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/block_int.h"
29 #include "block/qcow2.h"
32 int qcow2_grow_l1_table(BlockDriverState
*bs
, uint64_t min_size
,
35 BDRVQcowState
*s
= bs
->opaque
;
36 int new_l1_size2
, ret
, i
;
37 uint64_t *new_l1_table
;
38 int64_t old_l1_table_offset
, old_l1_size
;
39 int64_t new_l1_table_offset
, new_l1_size
;
42 if (min_size
<= s
->l1_size
)
45 /* Do a sanity check on min_size before trying to calculate new_l1_size
46 * (this prevents overflows during the while loop for the calculation of
48 if (min_size
> INT_MAX
/ sizeof(uint64_t)) {
53 new_l1_size
= min_size
;
55 /* Bump size up to reduce the number of times we have to grow */
56 new_l1_size
= s
->l1_size
;
57 if (new_l1_size
== 0) {
60 while (min_size
> new_l1_size
) {
61 new_l1_size
= (new_l1_size
* 3 + 1) / 2;
65 if (new_l1_size
> INT_MAX
/ sizeof(uint64_t)) {
70 fprintf(stderr
, "grow l1_table from %d to %" PRId64
"\n",
71 s
->l1_size
, new_l1_size
);
74 new_l1_size2
= sizeof(uint64_t) * new_l1_size
;
75 new_l1_table
= g_malloc0(align_offset(new_l1_size2
, 512));
76 memcpy(new_l1_table
, s
->l1_table
, s
->l1_size
* sizeof(uint64_t));
78 /* write new table (align to cluster) */
79 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_GROW_ALLOC_TABLE
);
80 new_l1_table_offset
= qcow2_alloc_clusters(bs
, new_l1_size2
);
81 if (new_l1_table_offset
< 0) {
83 return new_l1_table_offset
;
86 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
91 /* the L1 position has not yet been updated, so these clusters must
92 * indeed be completely free */
93 ret
= qcow2_pre_write_overlap_check(bs
, 0, new_l1_table_offset
,
99 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_GROW_WRITE_TABLE
);
100 for(i
= 0; i
< s
->l1_size
; i
++)
101 new_l1_table
[i
] = cpu_to_be64(new_l1_table
[i
]);
102 ret
= bdrv_pwrite_sync(bs
->file
, new_l1_table_offset
, new_l1_table
, new_l1_size2
);
105 for(i
= 0; i
< s
->l1_size
; i
++)
106 new_l1_table
[i
] = be64_to_cpu(new_l1_table
[i
]);
109 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_GROW_ACTIVATE_TABLE
);
110 cpu_to_be32w((uint32_t*)data
, new_l1_size
);
111 stq_be_p(data
+ 4, new_l1_table_offset
);
112 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_size
), data
,sizeof(data
));
117 old_l1_table_offset
= s
->l1_table_offset
;
118 s
->l1_table_offset
= new_l1_table_offset
;
119 s
->l1_table
= new_l1_table
;
120 old_l1_size
= s
->l1_size
;
121 s
->l1_size
= new_l1_size
;
122 qcow2_free_clusters(bs
, old_l1_table_offset
, old_l1_size
* sizeof(uint64_t),
123 QCOW2_DISCARD_OTHER
);
126 g_free(new_l1_table
);
127 qcow2_free_clusters(bs
, new_l1_table_offset
, new_l1_size2
,
128 QCOW2_DISCARD_OTHER
);
135 * Loads a L2 table into memory. If the table is in the cache, the cache
136 * is used; otherwise the L2 table is loaded from the image file.
138 * Returns a pointer to the L2 table on success, or NULL if the read from
139 * the image file failed.
142 static int l2_load(BlockDriverState
*bs
, uint64_t l2_offset
,
145 BDRVQcowState
*s
= bs
->opaque
;
148 ret
= qcow2_cache_get(bs
, s
->l2_table_cache
, l2_offset
, (void**) l2_table
);
154 * Writes one sector of the L1 table to the disk (can't update single entries
155 * and we really don't want bdrv_pread to perform a read-modify-write)
157 #define L1_ENTRIES_PER_SECTOR (512 / 8)
158 int qcow2_write_l1_entry(BlockDriverState
*bs
, int l1_index
)
160 BDRVQcowState
*s
= bs
->opaque
;
161 uint64_t buf
[L1_ENTRIES_PER_SECTOR
];
165 l1_start_index
= l1_index
& ~(L1_ENTRIES_PER_SECTOR
- 1);
166 for (i
= 0; i
< L1_ENTRIES_PER_SECTOR
; i
++) {
167 buf
[i
] = cpu_to_be64(s
->l1_table
[l1_start_index
+ i
]);
170 ret
= qcow2_pre_write_overlap_check(bs
, QCOW2_OL_ACTIVE_L1
,
171 s
->l1_table_offset
+ 8 * l1_start_index
, sizeof(buf
));
176 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
177 ret
= bdrv_pwrite_sync(bs
->file
, s
->l1_table_offset
+ 8 * l1_start_index
,
189 * Allocate a new l2 entry in the file. If l1_index points to an already
190 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
191 * table) copy the contents of the old L2 table into the newly allocated one.
192 * Otherwise the new table is initialized with zeros.
