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
25 #include "block_int.h"
31 Differences with QCOW:
33 - Support for multiple incremental snapshots.
34 - Memory management by reference counts.
35 - Clusters which have a reference count of one have the bit
36 QCOW_OFLAG_COPIED to optimize write performance.
37 - Size of compressed clusters is stored in sectors to reduce bit usage
38 in the cluster offsets.
39 - Support for storing additional data (such as the VM state) in the
41 - If a backing store is used, the cluster size is not constrained
42 (could be backported to QCOW).
43 - L2 tables have always a size of one cluster.
47 //#define DEBUG_ALLOC2
49 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
50 #define QCOW_VERSION 2
52 #define QCOW_CRYPT_NONE 0
53 #define QCOW_CRYPT_AES 1
55 /* indicate that the refcount of the referenced cluster is exactly one. */
56 #define QCOW_OFLAG_COPIED (1LL << 63)
57 /* indicate that the cluster is compressed (they never have the copied flag) */
58 #define QCOW_OFLAG_COMPRESSED (1LL << 62)
60 #define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
63 #define offsetof(type, field) ((size_t) &((type *)0)->field)
66 typedef struct QCowHeader
{
69 uint64_t backing_file_offset
;
70 uint32_t backing_file_size
;
71 uint32_t cluster_bits
;
72 uint64_t size
; /* in bytes */
73 uint32_t crypt_method
;
74 uint32_t l1_size
; /* XXX: save number of clusters instead ? */
75 uint64_t l1_table_offset
;
76 uint64_t refcount_table_offset
;
77 uint32_t refcount_table_clusters
;
78 uint32_t nb_snapshots
;
79 uint64_t snapshots_offset
;
82 typedef struct __attribute__((packed
)) QCowSnapshotHeader
{
83 /* header is 8 byte aligned */
84 uint64_t l1_table_offset
;
93 uint64_t vm_clock_nsec
;
95 uint32_t vm_state_size
;
96 uint32_t extra_data_size
; /* for extension */
97 /* extra data follows */
100 } QCowSnapshotHeader
;
102 #define L2_CACHE_SIZE 16
104 typedef struct QCowSnapshot
{
105 uint64_t l1_table_offset
;
109 uint32_t vm_state_size
;
112 uint64_t vm_clock_nsec
;
115 typedef struct BDRVQcowState
{
116 BlockDriverState
*hd
;
123 int l1_vm_state_index
;
126 uint64_t cluster_offset_mask
;
127 uint64_t l1_table_offset
;
130 uint64_t l2_cache_offsets
[L2_CACHE_SIZE
];
131 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
132 uint8_t *cluster_cache
;
133 uint8_t *cluster_data
;
134 uint64_t cluster_cache_offset
;
136 uint64_t *refcount_table
;
137 uint64_t refcount_table_offset
;
138 uint32_t refcount_table_size
;
139 uint64_t refcount_block_cache_offset
;
140 uint16_t *refcount_block_cache
;
141 int64_t free_cluster_index
;
142 int64_t free_byte_offset
;
144 uint32_t crypt_method
; /* current crypt method, 0 if no key yet */
145 uint32_t crypt_method_header
;
146 AES_KEY aes_encrypt_key
;
147 AES_KEY aes_decrypt_key
;
148 uint64_t snapshots_offset
;
151 QCowSnapshot
*snapshots
;
154 static int decompress_cluster(BDRVQcowState
*s
, uint64_t cluster_offset
);
155 static int qcow_read(BlockDriverState
*bs
, int64_t sector_num
,
156 uint8_t *buf
, int nb_sectors
);
157 static int qcow_read_snapshots(BlockDriverState
*bs
);
158 static void qcow_free_snapshots(BlockDriverState
*bs
);
159 static int refcount_init(BlockDriverState
*bs
);
160 static void refcount_close(BlockDriverState
*bs
);
161 static int get_refcount(BlockDriverState
*bs
, int64_t cluster_index
);
162 static int update_cluster_refcount(BlockDriverState
*bs
,
163 int64_t cluster_index
,
165 static void update_refcount(BlockDriverState
*bs
,
166 int64_t offset
, int64_t length
,
168 static int64_t alloc_clusters(BlockDriverState
*bs
, int64_t size
);
169 static int64_t alloc_bytes(BlockDriverState
*bs
, int size
);
170 static void free_clusters(BlockDriverState
*bs
,
171 int64_t offset
, int64_t size
);
173 static void check_refcounts(BlockDriverState
*bs
);
176 static int qcow_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
178 const QCowHeader
*cow_header
= (const void *)buf
;
180 if (buf_size
>= sizeof(QCowHeader
) &&
181 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
182 be32_to_cpu(cow_header
->version
) == QCOW_VERSION
)
188 static int qcow_open(BlockDriverState
*bs
, const char *filename
, int flags
)
190 BDRVQcowState
*s
= bs
->opaque
;
191 int len
, i
, shift
, ret
;
194 ret
= bdrv_file_open(&s
->hd
, filename
, flags
);
197 if (bdrv_pread(s
->hd
, 0, &header
, sizeof(header
)) != sizeof(header
))
199 be32_to_cpus(&header
.magic
);
200 be32_to_cpus(&header
.version
);
201 be64_to_cpus(&header
.backing_file_offset
);
202 be32_to_cpus(&header
.backing_file_size
);
203 be64_to_cpus(&header
.size
);
204 be32_to_cpus(&header
.cluster_bits
);
205 be32_to_cpus(&header
.crypt_method
);
206 be64_to_cpus(&header
.l1_table_offset
);
207 be32_to_cpus(&header
.l1_size
);
208 be64_to_cpus(&header
.refcount_table_offset
);
209 be32_to_cpus(&header
.refcount_table_clusters
);
210 be64_to_cpus(&header
.snapshots_offset
);
211 be32_to_cpus(&header
.nb_snapshots
);
213 if (header
.magic
!= QCOW_MAGIC
|| header
.version
!= QCOW_VERSION
)
215 if (header
.size
<= 1 ||
216 header
.cluster_bits
< 9 ||
217 header
.cluster_bits
> 16)
219 if (header
.crypt_method
> QCOW_CRYPT_AES
)
221 s
->crypt_method_header
= header
.crypt_method
;
222 if (s
->crypt_method_header
)
224 s
->cluster_bits
= header
.cluster_bits
;
225 s
->cluster_size
= 1 << s
->cluster_bits
;
226 s
->cluster_sectors
= 1 << (s
->cluster_bits
- 9);
227 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
228 s
->l2_size
= 1 << s
->l2_bits
;
229 bs
->total_sectors
= header
.size
/ 512;
230 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
231 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
232 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
233 s
->refcount_table_offset
= header
.refcount_table_offset
;
234 s
->refcount_table_size
=
235 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
237 s
->snapshots_offset
= header
.snapshots_offset
;
238 s
->nb_snapshots
= header
.nb_snapshots
;
240 /* read the level 1 table */
241 s
->l1_size
= header
.l1_size
;
242 shift
= s
->cluster_bits
+ s
->l2_bits
;
243 s
->l1_vm_state_index
= (header
.size
+ (1LL << shift
) - 1) >> shift
;
244 /* the L1 table must contain at least enough entries to put
246 if (s
->l1_size
< s
->l1_vm_state_index
)
248 s
->l1_table_offset
= header
.l1_table_offset
;
249 s
->l1_table
= qemu_malloc(s
->l1_size
* sizeof(uint64_t));
252 if (bdrv_pread(s
->hd
, s
->l1_table_offset
, s
->l1_table
, s
->l1_size
* sizeof(uint64_t)) !=
253 s
->l1_size
* sizeof(uint64_t))
255 for(i
= 0;i
< s
->l1_size
; i
++) {
256 be64_to_cpus(&s
->l1_table
[i
]);
259 s
->l2_cache
= qemu_malloc(s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
262 s
->cluster_cache
= qemu_malloc(s
->cluster_size
);
263 if (!s
->cluster_cache
)
265 /* one more sector for decompressed data alignment */
266 s
->cluster_data
= qemu_malloc(s
->cluster_size
+ 512);
267 if (!s
->cluster_data
)
269 s
->cluster_cache_offset
= -1;
271 if (refcount_init(bs
) < 0)
274 /* read the backing file name */
275 if (header
.backing_file_offset
!= 0) {
276 len
= header
.backing_file_size
;
279 if (bdrv_pread(s
->hd
, header
.backing_file_offset
, bs
->backing_file
, len
) != len
)
281 bs
->backing_file
[len
] = '\0';
283 if (qcow_read_snapshots(bs
) < 0)
292 qcow_free_snapshots(bs
);
294 qemu_free(s
->l1_table
);
295 qemu_free(s
->l2_cache
);
296 qemu_free(s
->cluster_cache
);
297 qemu_free(s
->cluster_data
);
302 static int qcow_set_key(BlockDriverState
*bs
, const char *key
)
304 BDRVQcowState
*s
= bs
->opaque
;
308 memset(keybuf
, 0, 16);
312 /* XXX: we could compress the chars to 7 bits to increase
314 for(i
= 0;i
< len
;i
++) {
317 s
->crypt_method
= s
->crypt_method_header
;
319 if (AES_set_encrypt_key(keybuf
, 128, &s
->aes_encrypt_key
) != 0)
321 if (AES_set_decrypt_key(keybuf
, 128, &s
->aes_decrypt_key
) != 0)
331 AES_encrypt(in
, tmp
, &s
->aes_encrypt_key
);
332 AES_decrypt(tmp
, out
, &s
->aes_decrypt_key
);
333 for(i
= 0; i
< 16; i
++)
334 printf(" %02x", tmp
[i
]);
336 for(i
= 0; i
< 16; i
++)
337 printf(" %02x", out
[i
]);
344 /* The crypt function is compatible with the linux cryptoloop
345 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
347 static void encrypt_sectors(BDRVQcowState
*s
, int64_t sector_num
,
348 uint8_t *out_buf
, const uint8_t *in_buf
,
349 int nb_sectors
, int enc
,
358 for(i
= 0; i
< nb_sectors
; i
++) {
359 ivec
.ll
[0] = cpu_to_le64(sector_num
);
361 AES_cbc_encrypt(in_buf
, out_buf
, 512, key
,
369 static int copy_sectors(BlockDriverState
*bs
, uint64_t start_sect
,
370 uint64_t cluster_offset
, int n_start
, int n_end
)
372 BDRVQcowState
*s
= bs
->opaque
;
378 ret
= qcow_read(bs
, start_sect
+ n_start
, s
->cluster_data
, n
);
381 if (s
->crypt_method
) {
382 encrypt_sectors(s
, start_sect
+ n_start
,
384 s
->cluster_data
, n
, 1,
385 &s
->aes_encrypt_key
);
387 ret
= bdrv_write(s
->hd
, (cluster_offset
>> 9) + n_start
,
394 static void l2_cache_reset(BlockDriverState
*bs
)
396 BDRVQcowState
*s
= bs
->opaque
;
398 memset(s
->l2_cache
, 0, s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
399 memset(s
->l2_cache_offsets
, 0, L2_CACHE_SIZE
* sizeof(uint64_t));
400 memset(s
->l2_cache_counts
, 0, L2_CACHE_SIZE
* sizeof(uint32_t));
403 static inline int l2_cache_new_entry(BlockDriverState
*bs
)
405 BDRVQcowState
*s
= bs
->opaque
;
409 /* find a new entry in the least used one */
411 min_count
= 0xffffffff;
412 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
413 if (s
->l2_cache_counts
[i
] < min_count
) {
414 min_count
= s
->l2_cache_counts
[i
];
421 static int64_t align_offset(int64_t offset
, int n
)
423 offset
= (offset
+ n
- 1) & ~(n
- 1);
427 static int grow_l1_table(BlockDriverState
*bs
, int min_size
)
429 BDRVQcowState
*s
= bs
->opaque
;
430 int new_l1_size
, new_l1_size2
, ret
, i
;
431 uint64_t *new_l1_table
;
432 uint64_t new_l1_table_offset
;
436 new_l1_size
= s
->l1_size
;
437 if (min_size
<= new_l1_size
)
439 while (min_size
> new_l1_size
) {
440 new_l1_size
= (new_l1_size
* 3 + 1) / 2;
443 printf("grow l1_table from %d to %d\n", s
->l1_size
, new_l1_size
);
446 new_l1_size2
= sizeof(uint64_t) * new_l1_size
;
447 new_l1_table
= qemu_mallocz(new_l1_size2
);
450 memcpy(new_l1_table
, s
->l1_table
, s
->l1_size
* sizeof(uint64_t));
452 /* write new table (align to cluster) */
453 new_l1_table_offset
= alloc_clusters(bs
, new_l1_size2
);
455 for(i
= 0; i
< s
->l1_size
; i
++)
456 new_l1_table
[i
] = cpu_to_be64(new_l1_table
[i
]);
457 ret
= bdrv_pwrite(s
->hd
, new_l1_table_offset
, new_l1_table
, new_l1_size2
);
458 if (ret
!= new_l1_size2
)
460 for(i
= 0; i
< s
->l1_size
; i
++)
461 new_l1_table
[i
] = be64_to_cpu(new_l1_table
[i
]);
464 data64
= cpu_to_be64(new_l1_table_offset
);
465 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, l1_table_offset
),
466 &data64
, sizeof(data64
)) != sizeof(data64
))
468 data32
= cpu_to_be32(new_l1_size
);
469 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, l1_size
),
470 &data32
, sizeof(data32
)) != sizeof(data32
))
472 qemu_free(s
->l1_table
);
473 free_clusters(bs
, s
->l1_table_offset
, s
->l1_size
* sizeof(uint64_t));
474 s
->l1_table_offset
= new_l1_table_offset
;
475 s
->l1_table
= new_l1_table
;
476 s
->l1_size
= new_l1_size
;
479 qemu_free(s
->l1_table
);
487 * 1 to allocate a normal cluster (for sector indexes 'n_start' to
490 * 2 to allocate a compressed cluster of size
491 * 'compressed_size'. 'compressed_size' must be > 0 and <
494 * return 0 if not allocated.
