2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "qemu-common.h"
25 #include "block_int.h"
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
50 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
51 #define QCOW_VERSION 2
53 #define QCOW_CRYPT_NONE 0
54 #define QCOW_CRYPT_AES 1
56 #define QCOW_MAX_CRYPT_CLUSTERS 32
58 /* indicate that the refcount of the referenced cluster is exactly one. */
59 #define QCOW_OFLAG_COPIED (1LL << 63)
60 /* indicate that the cluster is compressed (they never have the copied flag) */
61 #define QCOW_OFLAG_COMPRESSED (1LL << 62)
63 #define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
65 typedef struct QCowHeader
{
68 uint64_t backing_file_offset
;
69 uint32_t backing_file_size
;
70 uint32_t cluster_bits
;
71 uint64_t size
; /* in bytes */
72 uint32_t crypt_method
;
73 uint32_t l1_size
; /* XXX: save number of clusters instead ? */
74 uint64_t l1_table_offset
;
75 uint64_t refcount_table_offset
;
76 uint32_t refcount_table_clusters
;
77 uint32_t nb_snapshots
;
78 uint64_t snapshots_offset
;
86 #define QCOW_EXT_MAGIC_END 0
87 #define QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
90 typedef struct __attribute__((packed
)) QCowSnapshotHeader
{
91 /* header is 8 byte aligned */
92 uint64_t l1_table_offset
;
101 uint64_t vm_clock_nsec
;
103 uint32_t vm_state_size
;
104 uint32_t extra_data_size
; /* for extension */
105 /* extra data follows */
108 } QCowSnapshotHeader
;
110 #define L2_CACHE_SIZE 16
112 typedef struct QCowSnapshot
{
113 uint64_t l1_table_offset
;
117 uint32_t vm_state_size
;
120 uint64_t vm_clock_nsec
;
123 typedef struct BDRVQcowState
{
124 BlockDriverState
*hd
;
131 int l1_vm_state_index
;
134 uint64_t cluster_offset_mask
;
135 uint64_t l1_table_offset
;
138 uint64_t l2_cache_offsets
[L2_CACHE_SIZE
];
139 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
140 uint8_t *cluster_cache
;
141 uint8_t *cluster_data
;
142 uint64_t cluster_cache_offset
;
144 uint64_t *refcount_table
;
145 uint64_t refcount_table_offset
;
146 uint32_t refcount_table_size
;
147 uint64_t refcount_block_cache_offset
;
148 uint16_t *refcount_block_cache
;
149 int64_t free_cluster_index
;
150 int64_t free_byte_offset
;
152 uint32_t crypt_method
; /* current crypt method, 0 if no key yet */
153 uint32_t crypt_method_header
;
154 AES_KEY aes_encrypt_key
;
155 AES_KEY aes_decrypt_key
;
156 uint64_t snapshots_offset
;
159 QCowSnapshot
*snapshots
;
162 static int decompress_cluster(BDRVQcowState
*s
, uint64_t cluster_offset
);
163 static int qcow_read(BlockDriverState
*bs
, int64_t sector_num
,
164 uint8_t *buf
, int nb_sectors
);
165 static int qcow_read_snapshots(BlockDriverState
*bs
);
166 static void qcow_free_snapshots(BlockDriverState
*bs
);
167 static int refcount_init(BlockDriverState
*bs
);
168 static void refcount_close(BlockDriverState
*bs
);
169 static int get_refcount(BlockDriverState
*bs
, int64_t cluster_index
);
170 static int update_cluster_refcount(BlockDriverState
*bs
,
171 int64_t cluster_index
,
173 static void update_refcount(BlockDriverState
*bs
,
174 int64_t offset
, int64_t length
,
176 static int64_t alloc_clusters(BlockDriverState
*bs
, int64_t size
);
177 static int64_t alloc_bytes(BlockDriverState
*bs
, int size
);
178 static void free_clusters(BlockDriverState
*bs
,
179 int64_t offset
, int64_t size
);
181 static void check_refcounts(BlockDriverState
*bs
);
184 static int qcow_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
186 const QCowHeader
*cow_header
= (const void *)buf
;
188 if (buf_size
>= sizeof(QCowHeader
) &&
189 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
190 be32_to_cpu(cow_header
->version
) == QCOW_VERSION
)
198 * read qcow2 extension and fill bs
199 * start reading from start_offset
200 * finish reading upon magic of value 0 or when end_offset reached
201 * unknown magic is skipped (future extension this version knows nothing about)
202 * return 0 upon success, non-0 otherwise
204 static int qcow_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
207 BDRVQcowState
*s
= bs
->opaque
;
212 printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
214 offset
= start_offset
;
215 while (offset
< end_offset
) {
219 if (offset
> s
->cluster_size
)
220 printf("qcow_handle_extension: suspicious offset %lu\n", offset
);
222 printf("attemting to read extended header in offset %lu\n", offset
);
225 if (bdrv_pread(s
->hd
, offset
, &ext
, sizeof(ext
)) != sizeof(ext
)) {
226 fprintf(stderr
, "qcow_handle_extension: ERROR: pread fail from offset %llu\n",
227 (unsigned long long)offset
);
230 be32_to_cpus(&ext
.magic
);
231 be32_to_cpus(&ext
.len
);
232 offset
+= sizeof(ext
);
234 printf("ext.magic = 0x%x\n", ext
.magic
);
237 case QCOW_EXT_MAGIC_END
:
240 case QCOW_EXT_MAGIC_BACKING_FORMAT
:
241 if (ext
.len
>= sizeof(bs
->backing_format
)) {
242 fprintf(stderr
, "ERROR: ext_backing_format: len=%u too large"
244 ext
.len
, sizeof(bs
->backing_format
));
247 if (bdrv_pread(s
->hd
, offset
, bs
->backing_format
,
250 bs
->backing_format
[ext
.len
] = '\0';
252 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
254 offset
+= ((ext
.len
+ 7) & ~7);
258 /* unknown magic -- just skip it */
259 offset
+= ((ext
.len
+ 7) & ~7);
268 static int qcow_open(BlockDriverState
*bs
, const char *filename
, int flags
)
270 BDRVQcowState
*s
= bs
->opaque
;
271 int len
, i
, shift
, ret
;
275 /* Performance is terrible right now with cache=writethrough due mainly
276 * to reference count updates. If the user does not explicitly specify
277 * a caching type, force to writeback caching.
279 if ((flags
& BDRV_O_CACHE_DEF
)) {
280 flags
|= BDRV_O_CACHE_WB
;
281 flags
&= ~BDRV_O_CACHE_DEF
;
283 ret
= bdrv_file_open(&s
->hd
, filename
, flags
);
286 if (bdrv_pread(s
->hd
, 0, &header
, sizeof(header
)) != sizeof(header
))
288 be32_to_cpus(&header
.magic
);
289 be32_to_cpus(&header
.version
);
290 be64_to_cpus(&header
.backing_file_offset
);
291 be32_to_cpus(&header
.backing_file_size
);
292 be64_to_cpus(&header
.size
);
293 be32_to_cpus(&header
.cluster_bits
);
294 be32_to_cpus(&header
.crypt_method
);
295 be64_to_cpus(&header
.l1_table_offset
);
296 be32_to_cpus(&header
.l1_size
);
297 be64_to_cpus(&header
.refcount_table_offset
);
298 be32_to_cpus(&header
.refcount_table_clusters
);
299 be64_to_cpus(&header
.snapshots_offset
);
300 be32_to_cpus(&header
.nb_snapshots
);
302 if (header
.magic
!= QCOW_MAGIC
|| header
.version
!= QCOW_VERSION
)
304 if (header
.size
<= 1 ||
305 header
.cluster_bits
< 9 ||
306 header
.cluster_bits
> 16)
308 if (header
.crypt_method
> QCOW_CRYPT_AES
)
310 s
->crypt_method_header
= header
.crypt_method
;
311 if (s
->crypt_method_header
)
313 s
->cluster_bits
= header
.cluster_bits
;
314 s
->cluster_size
= 1 << s
->cluster_bits
;
315 s
->cluster_sectors
= 1 << (s
->cluster_bits
- 9);
316 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
317 s
->l2_size
= 1 << s
->l2_bits
;
318 bs
->total_sectors
= header
.size
/ 512;
319 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
320 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
321 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
322 s
->refcount_table_offset
= header
.refcount_table_offset
;
323 s
->refcount_table_size
=
324 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
326 s
->snapshots_offset
= header
.snapshots_offset
;
327 s
->nb_snapshots
= header
.nb_snapshots
;
329 /* read the level 1 table */
330 s
->l1_size
= header
.l1_size
;
331 shift
= s
->cluster_bits
+ s
->l2_bits
;
332 s
->l1_vm_state_index
= (header
.size
+ (1LL << shift
) - 1) >> shift
;
333 /* the L1 table must contain at least enough entries to put
335 if (s
->l1_size
< s
->l1_vm_state_index
)
337 s
->l1_table_offset
= header
.l1_table_offset
;
338 s
->l1_table
= qemu_malloc(s
->l1_size
* sizeof(uint64_t));
339 if (bdrv_pread(s
->hd
, s
->l1_table_offset
, s
->l1_table
, s
->l1_size
* sizeof(uint64_t)) !=
340 s
->l1_size
* sizeof(uint64_t))
342 for(i
= 0;i
< s
->l1_size
; i
++) {
343 be64_to_cpus(&s
->l1_table
[i
]);
346 s
->l2_cache
= qemu_malloc(s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
347 s
->cluster_cache
= qemu_malloc(s
->cluster_size
);
348 /* one more sector for decompressed data alignment */
349 s
->cluster_data
= qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
351 s
->cluster_cache_offset
= -1;
353 if (refcount_init(bs
) < 0)
356 /* read qcow2 extensions */
357 if (header
.backing_file_offset
)
358 ext_end
= header
.backing_file_offset
;
360 ext_end
= s
->cluster_size
;
361 if (qcow_read_extensions(bs
, sizeof(header
), ext_end
))
364 /* read the backing file name */
365 if (header
.backing_file_offset
!= 0) {
366 len
= header
.backing_file_size
;
369 if (bdrv_pread(s
->hd
, header
.backing_file_offset
, bs
->backing_file
, len
) != len
)
371 bs
->backing_file
[len
] = '\0';
373 if (qcow_read_snapshots(bs
) < 0)
382 qcow_free_snapshots(bs
);
384 qemu_free(s
->l1_table
);
385 qemu_free(s
->l2_cache
);
386 qemu_free(s
->cluster_cache
);
387 qemu_free(s
->cluster_data
);
392 static int qcow_set_key(BlockDriverState
*bs
, const char *key
)
394 BDRVQcowState
*s
= bs
->opaque
;
398 memset(keybuf
, 0, 16);
402 /* XXX: we could compress the chars to 7 bits to increase
404 for(i
= 0;i
< len
;i
++) {
407 s
->crypt_method
= s
->crypt_method_header
;
409 if (AES_set_encrypt_key(keybuf
, 128, &s
->aes_encrypt_key
) != 0)
411 if (AES_set_decrypt_key(keybuf
, 128, &s
->aes_decrypt_key
) != 0)
421 AES_encrypt(in
, tmp
, &s
->aes_encrypt_key
);
422 AES_decrypt(tmp
, out
, &s
->aes_decrypt_key
);
423 for(i
= 0; i
< 16; i
++)
424 printf(" %02x", tmp
[i
]);
426 for(i
= 0; i
< 16; i
++)
427 printf(" %02x", out
[i
]);