196 static int l2_allocate(BlockDriverState
*bs
, int l1_index
, uint64_t **table
)
198 BDRVQcowState
*s
= bs
->opaque
;
199 uint64_t old_l2_offset
;
200 uint64_t *l2_table
= NULL
;
204 old_l2_offset
= s
->l1_table
[l1_index
];
206 trace_qcow2_l2_allocate(bs
, l1_index
);
208 /* allocate a new l2 entry */
210 l2_offset
= qcow2_alloc_clusters(bs
, s
->l2_size
* sizeof(uint64_t));
216 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
221 /* allocate a new entry in the l2 cache */
223 trace_qcow2_l2_allocate_get_empty(bs
, l1_index
);
224 ret
= qcow2_cache_get_empty(bs
, s
->l2_table_cache
, l2_offset
, (void**) table
);
231 if ((old_l2_offset
& L1E_OFFSET_MASK
) == 0) {
232 /* if there was no old l2 table, clear the new table */
233 memset(l2_table
, 0, s
->l2_size
* sizeof(uint64_t));
237 /* if there was an old l2 table, read it from the disk */
238 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_ALLOC_COW_READ
);
239 ret
= qcow2_cache_get(bs
, s
->l2_table_cache
,
240 old_l2_offset
& L1E_OFFSET_MASK
,
241 (void**) &old_table
);
246 memcpy(l2_table
, old_table
, s
->cluster_size
);
248 ret
= qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &old_table
);
254 /* write the l2 table to the file */
255 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_ALLOC_WRITE
);
257 trace_qcow2_l2_allocate_write_l2(bs
, l1_index
);
258 qcow2_cache_entry_mark_dirty(s
->l2_table_cache
, l2_table
);
259 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
264 /* update the L1 entry */
265 trace_qcow2_l2_allocate_write_l1(bs
, l1_index
);
266 s
->l1_table
[l1_index
] = l2_offset
| QCOW_OFLAG_COPIED
;
267 ret
= qcow2_write_l1_entry(bs
, l1_index
);
273 trace_qcow2_l2_allocate_done(bs
, l1_index
, 0);
277 trace_qcow2_l2_allocate_done(bs
, l1_index
, ret
);
278 if (l2_table
!= NULL
) {
279 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) table
);
281 s
->l1_table
[l1_index
] = old_l2_offset
;
283 qcow2_free_clusters(bs
, l2_offset
, s
->l2_size
* sizeof(uint64_t),
284 QCOW2_DISCARD_ALWAYS
);
290 * Checks how many clusters in a given L2 table are contiguous in the image
291 * file. As soon as one of the flags in the bitmask stop_flags changes compared
292 * to the first cluster, the search is stopped and the cluster is not counted
293 * as contiguous. (This allows it, for example, to stop at the first compressed
294 * cluster which may require a different handling)
296 static int count_contiguous_clusters(uint64_t nb_clusters
, int cluster_size
,
297 uint64_t *l2_table
, uint64_t stop_flags
)
300 uint64_t mask
= stop_flags
| L2E_OFFSET_MASK
| QCOW_OFLAG_COMPRESSED
;
301 uint64_t first_entry
= be64_to_cpu(l2_table
[0]);
302 uint64_t offset
= first_entry
& mask
;
307 assert(qcow2_get_cluster_type(first_entry
) != QCOW2_CLUSTER_COMPRESSED
);
309 for (i
= 0; i
< nb_clusters
; i
++) {
310 uint64_t l2_entry
= be64_to_cpu(l2_table
[i
]) & mask
;
311 if (offset
+ (uint64_t) i
* cluster_size
!= l2_entry
) {
319 static int count_contiguous_free_clusters(uint64_t nb_clusters
, uint64_t *l2_table
)
323 for (i
= 0; i
< nb_clusters
; i
++) {
324 int type
= qcow2_get_cluster_type(be64_to_cpu(l2_table
[i
]));
326 if (type
!= QCOW2_CLUSTER_UNALLOCATED
) {
334 /* The crypt function is compatible with the linux cryptoloop
335 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
337 void qcow2_encrypt_sectors(BDRVQcowState
*s
, int64_t sector_num
,
338 uint8_t *out_buf
, const uint8_t *in_buf
,
339 int nb_sectors
, int enc
,
348 for(i
= 0; i
< nb_sectors
; i
++) {
349 ivec
.ll
[0] = cpu_to_le64(sector_num
);
351 AES_cbc_encrypt(in_buf
, out_buf
, 512, key
,
359 static int coroutine_fn
copy_sectors(BlockDriverState
*bs
,
361 uint64_t cluster_offset
,
362 int n_start
, int n_end
)
364 BDRVQcowState
*s
= bs
->opaque
;
374 iov
.iov_len
= n
* BDRV_SECTOR_SIZE
;
375 iov
.iov_base
= qemu_blockalign(bs
, iov
.iov_len
);
377 qemu_iovec_init_external(&qiov
, &iov
, 1);
379 BLKDBG_EVENT(bs
->file
, BLKDBG_COW_READ
);
385 /* Call .bdrv_co_readv() directly instead of using the public block-layer
386 * interface. This avoids double I/O throttling and request tracking,
387 * which can lead to deadlock when block layer copy-on-read is enabled.
389 ret
= bs
->drv
->bdrv_co_readv(bs
, start_sect
+ n_start
, n
, &qiov
);
394 if (s
->crypt_method
) {
395 qcow2_encrypt_sectors(s
, start_sect
+ n_start
,
396 iov
.iov_base
, iov
.iov_base
, n
, 1,
397 &s
->aes_encrypt_key
);
400 ret
= qcow2_pre_write_overlap_check(bs
, 0,
401 cluster_offset
+ n_start
* BDRV_SECTOR_SIZE
, n
* BDRV_SECTOR_SIZE
);
406 BLKDBG_EVENT(bs
->file
, BLKDBG_COW_WRITE
);
407 ret
= bdrv_co_writev(bs
->file
, (cluster_offset
>> 9) + n_start
, n
, &qiov
);
414 qemu_vfree(iov
.iov_base
);
422 * For a given offset of the disk image, find the cluster offset in
423 * qcow2 file. The offset is stored in *cluster_offset.
425 * on entry, *num is the number of contiguous sectors we'd like to
426 * access following offset.
428 * on exit, *num is the number of contiguous sectors we can read.