496 static uint64_t get_cluster_offset(BlockDriverState
*bs
,
497 uint64_t offset
, int allocate
,
499 int n_start
, int n_end
)
501 BDRVQcowState
*s
= bs
->opaque
;
502 int min_index
, i
, j
, l1_index
, l2_index
, ret
;
503 uint64_t l2_offset
, *l2_table
, cluster_offset
, tmp
, old_l2_offset
;
505 l1_index
= offset
>> (s
->l2_bits
+ s
->cluster_bits
);
506 if (l1_index
>= s
->l1_size
) {
507 /* outside l1 table is allowed: we grow the table if needed */
510 if (grow_l1_table(bs
, l1_index
+ 1) < 0)
513 l2_offset
= s
->l1_table
[l1_index
];
518 old_l2_offset
= l2_offset
;
519 /* allocate a new l2 entry */
520 l2_offset
= alloc_clusters(bs
, s
->l2_size
* sizeof(uint64_t));
521 /* update the L1 entry */
522 s
->l1_table
[l1_index
] = l2_offset
| QCOW_OFLAG_COPIED
;
523 tmp
= cpu_to_be64(l2_offset
| QCOW_OFLAG_COPIED
);
524 if (bdrv_pwrite(s
->hd
, s
->l1_table_offset
+ l1_index
* sizeof(tmp
),
525 &tmp
, sizeof(tmp
)) != sizeof(tmp
))
527 min_index
= l2_cache_new_entry(bs
);
528 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
530 if (old_l2_offset
== 0) {
531 memset(l2_table
, 0, s
->l2_size
* sizeof(uint64_t));
533 if (bdrv_pread(s
->hd
, old_l2_offset
,
534 l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
535 s
->l2_size
* sizeof(uint64_t))
538 if (bdrv_pwrite(s
->hd
, l2_offset
,
539 l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
540 s
->l2_size
* sizeof(uint64_t))
543 if (!(l2_offset
& QCOW_OFLAG_COPIED
)) {
545 free_clusters(bs
, l2_offset
, s
->l2_size
* sizeof(uint64_t));
549 l2_offset
&= ~QCOW_OFLAG_COPIED
;
551 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
552 if (l2_offset
== s
->l2_cache_offsets
[i
]) {
553 /* increment the hit count */
554 if (++s
->l2_cache_counts
[i
] == 0xffffffff) {
555 for(j
= 0; j
< L2_CACHE_SIZE
; j
++) {
556 s
->l2_cache_counts
[j
] >>= 1;
559 l2_table
= s
->l2_cache
+ (i
<< s
->l2_bits
);
563 /* not found: load a new entry in the least used one */
564 min_index
= l2_cache_new_entry(bs
);
565 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
566 if (bdrv_pread(s
->hd
, l2_offset
, l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
567 s
->l2_size
* sizeof(uint64_t))
570 s
->l2_cache_offsets
[min_index
] = l2_offset
;
571 s
->l2_cache_counts
[min_index
] = 1;
573 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
574 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
575 if (!cluster_offset
) {
577 return cluster_offset
;
578 } else if (!(cluster_offset
& QCOW_OFLAG_COPIED
)) {
580 return cluster_offset
;
581 /* free the cluster */
582 if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
584 nb_csectors
= ((cluster_offset
>> s
->csize_shift
) &
586 free_clusters(bs
, (cluster_offset
& s
->cluster_offset_mask
) & ~511,
589 free_clusters(bs
, cluster_offset
, s
->cluster_size
);
592 cluster_offset
&= ~QCOW_OFLAG_COPIED
;
593 return cluster_offset
;
596 /* allocate a new cluster */
597 cluster_offset
= alloc_clusters(bs
, s
->cluster_size
);
599 /* we must initialize the cluster content which won't be
601 if ((n_end
- n_start
) < s
->cluster_sectors
) {
604 start_sect
= (offset
& ~(s
->cluster_size
- 1)) >> 9;
605 ret
= copy_sectors(bs
, start_sect
,
606 cluster_offset
, 0, n_start
);
609 ret
= copy_sectors(bs
, start_sect
,
610 cluster_offset
, n_end
, s
->cluster_sectors
);
614 tmp
= cpu_to_be64(cluster_offset
| QCOW_OFLAG_COPIED
);
617 cluster_offset
= alloc_bytes(bs
, compressed_size
);
618 nb_csectors
= ((cluster_offset
+ compressed_size
- 1) >> 9) -
619 (cluster_offset
>> 9);
620 cluster_offset
|= QCOW_OFLAG_COMPRESSED
|
621 ((uint64_t)nb_csectors
<< s
->csize_shift
);
622 /* compressed clusters never have the copied flag */
623 tmp
= cpu_to_be64(cluster_offset
);
625 /* update L2 table */
626 l2_table
[l2_index
] = tmp
;
627 if (bdrv_pwrite(s
->hd
,
628 l2_offset
+ l2_index
* sizeof(tmp
), &tmp
, sizeof(tmp
)) != sizeof(tmp
))
630 return cluster_offset
;
633 static int qcow_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
634 int nb_sectors
, int *pnum
)
636 BDRVQcowState
*s
= bs
->opaque
;
637 int index_in_cluster
, n
;
638 uint64_t cluster_offset
;
640 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 0, 0, 0, 0);
641 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
642 n
= s
->cluster_sectors
- index_in_cluster
;
646 return (cluster_offset
!= 0);
649 static int decompress_buffer(uint8_t *out_buf
, int out_buf_size
,
650 const uint8_t *buf
, int buf_size
)
652 z_stream strm1
, *strm
= &strm1
;
655 memset(strm
, 0, sizeof(*strm
));
657 strm
->next_in
= (uint8_t *)buf
;
658 strm
->avail_in
= buf_size
;
659 strm
->next_out
= out_buf
;
660 strm
->avail_out
= out_buf_size
;
662 ret
= inflateInit2(strm
, -12);
665 ret
= inflate(strm
, Z_FINISH
);
666 out_len
= strm
->next_out
- out_buf
;
667 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) ||
668 out_len
!= out_buf_size
) {
676 static int decompress_cluster(BDRVQcowState
*s
, uint64_t cluster_offset
)
678 int ret
, csize
, nb_csectors
, sector_offset
;
681 coffset
= cluster_offset
& s
->cluster_offset_mask
;
682 if (s
->cluster_cache_offset
!= coffset
) {
683 nb_csectors
= ((cluster_offset
>> s
->csize_shift
) & s
->csize_mask
) + 1;
684 sector_offset
= coffset
& 511;
685 csize
= nb_csectors
* 512 - sector_offset
;
686 ret
= bdrv_read(s
->hd
, coffset
>> 9, s
->cluster_data
, nb_csectors
);
690 if (decompress_buffer(s
->cluster_cache
, s
->cluster_size
,
691 s
->cluster_data
+ sector_offset
, csize
) < 0) {
694 s
->cluster_cache_offset
= coffset
;
699 /* handle reading after the end of the backing file */
700 static int backing_read1(BlockDriverState
*bs
,
701 int64_t sector_num
, uint8_t *buf
, int nb_sectors
)
704 if ((sector_num
+ nb_sectors
) <= bs
->total_sectors
)
706 if (sector_num
>= bs
->total_sectors
)
709 n1
= bs
->total_sectors
- sector_num
;
710 memset(buf
+ n1
* 512, 0, 512 * (nb_sectors
- n1
));
714 static int qcow_read(BlockDriverState
*bs
, int64_t sector_num
,
715 uint8_t *buf
, int nb_sectors
)
717 BDRVQcowState
*s
= bs
->opaque
;
718 int ret
, index_in_cluster
, n
, n1
;
719 uint64_t cluster_offset
;
721 while (nb_sectors
> 0) {
722 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 0, 0, 0, 0);
723 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
724 n
= s
->cluster_sectors
- index_in_cluster
;
727 if (!cluster_offset
) {
728 if (bs
->backing_hd
) {
729 /* read from the base image */
730 n1
= backing_read1(bs
->backing_hd
, sector_num
, buf
, n
);
732 ret
= bdrv_read(bs
->backing_hd
, sector_num
, buf
, n1
);
737 memset(buf
, 0, 512 * n
);
739 } else if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
740 if (decompress_cluster(s
, cluster_offset
) < 0)
742 memcpy(buf
, s
->cluster_cache
+ index_in_cluster
* 512, 512 * n
);
744 ret
= bdrv_pread(s
->hd
, cluster_offset
+ index_in_cluster
* 512, buf
, n
* 512);
747 if (s
->crypt_method
) {
748 encrypt_sectors(s
, sector_num
, buf
, buf
, n
, 0,
749 &s
->aes_decrypt_key
);
759 static int qcow_write(BlockDriverState
*bs
, int64_t sector_num
,
760 const uint8_t *buf
, int nb_sectors
)
762 BDRVQcowState
*s
= bs
->opaque
;
763 int ret
, index_in_cluster
, n
;
764 uint64_t cluster_offset
;
766 while (nb_sectors
> 0) {
767 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
768 n
= s
->cluster_sectors
- index_in_cluster
;
771 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 1, 0,
773 index_in_cluster
+ n
);
776 if (s
->crypt_method
) {
777 encrypt_sectors(s
, sector_num
, s
->cluster_data
, buf
, n
, 1,
778 &s
->aes_encrypt_key
);
779 ret
= bdrv_pwrite(s
->hd
, cluster_offset
+ index_in_cluster
* 512,
780 s
->cluster_data
, n
* 512);
782 ret
= bdrv_pwrite(s
->hd
, cluster_offset
+ index_in_cluster
* 512, buf
, n
* 512);
790 s
->cluster_cache_offset
= -1; /* disable compressed cache */
794 typedef struct QCowAIOCB
{
795 BlockDriverAIOCB common
;
800 uint64_t cluster_offset
;
801 uint8_t *cluster_data
;
802 BlockDriverAIOCB
*hd_aiocb
;
805 static void qcow_aio_read_cb(void *opaque
, int ret
)
807 QCowAIOCB
*acb
= opaque
;
808 BlockDriverState
*bs
= acb
->common
.bs
;
809 BDRVQcowState
*s
= bs
->opaque
;
810 int index_in_cluster
, n1
;
812 acb
->hd_aiocb
= NULL
;
815 acb
->common
.cb(acb
->common
.opaque
, ret
);
816 qemu_aio_release(acb
);
821 /* post process the read buffer */
822 if (!acb
->cluster_offset
) {
824 } else if (acb
->cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
827 if (s
->crypt_method
) {
828 encrypt_sectors(s
, acb
->sector_num
, acb
->buf
, acb
->buf
,
830 &s
->aes_decrypt_key
);
834 acb
->nb_sectors
-= acb
->n
;
835 acb
->sector_num
+= acb
->n
;
836 acb
->buf
+= acb
->n
* 512;
838 if (acb
->nb_sectors
== 0) {
839 /* request completed */
840 acb
->common
.cb(acb
->common
.opaque
, 0);
841 qemu_aio_release(acb
);
845 /* prepare next AIO request */
846 acb
->cluster_offset
= get_cluster_offset(bs
, acb
->sector_num
<< 9,
848 index_in_cluster
= acb
->sector_num
& (s
->cluster_sectors
- 1);
849 acb
->n
= s
->cluster_sectors
- index_in_cluster
;
850 if (acb
->n
> acb
->nb_sectors
)
851 acb
->n
= acb
->nb_sectors
;
853 if (!acb
->cluster_offset
) {
854 if (bs
->backing_hd
) {
855 /* read from the base image */
856 n1
= backing_read1(bs
->backing_hd
, acb
->sector_num
,
859 acb
->hd_aiocb
= bdrv_aio_read(bs
->backing_hd
, acb
->sector_num
,
860 acb
->buf
, acb
->n
, qcow_aio_read_cb
, acb
);
861 if (acb
->hd_aiocb
== NULL
)
867 /* Note: in this case, no need to wait */
868 memset(acb
->buf
, 0, 512 * acb
->n
);
871 } else if (acb
->cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
872 /* add AIO support for compressed blocks ? */
873 if (decompress_cluster(s
, acb
->cluster_offset
) < 0)
876 s
->cluster_cache
+ index_in_cluster
* 512, 512 * acb
->n
);
879 if ((acb
->cluster_offset
& 511) != 0) {
883 acb
->hd_aiocb
= bdrv_aio_read(s
->hd
,
884 (acb
->cluster_offset
>> 9) + index_in_cluster
,
885 acb
->buf
, acb
->n
, qcow_aio_read_cb
, acb
);
886 if (acb
->hd_aiocb
== NULL
)
891 static QCowAIOCB
*qcow_aio_setup(BlockDriverState
*bs
,
892 int64_t sector_num
, uint8_t *buf
, int nb_sectors
,
893 BlockDriverCompletionFunc
*cb
, void *opaque
)
897 acb
= qemu_aio_get(bs
, cb
, opaque
);
900 acb
->hd_aiocb
= NULL
;
901 acb
->sector_num
= sector_num
;
903 acb
->nb_sectors
= nb_sectors
;
905 acb
->cluster_offset
= 0;
909 static BlockDriverAIOCB
*qcow_aio_read(BlockDriverState
*bs
,
910 int64_t sector_num
, uint8_t *buf
, int nb_sectors
,
911 BlockDriverCompletionFunc
*cb
, void *opaque
)
915 acb
= qcow_aio_setup(bs
, sector_num
, buf
, nb_sectors
, cb
, opaque
);
919 qcow_aio_read_cb(acb
, 0);
923 static void qcow_aio_write_cb(void *opaque
, int ret
)
925 QCowAIOCB
*acb
= opaque
;
926 BlockDriverState
*bs
= acb
->common
.bs
;
927 BDRVQcowState
*s
= bs
->opaque
;
928 int index_in_cluster
;
929 uint64_t cluster_offset
;
930 const uint8_t *src_buf
;
932 acb
->hd_aiocb
= NULL
;
936 acb
->common
.cb(acb
->common
.opaque
, ret
);
937 qemu_aio_release(acb
);
941 acb
->nb_sectors
-= acb
->n
;
942 acb
->sector_num
+= acb
->n
;
943 acb
->buf
+= acb
->n
* 512;
945 if (acb
->nb_sectors
== 0) {
946 /* request completed */
947 acb
->common
.cb(acb
->common
.opaque
, 0);
948 qemu_aio_release(acb
);
952 index_in_cluster
= acb
->sector_num
& (s
->cluster_sectors
- 1);
953 acb
->n
= s
->cluster_sectors
- index_in_cluster
;
954 if (acb
->n
> acb
->nb_sectors
)
955 acb
->n
= acb
->nb_sectors
;
956 cluster_offset
= get_cluster_offset(bs
, acb
->sector_num
<< 9, 1, 0,
958 index_in_cluster
+ acb
->n
);
959 if (!cluster_offset
|| (cluster_offset
& 511) != 0) {
963 if (s
->crypt_method
) {
964 if (!acb
->cluster_data
) {
965 acb
->cluster_data
= qemu_mallocz(s
->cluster_size
);
966 if (!acb
->cluster_data
) {
971 encrypt_sectors(s
, acb
->sector_num
, acb
->cluster_data
, acb
->buf
,
972 acb
->n
, 1, &s
->aes_encrypt_key
);
973 src_buf
= acb
->cluster_data
;
977 acb
->hd_aiocb
= bdrv_aio_write(s
->hd
,
978 (cluster_offset
>> 9) + index_in_cluster
,
980 qcow_aio_write_cb
, acb
);
981 if (acb
->hd_aiocb
== NULL
)
985 static BlockDriverAIOCB
*qcow_aio_write(BlockDriverState
*bs
,
986 int64_t sector_num
, const uint8_t *buf
, int nb_sectors
,
987 BlockDriverCompletionFunc
*cb
, void *opaque
)
989 BDRVQcowState
*s
= bs
->opaque
;
992 s
->cluster_cache_offset
= -1; /* disable compressed cache */
994 acb
= qcow_aio_setup(bs
, sector_num
, (uint8_t*)buf
, nb_sectors
, cb
, opaque
);
998 qcow_aio_write_cb(acb
, 0);
1002 static void qcow_aio_cancel(BlockDriverAIOCB
*blockacb
)
1004 QCowAIOCB
*acb
= (QCowAIOCB
*)blockacb
;
1006 bdrv_aio_cancel(acb
->hd_aiocb
);
1007 qemu_aio_release(acb
);
1010 static void qcow_close(BlockDriverState
*bs
)
1012 BDRVQcowState
*s
= bs
->opaque
;
1013 qemu_free(s
->l1_table
);
1014 qemu_free(s
->l2_cache
);
1015 qemu_free(s
->cluster_cache
);
1016 qemu_free(s
->cluster_data
);
1021 /* XXX: use std qcow open function ? */
1022 typedef struct QCowCreateState
{
1025 uint16_t *refcount_block
;
1026 uint64_t *refcount_table
;
1027 int64_t l1_table_offset
;
1028 int64_t refcount_table_offset
;
1029 int64_t refcount_block_offset
;
1032 static void create_refcount_update(QCowCreateState
*s
,
1033 int64_t offset
, int64_t size
)
1036 int64_t start
, last
, cluster_offset
;
1039 start
= offset
& ~(s
->cluster_size
- 1);
1040 last
= (offset
+ size
- 1) & ~(s
->cluster_size
- 1);
1041 for(cluster_offset
= start
; cluster_offset
<= last
;
1042 cluster_offset
+= s
->cluster_size
) {
1043 p
= &s
->refcount_block
[cluster_offset
>> s
->cluster_bits
];
1044 refcount
= be16_to_cpu(*p
);
1046 *p
= cpu_to_be16(refcount
);
1050 static int qcow_create(const char *filename
, int64_t total_size
,
1051 const char *backing_file
, int flags
)
1053 int fd
, header_size
, backing_filename_len
, l1_size
, i
, shift
, l2_bits
;
1055 uint64_t tmp
, offset
;
1056 QCowCreateState s1
, *s
= &s1
;
1058 memset(s
, 0, sizeof(*s
));
1060 fd
= open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, 0644);
1063 memset(&header
, 0, sizeof(header
));
1064 header
.magic
= cpu_to_be32(QCOW_MAGIC
);
1065 header
.version
= cpu_to_be32(QCOW_VERSION
);
1066 header
.size
= cpu_to_be64(total_size
* 512);
1067 header_size
= sizeof(header
);
1068 backing_filename_len
= 0;
1070 header
.backing_file_offset
= cpu_to_be64(header_size
);
1071 backing_filename_len
= strlen(backing_file
);
1072 header
.backing_file_size
= cpu_to_be32(backing_filename_len
);
1073 header_size
+= backing_filename_len
;
1075 s
->cluster_bits
= 12; /* 4 KB clusters */
1076 s
->cluster_size
= 1 << s
->cluster_bits
;
1077 header
.cluster_bits
= cpu_to_be32(s
->cluster_bits
);
1078 header_size
= (header_size
+ 7) & ~7;
1080 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
1082 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
1084 l2_bits
= s
->cluster_bits
- 3;
1085 shift
= s
->cluster_bits
+ l2_bits
;
1086 l1_size
= (((total_size
* 512) + (1LL << shift
) - 1) >> shift
);
1087 offset
= align_offset(header_size
, s
->cluster_size
);
1088 s
->l1_table_offset
= offset
;
1089 header
.l1_table_offset
= cpu_to_be64(s
->l1_table_offset
);
1090 header
.l1_size
= cpu_to_be32(l1_size
);
1091 offset
+= align_offset(l1_size
* sizeof(uint64_t), s
->cluster_size
);
1093 s
->refcount_table
= qemu_mallocz(s
->cluster_size
);
1094 if (!s
->refcount_table
)
1096 s
->refcount_block
= qemu_mallocz(s
->cluster_size
);
1097 if (!s
->refcount_block
)
1100 s
->refcount_table_offset
= offset
;
1101 header
.refcount_table_offset
= cpu_to_be64(offset
);
1102 header
.refcount_table_clusters
= cpu_to_be32(1);
1103 offset
+= s
->cluster_size
;
1105 s
->refcount_table
[0] = cpu_to_be64(offset
);
1106 s
->refcount_block_offset
= offset
;
1107 offset
+= s
