]>
git.proxmox.com Git - qemu.git/blob - block/qcow2-cluster.c
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 new_l1_table_offset
= qcow2_alloc_clusters(bs
, new_l1_size2
);
58 if (new_l1_table_offset
< 0) {
59 qemu_free(new_l1_table
);
60 return new_l1_table_offset
;
63 for(i
= 0; i
< s
->l1_size
; i
++)
64 new_l1_table
[i
] = cpu_to_be64(new_l1_table
[i
]);
65 ret
= bdrv_pwrite(s
->hd
, new_l1_table_offset
, new_l1_table
, new_l1_size2
);
66 if (ret
!= new_l1_size2
)
68 for(i
= 0; i
< s
->l1_size
; i
++)
69 new_l1_table
[i
] = be64_to_cpu(new_l1_table
[i
]);
72 cpu_to_be32w((uint32_t*)data
, new_l1_size
);
73 cpu_to_be64w((uint64_t*)(data
+ 4), new_l1_table_offset
);
74 ret
= bdrv_pwrite(s
->hd
, offsetof(QCowHeader
, l1_size
), data
,sizeof(data
));
75 if (ret
!= sizeof(data
)) {
78 qemu_free(s
->l1_table
);
79 qcow2_free_clusters(bs
, s
->l1_table_offset
, s
->l1_size
* sizeof(uint64_t));
80 s
->l1_table_offset
= new_l1_table_offset
;
81 s
->l1_table
= new_l1_table
;
82 s
->l1_size
= new_l1_size
;
85 qemu_free(new_l1_table
);
86 qcow2_free_clusters(bs
, new_l1_table_offset
, new_l1_size2
);
87 return ret
< 0 ? ret
: -EIO
;
90 void qcow2_l2_cache_reset(BlockDriverState
*bs
)
92 BDRVQcowState
*s
= bs
->opaque
;
94 memset(s
->l2_cache
, 0, s
->l2_size
* L2_CACHE_SIZE
* sizeof(uint64_t));
95 memset(s
->l2_cache_offsets
, 0, L2_CACHE_SIZE
* sizeof(uint64_t));
96 memset(s
->l2_cache_counts
, 0, L2_CACHE_SIZE
* sizeof(uint32_t));
99 static inline int l2_cache_new_entry(BlockDriverState
*bs
)
101 BDRVQcowState
*s
= bs
->opaque
;
105 /* find a new entry in the least used one */
107 min_count
= 0xffffffff;
108 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
109 if (s
->l2_cache_counts
[i
] < min_count
) {
110 min_count
= s
->l2_cache_counts
[i
];
120 * seek l2_offset in the l2_cache table
121 * if not found, return NULL,
123 * increments the l2 cache hit count of the entry,
124 * if counter overflow, divide by two all counters
125 * return the pointer to the l2 cache entry
129 static uint64_t *seek_l2_table(BDRVQcowState
*s
, uint64_t l2_offset
)
133 for(i
= 0; i
< L2_CACHE_SIZE
; i
++) {
134 if (l2_offset
== s
->l2_cache_offsets
[i
]) {
135 /* increment the hit count */
136 if (++s
->l2_cache_counts
[i
] == 0xffffffff) {
137 for(j
= 0; j
< L2_CACHE_SIZE
; j
++) {
138 s
->l2_cache_counts
[j
] >>= 1;
141 return s
->l2_cache
+ (i
<< s
->l2_bits
);
150 * Loads a L2 table into memory. If the table is in the cache, the cache
151 * is used; otherwise the L2 table is loaded from the image file.
153 * Returns a pointer to the L2 table on success, or NULL if the read from
154 * the image file failed.