430 * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error
433 int qcow2_get_cluster_offset(BlockDriverState
*bs
, uint64_t offset
,
434 int *num
, uint64_t *cluster_offset
)
436 BDRVQcowState
*s
= bs
->opaque
;
437 unsigned int l2_index
;
438 uint64_t l1_index
, l2_offset
, *l2_table
;
440 unsigned int index_in_cluster
, nb_clusters
;
441 uint64_t nb_available
, nb_needed
;
444 index_in_cluster
= (offset
>> 9) & (s
->cluster_sectors
- 1);
445 nb_needed
= *num
+ index_in_cluster
;
447 l1_bits
= s
->l2_bits
+ s
->cluster_bits
;
449 /* compute how many bytes there are between the offset and
450 * the end of the l1 entry
453 nb_available
= (1ULL << l1_bits
) - (offset
& ((1ULL << l1_bits
) - 1));
455 /* compute the number of available sectors */
457 nb_available
= (nb_available
>> 9) + index_in_cluster
;
459 if (nb_needed
> nb_available
) {
460 nb_needed
= nb_available
;
465 /* seek the the l2 offset in the l1 table */
467 l1_index
= offset
>> l1_bits
;
468 if (l1_index
>= s
->l1_size
) {
469 ret
= QCOW2_CLUSTER_UNALLOCATED
;
473 l2_offset
= s
->l1_table
[l1_index
] & L1E_OFFSET_MASK
;
475 ret
= QCOW2_CLUSTER_UNALLOCATED
;
479 /* load the l2 table in memory */
481 ret
= l2_load(bs
, l2_offset
, &l2_table
);
486 /* find the cluster offset for the given disk offset */
488 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
489 *cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
490 nb_clusters
= size_to_clusters(s
, nb_needed
<< 9);
492 ret
= qcow2_get_cluster_type(*cluster_offset
);
494 case QCOW2_CLUSTER_COMPRESSED
:
495 /* Compressed clusters can only be processed one by one */
497 *cluster_offset
&= L2E_COMPRESSED_OFFSET_SIZE_MASK
;
499 case QCOW2_CLUSTER_ZERO
:
500 if (s
->qcow_version
< 3) {
501 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
504 c
= count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
505 &l2_table
[l2_index
], QCOW_OFLAG_ZERO
);
508 case QCOW2_CLUSTER_UNALLOCATED
:
509 /* how many empty clusters ? */
510 c
= count_contiguous_free_clusters(nb_clusters
, &l2_table
[l2_index
]);
513 case QCOW2_CLUSTER_NORMAL
:
514 /* how many allocated clusters ? */
515 c
= count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
516 &l2_table
[l2_index
], QCOW_OFLAG_ZERO
);
517 *cluster_offset
&= L2E_OFFSET_MASK
;
523 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
525 nb_available
= (c
* s
->cluster_sectors
);
528 if (nb_available
> nb_needed
)
529 nb_available
= nb_needed
;
531 *num
= nb_available
- index_in_cluster
;
539 * for a given disk offset, load (and allocate if needed)
542 * the l2 table offset in the qcow2 file and the cluster index
543 * in the l2 table are given to the caller.
545 * Returns 0 on success, -errno in failure case
547 static int get_cluster_table(BlockDriverState
*bs
, uint64_t offset
,
548 uint64_t **new_l2_table
,
551 BDRVQcowState
*s
= bs
->opaque
;
552 unsigned int l2_index
;
553 uint64_t l1_index
, l2_offset
;
554 uint64_t *l2_table
= NULL
;
557 /* seek the the l2 offset in the l1 table */
559 l1_index
= offset
>> (s
->l2_bits
+ s
->cluster_bits
);
560 if (l1_index
>= s
->l1_size
) {
561 ret
= qcow2_grow_l1_table(bs
, l1_index
+ 1, false);
567 assert(l1_index
< s
->l1_size
);
568 l2_offset
= s
->l1_table
[l1_index
] & L1E_OFFSET_MASK
;
570 /* seek the l2 table of the given l2 offset */
572 if (s
->l1_table
[l1_index
] & QCOW_OFLAG_COPIED
) {
573 /* load the l2 table in memory */
574 ret
= l2_load(bs
, l2_offset
, &l2_table
);
579 /* First allocate a new L2 table (and do COW if needed) */
580 ret
= l2_allocate(bs
, l1_index
, &l2_table
);
585 /* Then decrease the refcount of the old table */
587 qcow2_free_clusters(bs
, l2_offset
, s
->l2_size
* sizeof(uint64_t),
588 QCOW2_DISCARD_OTHER
);
592 /* find the cluster offset for the given disk offset */
594 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
596 *new_l2_table
= l2_table
;
597 *new_l2_index
= l2_index
;
603 * alloc_compressed_cluster_offset
605 * For a given offset of the disk image, return cluster offset in
608 * If the offset is not found, allocate a new compressed cluster.
610 * Return the cluster offset if successful,
611 * Return 0, otherwise.
615 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState
*bs
,
619 BDRVQcowState
*s
= bs
->opaque
;
622 int64_t cluster_offset
;
625 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_index
);
630 /* Compression can't overwrite anything. Fail if the cluster was already
632 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
633 if (cluster_offset
& L2E_OFFSET_MASK
) {
634 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
638 cluster_offset
= qcow2_alloc_bytes(bs
, compressed_size
);
639 if (cluster_offset
< 0) {
640 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
644 nb_csectors
= ((cluster_offset
+ compressed_size
- 1) >> 9) -
645 (cluster_offset
>> 9);
647 cluster_offset
|= QCOW_OFLAG_COMPRESSED
|
648 ((uint64_t)nb_csectors
<< s
->csize_shift
);
650 /* update L2 table */
652 /* compressed clusters never have the copied flag */
654 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_UPDATE_COMPRESSED
);
655 qcow2_cache_entry_mark_dirty(s
->l2_table_cache
, l2_table
);
656 l2_table
[l2_index
] = cpu_to_be64(cluster_offset
);
657 ret
= qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
662 return cluster_offset
;
665 static int perform_cow(BlockDriverState
*bs
, QCowL2Meta
*m
, Qcow2COWRegion
*r
)
667 BDRVQcowState
*s
= bs
->opaque
;
670 if (r
->nb_sectors
== 0) {
674 qemu_co_mutex_unlock(&s
->lock
);
675 ret
= copy_sectors(bs
, m
->offset
/ BDRV_SECTOR_SIZE
, m
->alloc_offset
,
676 r
->offset
/ BDRV_SECTOR_SIZE
,
677 r
->offset
/ BDRV_SECTOR_SIZE
+ r
->nb_sectors
);
678 qemu_co_mutex_lock(&s
->lock
);
685 * Before we update the L2 table to actually point to the new cluster, we
686 * need to be sure that the refcounts have been increased and COW was
689 qcow2_cache_depends_on_flush(s
->l2_table_cache
);
694 int qcow2_alloc_cluster_link_l2(BlockDriverState
*bs
, QCowL2Meta
*m
)
696 BDRVQcowState
*s
= bs
->opaque
;
697 int i
, j
= 0, l2_index
, ret
;
698 uint64_t *old_cluster
, *l2_table
;
699 uint64_t cluster_offset
= m
->alloc_offset
;
701 trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m
->nb_clusters
);
702 assert(m
->nb_clusters
> 0);
704 old_cluster
= g_malloc(m
->nb_clusters
* sizeof(uint64_t));
706 /* copy content of unmodified sectors */
707 ret
= perform_cow(bs
, m
, &m
->cow_start
);
712 ret
= perform_cow(bs
, m
, &m
->cow_end
);
717 /* Update L2 table. */
718 if (s
->use_lazy_refcounts
) {
719 qcow2_mark_dirty(bs
);
721 if (qcow2_need_accurate_refcounts(s
)) {
722 qcow2_cache_set_dependency(bs
, s
->l2_table_cache
,
723 s
->refcount_block_cache
);
726 ret
= get_cluster_table(bs
, m
->offset
, &l2_table
, &l2_index
);
730 qcow2_cache_entry_mark_dirty(s
->l2_table_cache
, l2_table
);
732 assert(l2_index
+ m
->nb_clusters
<= s
->l2_size
);
733 for (i
= 0; i
< m
->nb_clusters
; i
++) {
734 /* if two concurrent writes happen to the same unallocated cluster
735 * each write allocates separate cluster and writes data concurrently.
