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
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu/error-report.h"
28 #include "block/block_int.h"
29 #include "block/qdict.h"
30 #include "sysemu/block-backend.h"
31 #include "qemu/module.h"
32 #include "qemu/option.h"
33 #include "qemu/bswap.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qstring.h"
37 #include "qapi/qobject-input-visitor.h"
38 #include "qapi/qapi-visit-block-core.h"
39 #include "crypto/block.h"
40 #include "migration/blocker.h"
43 /**************************************************************/
44 /* QEMU COW block driver with compression and encryption support */
46 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
47 #define QCOW_VERSION 1
49 #define QCOW_CRYPT_NONE 0
50 #define QCOW_CRYPT_AES 1
52 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
54 typedef struct QCowHeader
{
57 uint64_t backing_file_offset
;
58 uint32_t backing_file_size
;
60 uint64_t size
; /* in bytes */
64 uint32_t crypt_method
;
65 uint64_t l1_table_offset
;
66 } QEMU_PACKED QCowHeader
;
68 #define L2_CACHE_SIZE 16
70 typedef struct BDRVQcowState
{
76 uint64_t cluster_offset_mask
;
77 uint64_t l1_table_offset
;
80 uint64_t l2_cache_offsets
[L2_CACHE_SIZE
];
81 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
82 uint8_t *cluster_cache
;
83 uint8_t *cluster_data
;
84 uint64_t cluster_cache_offset
;
85 QCryptoBlock
*crypto
; /* Disk encryption format driver */
86 uint32_t crypt_method_header
;
88 Error
*migration_blocker
;
91 static QemuOptsList qcow_create_opts
;
93 static int decompress_cluster(BlockDriverState
*bs
, uint64_t cluster_offset
);
95 static int qcow_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
97 const QCowHeader
*cow_header
= (const void *)buf
;
99 if (buf_size
>= sizeof(QCowHeader
) &&
100 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
101 be32_to_cpu(cow_header
->version
) == QCOW_VERSION
)
107 static QemuOptsList qcow_runtime_opts
= {
109 .head
= QTAILQ_HEAD_INITIALIZER(qcow_runtime_opts
.head
),
111 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
112 { /* end of list */ }
116 static int qcow_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
119 BDRVQcowState
*s
= bs
->opaque
;
120 unsigned int len
, i
, shift
;
123 Error
*local_err
= NULL
;
124 QCryptoBlockOpenOptions
*crypto_opts
= NULL
;
125 unsigned int cflags
= 0;
126 QDict
*encryptopts
= NULL
;
127 const char *encryptfmt
;
129 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
130 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
132 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_file
,
139 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
143 header
.magic
= be32_to_cpu(header
.magic
);
144 header
.version
= be32_to_cpu(header
.version
);
145 header
.backing_file_offset
= be64_to_cpu(header
.backing_file_offset
);
146 header
.backing_file_size
= be32_to_cpu(header
.backing_file_size
);
147 header
.mtime
= be32_to_cpu(header
.mtime
);
148 header
.size
= be64_to_cpu(header
.size
);
149 header
.crypt_method
= be32_to_cpu(header
.crypt_method
);
150 header
.l1_table_offset
= be64_to_cpu(header
.l1_table_offset
);
152 if (header
.magic
!= QCOW_MAGIC
) {
153 error_setg(errp
, "Image not in qcow format");
157 if (header
.version
!= QCOW_VERSION
) {
158 error_setg(errp
, "Unsupported qcow version %" PRIu32
, header
.version
);
163 if (header
.size
<= 1) {
164 error_setg(errp