157 static uint64_t *l2_load(BlockDriverState
*bs
, uint64_t l2_offset
)
159 BDRVQcowState
*s
= bs
->opaque
;
163 /* seek if the table for the given offset is in the cache */
165 l2_table
= seek_l2_table(s
, l2_offset
);
166 if (l2_table
!= NULL
)
169 /* not found: load a new entry in the least used one */
171 min_index
= l2_cache_new_entry(bs
);
172 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
173 if (bdrv_pread(s
->hd
, l2_offset
, l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
174 s
->l2_size
* sizeof(uint64_t))
176 s
->l2_cache_offsets
[min_index
] = l2_offset
;
177 s
->l2_cache_counts
[min_index
] = 1;
183 * Writes one sector of the L1 table to the disk (can't update single entries
184 * and we really don't want bdrv_pread to perform a read-modify-write)
186 #define L1_ENTRIES_PER_SECTOR (512 / 8)
187 static int write_l1_entry(BDRVQcowState
*s
, int l1_index
)
189 uint64_t buf
[L1_ENTRIES_PER_SECTOR
];
193 l1_start_index
= l1_index
& ~(L1_ENTRIES_PER_SECTOR
- 1);
194 for (i
= 0; i
< L1_ENTRIES_PER_SECTOR
; i
++) {
195 buf
[i
] = cpu_to_be64(s
->l1_table
[l1_start_index
+ i
]);
198 if (bdrv_pwrite(s
->hd
, s
->l1_table_offset
+ 8 * l1_start_index
,
199 buf
, sizeof(buf
)) != sizeof(buf
))
210 * Allocate a new l2 entry in the file. If l1_index points to an already
211 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
212 * table) copy the contents of the old L2 table into the newly allocated one.
213 * Otherwise the new table is initialized with zeros.
217 static uint64_t *l2_allocate(BlockDriverState
*bs
, int l1_index
)
219 BDRVQcowState
*s
= bs
->opaque
;
221 uint64_t old_l2_offset
;
225 old_l2_offset
= s
->l1_table
[l1_index
];
227 /* allocate a new l2 entry */
229 l2_offset
= qcow2_alloc_clusters(bs
, s
->l2_size
* sizeof(uint64_t));
234 /* allocate a new entry in the l2 cache */
236 min_index
= l2_cache_new_entry(bs
);
237 l2_table
= s
->l2_cache
+ (min_index
<< s
->l2_bits
);
239 if (old_l2_offset
== 0) {
240 /* if there was no old l2 table, clear the new table */
241 memset(l2_table
, 0, s
->l2_size
* sizeof(uint64_t));
243 /* if there was an old l2 table, read it from the disk */
244 if (bdrv_pread(s
->hd
, old_l2_offset
,
245 l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
246 s
->l2_size
* sizeof(uint64_t))
249 /* write the l2 table to the file */
250 if (bdrv_pwrite(s
->hd
, l2_offset
,
251 l2_table
, s
->l2_size
* sizeof(uint64_t)) !=
252 s
->l2_size
* sizeof(uint64_t))
255 /* update the L1 entry */
256 s
->l1_table
[l1_index
] = l2_offset
| QCOW_OFLAG_COPIED
;
257 if (write_l1_entry(s
, l1_index
) < 0) {
261 /* update the l2 cache entry */
263 s
->l2_cache_offsets
[min_index
] = l2_offset
;
264 s
->l2_cache_counts
[min_index
] = 1;
269 qcow2_l2_cache_reset(bs
);
273 static int count_contiguous_clusters(uint64_t nb_clusters
, int cluster_size
,
274 uint64_t *l2_table
, uint64_t start
, uint64_t mask
)
277 uint64_t offset
= be64_to_cpu(l2_table
[0]) & ~mask
;
282 for (i
= start
; i
< start
+ nb_clusters
; i
++)
283 if (offset
+ (uint64_t) i
* cluster_size
!= (be64_to_cpu(l2_table
[i
]) & ~mask
))
289 static int count_contiguous_free_clusters(uint64_t nb_clusters
, uint64_t *l2_table
)
293 while(nb_clusters
-- && l2_table
[i
] == 0)
299 /* The crypt function is compatible with the linux cryptoloop
300 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
302 void qcow2_encrypt_sectors(BDRVQcowState
*s
, int64_t sector_num
,
303 uint8_t *out_buf
, const uint8_t *in_buf
,
304 int nb_sectors
, int enc
,
313 for(i
= 0; i
< nb_sectors
; i
++) {
314 ivec
.ll
[0] = cpu_to_le64(sector_num
);
316 AES_cbc_encrypt(in_buf
, out_buf
, 512, key
,
325 static int qcow_read(BlockDriverState
*bs
, int64_t sector_num
,
326 uint8_t *buf
, int nb_sectors