->cluster_size
;
1109 /* update refcounts */
1110 create_refcount_update(s
, 0, header_size
);
1111 create_refcount_update(s
, s
->l1_table_offset
, l1_size
* sizeof(uint64_t));
1112 create_refcount_update(s
, s
->refcount_table_offset
, s
->cluster_size
);
1113 create_refcount_update(s
, s
->refcount_block_offset
, s
->cluster_size
);
1115 /* write all the data */
1116 write(fd
, &header
, sizeof(header
));
1118 write(fd
, backing_file
, backing_filename_len
);
1120 lseek(fd
, s
->l1_table_offset
, SEEK_SET
);
1122 for(i
= 0;i
< l1_size
; i
++) {
1123 write(fd
, &tmp
, sizeof(tmp
));
1125 lseek(fd
, s
->refcount_table_offset
, SEEK_SET
);
1126 write(fd
, s
->refcount_table
, s
->cluster_size
);
1128 lseek(fd
, s
->refcount_block_offset
, SEEK_SET
);
1129 write(fd
, s
->refcount_block
, s
->cluster_size
);
1131 qemu_free(s
->refcount_table
);
1132 qemu_free(s
->refcount_block
);
1136 qemu_free(s
->refcount_table
);
1137 qemu_free(s
->refcount_block
);
1142 static int qcow_make_empty(BlockDriverState
*bs
)
1145 /* XXX: not correct */
1146 BDRVQcowState
*s
= bs
->opaque
;
1147 uint32_t l1_length
= s
->l1_size
* sizeof(uint64_t);
1150 memset(s
->l1_table
, 0, l1_length
);
1151 if (bdrv_pwrite(s
->hd
, s
->l1_table_offset
, s
->l1_table
, l1_length
) < 0)
1153 ret
= bdrv_truncate(s
->hd
, s
->l1_table_offset
+ l1_length
);
1162 /* XXX: put compressed sectors first, then all the cluster aligned
1163 tables to avoid losing bytes in alignment */
1164 static int qcow_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1165 const uint8_t *buf
, int nb_sectors
)
1167 BDRVQcowState
*s
= bs
->opaque
;
1171 uint64_t cluster_offset
;
1173 if (nb_sectors
== 0) {
1174 /* align end of file to a sector boundary to ease reading with
1175 sector based I/Os */
1176 cluster_offset
= bdrv_getlength(s
->hd
);
1177 cluster_offset
= (cluster_offset
+ 511) & ~511;
1178 bdrv_truncate(s
->hd
, cluster_offset
);
1182 if (nb_sectors
!= s
->cluster_sectors
)
1185 out_buf
= qemu_malloc(s
->cluster_size
+ (s
->cluster_size
/ 1000) + 128);
1189 /* best compression, small window, no zlib header */
1190 memset(&strm
, 0, sizeof(strm
));
1191 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
1193 9, Z_DEFAULT_STRATEGY
);
1199 strm
.avail_in
= s
->cluster_size
;
1200 strm
.next_in
= (uint8_t *)buf
;
1201 strm
.avail_out
= s
->cluster_size
;
1202 strm
.next_out
= out_buf
;
1204 ret
= deflate(&strm
, Z_FINISH
);
1205 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
1210 out_len
= strm
.next_out
- out_buf
;
1214 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
1215 /* could not compress: write normal cluster */
1216 qcow_write(bs
, sector_num
, buf
, s
->cluster_sectors
);
1218 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 2,
1220 cluster_offset
&= s
->cluster_offset_mask
;
1221 if (bdrv_pwrite(s
->hd
, cluster_offset
, out_buf
, out_len
) != out_len
) {
1231 static void qcow_flush(BlockDriverState
*bs
)
1233 BDRVQcowState
*s
= bs
->opaque
;
1237 static int qcow_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1239 BDRVQcowState
*s
= bs
->opaque
;
1240 bdi
->cluster_size
= s
->cluster_size
;
1241 bdi
->vm_state_offset
= (int64_t)s
->l1_vm_state_index
<<
1242 (s
->cluster_bits
+ s
->l2_bits
);
1246 /*********************************************************/
1247 /* snapshot support */
1249 /* update the refcounts of snapshots and the copied flag */
1250 static int update_snapshot_refcount(BlockDriverState
*bs
,
1251 int64_t l1_table_offset
,
1255 BDRVQcowState
*s
= bs
->opaque
;
1256 uint64_t *l1_table
, *l2_table
, l2_offset
, offset
, l1_size2
, l1_allocated
;
1257 int64_t old_offset
, old_l2_offset
;
1258 int l2_size
, i
, j
, l1_modified
, l2_modified
, nb_csectors
, refcount
;
1264 l1_size2
= l1_size
* sizeof(uint64_t);
1266 if (l1_table_offset
!= s
->l1_table_offset
) {
1267 l1_table
= qemu_malloc(l1_size2
);
1271 if (bdrv_pread(s
->hd
, l1_table_offset
,
1272 l1_table
, l1_size2
) != l1_size2
)
1274 for(i
= 0;i
< l1_size
; i
++)
1275 be64_to_cpus(&l1_table
[i
]);
1277 assert(l1_size
== s
->l1_size
);
1278 l1_table
= s
->l1_table
;
1282 l2_size
= s
->l2_size
* sizeof(uint64_t);
1283 l2_table
= qemu_malloc(l2_size
);
1287 for(i
= 0; i
< l1_size
; i
++) {
1288 l2_offset
= l1_table
[i
];
1290 old_l2_offset
= l2_offset
;
1291 l2_offset
&= ~QCOW_OFLAG_COPIED
;
1293 if (bdrv_pread(s
->hd
, l2_offset
, l2_table
, l2_size
) != l2_size
)
1295 for(j
= 0; j
< s
->l2_size
; j
++) {
1296 offset
= be64_to_cpu(l2_table
[j
]);
1298 old_offset
= offset
;
1299 offset
&= ~QCOW_OFLAG_COPIED
;
1300 if (offset
& QCOW_OFLAG_COMPRESSED
) {
1301 nb_csectors
= ((offset
>> s
->csize_shift
) &
1304 update_refcount(bs
, (offset
& s
->cluster_offset_mask
) & ~511,
1305 nb_csectors
* 512, addend
);
1306 /* compressed clusters are never modified */
1310 refcount
= update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, addend
);
1312 refcount
= get_refcount(bs
, offset
>> s
->cluster_bits
);
1316 if (refcount
== 1) {
1317 offset
|= QCOW_OFLAG_COPIED
;
1319 if (offset
!= old_offset
) {
1320 l2_table
[j
] = cpu_to_be64(offset
);
1326 if (bdrv_pwrite(s
->hd
,
1327 l2_offset
, l2_table
, l2_size
) != l2_size
)
1332 refcount
= update_cluster_refcount(bs
, l2_offset
>> s
->cluster_bits
, addend
);
1334 refcount
= get_refcount(bs
, l2_offset
>> s
->cluster_bits
);
1336 if (refcount
== 1) {
1337 l2_offset
|= QCOW_OFLAG_COPIED
;
1339 if (l2_offset
!= old_l2_offset
) {
1340 l1_table
[i
] = l2_offset
;
1346 for(i
= 0; i
< l1_size
; i
++)
1347 cpu_to_be64s(&l1_table
[i
]);
1348 if (bdrv_pwrite(s
->hd
, l1_table_offset
, l1_table
,
1349 l1_size2
) != l1_size2
)
1351 for(i
= 0; i
< l1_size
; i
++)
1352 be64_to_cpus(&l1_table
[i
]);
1355 qemu_free(l1_table
);
1356 qemu_free(l2_table
);
1360 qemu_free(l1_table
);
1361 qemu_free(l2_table
);
1365 static void qcow_free_snapshots(BlockDriverState
*bs
)
1367 BDRVQcowState
*s
= bs
->opaque
;
1370 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1371 qemu_free(s
->snapshots
[i
].name
);
1372 qemu_free(s
->snapshots
[i
].id_str
);
1374 qemu_free(s
->snapshots
);
1375 s
->snapshots
= NULL
;
1376 s
->nb_snapshots
= 0;
1379 static int qcow_read_snapshots(BlockDriverState
*bs
)
1381 BDRVQcowState
*s
= bs
->opaque
;
1382 QCowSnapshotHeader h
;
1384 int i
, id_str_size
, name_size
;
1386 uint32_t extra_data_size
;
1388 offset
= s
->snapshots_offset
;
1389 s
->snapshots
= qemu_mallocz(s
->nb_snapshots
* sizeof(QCowSnapshot
));
1392 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1393 offset
= align_offset(offset
, 8);
1394 if (bdrv_pread(s
->hd
, offset
, &h
, sizeof(h
)) != sizeof(h
))
1396 offset
+= sizeof(h
);
1397 sn
= s
->snapshots
+ i
;
1398 sn
->l1_table_offset
= be64_to_cpu(h
.l1_table_offset
);
1399 sn
->l1_size
= be32_to_cpu(h
.l1_size
);
1400 sn
->vm_state_size
= be32_to_cpu(h
.vm_state_size
);
1401 sn
->date_sec
= be32_to_cpu(h
.date_sec
);
1402 sn
->date_nsec
= be32_to_cpu(h
.date_nsec
);
1403 sn
->vm_clock_nsec
= be64_to_cpu(h
.vm_clock_nsec
);
1404 extra_data_size
= be32_to_cpu(h
.extra_data_size
);
1406 id_str_size
= be16_to_cpu(h
.id_str_size
);
1407 name_size
= be16_to_cpu(h
.name_size
);
1409 offset
+= extra_data_size
;
1411 sn
->id_str
= qemu_malloc(id_str_size
+ 1);
1414 if (bdrv_pread(s
->hd
, offset
, sn
->id_str
, id_str_size
) != id_str_size
)
1416 offset
+= id_str_size
;
1417 sn
->id_str
[id_str_size
] = '\0';
1419 sn
->name
= qemu_malloc(name_size
+ 1);
1422 if (bdrv_pread(s
->hd
, offset
, sn
->name
, name_size
) != name_size
)
1424 offset
+= name_size
;
1425 sn
->name
[name_size
] = '\0';
1427 s
->snapshots_size
= offset
- s
->snapshots_offset
;
1430 qcow_free_snapshots(bs
);
1434 /* add at the end of the file a new list of snapshots */
1435 static int qcow_write_snapshots(BlockDriverState
*bs
)
1437 BDRVQcowState
*s
= bs
->opaque
;
1439 QCowSnapshotHeader h
;
1440 int i
, name_size
, id_str_size
, snapshots_size
;
1443 int64_t offset
, snapshots_offset
;
1445 /* compute the size of the snapshots */
1447 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1448 sn
= s
->snapshots
+ i
;
1449 offset
= align_offset(offset
, 8);
1450 offset
+= sizeof(h
);
1451 offset
+= strlen(sn
->id_str
);
1452 offset
+= strlen(sn
->name
);
1454 snapshots_size
= offset
;
1456 snapshots_offset
= alloc_clusters(bs
, snapshots_size
);
1457 offset
= snapshots_offset
;
1459 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1460 sn
= s
->snapshots
+ i
;
1461 memset(&h
, 0, sizeof(h
));
1462 h
.l1_table_offset
= cpu_to_be64(sn
->l1_table_offset
);
1463 h
.l1_size
= cpu_to_be32(sn
->l1_size
);
1464 h
.vm_state_size
= cpu_to_be32(sn
->vm_state_size
);
1465 h
.date_sec
= cpu_to_be32(sn
->date_sec
);
1466 h
.date_nsec
= cpu_to_be32(sn
->date_nsec
);
1467 h
.vm_clock_nsec
= cpu_to_be64(sn
->vm_clock_nsec
);
1469 id_str_size
= strlen(sn
->id_str
);
1470 name_size
= strlen(sn
->name
);
1471 h
.id_str_size
= cpu_to_be16(id_str_size
);
1472 h
.name_size
= cpu_to_be16(name_size
);
1473 offset
= align_offset(offset
, 8);
1474 if (bdrv_pwrite(s
->hd
, offset
, &h
, sizeof(h
)) != sizeof(h
))
1476 offset
+= sizeof(h
);
1477 if (bdrv_pwrite(s
->hd
, offset
, sn
->id_str
, id_str_size
) != id_str_size
)
1479 offset
+= id_str_size
;
1480 if (bdrv_pwrite(s
->hd
, offset
, sn
->name
, name_size
) != name_size
)
1482 offset
+= name_size
;
1485 /* update the various header fields */
1486 data64
= cpu_to_be64(snapshots_offset
);
1487 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, snapshots_offset
),
1488 &data64
, sizeof(data64
)) != sizeof(data64
))
1490 data32
= cpu_to_be32(s
->nb_snapshots
);
1491 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, nb_snapshots
),
1492 &data32
, sizeof(data32
)) != sizeof(data32
))
1495 /* free the old snapshot table */
1496 free_clusters(bs
, s
->snapshots_offset
, s
->snapshots_size
);
1497 s
->snapshots_offset
= snapshots_offset
;
1498 s
->snapshots_size
= snapshots_size
;
1504 static void find_new_snapshot_id(BlockDriverState
*bs
,
1505 char *id_str
, int id_str_size
)
1507 BDRVQcowState
*s
= bs
->opaque
;
1509 int i
, id
, id_max
= 0;
1511 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1512 sn
= s
->snapshots
+ i
;
1513 id
= strtoul(sn
->id_str
, NULL
, 10);
1517 snprintf(id_str
, id_str_size
, "%d", id_max
+ 1);
1520 static int find_snapshot_by_id(BlockDriverState
*bs
, const char *id_str
)
1522 BDRVQcowState
*s
= bs
->opaque
;
1525 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1526 if (!strcmp(s
->snapshots
[i
].id_str
, id_str
))
1532 static int find_snapshot_by_id_or_name(BlockDriverState
*bs
, const char *name
)
1534 BDRVQcowState
*s
= bs
->opaque
;
1537 ret
= find_snapshot_by_id(bs
, name
);
1540 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1541 if (!strcmp(s
->snapshots
[i
].name
, name
))
1547 /* if no id is provided, a new one is constructed */
1548 static int qcow_snapshot_create(BlockDriverState
*bs
,
1549 QEMUSnapshotInfo
*sn_info
)
1551 BDRVQcowState
*s
= bs
->opaque
;
1552 QCowSnapshot
*snapshots1
, sn1
, *sn
= &sn1
;
1554 uint64_t *l1_table
= NULL
;
1556 memset(sn
, 0, sizeof(*sn
));
1558 if (sn_info
->id_str
[0] == '\0') {
1559 /* compute a new id */
1560 find_new_snapshot_id(bs
, sn_info
->id_str
, sizeof(sn_info
->id_str
));
1563 /* check that the ID is unique */
1564 if (find_snapshot_by_id(bs
, sn_info
->id_str
) >= 0)
1567 sn
->id_str
= qemu_strdup(sn_info
->id_str
);
1570 sn
->name
= qemu_strdup(sn_info
->name
);
1573 sn
->vm_state_size
= sn_info
->vm_state_size
;
1574 sn
->date_sec
= sn_info
->date_sec
;
1575 sn
->date_nsec
= sn_info
->date_nsec
;
1576 sn
->vm_clock_nsec
= sn_info
->vm_clock_nsec
;
1578 ret
= update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 1);
1582 /* create the L1 table of the snapshot */
1583 sn
->l1_table_offset
= alloc_clusters(bs
, s
->l1_size
* sizeof(uint64_t));
1584 sn
->l1_size
= s
->l1_size
;
1586 l1_table
= qemu_malloc(s
->l1_size
* sizeof(uint64_t));
1589 for(i
= 0; i
< s
->l1_size
; i
++) {
1590 l1_table
[i
] = cpu_to_be64(s
->l1_table
[i
]);
1592 if (bdrv_pwrite(s
->hd
, sn
->l1_table_offset
,
1593 l1_table
, s
->l1_size
* sizeof(uint64_t)) !=
1594 (s
->l1_size
* sizeof(uint64_t)))
1596 qemu_free(l1_table
);
1599 snapshots1
= qemu_malloc((s
->nb_snapshots
+ 1) * sizeof(QCowSnapshot
));
1602 memcpy(snapshots1
, s
->snapshots
, s
->nb_snapshots
* sizeof(QCowSnapshot
));
1603 s
->snapshots
= snapshots1
;
1604 s
->snapshots
[s
->nb_snapshots
++] = *sn
;
1606 if (qcow_write_snapshots(bs
) < 0)
1609 check_refcounts(bs
);
1613 qemu_free(sn
->name
);
1614 qemu_free(l1_table
);
1618 /* copy the snapshot 'snapshot_name' into the current disk image */
1619 static int qcow_snapshot_goto(BlockDriverState
*bs
,
1620 const char *snapshot_id
)
1622 BDRVQcowState
*s
= bs
->opaque
;
1624 int i
, snapshot_index
, l1_size2
;
1626 snapshot_index
= find_snapshot_by_id_or_name(bs
, snapshot_id
);
1627 if (snapshot_index
< 0)
1629 sn
= &s
->snapshots
[snapshot_index
];
1631 if (update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, -1) < 0)
1634 if (grow_l1_table(bs
, sn
->l1_size
) < 0)
1637 s
->l1_size
= sn
->l1_size
;
1638 l1_size2
= s
->l1_size
* sizeof(uint64_t);
1639 /* copy the snapshot l1 table to the current l1 table */
1640 if (bdrv_pread(s
->hd
, sn
->l1_table_offset
,
1641 s
->l1_table
, l1_size2
) != l1_size2
)
1643 if (bdrv_pwrite(s
->hd
, s
->l1_table_offset
,
1644 s
->l1_table
, l1_size2
) != l1_size2
)
1646 for(i
= 0;i
< s
->l1_size
; i
++) {
1647 be64_to_cpus(&s
->l1_table
[i
]);
1650 if (update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 1) < 0)
1654 check_refcounts(bs
);
1661 static int qcow_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
1663 BDRVQcowState
*s
= bs
->opaque
;
1665 int snapshot_index
, ret
;
1667 snapshot_index
= find_snapshot_by_id_or_name(bs
, snapshot_id
);
1668 if (snapshot_index
< 0)
1670 sn
= &s
->snapshots
[snapshot_index
];
1672 ret
= update_snapshot_refcount(bs
, sn
->l1_table_offset
, sn
->l1_size
, -1);
1675 /* must update the copied flag on the current cluster offsets */
1676 ret
= update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 0);
1679 free_clusters(bs
, sn
->l1_table_offset
, sn
->l1_size
* sizeof(uint64_t));
1681 qemu_free(sn
->id_str
);
1682 qemu_free(sn
->name
);
1683 memmove(sn
, sn
+ 1, (s
->nb_snapshots
- snapshot_index
- 1) * sizeof(*sn
));
1685 ret
= qcow_write_snapshots(bs
);
1687 /* XXX: restore snapshot if error ? */
1691 check_refcounts(bs
);
1696 static int qcow_snapshot_list(BlockDriverState
*bs
,
1697 QEMUSnapshotInfo
**psn_tab
)
1699 BDRVQcowState
*s
= bs
->opaque
;
1700 QEMUSnapshotInfo
*sn_tab
, *sn_info
;
1704 sn_tab
= qemu_mallocz(s
->nb_snapshots
* sizeof(QEMUSnapshotInfo
));
1707 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1708 sn_info
= sn_tab
+ i
;
1709 sn
= s
->snapshots
+ i
;
1710 pstrcpy(sn_info
->id_str
, sizeof(sn_info
->id_str
),
1712 pstrcpy(sn_info
->name
, sizeof(sn_info
->name
),
1714 sn_info
->vm_state_size
= sn
->vm_state_size
;
1715 sn_info
->date_sec
= sn
->date_sec
;
1716 sn_info
->date_nsec
= sn
->date_nsec
;
1717 sn_info
->vm_clock_nsec
= sn
->vm_clock_nsec
;
1720 return s
->nb_snapshots
;
1727 /*********************************************************/
1728 /* refcount handling */
1730 static int refcount_init(BlockDriverState
*bs
)
1732 BDRVQcowState
*s
= bs
->opaque
;
1733 int ret
, refcount_table_size2
, i
;
1735 s
->refcount_block_cache
= qemu_malloc(s
->cluster_size
);
1736 if (!s
->refcount_block_cache
)
1738 refcount_table_size2
= s
->refcount_table_size
* sizeof(uint64_t);
1739 s
->refcount_table
= qemu_malloc(refcount_table_size2
);
1740 if (!s
->refcount_table
)