, "Image size is too small (must be at least 2 bytes)");
168 if (header
.cluster_bits
< 9 || header
.cluster_bits
> 16) {
169 error_setg(errp
, "Cluster size must be between 512 and 64k");
174 /* l2_bits specifies number of entries; storing a uint64_t in each entry,
175 * so bytes = num_entries << 3. */
176 if (header
.l2_bits
< 9 - 3 || header
.l2_bits
> 16 - 3) {
177 error_setg(errp
, "L2 table size must be between 512 and 64k");
182 s
->crypt_method_header
= header
.crypt_method
;
183 if (s
->crypt_method_header
) {
184 if (bdrv_uses_whitelist() &&
185 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
187 "Use of AES-CBC encrypted qcow images is no longer "
188 "supported in system emulators");
189 error_append_hint(errp
,
190 "You can use 'qemu-img convert' to convert your "
191 "image to an alternative supported format, such "
192 "as unencrypted qcow, or raw with the LUKS "
193 "format instead.\n");
197 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
198 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
200 "Header reported 'aes' encryption format but "
201 "options specify '%s'", encryptfmt
);
205 qdict_put_str(encryptopts
, "format", "qcow");
206 crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
212 if (flags
& BDRV_O_NO_IO
) {
213 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
215 s
->crypto
= qcrypto_block_open(crypto_opts
, "encrypt.",
216 NULL
, NULL
, cflags
, 1, errp
);
222 error_setg(errp
, "invalid encryption method in qcow header");
226 bs
->encrypted
= true;
229 error_setg(errp
, "No encryption in image header, but options "
230 "specified format '%s'", encryptfmt
);
235 s
->cluster_bits
= header
.cluster_bits
;
236 s
->cluster_size
= 1 << s
->cluster_bits
;
237 s
->l2_bits
= header
.l2_bits
;
238 s
->l2_size
= 1 << s
->l2_bits
;
239 bs
->total_sectors
= header
.size
/ 512;
240 s
->cluster_offset_mask
= (1LL << (63 - s
->cluster_bits
)) - 1;
242 /* read the level 1 table */
243 shift
= s
->cluster_bits
+ s
->l2_bits
;
244 if (header
.size
> UINT64_MAX
- (1LL << shift
)) {
245 error_setg(errp
, "Image too large");
249 uint64_t l1_size
= (header
.size
+ (1LL << shift
) - 1) >> shift
;
250 if (l1_size
> INT_MAX
/ sizeof(uint64_t)) {
251 error_setg(errp
, "Image too large");
255 s
->l1_size
= l1_size
;
258 s
->l1_table_offset
= header
.l1_table_offset
;
259 s
->l1_table
= g_try_new(uint64_t, s
->l1_size
);
260 if (s
->l1_table
== NULL
) {
261 error_setg(errp
, "Could not allocate memory for L1 table");
266 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
267 s
->l1_size
* sizeof(uint64_t));
272 for(i
= 0;i
< s
->l1_size
; i
++) {
273 s
->l1_table
[i
] = be64_to_cpu(s
->l1_table
[i
]);
276 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
278 qemu_try_blockalign(bs
->file
->bs
,
279 s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
280 if (s
->l2_cache
== NULL
) {
281 error_setg(errp
, "Could not allocate L2 table cache");
285 s
->cluster_cache
= g_malloc(s
->cluster_size
);
286 s
->cluster_data
= g_malloc(s
->cluster_size
);
287 s
->cluster_cache_offset
= -1;
289 /* read the backing file name */
290 if (header
.backing_file_offset
!= 0) {
291 len
= header
.backing_file_size
;
292 if (len
> 1023 || len
>= sizeof(bs
->backing_file
)) {
293 error_setg(errp
, "Backing file name too long");
297 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
298 bs
->backing_file
, len
);
302 bs
->backing_file
