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 "qemu/module.h"
29 #include "qapi/qmp/qerror.h"
30 #include "crypto/cipher.h"
31 #include "migration/migration.h"
33 /**************************************************************/
34 /* QEMU COW block driver with compression and encryption support */
36 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
37 #define QCOW_VERSION 1
39 #define QCOW_CRYPT_NONE 0
40 #define QCOW_CRYPT_AES 1
42 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
44 typedef struct QCowHeader
{
47 uint64_t backing_file_offset
;
48 uint32_t backing_file_size
;
50 uint64_t size
; /* in bytes */
54 uint32_t crypt_method
;
55 uint64_t l1_table_offset
;
56 } QEMU_PACKED QCowHeader
;
58 #define L2_CACHE_SIZE 16
60 typedef struct BDRVQcowState
{
67 uint64_t cluster_offset_mask
;
68 uint64_t l1_table_offset
;
71 uint64_t l2_cache_offsets
[L2_CACHE_SIZE
];
72 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
73 uint8_t *cluster_cache
;
74 uint8_t *cluster_data
;
75 uint64_t cluster_cache_offset
;
76 QCryptoCipher
*cipher
; /* NULL if no key yet */
77 uint32_t crypt_method_header
;
79 Error
*migration_blocker
;
82 static int decompress_cluster(BlockDriverState
*bs
, uint64_t cluster_offset
);
84 static int qcow_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
86 const QCowHeader
*cow_header
= (const void *)buf
;
88 if (buf_size
>= sizeof(QCowHeader
) &&
89 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
90 be32_to_cpu(cow_header
->version
) == QCOW_VERSION
)
96 static int qcow_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
99 BDRVQcowState
*s
= bs
->opaque
;
100 unsigned int len
, i
, shift
;
104 ret
= bdrv_pread(bs
->file
->bs
, 0, &header
, sizeof(header
));
108 be32_to_cpus(&header
.magic
);
109 be32_to_cpus(&header
.version
);
110 be64_to_cpus(&header
.backing_file_offset
);
111 be32_to_cpus(&header
.backing_file_size
);
112 be32_to_cpus(&header
.mtime
);
113 be64_to_cpus(&header
.size
);
114 be32_to_cpus(&header
.crypt_method
);
115 be64_to_cpus(&header
.l1_table_offset
);
117 if (header
.magic
!= QCOW_MAGIC
) {
118 error_setg(errp
, "Image not in qcow format");
122 if (header
.version
!= QCOW_VERSION
) {
124 snprintf(version
, sizeof(version
), "QCOW version %" PRIu32
,
126 error_setg(errp
, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
127 bdrv_get_device_or_node_name(bs
), "qcow", version
);
132 if (header
.size
<= 1) {
133 error_setg(errp
, "Image size is too small (must be at least 2 bytes)");
137 if (header
.cluster_bits
< 9 || header
.cluster_bits
> 16) {
138 error_setg(errp
, "Cluster size must be between 512 and 64k");
143 /* l2_bits specifies number of entries; storing a uint64_t in each entry,
144 * so bytes = num_entries << 3. */
145 if (header
.l2_bits
< 9 - 3 || header
.l2_bits
> 16 - 3) {
146 error_setg(errp
, "L2 table size must be between 512 and 64k");
151 if (header
.crypt_method
> QCOW_CRYPT_AES
) {
152 error_setg(errp
, "invalid encryption method in qcow header");
156 if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128
)) {
157 error_setg(errp
, "AES cipher not available");
161 s
->crypt_method_header
= header
.crypt_method
;
162 if (s
->crypt_method_header
) {
165 s
->cluster_bits
= header
.cluster_bits
;
166 s
->cluster_size
= 1 << s
->cluster_bits
;
167 s
->cluster_sectors
= 1 << (s
->cluster_bits
- 9);
168 s
->l2_bits
= header
.l2_bits
;
169 s
->l2_size
= 1 << s
->l2_bits
;
170 bs
->total_sectors
= header
.size
/ 512;
171 s