434 /* The crypt function is compatible with the linux cryptoloop
435 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
437 static void encrypt_sectors(BDRVQcowState
*s
, int64_t sector_num
,
438 uint8_t *out_buf
, const uint8_t *in_buf
,
439 int nb_sectors
, int enc
,
448 for(i
= 0; i
< nb_sectors
; i
++) {
449 ivec
.ll
[0] = cpu_to_le64(sector_num
);
451 AES_cbc_encrypt(in_buf
, out_buf
, 512, key
,
459 static int copy_sectors(BlockDriverState
*bs
, uint64_t start_sect
,
460 uint64_t cluster_offset
, int n_start
, int n_end
)
462 BDRVQcowState
*s
= bs
->opaque
;
468 ret
= qcow_read(bs
, start_sect
+ n_start
, s
->cluster_data
, n
);
471 if (s
->crypt_method
) {
472 encrypt_sectors(s
, start_sect
+ n_start
,
474 s
->cluster_data
, n
, 1,
475 &s
->aes_encrypt_key
);
477 ret
= bdrv_write(s
->hd
, (cluster_offset
>> 9) + n_start
,
484 static void l2_cache_reset(BlockDriverState
*bs
)
486 BDRVQcowState
*s
= bs
->opaque
;
488 memset(s
->l2_cache
, 0, s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
489 memset(s
->l2_cache_offsets
, 0, L2_CACHE_SIZE
* sizeof(uint64_t));
490 memset(s
->l2_cache_counts
, 0, L2_CACHE_SIZE
* sizeof(uint32_t));
493 static inline int l2_cache_new_entry(BlockDriverState
*bs
)
495 BDRVQcowState
*s
= bs
->opaque
;
499 /* find a new entry in the least used one */
501 min_count
= 0xffffffff;
502 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
503 if (s
->l2_cache_counts
[i
] < min_count
) {
504 min_count
= s
->l2_cache_counts
[i
];
511 static int64_t align_offset(int64_t offset
, int n
)
513 offset
= (offset
+ n
- 1) & ~(n
- 1);
517 static int grow_l1_table(BlockDriverState
*bs
, int min_size
)
519 BDRVQcowState
*s
= bs
->opaque
;
520 int new_l1_size
, new_l1_size2
, ret
, i
;
521 uint64_t *new_l1_table
;
522 uint64_t new_l1_table_offset
;
525 new_l1_size
= s
->l1_size
;
526 if (min_size
<= new_l1_size
)
528 while (min_size
> new_l1_size
) {
529 new_l1_size
= (new_l1_size
* 3 + 1) / 2;
532 printf("grow l1_table from %d to %d\n", s
->l1_size
, new_l1_size
);
535 new_l1_size2
= sizeof(uint64_t) * new_l1_size
;
536 new_l1_table
= qemu_mallocz(new_l1_size2
);
537 memcpy(new_l1_table
, s
->l1_table
, s
->l1_size
* sizeof(uint64_t));
539 /* write new table (align to cluster) */
540 new_l1_table_offset
= alloc_clusters(bs
, new_l1_size2
);
542 for(i
= 0; i
< s
->l1_size
; i
++)
543 new_l1_table
[i
] = cpu_to_be64(new_l1_table
[i
]);
544 ret
= bdrv_pwrite(s
->hd
, new_l1_table_offset
, new_l1_table
, new_l1_size2
);
545 if (ret
!= new_l1_size2
)
547 for(i
= 0; i
< s
->l1_size
; i
++)
548 new_l1_table
[i
] = be64_to_cpu(new_l1_table
[i
]);
551 cpu_to_be32w((uint32_t*)data
, new_l1_size
);
552 cpu_to_be64w((uint64_t*)(data
+ 4), new_l1_table_offset
);
553 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, l1_size
), data
,
554 sizeof(data
)) != sizeof(data
))
556 qemu_free(s
->l1_table
);
557 free_clusters(bs
, s
->l1_table_offset
, s
->l1_size
* sizeof(uint64_t));
558 s
->l1_table_offset
= new_l1_table_offset
;
559 s
->l1_table
= new_l1_table
;
560 s
->l1_size
= new_l1_size
;
563 qemu_free(s
->l1_table
);
570 * seek l2_offset in the l2_cache table
571 * if not found, return NULL,
573 * increments the l2 cache hit count of the entry,
574 * if counter overflow, divide by two all counters
575 * return the pointer to the l2 cache entry
579 static uint64_t *seek_l2_table(BDRVQcowState
*s
, uint64_t l2_offset
)
583 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
584 if (l2_offset
== s
->l2_cache_offsets
[i
]) {
585 /* increment the hit count */
586 if (++s
->l2_cache_counts
[i
] == 0xffffffff) {
587 for(j
= 0; j
< L2_CACHE_SIZE
; j
++) {
588 s
->l2_cache_counts
[j
] >>= 1;
591 return s
->l2_cache
+ (i
<< s
->l2_bits
);
600 * Loads a L2 table into memory. If the table is in the cache, the cache
601 * is used; otherwise the L2 table is loaded from the image file.
603 * Returns a pointer to the L2 table on success, or NULL if the read from
604 * the image file failed.
607 static uint64_t *l2_load(BlockDriverState
*bs
, uint64_t l2_offset
)
609 BDRVQcowState
*s
= bs
->opaque
;
613 /* seek if the table for the given offset is in the cache */
615 l2_table
= seek_l2_table(s
, l2_offset
);
616 if (l2_table
!= NULL
)
619 /* not found: load a new entry in the least used one */
621 min_index
= l2_cache_new_entry(bs
);
622 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
623 if (bdrv_pread(s
->hd
, l2_offset
, l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
624 s
->l2_size
* sizeof(uint64_t))
626 s
->l2_cache_offsets
[min_index
] = l2_offset
;
627 s
->l2_cache_counts
[min_index
] = 1;
635 * Allocate a new l2 entry in the file. If l1_index points to an already
636 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
637 * table) copy the contents of the old L2 table into the newly allocated one.
638 * Otherwise the new table is initialized with zeros.
642 static uint64_t *l2_allocate(BlockDriverState
*bs
, int l1_index
)
644 BDRVQcowState
*s
= bs
->opaque
;
646 uint64_t old_l2_offset
, tmp
;
647 uint64_t *l2_table
, l2_offset
;
649 old_l2_offset
= s
->l1_table
[l1_index
];
651 /* allocate a new l2 entry */
653 l2_offset
= alloc_clusters(bs
, s
->l2_size
* sizeof(uint64_t));
655 /* update the L1 entry */
657 s
->l1_table
[l1_index
] = l2_offset
| QCOW_OFLAG_COPIED
;
659 tmp
= cpu_to_be64(l2_offset
| QCOW_OFLAG_COPIED
);
660 if (bdrv_pwrite(s
->hd
, s
->l1_table_offset
+ l1_index
* sizeof(tmp
),
661 &tmp
, sizeof(tmp
)) != sizeof(tmp
))
664 /* allocate a new entry in the l2 cache */
666 min_index
= l2_cache_new_entry(bs
);
667 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
669 if (old_l2_offset
== 0) {
670 /* if there was no old l2 table, clear the new table */
671 memset(l2_table
, 0, s
->l2_size
* sizeof(uint64_t));
673 /* if there was an old l2 table, read it from the disk */
674 if (bdrv_pread(s
->hd
, old_l2_offset
,
675 l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
676 s
->l2_size
* sizeof(uint64_t))
679 /* write the l2 table to the file */
680 if (bdrv_pwrite(s
->hd
, l2_offset
,
681 l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
682 s
->l2_size
* sizeof(uint64_t))
685 /* update the l2 cache entry */
687 s
->l2_cache_offsets
[min_index
] = l2_offset
;
688 s
->l2_cache_counts
[min_index
] = 1;
693 static int size_to_clusters(BDRVQcowState
*s
, int64_t size
)
695 return (size
+ (s
->cluster_size
- 1)) >> s
->cluster_bits
;
698 static int count_contiguous_clusters(uint64_t nb_clusters
, int cluster_size
,
699 uint64_t *l2_table
, uint64_t start
, uint64_t mask
)
702 uint64_t offset
= be64_to_cpu(l2_table
[0]) & ~mask
;
707 for (i
= start
; i
< start
+ nb_clusters
; i
++)
708 if (offset
+ i
* cluster_size
!= (be64_to_cpu(l2_table
[i
]) & ~mask
))
714 static int count_contiguous_free_clusters(uint64_t nb_clusters
, uint64_t *l2_table
)
718 while(nb_clusters
-- && l2_table
[i
] == 0)
727 * For a given offset of the disk image, return cluster offset in
730 * on entry, *num is the number of contiguous clusters we'd like to
731 * access following offset.
733 * on exit, *num is the number of contiguous clusters we can read.
735 * Return 1, if the offset is found
736 * Return 0, otherwise.
740 static uint64_t get_cluster_offset(BlockDriverState
*bs
,
741 uint64_t offset
, int *num
)
743 BDRVQcowState
*s
= bs
->opaque
;
744 int l1_index
, l2_index
;
745 uint64_t l2_offset
, *l2_table
, cluster_offset
;
747 int index_in_cluster
, nb_available
, nb_needed
, nb_clusters
;
749 index_in_cluster
= (offset
>> 9) & (s
->cluster_sectors
- 1);
750 nb_needed
= *num
+ index_in_cluster
;
752 l1_bits
= s
->l2_bits
+ s
->cluster_bits
;
754 /* compute how many bytes there are between the offset and
755 * the end of the l1 entry
758 nb_available
= (1 << l1_bits
) - (offset
& ((1 << l1_bits
) - 1));
760 /* compute the number of available sectors */
762 nb_available
= (nb_available
>> 9) + index_in_cluster
;
766 /* seek the the l2 offset in the l1 table */
768 l1_index
= offset
>> l1_bits
;
769 if (l1_index
>= s
->l1_size
)
772 l2_offset
= s
->l1_table
[l1_index
];
774 /* seek the l2 table of the given l2 offset */
779 /* load the l2 table in memory */
781 l2_offset
&= ~QCOW_OFLAG_COPIED
;
782 l2_table
= l2_load(bs
, l2_offset
);
783 if (l2_table
== NULL
)
786 /* find the cluster offset for the given disk offset */
788 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
789 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
790 nb_clusters
= size_to_clusters(s
, nb_needed
<< 9);
792 if (!cluster_offset
) {
793 /* how many empty clusters ? */
794 c
= count_contiguous_free_clusters(nb_clusters
, &l2_table
[l2_index
]);
796 /* how many allocated clusters ? */
797 c
= count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
798 &l2_table
[l2_index
], 0, QCOW_OFLAG_COPIED
);
801 nb_available
= (c
* s
->cluster_sectors
);
803 if (nb_available
> nb_needed
)
804 nb_available
= nb_needed
;
806 *num
= nb_available
- index_in_cluster
;
808 return cluster_offset
& ~QCOW_OFLAG_COPIED
;
814 * free clusters according to its type: compressed or not
818 static void free_any_clusters(BlockDriverState
*bs
,
819 uint64_t cluster_offset
, int nb_clusters
)
821 BDRVQcowState
*s
= bs
->opaque
;
823 /* free the cluster */
825 if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
827 nb_csectors
= ((cluster_offset
>> s
->csize_shift
) &
829 free_clusters(bs
, (cluster_offset
& s
->cluster_offset_mask
) & ~511,
834 free_clusters(bs
, cluster_offset
, nb_clusters
<< s
->cluster_bits
);
842 * for a given disk offset, load (and allocate if needed)
845 * the l2 table offset in the qcow2 file and the cluster index
846 * in the l2 table are given to the caller.