[len
] = '\0';
305 /* Disable migration when qcow images are used */
306 error_setg(&s
->migration_blocker
, "The qcow format used by node '%s' "
307 "does not support live migration",
308 bdrv_get_device_or_node_name(bs
));
309 ret
= migrate_add_blocker(s
->migration_blocker
, &local_err
);
311 error_propagate(errp
, local_err
);
312 error_free(s
->migration_blocker
);
316 qobject_unref(encryptopts
);
317 qapi_free_QCryptoBlockOpenOptions(crypto_opts
);
318 qemu_co_mutex_init(&s
->lock
);
323 qemu_vfree(s
->l2_cache
);
324 g_free(s
->cluster_cache
);
325 g_free(s
->cluster_data
);
326 qcrypto_block_free(s
->crypto
);
327 qobject_unref(encryptopts
);
328 qapi_free_QCryptoBlockOpenOptions(crypto_opts
);
333 /* We have nothing to do for QCOW reopen, stubs just return
335 static int qcow_reopen_prepare(BDRVReopenState
*state
,
336 BlockReopenQueue
*queue
, Error
**errp
)
346 * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
347 * to 'n_end' within the cluster)
349 * 2 to allocate a compressed cluster of size
350 * 'compressed_size'. 'compressed_size' must be > 0 and <
353 * return 0 if not allocated, 1 if *result is assigned, and negative
356 static int get_cluster_offset(BlockDriverState
*bs
,
357 uint64_t offset
, int allocate
,
359 int n_start
, int n_end
, uint64_t *result
)
361 BDRVQcowState
*s
= bs
->opaque
;
362 int min_index
, i
, j
, l1_index
, l2_index
, ret
;
364 uint64_t *l2_table
, cluster_offset
, tmp
;
369 l1_index
= offset
>> (s
->l2_bits
+ s
->cluster_bits
);
370 l2_offset
= s
->l1_table
[l1_index
];
375 /* allocate a new l2 entry */
376 l2_offset
= bdrv_getlength(bs
->file
->bs
);
380 /* round to cluster size */
381 l2_offset
= QEMU_ALIGN_UP(l2_offset
, s
->cluster_size
);
382 /* update the L1 entry */
383 s
->l1_table
[l1_index
] = l2_offset
;
384 tmp
= cpu_to_be64(l2_offset
);
385 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
386 ret
= bdrv_pwrite_sync(bs
->file
,
387 s
->l1_table_offset
+ l1_index
* sizeof(tmp
),
394 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
395 if (l2_offset
== s
->l2_cache_offsets
[i
]) {
396 /* increment the hit count */
397 if (++s
->l2_cache_counts
[i
] == 0xffffffff) {
398 for(j
= 0; j
< L2_CACHE_SIZE
; j
++) {
399 s
->l2_cache_counts
[j
] >>= 1;
402 l2_table
= s
->l2_cache
+ (i
<< s
->l2_bits
);
406 /* not found: load a new entry in the least used one */
408 min_count
= 0xffffffff;
409 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
410 if (s
->l2_cache_counts
[i
] < min_count
) {
411 min_count
= s
->l2_cache_counts
[i
];
415 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
416 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_LOAD
);
418 memset(l2_table
, 0, s
->l2_size
* sizeof(uint64_t));
419 ret
= bdrv_pwrite_sync(bs
->file
, l2_offset
, l2_table
,
420 s
->l2_size
* sizeof(uint64_t));
425 ret
= bdrv_pread(bs
->file
, l2_offset
, l2_table
,
426 s
->l2_size
* sizeof(uint64_t));
431 s
->l2_cache_offsets
[min_index
] = l2_offset
;
432 s
->l2_cache_counts
[min_index
] = 1;
434 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
435 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
436 if (!cluster_offset
||
437 ((cluster_offset
& QCOW_OFLAG_COMPRESSED
) && allocate
== 1)) {
440 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC
);
441 assert(QEMU_IS_ALIGNED(n_start
| n_end
, BDRV_SECTOR_SIZE
));
442 /* allocate a new cluster */
443 if ((cluster_offset