->cluster_offset_mask
= (1LL << (63 - s
->cluster_bits
)) - 1;
173 /* read the level 1 table */
174 shift
= s
->cluster_bits
+ s
->l2_bits
;
175 if (header
.size
> UINT64_MAX
- (1LL << shift
)) {
176 error_setg(errp
, "Image too large");
180 uint64_t l1_size
= (header
.size
+ (1LL << shift
) - 1) >> shift
;
181 if (l1_size
> INT_MAX
/ sizeof(uint64_t)) {
182 error_setg(errp
, "Image too large");
186 s
->l1_size
= l1_size
;
189 s
->l1_table_offset
= header
.l1_table_offset
;
190 s
->l1_table
= g_try_new(uint64_t, s
->l1_size
);
191 if (s
->l1_table
== NULL
) {
192 error_setg(errp
, "Could not allocate memory for L1 table");
197 ret
= bdrv_pread(bs
->file
->bs
, s
->l1_table_offset
, s
->l1_table
,
198 s
->l1_size
* sizeof(uint64_t));
203 for(i
= 0;i
< s
->l1_size
; i
++) {
204 be64_to_cpus(&s
->l1_table
[i
]);
207 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
209 qemu_try_blockalign(bs
->file
->bs
,
210 s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
211 if (s
->l2_cache
== NULL
) {
212 error_setg(errp
, "Could not allocate L2 table cache");
216 s
->cluster_cache
= g_malloc(s
->cluster_size
);
217 s
->cluster_data
= g_malloc(s
->cluster_size
);
218 s
->cluster_cache_offset
= -1;
220 /* read the backing file name */
221 if (header
.backing_file_offset
!= 0) {
222 len
= header
.backing_file_size
;
223 if (len
> 1023 || len
>= sizeof(bs
->backing_file
)) {
224 error_setg(errp
, "Backing file name too long");
228 ret
= bdrv_pread(bs
->file
->bs
, header
.backing_file_offset
,
229 bs
->backing_file
, len
);
233 bs
->backing_file
[len
] = '\0';
236 /* Disable migration when qcow images are used */
237 error_setg(&s
->migration_blocker
, "The qcow format used by node '%s' "
238 "does not support live migration",
239 bdrv_get_device_or_node_name(bs
));
240 migrate_add_blocker(s
->migration_blocker
);
242 qemu_co_mutex_init(&s
->lock
);
247 qemu_vfree(s
->l2_cache
);
248 g_free(s
->cluster_cache
);
249 g_free(s
->cluster_data
);
254 /* We have nothing to do for QCOW reopen, stubs just return
256 static int qcow_reopen_prepare(BDRVReopenState
*state
,
257 BlockReopenQueue
*queue
, Error
**errp
)
262 static int qcow_set_key(BlockDriverState
*bs
, const char *key
)
264 BDRVQcowState
*s
= bs
->opaque
;
269 memset(keybuf
, 0, 16);
273 /* XXX: we could compress the chars to 7 bits to increase
275 for(i
= 0;i
< len
;i
++) {
278 assert(bs
->encrypted
);
280 qcrypto_cipher_free(s
->cipher
);
281 s
->cipher
= qcrypto_cipher_new(
282 QCRYPTO_CIPHER_ALG_AES_128
,
283 QCRYPTO_CIPHER_MODE_CBC
,
284 keybuf
, G_N_ELEMENTS(keybuf
),
288 /* XXX would be nice if errors in this method could
289 * be properly propagate to the caller. Would need
290 * the bdrv_set_key() API signature to be fixed. */
297 /* The crypt function is compatible with the linux cryptoloop
298 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
300 static int encrypt_sectors(BDRVQcowState
*s
, int64_t sector_num
,
301 uint8_t *out_buf
, const uint8_t *in_buf
,
302 int nb_sectors
, bool enc
, Error
**errp
)
311 for(i
= 0; i
< nb_sectors
; i
++) {
312 ivec
.ll
[0] = cpu_to_le64(sector_num
);
314 if (qcrypto_cipher_setiv(s
->cipher
,
315 ivec
.b
, G_N_ELEMENTS(ivec
.b
),
320 ret
= qcrypto_cipher_encrypt(s
->cipher
,
326 ret
= qcrypto_cipher_decrypt(s
->cipher
,
346 * 1 to allocate a normal cluster (for sector indexes 'n_start' to
349 * 2 to allocate a compressed cluster of size
350 * 'compressed_size'. 'compressed_size' must be > 0 and <
353 * return 0 if not allocated.