850 static int get_cluster_table(BlockDriverState
*bs
, uint64_t offset
,
851 uint64_t **new_l2_table
,
852 uint64_t *new_l2_offset
,
855 BDRVQcowState
*s
= bs
->opaque
;
856 int l1_index
, l2_index
, ret
;
857 uint64_t l2_offset
, *l2_table
;
859 /* seek the the l2 offset in the l1 table */
861 l1_index
= offset
>> (s
->l2_bits
+ s
->cluster_bits
);
862 if (l1_index
>= s
->l1_size
) {
863 ret
= grow_l1_table(bs
, l1_index
+ 1);
867 l2_offset
= s
->l1_table
[l1_index
];
869 /* seek the l2 table of the given l2 offset */
871 if (l2_offset
& QCOW_OFLAG_COPIED
) {
872 /* load the l2 table in memory */
873 l2_offset
&= ~QCOW_OFLAG_COPIED
;
874 l2_table
= l2_load(bs
, l2_offset
);
875 if (l2_table
== NULL
)
879 free_clusters(bs
, l2_offset
, s
->l2_size
* sizeof(uint64_t));
880 l2_table
= l2_allocate(bs
, l1_index
);
881 if (l2_table
== NULL
)
883 l2_offset
= s
->l1_table
[l1_index
] & ~QCOW_OFLAG_COPIED
;
886 /* find the cluster offset for the given disk offset */
888 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
890 *new_l2_table
= l2_table
;
891 *new_l2_offset
= l2_offset
;
892 *new_l2_index
= l2_index
;
898 * alloc_compressed_cluster_offset
900 * For a given offset of the disk image, return cluster offset in
903 * If the offset is not found, allocate a new compressed cluster.
905 * Return the cluster offset if successful,
906 * Return 0, otherwise.
910 static uint64_t alloc_compressed_cluster_offset(BlockDriverState
*bs
,
914 BDRVQcowState
*s
= bs
->opaque
;
916 uint64_t l2_offset
, *l2_table
, cluster_offset
;
919 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_offset
, &l2_index
);
923 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
924 if (cluster_offset
& QCOW_OFLAG_COPIED
)
925 return cluster_offset
& ~QCOW_OFLAG_COPIED
;
928 free_any_clusters(bs
, cluster_offset
, 1);
930 cluster_offset
= alloc_bytes(bs
, compressed_size
);
931 nb_csectors
= ((cluster_offset
+ compressed_size
- 1) >> 9) -
932 (cluster_offset
>> 9);
934 cluster_offset
|= QCOW_OFLAG_COMPRESSED
|
935 ((uint64_t)nb_csectors
<< s
->csize_shift
);
937 /* update L2 table */
939 /* compressed clusters never have the copied flag */
941 l2_table
[l2_index
] = cpu_to_be64(cluster_offset
);
942 if (bdrv_pwrite(s
->hd
,
943 l2_offset
+ l2_index
* sizeof(uint64_t),
945 sizeof(uint64_t)) != sizeof(uint64_t))
948 return cluster_offset
;
951 typedef struct QCowL2Meta
959 static int alloc_cluster_link_l2(BlockDriverState
*bs
, uint64_t cluster_offset
,
962 BDRVQcowState
*s
= bs
->opaque
;
963 int i
, j
= 0, l2_index
, ret
;
964 uint64_t *old_cluster
, start_sect
, l2_offset
, *l2_table
;
966 if (m
->nb_clusters
== 0)
969 old_cluster
= qemu_malloc(m
->nb_clusters
* sizeof(uint64_t));
971 /* copy content of unmodified sectors */
972 start_sect
= (m
->offset
& ~(s
->cluster_size
- 1)) >> 9;
974 ret
= copy_sectors(bs
, start_sect
, cluster_offset
, 0, m
->n_start
);
979 if (m
->nb_available
& (s
->cluster_sectors
- 1)) {
980 uint64_t end
= m
->nb_available
& ~(uint64_t)(s
->cluster_sectors
- 1);
981 ret
= copy_sectors(bs
, start_sect
+ end
, cluster_offset
+ (end
<< 9),
982 m
->nb_available
- end
, s
->cluster_sectors
);
988 /* update L2 table */
989 if (!get_cluster_table(bs
, m
->offset
, &l2_table
, &l2_offset
, &l2_index
))
992 for (i
= 0; i
< m
->nb_clusters
; i
++) {
993 if(l2_table
[l2_index
+ i
] != 0)
994 old_cluster
[j
++] = l2_table
[l2_index
+ i
];
996 l2_table
[l2_index
+ i
] = cpu_to_be64((cluster_offset
+
997 (i
<< s
->cluster_bits
)) | QCOW_OFLAG_COPIED
);
1000 if (bdrv_pwrite(s
->hd
, l2_offset
+ l2_index
* sizeof(uint64_t),
1001 l2_table
+ l2_index
, m
->nb_clusters
* sizeof(uint64_t)) !=
1002 m
->nb_clusters
* sizeof(uint64_t))
1005 for (i
= 0; i
< j
; i
++)
1006 free_any_clusters(bs
, old_cluster
[i
], 1);
1010 qemu_free(old_cluster
);
1015 * alloc_cluster_offset
1017 * For a given offset of the disk image, return cluster offset in
1020 * If the offset is not found, allocate a new cluster.
1022 * Return the cluster offset if successful,
1023 * Return 0, otherwise.
1027 static uint64_t alloc_cluster_offset(BlockDriverState
*bs
,
1029 int n_start
, int n_end
,
1030 int *num
, QCowL2Meta
*m
)
1032 BDRVQcowState
*s
= bs
->opaque
;
1034 uint64_t l2_offset
, *l2_table
, cluster_offset
;
1035 int nb_clusters
, i
= 0;
1037 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_offset
, &l2_index
);
1041 nb_clusters
= size_to_clusters(s
, n_end
<< 9);
1043 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
1045 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
1047 /* We keep all QCOW_OFLAG_COPIED clusters */
1049 if (cluster_offset
& QCOW_OFLAG_COPIED
) {
1050 nb_clusters
= count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
1051 &l2_table
[l2_index
], 0, 0);
1053 cluster_offset
&= ~QCOW_OFLAG_COPIED
;
1059 /* for the moment, multiple compressed clusters are not managed */
1061 if (cluster_offset
& QCOW_OFLAG_COMPRESSED
)
1064 /* how many available clusters ? */
1066 while (i
< nb_clusters
) {
1067 i
+= count_contiguous_clusters(nb_clusters
- i
, s
->cluster_size
,
1068 &l2_table
[l2_index
], i
, 0);
1070 if(be64_to_cpu(l2_table
[l2_index
+ i
]))
1073 i
+= count_contiguous_free_clusters(nb_clusters
- i
,
1074 &l2_table
[l2_index
+ i
]);
1076 cluster_offset
= be64_to_cpu(l2_table
[l2_index
+ i
]);
1078 if ((cluster_offset
& QCOW_OFLAG_COPIED
) ||
1079 (cluster_offset
& QCOW_OFLAG_COMPRESSED
))
1084 /* allocate a new cluster */
1086 cluster_offset
= alloc_clusters(bs
, nb_clusters
* s
->cluster_size
);
1088 /* save info needed for meta data update */
1090 m
->n_start
= n_start
;
1091 m
->nb_clusters
= nb_clusters
;
1094 m
->nb_available
= MIN(nb_clusters
<< (s
->cluster_bits
- 9), n_end
);
1096 *num
= m
->nb_available
- n_start
;
1098 return cluster_offset
;
1101 static int qcow_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1102 int nb_sectors
, int *pnum
)
1104 uint64_t cluster_offset
;
1107 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, pnum
);
1109 return (cluster_offset
!= 0);
1112 static int decompress_buffer(uint8_t *out_buf
, int out_buf_size
,
1113 const uint8_t *buf
, int buf_size
)
1115 z_stream strm1
, *strm
= &strm1
;
1118 memset(strm
, 0, sizeof(*strm
));
1120 strm
->next_in
= (uint8_t *)buf
;
1121 strm
->avail_in
= buf_size
;
1122 strm
->next_out
= out_buf
;
1123 strm
->avail_out
= out_buf_size
;
1125 ret
= inflateInit2(strm
, -12);
1128 ret
= inflate(strm
, Z_FINISH
);
1129 out_len
= strm
->next_out
- out_buf
;
1130 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) ||
1131 out_len
!= out_buf_size
) {
1139 static int decompress_cluster(BDRVQcowState
*s
, uint64_t cluster_offset
)
1141 int ret
, csize
, nb_csectors
, sector_offset
;
1144 coffset
= cluster_offset
& s
->cluster_offset_mask
;
1145 if (s
->cluster_cache_offset
!= coffset
) {
1146 nb_csectors
= ((cluster_offset
>> s
->csize_shift
) & s
->csize_mask
) + 1;
1147 sector_offset
= coffset
& 511;
1148 csize
= nb_csectors
* 512 - sector_offset
;
1149 ret
= bdrv_read(s
->hd
, coffset
>> 9, s
->cluster_data
, nb_csectors
);
1153 if (decompress_buffer(s
->cluster_cache
, s
->cluster_size
,
1154 s
->cluster_data
+ sector_offset
, csize
) < 0) {
1157 s
->cluster_cache_offset
= coffset
;
1162 /* handle reading after the end of the backing file */
1163 static int backing_read1(BlockDriverState
*bs
,
1164 int64_t sector_num
, uint8_t *buf
, int nb_sectors
)
1167 if ((sector_num
+ nb_sectors
) <= bs
->total_sectors
)
1169 if (sector_num
>= bs
->total_sectors
)
1172 n1
= bs
->total_sectors
- sector_num
;
1173 memset(buf
+ n1
* 512, 0, 512 * (nb_sectors
- n1
));
1177 static int qcow_read(BlockDriverState
*bs
, int64_t sector_num
,
1178 uint8_t *buf
, int nb_sectors
)
1180 BDRVQcowState
*s
= bs
->opaque
;
1181 int ret
, index_in_cluster
, n
, n1
;
1182 uint64_t cluster_offset
;
1184 while (nb_sectors
> 0) {
1186 cluster_offset
= get_cluster_offset(bs
, sector_num
<< 9, &n
);
1187 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
1188 if (!cluster_offset
) {
1189 if (bs
->backing_hd
) {
1190 /* read from the base image */
1191 n1
= backing_read1(bs
->backing_hd
, sector_num
, buf
, n
);
1193 ret
= bdrv_read(bs
->backing_hd
, sector_num
, buf
, n1
);
1198 memset(buf
