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
27 #include "qemu-common.h"
28 #include "block_int.h"
29 #include "block/qcow2.h"
31 int qcow2_grow_l1_table(BlockDriverState
*bs
, int min_size
)
33 BDRVQcowState
*s
= bs
->opaque
;
34 int new_l1_size
, new_l1_size2
, ret
, i
;
35 uint64_t *new_l1_table
;
36 int64_t new_l1_table_offset
;
39 new_l1_size
= s
->l1_size
;
40 if (min_size
<= new_l1_size
)
42 if (new_l1_size
== 0) {
45 while (min_size
> new_l1_size
) {
46 new_l1_size
= (new_l1_size
* 3 + 1) / 2;
49 printf("grow l1_table from %d to %d\n", s
->l1_size
, new_l1_size
);
52 new_l1_size2
= sizeof(uint64_t) * new_l1_size
;
53 new_l1_table
= qemu_mallocz(align_offset(new_l1_size2
, 512));
54 memcpy(new_l1_table
, s
->l1_table
, s
->l1_size
* sizeof(uint64_t));
56 /* write new table (align to cluster) */
57 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_GROW_ALLOC_TABLE
);
58 new_l1_table_offset
= qcow2_alloc_clusters(bs
, new_l1_size2
);
59 if (new_l1_table_offset
< 0) {
60 qemu_free(new_l1_table
);
61 return new_l1_table_offset
;
65 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_GROW_WRITE_TABLE
);
66 for(i
= 0; i
< s
->l1_size
; i
++)
67 new_l1_table
[i
] = cpu_to_be64(new_l1_table
[i
]);
68 ret
= bdrv_pwrite_sync(bs
->file
, new_l1_table_offset
, new_l1_table
, new_l1_size2
);
71 for(i
= 0; i
< s
->l1_size
; i
++)
72 new_l1_table
[i
] = be64_to_cpu(new_l1_table
[i
]);
75 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_GROW_ACTIVATE_TABLE
);
76 cpu_to_be32w((uint32_t*)data
, new_l1_size
);
77 cpu_to_be64w((uint64_t*)(data
+ 4), new_l1_table_offset
);
78 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_size
), data
,sizeof(data
));
82 qemu_free(s
->l1_table
);
83 qcow2_free_clusters(bs
, s
->l1_table_offset
, s
->l1_size
* sizeof(uint64_t));
84 s
->l1_table_offset
= new_l1_table_offset
;
85 s
->l1_table
= new_l1_table
;
86 s
->l1_size
= new_l1_size
;
89 qemu_free(new_l1_table
);
90 qcow2_free_clusters(bs
, new_l1_table_offset
, new_l1_size2
);
94 void qcow2_l2_cache_reset(BlockDriverState
*bs
)
96 BDRVQcowState
*s
= bs
->opaque
;
98 memset(s
->l2_cache
, 0, s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
99 memset(s
->l2_cache_offsets
, 0, L2_CACHE_SIZE
* sizeof(uint64_t));
100 memset(s
->l2_cache_counts
, 0, L2_CACHE_SIZE
* sizeof(uint32_t));
103 static inline int l2_cache_new_entry(BlockDriverState
*bs
)
105 BDRVQcowState
*s
= bs
->opaque
;
109 /* find a new entry in the least used one */
111 min_count
= 0xffffffff;
112 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
113 if (s
->l2_cache_counts
[i
] < min_count
) {
114 min_count
= s
->l2_cache_counts
[i
];
124 * seek l2_offset in the l2_cache table
125 * if not found, return NULL,
127 * increments the l2 cache hit count of the entry,
128 * if counter overflow, divide by two all counters
129 * return the pointer to the l2 cache entry
133 static uint64_t *seek_l2_table(BDRVQcowState
*s
, uint64_t l2_offset
)
137 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
138 if (l2_offset
== s
->l2_cache_offsets
[i
]) {
139 /* increment the hit count */
140 if (++s
->l2_cache_counts
[i
] == 0xffffffff) {
141 for(j
= 0; j
< L2_CACHE_SIZE
; j
++) {
142 s
->l2_cache_counts
[j
] >>= 1;
145 return s
->l2_cache
+ (i
<< s
->l2_bits
);
154 * Loads a L2 table into memory. If the table is in the cache, the cache
155 * is used; otherwise the L2 table is loaded from the image file.