& QCOW_OFLAG_COMPRESSED
) &&
444 (n_end
- n_start
) < s
->cluster_size
) {
445 /* if the cluster is already compressed, we must
446 decompress it in the case it is not completely
448 if (decompress_cluster(bs
, cluster_offset
) < 0) {
451 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
452 if ((int64_t) cluster_offset
< 0) {
453 return cluster_offset
;
455 cluster_offset
= QEMU_ALIGN_UP(cluster_offset
, s
->cluster_size
);
456 /* write the cluster content */
457 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
458 ret
= bdrv_pwrite(bs
->file
, cluster_offset
, s
->cluster_cache
,
464 cluster_offset
= bdrv_getlength(bs
->file
->bs
);
465 if ((int64_t) cluster_offset
< 0) {
466 return cluster_offset
;
469 /* round to cluster size */
470 cluster_offset
= QEMU_ALIGN_UP(cluster_offset
, s
->cluster_size
);
471 if (cluster_offset
+ s
->cluster_size
> INT64_MAX
) {
474 ret
= bdrv_truncate(bs
->file
, cluster_offset
+ s
->cluster_size
,
475 PREALLOC_MODE_OFF
, NULL
);
479 /* if encrypted, we must initialize the cluster
480 content which won't be written */
482 (n_end
- n_start
) < s
->cluster_size
) {
483 uint64_t start_offset
;
485 start_offset
= offset
& ~(s
->cluster_size
- 1);
486 for (i
= 0; i
< s
->cluster_size
; i
+= BDRV_SECTOR_SIZE
) {
487 if (i
< n_start
|| i
>= n_end
) {
488 memset(s
->cluster_data
, 0x00, BDRV_SECTOR_SIZE
);
489 if (qcrypto_block_encrypt(s
->crypto
,
496 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
497 ret
= bdrv_pwrite(bs
->file
,
507 } else if (allocate
== 2) {
508 cluster_offset
|= QCOW_OFLAG_COMPRESSED
|
509 (uint64_t)compressed_size
<< (63 - s
->cluster_bits
);
512 /* update L2 table */
513 tmp
= cpu_to_be64(cluster_offset
);
514 l2_table
[l2_index
] = tmp
;
516 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_UPDATE_COMPRESSED
);
518 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_UPDATE
);
520 ret
= bdrv_pwrite_sync(bs
->file
, l2_offset
+ l2_index
* sizeof(tmp
),
526 *result
= cluster_offset
;
530 static int coroutine_fn
qcow_co_block_status(BlockDriverState
*bs
,
532 int64_t offset
, int64_t bytes
,
533 int64_t *pnum
, int64_t *map
,
534 BlockDriverState
**file
)
536 BDRVQcowState
*s
= bs
->opaque
;
537 int index_in_cluster
, ret
;
539 uint64_t cluster_offset
;
541 qemu_co_mutex_lock(&s
->lock
);
542 ret
= get_cluster_offset(bs
, offset
, 0, 0, 0, 0, &cluster_offset
);
543 qemu_co_mutex_unlock(&s
->lock
);
547 index_in_cluster
= offset
& (s
->cluster_size
- 1);
548 n
= s
->cluster_size
- index_in_cluster
;
553 if (!cluster_offset
) {
556 if ((cluster_offset
& QCOW_OFLAG_COMPRESSED
) || s
->crypto
) {
557 return BDRV_BLOCK_DATA
;
559 *map
= cluster_offset
| index_in_cluster
;
560 *file
= bs
->file
->bs
;
561 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
564 static int decompress_buffer(uint8_t *out_buf
, int out_buf_size
,
565 const uint8_t *buf
, int buf_size
)
567 z_stream strm1
, *strm
= &strm1
;
570 memset(strm
, 0, sizeof(*strm
));
572 strm
->next_in
= (uint8_t *)buf
;
573 strm
->avail_in
= buf_size
;
574 strm
->next_out
= out_buf
;
575 strm
->avail_out
= out_buf_size
;
577 ret
= inflateInit2(strm
, -12);
580 ret
= inflate(strm
, Z_FINISH
);
581 out_len
= strm
->next_out
- out_buf
;
582 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) ||
583 out_len
!= out_buf_size
) {
591 static int decompress_cluster(BlockDriverState
*bs
, uint64_t cluster_offset
)
593 BDRVQcowState
*s
= bs
->opaque