)
328 BDRVQcowState
*s
= bs
->opaque
;
329 int ret
, index_in_cluster
, n
, n1
;
330 uint64_t cluster_offset
;
332 while (nb_sectors
> 0) {
334 cluster_offset
= qcow2_get_cluster_offset(bs
, sector_num
<< 9, &n
);
335 index_in_cluster
= sector_num
& (s
->cluster_sectors
- 1);
336 if (!cluster_offset
) {
337 if (bs
->backing_hd
) {
338 /* read from the base image */
339 n1
= qcow2_backing_read1(bs
->backing_hd
, sector_num
, buf
, n
);
341 ret
= bdrv_read(bs
->backing_hd
, sector_num
, buf
, n1
);
346 memset(buf
, 0, 512 * n
);
348 } else if (cluster_offset
& QCOW_OFLAG_COMPRESSED
) {
349 if (qcow2_decompress_cluster(s
, cluster_offset
) < 0)
351 memcpy(buf
, s
->cluster_cache
+ index_in_cluster
* 512, 512 * n
);
353 ret
= bdrv_pread(s
->hd
, cluster_offset
+ index_in_cluster
* 512, buf
, n
* 512);
356 if (s
->crypt_method
) {
357 qcow2_encrypt_sectors(s
, sector_num
, buf
, buf
, n
, 0,
358 &s
->aes_decrypt_key
);
368 static int copy_sectors(BlockDriverState
*bs
, uint64_t start_sect
,
369 uint64_t cluster_offset
, int n_start
, int n_end
)
371 BDRVQcowState
*s
= bs
->opaque
;
377 ret
= qcow_read(bs
, start_sect
+ n_start
, s
->cluster_data
, n
);
380 if (s
->crypt_method
) {
381 qcow2_encrypt_sectors(s
, start_sect
+ n_start
,
383 s
->cluster_data
, n
, 1,
384 &s
->aes_encrypt_key
);
386 ret
= bdrv_write(s
->hd
, (cluster_offset
>> 9) + n_start
,
397 * For a given offset of the disk image, return cluster offset in
400 * on entry, *num is the number of contiguous clusters we'd like to
401 * access following offset.
403 * on exit, *num is the number of contiguous clusters we can read.
405 * Return 1, if the offset is found
406 * Return 0, otherwise.
410 uint64_t qcow2_get_cluster_offset(BlockDriverState
*bs
, uint64_t offset
,
413 BDRVQcowState
*s
= bs
->opaque
;
414 unsigned int l1_index
, l2_index
;
415 uint64_t l2_offset
, *l2_table
, cluster_offset
;
417 unsigned int index_in_cluster
, nb_clusters
;
418 uint64_t nb_available
, nb_needed
;
420 index_in_cluster
= (offset
>> 9) & (s
->cluster_sectors
- 1);
421 nb_needed
= *num
+ index_in_cluster
;
423 l1_bits
= s
->l2_bits
+ s
->cluster_bits
;
425 /* compute how many bytes there are between the offset and
426 * the end of the l1 entry
429 nb_available
= (1ULL << l1_bits
) - (offset
& ((1ULL << l1_bits
) - 1));
431 /* compute the number of available sectors */
433 nb_available
= (nb_available
>> 9) + index_in_cluster
;
435 if (nb_needed
> nb_available
) {
436 nb_needed
= nb_available
;
441 /* seek the the l2 offset in the l1 table */
443 l1_index
= offset
>> l1_bits
;
444 if (l1_index
>= s
->l1_size
)
447 l2_offset
= s
->l1_table
[l1_index
];
449 /* seek the l2 table of the given l2 offset */
454 /* load the l2 table in memory */
456 l2_offset
&= ~QCOW_OFLAG_COPIED
;
457 l2_table
= l2_load(bs
, l2_offset
);
458 if (l2_table
== NULL
)
461 /* find the cluster offset for the given disk offset */
463 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
464 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
465 nb_clusters
= size_to_clusters(s
, nb_needed
<< 9);
467 if (!cluster_offset
) {
468 /* how many empty clusters ? */
469 c
= count_contiguous_free_clusters(nb_clusters
, &l2_table
[l2_index
]);
471 /* how many allocated clusters ? */
472 c
= count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
473 &l2_table
[l2_index
], 0, QCOW_OFLAG_COPIED
);
476 nb_available
= (c
* s
->cluster_sectors
);
478 if (nb_available
> nb_needed
)
479 nb_available
= nb_needed
;
481 *num
= nb_available
- index_in_cluster
;
483 return cluster_offset
& ~QCOW_OFLAG_COPIED
;
489 * for a given disk offset, load (and allocate if needed)
492 * the l2 table offset in the qcow2 file and the cluster index
493 * in the l2 table are given to the caller.