, 0, 512 * n
);
1200 } else if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
1201 if (decompress_cluster(s
, cluster_offset
) < 0)
1203 memcpy(buf
, s
->cluster_cache
+ index_in_cluster
* 512, 512 * n
);
1205 ret
= bdrv_pread(s
->hd
, cluster_offset
+ index_in_cluster
* 512, buf
, n
* 512);
1208 if (s
->crypt_method
) {
1209 encrypt_sectors(s
, sector_num
, buf
, buf
, n
, 0,
1210 &s
->aes_decrypt_key
);
1220 static int qcow_write(BlockDriverState
*bs
, int64_t sector_num
,
1221 const uint8_t *buf
, int nb_sectors
)
1223 BDRVQcowState
*s
= bs
->opaque
;
1224 int ret
, index_in_cluster
, n
;
1225 uint64_t cluster_offset
;
1229 while (nb_sectors
> 0) {
1230 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
1231 n_end
= index_in_cluster
+ nb_sectors
;
1232 if (s
->crypt_method
&&
1233 n_end
> QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
)
1234 n_end
= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
;
1235 cluster_offset
= alloc_cluster_offset(bs
, sector_num
<< 9,
1237 n_end
, &n
, &l2meta
);
1238 if (!cluster_offset
)
1240 if (s
->crypt_method
) {
1241 encrypt_sectors(s
, sector_num
, s
->cluster_data
, buf
, n
, 1,
1242 &s
->aes_encrypt_key
);
1243 ret
= bdrv_pwrite(s
->hd
, cluster_offset
+ index_in_cluster
* 512,
1244 s
->cluster_data
, n
* 512);
1246 ret
= bdrv_pwrite(s
->hd
, cluster_offset
+ index_in_cluster
* 512, buf
, n
* 512);
1248 if (ret
!= n
* 512 || alloc_cluster_link_l2(bs
, cluster_offset
, &l2meta
) < 0) {
1249 free_any_clusters(bs
, cluster_offset
, l2meta
.nb_clusters
);
1256 s
->cluster_cache_offset
= -1; /* disable compressed cache */
1260 typedef struct QCowAIOCB
{
1261 BlockDriverAIOCB common
;
1266 uint64_t cluster_offset
;
1267 uint8_t *cluster_data
;
1268 BlockDriverAIOCB
*hd_aiocb
;
1273 static void qcow_aio_read_cb(void *opaque
, int ret
);
1274 static void qcow_aio_read_bh(void *opaque
)
1276 QCowAIOCB
*acb
= opaque
;
1277 qemu_bh_delete(acb
->bh
);
1279 qcow_aio_read_cb(opaque
, 0);
1282 static int qcow_schedule_bh(QEMUBHFunc
*cb
, QCowAIOCB
*acb
)
1287 acb
->bh
= qemu_bh_new(cb
, acb
);
1291 qemu_bh_schedule(acb
->bh
);
1296 static void qcow_aio_read_cb(void *opaque
, int ret
)
1298 QCowAIOCB
*acb
= opaque
;
1299 BlockDriverState
*bs
= acb
->common
.bs
;
1300 BDRVQcowState
*s
= bs
->opaque
;
1301 int index_in_cluster
, n1
;
1303 acb
->hd_aiocb
= NULL
;
1306 acb
->common
.cb(acb
->common
.opaque
, ret
);
1307 qemu_aio_release(acb
);
1311 /* post process the read buffer */
1312 if (!acb
->cluster_offset
) {
1314 } else if (acb
->cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
1317 if (s
->crypt_method
) {
1318 encrypt_sectors(s
, acb
->sector_num
, acb
->buf
, acb
->buf
,
1320 &s
->aes_decrypt_key
);
1324 acb
->nb_sectors
-= acb
->n
;
1325 acb
->sector_num
+= acb
->n
;
1326 acb
->buf
+= acb
->n
* 512;
1328 if (acb
->nb_sectors
== 0) {
1329 /* request completed */
1330 acb
->common
.cb(acb
->common
.opaque
, 0);
1331 qemu_aio_release(acb
);
1335 /* prepare next AIO request */
1336 acb
->n
= acb
->nb_sectors
;
1337 acb
->cluster_offset
= get_cluster_offset(bs
, acb
->sector_num
<< 9, &acb
->n
);
1338 index_in_cluster
= acb
->sector_num
& (s
->cluster_sectors
- 1);
1340 if (!acb
->cluster_offset
) {
1341 if (bs
->backing_hd
) {
1342 /* read from the base image */
1343 n1
= backing_read1(bs
->backing_hd
, acb
->sector_num
,
1346 acb
->hd_aiocb
= bdrv_aio_read(bs
->backing_hd
, acb
->sector_num
,
1347 acb
->buf
, acb
->n
, qcow_aio_read_cb
, acb
);
1348 if (acb
->hd_aiocb
== NULL
)
1351 ret
= qcow_schedule_bh(qcow_aio_read_bh
, acb
);
1356 /* Note: in this case, no need to wait */
1357 memset(acb
->buf
, 0, 512 * acb
->n
);
1358 ret
= qcow_schedule_bh(qcow_aio_read_bh
, acb
);
1362 } else if (acb
->cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
1363 /* add AIO support for compressed blocks ? */
1364 if (decompress_cluster(s
, acb
->cluster_offset
) < 0)
1367 s
->cluster_cache
+ index_in_cluster
* 512, 512 * acb
->n
);
1368 ret
= qcow_schedule_bh(qcow_aio_read_bh
, acb
);
1372 if ((acb
->cluster_offset
& 511) != 0) {
1376 acb
->hd_aiocb
= bdrv_aio_read(s
->hd
,
1377 (acb
->cluster_offset
>> 9) + index_in_cluster
,
1378 acb
->buf
, acb
->n
, qcow_aio_read_cb
, acb
);
1379 if (acb
->hd_aiocb
== NULL
)
1384 static QCowAIOCB
*qcow_aio_setup(BlockDriverState
*bs
,
1385 int64_t sector_num
, uint8_t *buf
, int nb_sectors
,
1386 BlockDriverCompletionFunc
*cb
, void *opaque
)
1390 acb
= qemu_aio_get(bs
, cb
, opaque
);
1393 acb
->hd_aiocb
= NULL
;
1394 acb
->sector_num
= sector_num
;
1396 acb
->nb_sectors
= nb_sectors
;
1398 acb
->cluster_offset
= 0;
1399 acb
->l2meta
.nb_clusters
= 0;
1403 static BlockDriverAIOCB
*qcow_aio_read(BlockDriverState
*bs
,
1404 int64_t sector_num
, uint8_t *buf
, int nb_sectors
,
1405 BlockDriverCompletionFunc
*cb
, void *opaque
)
1409 acb
= qcow_aio_setup(bs
, sector_num
, buf
, nb_sectors
, cb
, opaque
);
1413 qcow_aio_read_cb(acb
, 0);
1414 return &acb
->common
;
1417 static void qcow_aio_write_cb(void *opaque
, int ret
)
1419 QCowAIOCB
*acb
= opaque
;
1420 BlockDriverState
*bs
= acb
->common
.bs
;
1421 BDRVQcowState
*s
= bs
->opaque
;
1422 int index_in_cluster
;
1423 const uint8_t *src_buf
;
1426 acb
->hd_aiocb
= NULL
;
1430 acb
->common
.cb(acb
->common
.opaque
, ret
);
1431 qemu_aio_release(acb
);
1435 if (alloc_cluster_link_l2(bs
, acb
->cluster_offset
, &acb
->l2meta
) < 0) {
1436 free_any_clusters(bs
, acb
->cluster_offset
, acb
->l2meta
.nb_clusters
);
1440 acb
->nb_sectors
-= acb
->n
;
1441 acb
->sector_num
+= acb
->n
;
1442 acb
->buf
+= acb
->n
* 512;
1444 if (acb
->nb_sectors
== 0) {
1445 /* request completed */
1446 acb
->common
.cb(acb
->common
.opaque
, 0);
1447 qemu_aio_release(acb
);
1451 index_in_cluster
= acb
->sector_num
& (s
->cluster_sectors
- 1);
1452 n_end
= index_in_cluster
+ acb
->nb_sectors
;
1453 if (s
->crypt_method
&&
1454 n_end
> QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
)
1455 n_end
= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_sectors
;
1457 acb
->cluster_offset
= alloc_cluster_offset(bs
, acb
->sector_num
<< 9,
1459 n_end
, &acb
->n
, &acb
->l2meta
);
1460 if (!acb
->cluster_offset
|| (acb
->cluster_offset
& 511) != 0) {
1464 if (s
->crypt_method
) {
1465 if (!acb
->cluster_data
) {
1466 acb
->cluster_data
= qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS
*
1469 encrypt_sectors(s
, acb
->sector_num
, acb
->cluster_data
, acb
->buf
,
1470 acb
->n
, 1, &s
->aes_encrypt_key
);
1471 src_buf
= acb
->cluster_data
;
1475 acb
->hd_aiocb
= bdrv_aio_write(s
->hd
,
1476 (acb
->cluster_offset
>> 9) + index_in_cluster
,
1478 qcow_aio_write_cb
, acb
);
1479 if (acb
->hd_aiocb
== NULL
)
1483 static BlockDriverAIOCB
*qcow_aio_write(BlockDriverState
*bs
,
1484 int64_t sector_num
, const uint8_t *buf
, int nb_sectors
,
1485 BlockDriverCompletionFunc
*cb
, void *opaque
)
1487 BDRVQcowState
*s
= bs
->opaque
;
1490 s
->cluster_cache_offset
= -1; /* disable compressed cache */
1492 acb
= qcow_aio_setup(bs
, sector_num
, (uint8_t*)buf
, nb_sectors
, cb
, opaque
);
1496 qcow_aio_write_cb(acb
, 0);
1497 return &acb
->common
;
1500 static void qcow_aio_cancel(BlockDriverAIOCB
*blockacb
)
1502 QCowAIOCB
*acb
= (QCowAIOCB
*)blockacb
;
1504 bdrv_aio_cancel(acb
->hd_aiocb
);
1505 qemu_aio_release(acb
);
1508 static void qcow_close(BlockDriverState
*bs
)
1510 BDRVQcowState
*s
= bs
->opaque
;
1511 qemu_free(s
->l1_table
);
1512 qemu_free(s
->l2_cache
);
1513 qemu_free(s
->cluster_cache
);
1514 qemu_free(s
->cluster_data
);
1519 /* XXX: use std qcow open function ? */
1520 typedef struct QCowCreateState
{
1523 uint16_t *refcount_block
;
1524 uint64_t *refcount_table
;
1525 int64_t l1_table_offset
;
1526 int64_t refcount_table_offset
;
1527 int64_t refcount_block_offset
;
1530 static void create_refcount_update(QCowCreateState
*s
,
1531 int64_t offset
, int64_t size
)
1534 int64_t start
, last
, cluster_offset
;
1537 start
= offset
& ~(s
->cluster_size
- 1);
1538 last
= (offset
+ size
- 1) & ~(s
->cluster_size
- 1);
1539 for(cluster_offset
= start
; cluster_offset
<= last
;
1540 cluster_offset
+= s
->cluster_size
) {
1541 p
= &s
->refcount_block
[cluster_offset
>> s
->cluster_bits
];
1542 refcount
= be16_to_cpu(*p
);
1544 *p