;
597 coffset
= cluster_offset
& s
->cluster_offset_mask
;
598 if (s
->cluster_cache_offset
!= coffset
) {
599 csize
= cluster_offset
>> (63 - s
->cluster_bits
);
600 csize
&= (s
->cluster_size
- 1);
601 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
602 ret
= bdrv_pread(bs
->file
, coffset
, s
->cluster_data
, csize
);
605 if (decompress_buffer(s
->cluster_cache
, s
->cluster_size
,
606 s
->cluster_data
, csize
) < 0) {
609 s
->cluster_cache_offset
= coffset
;
614 static void qcow_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
616 /* At least encrypted images require 512-byte alignment. Apply the
617 * limit universally, rather than just on encrypted images, as
618 * it's easier to let the block layer handle rounding than to
619 * audit this code further. */
620 bs
->bl
.request_alignment
= BDRV_SECTOR_SIZE
;
623 static coroutine_fn
int qcow_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
624 uint64_t bytes
, QEMUIOVector
*qiov
,
627 BDRVQcowState
*s
= bs
->opaque
;
628 int offset_in_cluster
;
630 uint64_t cluster_offset
;
631 QEMUIOVector hd_qiov
;
636 if (qiov
->niov
> 1) {
637 buf
= orig_buf
= qemu_try_blockalign(bs
, qiov
->size
);
643 buf
= (uint8_t *)qiov
->iov
->iov_base
;
646 qemu_co_mutex_lock(&s
->lock
);
649 /* prepare next request */
650 ret
= get_cluster_offset(bs
, offset
, 0, 0, 0, 0, &cluster_offset
);
654 offset_in_cluster
= offset
& (s
->cluster_size
- 1);
655 n
= s
->cluster_size
- offset_in_cluster
;
660 if (!cluster_offset
) {
662 /* read from the base image */
663 qemu_iovec_init_buf(&hd_qiov
, buf
, n
);
664 qemu_co_mutex_unlock(&s
->lock
);
665 /* qcow2 emits this on bs->file instead of bs->backing */
666 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
667 ret
= bdrv_co_preadv(bs
->backing
, offset
, n
, &hd_qiov
, 0);
668 qemu_co_mutex_lock(&s
->lock
);
673 /* Note: in this case, no need to wait */
676 } else if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
677 /* add AIO support for compressed blocks ? */
678 if (decompress_cluster(bs
, cluster_offset
) < 0) {
682 memcpy(buf
, s
->cluster_cache
+ offset_in_cluster
, n
);
684 if ((cluster_offset
& 511) != 0) {
688 qemu_iovec_init_buf(&hd_qiov
, buf
, n
);
689 qemu_co_mutex_unlock(&s
->lock
);
690 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
691 ret
= bdrv_co_preadv(bs
->file
, cluster_offset
+ offset_in_cluster
,
693 qemu_co_mutex_lock(&s
->lock
);
699 if (qcrypto_block_decrypt(s
->crypto
,
700 offset
, buf
, n
, NULL
) < 0) {
713 qemu_co_mutex_unlock(&s
->lock
);
715 if (qiov
->niov
> 1) {
716 qemu_iovec_from_buf(qiov
, 0, orig_buf
, qiov
->size
);
717 qemu_vfree(orig_buf
);
723 static coroutine_fn
int qcow_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
724 uint64_t bytes
, QEMUIOVector
*qiov
,
727 BDRVQcowState
*s
= bs
->opaque
;
728 int offset_in_cluster
;
729 uint64_t cluster_offset
;
731 QEMUIOVector hd_qiov
;
736 s
->cluster_cache_offset
= -1; /* disable compressed cache */
738 /* We must always copy the iov when encrypting, so we
739 * don't modify the original data buffer during encryption */
740 if (bs
->encrypted
|| qiov
->niov
> 1) {
741 buf
= orig_buf
= qemu_try_blockalign(bs
, qiov
->size
);
745 qemu_iovec_to_buf(qiov
, 0, buf
, qiov
->size
);
748 buf
= (uint8_t *)qiov
->iov
->iov_base
;
751 qemu_co_mutex_lock(&s
->lock
);
754 offset_in_cluster
= offset
& (s
->cluster_size
- 1);
755 n
= s
->cluster_size
- offset_in_cluster