157 * Returns a pointer to the L2 table on success, or NULL if the read from
158 * the image file failed.
161 static int l2_load(BlockDriverState
*bs
, uint64_t l2_offset
,
164 BDRVQcowState
*s
= bs
->opaque
;
168 /* seek if the table for the given offset is in the cache */
170 *l2_table
= seek_l2_table(s
, l2_offset
);
171 if (*l2_table
!= NULL
) {
175 /* not found: load a new entry in the least used one */
177 min_index
= l2_cache_new_entry(bs
);
178 *l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
180 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_LOAD
);
181 ret
= bdrv_pread(bs
->file
, l2_offset
, *l2_table
,
182 s
->l2_size
* sizeof(uint64_t));
187 s
->l2_cache_offsets
[min_index
] = l2_offset
;
188 s
->l2_cache_counts
[min_index
] = 1;
194 * Writes one sector of the L1 table to the disk (can't update single entries
195 * and we really don't want bdrv_pread to perform a read-modify-write)
197 #define L1_ENTRIES_PER_SECTOR (512 / 8)
198 static int write_l1_entry(BlockDriverState
*bs
, int l1_index
)
200 BDRVQcowState
*s
= bs
->opaque
;
201 uint64_t buf
[L1_ENTRIES_PER_SECTOR
];
205 l1_start_index
= l1_index
& ~(L1_ENTRIES_PER_SECTOR
- 1);
206 for (i
= 0; i
< L1_ENTRIES_PER_SECTOR
; i
++) {
207 buf
[i
] = cpu_to_be64(s
->l1_table
[l1_start_index
+ i
]);
210 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
211 ret
= bdrv_pwrite_sync(bs
->file
, s
->l1_table_offset
+ 8 * l1_start_index
,
223 * Allocate a new l2 entry in the file. If l1_index points to an already
224 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
225 * table) copy the contents of the old L2 table into the newly allocated one.
226 * Otherwise the new table is initialized with zeros.
230 static int l2_allocate(BlockDriverState
*bs
, int l1_index
, uint64_t **table
)
232 BDRVQcowState
*s
= bs
->opaque
;
234 uint64_t old_l2_offset
;
239 old_l2_offset
= s
->l1_table
[l1_index
];
241 /* allocate a new l2 entry */
243 l2_offset
= qcow2_alloc_clusters(bs
, s
->l2_size
* sizeof(uint64_t));
247 bdrv_flush(bs
->file
);
249 /* allocate a new entry in the l2 cache */
251 min_index
= l2_cache_new_entry(bs
);
252 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
254 if (old_l2_offset
== 0) {
255 /* if there was no old l2 table, clear the new table */
256 memset(l2_table
, 0, s
->l2_size
* sizeof(uint64_t));
258 /* if there was an old l2 table, read it from the disk */
259 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_ALLOC_COW_READ
);
260 ret
= bdrv_pread(bs
->file
, old_l2_offset
, l2_table
,
261 s
->l2_size
* sizeof(uint64_t));
266 /* write the l2 table to the file */
267 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_ALLOC_WRITE
);
268 ret
= bdrv_pwrite_sync(bs
->file
, l2_offset
, l2_table
,
269 s
->l2_size
* sizeof(uint64_t));
274 /* update the L1 entry */
275 s
->l1_table
[l1_index
] = l2_offset
| QCOW_OFLAG_COPIED
;
276 ret
= write_l1_entry(bs
, l1_index