1742 if (s
->refcount_table_size
> 0) {
1743 ret
= bdrv_pread(s
->hd
, s
->refcount_table_offset
,
1744 s
->refcount_table
, refcount_table_size2
);
1745 if (ret
!= refcount_table_size2
)
1747 for(i
= 0; i
< s
->refcount_table_size
; i
++)
1748 be64_to_cpus(&s
->refcount_table
[i
]);
1755 static void refcount_close(BlockDriverState
*bs
)
1757 BDRVQcowState
*s
= bs
->opaque
;
1758 qemu_free(s
->refcount_block_cache
);
1759 qemu_free(s
->refcount_table
);
1763 static int load_refcount_block(BlockDriverState
*bs
,
1764 int64_t refcount_block_offset
)
1766 BDRVQcowState
*s
= bs
->opaque
;
1768 ret
= bdrv_pread(s
->hd
, refcount_block_offset
, s
->refcount_block_cache
,
1770 if (ret
!= s
->cluster_size
)
1772 s
->refcount_block_cache_offset
= refcount_block_offset
;
1776 static int get_refcount(BlockDriverState
*bs
, int64_t cluster_index
)
1778 BDRVQcowState
*s
= bs
->opaque
;
1779 int refcount_table_index
, block_index
;
1780 int64_t refcount_block_offset
;
1782 refcount_table_index
= cluster_index
>> (s
->cluster_bits
- REFCOUNT_SHIFT
);
1783 if (refcount_table_index
>= s
->refcount_table_size
)
1785 refcount_block_offset
= s
->refcount_table
[refcount_table_index
];
1786 if (!refcount_block_offset
)
1788 if (refcount_block_offset
!= s
->refcount_block_cache_offset
) {
1789 /* better than nothing: return allocated if read error */
1790 if (load_refcount_block(bs
, refcount_block_offset
) < 0)
1793 block_index
= cluster_index
&
1794 ((1 << (s
->cluster_bits
- REFCOUNT_SHIFT
)) - 1);
1795 return be16_to_cpu(s
->refcount_block_cache
[block_index
]);
1798 /* return < 0 if error */
1799 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, int64_t size
)
1801 BDRVQcowState
*s
= bs
->opaque
;
1804 nb_clusters
= (size
+ s
->cluster_size
- 1) >> s
->cluster_bits
;
1806 if (get_refcount(bs
, s
->free_cluster_index
) == 0) {
1807 s
->free_cluster_index
++;
1808 for(i
= 1; i
< nb_clusters
; i
++) {
1809 if (get_refcount(bs
, s
->free_cluster_index
) != 0)
1811 s
->free_cluster_index
++;
1814 printf("alloc_clusters: size=%lld -> %lld\n",
1816 (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
);
1818 return (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
;
1821 s
->free_cluster_index
++;
1826 static int64_t alloc_clusters(BlockDriverState
*bs
, int64_t size
)
1830 offset
= alloc_clusters_noref(bs
, size
);
1831 update_refcount(bs
, offset
, size
, 1);
1835 /* only used to allocate compressed sectors. We try to allocate
1836 contiguous sectors. size must be <= cluster_size */
1837 static int64_t alloc_bytes(BlockDriverState
*bs
, int size
)
1839 BDRVQcowState
*s
= bs
->opaque
;
1840 int64_t offset
, cluster_offset
;
1841 int free_in_cluster
;
1843 assert(size
> 0 && size
<= s
->cluster_size
);
1844 if (s
->free_byte_offset
== 0) {
1845 s
->free_byte_offset
= alloc_clusters(bs
, s
->cluster_size
);
1848 free_in_cluster
= s
->cluster_size
-
1849 (s
->free_byte_offset
& (s
->cluster_size
- 1));
1850 if (size
<= free_in_cluster
) {
1851 /* enough space in current cluster */
1852 offset
= s
->free_byte_offset
;
1853 s
->free_byte_offset
+= size
;
1854 free_in_cluster
-= size
;
1855 if (free_in_cluster
== 0)
1856 s
->free_byte_offset
= 0;
1857 if ((offset
& (s
->cluster_size
- 1)) != 0)
1858 update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, 1);
1860 offset
= alloc_clusters(bs
, s
->cluster_size
);
1861 cluster_offset
= s
->free_byte_offset
& ~(s
->cluster_size
- 1);
1862 if ((cluster_offset
+ s
->cluster_size
) == offset
) {
1863 /* we are lucky: contiguous data */
1864 offset
= s
->free_byte_offset
;
1865 update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, 1);
1866 s
->free_byte_offset
+= size
;
1868 s
->free_byte_offset
= offset
;
1875 static void free_clusters(BlockDriverState
*bs
,
1876 int64_t offset
, int64_t size
)
1878 update_refcount(bs
, offset
, size
, -1);
1881 static int grow_refcount_table(BlockDriverState
*bs
, int min_size
)
1883 BDRVQcowState
*s
= bs
->opaque
;
1884 int new_table_size
, new_table_size2
, refcount_table_clusters
, i
, ret
;
1885 uint64_t *new_table
;
1886 int64_t table_offset
;
1890 if (min_size
<= s
->refcount_table_size
)
1892 /* compute new table size */
1893 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
1895 if (refcount_table_clusters
== 0) {
1896 refcount_table_clusters
= 1;
1898 refcount_table_clusters
= (refcount_table_clusters
* 3 + 1) / 2;
1900 new_table_size
= refcount_table_clusters
<< (s
->cluster_bits
- 3);
1901 if (min_size
<= new_table_size
)
1905 printf("grow_refcount_table from %d to %d\n",
1906 s
->refcount_table_size
,
1909 new_table_size2
= new_table_size
* sizeof(uint64_t);
1910 new_table
= qemu_mallocz(new_table_size2
);
1913 memcpy(new_table
, s
->refcount_table
,
1914 s
->refcount_table_size
* sizeof(uint64_t));
1915 for(i
= 0; i
< s
->refcount_table_size
; i
++)
1916 cpu_to_be64s(&new_table
[i
]);
1917 /* Note: we cannot update the refcount now to avoid recursion */
1918 table_offset
= alloc_clusters_noref(bs
, new_table_size2
);
1919 ret
= bdrv_pwrite(s
->hd
, table_offset
, new_table
, new_table_size2
);
1920 if (ret
!= new_table_size2
)
1922 for(i
= 0; i
< s
->refcount_table_size
; i
++)
1923 be64_to_cpus(&new_table
[i
]);
1925 data64
= cpu_to_be64(table_offset
);
1926 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, refcount_table_offset
),
1927 &data64
, sizeof(data64
)) != sizeof(data64
))
1929 data32
= cpu_to_be32(refcount_table_clusters
);
1930 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, refcount_table_clusters
),
1931 &data32
, sizeof(data32
)) != sizeof(data32
))
1933 qemu_free(s
->refcount_table
);
1934 s
->refcount_table
= new_table
;
1935 s
->refcount_table_size
= new_table_size
;
1937 update_refcount(bs
, table_offset
, new_table_size2
, 1);
1940 free_clusters(bs
, table_offset
, new_table_size2
);
1941 qemu_free(new_table
);
1945 /* addend must be 1 or -1 */
1946 /* XXX: cache several refcount block clusters ? */
1947 static int update_cluster_refcount(BlockDriverState
*bs
,
1948 int64_t cluster_index
,
1951 BDRVQcowState
*s
= bs
->opaque
;
1952 int64_t offset
, refcount_block_offset
;
1953 int ret
, refcount_table_index
, block_index
, refcount
;
1956 refcount_table_index
= cluster_index
>> (s
->cluster_bits
- REFCOUNT_SHIFT
);
1957 if (refcount_table_index
>= s
->refcount_table_size
) {
1960 ret
= grow_refcount_table(bs
, refcount_table_index
+ 1);
1964 refcount_block_offset
= s
->refcount_table
[refcount_table_index
];
1965 if (!refcount_block_offset
) {
1968 /* create a new refcount block */
1969 /* Note: we cannot update the refcount now to avoid recursion */
1970 offset
= alloc_clusters_noref(bs
, s
->cluster_size
);
1971 memset(s
->refcount_block_cache
, 0, s
->cluster_size
);
1972 ret
= bdrv_pwrite(s
->hd
, offset
, s
->refcount_block_cache
, s
->cluster_size
);
1973 if (ret
!= s
->cluster_size
)
1975 s
->refcount_table
[refcount_table_index
] = offset
;
1976 data64
= cpu_to_be64(offset
);
1977 ret
= bdrv_pwrite(s
->hd
, s
->refcount_table_offset
+
1978 refcount_table_index
* sizeof(uint64_t),
1979 &data64
, sizeof(data64
));
1980 if (ret
!= sizeof(data64
))
1983 refcount_block_offset
= offset
;
1984 s
->refcount_block_cache_offset
= offset
;
1985 update_refcount(bs
, offset
, s
->cluster_size
, 1);
1987 if (refcount_block_offset
!= s
->refcount_block_cache_offset
) {
1988 if (load_refcount_block(bs
, refcount_block_offset
) < 0)
1992 /* we can update the count and save it */
1993 block_index
= cluster_index
&
1994 ((1 << (s
->cluster_bits
- REFCOUNT_SHIFT
)) - 1);
1995 refcount
= be16_to_cpu(s
->refcount_block_cache
[block_index
]);
1997 if (refcount
< 0 || refcount
> 0xffff)
1999 if (refcount
== 0 && cluster_index
< s
->free_cluster_index
) {
2000 s
->free_cluster_index
= cluster_index
;
2002 s
->refcount_block_cache
[block_index
] = cpu_to_be16(refcount
);
2003 if (bdrv_pwrite(s
->hd
,
2004 refcount_block_offset
+ (block_index
<< REFCOUNT_SHIFT
),
2005 &s
->refcount_block_cache
[block_index
], 2) != 2)
2010 static void update_refcount(BlockDriverState
*bs
,
2011 int64_t offset
, int64_t length
,
2014 BDRVQcowState
*s
= bs
->opaque
;
2015 int64_t start
, last
, cluster_offset
;
2018 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2019 offset
, length
, addend
);
2023 start
= offset
& ~(s
->cluster_size
- 1);
2024 last
= (offset
+ length
- 1) & ~(s
->cluster_size
- 1);
2025 for(cluster_offset
= start
; cluster_offset
<= last
;
2026 cluster_offset
+= s
->cluster_size
) {
2027 update_cluster_refcount(bs
, cluster_offset
>> s
->cluster_bits
, addend
);
2032 static void inc_refcounts(BlockDriverState
*bs
,
2033 uint16_t *refcount_table
,
2034 int refcount_table_size
,
2035 int64_t offset
, int64_t size
)
2037 BDRVQcowState
*s
= bs
->opaque
;
2038 int64_t start
, last
, cluster_offset
;
2044 start
= offset
& ~(s
->cluster_size
- 1);
2045 last
= (offset
+ size
- 1) & ~(s
->cluster_size
- 1);
2046 for(cluster_offset
= start
; cluster_offset
<= last
;
2047 cluster_offset
+= s
->cluster_size
) {
2048 k
= cluster_offset
>> s
->cluster_bits
;
2049 if (k
< 0 || k
>= refcount_table_size
) {
2050 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset
);
2052 if (++refcount_table
[k
] == 0) {
2053 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset
);
2059 static int check_refcounts_l1(BlockDriverState
*bs
,
2060 uint16_t *refcount_table
,
2061 int refcount_table_size
,
2062 int64_t l1_table_offset
, int l1_size
,
2065 BDRVQcowState
*s
= bs
->opaque
;
2066 uint64_t *l1_table
, *l2_table
, l2_offset
, offset
, l1_size2
;
2067 int l2_size
, i
, j
, nb_csectors
, refcount
;
2070 l1_size2
= l1_size
* sizeof(uint64_t);
2072 inc_refcounts(bs
, refcount_table
, refcount_table_size
,
2073 l1_table_offset
, l1_size2
);
2075 l1_table
= qemu_malloc(l1_size2
);
2078 if (bdrv_pread(s
->hd
, l1_table_offset
,
2079 l1_table
, l1_size2
) != l1_size2
)
2081 for(i
= 0;i
< l1_size
; i
++)
2082 be64_to_cpus(&l1_table
[i
]);
2084 l2_size
= s
->l2_size
* sizeof(uint64_t);
2085 l2_table
= qemu_malloc(l2_size
);
2088 for(i
= 0; i
< l1_size
; i
++) {
2089 l2_offset
= l1_table
[i
];
2092 refcount
= get_refcount(bs
, (l2_offset
& ~QCOW_OFLAG_COPIED
) >> s
->cluster_bits
);
2093 if ((refcount
== 1) != ((l2_offset
& QCOW_OFLAG_COPIED
) != 0)) {
2094 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2095 l2_offset
, refcount
);
2098 l2_offset
&= ~QCOW_OFLAG_COPIED
;
2099 if (bdrv_pread(s
->hd
, l2_offset
, l2_table
, l2_size
) != l2_size
)
2101 for(j
= 0; j
< s
->l2_size
; j
++) {
2102 offset
= be64_to_cpu(l2_table
[j
]);
2104 if (offset
& QCOW_OFLAG_COMPRESSED
) {
2105 if (offset
& QCOW_OFLAG_COPIED
) {
2106 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2107 offset
>> s
->cluster_bits
);
2108 offset
&= ~QCOW_OFLAG_COPIED
;
2110 nb_csectors
= ((offset
>> s
->csize_shift
) &
2112 offset
&= s
->cluster_offset_mask
;
2113 inc_refcounts(bs
, refcount_table
,
2114 refcount_table_size
,
2115 offset
& ~511, nb_csectors
* 512);
2118 refcount
= get_refcount(bs
, (offset
& ~QCOW_OFLAG_COPIED
) >> s
->cluster_bits
);
2119 if ((refcount
== 1) != ((offset
& QCOW_OFLAG_COPIED
) != 0)) {
2120 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2124 offset
&= ~QCOW_OFLAG_COPIED
;
2125 inc_refcounts(bs
, refcount_table
,
2126 refcount_table_size
,
2127 offset
, s
->cluster_size
);
2131 inc_refcounts(bs
, refcount_table
,
2132 refcount_table_size
,
2137 qemu_free(l1_table
);
2138 qemu_free(l2_table
);
2141 printf("ERROR: I/O error in check_refcounts_l1\n");
2142 qemu_free(l1_table
);
2143 qemu_free(l2_table
);
2147 static void check_refcounts(BlockDriverState
*bs
)
2149 BDRVQcowState
*s
= bs
->opaque
;
2151 int nb_clusters
, refcount1
, refcount2
, i
;
2153 uint16_t *refcount_table
;
2155 size
= bdrv_getlength(s
->hd
);
2156 nb_clusters
= (size
+ s
->cluster_size
- 1) >> s
->cluster_bits
;
2157 refcount_table
= qemu_mallocz(nb_clusters
* sizeof(uint16_t));
2160 inc_refcounts(bs
, refcount_table
, nb_clusters
,
2161 0, s
->cluster_size
);
2163 check_refcounts_l1(bs
, refcount_table
, nb_clusters
,
2164 s
->l1_table_offset
, s
->l1_size
, 1);
2167 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
2168 sn
= s
->snapshots
+ i
;
2169 check_refcounts_l1(bs
, refcount_table
, nb_clusters
,
2170 sn
->l1_table_offset
, sn
->l1_size
, 0);
2172 inc_refcounts(bs
, refcount_table
, nb_clusters
,
2173 s
->snapshots_offset
, s
->snapshots_size
);
2176 inc_refcounts(bs
, refcount_table
, nb_clusters
,
2177 s
->refcount_table_offset
,
2178 s
->refcount_table_size
* sizeof(uint64_t));
2179 for(i
= 0; i
< s
->refcount_table_size
; i
++) {
2181 offset
= s
->refcount_table
[i
];
2183 inc_refcounts(bs
, refcount_table
, nb_clusters
,
2184 offset
, s
->cluster_size
);
2188 /* compare ref counts */
2189 for(i
= 0; i
< nb_clusters
; i
++) {
2190 refcount1
= get_refcount(bs
, i
);
2191 refcount2
= refcount_table
[i
];
2192 if (refcount1
!= refcount2
)
2193 printf("ERROR cluster %d refcount=%d reference=%d\n",
2194 i
, refcount1
, refcount2
);
2197 qemu_free(refcount_table
);
2201 static void dump_refcounts(BlockDriverState
*bs
)
2203 BDRVQcowState
*s
= bs
->opaque
;
2204 int64_t nb_clusters
, k
, k1
, size
;
2207 size
= bdrv_getlength(s
->hd
);
2208 nb_clusters
= (size
+ s
->cluster_size
- 1) >> s
->cluster_bits
;
2209 for(k
= 0; k
< nb_clusters
;) {
2211 refcount
= get_refcount(bs
, k
);
2213 while (k
< nb_clusters
&& get_refcount(bs
, k
) == refcount
)
2215 printf("%lld: refcount=%d nb=%lld\n", k
, refcount
, k
- k1
);
2221 BlockDriver bdrv_qcow2
= {
2223 sizeof(BDRVQcowState
),
2235 .bdrv_aio_read
= qcow_aio_read
,
2236 .bdrv_aio_write
= qcow_aio_write
,
2237 .bdrv_aio_cancel
= qcow_aio_cancel
,
2238 .aiocb_size
= sizeof(QCowAIOCB
),
2239 .bdrv_write_compressed
= qcow_write_compressed
,
2241 .bdrv_snapshot_create
= qcow_snapshot_create
,
2242 .bdrv_snapshot_goto
= qcow_snapshot_goto
,
2243 .bdrv_snapshot_delete
= qcow_snapshot_delete
,
2244 .bdrv_snapshot_list
= qcow_snapshot_list
,
2245 .bdrv_get_info
= qcow_get_info
,
2248 * Block driver for the QCOW version 2 format
2250 * Copyright (c) 2004-2006 Fabrice Bellard
2252 * Permission is hereby granted, free of charge, to any person obtaining a copy
2253 * of this software and associated documentation files (the "Software"), to deal
2254 * in the Software without restriction, including without limitation the rights
2255 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
2256 * copies of the Software, and to permit persons to whom the Software is
2257 * furnished to do so, subject to the following conditions:
2259 * The above copyright notice and this permission notice shall be included in
2260 * all copies or substantial portions of the Software.
2262 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
2263 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
2264 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
2265 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
2266 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
2267 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
2271 #include "block_int.h"
2277 Differences with QCOW:
2279 - Support for multiple incremental snapshots.
2280 - Memory management by reference counts.
2281 - Clusters which have a reference count of one have the bit
2282 QCOW_OFLAG_COPIED to optimize write performance.
2283 - Size of compressed clusters is stored in sectors to reduce bit usage
2284 in the cluster offsets.
2285 - Support for storing additional data (such as the VM state) in the
2287 - If a backing store is used, the cluster size is not constrained
2288 (could be backported to QCOW).
2289 - L2 tables have always a size of one cluster.