;
759 ret
= get_cluster_offset(bs
, offset
, 1, 0, offset_in_cluster
,
760 offset_in_cluster
+ n
, &cluster_offset
);
764 if (!cluster_offset
|| (cluster_offset
& 511) != 0) {
770 if (qcrypto_block_encrypt(s
->crypto
, offset
, buf
, n
, NULL
) < 0) {
776 qemu_iovec_init_buf(&hd_qiov
, buf
, n
);
777 qemu_co_mutex_unlock(&s
->lock
);
778 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
779 ret
= bdrv_co_pwritev(bs
->file
, cluster_offset
+ offset_in_cluster
,
781 qemu_co_mutex_lock(&s
->lock
);
791 qemu_co_mutex_unlock(&s
->lock
);
793 qemu_vfree(orig_buf
);
798 static void qcow_close(BlockDriverState
*bs
)
800 BDRVQcowState
*s
= bs
->opaque
;
802 qcrypto_block_free(s
->crypto
);
805 qemu_vfree(s
->l2_cache
);
806 g_free(s
->cluster_cache
);
807 g_free(s
->cluster_data
);
809 migrate_del_blocker(s
->migration_blocker
);
810 error_free(s
->migration_blocker
);
813 static int coroutine_fn
qcow_co_create(BlockdevCreateOptions
*opts
,
816 BlockdevCreateOptionsQcow
*qcow_opts
;
817 int header_size
, backing_filename_len
, l1_size
, shift
, i
;
820 int64_t total_size
= 0;
822 BlockDriverState
*bs
;
823 BlockBackend
*qcow_blk
;
824 QCryptoBlock
*crypto
= NULL
;
826 assert(opts
->driver
== BLOCKDEV_DRIVER_QCOW
);
827 qcow_opts
= &opts
->u
.qcow
;
830 total_size
= qcow_opts
->size
;
831 if (total_size
== 0) {
832 error_setg(errp
, "Image size is too small, cannot be zero length");
836 if (qcow_opts
->has_encrypt
&&
837 qcow_opts
->encrypt
->format
!= Q_CRYPTO_BLOCK_FORMAT_QCOW
)
839 error_setg(errp
, "Unsupported encryption format");
843 /* Create BlockBackend to write to the image */
844 bs
= bdrv_open_blockdev_ref(qcow_opts
->file
, errp
);
849 qcow_blk
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
);
850 ret
= blk_insert_bs(qcow_blk
, bs
, errp
);
854 blk_set_allow_write_beyond_eof(qcow_blk
, true);
856 /* Create image format */
857 ret
= blk_truncate(qcow_blk
, 0, PREALLOC_MODE_OFF
, errp
);
862 memset(&header
, 0, sizeof(header
));
863 header
.magic
= cpu_to_be32(QCOW_MAGIC
);
864 header
.version
= cpu_to_be32(QCOW_VERSION
);
865 header
.size
= cpu_to_be64(total_size
);
866 header_size
= sizeof(header
);
867 backing_filename_len
= 0;
868 if (qcow_opts
->has_backing_file
) {
869 if (strcmp(qcow_opts
->backing_file
, "fat:")) {
870 header
.backing_file_offset
= cpu_to_be64(header_size
);
871 backing_filename_len
= strlen(qcow_opts
->backing_file
);
872 header
.backing_file_size
= cpu_to_be32(backing_filename_len
);
873 header_size
+= backing_filename_len
;
875 /* special backing file for vvfat */
876 qcow_opts
->has_backing_file
= false;
878 header
.cluster_bits
= 9; /* 512 byte cluster to avoid copying
879 unmodified sectors */
880 header
.l2_bits
= 12; /* 32 KB L2 tables */
882 header
.cluster_bits
= 12; /* 4 KB clusters */
883 header
.l2_bits
= 9; /* 4 KB L2 tables */
885 header_size
= (header_size
+ 7) & ~7;
886 shift
= header
.cluster_bits
+ header
.l2_bits
;
887 l1_size
= (total_size
+ (1LL << shift
) - 1) >> shift
;
889 header
.l1_table_offset
= cpu_to_be64(header_size
);
891 if (qcow_opts
->has_encrypt
) {
892 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
894 crypto
= qcrypto_block_create(qcow_opts
->encrypt
, "encrypt.",
895 NULL
, NULL
, NULL
, errp
);
901 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
904 /* write all the data */
905 ret
= blk_pwrite(qcow_blk
, 0, &header
, sizeof(header
), 0);
906 if (ret