= cpu_to_be16(refcount
);
1548 static int qcow_create2(const char *filename
, int64_t total_size
,
1549 const char *backing_file
, const char *backing_format
,
1553 int fd
, header_size
, backing_filename_len
, l1_size
, i
, shift
, l2_bits
;
1554 int backing_format_len
= 0;
1556 uint64_t tmp
, offset
;
1557 QCowCreateState s1
, *s
= &s1
;
1558 QCowExtension ext_bf
= {0, 0};
1561 memset(s
, 0, sizeof(*s
));
1563 fd
= open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, 0644);
1566 memset(&header
, 0, sizeof(header
));
1567 header
.magic
= cpu_to_be32(QCOW_MAGIC
);
1568 header
.version
= cpu_to_be32(QCOW_VERSION
);
1569 header
.size
= cpu_to_be64(total_size
* 512);
1570 header_size
= sizeof(header
);
1571 backing_filename_len
= 0;
1573 if (backing_format
) {
1574 ext_bf
.magic
= QCOW_EXT_MAGIC_BACKING_FORMAT
;
1575 backing_format_len
= strlen(backing_format
);
1576 ext_bf
.len
= (backing_format_len
+ 7) & ~7;
1577 header_size
+= ((sizeof(ext_bf
) + ext_bf
.len
+ 7) & ~7);
1579 header
.backing_file_offset
= cpu_to_be64(header_size
);
1580 backing_filename_len
= strlen(backing_file
);
1581 header
.backing_file_size
= cpu_to_be32(backing_filename_len
);
1582 header_size
+= backing_filename_len
;
1584 s
->cluster_bits
= 12; /* 4 KB clusters */
1585 s
->cluster_size
= 1 << s
->cluster_bits
;
1586 header
.cluster_bits
= cpu_to_be32(s
->cluster_bits
);
1587 header_size
= (header_size
+ 7) & ~7;
1588 if (flags
& BLOCK_FLAG_ENCRYPT
) {
1589 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_AES
);
1591 header
.crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
1593 l2_bits
= s
->cluster_bits
- 3;
1594 shift
= s
->cluster_bits
+ l2_bits
;
1595 l1_size
= (((total_size
* 512) + (1LL << shift
) - 1) >> shift
);
1596 offset
= align_offset(header_size
, s
->cluster_size
);
1597 s
->l1_table_offset
= offset
;
1598 header
.l1_table_offset
= cpu_to_be64(s
->l1_table_offset
);
1599 header
.l1_size
= cpu_to_be32(l1_size
);
1600 offset
+= align_offset(l1_size
* sizeof(uint64_t), s
->cluster_size
);
1602 s
->refcount_table
= qemu_mallocz(s
->cluster_size
);
1603 s
->refcount_block
= qemu_mallocz(s
->cluster_size
);
1605 s
->refcount_table_offset
= offset
;
1606 header
.refcount_table_offset
= cpu_to_be64(offset
);
1607 header
.refcount_table_clusters
= cpu_to_be32(1);
1608 offset
+= s
->cluster_size
;
1610 s
->refcount_table
[0] = cpu_to_be64(offset
);
1611 s
->refcount_block_offset
= offset
;
1612 offset
+= s
->cluster_size
;
1614 /* update refcounts */
1615 create_refcount_update(s
, 0, header_size
);
1616 create_refcount_update(s
, s
->l1_table_offset
, l1_size
* sizeof(uint64_t));
1617 create_refcount_update(s
, s
->refcount_table_offset
, s
->cluster_size
);
1618 create_refcount_update(s
, s
->refcount_block_offset
, s
->cluster_size
);
1620 /* write all the data */
1621 write(fd
, &header
, sizeof(header
));
1623 if (backing_format_len
) {
1625 int d
= ext_bf
.len
- backing_format_len
;
1627 memset(zero
, 0, sizeof(zero
));
1628 cpu_to_be32s(&ext_bf
.magic
);
1629 cpu_to_be32s(&ext_bf
.len
);
1630 write(fd
, &ext_bf
, sizeof(ext_bf
));
1631 write(fd
, backing_format
, backing_format_len
);
1636 write(fd
, backing_file
, backing_filename_len
);
1638 lseek(fd
, s
->l1_table_offset
, SEEK_SET
);
1640 for(i
= 0;i
< l1_size
; i
++) {
1641 write(fd
, &tmp
, sizeof(tmp
));
1643 lseek(fd
, s
->refcount_table_offset
, SEEK_SET
);
1644 write(fd
, s
->refcount_table
, s
->cluster_size
);
1646 lseek(fd
, s
->refcount_block_offset
, SEEK_SET
);
1647 write(fd
, s
->refcount_block
, s
->cluster_size
);
1649 qemu_free(s
->refcount_table
);
1650 qemu_free(s
->refcount_block
);
1655 static int qcow_create(const char *filename
, int64_t total_size
,
1656 const char *backing_file
, int flags
)
1658 return qcow_create2(filename
, total_size
, backing_file
, NULL
, flags
);
1661 static int qcow_make_empty(BlockDriverState
*bs
)
1664 /* XXX: not correct */
1665 BDRVQcowState
*s
= bs
->opaque
;
1666 uint32_t l1_length
= s
->l1_size
* sizeof(uint64_t);
1669 memset(s
->l1_table
, 0, l1_length
);
1670 if (bdrv_pwrite(s
->hd
, s
->l1_table_offset
, s
->l1_table
, l1_length
) < 0)
1672 ret
= bdrv_truncate(s
->hd
, s
->l1_table_offset
+ l1_length
);
1681 /* XXX: put compressed sectors first, then all the cluster aligned
1682 tables to avoid losing bytes in alignment */
1683 static int qcow_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1684 const uint8_t *buf
, int nb_sectors
)
1686 BDRVQcowState
*s
= bs
->opaque
;
1690 uint64_t cluster_offset
;
1692 if (nb_sectors
== 0) {
1693 /* align end of file to a sector boundary to ease reading with
1694 sector based I/Os */
1695 cluster_offset
= bdrv_getlength(s
->hd
);
1696 cluster_offset
= (cluster_offset
+ 511) & ~511;
1697 bdrv_truncate(s
->hd
, cluster_offset
);
1701 if (nb_sectors
!= s
->cluster_sectors
)
1704 out_buf
= qemu_malloc(s
->cluster_size
+ (s
->cluster_size
/ 1000) + 128);
1706 /* best compression, small window, no zlib header */
1707 memset(&strm
, 0, sizeof(strm
));
1708 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
,
1710 9, Z_DEFAULT_STRATEGY
);
1716 strm
.avail_in
= s
->cluster_size
;
1717 strm
.next_in
= (uint8_t *)buf
;
1718 strm
.avail_out
= s
->cluster_size
;
1719 strm
.next_out
= out_buf
;
1721 ret
= deflate(&strm
, Z_FINISH
);
1722 if (ret
!= Z_STREAM_END
&& ret
!= Z_OK
) {
1727 out_len
= strm
.next_out
- out_buf
;
1731 if (ret
!= Z_STREAM_END
|| out_len
>= s
->cluster_size
) {
1732 /* could not compress: write normal cluster */
1733 qcow_write(bs
, sector_num
, buf
, s
->cluster_sectors
);
1735 cluster_offset
= alloc_compressed_cluster_offset(bs
, sector_num
<< 9,
1737 if (!cluster_offset
)
1739 cluster_offset
&= s
->cluster_offset_mask
;
1740 if (bdrv_pwrite(s
->hd
, cluster_offset
, out_buf
, out_len
) != out_len
) {
1750 static void qcow_flush(BlockDriverState
*bs
)
1752 BDRVQcowState
*s
= bs
->opaque
;
1756 static int qcow_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1758 BDRVQcowState
*s
= bs
->opaque
;
1759 bdi
->cluster_size
= s
->cluster_size
;
1760 bdi
->vm_state_offset
= (int64_t)s
->l1_vm_state_index
<<
1761 (s
->cluster_bits
+ s
->l2_bits
);
1765 /*********************************************************/
1766 /* snapshot support */
1768 /* update the refcounts of snapshots and the copied flag */
1769 static int update_snapshot_refcount(BlockDriverState
*bs
,
1770 int64_t l1_table_offset
,
1774 BDRVQcowState
*s
= bs
->opaque
;
1775 uint64_t *l1_table
, *l2_table
, l2_offset
, offset
, l1_size2
, l1_allocated
;
1776 int64_t old_offset
, old_l2_offset
;
1777 int l2_size
, i
, j
, l1_modified
, l2_modified
, nb_csectors
, refcount
;
1783 l1_size2
= l1_size
* sizeof(uint64_t);
1785 if (l1_table_offset
!= s
->l1_table_offset
) {
1786 l1_table
= qemu_malloc(l1_size2
);
1788 if (bdrv_pread(s
->hd
, l1_table_offset
,
1789 l1_table
, l1_size2
) != l1_size2
)
1791 for(i
= 0;i
< l1_size
; i
++)
1792 be64_to_cpus(&l1_table
[i
]);
1794 assert(l1_size
== s
->l1_size
);
1795 l1_table
= s
->l1_table
;
1799 l2_size
= s
->l2_size
* sizeof(uint64_t);
1800 l2_table
= qemu_malloc(l2_size
);
1802 for(i
= 0; i
< l1_size
; i
++) {
1803 l2_offset
= l1_table
[i
];
1805 old_l2_offset
= l2_offset
;
1806 l2_offset
&= ~QCOW_OFLAG_COPIED
;
1808 if (bdrv_pread(s
->hd
, l2_offset
, l2_table
, l2_size
) != l2_size
)
1810 for(j
= 0; j
< s
->l2_size
; j
++) {
1811 offset
= be64_to_cpu(l2_table
[j
]);
1813 old_offset
= offset
;
1814 offset
&= ~QCOW_OFLAG_COPIED
;
1815 if (offset
& QCOW_OFLAG_COMPRESSED
) {
1816 nb_csectors
= ((offset
>> s
->csize_shift
) &
1819 update_refcount(bs
, (offset
& s
->cluster_offset_mask
) & ~511,
1820 nb_csectors
* 512, addend
);
1821 /* compressed clusters are never modified */
1825 refcount
= update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, addend
);
1827 refcount
= get_refcount(bs
, offset
>> s
->cluster_bits
);
1831 if (refcount
== 1) {
1832 offset
|= QCOW_OFLAG_COPIED
;
1834 if (offset
!= old_offset
) {
1835 l2_table
[j
] = cpu_to_be64(offset
);
1841 if (bdrv_pwrite(s
->hd
,
1842 l2_offset
, l2_table
, l2_size
) != l2_size
)
1847 refcount
= update_cluster_refcount(bs
, l2_offset
>> s
->cluster_bits
, addend
);
1849 refcount
= get_refcount(bs
, l2_offset
>> s
->cluster_bits