736 * The first one to complete updates l2 table with pointer to its
737 * cluster the second one has to do RMW (which is done above by
738 * copy_sectors()), update l2 table with its cluster pointer and free
739 * old cluster. This is what this loop does */
740 if(l2_table
[l2_index
+ i
] != 0)
741 old_cluster
[j
++] = l2_table
[l2_index
+ i
];
743 l2_table
[l2_index
+ i
] = cpu_to_be64((cluster_offset
+
744 (i
<< s
->cluster_bits
)) | QCOW_OFLAG_COPIED
);
748 ret
= qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
754 * If this was a COW, we need to decrease the refcount of the old cluster.
755 * Also flush bs->file to get the right order for L2 and refcount update.
757 * Don't discard clusters that reach a refcount of 0 (e.g. compressed
758 * clusters), the next write will reuse them anyway.
761 for (i
= 0; i
< j
; i
++) {
762 qcow2_free_any_clusters(bs
, be64_to_cpu(old_cluster
[i
]), 1,
763 QCOW2_DISCARD_NEVER
);
774 * Returns the number of contiguous clusters that can be used for an allocating
775 * write, but require COW to be performed (this includes yet unallocated space,
776 * which must copy from the backing file)
778 static int count_cow_clusters(BDRVQcowState
*s
, int nb_clusters
,
779 uint64_t *l2_table
, int l2_index
)
783 for (i
= 0; i
< nb_clusters
; i
++) {
784 uint64_t l2_entry
= be64_to_cpu(l2_table
[l2_index
+ i
]);
785 int cluster_type
= qcow2_get_cluster_type(l2_entry
);
787 switch(cluster_type
) {
788 case QCOW2_CLUSTER_NORMAL
:
789 if (l2_entry
& QCOW_OFLAG_COPIED
) {
793 case QCOW2_CLUSTER_UNALLOCATED
:
794 case QCOW2_CLUSTER_COMPRESSED
:
795 case QCOW2_CLUSTER_ZERO
:
803 assert(i
<= nb_clusters
);
808 * Check if there already is an AIO write request in flight which allocates
809 * the same cluster. In this case we need to wait until the previous
810 * request has completed and updated the L2 table accordingly.
813 * 0 if there was no dependency. *cur_bytes indicates the number of
814 * bytes from guest_offset that can be read before the next
815 * dependency must be processed (or the request is complete)
817 * -EAGAIN if we had to wait for another request, previously gathered
818 * information on cluster allocation may be invalid now. The caller
819 * must start over anyway, so consider *cur_bytes undefined.
821 static int handle_dependencies(BlockDriverState
*bs
, uint64_t guest_offset
,
822 uint64_t *cur_bytes
, QCowL2Meta
**m
)
824 BDRVQcowState
*s
= bs
->opaque
;
825 QCowL2Meta
*old_alloc
;
826 uint64_t bytes
= *cur_bytes
;
828 QLIST_FOREACH(old_alloc
, &s
->cluster_allocs
, next_in_flight
) {
830 uint64_t start
= guest_offset
;
831 uint64_t end
= start
+ bytes
;
832 uint64_t old_start
= l2meta_cow_start(old_alloc
);
833 uint64_t old_end
= l2meta_cow_end(old_alloc
);
835 if (end
<= old_start
|| start
>= old_end
) {
836 /* No intersection */
838 if (start
< old_start
) {
839 /* Stop at the start of a running allocation */
840 bytes
= old_start
- start
;
845 /* Stop if already an l2meta exists. After yielding, it wouldn't
846 * be valid any more, so we'd have to clean up the old L2Metas
847 * and deal with requests depending on them before starting to
848 * gather new ones. Not worth the trouble. */
849 if (bytes
== 0 && *m
) {
855 /* Wait for the dependency to complete. We need to recheck
856 * the free/allocated clusters when we continue. */
857 qemu_co_mutex_unlock(&s
->lock
);
858 qemu_co_queue_wait(&old_alloc
->dependent_requests
);
859 qemu_co_mutex_lock(&s
->lock
);
865 /* Make sure that existing clusters and new allocations are only used up to
866 * the next dependency if we shortened the request above */
873 * Checks how many already allocated clusters that don't require a copy on
874 * write there are at the given guest_offset (up to *bytes). If
875 * *host_offset is not zero, only physically contiguous clusters beginning at
876 * this host offset are counted.
878 * Note that guest_offset may not be cluster aligned. In this case, the
879 * returned *host_offset points to exact byte referenced by guest_offset and
880 * therefore isn't cluster aligned as well.
883 * 0: if no allocated clusters are available at the given offset.
884 * *bytes is normally unchanged. It is set to 0 if the cluster
885 * is allocated and doesn't need COW, but doesn't have the right
888 * 1: if allocated clusters that don't require a COW are available at
889 * the requested offset. *bytes may have decreased and describes
890 * the length of the area that can be written to.
892 * -errno: in error cases
894 static int handle_copied(BlockDriverState
*bs
, uint64_t guest_offset
,
895 uint64_t *host_offset
, uint64_t *bytes
, QCowL2Meta
**m
)
897 BDRVQcowState
*s
= bs
->opaque
;
899 uint64_t cluster_offset
;
901 unsigned int nb_clusters
;
902 unsigned int keep_clusters
;
905 trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset
, *host_offset
,
908 assert(*host_offset
== 0 || offset_into_cluster(s
, guest_offset
)
909 == offset_into_cluster(s
, *host_offset
));
912 * Calculate the number of clusters to look for. We stop at L2 table
913 * boundaries to keep things simple.