355 static uint64_t get_cluster_offset(BlockDriverState
*bs
,
356 uint64_t offset
, int allocate
,
358 int n_start
, int n_end
)
360 BDRVQcowState
*s
= bs
->opaque
;
361 int min_index
, i
, j
, l1_index
, l2_index
;
362 uint64_t l2_offset
, *l2_table
, cluster_offset
, tmp
;
366 l1_index
= offset
>> (s
->l2_bits
+ s
->cluster_bits
);
367 l2_offset
= s
->l1_table
[l1_index
];
372 /* allocate a new l2 entry */
373 l2_offset
= bdrv_getlength(bs
->file
->bs
);
374 /* round to cluster size */
375 l2_offset
= (l2_offset
+ s
->cluster_size
- 1) & ~(s
->cluster_size
- 1);
376 /* update the L1 entry */
377 s
->l1_table
[l1_index
] = l2_offset
;
378 tmp
= cpu_to_be64(l2_offset
);
379 if (bdrv_pwrite_sync(bs
->file
->bs
,
380 s
->l1_table_offset
+ l1_index
* sizeof(tmp
),
381 &tmp
, sizeof(tmp
)) < 0)
385 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
386 if (l2_offset
== s
->l2_cache_offsets
[i
]) {
387 /* increment the hit count */
388 if (++s
->l2_cache_counts
[i
] == 0xffffffff) {
389 for(j
= 0; j
< L2_CACHE_SIZE
; j
++) {
390 s
->l2_cache_counts
[j
] >>= 1;
393 l2_table
= s
->l2_cache
+ (i
<< s
->l2_bits
);
397 /* not found: load a new entry in the least used one */
399 min_count
= 0xffffffff;
400 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
401 if (s
->l2_cache_counts
[i
] < min_count
) {
402 min_count
= s
->l2_cache_counts
[i
];
406 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
408 memset(l2_table
, 0, s
->l2_size
* sizeof(uint64_t));
409 if (bdrv_pwrite_sync(bs
->file
->bs
, l2_offset
, l2_table
,
410 s
->l2_size
* sizeof(uint64_t)) < 0)
413 if (bdrv_pread(bs
->file
->bs
, l2_offset
, l2_table
,
414 s
->l2_size
* sizeof(uint64_t)) !=
415 s
->l2_size
* sizeof(uint64_t))
418 s
->l2_cache_offsets
[min_index
] = l2_offset
;
419 s
->l2_cache_counts
[min_index
] = 1;
421 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
422 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
423 if (!cluster_offset
||
424 ((cluster_offset
& QCOW_OFLAG_COMPRESSED
) && allocate
== 1)) {
427 /* allocate a new cluster */
428 if ((cluster_offset
& QCOW_OFLAG_COMPRESSED
) &&
429 (n_end
- n_start
) < s
->cluster_sectors
) {
430 /* if the cluster is already compressed, we must
431 decompress it in the case it is not completely
433 if (decompress_cluster(bs
, cluster_offset
) < 0)
435 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
436 cluster_offset
= (cluster_offset
+ s
->cluster_size
- 1) &
437 ~(s
->cluster_size
- 1);
438 /* write the cluster content */
439 if (bdrv_pwrite(bs
->file
->bs
, cluster_offset
, s
->cluster_cache
,
444 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
446 /* round to cluster size */
447 cluster_offset
= (cluster_offset
+ s
->cluster_size
- 1) &
448 ~(s
->cluster_size
- 1);
449 bdrv_truncate(bs
->file
->bs
, cluster_offset
+ s
->cluster_size
);
450 /* if encrypted, we must initialize the cluster
451 content which won't be written */
453 (n_end
- n_start
) < s
->cluster_sectors
) {
456 start_sect
= (offset
& ~(s
->cluster_size
- 1)) >> 9;
457 memset(s
->cluster_data
+ 512, 0x00, 512);
458 for(i
= 0; i