);
1851 if (refcount
== 1) {
1852 l2_offset
|= QCOW_OFLAG_COPIED
;
1854 if (l2_offset
!= old_l2_offset
) {
1855 l1_table
[i
] = l2_offset
;
1861 for(i
= 0; i
< l1_size
; i
++)
1862 cpu_to_be64s(&l1_table
[i
]);
1863 if (bdrv_pwrite(s
->hd
, l1_table_offset
, l1_table
,
1864 l1_size2
) != l1_size2
)
1866 for(i
= 0; i
< l1_size
; i
++)
1867 be64_to_cpus(&l1_table
[i
]);
1870 qemu_free(l1_table
);
1871 qemu_free(l2_table
);
1875 qemu_free(l1_table
);
1876 qemu_free(l2_table
);
1880 static void qcow_free_snapshots(BlockDriverState
*bs
)
1882 BDRVQcowState
*s
= bs
->opaque
;
1885 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1886 qemu_free(s
->snapshots
[i
].name
);
1887 qemu_free(s
->snapshots
[i
].id_str
);
1889 qemu_free(s
->snapshots
);
1890 s
->snapshots
= NULL
;
1891 s
->nb_snapshots
= 0;
1894 static int qcow_read_snapshots(BlockDriverState
*bs
)
1896 BDRVQcowState
*s
= bs
->opaque
;
1897 QCowSnapshotHeader h
;
1899 int i
, id_str_size
, name_size
;
1901 uint32_t extra_data_size
;
1903 if (!s
->nb_snapshots
) {
1904 s
->snapshots
= NULL
;
1905 s
->snapshots_size
= 0;
1909 offset
= s
->snapshots_offset
;
1910 s
->snapshots
= qemu_mallocz(s
->nb_snapshots
* sizeof(QCowSnapshot
));
1911 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1912 offset
= align_offset(offset
, 8);
1913 if (bdrv_pread(s
->hd
, offset
, &h
, sizeof(h
)) != sizeof(h
))
1915 offset
+= sizeof(h
);
1916 sn
= s
->snapshots
+ i
;
1917 sn
->l1_table_offset
= be64_to_cpu(h
.l1_table_offset
);
1918 sn
->l1_size
= be32_to_cpu(h
.l1_size
);
1919 sn
->vm_state_size
= be32_to_cpu(h
.vm_state_size
);
1920 sn
->date_sec
= be32_to_cpu(h
.date_sec
);
1921 sn
->date_nsec
= be32_to_cpu(h
.date_nsec
);
1922 sn
->vm_clock_nsec
= be64_to_cpu(h
.vm_clock_nsec
);
1923 extra_data_size
= be32_to_cpu(h
.extra_data_size
);
1925 id_str_size
= be16_to_cpu(h
.id_str_size
);
1926 name_size
= be16_to_cpu(h
.name_size
);
1928 offset
+= extra_data_size
;
1930 sn
->id_str
= qemu_malloc(id_str_size
+ 1);
1931 if (bdrv_pread(s
->hd
, offset
, sn
->id_str
, id_str_size
) != id_str_size
)
1933 offset
+= id_str_size
;
1934 sn
->id_str
[id_str_size
] = '\0';
1936 sn
->name
= qemu_malloc(name_size
+ 1);
1937 if (bdrv_pread(s
->hd
, offset
, sn
->name
, name_size
) != name_size
)
1939 offset
+= name_size
;
1940 sn
->name
[name_size
] = '\0';
1942 s
->snapshots_size
= offset
- s
->snapshots_offset
;
1945 qcow_free_snapshots(bs
);
1949 /* add at the end of the file a new list of snapshots */
1950 static int qcow_write_snapshots(BlockDriverState
*bs
)
1952 BDRVQcowState
*s
= bs
->opaque
;
1954 QCowSnapshotHeader h
;
1955 int i
, name_size
, id_str_size
, snapshots_size
;
1958 int64_t offset
, snapshots_offset
;
1960 /* compute the size of the snapshots */
1962 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1963 sn
= s
->snapshots
+ i
;
1964 offset
= align_offset(offset
, 8);
1965 offset
+= sizeof(h
);
1966 offset
+= strlen(sn
->id_str
);
1967 offset
+= strlen(sn
->name
);
1969 snapshots_size
= offset
;
1971 snapshots_offset
= alloc_clusters(bs
, snapshots_size
);
1972 offset
= snapshots_offset
;
1974 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1975 sn
= s
->snapshots
+ i
;
1976 memset(&h
, 0, sizeof(h
));
1977 h
.l1_table_offset
= cpu_to_be64(sn
->l1_table_offset
);
1978 h
.l1_size
= cpu_to_be32(sn
->l1_size
);
1979 h
.vm_state_size
= cpu_to_be32(sn
->vm_state_size
);
1980 h
.date_sec
= cpu_to_be32(sn
->date_sec
);
1981 h
.date_nsec
= cpu_to_be32(sn
->date_nsec
);
1982 h
.vm_clock_nsec
= cpu_to_be64(sn
->vm_clock_nsec
);
1984 id_str_size
= strlen(sn
->id_str
);
1985 name_size
= strlen(sn
->name
);
1986 h
.id_str_size
= cpu_to_be16(id_str_size
);
1987 h
.name_size
= cpu_to_be16(name_size
);
1988 offset
= align_offset(offset
, 8);
1989 if (bdrv_pwrite(s
->hd
, offset
, &h
, sizeof(h
)) != sizeof(h
))
1991 offset
+= sizeof(h
);
1992 if (bdrv_pwrite(s
->hd
, offset
, sn
->id_str
, id_str_size
) != id_str_size
)
1994 offset
+= id_str_size
;
1995 if (bdrv_pwrite(s
->hd
, offset
, sn
->name
, name_size
) != name_size
)
1997 offset
+= name_size
;
2000 /* update the various header fields */
2001 data64
= cpu_to_be64(snapshots_offset
);
2002 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, snapshots_offset
),
2003 &data64
, sizeof(data64
)) != sizeof(data64
))
2005 data32
= cpu_to_be32(s
->nb_snapshots
);
2006 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, nb_snapshots
),
2007 &data32
, sizeof(data32
)) != sizeof(data32
))
2010 /* free the old snapshot table */
2011 free_clusters(bs
, s
->snapshots_offset
, s
->snapshots_size
);
2012 s
->snapshots_offset
= snapshots_offset
;
2013 s
->snapshots_size
= snapshots_size
;
2019 static void find_new_snapshot_id(BlockDriverState
*bs
,
2020 char *id_str
, int id_str_size
)
2022 BDRVQcowState
*s
= bs
->opaque
;
2024 int i
, id
, id_max
= 0;
2026 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
2027 sn
= s
->snapshots
+ i
;
2028 id
= strtoul(sn
->id_str
, NULL
, 10);
2032 snprintf(id_str
, id_str_size
, "%d", id_max
+ 1);
2035 static int find_snapshot_by_id(BlockDriverState
*bs
, const char *id_str
)
2037 BDRVQcowState
*s
= bs
->opaque
;
2040 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
2041 if (!strcmp(s
->snapshots
[i
].id_str
, id_str
))
2047 static int find_snapshot_by_id_or_name(BlockDriverState
*bs
, const char *name
)
2049 BDRVQcowState
*s
= bs
->opaque
;
2052 ret
= find_snapshot_by_id(bs
, name
);
2055 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
2056 if (!strcmp(s
->snapshots
[i
].name
, name
))
2062 /* if no id is provided, a new one is constructed */
2063 static int qcow_snapshot_create(BlockDriverState
*bs
,
2064 QEMUSnapshotInfo
*sn_info
)
2066 BDRVQcowState
*s
= bs
->opaque
;
2067 QCowSnapshot
*snapshots1
, sn1
, *sn
= &sn1
;
2069 uint64_t *l1_table
= NULL
;
2071 memset(sn
, 0, sizeof(*sn
));
2073 if (sn_info
->id_str
[0] == '\0') {
2074 /* compute a new id */
2075 find_new_snapshot_id(bs
, sn_info
->id_str
, sizeof(sn_info
->id_str
));
2078 /* check that the ID is unique */
2079 if (find_snapshot_by_id(bs
, sn_info
->id_str
) >= 0)
2082 sn
->id_str
= qemu_strdup(sn_info
->id_str
);
2085 sn
->name
= qemu_strdup(sn_info
->name
);
2088 sn
->vm_state_size
= sn_info
->vm_state_size
;
2089 sn
->date_sec
= sn_info
->date_sec
;
2090 sn
->date_nsec
= sn_info
->date_nsec
;
2091 sn
->vm_clock_nsec
= sn_info
->vm_clock_nsec
;
2093 ret
= update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 1);
2097 /* create the L1 table of the snapshot */
2098 sn
->l1_table_offset
= alloc_clusters(bs
, s
->l1_size
* sizeof(uint64_t));
2099 sn
->l1_size
= s
->l1_size
;
2101 l1_table
= qemu_malloc(s
->l1_size
* sizeof(uint64_t));
2102 for(i
= 0; i
< s
->l1_size
; i
++) {
2103 l1_table
[i
] = cpu_to_be64(s
->l1_table
[i
]);
2105 if (bdrv_pwrite(s
->hd
, sn
->l1_table_offset
,
2106 l1_table
, s
->l1_size
* sizeof(uint64_t)) !=
2107 (s
->l1_size
* sizeof(uint64_t)))
2109 qemu_free(l1_table
);
2112 snapshots1
= qemu_malloc((s
->nb_snapshots
+ 1) * sizeof(QCowSnapshot
));
2114 memcpy(snapshots1
, s
->snapshots
, s
->nb_snapshots
* sizeof(QCowSnapshot
));
2115 qemu_free(s
->snapshots
);
2117 s
->snapshots
= snapshots1
;
2118 s
->snapshots
[s
->nb_snapshots
++] = *sn
;
2120 if (qcow_write_snapshots(bs
) < 0)
2123 check_refcounts(bs
);
2127 qemu_free(sn
->name
);
2128 qemu_free(l1_table
);
2132 /* copy the snapshot 'snapshot_name' into the current disk image */
2133 static int qcow_snapshot_goto(BlockDriverState
*bs
,
2134 const char *snapshot_id
)
2136 BDRVQcowState
*s
= bs
->opaque
;
2138 int i
, snapshot_index
, l1_size2
;
2140 snapshot_index
= find_snapshot_by_id_or_name(bs
, snapshot_id
);
2141 if (snapshot_index
< 0)
2143 sn
= &s
->snapshots
[snapshot_index
];
2145 if (update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, -1) < 0)
2148 if (grow_l1_table(bs
, sn
->l1_size
) < 0)
2151 s
->l1_size
= sn
->l1_size
;
2152 l1_size2
= s
->l1_size
* sizeof(uint64_t);
2153 /* copy the snapshot l1 table to the current l1 table */
2154 if (bdrv_pread(s
->hd
, sn
->l1_table_offset
,
2155 s
->l1_table
, l1_size2
) != l1_size2
)
2157 if (bdrv_pwrite(s
->hd
, s
->l1_table_offset
,
2158 s
->l1_table
, l1_size2
) != l1_size2
)
2160 for(i
= 0;i