916 size_to_clusters(s
, offset_into_cluster(s
, guest_offset
) + *bytes
);
918 l2_index
= offset_to_l2_index(s
, guest_offset
);
919 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
921 /* Find L2 entry for the first involved cluster */
922 ret
= get_cluster_table(bs
, guest_offset
, &l2_table
, &l2_index
);
927 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
929 /* Check how many clusters are already allocated and don't need COW */
930 if (qcow2_get_cluster_type(cluster_offset
) == QCOW2_CLUSTER_NORMAL
931 && (cluster_offset
& QCOW_OFLAG_COPIED
))
933 /* If a specific host_offset is required, check it */
934 bool offset_matches
=
935 (cluster_offset
& L2E_OFFSET_MASK
) == *host_offset
;
937 if (*host_offset
!= 0 && !offset_matches
) {
943 /* We keep all QCOW_OFLAG_COPIED clusters */
945 count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
947 QCOW_OFLAG_COPIED
| QCOW_OFLAG_ZERO
);
948 assert(keep_clusters
<= nb_clusters
);
951 keep_clusters
* s
->cluster_size
952 - offset_into_cluster(s
, guest_offset
));
961 pret
= qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
966 /* Only return a host offset if we actually made progress. Otherwise we
967 * would make requirements for handle_alloc() that it can't fulfill */
969 *host_offset
= (cluster_offset
& L2E_OFFSET_MASK
)
970 + offset_into_cluster(s
, guest_offset
);
977 * Allocates new clusters for the given guest_offset.
979 * At most *nb_clusters are allocated, and on return *nb_clusters is updated to
980 * contain the number of clusters that have been allocated and are contiguous
983 * If *host_offset is non-zero, it specifies the offset in the image file at
984 * which the new clusters must start. *nb_clusters can be 0 on return in this
985 * case if the cluster at host_offset is already in use. If *host_offset is
986 * zero, the clusters can be allocated anywhere in the image file.
988 * *host_offset is updated to contain the offset into the image file at which
989 * the first allocated cluster starts.
991 * Return 0 on success and -errno in error cases. -EAGAIN means that the
992 * function has been waiting for another request and the allocation must be
993 * restarted, but the whole request should not be failed.
995 static int do_alloc_cluster_offset(BlockDriverState
*bs
, uint64_t guest_offset
,
996 uint64_t *host_offset
, unsigned int *nb_clusters
)
998 BDRVQcowState
*s
= bs
->opaque
;
1000 trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset
,
1001 *host_offset
, *nb_clusters
);
1003 /* Allocate new clusters */
1004 trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
1005 if (*host_offset
== 0) {
1006 int64_t cluster_offset
=
1007 qcow2_alloc_clusters(bs
, *nb_clusters
* s
->cluster_size
);
1008 if (cluster_offset
< 0) {
1009 return cluster_offset
;
1011 *host_offset
= cluster_offset
;
1014 int ret
= qcow2_alloc_clusters_at(bs
, *host_offset
, *nb_clusters
);
1024 * Allocates new clusters for an area that either is yet unallocated or needs a
1025 * copy on write. If *host_offset is non-zero, clusters are only allocated if
1026 * the new allocation can match the specified host offset.
1028 * Note that guest_offset may not be cluster aligned. In this case, the
1029 * returned *host_offset points to exact byte referenced by guest_offset and
1030 * therefore isn't cluster aligned as well.
1033 * 0: if no clusters could be allocated. *bytes is set to 0,
1034 * *host_offset is left unchanged.
1036 * 1: if new clusters were allocated. *bytes may be decreased if the
1037 * new allocation doesn't cover all of the requested area.
1038 * *host_offset is updated to contain the host offset of the first
1039 * newly allocated cluster.
1041 * -errno: in error cases
1043 static int handle_alloc(BlockDriverState
*bs
, uint64_t guest_offset
,
1044 uint64_t *host_offset
, uint64_t *bytes
, QCowL2Meta
**m
)
1046 BDRVQcowState
*s
= bs
->opaque
;
1050 unsigned int nb_clusters
;
1053 uint64_t alloc_cluster_offset
;
1055 trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset
, *host_offset
,
1060 * Calculate the number of clusters to look for. We stop at L2 table
1061 * boundaries to keep things simple.
1064 size_to_clusters(s
, offset_into_cluster(s
, guest_offset
) + *bytes
);
1066 l2_index
= offset_to_l2_index(s
, guest_offset
);
1067 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
1069 /* Find L2 entry for the first involved cluster */
1070 ret
= get_cluster_table(bs
, guest_offset
, &l2_table
, &l2_index
);
1075 entry
= be64_to_cpu(l2_table
[l2_index
]);
1077 /* For the moment, overwrite compressed clusters one by one */
1078 if (entry
& QCOW_OFLAG_COMPRESSED
) {
1081 nb_clusters
= count_cow_clusters(s
, nb_clusters
, l2_table
, l2_index
);
1084 /* This function is only called when there were no non-COW clusters, so if
1085 * we can't find any unallocated or COW clusters either, something is
1086 * wrong with our code. */
1087 assert(nb_clusters
> 0);
1089 ret
= qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
1094 /* Allocate, if necessary at a given offset in the image file */
1095 alloc_cluster_offset
= start_of_cluster(s
, *host_offset
);
1096 ret
= do_alloc_cluster_offset(bs
, guest_offset
, &alloc_cluster_offset
,
1102 /* Can't extend contiguous allocation */
1103 if (nb_clusters
== 0) {
1109 * Save info needed for meta data update.
1111 * requested_sectors: Number of sectors from the start of the first
1112 * newly allocated cluster to the end of the (possibly shortened
1113 * before) write request.
1115 * avail_sectors: Number of sectors from the start of the first
1116 * newly allocated to the end of the last newly allocated cluster.
1118 * nb_sectors: The number of sectors from the start of the first
1119 * newly allocated cluster to the end of the area that the write
1120 * request actually writes to (excluding COW at the end)
1122 int requested_sectors
=
1123 (*bytes
+ offset_into_cluster(s
, guest_offset
))
1124 >> BDRV_SECTOR_BITS
;
1125 int avail_sectors
= nb_clusters
1126 << (s
->cluster_bits
- BDRV_SECTOR_BITS
);
1127 int alloc_n_start
= offset_into_cluster(s
, guest_offset
)
1128 >> BDRV_SECTOR_BITS
;
1129 int nb_sectors
= MIN(requested_sectors
, avail_sectors
);
1130 QCowL2Meta
*old_m
= *m
;
1132 *m
= g_malloc0(sizeof(**m
));
1134 **m
= (QCowL2Meta
) {
1137 .alloc_offset
= alloc_cluster_offset
,
1138 .offset
= start_of_cluster(s
, guest_offset
),
1139 .nb_clusters
= nb_clusters
,
1140 .nb_available
= nb_sectors
,
1144 .nb_sectors
= alloc_n_start
,
1147 .offset
= nb_sectors
* BDRV_SECTOR_SIZE
,
1148 .nb_sectors
= avail_sectors
- nb_sectors
,
1151 qemu_co_queue_init(&(*m
)->dependent_requests
);
1152 QLIST_INSERT_HEAD(&s
->cluster_allocs
, *m
, next_in_flight
);
1154 *host_offset
= alloc_cluster_offset
+ offset_into_cluster(s
, guest_offset
);
1155 *bytes
= MIN(*bytes
, (nb_sectors
* BDRV_SECTOR_SIZE
)
1156 - offset_into_cluster(s
, guest_offset
));
1157 assert(*bytes
!= 0);
1162 if (*m
&& (*m
)->nb_clusters
> 0) {
1163 QLIST_REMOVE(*m
, next_in_flight
);
1169 * alloc_cluster_offset
1171 * For a given offset on the virtual disk, find the cluster offset in qcow2
1172 * file. If the offset is not found, allocate a new cluster.