< s
->cluster_sectors
; i
++) {
459 if (i
< n_start
|| i
>= n_end
) {
461 if (encrypt_sectors(s
, start_sect
+ i
,
463 s
->cluster_data
+ 512, 1,
469 if (bdrv_pwrite(bs
->file
->bs
,
470 cluster_offset
+ i
* 512,
471 s
->cluster_data
, 512) != 512)
476 } else if (allocate
== 2) {
477 cluster_offset
|= QCOW_OFLAG_COMPRESSED
|
478 (uint64_t)compressed_size
<< (63 - s
->cluster_bits
);
481 /* update L2 table */
482 tmp
= cpu_to_be64(cluster_offset
);
483 l2_table
[l2_index
] = tmp
;
484 if (bdrv_pwrite_sync(bs
->file
->bs
, l2_offset
+ l2_index
* sizeof(tmp
),
485 &tmp
, sizeof(tmp
)) < 0)
488 return cluster_offset
;
491 static int64_t coroutine_fn
qcow_co_get_block_status(BlockDriverState
*bs
,
492 int64_t sector_num
, int nb_sectors
, int *pnum
)
494 BDRVQcowState
*s
= bs
->opaque
;
495 int index_in_cluster
, n
;
496 uint64_t cluster_offset
;
498 qemu_co_mutex_lock(&s
->lock
);
499 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 0, 0, 0, 0);
500 qemu_co_mutex_unlock(&s
->lock
);
501 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
502 n
= s
->cluster_sectors
- index_in_cluster
;
506 if (!cluster_offset
) {
509 if ((cluster_offset
& QCOW_OFLAG_COMPRESSED
) || s
->cipher
) {
510 return BDRV_BLOCK_DATA
;
512 cluster_offset
|= (index_in_cluster
<< BDRV_SECTOR_BITS
);
513 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| cluster_offset
;
516 static int decompress_buffer(uint8_t *out_buf
, int out_buf_size
,
517 const uint8_t *buf
, int buf_size
)
519 z_stream strm1
, *strm
= &strm1
;
522 memset(strm
, 0, sizeof(*strm
));
524 strm
->next_in
= (uint8_t *)buf
;
525 strm
->avail_in
= buf_size
;
526 strm
->next_out
= out_buf
;
527 strm
->avail_out
= out_buf_size
;
529 ret
= inflateInit2(strm
, -12);
532 ret
= inflate(strm
, Z_FINISH
);
533 out_len
= strm
->next_out
- out_buf
;
534 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) ||
535 out_len
!= out_buf_size
) {
543 static int decompress_cluster(BlockDriverState
*bs
, uint64_t cluster_offset
)
545 BDRVQcowState
*s
= bs
->opaque
;
549 coffset
= cluster_offset
& s
->cluster_offset_mask
;
550 if (s
->cluster_cache_offset
!= coffset
) {
551 csize
= cluster_offset
>> (63 - s
->cluster_bits
);
552 csize
&= (s
->cluster_size
- 1);
553 ret
= bdrv_pread(bs
->file
->bs
, coffset
, s
->cluster_data
, csize
);
556 if (decompress_buffer(s
->cluster_cache
, s
->cluster_size
,
557 s
->cluster_data
, csize
) < 0) {
560 s
->cluster_cache_offset
= coffset
;
565 static coroutine_fn
int qcow_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
566 int nb_sectors
, QEMUIOVector
*qiov
)
568 BDRVQcowState
*s
= bs
->opaque
;
569 int index_in_cluster
;
571 uint64_t cluster_offset
;
573 QEMUIOVector hd_qiov
;
578 if (qiov
->niov
> 1) {
579 buf
= orig_buf
= qemu_try_blockalign(bs
, qiov
->size
);
585 buf
= (uint8_t *)qiov
->iov
->iov_base
;
588 qemu_co_mutex_lock(&s
->lock
);
590 while (nb_sectors
!= 0) {
591 /* prepare next request */
592 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9,
594 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
595 n
= s
->cluster_sectors
- index_in_cluster
;