495 * Returns 0 on success, -errno in failure case
497 static int get_cluster_table(BlockDriverState
*bs
, uint64_t offset
,
498 uint64_t **new_l2_table
,
499 uint64_t *new_l2_offset
,
502 BDRVQcowState
*s
= bs
->opaque
;
503 unsigned int l1_index
, l2_index
;
504 uint64_t l2_offset
, *l2_table
;
507 /* seek the the l2 offset in the l1 table */
509 l1_index
= offset
>> (s
->l2_bits
+ s
->cluster_bits
);
510 if (l1_index
>= s
->l1_size
) {
511 ret
= qcow2_grow_l1_table(bs
, l1_index
+ 1);
516 l2_offset
= s
->l1_table
[l1_index
];
518 /* seek the l2 table of the given l2 offset */
520 if (l2_offset
& QCOW_OFLAG_COPIED
) {
521 /* load the l2 table in memory */
522 l2_offset
&= ~QCOW_OFLAG_COPIED
;
523 l2_table
= l2_load(bs
, l2_offset
);
524 if (l2_table
== NULL
) {
529 qcow2_free_clusters(bs
, l2_offset
, s
->l2_size
* sizeof(uint64_t));
530 l2_table
= l2_allocate(bs
, l1_index
);
531 if (l2_table
== NULL
) {
534 l2_offset
= s
->l1_table
[l1_index
] & ~QCOW_OFLAG_COPIED
;
537 /* find the cluster offset for the given disk offset */
539 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
541 *new_l2_table
= l2_table
;
542 *new_l2_offset
= l2_offset
;
543 *new_l2_index
= l2_index
;
549 * alloc_compressed_cluster_offset
551 * For a given offset of the disk image, return cluster offset in
554 * If the offset is not found, allocate a new compressed cluster.
556 * Return the cluster offset if successful,
557 * Return 0, otherwise.
561 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState
*bs
,
565 BDRVQcowState
*s
= bs
->opaque
;
567 uint64_t l2_offset
, *l2_table
;
568 int64_t cluster_offset
;
571 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_offset
, &l2_index
);
576 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
577 if (cluster_offset
& QCOW_OFLAG_COPIED
)
578 return cluster_offset
& ~QCOW_OFLAG_COPIED
;
581 qcow2_free_any_clusters(bs
, cluster_offset
, 1);
583 cluster_offset
= qcow2_alloc_bytes(bs
, compressed_size
);
584 if (cluster_offset
< 0) {
588 nb_csectors
= ((cluster_offset
+ compressed_size
- 1) >> 9) -
589 (cluster_offset
>> 9);
591 cluster_offset
|= QCOW_OFLAG_COMPRESSED
|
592 ((uint64_t)nb_csectors
<< s
->csize_shift
);
594 /* update L2 table */
596 /* compressed clusters never have the copied flag */
598 l2_table
[l2_index
] = cpu_to_be64(cluster_offset
);
599 if (bdrv_pwrite(s
->hd
,
600 l2_offset
+ l2_index
* sizeof(uint64_t),
602 sizeof(uint64_t)) != sizeof(uint64_t))
605 return cluster_offset
;
609 * Write L2 table updates to disk, writing whole sectors to avoid a
610 * read-modify-write in bdrv_pwrite
612 #define L2_ENTRIES_PER_SECTOR (512 / 8)
613 static int write_l2_entries(BDRVQcowState
*s
, uint64_t *l2_table
,
614 uint64_t l2_offset
, int l2_index
, int num
)
616 int l2_start_index
= l2_index
& ~(L1_ENTRIES_PER_SECTOR
- 1);