< s
->l1_size
; i
++) {
2161 be64_to_cpus(&s
->l1_table
[i
]);
2164 if (update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 1) < 0)
2168 check_refcounts(bs
);
2175 static int qcow_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
2177 BDRVQcowState
*s
= bs
->opaque
;
2179 int snapshot_index
, ret
;
2181 snapshot_index
= find_snapshot_by_id_or_name(bs
, snapshot_id
);
2182 if (snapshot_index
< 0)
2184 sn
= &s
->snapshots
[snapshot_index
];
2186 ret
= update_snapshot_refcount(bs
, sn
->l1_table_offset
, sn
->l1_size
, -1);
2189 /* must update the copied flag on the current cluster offsets */
2190 ret
= update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 0);
2193 free_clusters(bs
, sn
->l1_table_offset
, sn
->l1_size
* sizeof(uint64_t));
2195 qemu_free(sn
->id_str
);
2196 qemu_free(sn
->name
);
2197 memmove(sn
, sn
+ 1, (s
->nb_snapshots
- snapshot_index
- 1) * sizeof(*sn
));
2199 ret
= qcow_write_snapshots(bs
);
2201 /* XXX: restore snapshot if error ? */
2205 check_refcounts(bs
);
2210 static int qcow_snapshot_list(BlockDriverState
*bs
,
2211 QEMUSnapshotInfo
**psn_tab
)
2213 BDRVQcowState
*s
= bs
->opaque
;
2214 QEMUSnapshotInfo
*sn_tab
, *sn_info
;
2218 sn_tab
= qemu_mallocz(s
->nb_snapshots
* sizeof(QEMUSnapshotInfo
));
2219 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
2220 sn_info
= sn_tab
+ i
;
2221 sn
= s
->snapshots
+ i
;
2222 pstrcpy(sn_info
->id_str
, sizeof(sn_info
->id_str
),
2224 pstrcpy(sn_info
->name
, sizeof(sn_info
->name
),
2226 sn_info
->vm_state_size
= sn
->vm_state_size
;
2227 sn_info
->date_sec
= sn
->date_sec
;
2228 sn_info
->date_nsec
= sn
->date_nsec
;
2229 sn_info
->vm_clock_nsec
= sn
->vm_clock_nsec
;
2232 return s
->nb_snapshots
;
2235 /*********************************************************/
2236 /* refcount handling */
2238 static int refcount_init(BlockDriverState
*bs
)
2240 BDRVQcowState
*s
= bs
->opaque
;
2241 int ret
, refcount_table_size2
, i
;
2243 s
->refcount_block_cache
= qemu_malloc(s
->cluster_size
);
2244 refcount_table_size2
= s
->refcount_table_size
* sizeof(uint64_t);
2245 s
->refcount_table
= qemu_malloc(refcount_table_size2
);
2246 if (s
->refcount_table_size
> 0) {
2247 ret
= bdrv_pread(s
->hd
, s
->refcount_table_offset
,
2248 s
->refcount_table
, refcount_table_size2
);
2249 if (ret
!= refcount_table_size2
)
2251 for(i
= 0; i
< s
->refcount_table_size
; i
++)
2252 be64_to_cpus(&s
->refcount_table
[i
]);
2259 static void refcount_close(BlockDriverState
*bs
)
2261 BDRVQcowState
*s
= bs
->opaque
;
2262 qemu_free(s
->refcount_block_cache
);
2263 qemu_free(s
->refcount_table
);
2267 static int load_refcount_block(BlockDriverState
*bs
,
2268 int64_t refcount_block_offset
)
2270 BDRVQcowState
*s
= bs
->opaque
;
2272 ret
= bdrv_pread(s
->hd
, refcount_block_offset
, s
->refcount_block_cache
,
2274 if (ret
!= s
->cluster_size
)
2276 s
->refcount_block_cache_offset
= refcount_block_offset
;
2280 static int get_refcount(BlockDriverState
*bs
, int64_t cluster_index
)
2282 BDRVQcowState
*s
= bs
->opaque
;
2283 int refcount_table_index
, block_index
;
2284 int64_t refcount_block_offset
;
2286 refcount_table_index
= cluster_index
>> (s
->cluster_bits
- REFCOUNT_SHIFT
);
2287 if (refcount_table_index
>= s
->refcount_table_size
)
2289 refcount_block_offset
= s
->refcount_table
[refcount_table_index
];
2290 if (!refcount_block_offset
)
2292 if (refcount_block_offset
!= s
->refcount_block_cache_offset
) {
2293 /* better than nothing: return allocated if read error */
2294 if (load_refcount_block(bs
, refcount_block_offset
) < 0)
2297 block_index
= cluster_index
&
2298 ((1 << (s
->cluster_bits
- REFCOUNT_SHIFT
)) - 1);
2299 return be16_to_cpu(s
->refcount_block_cache
[block_index
]);
2302 /* return < 0 if error */
2303 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, int64_t size
)
2305 BDRVQcowState
*s
= bs
->opaque
;
2308 nb_clusters
= size_to_clusters(s
, size
);
2310 for(i
= 0; i
< nb_clusters
; i
++) {
2311 int64_t i
= s
->free_cluster_index
++;
2312 if (get_refcount(bs
, i
) != 0)
2316 printf("alloc_clusters: size=%lld -> %lld\n",
2318 (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
);
2320 return (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
;
2323 static int64_t alloc_clusters(BlockDriverState
*bs
, int64_t size
)
2327 offset
= alloc_clusters_noref(bs
, size
);
2328 update_refcount(bs
, offset
, size
, 1);
2332 /* only used to allocate compressed sectors. We try to allocate
2333 contiguous sectors. size must be <= cluster_size */
2334 static int64_t alloc_bytes(BlockDriverState
*bs
, int size
)
2336 BDRVQcowState
*s
= bs
->opaque
;
2337 int64_t offset
, cluster_offset
;
2338 int free_in_cluster
;
2340 assert(size
> 0 && size
<= s
->cluster_size
);
2341 if (s
->free_byte_offset
== 0) {
2342 s
->free_byte_offset
= alloc_clusters(bs
, s
->cluster_size
);
2345 free_in_cluster
= s
->cluster_size
-
2346 (s
->free_byte_offset
& (s
->cluster_size
- 1));
2347 if (size
<= free_in_cluster
) {
2348 /* enough space in current cluster */
2349 offset
= s
->free_byte_offset
;
2350 s
->free_byte_offset
+= size
;
2351 free_in_cluster
-= size
;
2352 if (free_in_cluster
== 0)
2353 s
->free_byte_offset
= 0;
2354 if ((offset
& (s
->cluster_size
- 1)) != 0)
2355 update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, 1);
2357 offset
= alloc_clusters(bs
, s
->cluster_size
);
2358 cluster_offset
= s
->free_byte_offset
& ~(s
->cluster_size
- 1);
2359 if ((cluster_offset
+ s
->cluster_size
) == offset
) {
2360 /* we are lucky: contiguous data */
2361 offset
= s
->free_byte_offset
;
2362 update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, 1);
2363 s
->free_byte_offset
+= size
;
2365 s
->free_byte_offset
= offset
;
2372 static void free_clusters(BlockDriverState
*bs
,
2373 int64_t offset
, int64_t size
)
2375 update_refcount(bs
, offset
, size
, -1);
2378 static int grow_refcount_table(BlockDriverState
*bs
, int min_size
)
2380 BDRVQcowState
*s
= bs
->opaque
;
2381 int new_table_size
, new_table_size2
, refcount_table_clusters
, i
, ret
;
2382 uint64_t *new_table
;
2383 int64_t table_offset
;
2386 int64_t old_table_offset
;
2388 if (min_size
<= s
->refcount_table_size
)
2390 /* compute new table size */
2391 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2393 if (refcount_table_clusters
== 0) {
2394 refcount_table_clusters
= 1;
2396 refcount_table_clusters
= (refcount_table_clusters
* 3 + 1) / 2;
2398 new_table_size
= refcount_table_clusters
<< (s
->cluster_bits
- 3);
2399 if (min_size
<= new_table_size
)
2403 printf("grow_refcount_table from %d to %d\n",
2404 s
->refcount_table_size
,
2407 new_table_size2
= new_table_size
* sizeof(uint64_t);
2408 new_table
= qemu_mallocz(new_table_size2
);
2409 memcpy(new_table
, s
->refcount_table
,
2410 s
->refcount_table_size
* sizeof(uint64_t));
2411 for(i
= 0; i
< s
->refcount_table_size
; i
++)
2412 cpu_to_be64s(&new_table
[i
]);
2413 /* Note: we cannot update the refcount now to avoid recursion */
2414 table_offset
= alloc_clusters_noref(bs
, new_table_size2
);
2415 ret
= bdrv_pwrite(s
->hd
, table_offset
, new_table
, new_table_size2
);
2416 if (ret
!= new_table_size2
)
2418 for(i
= 0; i
< s
->refcount_table_size
; i
++)
2419 be64_to_cpus(&new_table
[i
]);
2421 cpu_to_be64w((uint64_t*)data
, table_offset
);
2422 cpu_to_be32w((uint32_t*)(data
+ 8), refcount_table_clusters
);
2423 if (bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, refcount_table_offset
),
2424 data
, sizeof(data
)) != sizeof(data
))
2426 qemu_free(s
->refcount_table
);
2427 old_table_offset
= s
->refcount_table_offset
;
2428 old_table_size
= s
->refcount_table_size
;
2429 s
->refcount_table
= new_table
;
2430 s
->refcount_table_size
= new_table_size
;
2431 s
->refcount_table_offset
= table_offset
;
2433 update_refcount(bs
, table_offset
, new_table_size2
, 1);
2434 free_clusters(bs
, old_table_offset
, old_table_size
* sizeof(uint64_t));
2437 free_clusters(bs
, table_offset
, new_table_size2
);
2438 qemu_free(new_table
);
2442 /* addend must be 1 or -1 */
2443 /* XXX: cache several refcount block clusters ? */
2444 static int update_cluster_refcount(BlockDriverState
*bs
,
2445 int64_t cluster_index
,
2448 BDRVQcowState
*s
= bs
->opaque
;