596 if (n
> nb_sectors
) {
600 if (!cluster_offset
) {
602 /* read from the base image */
603 hd_iov
.iov_base
= (void *)buf
;
604 hd_iov
.iov_len
= n
* 512;
605 qemu_iovec_init_external(&hd_qiov
, &hd_iov
, 1);
606 qemu_co_mutex_unlock(&s
->lock
);
607 ret
= bdrv_co_readv(bs
->backing
->bs
, sector_num
,
609 qemu_co_mutex_lock(&s
->lock
);
614 /* Note: in this case, no need to wait */
615 memset(buf
, 0, 512 * n
);
617 } else if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
618 /* add AIO support for compressed blocks ? */
619 if (decompress_cluster(bs
, cluster_offset
) < 0) {
623 s
->cluster_cache
+ index_in_cluster
* 512, 512 * n
);
625 if ((cluster_offset
& 511) != 0) {
628 hd_iov
.iov_base
= (void *)buf
;
629 hd_iov
.iov_len
= n
* 512;
630 qemu_iovec_init_external(&hd_qiov
, &hd_iov
, 1);
631 qemu_co_mutex_unlock(&s
->lock
);
632 ret
= bdrv_co_readv(bs
->file
->bs
,
633 (cluster_offset
>> 9) + index_in_cluster
,
635 qemu_co_mutex_lock(&s
->lock
);
641 if (encrypt_sectors(s
, sector_num
, buf
, buf
,
642 n
, false, &err
) < 0) {
655 qemu_co_mutex_unlock(&s
->lock
);
657 if (qiov
->niov
> 1) {
658 qemu_iovec_from_buf(qiov
, 0, orig_buf
, qiov
->size
);
659 qemu_vfree(orig_buf
);
670 static coroutine_fn
int qcow_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
671 int nb_sectors
, QEMUIOVector
*qiov
)
673 BDRVQcowState
*s
= bs
->opaque
;
674 int index_in_cluster
;
675 uint64_t cluster_offset
;
676 const uint8_t *src_buf
;
678 uint8_t *cluster_data
= NULL
;
680 QEMUIOVector hd_qiov
;
684 s
->cluster_cache_offset
= -1; /* disable compressed cache */
686 if (qiov
->niov
> 1) {
687 buf
= orig_buf
= qemu_try_blockalign(bs
, qiov
->size
);
691 qemu_iovec_to_buf(qiov
, 0, buf
, qiov
->size
);
694 buf
= (uint8_t *)qiov
->iov
->iov_base
;
697 qemu_co_mutex_lock(&s
->lock
);
699 while (nb_sectors
!= 0) {
701 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
702 n
= s
->cluster_sectors
- index_in_cluster
;
703 if (n
> nb_sectors
) {
706 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 1, 0,
708 index_in_cluster
+ n
);
709 if (!cluster_offset
|| (cluster_offset
& 511) != 0) {
717 cluster_data
= g_malloc0(s
->cluster_size
);
719 if (encrypt_sectors(s
, sector_num
, cluster_data
, buf
,
720 n
, true, &err
) < 0) {
725 src_buf
= cluster_data
;
730 hd_iov
.iov_base
= (void *)src_buf
;
731 hd_iov
.iov_len
= n
* 512;
732 qemu_iovec_init_external(&hd_qiov
, &hd_iov
, 1);
733 qemu_co_mutex_unlock(&s
->lock
);
734 ret
= bdrv_co_writev(bs
->file
->bs
,
735 (cluster_offset
>> 9) + index_in_cluster
,
737 qemu_co_mutex_lock(&s
->lock
);
747 qemu_co_mutex_unlock(&s
->lock
);
749 if (qiov
->niov
> 1) {
750 qemu_vfree(orig_buf
);
752 g_free(cluster_data
);
757 static void qcow_close(BlockDriverState
*bs
)
759 BDRVQcowState
*s
= bs
->opaque
;
761 qcrypto_cipher_free(s
->cipher
);
764 qemu_vfree(s
->l2_cache
);
765 g_free(s
->cluster_cache
);
766 g_free(s
->cluster_data
);
768 migrate_del_blocker(s
->migration_blocker
);
769 error_free(s
->migration_blocker
);
772 static int qcow_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
774 int header_size