1174 * If the cluster was already allocated, m->nb_clusters is set to 0 and
1175 * other fields in m are meaningless.
1177 * If the cluster is newly allocated, m->nb_clusters is set to the number of
1178 * contiguous clusters that have been allocated. In this case, the other
1179 * fields of m are valid and contain information about the first allocated
1182 * If the request conflicts with another write request in flight, the coroutine
1183 * is queued and will be reentered when the dependency has completed.
1185 * Return 0 on success and -errno in error cases
1187 int qcow2_alloc_cluster_offset(BlockDriverState
*bs
, uint64_t offset
,
1188 int *num
, uint64_t *host_offset
, QCowL2Meta
**m
)
1190 BDRVQcowState
*s
= bs
->opaque
;
1191 uint64_t start
, remaining
;
1192 uint64_t cluster_offset
;
1196 trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset
, *num
);
1198 assert((offset
& ~BDRV_SECTOR_MASK
) == 0);
1202 remaining
= *num
<< BDRV_SECTOR_BITS
;
1210 if (!*host_offset
) {
1211 *host_offset
= start_of_cluster(s
, cluster_offset
);
1214 assert(remaining
>= cur_bytes
);
1217 remaining
-= cur_bytes
;
1218 cluster_offset
+= cur_bytes
;
1220 if (remaining
== 0) {
1224 cur_bytes
= remaining
;
1227 * Now start gathering as many contiguous clusters as possible:
1229 * 1. Check for overlaps with in-flight allocations
1231 * a) Overlap not in the first cluster -> shorten this request and
1232 * let the caller handle the rest in its next loop iteration.
1234 * b) Real overlaps of two requests. Yield and restart the search
1235 * for contiguous clusters (the situation could have changed
1236 * while we were sleeping)
1238 * c) TODO: Request starts in the same cluster as the in-flight
1239 * allocation ends. Shorten the COW of the in-fight allocation,
1240 * set cluster_offset to write to the same cluster and set up
1241 * the right synchronisation between the in-flight request and
1244 ret
= handle_dependencies(bs
, start
, &cur_bytes
, m
);
1245 if (ret
== -EAGAIN
) {
1246 /* Currently handle_dependencies() doesn't yield if we already had
1247 * an allocation. If it did, we would have to clean up the L2Meta
1248 * structs before starting over. */
1251 } else if (ret
< 0) {
1253 } else if (cur_bytes
== 0) {
1256 /* handle_dependencies() may have decreased cur_bytes (shortened
1257 * the allocations below) so that the next dependency is processed
1258 * correctly during the next loop iteration. */
1262 * 2. Count contiguous COPIED clusters.
1264 ret
= handle_copied(bs
, start
, &cluster_offset
, &cur_bytes
, m
);
1269 } else if (cur_bytes
== 0) {
1274 * 3. If the request still hasn't completed, allocate new clusters,
1275 * considering any cluster_offset of steps 1c or 2.
1277 ret
= handle_alloc(bs
, start
, &cluster_offset
, &cur_bytes
, m
);
1283 assert(cur_bytes
== 0);
1288 *num
-= remaining
>> BDRV_SECTOR_BITS
;
1290 assert(*host_offset
!= 0);
1295 static int decompress_buffer(uint8_t *out_buf
, int out_buf_size
,
1296 const uint8_t *buf
, int buf_size
)
1298 z_stream strm1
, *strm
= &strm1
;
1301 memset(strm
, 0, sizeof(*strm
));
1303 strm
->next_in
= (uint8_t *)buf
;
1304 strm
->avail_in
= buf_size
;
1305 strm
->next_out
= out_buf
;
1306 strm
->avail_out
= out_buf_size
;
1308 ret
= inflateInit2(strm
, -12);
1311 ret
= inflate(strm
, Z_FINISH
);
1312 out_len
= strm
->next_out
- out_buf
;
1313 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) ||
1314 out_len
!= out_buf_size
) {
1322 int qcow2_decompress_cluster(BlockDriverState
*bs
, uint64_t cluster_offset
)
1324 BDRVQcowState
*s
= bs
->opaque
;
1325 int ret
, csize
, nb_csectors
, sector_offset
;
1328 coffset
= cluster_offset
& s
->cluster_offset_mask
;
1329 if (s
->cluster_cache_offset
!= coffset
) {
1330 nb_csectors
= ((cluster_offset
>> s
->csize_shift
) & s
->csize_mask
) + 1;
1331 sector_offset
= coffset
& 511;
1332 csize
= nb_csectors
* 512 - sector_offset
;
1333 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
1334 ret
= bdrv_read(bs
->file
, coffset
>> 9, s
->cluster_data
, nb_csectors
);
1338 if (decompress_buffer(s
->cluster_cache
, s
->cluster_size
,
1339 s
->cluster_data
+ sector_offset
, csize
) < 0) {
1342 s
->cluster_cache_offset
= coffset
;
1348 * This discards as many clusters of nb_clusters as possible at once (i.e.
1349 * all clusters in the same L2 table) and returns the number of discarded
1352 static int discard_single_l2(BlockDriverState
*bs
, uint64_t offset
,
1353 unsigned int nb_clusters
, enum qcow2_discard_type type
)
1355 BDRVQcowState
*s
= bs
->opaque
;
1361 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_index
);
1366 /* Limit nb_clusters to one L2 table */
1367 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
1369 for (i
= 0; i
< nb_clusters
; i
++) {
1370 uint64_t old_l2_entry
;
1372 old_l2_entry
= be64_to_cpu(l2_table
[l2_index
+ i
]);
1375 * Make sure that a discarded area reads back as zeroes for v3 images
1376 * (we cannot do it for v2 without actually writing a zero-filled
1377 * buffer). We can skip the operation if the cluster is already marked
1378 * as zero, or if it's unallocated and we don't have a backing file.