2449 int64_t offset
, refcount_block_offset
;
2450 int ret
, refcount_table_index
, block_index
, refcount
;
2453 refcount_table_index
= cluster_index
>> (s
->cluster_bits
- REFCOUNT_SHIFT
);
2454 if (refcount_table_index
>= s
->refcount_table_size
) {
2457 ret
= grow_refcount_table(bs
, refcount_table_index
+ 1);
2461 refcount_block_offset
= s
->refcount_table
[refcount_table_index
];
2462 if (!refcount_block_offset
) {
2465 /* create a new refcount block */
2466 /* Note: we cannot update the refcount now to avoid recursion */
2467 offset
= alloc_clusters_noref(bs
, s
->cluster_size
);
2468 memset(s
->refcount_block_cache
, 0, s
->cluster_size
);
2469 ret
= bdrv_pwrite(s
->hd
, offset
, s
->refcount_block_cache
, s
->cluster_size
);
2470 if (ret
!= s
->cluster_size
)
2472 s
->refcount_table
[refcount_table_index
] = offset
;
2473 data64
= cpu_to_be64(offset
);
2474 ret
= bdrv_pwrite(s
->hd
, s
->refcount_table_offset
+
2475 refcount_table_index
* sizeof(uint64_t),
2476 &data64
, sizeof(data64
));
2477 if (ret
!= sizeof(data64
))
2480 refcount_block_offset
= offset
;
2481 s
->refcount_block_cache_offset
= offset
;
2482 update_refcount(bs
, offset
, s
->cluster_size
, 1);
2484 if (refcount_block_offset
!= s
->refcount_block_cache_offset
) {
2485 if (load_refcount_block(bs
, refcount_block_offset
) < 0)
2489 /* we can update the count and save it */
2490 block_index
= cluster_index
&
2491 ((1 << (s
->cluster_bits
- REFCOUNT_SHIFT
)) - 1);
2492 refcount
= be16_to_cpu(s
->refcount_block_cache
[block_index
]);
2494 if (refcount
< 0 || refcount
> 0xffff)
2496 if (refcount
== 0 && cluster_index
< s
->free_cluster_index
) {
2497 s
->free_cluster_index
= cluster_index
;
2499 s
->refcount_block_cache
[block_index
] = cpu_to_be16(refcount
);
2500 if (bdrv_pwrite(s
->hd
,
2501 refcount_block_offset
+ (block_index
<< REFCOUNT_SHIFT
),
2502 &s
->refcount_block_cache
[block_index
], 2) != 2)
2507 static void update_refcount(BlockDriverState
*bs
,
2508 int64_t offset
, int64_t length
,
2511 BDRVQcowState
*s
= bs
->opaque
;
2512 int64_t start
, last
, cluster_offset
;
2515 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2516 offset
, length
, addend
);
2520 start
= offset
& ~(s
->cluster_size
- 1);
2521 last
= (offset
+ length
- 1) & ~(s
->cluster_size
- 1);
2522 for(cluster_offset
= start
; cluster_offset
<= last
;
2523 cluster_offset
+= s
->cluster_size
) {
2524 update_cluster_refcount(bs
, cluster_offset
>> s
->cluster_bits
, addend
);
2529 static void inc_refcounts(BlockDriverState
*bs
,
2530 uint16_t *refcount_table
,
2531 int refcount_table_size
,
2532 int64_t offset
, int64_t size
)
2534 BDRVQcowState
*s
= bs
->opaque
;
2535 int64_t start
, last
, cluster_offset
;
2541 start
= offset
& ~(s
->cluster_size
- 1);
2542 last
= (offset
+ size
- 1) & ~(s
->cluster_size
- 1);
2543 for(cluster_offset
= start
; cluster_offset
<= last
;
2544 cluster_offset
+= s
->cluster_size
) {
2545 k
= cluster_offset
>> s
->cluster_bits
;
2546 if (k
< 0 || k
>= refcount_table_size
) {
2547 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset
);
2549 if (++refcount_table
[k
] == 0) {
2550 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset
);
2556 static int check_refcounts_l1(BlockDriverState
*bs
,
2557 uint16_t *refcount_table
,
2558 int refcount_table_size
,
2559 int64_t l1_table_offset
, int l1_size
,
2562 BDRVQcowState
*s
= bs
->opaque
;
2563 uint64_t *l1_table
, *l2_table
, l2_offset
, offset
, l1_size2
;
2564 int l2_size
, i
, j
, nb_csectors
, refcount
;
2567 l1_size2
= l1_size
* sizeof(uint64_t);
2569 inc_refcounts(bs
, refcount_table
, refcount_table_size
,
2570 l1_table_offset
, l1_size2
);
2572 l1_table
= qemu_malloc(l1_size2
);
2573 if (bdrv_pread(s
->hd
, l1_table_offset
,
2574 l1_table
, l1_size2
) != l1_size2
)
2576 for(i
= 0;i
< l1_size
; i
++)
2577 be64_to_cpus(&l1_table
[i
]);
2579 l2_size
= s
->l2_size
* sizeof(uint64_t);
2580 l2_table
= qemu_malloc(l2_size
);
2581 for(i
= 0; i
< l1_size
; i
++) {
2582 l2_offset
= l1_table
[i
];
2585 refcount
= get_refcount(bs
, (l2_offset
& ~QCOW_OFLAG_COPIED
) >> s
->cluster_bits
);
2586 if ((refcount
== 1) != ((l2_offset
& QCOW_OFLAG_COPIED
) != 0)) {
2587 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2588 l2_offset
, refcount
);
2591 l2_offset
&= ~QCOW_OFLAG_COPIED
;
2592 if (bdrv_pread(s
->hd
, l2_offset
, l2_table
, l2_size
) != l2_size
)
2594 for(j
= 0; j
< s
->l2_size
; j
++) {
2595 offset
= be64_to_cpu(l2_table
[j
]);
2597 if (offset
& QCOW_OFLAG_COMPRESSED
) {
2598 if (offset
& QCOW_OFLAG_COPIED
) {
2599 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2600 offset
>> s
->cluster_bits
);
2601 offset
&= ~QCOW_OFLAG_COPIED
;
2603 nb_csectors
= ((offset
>> s
->csize_shift
) &
2605 offset
&= s
->cluster_offset_mask
;
2606 inc_refcounts(bs
, refcount_table
,
2607 refcount_table_size
,
2608 offset
& ~511, nb_csectors
* 512);
2611 refcount
= get_refcount(bs
, (offset
& ~QCOW_OFLAG_COPIED
) >> s
->cluster_bits
);
2612 if ((refcount
== 1) != ((offset
& QCOW_OFLAG_COPIED
) != 0)) {
2613 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2617 offset
&= ~QCOW_OFLAG_COPIED
;
2618 inc_refcounts(bs
, refcount_table
,
2619 refcount_table_size
,
2620 offset
, s
->cluster_size
);
2624 inc_refcounts(bs
, refcount_table
,
2625 refcount_table_size
,
2630 qemu_free(l1_table
);
2631 qemu_free(l2_table
);
2634 printf("ERROR: I/O error in check_refcounts_l1\n");
2635 qemu_free(l1_table
);
2636 qemu_free(l2_table
);
2640 static void check_refcounts(BlockDriverState
*bs
)
2642 BDRVQcowState
*s
= bs
->opaque
;
2644 int nb_clusters
, refcount1
, refcount2
, i
;
2646 uint16_t *refcount_table
;
2648 size
= bdrv_getlength(s
->hd
);
2649 nb_clusters
= size_to_clusters(s
, size
);
2650 refcount_table
= qemu_mallocz(nb_clusters
* sizeof(uint16_t));
2653 inc_refcounts(bs
, refcount_table
, nb_clusters
,
2654 0, s
->cluster_size
);
2656 check_refcounts_l1(bs
, refcount_table
, nb_clusters
,
2657 s
->l1_table_offset
, s
->l1_size
, 1);
2660 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
2661 sn
= s
->snapshots
+ i
;
2662 check_refcounts_l1(bs
, refcount_table
, nb_clusters
,
2663 sn
->l1_table_offset
, sn
->l1_size
, 0);
2665 inc_refcounts(bs
, refcount_table
, nb_clusters
,
2666 s
->snapshots_offset
, s
->snapshots_size
);
2669 inc_refcounts(bs
, refcount_table
, nb_clusters
,
2670 s
->refcount_table_offset
,
2671 s
->refcount_table_size
* sizeof(uint64_t));
2672 for(i
= 0; i
< s
->refcount_table_size
; i
++) {
2674 offset
= s
->refcount_table
[i
];
2676 inc_refcounts(bs
, refcount_table
, nb_clusters
,
2677 offset
, s
->cluster_size
);
2681 /* compare ref counts */
2682 for(i
= 0; i
< nb_clusters
; i
++) {
2683 refcount1
= get_refcount(bs
, i
);
2684 refcount2
= refcount_table
[i
];
2685 if (refcount1
!= refcount2
)
2686 printf("ERROR cluster %d refcount=%d reference=%d\n",
2687 i
, refcount1
, refcount2
);
2690 qemu_free(refcount_table
);
2694 static void dump_refcounts(BlockDriverState
*bs
)
2696 BDRVQcowState
*s
= bs
->opaque
;
2697 int64_t nb_clusters
, k
, k1
, size
;
2700 size
= bdrv_getlength(s
->hd
);
2701 nb_clusters
= size_to_clusters(s
, size
);
2702 for(k
= 0; k
< nb_clusters
;) {
2704 refcount
= get_refcount(bs
, k
);
2706 while (k
< nb_clusters
&& get_refcount(bs
, k
) == refcount
)
2708 printf("%lld: refcount=%d nb=%lld\n", k
, refcount
, k
- k1
);
2714 BlockDriver bdrv_qcow2
= {
2715 .format_name
= "qcow2",
2716 .instance_size
= sizeof(BDRVQcowState
),
2717 .bdrv_probe
= qcow_probe
,
2718 .bdrv_open
= qcow_open
,
2719 .bdrv_close
= qcow_close
,
2720 .bdrv_create
= qcow_create
,
2721 .bdrv_flush
= qcow_flush
,
2722 .bdrv_is_allocated
= qcow_is_allocated
,
2723 .bdrv_set_key
= qcow_set_key
,
2724 .bdrv_make_empty
= qcow_make_empty
,
2726 .bdrv_aio_read
= qcow_aio_read
,
2727 .bdrv_aio_write
= qcow_aio_write
,
2728 .bdrv_aio_cancel
= qcow_aio_cancel
,
2729 .aiocb_size
= sizeof(QCowAIOCB
),
2730 .bdrv_write_compressed
= qcow_write_compressed
,
2732 .bdrv_snapshot_create
= qcow_snapshot_create
,
2733 .bdrv_snapshot_goto
= qcow_snapshot_goto
,
2734 .bdrv_snapshot_delete
= qcow_snapshot_delete
,
2735 .bdrv_snapshot_list
= qcow_snapshot_list
,
2736 .bdrv_get_info
= qcow_get_info
,
2738 .bdrv_create2
= qcow_create2
,