, backing_filename_len
, l1_size
, shift
, i
;
777 int64_t total_size
= 0;
778 char *backing_file
= NULL
;
780 Error
*local_err
= NULL
;
782 BlockDriverState
*qcow_bs
;
784 /* Read out options */
785 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
787 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
788 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_ENCRYPT
, false)) {
789 flags
|= BLOCK_FLAG_ENCRYPT
;
792 ret
= bdrv_create_file(filename
, opts
, &local_err
);
794 error_propagate(errp
, local_err
);
799 ret
= bdrv_open(&qcow_bs
, filename
, NULL
, NULL
,
800 BDRV_O_RDWR
| BDRV_O_PROTOCOL
, &local_err
);
802 error_propagate(errp
, local_err
);
806 ret
= bdrv_truncate(qcow_bs
, 0);
811 memset(&header
, 0, sizeof(header
));
812 header
.magic
= cpu_to_be32(QCOW_MAGIC
);
813 header
.version
= cpu_to_be32(QCOW_VERSION
);
814 header
.size
= cpu_to_be64(total_size
);
815 header_size
= sizeof(header
);
816 backing_filename_len
= 0;
818 if (strcmp(backing_file
, "fat:")) {
819 header
.backing_file_offset
= cpu_to_be64(header_size
);
820 backing_filename_len
= strlen(backing_file
);
821 header
.backing_file_size
= cpu_to_be32(backing_filename_len
);
822 header_size
+= backing_filename_len
;
824 /* special backing file for vvfat */
827 header
.cluster_bits
= 9; /* 512 byte cluster to avoid copying
828 unmodified sectors */
829 header
.l2_bits
= 12; /* 32 KB L2 tables */
831 header
.cluster_bits
= 12; /* 4 KB clusters */
832 header
.l2_bits
= 9; /* 4 KB L2 tables */
834 header_size
= (header_size
+ 7) & ~7;
835 shift
= header
.cluster_bits
+ header
.l2_bits
;
836 l1_size
= (total_size
+ (1LL << shift
) - 1) >> shift
;
838 header
.l1_table_offset
= cpu_to_be64(header_size
);
839 if (flags
& BLOCK_FLAG_ENCRYPT
) {
840 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
842 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
845 /* write all the data */
846 ret
= bdrv_pwrite(qcow_bs
, 0, &header
, sizeof(header
));
847 if (ret
!= sizeof(header
)) {
852 ret
= bdrv_pwrite(qcow_bs
, sizeof(header
),
853 backing_file
, backing_filename_len
);
854 if (ret
!= backing_filename_len
) {
859 tmp
= g_malloc0(BDRV_SECTOR_SIZE
);
860 for (i
= 0; i
< ((sizeof(uint64_t)*l1_size
+ BDRV_SECTOR_SIZE
- 1)/
861 BDRV_SECTOR_SIZE
); i
++) {
862 ret
= bdrv_pwrite(qcow_bs
, header_size
+
863 BDRV_SECTOR_SIZE
*i
, tmp
, BDRV_SECTOR_SIZE
);
864 if (ret
!= BDRV_SECTOR_SIZE
) {
875 g_free(backing_file
);
879 static int qcow_make_empty(BlockDriverState
*bs
)
881 BDRVQcowState
*s
= bs
->opaque
;
882 uint32_t l1_length
= s
->l1_size
* sizeof(uint64_t);
885 memset(s
->l1_table
, 0, l1_length
);
886 if (bdrv_pwrite_sync(bs
->file
->bs
, s
->l1_table_offset
, s
->l1_table
,
889 ret
= bdrv_truncate(bs
->file
->bs
, s
->l1_table_offset
+ l1_length
);
893 memset(s
->l2_cache
, 0, s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
894 memset(s
->l2_cache_offsets
, 0, L2_CACHE_SIZE
* sizeof(uint64_t));
895 memset(s
->l2_cache_counts
, 0, L2_CACHE_SIZE
* sizeof(uint32_t));
900 /* XXX: put compressed sectors first, then all the cluster aligned