2292 //#define DEBUG_ALLOC
2293 //#define DEBUG_ALLOC2
2295 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
2296 #define QCOW_VERSION 2
2298 #define QCOW_CRYPT_NONE 0
2299 #define QCOW_CRYPT_AES 1
2301 /* indicate that the refcount of the referenced cluster is exactly one. */
2302 #define QCOW_OFLAG_COPIED (1LL << 63)
2303 /* indicate that the cluster is compressed (they never have the copied flag) */
2304 #define QCOW_OFLAG_COMPRESSED (1LL << 62)
2306 #define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
2309 #define offsetof(type, field) ((size_t) &((type *)0)->field)
2312 typedef struct QCowHeader
{
2315 uint64_t backing_file_offset
;
2316 uint32_t backing_file_size
;
2317 uint32_t cluster_bits
;
2318 uint64_t size
; /* in bytes */
2319 uint32_t crypt_method
;
2320 uint32_t l1_size
; /* XXX: save number of clusters instead ? */
2321 uint64_t l1_table_offset
;
2322 uint64_t refcount_table_offset
;
2323 uint32_t refcount_table_clusters
;
2324 uint32_t nb_snapshots
;
2325 uint64_t snapshots_offset
;
2328 typedef struct __attribute__((packed
)) QCowSnapshotHeader
{
2329 /* header is 8 byte aligned */
2330 uint64_t l1_table_offset
;
2333 uint16_t id_str_size
;
2339 uint64_t vm_clock_nsec
;
2341 uint32_t vm_state_size
;
2342 uint32_t extra_data_size
; /* for extension */
2343 /* extra data follows */
2344 /* id_str follows */
2346 } QCowSnapshotHeader
;
2348 #define L2_CACHE_SIZE 16
2350 typedef struct QCowSnapshot
{
2351 uint64_t l1_table_offset
;
2355 uint32_t vm_state_size
;
2358 uint64_t vm_clock_nsec
;
2361 typedef struct BDRVQcowState
{
2362 BlockDriverState
*hd
;
2365 int cluster_sectors
;
2369 int l1_vm_state_index
;
2372 uint64_t cluster_offset_mask
;
2373 uint64_t l1_table_offset
;
2376 uint64_t l2_cache_offsets
[L2_CACHE_SIZE
];
2377 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
2378 uint8_t *cluster_cache
;
2379 uint8_t *cluster_data
;
2380 uint64_t cluster_cache_offset
;
2382 uint64_t *refcount_table
;
2383 uint64_t refcount_table_offset
;
2384 uint32_t refcount_table_size
;
2385 uint64_t refcount_block_cache_offset
;
2386 uint16_t *refcount_block_cache
;
2387 int64_t free_cluster_index
;
2388 int64_t free_byte_offset
;
2390 uint32_t crypt_method
; /* current crypt method, 0 if no key yet */
2391 uint32_t crypt_method_header
;
2392 AES_KEY aes_encrypt_key
;
2393 AES_KEY aes_decrypt_key
;
2394 uint64_t snapshots_offset
;
2397 QCowSnapshot
*snapshots
;
2400 static int decompress_cluster(BDRVQcowState
*s
, uint64_t cluster_offset
);
2401 static int qcow_read(BlockDriverState
*bs
, int64_t sector_num
,
2402 uint8_t *buf
, int nb_sectors
);
2403 static int qcow_read_snapshots(BlockDriverState
*bs
);
2404 static void qcow_free_snapshots(BlockDriverState
*bs
);
2405 static int refcount_init(BlockDriverState
*bs
);
2406 static void refcount_close(BlockDriverState
*bs
);
2407 static int get_refcount(BlockDriverState
*bs
, int64_t cluster_index
);
2408 static int update_cluster_refcount(BlockDriverState
*bs
,
2409 int64_t cluster_index
,
2411 static void update_refcount(BlockDriverState
*bs
,
2412 int64_t offset
, int64_t length
,
2414 static int64_t alloc_clusters(BlockDriverState
*bs
, int64_t size
);
2415 static int64_t alloc_bytes(BlockDriverState
*bs
, int size
);
2416 static void free_clusters(BlockDriverState
*bs
,
2417 int64_t offset
, int64_t size
);
2419 static void check_refcounts(BlockDriverState
*bs
);
2422 static int qcow_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
2424 const QCowHeader
*cow_header
= (const void *)buf
;
2426 if (buf_size
>= sizeof(QCowHeader
) &&
2427 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
2428 be32_to_cpu(cow_header
->version
) == QCOW_VERSION
)
2434 static int qcow_open(BlockDriverState
*bs
, const char *filename
, int flags
)
2436 BDRVQcowState
*s
= bs
->opaque
;
2437 int len
, i
, shift
, ret
;
2440 ret
= bdrv_file_open(&s
->hd
, filename
, flags
);
2443 if (bdrv_pread(s
->hd
, 0, &header
, sizeof(header
)) != sizeof(header
))
2445 be32_to_cpus(&header
.magic
);
2446 be32_to_cpus(&header
.version
);
2447 be64_to_cpus(&header
.backing_file_offset
);
2448 be32_to_cpus(&header
.backing_file_size
);
2449 be64_to_cpus(&header
.size
);
2450 be32_to_cpus(&header
.cluster_bits
);
2451 be32_to_cpus(&header
.crypt_method
);
2452 be64_to_cpus(&header
.l1_table_offset
);
2453 be32_to_cpus(&header
.l1_size
);
2454 be64_to_cpus(&header
.refcount_table_offset
);
2455 be32_to_cpus(&header
.refcount_table_clusters
);
2456 be64_to_cpus(&header
.snapshots_offset
);
2457 be32_to_cpus(&header
.nb_snapshots
);
2459 if (header
.magic
!= QCOW_MAGIC
|| header
.version
!= QCOW_VERSION
)
2461 if (header
.size
<= 1 ||
2462 header
.cluster_bits
< 9 ||
2463 header
.cluster_bits
> 16)
2465 if (header
.crypt_method
> QCOW_CRYPT_AES
)
2467 s
->crypt_method_header
= header
.crypt_method
;
2468 if (s
->crypt_method_header
)
2470 s
->cluster_bits
= header
.cluster_bits
;
2471 s
->cluster_size
= 1 << s
->cluster_bits
;
2472 s
->cluster_sectors
= 1 << (s
->cluster_bits
- 9);
2473 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
2474 s
->l2_size
= 1 << s
->l2_bits
;
2475 bs
->total_sectors
= header
.size
/ 512;
2476 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
2477 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
2478 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
2479 s
->refcount_table_offset
= header
.refcount_table_offset
;
2480 s
->refcount_table_size
=
2481 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
2483 s
->snapshots_offset
= header
.snapshots_offset
;
2484 s
->nb_snapshots
= header
.nb_snapshots
;
2486 /* read the level 1 table */
2487 s
->l1_size
= header
.l1_size
;
2488 shift
= s
->cluster_bits
+ s
->l2_bits
;
2489 s
->l1_vm_state_index
= (header
.size
+ (1LL << shift
) - 1) >> shift
;
2490 /* the L1 table must contain at least enough entries to put
2491 header.size bytes */
2492 if (s
->l1_size
< s
->l1_vm_state_index
)
2494 s
->l1_table_offset
= header
.l1_table_offset
;
2495 s
->l1_table
= qemu_malloc(s
->l1_size
* sizeof(uint64_t));
2498 if (bdrv_pread(s
->hd
, s
->l1_table_offset
, s
->l1_table
, s
->l1_size
* sizeof(uint64_t)) !=
2499 s
->l1_size
* sizeof(uint64_t))
2501 for(i
= 0;i
< s
->l1_size
; i
++) {
2502 be64_to_cpus(&s
->l1_table
[i
]);
2504 /* alloc L2 cache */
2505 s
->l2_cache
= qemu_malloc(s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
2508 s
->cluster_cache
= qemu_malloc(s
->cluster_size
);
2509 if (!s
->cluster_cache
)
2511 /* one more sector for decompressed data alignment */
2512 s
->cluster_data
= qemu_malloc(s
->cluster_size
+ 512);
2513 if (!s
->cluster_data
)
2515 s
->cluster_cache_offset
= -1;
2517 if (refcount_init(bs
) < 0)
2520 /* read the backing file name */
2521 if (header
.backing_file_offset
!= 0) {
2522 len
= header
.backing_file_size
;
2525 if (bdrv_pread(s
->hd
, header
.backing_file_offset
, bs
->backing_file
, len
) != len
)
2527 bs
->backing_file
[len
] = '\0';
2529 if (qcow_read_snapshots(bs
) < 0)
2533 check_refcounts(bs
);
2538 qcow_free_snapshots(bs
);
2540 qemu_free(s
->l1_table
);
2541 qemu_free(s
->l2_cache
);
2542 qemu_free(s
->cluster_cache
);
2543 qemu_free(s
->cluster_data
);
2548 static int qcow_set_key(BlockDriverState
*bs
, const char *key
)
2550 BDRVQcowState
*s
= bs
->opaque
;
2554 memset(keybuf
, 0, 16);
2558 /* XXX: we could compress the chars to 7 bits to increase
2560 for(i
= 0;i
< len
;i
++) {
2563 s
->crypt_method
= s
->crypt_method_header
;
2565 if (AES_set_encrypt_key(keybuf
, 128, &s
->aes_encrypt_key
) != 0)
2567 if (AES_set_decrypt_key(keybuf
, 128, &s
->aes_decrypt_key
) != 0)
2577 AES_encrypt(in
, tmp
, &s
->aes_encrypt_key
);
2578 AES_decrypt(tmp
, out
, &s
->aes_decrypt_key
);
2579 for(i
= 0; i
< 16; i
++)
2580 printf(" %02x", tmp
[i
]);
2582 for(i
= 0; i
< 16; i
++)
2583 printf(" %02x", out
[i
]);
2590 /* The crypt function is compatible with the linux cryptoloop
2591 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
2593 static void encrypt_sectors(BDRVQcowState
*s
, int64_t sector_num
,
2594 uint8_t *out_buf
, const uint8_t *in_buf
,
2595 int nb_sectors
, int enc
,
2604 for(i
= 0; i
< nb_sectors
; i
++) {
2605 ivec
.ll
[0] = cpu_to_le64(sector_num
);
2607 AES_cbc_encrypt(in_buf
, out_buf
, 512, key
,
2615 static int copy_sectors(BlockDriverState
*bs
, uint64_t start_sect
,
2616 uint64_t cluster_offset
, int n_start
, int n_end
)
2618 BDRVQcowState
*s
= bs
->opaque
;
2621 n
= n_end
- n_start
;
2624 ret
= qcow_read(bs
, start_sect
+ n_start
, s
->cluster_data
, n
);
2627 if (s
->crypt_method
) {
2628 encrypt_sectors(s
, start_sect
+ n_start
,
2630 s
->cluster_data
, n
, 1,
2631 &s
->aes_encrypt_key
);
2633 ret
= bdrv_write(s
->hd
, (cluster_offset
>> 9) + n_start
,
2634 s
->cluster_data
, n
);
2640 static void l2_cache_reset(BlockDriverState
*bs
)
2642 BDRVQcowState
*s
= bs
->opaque
;
2644 memset(s
->l2_cache
, 0, s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
2645 memset(s
->l2_cache_offsets
, 0, L2_CACHE_SIZE
* sizeof(uint64_t));
2646 memset(s
->l2_cache_counts
, 0, L2_CACHE_SIZE
* sizeof(uint32_t));
2649 static inline int l2_cache_new_entry(BlockDriverState
*bs
)
2651 BDRVQcowState
*s
= bs
->opaque
;
2655 /* find a new entry in the least used one */
2657 min_count
= 0xffffffff;
2658 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
2659 if (s
->l2_cache_counts
[i
] < min_count
) {
2660 min_count
= s
->l2_cache_counts
[i
];
2667 static int64_t align_offset(int64_t offset
, int n
)
2669 offset
= (offset
+ n
- 1) & ~(n
- 1);
2673 static int grow_l1_table(BlockDriverState
*bs
, int min_size
)
2675 BDRVQcowState
*s
= bs
->opaque
;
2676 int new_l1_size
, new_l1_size2
, ret
, i
;
2677 uint64_t *new_l1_table
;
2678 uint64_t new_l1_table_offset
;
2682 new_l1_size
= s
->l1_size
;
2683 if (min_size
<= new_l1_size
)
2685 while (min_size
> new_l1_size
) {
2686 new_l1_size
= (new_l1_size
* 3 + 1) / 2;
2689 printf("grow l1_table from %d to %d\n", s
->l1_size
, new_l1_size
);
2692 new_l1_size2
= sizeof(uint64_t) * new_l1_size
;
2693 new_l1_table
= qemu_mallocz(new_l1_size2
);
2696 memcpy(new_l1_table
, s
->l1_table
, s
->l1_size
* sizeof(uint64_t));
2698 /* write new table (align to cluster) */
2699 new_l1_table_offset
= alloc_clusters(bs
, new_l1_size2
);
2701 for(i
= 0; i
< s
->l1_size
; i
++)
2702 new_l1_table
[i
] = cpu_to_be64(new_l1_table
[i
]);
2703 ret
= bdrv_pwrite(s
->hd
, new_l1_table_offset
, new_l1_table
, new_l1_size2
);
2704 if (ret
!= new_l1_size2
)
2706 for(i
= 0; i
< s
->l1_size
; i
++)
2707 new_l1_table
[i
] = be64_to_cpu(new_l1_table
[i
]);
2710 data64
= cpu_to_be64(new_l1_table_offset
);
2711 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, l1_table_offset
),
2712 &data64
, sizeof(data64
)) != sizeof(data64
))
2714 data32
= cpu_to_be32(new_l1_size
);
2715 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, l1_size
),
2716 &data32
, sizeof(data32
)) != sizeof(data32
))
2718 qemu_free(s
->l1_table
);
2719 free_clusters(bs
, s
->l1_table_offset
, s
->l1_size
* sizeof(uint64_t));
2720 s
->l1_table_offset
= new_l1_table_offset
;
2721 s
->l1_table
= new_l1_table
;
2722 s
->l1_size
= new_l1_size
;
2725 qemu_free(s
->l1_table
);
2731 * 0 not to allocate.
2733 * 1 to allocate a normal cluster (for sector indexes 'n_start' to
2736 * 2 to allocate a compressed cluster of size
2737 * 'compressed_size'. 'compressed_size' must be > 0 and <
2740 * return 0 if not allocated.