617 int start_offset
= (8 * l2_index
) & ~511;
618 int end_offset
= (8 * (l2_index
+ num
) + 511) & ~511;
619 size_t len
= end_offset
- start_offset
;
621 if (bdrv_pwrite(s
->hd
, l2_offset
+ start_offset
, &l2_table
[l2_start_index
],
630 int qcow2_alloc_cluster_link_l2(BlockDriverState
*bs
, QCowL2Meta
*m
)
632 BDRVQcowState
*s
= bs
->opaque
;
633 int i
, j
= 0, l2_index
, ret
;
634 uint64_t *old_cluster
, start_sect
, l2_offset
, *l2_table
;
635 uint64_t cluster_offset
= m
->cluster_offset
;
637 if (m
->nb_clusters
== 0)
640 old_cluster
= qemu_malloc(m
->nb_clusters
* sizeof(uint64_t));
642 /* copy content of unmodified sectors */
643 start_sect
= (m
->offset
& ~(s
->cluster_size
- 1)) >> 9;
645 ret
= copy_sectors(bs
, start_sect
, cluster_offset
, 0, m
->n_start
);
650 if (m
->nb_available
& (s
->cluster_sectors
- 1)) {
651 uint64_t end
= m
->nb_available
& ~(uint64_t)(s
->cluster_sectors
- 1);
652 ret
= copy_sectors(bs
, start_sect
+ end
, cluster_offset
+ (end
<< 9),
653 m
->nb_available
- end
, s
->cluster_sectors
);
658 /* update L2 table */
659 ret
= get_cluster_table(bs
, m
->offset
, &l2_table
, &l2_offset
, &l2_index
);
664 for (i
= 0; i
< m
->nb_clusters
; i
++) {
665 /* if two concurrent writes happen to the same unallocated cluster
666 * each write allocates separate cluster and writes data concurrently.
667 * The first one to complete updates l2 table with pointer to its
668 * cluster the second one has to do RMW (which is done above by
669 * copy_sectors()), update l2 table with its cluster pointer and free
670 * old cluster. This is what this loop does */
671 if(l2_table
[l2_index
+ i
] != 0)
672 old_cluster
[j
++] = l2_table
[l2_index
+ i
];
674 l2_table
[l2_index
+ i
] = cpu_to_be64((cluster_offset
+
675 (i
<< s
->cluster_bits
)) | QCOW_OFLAG_COPIED
);
678 ret
= write_l2_entries(s
, l2_table
, l2_offset
, l2_index
, m
->nb_clusters
);
680 qcow2_l2_cache_reset(bs
);
684 for (i
= 0; i
< j
; i
++)
685 qcow2_free_any_clusters(bs
,
686 be64_to_cpu(old_cluster
[i
]) & ~QCOW_OFLAG_COPIED
, 1);
690 qemu_free(old_cluster
);
695 * alloc_cluster_offset
697 * For a given offset of the disk image, return cluster offset in qcow2 file.
698 * If the offset is not found, allocate a new cluster.
700 * If the cluster was already allocated, m->nb_clusters is set to 0,
701 * m->depends_on is set to NULL and the other fields in m are meaningless.
703 * If the cluster is newly allocated, m->nb_clusters is set to the number of
704 * contiguous clusters that have been allocated. This may be 0 if the request
705 * conflict with another write request in flight; in this case, m->depends_on
706 * is set and the remaining fields of m are meaningless.
708 * If m->nb_clusters is non-zero, the other fields of m are valid and contain
709 * information about the first allocated cluster.