);
281 /* update the l2 cache entry */
283 s
->l2_cache_offsets
[min_index
] = l2_offset
;
284 s
->l2_cache_counts
[min_index
] = 1;
290 s
->l1_table
[l1_index
] = old_l2_offset
;
291 qcow2_l2_cache_reset(bs
);
295 static int count_contiguous_clusters(uint64_t nb_clusters
, int cluster_size
,
296 uint64_t *l2_table
, uint64_t start
, uint64_t mask
)
299 uint64_t offset
= be64_to_cpu(l2_table
[0]) & ~mask
;
304 for (i
= start
; i
< start
+ nb_clusters
; i
++)
305 if (offset
+ (uint64_t) i
* cluster_size
!= (be64_to_cpu(l2_table
[i
]) & ~mask
))
311 static int count_contiguous_free_clusters(uint64_t nb_clusters
, uint64_t *l2_table
)
315 while(nb_clusters
-- && l2_table
[i
] == 0)
321 /* The crypt function is compatible with the linux cryptoloop
322 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
324 void qcow2_encrypt_sectors(BDRVQcowState
*s
, int64_t sector_num
,
325 uint8_t *out_buf
, const uint8_t *in_buf
,
326 int nb_sectors
, int enc
,
335 for(i
= 0; i
< nb_sectors
; i
++) {
336 ivec
.ll
[0] = cpu_to_le64(sector_num
);
338 AES_cbc_encrypt(in_buf
, out_buf
, 512, key
,
347 static int qcow_read(BlockDriverState
*bs
, int64_t sector_num
,
348 uint8_t *buf
, int nb_sectors
)
350 BDRVQcowState
*s
= bs
->opaque
;
351 int ret
, index_in_cluster
, n
, n1
;
352 uint64_t cluster_offset
;
354 while (nb_sectors
> 0) {
357 ret
= qcow2_get_cluster_offset(bs
, sector_num
<< 9, &n
,
363 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
364 if (!cluster_offset
) {
365 if (bs
->backing_hd
) {
366 /* read from the base image */
367 n1
= qcow2_backing_read1(bs
->backing_hd
, sector_num
, buf
, n
);
369 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING
);
370 ret
= bdrv_read(bs
->backing_hd
, sector_num
, buf
, n1
);
375 memset(buf
, 0, 512 * n
);
377 } else if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
378 if (qcow2_decompress_cluster(bs
, cluster_offset
) < 0)
380 memcpy(buf
, s
->cluster_cache
+ index_in_cluster
* 512, 512 * n
);
382 BLKDBG_EVENT(bs
->file
, BLKDBG_READ
);
383 ret
= bdrv_pread(bs
->file
, cluster_offset
+ index_in_cluster
* 512, buf
, n
* 512);
386 if (s
->crypt_method
) {
387 qcow2_encrypt_sectors(s
, sector_num
, buf
, buf
, n
, 0,
388 &s
->aes_decrypt_key
);
398 static int copy_sectors(BlockDriverState
*bs
, uint64_t start_sect
,
399 uint64_t cluster_offset
, int n_start
, int n_end
)
401 BDRVQcowState
*s
= bs
->opaque
;
407 BLKDBG_EVENT(bs
->file
, BLKDBG_COW_READ
);
408 ret
= qcow_read(bs
, start_sect
+ n_start
, s
->cluster_data
, n
);
411 if (s
->crypt_method
) {
412 qcow2_encrypt_sectors(s
, start_sect
+ n_start
,
414 s
->cluster_data
, n
, 1,
415 &s
->aes_encrypt_key
);
417 BLKDBG_EVENT(bs
->file
, BLKDBG_COW_WRITE
);
418 ret
= bdrv_write(bs
->file
, (cluster_offset
>> 9) + n_start
,
429 * For a given offset of the disk image, find the cluster offset in
430 * qcow2 file. The offset is stored in *cluster_offset.