2742 static uint64_t get_cluster_offset(BlockDriverState
*bs
,
2743 uint64_t offset
, int allocate
,
2744 int compressed_size
,
2745 int n_start
, int n_end
)
2747 BDRVQcowState
*s
= bs
->opaque
;
2748 int min_index
, i
, j
, l1_index
, l2_index
, ret
;
2749 uint64_t l2_offset
, *l2_table
, cluster_offset
, tmp
, old_l2_offset
;
2751 l1_index
= offset
>> (s
->l2_bits
+ s
->cluster_bits
);
2752 if (l1_index
>= s
->l1_size
) {
2753 /* outside l1 table is allowed: we grow the table if needed */
2756 if (grow_l1_table(bs
, l1_index
+ 1) < 0)
2759 l2_offset
= s
->l1_table
[l1_index
];
2764 old_l2_offset
= l2_offset
;
2765 /* allocate a new l2 entry */
2766 l2_offset
= alloc_clusters(bs
, s
->l2_size
* sizeof(uint64_t));
2767 /* update the L1 entry */
2768 s
->l1_table
[l1_index
] = l2_offset
| QCOW_OFLAG_COPIED
;
2769 tmp
= cpu_to_be64(l2_offset
| QCOW_OFLAG_COPIED
);
2770 if (bdrv_pwrite(s
->hd
, s
->l1_table_offset
+ l1_index
* sizeof(tmp
),
2771 &tmp
, sizeof(tmp
)) != sizeof(tmp
))
2773 min_index
= l2_cache_new_entry(bs
);
2774 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
2776 if (old_l2_offset
== 0) {
2777 memset(l2_table
, 0, s
->l2_size
* sizeof(uint64_t));
2779 if (bdrv_pread(s
->hd
, old_l2_offset
,
2780 l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
2781 s
->l2_size
* sizeof(uint64_t))
2784 if (bdrv_pwrite(s
->hd
, l2_offset
,
2785 l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
2786 s
->l2_size
* sizeof(uint64_t))
2789 if (!(l2_offset
& QCOW_OFLAG_COPIED
)) {
2791 free_clusters(bs
, l2_offset
, s
->l2_size
* sizeof(uint64_t));
2795 l2_offset
&= ~QCOW_OFLAG_COPIED
;
2797 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
2798 if (l2_offset
== s
->l2_cache_offsets
[i
]) {
2799 /* increment the hit count */
2800 if (++s
->l2_cache_counts
[i
] == 0xffffffff) {
2801 for(j
= 0; j
< L2_CACHE_SIZE
; j
++) {
2802 s
->l2_cache_counts
[j
] >>= 1;
2805 l2_table
= s
->l2_cache
+ (i
<< s
->l2_bits
);
2809 /* not found: load a new entry in the least used one */
2810 min_index
= l2_cache_new_entry(bs
);
2811 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
2812 if (bdrv_pread(s
->hd
, l2_offset
, l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
2813 s
->l2_size
* sizeof(uint64_t))
2816 s
->l2_cache_offsets
[min_index
] = l2_offset
;
2817 s
->l2_cache_counts
[min_index
] = 1;
2819 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
2820 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
2821 if (!cluster_offset
) {
2823 return cluster_offset
;
2824 } else if (!(cluster_offset
& QCOW_OFLAG_COPIED
)) {
2826 return cluster_offset
;
2827 /* free the cluster */
2828 if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
2830 nb_csectors
= ((cluster_offset
>> s
->csize_shift
) &
2832 free_clusters(bs
, (cluster_offset
& s
->cluster_offset_mask
) & ~511,
2835 free_clusters(bs
, cluster_offset
, s
->cluster_size
);
2838 cluster_offset
&= ~QCOW_OFLAG_COPIED
;
2839 return cluster_offset
;
2841 if (allocate
== 1) {
2842 /* allocate a new cluster */
2843 cluster_offset
= alloc_clusters(bs
, s
->cluster_size
);
2845 /* we must initialize the cluster content which won't be
2847 if ((n_end
- n_start
) < s
->cluster_sectors
) {
2848 uint64_t start_sect
;
2850 start_sect
= (offset
& ~(s
->cluster_size
- 1)) >> 9;
2851 ret
= copy_sectors(bs
, start_sect
,
2852 cluster_offset
, 0, n_start
);
2855 ret
= copy_sectors(bs
, start_sect
,
2856 cluster_offset
, n_end
, s
->cluster_sectors
);
2860 tmp
= cpu_to_be64(cluster_offset
| QCOW_OFLAG_COPIED
);
2863 cluster_offset
= alloc_bytes(bs
, compressed_size
);
2864 nb_csectors
= ((cluster_offset
+ compressed_size
- 1) >> 9) -
2865 (cluster_offset
>> 9);
2866 cluster_offset
|= QCOW_OFLAG_COMPRESSED
|
2867 ((uint64_t)nb_csectors
<< s
->csize_shift
);
2868 /* compressed clusters never have the copied flag */
2869 tmp
= cpu_to_be64(cluster_offset
);
2871 /* update L2 table */
2872 l2_table
[l2_index
] = tmp
;
2873 if (bdrv_pwrite(s
->hd
,
2874 l2_offset
+ l2_index
* sizeof(tmp
), &tmp
, sizeof(tmp
)) != sizeof(tmp
))
2876 return cluster_offset
;
2879 static int qcow_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
2880 int nb_sectors
, int *pnum
)
2882 BDRVQcowState
*s
= bs
->opaque
;
2883 int index_in_cluster
, n
;
2884 uint64_t cluster_offset
;
2886 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 0, 0, 0, 0);
2887 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
2888 n
= s
->cluster_sectors
- index_in_cluster
;
2892 return (cluster_offset
!= 0);
2895 static int decompress_buffer(uint8_t *out_buf
, int out_buf_size
,
2896 const uint8_t *buf
, int buf_size
)
2898 z_stream strm1
, *strm
= &strm1
;
2901 memset(strm
, 0, sizeof(*strm
));
2903 strm
->next_in
= (uint8_t *)buf
;
2904 strm
->avail_in
= buf_size
;
2905 strm
->next_out
= out_buf
;
2906 strm
->avail_out
= out_buf_size
;
2908 ret
= inflateInit2(strm
, -12);
2911 ret
= inflate(strm
, Z_FINISH
);
2912 out_len
= strm
->next_out
- out_buf
;
2913 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) ||
2914 out_len
!= out_buf_size
) {
2922 static int decompress_cluster(BDRVQcowState
*s
, uint64_t cluster_offset
)
2924 int ret
, csize
, nb_csectors
, sector_offset
;
2927 coffset
= cluster_offset
& s
->cluster_offset_mask
;
2928 if (s
->cluster_cache_offset
!= coffset
) {
2929 nb_csectors
= ((cluster_offset
>> s
->csize_shift
) & s
->csize_mask
) + 1;
2930 sector_offset
= coffset
& 511;
2931 csize
= nb_csectors
* 512 - sector_offset
;
2932 ret
= bdrv_read(s
->hd
, coffset
>> 9, s
->cluster_data
, nb_csectors
);
2936 if (decompress_buffer(s
->cluster_cache
, s
->cluster_size
,
2937 s
->cluster_data
+ sector_offset
, csize
) < 0) {
2940 s
->cluster_cache_offset
= coffset
;
2945 /* handle reading after the end of the backing file */
2946 static int backing_read1(BlockDriverState
*bs
,
2947 int64_t sector_num
, uint8_t *buf
, int nb_sectors
)
2950 if ((sector_num
+ nb_sectors
) <= bs
->total_sectors
)
2952 if (sector_num
>= bs
->total_sectors
)
2955 n1
= bs
->total_sectors
- sector_num
;
2956 memset(buf
+ n1
* 512, 0, 512 * (nb_sectors
- n1
));
2960 static int qcow_read(BlockDriverState
*bs
, int64_t sector_num
,
2961 uint8_t *buf
, int nb_sectors
)
2963 BDRVQcowState
*s
= bs
->opaque
;
2964 int ret
, index_in_cluster
, n
, n1
;
2965 uint64_t cluster_offset
;
2967 while (nb_sectors
> 0) {
2968 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 0, 0, 0, 0);
2969 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
2970 n
= s
->cluster_sectors
- index_in_cluster
;
2973 if (!cluster_offset
) {
2974 if (bs
->backing_hd
) {
2975 /* read from the base image */
2976 n1
= backing_read1(bs
->backing_hd
, sector_num
, buf
, n
);
2978 ret
= bdrv_read(bs
->backing_hd
, sector_num
, buf
, n1
);
2983 memset(buf
, 0, 512 * n
);
2985 } else if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
2986 if (decompress_cluster(s
, cluster_offset
) < 0)
2988 memcpy(buf
, s
->cluster_cache
+ index_in_cluster
* 512, 512 * n
);
2990 ret
= bdrv_pread(s
->hd
, cluster_offset
+ index_in_cluster
* 512, buf
, n
* 512);
2993 if (s
->crypt_method
) {
2994 encrypt_sectors(s
, sector_num
, buf
, buf
, n
, 0,
2995 &s
->aes_decrypt_key
);
3005 static int qcow_write(BlockDriverState
*bs
, int64_t sector_num
,
3006 const uint8_t *buf
, int nb_sectors
)
3008 BDRVQcowState
*s
= bs
->opaque
;
3009 int ret
, index_in_cluster
, n
;
3010 uint64_t cluster_offset
;
3012 while (nb_sectors
> 0) {
3013 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
3014 n
= s
->cluster_sectors
- index_in_cluster
;
3017 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 1, 0,
3019 index_in_cluster
+ n
);
3020 if (!cluster_offset
)
3022 if (s
->crypt_method
) {
3023 encrypt_sectors(s
, sector_num
, s
->cluster_data
, buf
, n
, 1,
3024 &s
->aes_encrypt_key
);
3025 ret
= bdrv_pwrite(s
->hd
, cluster_offset
+ index_in_cluster
* 512,
3026 s
->cluster_data
, n
* 512);
3028 ret
= bdrv_pwrite(s
->hd
, cluster_offset
+ index_in_cluster
* 512, buf
, n
* 512);
3036 s
->cluster_cache_offset
= -1; /* disable compressed cache */
3040 typedef struct QCowAIOCB
{
3041 BlockDriverAIOCB common
;
3046 uint64_t cluster_offset
;
3047 uint8_t *cluster_data
;
3048 BlockDriverAIOCB
*hd_aiocb
;
3051 static void qcow_aio_read_cb(void *opaque
, int ret
)
3053 QCowAIOCB
*acb
= opaque
;
3054 BlockDriverState
*bs
= acb
->common
.bs
;
3055 BDRVQcowState
*s
= bs
->opaque
;
3056 int index_in_cluster
, n1
;
3058 acb
->hd_aiocb
= NULL
;
3061 acb
->common
.cb(acb
->common
.opaque
, ret
);
3062 qemu_aio_release(acb
);
3067 /* post process the read buffer */
3068 if (!acb
->cluster_offset
) {
3070 } else if (acb
->cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
3073 if (s
->crypt_method
) {
3074 encrypt_sectors(s
, acb
->sector_num
, acb
->buf
, acb
->buf
,
3076 &s
->aes_decrypt_key
);
3080 acb
->nb_sectors
-= acb
->n
;
3081 acb
->sector_num
+= acb
->n
;
3082 acb
->buf
+= acb
->n
* 512;
3084 if (acb
->nb_sectors
== 0) {
3085 /* request completed */
3086 acb
->common
.cb(acb
->common
.opaque
, 0);
3087 qemu_aio_release(acb
);
3091 /* prepare next AIO request */
3092 acb
->cluster_offset
= get_cluster_offset(bs
, acb
->sector_num
<< 9,
3094 index_in_cluster
= acb
->sector_num
& (s
->cluster_sectors
- 1);
3095 acb
->n
= s
->cluster_sectors
- index_in_cluster
;
3096 if (acb
->n
> acb
->nb_sectors
)
3097 acb
->n
= acb
->nb_sectors
;
3099 if (!acb
->cluster_offset
) {
3100 if (bs
->backing_hd
) {
3101 /* read from the base image */
3102 n1
= backing_read1(bs
->backing_hd
, acb
->sector_num
,
3105 acb
->hd_aiocb
= bdrv_aio_read(bs
->backing_hd
, acb
->sector_num
,
3106 acb
->buf
, acb
->n
, qcow_aio_read_cb
, acb
);
3107 if (acb
->hd_aiocb
== NULL
)
3113 /* Note: in this case, no need to wait */
3114 memset(acb
->buf
, 0, 512 * acb
->n
);
3117 } else if (acb
->cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
3118 /* add AIO support for compressed blocks ? */
3119 if (decompress_cluster(s
, acb
->cluster_offset
) < 0)
3122 s
->cluster_cache
+ index_in_cluster
* 512, 512 * acb
->n
);
3125 if ((acb
->cluster_offset
& 511) != 0) {
3129 acb
->hd_aiocb
= bdrv_aio_read(s
->hd
,
3130 (acb
->cluster_offset
>> 9) + index_in_cluster
,
3131 acb
->buf
, acb
->n
, qcow_aio_read_cb
, acb
);
3132 if (acb
->hd_aiocb
== NULL
)
3137 static QCowAIOCB
*qcow_aio_setup(BlockDriverState
*bs
,
3138 int64_t sector_num
, uint8_t *buf
, int nb_sectors
,
3139 BlockDriverCompletionFunc
*cb
, void *opaque
)
3143 acb
= qemu_aio_get(bs
, cb
, opaque
);
3146 acb
->hd_aiocb
= NULL
;
3147 acb
->sector_num
= sector_num
;
3149 acb
->nb_sectors
= nb_sectors
;
3151 acb
->cluster_offset
= 0;
3155 static BlockDriverAIOCB
*qcow_aio_read(BlockDriverState
*bs
,
3156 int64_t sector_num
, uint8_t *buf
, int nb_sectors
,
3157 BlockDriverCompletionFunc
*cb
, void *opaque
)
3161 acb
= qcow_aio_setup(bs
, sector_num
, buf
, nb_sectors
, cb
, opaque
);
3165 qcow_aio_read_cb(acb
, 0);
3166 return &acb
->common
;
3169 static void qcow_aio_write_cb(void *opaque
, int ret
)
3171 QCowAIOCB
*acb
= opaque
;
3172 BlockDriverState
*bs
= acb
->common
.bs
;
3173 BDRVQcowState
*s
= bs
->opaque
;
3174 int index_in_cluster
;
3175 uint64_t cluster_offset
;
3176 const uint8_t *src_buf
;
3178 acb
->hd_aiocb
= NULL
;
3182 acb
->common
.cb(acb
->common
.opaque
, ret
);
3183 qemu_aio_release(acb
);
3187 acb
->nb_sectors
-= acb
->n
;
3188 acb
->sector_num
+= acb
->n
;
3189 acb
->buf
+= acb
->n
* 512;
3191 if (acb
->nb_sectors
== 0) {
3192 /* request completed */
3193 acb
->common
.cb(acb
->common
.opaque
, 0);
3194 qemu_aio_release(acb
);
3198 index_in_cluster
= acb
->sector_num
& (s
->cluster_sectors
- 1);
3199 acb
->n
= s
->cluster_sectors
- index_in_cluster
;
3200 if (acb
->n
> acb
->nb_sectors
)
3201 acb
->n
= acb
->nb_sectors
;
3202 cluster_offset
= get_cluster_offset(bs
, acb
->sector_num
<< 9, 1, 0,
3204 index_in_cluster
+ acb
->n
);
3205 if (!cluster_offset
|| (cluster_offset
& 511) != 0) {
3209 if (s
->crypt_method
) {
3210 if (!acb
->cluster_data
) {
3211 acb
->cluster_data
= qemu_mallocz(s
->cluster_size
);
3212 if (!acb
->cluster_data
) {
3217 encrypt_sectors(s
, acb
->sector_num
, acb
->cluster_data
, acb
->buf
,
3218 acb
->n
, 1, &s
->aes_encrypt_key
);
3219 src_buf
= acb
->cluster_data
;
3223 acb
->hd_aiocb
= bdrv_aio_write(s
->hd
,
3224 (cluster_offset
>> 9) + index_in_cluster
,
3226 qcow_aio_write_cb
, acb
);
3227 if (acb
->hd_aiocb
== NULL
)
3231 static BlockDriverAIOCB
*qcow_aio_write(BlockDriverState
*bs
,
3232 int64_t sector_num
, const uint8_t *buf
, int nb_sectors
,
3233 BlockDriverCompletionFunc
*cb
, void *opaque
)
3235 BDRVQcowState
*s
= bs
->opaque
;
3238 s
->cluster_cache_offset
= -1; /* disable compressed cache */
3240 acb
= qcow_aio_setup(bs
, sector_num
, (uint8_t*)buf
, nb_sectors
, cb
, opaque
);
3244 qcow_aio_write_cb(acb
, 0);
3245 return &acb
->common
;
3248 static void qcow_aio_cancel(BlockDriverAIOCB
*blockacb
)
3250 QCowAIOCB
*acb
= (QCowAIOCB
*)blockacb
;
3252 bdrv_aio_cancel(acb
->hd_aiocb
);
3253 qemu_aio_release(acb
);
3256 static void qcow_close(BlockDriverState
*bs
)
3258 BDRVQcowState
*s
= bs
->opaque
;
3259 qemu_free(s
->l1_table
);
3260 qemu_free(s
->l2_cache
);
3261 qemu_free(s
->cluster_cache
);
3262 qemu_free(s
->cluster_data
);
3267 /* XXX: use std qcow open function ? */
3268 typedef struct QCowCreateState
{
3271 uint16_t *refcount_block
;
3272 uint64_t *refcount_table
;
3273 int64_t l1_table_offset
;
3274 int64_t refcount_table_offset
;
3275 int64_t refcount_block_offset
;
3278 static void create_refcount_update(QCowCreateState
*s
,
3279 int64_t offset
, int64_t size
)
3282 int64_t start
, last
, cluster_offset
;
3285 start
= offset
& ~(s
->cluster_size
- 1);
3286 last
= (offset
+ size
- 1) & ~(s
->cluster_size
- 1);
3287 for(cluster_offset
= start
; cluster_offset
<= last
;
3288 cluster_offset
+= s
->cluster_size
) {
3289 p
= &s
->refcount_block
[cluster_offset
>> s
->cluster_bits
];
3290 refcount
= be16_to_cpu(*p
);
3292 *p
= cpu_to_be16(refcount
);
3296 static int qcow_create(const char *filename
, int64_t total_size
,
3297 const char *backing_file
, int flags
)
3299 int fd
, header_size
, backing_filename_len
, l1_size
, i
, shift
, l2_bits
;
3301 uint64_t tmp
, offset
;
3302 QCowCreateState s1
, *s
= &s1
;
3304 memset(s
, 0, sizeof(*s
));
3306 fd
= open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, 0644);
3309 memset(&header
, 0, sizeof(header
));
3310 header
.magic
= cpu_to_be32(QCOW_MAGIC
);
3311 header
.version
= cpu_to_be32(QCOW_VERSION
);
3312 header
.size
= cpu_to_be64(total_size
* 512);
3313 header_size
= sizeof(header
);
3314 backing_filename_len
= 0;
3316 header
.backing_file_offset
= cpu_to_be64(header_size
);
3317 backing_filename_len
= strlen(backing_file
);
3318 header
.backing_file_size
= cpu_to_be32(backing_filename_len
);
3319 header_size
+= backing_filename_len
;
3321 s
->cluster_bits
= 12; /* 4 KB clusters */
3322 s
->cluster_size
= 1 << s
->cluster_bits
;
3323 header
.cluster_bits
= cpu_to_be32(s
->cluster_bits
);
3324 header_size
= (header_size
+ 7) & ~7;
3326 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
3328 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
3330 l2_bits
= s
->cluster_bits
- 3;
3331 shift
= s
->cluster_bits
+ l2_bits
;
3332 l1_size