711 * Return 0 on success and -errno in error cases
713 int qcow2_alloc_cluster_offset(BlockDriverState
*bs
, uint64_t offset
,
714 int n_start
, int n_end
, int *num
, QCowL2Meta
*m
)
716 BDRVQcowState
*s
= bs
->opaque
;
718 uint64_t l2_offset
, *l2_table
;
719 int64_t cluster_offset
;
720 unsigned int nb_clusters
, i
= 0;
721 QCowL2Meta
*old_alloc
;
723 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_offset
, &l2_index
);
728 nb_clusters
= size_to_clusters(s
, n_end
<< 9);
730 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
732 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
734 /* We keep all QCOW_OFLAG_COPIED clusters */
736 if (cluster_offset
& QCOW_OFLAG_COPIED
) {
737 nb_clusters
= count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
738 &l2_table
[l2_index
], 0, 0);
740 cluster_offset
&= ~QCOW_OFLAG_COPIED
;
742 m
->depends_on
= NULL
;
747 /* for the moment, multiple compressed clusters are not managed */
749 if (cluster_offset
& QCOW_OFLAG_COMPRESSED
)
752 /* how many available clusters ? */
754 while (i
< nb_clusters
) {
755 i
+= count_contiguous_clusters(nb_clusters
- i
, s
->cluster_size
,
756 &l2_table
[l2_index
], i
, 0);
757 if ((i
>= nb_clusters
) || be64_to_cpu(l2_table
[l2_index
+ i
])) {
761 i
+= count_contiguous_free_clusters(nb_clusters
- i
,
762 &l2_table
[l2_index
+ i
]);
763 if (i
>= nb_clusters
) {
767 cluster_offset
= be64_to_cpu(l2_table
[l2_index
+ i
]);
769 if ((cluster_offset
& QCOW_OFLAG_COPIED
) ||
770 (cluster_offset
& QCOW_OFLAG_COMPRESSED
))
773 assert(i
<= nb_clusters
);
777 * Check if there already is an AIO write request in flight which allocates
778 * the same cluster. In this case we need to wait until the previous
779 * request has completed and updated the L2 table accordingly.
781 QLIST_FOREACH(old_alloc
, &s
->cluster_allocs
, next_in_flight
) {
783 uint64_t end_offset
= offset
+ nb_clusters
* s
->cluster_size
;
784 uint64_t old_offset
= old_alloc
->offset
;
785 uint64_t old_end_offset
= old_alloc
->offset
+
786 old_alloc
->nb_clusters
* s
->cluster_size
;
788 if (end_offset
< old_offset
|| offset
> old_end_offset
) {
789 /* No intersection */
791 if (offset
< old_offset
) {
792 /* Stop at the start of a running allocation */
793 nb_clusters
= (old_offset
- offset
) >> s
->cluster_bits
;
798 if (nb_clusters
== 0) {
799 /* Set dependency and wait for a callback */
800 m
->depends_on
= old_alloc
;
812 QLIST_INSERT_HEAD(&s
->cluster_allocs
, m
, next_in_flight
);
814 /* allocate a new cluster */
816 cluster_offset
= qcow2_alloc_clusters(bs
, nb_clusters
* s
->cluster_size
);
817 if (cluster_offset
< 0) {
818 QLIST_REMOVE(m
, next_in_flight
);
819 return cluster_offset
;
822 /* save info needed for meta data update */
824 m
->n_start
= n_start
;
825 m
->nb_clusters
= nb_clusters
;
828 m
->nb_available
= MIN(nb_clusters
<< (s
->cluster_bits
- 9), n_end
);
829 m
->cluster_offset
= cluster_offset
;
831 *num
= m
->nb_available
- n_start
;
836 static int decompress_buffer(uint8_t *out_buf
, int out_buf_size
,
837 const uint8_t *buf
, int buf_size
)
839 z_stream strm1
, *strm
= &strm1
;
842 memset(strm
, 0, sizeof(*strm
));
844 strm
->next_in
= (uint8_t *)buf
;
845 strm
->avail_in
= buf_size
;
846 strm
->next_out
= out_buf
;
847 strm
->avail_out
= out_buf_size
;
849 ret
= inflateInit2(strm
, -12);
852 ret
= inflate(strm
, Z_FINISH
);
853 out_len
= strm
->next_out
- out_buf
;
854 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) ||
855 out_len
!= out_buf_size
) {
863 int qcow2_decompress_cluster(BDRVQcowState
*s
, uint64_t cluster_offset
)
865 int ret
, csize
, nb_csectors
, sector_offset
;
868 coffset
= cluster_offset
& s
->cluster_offset_mask
;
869 if (s
->cluster_cache_offset
!= coffset
) {
870 nb_csectors
= ((cluster_offset
>> s
->csize_shift
) & s
->csize_mask
) + 1;
871 sector_offset
= coffset
& 511;
872 csize
= nb_csectors
* 512 - sector_offset
;
873 ret
= bdrv_read(s
->hd
, coffset
>> 9, s
->cluster_data
, nb_csectors
);
877 if (decompress_buffer(s
->cluster_cache
, s
->cluster_size
,
878 s
->cluster_data
+ sector_offset
, csize
) < 0) {
881 s
->cluster_cache_offset
= coffset
;