432 * on entry, *num is the number of contiguous clusters we'd like to
433 * access following offset.
435 * on exit, *num is the number of contiguous clusters we can read.
437 * Return 0, if the offset is found
438 * Return -errno, otherwise.
442 int qcow2_get_cluster_offset(BlockDriverState
*bs
, uint64_t offset
,
443 int *num
, uint64_t *cluster_offset
)
445 BDRVQcowState
*s
= bs
->opaque
;
446 unsigned int l1_index
, l2_index
;
447 uint64_t l2_offset
, *l2_table
;
449 unsigned int index_in_cluster
, nb_clusters
;
450 uint64_t nb_available
, nb_needed
;
453 index_in_cluster
= (offset
>> 9) & (s
->cluster_sectors
- 1);
454 nb_needed
= *num
+ index_in_cluster
;
456 l1_bits
= s
->l2_bits
+ s
->cluster_bits
;
458 /* compute how many bytes there are between the offset and
459 * the end of the l1 entry
462 nb_available
= (1ULL << l1_bits
) - (offset
& ((1ULL << l1_bits
) - 1));
464 /* compute the number of available sectors */
466 nb_available
= (nb_available
>> 9) + index_in_cluster
;
468 if (nb_needed
> nb_available
) {
469 nb_needed
= nb_available
;
474 /* seek the the l2 offset in the l1 table */
476 l1_index
= offset
>> l1_bits
;
477 if (l1_index
>= s
->l1_size
)
480 l2_offset
= s
->l1_table
[l1_index
];
482 /* seek the l2 table of the given l2 offset */
487 /* load the l2 table in memory */
489 l2_offset
&= ~QCOW_OFLAG_COPIED
;
490 ret
= l2_load(bs
, l2_offset
, &l2_table
);
495 /* find the cluster offset for the given disk offset */
497 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
498 *cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
499 nb_clusters
= size_to_clusters(s
, nb_needed
<< 9);
501 if (!*cluster_offset
) {
502 /* how many empty clusters ? */
503 c
= count_contiguous_free_clusters(nb_clusters
, &l2_table
[l2_index
]);
505 /* how many allocated clusters ? */
506 c
= count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
507 &l2_table
[l2_index
], 0, QCOW_OFLAG_COPIED
);
510 nb_available
= (c
* s
->cluster_sectors
);
512 if (nb_available
> nb_needed
)
513 nb_available
= nb_needed
;
515 *num
= nb_available
- index_in_cluster
;
517 *cluster_offset
&=~QCOW_OFLAG_COPIED
;
524 * for a given disk offset, load (and allocate if needed)
527 * the l2 table offset in the qcow2 file and the cluster index
528 * in the l2 table are given to the caller.
530 * Returns 0 on success, -errno in failure case
532 static int get_cluster_table(BlockDriverState
*bs
, uint64_t offset
,
533 uint64_t **new_l2_table
,
534 uint64_t *new_l2_offset
,
537 BDRVQcowState
*s
= bs
->opaque
;
538 unsigned int l1_index
, l2_index
;
540 uint64_t *l2_table
= NULL
;
543 /* seek the the l2 offset in the l1 table */
545 l1_index
= offset
>> (s
->l2_bits
+ s
->cluster_bits
);
546 if (l1_index
>= s
->l1_size
) {
547 ret
= qcow2_grow_l1_table(bs
, l1_index
+ 1);
552 l2_offset
= s
->l1_table
[l1_index
];
554 /* seek the l2 table of the given l2 offset */
556 if (l2_offset
& QCOW_OFLAG_COPIED
) {
557 /* load the l2 table in memory */
558 l2_offset
&= ~QCOW_OFLAG_COPIED
;
559 ret
= l2_load(bs
, l2_offset
, &l2_table
);
565 qcow2_free_clusters(bs
, l2_offset
, s
->l2_size
* sizeof(uint64_t));
566 ret
= l2_allocate(bs
, l1_index
, &l2_table
);
570 l2_offset
= s
->l1_table
[l1_index
] & ~QCOW_OFLAG_COPIED
;
573 /* find the cluster offset for the given disk offset */
575 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
577 *new_l2_table
= l2_table
;
578 *new_l2_offset
= l2_offset
;
579 *new_l2_index
= l2_index
;
585 * alloc_compressed_cluster_offset
587 * For a given offset of the disk image, return cluster offset in
590 * If the offset is not found, allocate a new compressed cluster.