= (((total_size
* 512) + (1LL << shift
) - 1) >> shift
);
3333 offset
= align_offset(header_size
, s
->cluster_size
);
3334 s
->l1_table_offset
= offset
;
3335 header
.l1_table_offset
= cpu_to_be64(s
->l1_table_offset
);
3336 header
.l1_size
= cpu_to_be32(l1_size
);
3337 offset
+= align_offset(l1_size
* sizeof(uint64_t), s
->cluster_size
);
3339 s
->refcount_table
= qemu_mallocz(s
->cluster_size
);
3340 if (!s
->refcount_table
)
3342 s
->refcount_block
= qemu_mallocz(s
->cluster_size
);
3343 if (!s
->refcount_block
)
3346 s
->refcount_table_offset
= offset
;
3347 header
.refcount_table_offset
= cpu_to_be64(offset
);
3348 header
.refcount_table_clusters
= cpu_to_be32(1);
3349 offset
+= s
->cluster_size
;
3351 s
->refcount_table
[0] = cpu_to_be64(offset
);
3352 s
->refcount_block_offset
= offset
;
3353 offset
+= s
->cluster_size
;
3355 /* update refcounts */
3356 create_refcount_update(s
, 0, header_size
);
3357 create_refcount_update(s
, s
->l1_table_offset
, l1_size
* sizeof(uint64_t));
3358 create_refcount_update(s
, s
->refcount_table_offset
, s
->cluster_size
);
3359 create_refcount_update(s
, s
->refcount_block_offset
, s
->cluster_size
);
3361 /* write all the data */
3362 write(fd
, &header
, sizeof(header
));
3364 write(fd
, backing_file
, backing_filename_len
);
3366 lseek(fd
, s
->l1_table_offset
, SEEK_SET
);
3368 for(i
= 0;i
< l1_size
; i
++) {
3369 write(fd
, &tmp
, sizeof(tmp
));
3371 lseek(fd
, s
->refcount_table_offset
, SEEK_SET
);
3372 write(fd
, s
->refcount_table
, s
->cluster_size
);
3374 lseek(fd
, s
->refcount_block_offset
, SEEK_SET
);
3375 write(fd
, s
->refcount_block
, s
->cluster_size
);
3377 qemu_free(s
->refcount_table
);
3378 qemu_free(s
->refcount_block
);
3382 qemu_free(s
->refcount_table
);
3383 qemu_free(s
->refcount_block
);
3388 static int qcow_make_empty(BlockDriverState
*bs
)
3391 /* XXX: not correct */
3392 BDRVQcowState
*s
= bs
->opaque
;
3393 uint32_t l1_length
= s
->l1_size
* sizeof(uint64_t);
3396 memset(s
->l1_table
, 0, l1_length
);
3397 if (bdrv_pwrite(s
->hd
, s
->l1_table_offset
, s
->l1_table
, l1_length
) < 0)
3399 ret
= bdrv_truncate(s
->hd
, s
->l1_table_offset
+ l1_length
);
3408 /* XXX: put compressed sectors first, then all the cluster aligned
3409 tables to avoid losing bytes in alignment */
3410 static int qcow_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
3411 const uint8_t *buf
, int nb_sectors
)
3413 BDRVQcowState
*s
= bs
->opaque
;
3417 uint64_t cluster_offset
;
3419 if (nb_sectors
== 0) {
3420 /* align end of file to a sector boundary to ease reading with
3421 sector based I/Os */
3422 cluster_offset
= bdrv_getlength(s
->hd
);
3423 cluster_offset
= (cluster_offset
+ 511) & ~511;
3424 bdrv_truncate(s
->hd
, cluster_offset
);
3428 if (nb_sectors
!= s
->cluster_sectors
)
3431 out_buf
= qemu_malloc(s
->cluster_size
+ (s
->cluster_size
/ 1000) + 128);
3435 /* best compression, small window, no zlib header */
3436 memset(&strm
, 0, sizeof(strm
));
3437 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
3439 9, Z_DEFAULT_STRATEGY
);
3445 strm
.avail_in
= s
->cluster_size
;
3446 strm
.next_in
= (uint8_t *)buf
;
3447 strm
.avail_out
= s
->cluster_size
;
3448 strm
.next_out
= out_buf
;
3450 ret
= deflate(&strm
, Z_FINISH
);
3451 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
3456 out_len
= strm
.next_out
- out_buf
;
3460 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
3461 /* could not compress: write normal cluster */
3462 qcow_write(bs
, sector_num
, buf
, s
->cluster_sectors
);
3464 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, 2,
3466 cluster_offset
&= s
->cluster_offset_mask
;
3467 if (bdrv_pwrite(s
->hd
, cluster_offset
, out_buf
, out_len
) != out_len
) {
3477 static void qcow_flush(BlockDriverState
*bs
)
3479 BDRVQcowState
*s
= bs
->opaque
;
3483 static int qcow_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3485 BDRVQcowState
*s
= bs
->opaque
;
3486 bdi
->cluster_size
= s
->cluster_size
;
3487 bdi
->vm_state_offset
= (int64_t)s
->l1_vm_state_index
<<
3488 (s
->cluster_bits
+ s
->l2_bits
);
3492 /*********************************************************/
3493 /* snapshot support */
3495 /* update the refcounts of snapshots and the copied flag */
3496 static int update_snapshot_refcount(BlockDriverState
*bs
,
3497 int64_t l1_table_offset
,
3501 BDRVQcowState
*s
= bs
->opaque
;
3502 uint64_t *l1_table
, *l2_table
, l2_offset
, offset
, l1_size2
, l1_allocated
;
3503 int64_t old_offset
, old_l2_offset
;
3504 int l2_size
, i
, j
, l1_modified
, l2_modified
, nb_csectors
, refcount
;
3510 l1_size2
= l1_size
* sizeof(uint64_t);
3512 if (l1_table_offset
!= s
->l1_table_offset
) {
3513 l1_table
= qemu_malloc(l1_size2
);
3517 if (bdrv_pread(s
->hd
, l1_table_offset
,
3518 l1_table
, l1_size2
) != l1_size2
)
3520 for(i
= 0;i
< l1_size
; i
++)
3521 be64_to_cpus(&l1_table
[i
]);
3523 assert(l1_size
== s
->l1_size
);
3524 l1_table
= s
->l1_table
;
3528 l2_size
= s
->l2_size
* sizeof(uint64_t);
3529 l2_table
= qemu_malloc(l2_size
);
3533 for(i
= 0; i
< l1_size
; i
++) {
3534 l2_offset
= l1_table
[i
];
3536 old_l2_offset
= l2_offset
;
3537 l2_offset
&= ~QCOW_OFLAG_COPIED
;
3539 if (bdrv_pread(s
->hd
, l2_offset
, l2_table
, l2_size
) != l2_size
)
3541 for(j
= 0; j
< s
->l2_size
; j
++) {
3542 offset
= be64_to_cpu(l2_table
[j
]);
3544 old_offset
= offset
;
3545 offset
&= ~QCOW_OFLAG_COPIED
;
3546 if (offset
& QCOW_OFLAG_COMPRESSED
) {
3547 nb_csectors
= ((offset
>> s
->csize_shift
) &
3550 update_refcount(bs
, (offset
& s
->cluster_offset_mask
) & ~511,
3551 nb_csectors
* 512, addend
);
3552 /* compressed clusters are never modified */
3556 refcount
= update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, addend
);
3558 refcount
= get_refcount(bs
, offset
>> s
->cluster_bits
);
3562 if (refcount
== 1) {
3563 offset
|= QCOW_OFLAG_COPIED
;
3565 if (offset
!= old_offset
) {
3566 l2_table
[j
] = cpu_to_be64(offset
);
3572 if (bdrv_pwrite(s
->hd
,
3573 l2_offset
, l2_table
, l2_size
) != l2_size
)
3578 refcount
= update_cluster_refcount(bs
, l2_offset
>> s
->cluster_bits
, addend
);
3580 refcount
= get_refcount(bs
, l2_offset
>> s
->cluster_bits
);
3582 if (refcount
== 1) {
3583 l2_offset
|= QCOW_OFLAG_COPIED
;
3585 if (l2_offset
!= old_l2_offset
) {
3586 l1_table
[i
] = l2_offset
;
3592 for(i
= 0; i
< l1_size
; i
++)
3593 cpu_to_be64s(&l1_table
[i
]);
3594 if (bdrv_pwrite(s
->hd
, l1_table_offset
, l1_table
,
3595 l1_size2
) != l1_size2
)
3597 for(i
= 0; i
< l1_size
; i
++)
3598 be64_to_cpus(&l1_table
[i
]);
3601 qemu_free(l1_table
);
3602 qemu_free(l2_table
);
3606 qemu_free(l1_table
);
3607 qemu_free(l2_table
);
3611 static void qcow_free_snapshots(BlockDriverState
*bs
)
3613 BDRVQcowState
*s
= bs
->opaque
;
3616 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
3617 qemu_free(s
->snapshots
[i
].name
);
3618 qemu_free(s
->snapshots
[i
].id_str
);
3620 qemu_free(s
->snapshots
);
3621 s
->snapshots
= NULL
;
3622 s
->nb_snapshots
= 0;
3625 static int qcow_read_snapshots(BlockDriverState
*bs
)
3627 BDRVQcowState
*s
= bs
->opaque
;
3628 QCowSnapshotHeader h
;
3630 int i
, id_str_size
, name_size
;
3632 uint32_t extra_data_size
;
3634 offset
= s
->snapshots_offset
;
3635 s
->snapshots
= qemu_mallocz(s
->nb_snapshots
* sizeof(QCowSnapshot
));
3638 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
3639 offset
= align_offset(offset
, 8);
3640 if (bdrv_pread(s
->hd
, offset
, &h
, sizeof(h
)) != sizeof(h
))
3642 offset
+= sizeof(h
);
3643 sn
= s
->snapshots
+ i
;
3644 sn
->l1_table_offset
= be64_to_cpu(h
.l1_table_offset
);
3645 sn
->l1_size
= be32_to_cpu(h
.l1_size
);
3646 sn
->vm_state_size
= be32_to_cpu(h
.vm_state_size
);
3647 sn
->date_sec
= be32_to_cpu(h
.date_sec
);
3648 sn
->date_nsec
= be32_to_cpu(h
.date_nsec
);
3649 sn
->vm_clock_nsec
= be64_to_cpu(h
.vm_clock_nsec
);
3650 extra_data_size
= be32_to_cpu(h
.extra_data_size
);
3652 id_str_size
= be16_to_cpu(h
.id_str_size
);
3653 name_size
= be16_to_cpu(h
.name_size
);
3655 offset
+= extra_data_size
;
3657 sn
->id_str
= qemu_malloc(id_str_size
+ 1);
3660 if (bdrv_pread(s
->hd
, offset
, sn
->id_str
, id_str_size
) != id_str_size
)
3662 offset
+= id_str_size
;
3663 sn
->id_str
[id_str_size
] = '\0';
3665 sn
->name
= qemu_malloc(name_size
+ 1);
3668 if (bdrv_pread(s
->hd
, offset
, sn
->name
, name_size
) != name_size
)
3670 offset
+= name_size
;
3671 sn
->name
[name_size
] = '\0';
3673 s
->snapshots_size
= offset
- s
->snapshots_offset
;
3676 qcow_free_snapshots(bs
);
3680 /* add at the end of the file a new list of snapshots */
3681 static int qcow_write_snapshots(BlockDriverState
*bs
)
3683 BDRVQcowState
*s
= bs
->opaque
;
3685 QCowSnapshotHeader h
;
3686 int i
, name_size
, id_str_size
, snapshots_size
;
3689 int64_t offset
, snapshots_offset
;
3691 /* compute the size of the snapshots */
3693 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
3694 sn
= s
->snapshots
+ i
;
3695 offset
= align_offset(offset
, 8);
3696 offset
+= sizeof(h
);
3697 offset
+= strlen(sn
->id_str
);
3698 offset
+= strlen(sn
->name
);
3700 snapshots_size
= offset
;
3702 snapshots_offset
= alloc_clusters(bs
, snapshots_size
);
3703 offset
= snapshots_offset
;
3705 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
3706 sn
= s
->snapshots
+ i
;
3707 memset(&h
, 0, sizeof(h
));
3708 h
.l1_table_offset
= cpu_to_be64(sn
->l1_table_offset
);
3709 h
.l1_size
= cpu_to_be32(sn
->l1_size
);
3710 h
.vm_state_size
= cpu_to_be32(sn
->vm_state_size
);
3711 h
.date_sec
= cpu_to_be32(sn
->date_sec
);
3712 h
.date_nsec
= cpu_to_be32(sn
->date_nsec
);
3713 h
.vm_clock_nsec
= cpu_to_be64(sn
->vm_clock_nsec
);
3715 id_str_size
= strlen(sn
->id_str
);
3716 name_size
= strlen(sn
->name
);
3717 h
.id_str_size
= cpu_to_be16(id_str_size
);
3718 h
.name_size
= cpu_to_be16(name_size
);
3719 offset
= align_offset(offset
, 8);
3720 if (bdrv_pwrite(s
->hd
, offset
, &h
, sizeof(h
)) != sizeof(h
))
3722 offset
+= sizeof(h
);
3723 if (bdrv_pwrite(s
->hd
, offset
, sn
->id_str
, id_str_size
) != id_str_size
)
3725 offset
+= id_str_size
;
3726 if (bdrv_pwrite(s
->hd
, offset
, sn
->name
, name_size
) != name_size
)
3728 offset
+= name_size
;
3731 /* update the various header fields */
3732 data64
= cpu_to_be64(snapshots_offset
);
3733 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, snapshots_offset
),
3734 &data64
, sizeof(data64
)) != sizeof(data64
))
3736 data32
= cpu_to_be32(s
->nb_snapshots
);
3737 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, nb_snapshots
),
3738 &data32
, sizeof(data32
)) != sizeof(data32
))
3741 /* free the old snapshot table */
3742 free_clusters(bs
, s
->snapshots_offset
, s
->snapshots_size
);
3743 s
->snapshots_offset
= snapshots_offset
;
3744 s
->snapshots_size
= snapshots_size
;
3750 static void find_new_snapshot_id(BlockDriverState
*bs
,
3751 char *id_str
, int id_str_size
)
3753 BDRVQcowState
*s
= bs
->opaque
;
3755 int i
, id
, id_max
= 0;
3757 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
3758 sn
= s
->snapshots
+ i
;
3759 id
= strtoul(sn
->id_str
, NULL
, 10);
3763 snprintf(id_str
, id_str_size
, "%d", id_max
+ 1);
3766 static int find_snapshot_by_id(BlockDriverState
*bs
, const char *id_str
)
3768 BDRVQcowState
*s
= bs
->opaque
;
3771 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
3772 if (!strcmp(s
->snapshots
[i
].id_str
, id_str
))
3778 static int find_snapshot_by_id_or_name(BlockDriverState
*bs
, const char *name
)
3780 BDRVQcowState
*s
= bs
->opaque
;
3783 ret
= find_snapshot_by_id(bs
, name
);
3786 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
3787 if (!strcmp(s
->snapshots
[i
].name
, name
))
3793 /* if no id is provided, a new one is constructed */
3794 static int qcow_snapshot_create(BlockDriverState
*bs
,
3795 QEMUSnapshotInfo
*sn_info
)
3797 BDRVQcowState
*s
= bs
->opaque
;
3798 QCowSnapshot
*snapshots1
, sn1
, *sn
= &sn1
;
3800 uint64_t *l1_table
= NULL
;
3802 memset(sn
, 0, sizeof(*sn
));
3804 if (sn_info
->id_str
[0] == '\0') {
3805 /* compute a new id */
3806 find_new_snapshot_id(bs
, sn_info
->id_str
, sizeof(sn_info
->id_str
));
3809 /* check that the ID is unique */
3810 if (find_snapshot_by_id(bs
, sn_info
->id_str
) >= 0)
3813 sn
->id_str
= qemu_strdup(sn_info
->id_str
);
3816 sn
->name
= qemu_strdup(sn_info
->name
);
3819 sn
->vm_state_size
= sn_info
->vm_state_size
;
3820 sn
->date_sec
= sn_info
->date_sec
;
3821 sn
->date_nsec
= sn_info
->date_nsec
;
3822 sn
->vm_clock_nsec
= sn_info
->vm_clock_nsec
;
3824 ret
= update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 1);
3828 /* create the L1 table of the snapshot */
3829 sn
->l1_table_offset
= alloc_clusters(bs
, s
->l1_size
* sizeof(uint64_t));
3830 sn
->l1_size
= s
->l1_size
;
3832 l1_table
= qemu_malloc(s
->l1_size
* sizeof(uint64_t));
3835 for(i
= 0; i
< s
->l1_size
; i
++) {
3836 l1_table
[i
] = cpu_to_be64(s
->l1_table
[i
]);
3838 if (bdrv_pwrite(s
->hd
, sn
->l1_table_offset
,
3839 l1_table
, s
->l1_size
* sizeof(uint64_t)) !=
3840 (s
->l1_size
* sizeof(uint64_t)))
3842 qemu_free(l1_table
);
3845 snapshots1
= qemu_malloc((s
->nb_snapshots
+ 1) * sizeof(QCowSnapshot
));
3848 memcpy(snapshots1
, s
->snapshots
, s
->nb_snapshots
* sizeof(QCowSnapshot
));
3849 s
->snapshots
= snapshots1
;
3850 s
->snapshots
[s
->nb_snapshots
++] = *sn
;
3852 if (qcow_write_snapshots(bs
) < 0)
3855 check_refcounts(bs
);
3859 qemu_free(sn
->name
);
3860 qemu_free(l1_table
);
3864 /* copy the snapshot 'snapshot_name' into the current disk image */
3865 static int qcow_snapshot_goto(BlockDriverState
*bs
,
3866 const char *snapshot_id
)
3868 BDRVQcowState
*s
= bs
->opaque
;
3870 int i
, snapshot_index
, l1_size2
;
3872 snapshot_index
= find_snapshot_by_id_or_name(bs
, snapshot_id
);
3873 if (snapshot_index
< 0)
3875 sn
= &s
->snapshots
[snapshot_index
];
3877 if (update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, -1) < 0)
3880 if (grow_l1_table(bs
, sn
->l1_size
) < 0)
3883 s
->l1_size
= sn
->l1_size
;
3884 l1_size2
= s
->l1_size
* sizeof(uint64_t);
3885 /* copy the snapshot l1 table to the current l1 table */
3886 if (bdrv_pread(s
->hd
, sn
->l1_table_offset
,
3887 s
->l1_table
, l1_size2
) != l1_size2
)
3889 if (bdrv_pwrite(s
->hd
, s
->l1_table_offset
,
3890 s
->l1_table
, l1_size2
) != l1_size2
)
3892 for(i
= 0;i
< s
->l1_size
; i
++) {
3893 be64_to_cpus(&s
->l1_table
[i
]);
3896 if (update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 1) < 0)
3900 check_refcounts(bs
);
3907 static int qcow_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
3909 BDRVQcowState
*s
= bs
->opaque
;
3911 int snapshot_index
, ret
;
3913 snapshot_index
= find_snapshot_by_id_or_name(bs
, snapshot_id
);
3914 if (snapshot_index
< 0)
3916 sn
= &s
->snapshots
[snapshot_index
];
3918 ret
= update_snapshot_refcount(bs
, sn
->l1_table_offset
, sn
->l1_size
, -1);
3921 /* must update the copied flag on the current cluster offsets */
3922 ret
= update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 0);
3925 free_clusters(bs
, sn
->l1_table_offset
, sn
->l1_size
* sizeof(uint64_t));
3927 qemu_free(sn
->id_str
);
3928 qemu_free(sn
->name
);
3929 memmove(sn
, sn
+ 1, (s
->nb_snapshots
- snapshot_index
- 1) * sizeof(*sn
));
3931 ret
= qcow_write_snapshots(bs
);