901 tables to avoid losing bytes in alignment */
902 static int qcow_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
903 const uint8_t *buf
, int nb_sectors
)
905 BDRVQcowState
*s
= bs
->opaque
;
909 uint64_t cluster_offset
;
911 if (nb_sectors
!= s
->cluster_sectors
) {
914 /* Zero-pad last write if image size is not cluster aligned */
915 if (sector_num
+ nb_sectors
== bs
->total_sectors
&&
916 nb_sectors
< s
->cluster_sectors
) {
917 uint8_t *pad_buf
= qemu_blockalign(bs
, s
->cluster_size
);
918 memset(pad_buf
, 0, s
->cluster_size
);
919 memcpy(pad_buf
, buf
, nb_sectors
* BDRV_SECTOR_SIZE
);
920 ret
= qcow_write_compressed(bs
, sector_num
,
921 pad_buf
, s
->cluster_sectors
);
927 out_buf
= g_malloc(s
->cluster_size
+ (s
->cluster_size
/ 1000) + 128);
929 /* best compression, small window, no zlib header */
930 memset(&strm
, 0, sizeof(strm
));
931 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
933 9, Z_DEFAULT_STRATEGY
);
939 strm
.avail_in
= s
->cluster_size
;
940 strm
.next_in
= (uint8_t *)buf
;
941 strm
.avail_out
= s
->cluster_size
;
942 strm
.next_out
= out_buf
;
944 ret
= deflate(&strm
, Z_FINISH
);
945 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
950 out_len
= strm
.next_out
- out_buf
;
954 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
955 /* could not compress: write normal cluster */
956 ret
= bdrv_write(bs
, sector_num
, buf
, s
->cluster_sectors
);
961 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 2,
963 if (cluster_offset
== 0) {
968 cluster_offset
&= s
->cluster_offset_mask
;
969 ret
= bdrv_pwrite(bs
->file
->bs
, cluster_offset
, out_buf
, out_len
);
981 static int qcow_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
983 BDRVQcowState
*s
= bs
->opaque
;
984 bdi
->cluster_size
= s
->cluster_size
;
988 static QemuOptsList qcow_create_opts
= {
989 .name
= "qcow-create-opts",
990 .head
= QTAILQ_HEAD_INITIALIZER(qcow_create_opts
.head
),
993 .name
= BLOCK_OPT_SIZE
,
994 .type
= QEMU_OPT_SIZE
,
995 .help
= "Virtual disk size"
998 .name
= BLOCK_OPT_BACKING_FILE
,
999 .type
= QEMU_OPT_STRING
,
1000 .help
= "File name of a base image"
1003 .name
= BLOCK_OPT_ENCRYPT
,
1004 .type
= QEMU_OPT_BOOL
,
1005 .help
= "Encrypt the image",
1006 .def_value_str
= "off"
1008 { /* end of list */ }
1012 static BlockDriver bdrv_qcow
= {
1013 .format_name
= "qcow",
1014 .instance_size
= sizeof(BDRVQcowState
),
1015 .bdrv_probe
= qcow_probe
,
1016 .bdrv_open
= qcow_open
,
1017 .bdrv_close
= qcow_close
,
1018 .bdrv_reopen_prepare
= qcow_reopen_prepare
,
1019 .bdrv_create
= qcow_create
,
1020 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1021 .supports_backing
= true,
1023 .bdrv_co_readv
= qcow_co_readv
,
1024 .bdrv_co_writev
= qcow_co_writev
,
1025 .bdrv_co_get_block_status
= qcow_co_get_block_status
,
1027 .bdrv_set_key
= qcow_set_key
,
1028 .bdrv_make_empty
= qcow_make_empty
,
1029 .bdrv_write_compressed
= qcow_write_compressed
,
1030 .bdrv_get_info
= qcow_get_info
,
1032 .create_opts
= &qcow_create_opts
,
1035 static void bdrv_qcow_init(void)
1037 bdrv_register(&bdrv_qcow
);
1040 block_init(bdrv_qcow_init
);