
);
950 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
955 out_len
= strm
.next_out
- out_buf
;
959 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
960 /* could not compress: write normal cluster */
961 ret
= bdrv_write(bs
, sector_num
, buf
, s
->cluster_sectors
);
966 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 2,
968 if (cluster_offset
== 0) {
973 cluster_offset
&= s
->cluster_offset_mask
;
974 ret
= bdrv_pwrite(bs
->file
->bs
, cluster_offset
, out_buf
, out_len
);
986 static int qcow_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
988 BDRVQcowState
*s
= bs
->opaque
;
989 bdi
->cluster_size
= s
->cluster_size
;
993 static QemuOptsList qcow_create_opts
= {
994 .name
= "qcow-create-opts",
995 .head
= QTAILQ_HEAD_INITIALIZER(qcow_create_opts
.head
),
998 .name
= BLOCK_OPT_SIZE
,
999 .type
= QEMU_OPT_SIZE
,
1000 .help
= "Virtual disk size"
1003 .name
= BLOCK_OPT_BACKING_FILE
,
1004 .type
= QEMU_OPT_STRING
,
1005 .help
= "File name of a base image"
1008 .name
= BLOCK_OPT_ENCRYPT
,
1009 .type
= QEMU_OPT_BOOL
,
1010 .help
= "Encrypt the image",
1011 .def_value_str
= "off"
1013 { /* end of list */ }
1017 static BlockDriver bdrv_qcow
= {
1018 .format_name
= "qcow",
1019 .instance_size
= sizeof(BDRVQcowState
),
1020 .bdrv_probe
= qcow_probe
,
1021 .bdrv_open
= qcow_open
,
1022 .bdrv_close
= qcow_close
,
1023 .bdrv_reopen_prepare
= qcow_reopen_prepare
,
1024 .bdrv_create
= qcow_create
,
1025 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1026 .supports_backing
= true,
1028 .bdrv_co_readv
= qcow_co_readv
,
1029 .bdrv_co_writev
= qcow_co_writev
,
1030 .bdrv_co_get_block_status
= qcow_co_get_block_status
,
1032 .bdrv_set_key
= qcow_set_key
,
1033 .bdrv_make_empty
= qcow_make_empty
,
1034 .bdrv_write_compressed
= qcow_write_compressed
,
1035 .bdrv_get_info
= qcow_get_info
,
1037 .create_opts
= &qcow_create_opts
,
1040 static void bdrv_qcow_init(void)
1042 bdrv_register(&bdrv_qcow
);
1045 block_init(bdrv_qcow_init
);