!= sizeof(header
)) {
910 if (qcow_opts
->has_backing_file
) {
911 ret
= blk_pwrite(qcow_blk
, sizeof(header
),
912 qcow_opts
->backing_file
, backing_filename_len
, 0);
913 if (ret
!= backing_filename_len
) {
918 tmp
= g_malloc0(BDRV_SECTOR_SIZE
);
919 for (i
= 0; i
< DIV_ROUND_UP(sizeof(uint64_t) * l1_size
, BDRV_SECTOR_SIZE
);
921 ret
= blk_pwrite(qcow_blk
, header_size
+ BDRV_SECTOR_SIZE
* i
,
922 tmp
, BDRV_SECTOR_SIZE
, 0);
923 if (ret
!= BDRV_SECTOR_SIZE
) {
934 qcrypto_block_free(crypto
);
938 static int coroutine_fn
qcow_co_create_opts(const char *filename
,
939 QemuOpts
*opts
, Error
**errp
)
941 BlockdevCreateOptions
*create_options
= NULL
;
942 BlockDriverState
*bs
= NULL
;
946 Error
*local_err
= NULL
;
949 static const QDictRenames opt_renames
[] = {
950 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
951 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
955 /* Parse options and convert legacy syntax */
956 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, &qcow_create_opts
, true);
958 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
959 if (val
&& !strcmp(val
, "on")) {
960 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
961 } else if (val
&& !strcmp(val
, "off")) {
962 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
965 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
966 if (val
&& !strcmp(val
, "aes")) {
967 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
970 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
975 /* Create and open the file (protocol layer) */
976 ret
= bdrv_create_file(filename
, opts
, &local_err
);
978 error_propagate(errp
, local_err
);
982 bs
= bdrv_open(filename
, NULL
, NULL
,
983 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
989 /* Now get the QAPI type BlockdevCreateOptions */
990 qdict_put_str(qdict
, "driver", "qcow");
991 qdict_put_str(qdict
, "file", bs
->node_name
);
993 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
999 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, &local_err
);
1003 error_propagate(errp
, local_err
);
1008 /* Silently round up size */
1009 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW
);
1010 create_options
->u
.qcow
.size
=
1011 ROUND_UP(create_options
->u
.qcow
.size
, BDRV_SECTOR_SIZE
);
1013 /* Create the qcow image (format layer) */
1014 ret
= qcow_co_create(create_options
, errp
);
1021 qobject_unref(qdict
);
1023 qapi_free_BlockdevCreateOptions(create_options
);
1027 static int qcow_make_empty(BlockDriverState
*bs
)
1029 BDRVQcowState
*s
= bs
->opaque
;
1030 uint32_t l1_length
= s
->l1_size
* sizeof(uint64_t);
1033 memset(s
->l1_table
, 0, l1_length
);
1034 if (bdrv_pwrite_sync(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1037 ret
= bdrv_truncate(bs
->file
, s
->l1_table_offset
+ l1_length
,
1038 PREALLOC_MODE_OFF
, NULL
);
1042 memset(s
->l2_cache
, 0, s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
1043 memset(s
->l2_cache_offsets
, 0, L2_CACHE_SIZE
* sizeof(uint64_t));
1044 memset(s
->l2_cache_counts
, 0, L2_CACHE_SIZE
* sizeof(uint32_t));
1049 /* XXX: put compressed sectors first, then all the cluster aligned
1050 tables to avoid losing bytes in alignment */
1051 static coroutine_fn
int
1052 qcow_co_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
1053 uint64_t bytes
, QEMUIOVector
*qiov
)
1055 BDRVQcowState
*s
= bs
->opaque
;
1056 QEMUIOVector hd_qiov
;
1059 uint8_t *buf
, *out_buf