592 * Return the cluster offset if successful,
593 * Return 0, otherwise.
597 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState
*bs
,
601 BDRVQcowState
*s
= bs
->opaque
;
603 uint64_t l2_offset
, *l2_table
;
604 int64_t cluster_offset
;
607 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_offset
, &l2_index
);
612 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
613 if (cluster_offset
& QCOW_OFLAG_COPIED
)
614 return cluster_offset
& ~QCOW_OFLAG_COPIED
;
617 qcow2_free_any_clusters(bs
, cluster_offset
, 1);
619 cluster_offset
= qcow2_alloc_bytes(bs
, compressed_size
);
620 if (cluster_offset
< 0) {
624 nb_csectors
= ((cluster_offset
+ compressed_size
- 1) >> 9) -
625 (cluster_offset
>> 9);
627 cluster_offset
|= QCOW_OFLAG_COMPRESSED
|
628 ((uint64_t)nb_csectors
<< s
->csize_shift
);
630 /* update L2 table */
632 /* compressed clusters never have the copied flag */
634 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_UPDATE_COMPRESSED
);
635 l2_table
[l2_index
] = cpu_to_be64(cluster_offset
);
636 if (bdrv_pwrite_sync(bs
->file
,
637 l2_offset
+ l2_index
* sizeof(uint64_t),
639 sizeof(uint64_t)) < 0)
642 return cluster_offset
;
646 * Write L2 table updates to disk, writing whole sectors to avoid a
647 * read-modify-write in bdrv_pwrite
649 #define L2_ENTRIES_PER_SECTOR (512 / 8)
650 static int write_l2_entries(BlockDriverState
*bs
, uint64_t *l2_table
,
651 uint64_t l2_offset
, int l2_index
, int num
)
653 int l2_start_index
= l2_index
& ~(L1_ENTRIES_PER_SECTOR
- 1);
654 int start_offset
= (8 * l2_index
) & ~511;
655 int end_offset
= (8 * (l2_index
+ num
) + 511) & ~511;
656 size_t len
= end_offset
- start_offset
;
659 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_UPDATE
);
660 ret
= bdrv_pwrite(bs
->file
, l2_offset
+ start_offset
,
661 &l2_table
[l2_start_index
], len
);
669 int qcow2_alloc_cluster_link_l2(BlockDriverState
*bs
, QCowL2Meta
*m
)
671 BDRVQcowState
*s
= bs
->opaque
;
672 int i
, j
= 0, l2_index
, ret
;
673 uint64_t *old_cluster
, start_sect
, l2_offset
, *l2_table
;
674 uint64_t cluster_offset
= m
->cluster_offset
;
676 if (m
->nb_clusters
== 0)
679 old_cluster
= qemu_malloc(m
->nb_clusters
* sizeof(uint64_t));
681 /* copy content of unmodified sectors */
682 start_sect
= (m
->offset
& ~(s
->cluster_size
- 1)) >> 9;
684 ret
= copy_sectors(bs
, start_sect
, cluster_offset
, 0, m
->n_start
);
689 if (m
->nb_available
& (s
->cluster_sectors
- 1)) {
690 uint64_t end
= m
->nb_available
& ~(uint64_t)(s
->cluster_sectors
- 1);
691 ret
= copy_sectors(bs
, start_sect
+ end
, cluster_offset
+ (end
<< 9),
692 m
->nb_available
- end
, s
->cluster_sectors
);
697 /* update L2 table */
698 ret
= get_cluster_table(bs
, m
->offset
, &l2_table
, &l2_offset
, &l2_index
);
703 for (i
= 0; i
< m
->nb_clusters
; i
++) {
704 /* if two concurrent writes happen to the same unallocated cluster
705 * each write allocates separate cluster and writes data concurrently.