1380 * TODO We might want to use bdrv_get_block_status(bs) here, but we're
1381 * holding s->lock, so that doesn't work today.
1383 switch (qcow2_get_cluster_type(old_l2_entry
)) {
1384 case QCOW2_CLUSTER_UNALLOCATED
:
1385 if (!bs
->backing_hd
) {
1390 case QCOW2_CLUSTER_ZERO
:
1393 case QCOW2_CLUSTER_NORMAL
:
1394 case QCOW2_CLUSTER_COMPRESSED
:
1401 /* First remove L2 entries */
1402 qcow2_cache_entry_mark_dirty(s
->l2_table_cache
, l2_table
);
1403 if (s
->qcow_version
>= 3) {
1404 l2_table
[l2_index
+ i
] = cpu_to_be64(QCOW_OFLAG_ZERO
);
1406 l2_table
[l2_index
+ i
] = cpu_to_be64(0);
1409 /* Then decrease the refcount */
1410 qcow2_free_any_clusters(bs
, old_l2_entry
, 1, type
);
1413 ret
= qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
1421 int qcow2_discard_clusters(BlockDriverState
*bs
, uint64_t offset
,
1422 int nb_sectors
, enum qcow2_discard_type type
)
1424 BDRVQcowState
*s
= bs
->opaque
;
1425 uint64_t end_offset
;
1426 unsigned int nb_clusters
;
1429 end_offset
= offset
+ (nb_sectors
<< BDRV_SECTOR_BITS
);
1431 /* Round start up and end down */
1432 offset
= align_offset(offset
, s
->cluster_size
);
1433 end_offset
= start_of_cluster(s
, end_offset
);
1435 if (offset
> end_offset
) {
1439 nb_clusters
= size_to_clusters(s
, end_offset
- offset
);
1441 s
->cache_discards
= true;
1443 /* Each L2 table is handled by its own loop iteration */
1444 while (nb_clusters
> 0) {
1445 ret
= discard_single_l2(bs
, offset
, nb_clusters
, type
);
1451 offset
+= (ret
* s
->cluster_size
);
1456 s
->cache_discards
= false;
1457 qcow2_process_discards(bs
, ret
);
1463 * This zeroes as many clusters of nb_clusters as possible at once (i.e.
1464 * all clusters in the same L2 table) and returns the number of zeroed
1467 static int zero_single_l2(BlockDriverState
*bs
, uint64_t offset
,
1468 unsigned int nb_clusters
)
1470 BDRVQcowState
*s
= bs
->opaque
;
1476 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_index
);
1481 /* Limit nb_clusters to one L2 table */
1482 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
1484 for (i
= 0; i
< nb_clusters
; i
++) {
1485 uint64_t old_offset
;
1487 old_offset
= be64_to_cpu(l2_table
[l2_index
+ i
]);
1489 /* Update L2 entries */
1490 qcow2_cache_entry_mark_dirty(s
->l2_table_cache
, l2_table
);
1491 if (old_offset
& QCOW_OFLAG_COMPRESSED
) {
1492 l2_table
[l2_index
+ i
] = cpu_to_be64(QCOW_OFLAG_ZERO
);
1493 qcow2_free_any_clusters(bs
, old_offset
, 1, QCOW2_DISCARD_REQUEST
);
1495 l2_table
[l2_index
+ i
] |= cpu_to_be64(QCOW_OFLAG_ZERO
);
1499 ret
= qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
1507 int qcow2_zero_clusters(BlockDriverState
*bs
, uint64_t offset
, int nb_sectors
)
1509 BDRVQcowState
*s
= bs
->opaque
;
1510 unsigned int nb_clusters
;
1513 /* The zero flag is only supported by version 3 and newer */
1514 if (s
->qcow_version
< 3) {
1518 /* Each L2 table is handled by its own loop iteration */
1519 nb_clusters
= size_to_clusters(s
, nb_sectors
<< BDRV_SECTOR_BITS
);
1521 s
->cache_discards
= true;
1523 while (nb_clusters
> 0) {
1524 ret
= zero_single_l2(bs
, offset
, nb_clusters
);
1530 offset
+= (ret
* s
->cluster_size
);
1535 s
->cache_discards
= false;
1536 qcow2_process_discards(bs
, ret
);
1542 * Expands all zero clusters in a specific L1 table (or deallocates them, for
1543 * non-backed non-pre-allocated zero clusters).
1545 * expanded_clusters is a bitmap where every bit corresponds to one cluster in
1546 * the image file; a bit gets set if the corresponding cluster has been used for
1547 * zero expansion (i.e., has been filled with zeroes and is referenced from an
1548 * L2 table). nb_clusters contains the total cluster count of the image file,
1549 * i.e., the number of bits in expanded_clusters.