;
1060 uint64_t cluster_offset
;
1062 buf
= qemu_blockalign(bs
, s
->cluster_size
);
1063 if (bytes
!= s
->cluster_size
) {
1064 if (bytes
> s
->cluster_size
||
1065 offset
+ bytes
!= bs
->total_sectors
<< BDRV_SECTOR_BITS
)
1070 /* Zero-pad last write if image size is not cluster aligned */
1071 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
1073 qemu_iovec_to_buf(qiov
, 0, buf
, qiov
->size
);
1075 out_buf
= g_malloc(s
->cluster_size
);
1077 /* best compression, small window, no zlib header */
1078 memset(&strm
, 0, sizeof(strm
));
1079 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
1081 9, Z_DEFAULT_STRATEGY
);
1087 strm
.avail_in
= s
->cluster_size
;
1088 strm
.next_in
= (uint8_t *)buf
;
1089 strm
.avail_out
= s
->cluster_size
;
1090 strm
.next_out
= out_buf
;
1092 ret
= deflate(&strm
, Z_FINISH
);
1093 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
1098 out_len
= strm
.next_out
- out_buf
;
1102 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
1103 /* could not compress: write normal cluster */
1104 ret
= qcow_co_pwritev(bs
, offset
, bytes
, qiov
, 0);
1110 qemu_co_mutex_lock(&s
->lock
);
1111 ret
= get_cluster_offset(bs
, offset
, 2, out_len
, 0, 0, &cluster_offset
);
1112 qemu_co_mutex_unlock(&s
->lock
);
1116 if (cluster_offset
== 0) {
1120 cluster_offset
&= s
->cluster_offset_mask
;
1122 qemu_iovec_init_buf(&hd_qiov
, out_buf
, out_len
);
1123 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_COMPRESSED
);
1124 ret
= bdrv_co_pwritev(bs
->file
, cluster_offset
, out_len
, &hd_qiov
, 0);
1136 static int qcow_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1138 BDRVQcowState
*s
= bs
->opaque
;
1139 bdi
->cluster_size
= s
->cluster_size
;
1143 static QemuOptsList qcow_create_opts
= {
1144 .name
= "qcow-create-opts",
1145 .head
= QTAILQ_HEAD_INITIALIZER(qcow_create_opts
.head
),
1148 .name
= BLOCK_OPT_SIZE
,
1149 .type
= QEMU_OPT_SIZE
,
1150 .help
= "Virtual disk size"
1153 .name
= BLOCK_OPT_BACKING_FILE
,
1154 .type
= QEMU_OPT_STRING
,
1155 .help
= "File name of a base image"
1158 .name
= BLOCK_OPT_ENCRYPT
,
1159 .type
= QEMU_OPT_BOOL
,
1160 .help
= "Encrypt the image with format 'aes'. (Deprecated "
1161 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)",
1164 .name
= BLOCK_OPT_ENCRYPT_FORMAT
,
1165 .type
= QEMU_OPT_STRING
,
1166 .help
= "Encrypt the image, format choices: 'aes'",
1168 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1169 { /* end of list */ }
1173 static BlockDriver bdrv_qcow
= {
1174 .format_name
= "qcow",
1175 .instance_size
= sizeof(BDRVQcowState
),
1176 .bdrv_probe
= qcow_probe
,
1177 .bdrv_open
= qcow_open
,
1178 .bdrv_close
= qcow_close
,
1179 .bdrv_child_perm
= bdrv_format_default_perms
,
1180 .bdrv_reopen_prepare
= qcow_reopen_prepare
,
1181 .bdrv_co_create
= qcow_co_create
,
1182 .bdrv_co_create_opts
= qcow_co_create_opts
,
1183 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1184 .supports_backing
= true,
1185 .bdrv_refresh_limits
= qcow_refresh_limits
,
1187 .bdrv_co_preadv
= qcow_co_preadv
,
1188 .bdrv_co_pwritev
= qcow_co_pwritev
,
1189 .bdrv_co_block_status
= qcow_co_block_status
,
1191 .bdrv_make_empty
= qcow_make_empty
,
1192 .bdrv_co_pwritev_compressed
= qcow_co_pwritev_compressed
,
1193 .bdrv_get_info
= qcow_get_info
,
1195 .create_opts
= &qcow_create_opts
,
1198 static void bdrv_qcow_init(void)
1200 bdrv_register(&bdrv_qcow
);
1203 block_init(bdrv_qcow_init
);