3933 /* XXX: restore snapshot if error ? */
3937 check_refcounts(bs
);
3942 static int qcow_snapshot_list(BlockDriverState
*bs
,
3943 QEMUSnapshotInfo
**psn_tab
)
3945 BDRVQcowState
*s
= bs
->opaque
;
3946 QEMUSnapshotInfo
*sn_tab
, *sn_info
;
3950 sn_tab
= qemu_mallocz(s
->nb_snapshots
* sizeof(QEMUSnapshotInfo
));
3953 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
3954 sn_info
= sn_tab
+ i
;
3955 sn
= s
->snapshots
+ i
;
3956 pstrcpy(sn_info
->id_str
, sizeof(sn_info
->id_str
),
3958 pstrcpy(sn_info
->name
, sizeof(sn_info
->name
),
3960 sn_info
->vm_state_size
= sn
->vm_state_size
;
3961 sn_info
->date_sec
= sn
->date_sec
;
3962 sn_info
->date_nsec
= sn
->date_nsec
;
3963 sn_info
->vm_clock_nsec
= sn
->vm_clock_nsec
;
3966 return s
->nb_snapshots
;
3973 /*********************************************************/
3974 /* refcount handling */
3976 static int refcount_init(BlockDriverState
*bs
)
3978 BDRVQcowState
*s
= bs
->opaque
;
3979 int ret
, refcount_table_size2
, i
;
3981 s
->refcount_block_cache
= qemu_malloc(s
->cluster_size
);
3982 if (!s
->refcount_block_cache
)
3984 refcount_table_size2
= s
->refcount_table_size
* sizeof(uint64_t);
3985 s
->refcount_table
= qemu_malloc(refcount_table_size2
);
3986 if (!s
->refcount_table
)
3988 if (s
->refcount_table_size
> 0) {
3989 ret
= bdrv_pread(s
->hd
, s
->refcount_table_offset
,
3990 s
->refcount_table
, refcount_table_size2
);
3991 if (ret
!= refcount_table_size2
)
3993 for(i
= 0; i
< s
->refcount_table_size
; i
++)
3994 be64_to_cpus(&s
->refcount_table
[i
]);
4001 static void refcount_close(BlockDriverState
*bs
)
4003 BDRVQcowState
*s
= bs
->opaque
;
4004 qemu_free(s
->refcount_block_cache
);
4005 qemu_free(s
->refcount_table
);
4009 static int load_refcount_block(BlockDriverState
*bs
,
4010 int64_t refcount_block_offset
)
4012 BDRVQcowState
*s
= bs
->opaque
;
4014 ret
= bdrv_pread(s
->hd
, refcount_block_offset
, s
->refcount_block_cache
,
4016 if (ret
!= s
->cluster_size
)
4018 s
->refcount_block_cache_offset
= refcount_block_offset
;
4022 static int get_refcount(BlockDriverState
*bs
, int64_t cluster_index
)
4024 BDRVQcowState
*s
= bs
->opaque
;
4025 int refcount_table_index
, block_index
;
4026 int64_t refcount_block_offset
;
4028 refcount_table_index
= cluster_index
>> (s
->cluster_bits
- REFCOUNT_SHIFT
);
4029 if (refcount_table_index
>= s
->refcount_table_size
)
4031 refcount_block_offset
= s
->refcount_table
[refcount_table_index
];
4032 if (!refcount_block_offset
)
4034 if (refcount_block_offset
!= s
->refcount_block_cache_offset
) {
4035 /* better than nothing: return allocated if read error */
4036 if (load_refcount_block(bs
, refcount_block_offset
) < 0)
4039 block_index
= cluster_index
&
4040 ((1 << (s
->cluster_bits
- REFCOUNT_SHIFT
)) - 1);
4041 return be16_to_cpu(s
->refcount_block_cache
[block_index
]);
4044 /* return < 0 if error */
4045 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, int64_t size
)
4047 BDRVQcowState
*s
= bs
->opaque
;
4050 nb_clusters
= (size
+ s
->cluster_size
- 1) >> s
->cluster_bits
;
4052 if (get_refcount(bs
, s
->free_cluster_index
) == 0) {
4053 s
->free_cluster_index
++;
4054 for(i
= 1; i
< nb_clusters
; i
++) {
4055 if (get_refcount(bs
, s
->free_cluster_index
) != 0)
4057 s
->free_cluster_index
++;
4060 printf("alloc_clusters: size=%lld -> %lld\n",
4062 (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
);
4064 return (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
;
4067 s
->free_cluster_index
++;
4072 static int64_t alloc_clusters(BlockDriverState
*bs
, int64_t size
)
4076 offset
= alloc_clusters_noref(bs
, size
);
4077 update_refcount(bs
, offset
, size
, 1);
4081 /* only used to allocate compressed sectors. We try to allocate
4082 contiguous sectors. size must be <= cluster_size */
4083 static int64_t alloc_bytes(BlockDriverState
*bs
, int size
)
4085 BDRVQcowState
*s
= bs
->opaque
;
4086 int64_t offset
, cluster_offset
;
4087 int free_in_cluster
;
4089 assert(size
> 0 && size
<= s
->cluster_size
);
4090 if (s
->free_byte_offset
== 0) {
4091 s
->free_byte_offset
= alloc_clusters(bs
, s
->cluster_size
);
4094 free_in_cluster
= s
->cluster_size
-
4095 (s
->free_byte_offset
& (s
->cluster_size
- 1));
4096 if (size
<= free_in_cluster
) {
4097 /* enough space in current cluster */
4098 offset
= s
->free_byte_offset
;
4099 s
->free_byte_offset
+= size
;
4100 free_in_cluster
-= size
;
4101 if (free_in_cluster
== 0)
4102 s
->free_byte_offset
= 0;
4103 if ((offset
& (s
->cluster_size
- 1)) != 0)
4104 update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, 1);
4106 offset
= alloc_clusters(bs
, s
->cluster_size
);
4107 cluster_offset
= s
->free_byte_offset
& ~(s
->cluster_size
- 1);
4108 if ((cluster_offset
+ s
->cluster_size
) == offset
) {
4109 /* we are lucky: contiguous data */
4110 offset
= s
->free_byte_offset
;
4111 update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, 1);
4112 s
->free_byte_offset
+= size
;
4114 s
->free_byte_offset
= offset
;
4121 static void free_clusters(BlockDriverState
*bs
,
4122 int64_t offset
, int64_t size
)
4124 update_refcount(bs
, offset
, size
, -1);
4127 static int grow_refcount_table(BlockDriverState
*bs
, int min_size
)
4129 BDRVQcowState
*s
= bs
->opaque
;
4130 int new_table_size
, new_table_size2
, refcount_table_clusters
, i
, ret
;
4131 uint64_t *new_table
;
4132 int64_t table_offset
;
4136 if (min_size
<= s
->refcount_table_size
)
4138 /* compute new table size */
4139 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
4141 if (refcount_table_clusters
== 0) {
4142 refcount_table_clusters
= 1;
4144 refcount_table_clusters
= (refcount_table_clusters
* 3 + 1) / 2;
4146 new_table_size
= refcount_table_clusters
<< (s
->cluster_bits
- 3);
4147 if (min_size
<= new_table_size
)
4151 printf("grow_refcount_table from %d to %d\n",
4152 s
->refcount_table_size
,
4155 new_table_size2
= new_table_size
* sizeof(uint64_t);
4156 new_table
= qemu_mallocz(new_table_size2
);
4159 memcpy(new_table
, s
->refcount_table
,
4160 s
->refcount_table_size
* sizeof(uint64_t));
4161 for(i
= 0; i
< s
->refcount_table_size
; i
++)
4162 cpu_to_be64s(&new_table
[i
]);
4163 /* Note: we cannot update the refcount now to avoid recursion */
4164 table_offset
= alloc_clusters_noref(bs
, new_table_size2
);
4165 ret
= bdrv_pwrite(s
->hd
, table_offset
, new_table
, new_table_size2
);
4166 if (ret
!= new_table_size2
)
4168 for(i
= 0; i
< s
->refcount_table_size
; i
++)
4169 be64_to_cpus(&new_table
[i
]);
4171 data64
= cpu_to_be64(table_offset
);
4172 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, refcount_table_offset
),
4173 &data64
, sizeof(data64
)) != sizeof(data64
))
4175 data32
= cpu_to_be32(refcount_table_clusters
);
4176 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, refcount_table_clusters
),
4177 &data32
, sizeof(data32
)) != sizeof(data32
))
4179 qemu_free(s
->refcount_table
);
4180 s
->refcount_table
= new_table
;
4181 s
->refcount_table_size
= new_table_size
;
4183 update_refcount(bs
, table_offset
, new_table_size2
, 1);
4186 free_clusters(bs
, table_offset
, new_table_size2
);
4187 qemu_free(new_table
);
4191 /* addend must be 1 or -1 */
4192 /* XXX: cache several refcount block clusters ? */
4193 static int update_cluster_refcount(BlockDriverState
*bs
,
4194 int64_t cluster_index
,
4197 BDRVQcowState
*s
= bs
->opaque
;
4198 int64_t offset
, refcount_block_offset
;
4199 int ret
, refcount_table_index
, block_index
, refcount
;
4202 refcount_table_index
= cluster_index
>> (s
->cluster_bits
- REFCOUNT_SHIFT
);
4203 if (refcount_table_index
>= s
->refcount_table_size
) {
4206 ret
= grow_refcount_table(bs
, refcount_table_index
+ 1);
4210 refcount_block_offset
= s
->refcount_table
[refcount_table_index
];
4211 if (!refcount_block_offset
) {
4214 /* create a new refcount block */
4215 /* Note: we cannot update the refcount now to avoid recursion */
4216 offset
= alloc_clusters_noref(bs
, s
->cluster_size
);
4217 memset(s
->refcount_block_cache
, 0, s
->cluster_size
);
4218 ret
= bdrv_pwrite(s
->hd
, offset
, s
->refcount_block_cache
, s
->cluster_size
);
4219 if (ret
!= s
->cluster_size
)
4221 s
->refcount_table
[refcount_table_index
] = offset
;
4222 data64
= cpu_to_be64(offset
);
4223 ret
= bdrv_pwrite(s
->hd
, s
->refcount_table_offset
+
4224 refcount_table_index
* sizeof(uint64_t),
4225 &data64
, sizeof(data64
));
4226 if (ret
!= sizeof(data64
))
4229 refcount_block_offset
= offset
;
4230 s
->refcount_block_cache_offset
= offset
;
4231 update_refcount(bs
, offset
, s
->cluster_size
, 1);
4233 if (refcount_block_offset
!= s
->refcount_block_cache_offset
) {
4234 if (load_refcount_block(bs
, refcount_block_offset
) < 0)
4238 /* we can update the count and save it */
4239 block_index
= cluster_index
&
4240 ((1 << (s
->cluster_bits
- REFCOUNT_SHIFT
)) - 1);
4241 refcount
= be16_to_cpu(s
->refcount_block_cache
[block_index
]);
4243 if (refcount
< 0 || refcount
> 0xffff)
4245 if (refcount
== 0 && cluster_index
< s
->free_cluster_index
) {
4246 s
->free_cluster_index
= cluster_index
;
4248 s
->refcount_block_cache
[block_index
] = cpu_to_be16(refcount
);
4249 if (bdrv_pwrite(s
->hd
,
4250 refcount_block_offset
+ (block_index
<< REFCOUNT_SHIFT
),
4251 &s
->refcount_block_cache
[block_index
], 2) != 2)
4256 static void update_refcount(BlockDriverState
*bs
,
4257 int64_t offset
, int64_t length
,
4260 BDRVQcowState
*s
= bs
->opaque
;
4261 int64_t start
, last
, cluster_offset
;
4264 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
4265 offset
, length
, addend
);
4269 start
= offset
& ~(s
->cluster_size
- 1);
4270 last
= (offset
+ length
- 1) & ~(s
->cluster_size
- 1);
4271 for(cluster_offset
= start
; cluster_offset
<= last
;
4272 cluster_offset
+= s
->cluster_size
) {
4273 update_cluster_refcount(bs
, cluster_offset
>> s
->cluster_bits
, addend
);
4278 static void inc_refcounts(BlockDriverState
*bs
,
4279 uint16_t *refcount_table
,
4280 int refcount_table_size
,
4281 int64_t offset
, int64_t size
)
4283 BDRVQcowState
*s
= bs
->opaque
;
4284 int64_t start
, last
, cluster_offset
;
4290 start
= offset
& ~(s
->cluster_size
- 1);
4291 last
= (offset
+ size
- 1) & ~(s
->cluster_size
- 1);
4292 for(cluster_offset
= start
; cluster_offset
<= last
;
4293 cluster_offset
+= s
->cluster_size
) {
4294 k
= cluster_offset
>> s
->cluster_bits
;
4295 if (k
< 0 || k
>= refcount_table_size
) {
4296 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset
);
4298 if (++refcount_table
[k
] == 0) {
4299 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset
);
4305 static int check_refcounts_l1(BlockDriverState
*bs
,
4306 uint16_t *refcount_table
,
4307 int refcount_table_size
,
4308 int64_t l1_table_offset
, int l1_size
,
4311 BDRVQcowState
*s
= bs
->opaque
;
4312 uint64_t *l1_table
, *l2_table
, l2_offset
, offset
, l1_size2
;
4313 int l2_size
, i
, j
, nb_csectors
, refcount
;
4316 l1_size2
= l1_size
* sizeof(uint64_t);
4318 inc_refcounts(bs
, refcount_table
, refcount_table_size
,
4319 l1_table_offset
, l1_size2
);
4321 l1_table
= qemu_malloc(l1_size2
);
4324 if (bdrv_pread(s
->hd
, l1_table_offset
,
4325 l1_table
, l1_size2
) != l1_size2
)
4327 for(i
= 0;i
< l1_size
; i
++)
4328 be64_to_cpus(&l1_table
[i
]);
4330 l2_size
= s
->l2_size
* sizeof(uint64_t);
4331 l2_table
= qemu_malloc(l2_size
);
4334 for(i
= 0; i
< l1_size
; i
++) {
4335 l2_offset
= l1_table
[i
];
4338 refcount
= get_refcount(bs
, (l2_offset
& ~QCOW_OFLAG_COPIED
) >> s
->cluster_bits
);
4339 if ((refcount
== 1) != ((l2_offset
& QCOW_OFLAG_COPIED
) != 0)) {
4340 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
4341 l2_offset
, refcount
);
4344 l2_offset
&= ~QCOW_OFLAG_COPIED
;
4345 if (bdrv_pread(s
->hd
, l2_offset
, l2_table
, l2_size
) != l2_size
)
4347 for(j
= 0; j
< s
->l2_size
; j
++) {
4348 offset
= be64_to_cpu(l2_table
[j
]);
4350 if (offset
& QCOW_OFLAG_COMPRESSED
) {
4351 if (offset
& QCOW_OFLAG_COPIED
) {
4352 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
4353 offset
>> s
->cluster_bits
);
4354 offset
&= ~QCOW_OFLAG_COPIED
;
4356 nb_csectors
= ((offset
>> s
->csize_shift
) &
4358 offset
&= s
->cluster_offset_mask
;
4359 inc_refcounts(bs
, refcount_table
,
4360 refcount_table_size
,
4361 offset
& ~511, nb_csectors
* 512);
4364 refcount
= get_refcount(bs
, (offset
& ~QCOW_OFLAG_COPIED
) >> s
->cluster_bits
);
4365 if ((refcount
== 1) != ((offset
& QCOW_OFLAG_COPIED
) != 0)) {
4366 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
4370 offset
&= ~QCOW_OFLAG_COPIED
;
4371 inc_refcounts(bs
, refcount_table
,
4372 refcount_table_size
,
4373 offset
, s
->cluster_size
);
4377 inc_refcounts(bs
, refcount_table
,
4378 refcount_table_size
,
4383 qemu_free(l1_table
);
4384 qemu_free(l2_table
);
4387 printf("ERROR: I/O error in check_refcounts_l1\n");
4388 qemu_free(l1_table
);
4389 qemu_free(l2_table
);
4393 static void check_refcounts(BlockDriverState
*bs
)
4395 BDRVQcowState
*s
= bs
->opaque
;
4397 int nb_clusters
, refcount1
, refcount2
, i
;
4399 uint16_t *refcount_table
;
4401 size
= bdrv_getlength(s
->hd
);
4402 nb_clusters
= (size
+ s
->cluster_size
- 1) >> s
->cluster_bits
;
4403 refcount_table
= qemu_mallocz(nb_clusters
* sizeof(uint16_t));
4406 inc_refcounts(bs
, refcount_table
, nb_clusters
,
4407 0, s
->cluster_size
);
4409 check_refcounts_l1(bs
, refcount_table
, nb_clusters
,
4410 s
->l1_table_offset
, s
->l1_size
, 1);
4413 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
4414 sn
= s
->snapshots
+ i
;
4415 check_refcounts_l1(bs
, refcount_table
, nb_clusters
,
4416 sn
->l1_table_offset
, sn
->l1_size
, 0);
4418 inc_refcounts(bs
, refcount_table
, nb_clusters
,
4419 s
->snapshots_offset
, s
->snapshots_size
);
4422 inc_refcounts(bs
, refcount_table
, nb_clusters
,
4423 s
->refcount_table_offset
,
4424 s
->refcount_table_size
* sizeof(uint64_t));
4425 for(i
= 0; i
< s
->refcount_table_size
; i
++) {
4427 offset
= s
->refcount_table
[i
];
4429 inc_refcounts(bs
, refcount_table
, nb_clusters
,
4430 offset
, s
->cluster_size
);
4434 /* compare ref counts */
4435 for(i
= 0; i
< nb_clusters
; i
++) {
4436 refcount1
= get_refcount(bs
, i
);
4437 refcount2
= refcount_table
[i
];
4438 if (refcount1
!= refcount2
)
4439 printf("ERROR cluster %d refcount=%d reference=%d\n",
4440 i
, refcount1
, refcount2
);
4443 qemu_free(refcount_table
);
4447 static void dump_refcounts(BlockDriverState
*bs
)
4449 BDRVQcowState
*s
= bs
->opaque
;
4450 int64_t nb_clusters
, k
, k1
, size
;
4453 size
= bdrv_getlength(s
->hd
);
4454 nb_clusters
= (size
+ s
->cluster_size
- 1) >> s
->cluster_bits
;
4455 for(k
= 0; k
< nb_clusters
;) {
4457 refcount
= get_refcount(bs
, k
);
4459 while (k
< nb_clusters
&& get_refcount(bs
, k
) == refcount
)
4461 printf("%lld: refcount=%d nb=%lld\n", k
, refcount
, k
- k1
);
4467 BlockDriver bdrv_qcow2
= {
4469 sizeof(BDRVQcowState
),
4481 .bdrv_aio_read
= qcow_aio_read
,
4482 .bdrv_aio_write
= qcow_aio_write
,
4483 .bdrv_aio_cancel
= qcow_aio_cancel
,
4484 .aiocb_size
= sizeof(QCowAIOCB
),
4485 .bdrv_write_compressed
= qcow_write_compressed
,
4487 .bdrv_snapshot_create
= qcow_snapshot_create
,
4488 .bdrv_snapshot_goto
= qcow_snapshot_goto
,
4489 .bdrv_snapshot_delete
= qcow_snapshot_delete
,
4490 .bdrv_snapshot_list
= qcow_snapshot_list
,
4491 .bdrv_get_info
= qcow_get_info
,