1551 static int expand_zero_clusters_in_l1(BlockDriverState
*bs
, uint64_t *l1_table
,
1552 int l1_size
, uint8_t **expanded_clusters
,
1553 uint64_t *nb_clusters
)
1555 BDRVQcowState
*s
= bs
->opaque
;
1556 bool is_active_l1
= (l1_table
== s
->l1_table
);
1557 uint64_t *l2_table
= NULL
;
1561 if (!is_active_l1
) {
1562 /* inactive L2 tables require a buffer to be stored in when loading
1564 l2_table
= qemu_blockalign(bs
, s
->cluster_size
);
1567 for (i
= 0; i
< l1_size
; i
++) {
1568 uint64_t l2_offset
= l1_table
[i
] & L1E_OFFSET_MASK
;
1569 bool l2_dirty
= false;
1577 /* get active L2 tables from cache */
1578 ret
= qcow2_cache_get(bs
, s
->l2_table_cache
, l2_offset
,
1579 (void **)&l2_table
);
1581 /* load inactive L2 tables from disk */
1582 ret
= bdrv_read(bs
->file
, l2_offset
/ BDRV_SECTOR_SIZE
,
1583 (void *)l2_table
, s
->cluster_sectors
);
1589 for (j
= 0; j
< s
->l2_size
; j
++) {
1590 uint64_t l2_entry
= be64_to_cpu(l2_table
[j
]);
1591 int64_t offset
= l2_entry
& L2E_OFFSET_MASK
, cluster_index
;
1592 int cluster_type
= qcow2_get_cluster_type(l2_entry
);
1593 bool preallocated
= offset
!= 0;
1595 if (cluster_type
== QCOW2_CLUSTER_NORMAL
) {
1596 cluster_index
= offset
>> s
->cluster_bits
;
1597 assert((cluster_index
>= 0) && (cluster_index
< *nb_clusters
));
1598 if ((*expanded_clusters
)[cluster_index
/ 8] &
1599 (1 << (cluster_index
% 8))) {
1600 /* Probably a shared L2 table; this cluster was a zero
1601 * cluster which has been expanded, its refcount
1602 * therefore most likely requires an update. */
1603 ret
= qcow2_update_cluster_refcount(bs
, cluster_index
, 1,
1604 QCOW2_DISCARD_NEVER
);
1608 /* Since we just increased the refcount, the COPIED flag may
1609 * no longer be set. */
1610 l2_table
[j
] = cpu_to_be64(l2_entry
& ~QCOW_OFLAG_COPIED
);
1615 else if (qcow2_get_cluster_type(l2_entry
) != QCOW2_CLUSTER_ZERO
) {
1619 if (!preallocated
) {
1620 if (!bs
->backing_hd
) {
1621 /* not backed; therefore we can simply deallocate the
1628 offset
= qcow2_alloc_clusters(bs
, s
->cluster_size
);
1635 ret
= qcow2_pre_write_overlap_check(bs
, 0, offset
, s
->cluster_size
);
1637 if (!preallocated
) {
1638 qcow2_free_clusters(bs
, offset
, s
->cluster_size
,
1639 QCOW2_DISCARD_ALWAYS
);
1644 ret
= bdrv_write_zeroes(bs
->file
, offset
/ BDRV_SECTOR_SIZE
,
1645 s
->cluster_sectors
, 0);
1647 if (!preallocated
) {
1648 qcow2_free_clusters(bs
, offset
, s
->cluster_size
,
1649 QCOW2_DISCARD_ALWAYS
);
1654 l2_table
[j
] = cpu_to_be64(offset
| QCOW_OFLAG_COPIED
);
1657 cluster_index
= offset
>> s
->cluster_bits
;
1659 if (cluster_index
>= *nb_clusters
) {
1660 uint64_t old_bitmap_size
= (*nb_clusters
+ 7) / 8;
1661 uint64_t new_bitmap_size
;
1662 /* The offset may lie beyond the old end of the underlying image
1663 * file for growable files only */
1664 assert(bs
->file
->growable
);
1665 *nb_clusters
= size_to_clusters(s
, bs
->file
->total_sectors
*
1667 new_bitmap_size
= (*nb_clusters
+ 7) / 8;
1668 *expanded_clusters
= g_realloc(*expanded_clusters
,
1670 /* clear the newly allocated space */
1671 memset(&(*expanded_clusters
)[old_bitmap_size
], 0,
1672 new_bitmap_size
- old_bitmap_size
);
1675 assert((cluster_index
>= 0) && (cluster_index
< *nb_clusters
));
1676 (*expanded_clusters
)[cluster_index
/ 8] |= 1 << (cluster_index
% 8);
1681 qcow2_cache_entry_mark_dirty(s
->l2_table_cache
, l2_table
);
1682 qcow2_cache_depends_on_flush(s
->l2_table_cache
);
1684 ret
= qcow2_cache_put(bs
, s
->l2_table_cache
, (void **)&l2_table
);
1691 ret
= qcow2_pre_write_overlap_check(bs
,
1692 QCOW2_OL_INACTIVE_L2
| QCOW2_OL_ACTIVE_L2
, l2_offset
,
1698 ret
= bdrv_write(bs
->file
, l2_offset
/ BDRV_SECTOR_SIZE
,
1699 (void *)l2_table
, s
->cluster_sectors
);
1711 if (!is_active_l1
) {
1712 qemu_vfree(l2_table
);
1715 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **)&l2_table
);
1717 ret
= qcow2_cache_put(bs
, s
->l2_table_cache
,
1718 (void **)&l2_table
);
1726 * For backed images, expands all zero clusters on the image. For non-backed
1727 * images, deallocates all non-pre-allocated zero clusters (and claims the
1728 * allocation for pre-allocated ones). This is important for downgrading to a
1729 * qcow2 version which doesn't yet support metadata zero clusters.
1731 int qcow2_expand_zero_clusters(BlockDriverState
*bs
)
1733 BDRVQcowState
*s
= bs
->opaque
;
1734 uint64_t *l1_table
= NULL
;
1735 uint64_t nb_clusters
;
1736 uint8_t *expanded_clusters
;
1740 nb_clusters
= size_to_clusters(s
, bs
->file
->total_sectors
*
1742 expanded_clusters
= g_malloc0((nb_clusters
+ 7) / 8);
1744 ret
= expand_zero_clusters_in_l1(bs
, s
->l1_table
, s
->l1_size
,
1745 &expanded_clusters
, &nb_clusters
);
1750 /* Inactive L1 tables may point to active L2 tables - therefore it is
1751 * necessary to flush the L2 table cache before trying to access the L2
1752 * tables pointed to by inactive L1 entries (else we might try to expand
1753 * zero clusters that have already been expanded); furthermore, it is also
1754 * necessary to empty the L2 table cache, since it may contain tables which
1755 * are now going to be modified directly on disk, bypassing the cache.
1756 * qcow2_cache_empty() does both for us. */
1757 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
1762 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
1763 int l1_sectors
= (s
->snapshots
[i
].l1_size
* sizeof(uint64_t) +
1764 BDRV_SECTOR_SIZE
- 1) / BDRV_SECTOR_SIZE
;
1766 l1_table
= g_realloc(l1_table
, l1_sectors
* BDRV_SECTOR_SIZE
);
1768 ret
= bdrv_read(bs
->file
, s
->snapshots
[i
].l1_table_offset
/
1769 BDRV_SECTOR_SIZE
, (void *)l1_table
, l1_sectors
);
1774 for (j
= 0; j
< s
->snapshots
[i
].l1_size
; j
++) {
1775 be64_to_cpus(&l1_table
[j
]);
1778 ret
= expand_zero_clusters_in_l1(bs
, l1_table
, s
->snapshots
[i
].l1_size
,
1779 &expanded_clusters
, &nb_clusters
);
1788 g_free(expanded_clusters
);