706 * The first one to complete updates l2 table with pointer to its
707 * cluster the second one has to do RMW (which is done above by
708 * copy_sectors()), update l2 table with its cluster pointer and free
709 * old cluster. This is what this loop does */
710 if(l2_table
[l2_index
+ i
] != 0)
711 old_cluster
[j
++] = l2_table
[l2_index
+ i
];
713 l2_table
[l2_index
+ i
] = cpu_to_be64((cluster_offset
+
714 (i
<< s
->cluster_bits
)) | QCOW_OFLAG_COPIED
);
718 * Before we update the L2 table to actually point to the new cluster, we
719 * need to be sure that the refcounts have been increased and COW was
722 bdrv_flush(bs
->file
);
724 ret
= write_l2_entries(bs
, l2_table
, l2_offset
, l2_index
, m
->nb_clusters
);
726 qcow2_l2_cache_reset(bs
);
731 * If this was a COW, we need to decrease the refcount of the old cluster.
732 * Also flush bs->file to get the right order for L2 and refcount update.
735 bdrv_flush(bs
->file
);
736 for (i
= 0; i
< j
; i
++) {
737 qcow2_free_any_clusters(bs
,
738 be64_to_cpu(old_cluster
[i
]) & ~QCOW_OFLAG_COPIED
, 1);
744 qemu_free(old_cluster
);
749 * alloc_cluster_offset
751 * For a given offset of the disk image, return cluster offset in qcow2 file.
752 * If the offset is not found, allocate a new cluster.
754 * If the cluster was already allocated, m->nb_clusters is set to 0,
755 * m->depends_on is set to NULL and the other fields in m are meaningless.
757 * If the cluster is newly allocated, m->nb_clusters is set to the number of
758 * contiguous clusters that have been allocated. This may be 0 if the request
759 * conflict with another write request in flight; in this case, m->depends_on
760 * is set and the remaining fields of m are meaningless.
762 * If m->nb_clusters is non-zero, the other fields of m are valid and contain
763 * information about the first allocated cluster.
765 * Return 0 on success and -errno in error cases
767 int qcow2_alloc_cluster_offset(BlockDriverState
*bs
, uint64_t offset
,
768 int n_start
, int n_end
, int *num
, QCowL2Meta
*m
)
770 BDRVQcowState
*s
= bs
->opaque
;
772 uint64_t l2_offset
, *l2_table
;
773 int64_t cluster_offset
;
774 unsigned int nb_clusters
, i
= 0;
775 QCowL2Meta
*old_alloc
;
777 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_offset
, &l2_index
);
782 nb_clusters
= size_to_clusters(s
, n_end
<< 9);
784 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
786 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
788 /* We keep all QCOW_OFLAG_COPIED clusters */
790 if (cluster_offset
& QCOW_OFLAG_COPIED
) {
791 nb_clusters
= count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
792 &l2_table
[l2_index
], 0, 0);
794 cluster_offset
&= ~QCOW_OFLAG_COPIED
;
796 m
->depends_on
= NULL
;
801 /* for the moment, multiple compressed clusters are not managed */
803 if (cluster_offset
& QCOW_OFLAG_COMPRESSED
)
806 /* how many available clusters ? */
808 while (i
< nb_clusters
) {
809 i
+= count_contiguous_clusters(nb_clusters
- i
, s
->cluster_size
,
810 &l2_table
[l2_index
], i
, 0);
811 if ((i
>= nb_clusters
) || be64_to_cpu(l2_table
[l2_index
+ i
])) {
815 i
+= count_contiguous_free_clusters(nb_clusters
- i
,
816 &l2_table
[l2_index
+ i
]);
817 if (i
>= nb_clusters
) {
821 cluster_offset
= be64_to_cpu(l2_table
[l2_index
+ i
]);
823 if ((cluster_offset
& QCOW_OFLAG_COPIED
) ||
824 (cluster_offset
& QCOW_OFLAG_COMPRESSED
))
827 assert(i
<= nb_clusters
);
831 * Check if there already is an AIO write request in flight which allocates
832 * the same cluster. In this case we need to wait until the previous
833 * request has completed and updated the L2 table accordingly.
835 QLIST_FOREACH(old_alloc
, &s
->cluster_allocs
, next_in_flight
) {
837 uint64_t end_offset
= offset
+ nb_clusters
* s
->cluster_size
;
838 uint64_t old_offset
= old_alloc
->offset
;
839 uint64_t old_end_offset
= old_alloc
->offset
+
840 old_alloc
->nb_clusters
* s
->cluster_size
;
842 if (end_offset
< old_offset
|| offset
> old_end_offset
) {
843 /* No intersection */
845 if (offset
< old_offset
) {
846 /* Stop at the start of a running allocation */
847 nb_clusters
= (old_offset
- offset
) >> s
->cluster_bits
;
852 if (nb_clusters
== 0) {
853 /* Set dependency and wait for a callback */
854 m
->depends_on
= old_alloc
;
866 QLIST_INSERT_HEAD(&s
->cluster_allocs
, m
, next_in_flight
);
868 /* allocate a new cluster */
870 cluster_offset
= qcow2_alloc_clusters(bs
, nb_clusters
* s
->cluster_size
);
871 if (cluster_offset
< 0) {
872 QLIST_REMOVE(m
, next_in_flight
);
873 return cluster_offset
;
876 /* save info needed for meta data update */
878 m
->n_start
= n_start
;
879 m
->nb_clusters
= nb_clusters
;
882 m
->nb_available
= MIN(nb_clusters
<< (s
->cluster_bits
- 9), n_end
);
883 m
->cluster_offset
= cluster_offset
;
885 *num
= m
->nb_available
- n_start
;
890 static int decompress_buffer(uint8_t *out_buf
, int out_buf_size
,
891 const uint8_t *buf
, int buf_size
)
893 z_stream strm1
, *strm
= &strm1
;
896 memset(strm
, 0, sizeof(*strm
));
898 strm
->next_in
= (uint8_t *)buf
;
899 strm
->avail_in
= buf_size
;
900 strm
->next_out
= out_buf
;
901 strm
->avail_out
= out_buf_size
;
903 ret
= inflateInit2(strm
, -12);
906 ret
= inflate(strm
, Z_FINISH
);
907 out_len
= strm
->next_out
- out_buf
;
908 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) ||
909 out_len
!= out_buf_size
) {
917 int qcow2_decompress_cluster(BlockDriverState
*bs
, uint64_t cluster_offset
)
919 BDRVQcowState
*s
= bs
->opaque
;
920 int ret
, csize
, nb_csectors
, sector_offset
;
923 coffset
= cluster_offset
& s
->cluster_offset_mask
;
924 if (s
->cluster_cache_offset
!= coffset
) {
925 nb_csectors
= ((cluster_offset
>> s
->csize_shift
) & s
->csize_mask
) + 1;
926 sector_offset
= coffset
& 511;
927 csize
= nb_csectors
* 512 - sector_offset
;
928 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
929 ret
= bdrv_read(bs
->file
, coffset
>> 9, s
->cluster_data
, nb_csectors
);
933 if (decompress_buffer(s
->cluster_cache
, s
->cluster_size
,
934 s
->cluster_data
+ sector_offset
, csize
) < 0) {
937 s
->cluster_cache_offset
= coffset
;