2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu-common.h"
26 #include "block/block_int.h"
27 #include "block/qcow2.h"
28 #include "qemu/range.h"
29 #include "qapi/qmp/types.h"
31 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, int64_t size
);
32 static int QEMU_WARN_UNUSED_RESULT
update_refcount(BlockDriverState
*bs
,
33 int64_t offset
, int64_t length
,
34 int addend
, enum qcow2_discard_type type
);
37 /*********************************************************/
38 /* refcount handling */
40 int qcow2_refcount_init(BlockDriverState
*bs
)
42 BDRVQcowState
*s
= bs
->opaque
;
43 unsigned int refcount_table_size2
, i
;
46 assert(s
->refcount_table_size
<= INT_MAX
/ sizeof(uint64_t));
47 refcount_table_size2
= s
->refcount_table_size
* sizeof(uint64_t);
48 s
->refcount_table
= g_malloc(refcount_table_size2
);
49 if (s
->refcount_table_size
> 0) {
50 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_LOAD
);
51 ret
= bdrv_pread(bs
->file
, s
->refcount_table_offset
,
52 s
->refcount_table
, refcount_table_size2
);
53 if (ret
!= refcount_table_size2
)
55 for(i
= 0; i
< s
->refcount_table_size
; i
++)
56 be64_to_cpus(&s
->refcount_table
[i
]);
63 void qcow2_refcount_close(BlockDriverState
*bs
)
65 BDRVQcowState
*s
= bs
->opaque
;
66 g_free(s
->refcount_table
);
70 static int load_refcount_block(BlockDriverState
*bs
,
71 int64_t refcount_block_offset
,
72 void **refcount_block
)
74 BDRVQcowState
*s
= bs
->opaque
;
77 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_LOAD
);
78 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, refcount_block_offset
,
85 * Returns the refcount of the cluster given by its index. Any non-negative
86 * return value is the refcount of the cluster, negative values are -errno
87 * and indicate an error.
89 static int get_refcount(BlockDriverState
*bs
, int64_t cluster_index
)
91 BDRVQcowState
*s
= bs
->opaque
;
92 int refcount_table_index
, block_index
;
93 int64_t refcount_block_offset
;
95 uint16_t *refcount_block
;
98 refcount_table_index
= cluster_index
>> (s
->cluster_bits
- REFCOUNT_SHIFT
);
99 if (refcount_table_index
>= s
->refcount_table_size
)
101 refcount_block_offset
=
102 s
->refcount_table
[refcount_table_index
] & REFT_OFFSET_MASK
;
103 if (!refcount_block_offset
)
106 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, refcount_block_offset
,
107 (void**) &refcount_block
);
112 block_index
= cluster_index
&
113 ((1 << (s
->cluster_bits
- REFCOUNT_SHIFT
)) - 1);
114 refcount
= be16_to_cpu(refcount_block
[block_index
]);
116 ret
= qcow2_cache_put(bs
, s
->refcount_block_cache
,
117 (void**) &refcount_block
);
126 * Rounds the refcount table size up to avoid growing the table for each single
127 * refcount block that is allocated.
129 static unsigned int next_refcount_table_size(BDRVQcowState
*s
,
130 unsigned int min_size
)
132 unsigned int min_clusters
= (min_size
>> (s
->cluster_bits
- 3)) + 1;
133 unsigned int refcount_table_clusters
=
134 MAX(1, s
->refcount_table_size
>> (s
->cluster_bits
- 3));
136 while (min_clusters
> refcount_table_clusters
) {
137 refcount_table_clusters
= (refcount_table_clusters
* 3 + 1) / 2;
140 return refcount_table_clusters
<< (s
->cluster_bits
- 3);
144 /* Checks if two offsets are described by the same refcount block */
145 static int in_same_refcount_block(BDRVQcowState
*s
, uint64_t offset_a
,
148 uint64_t block_a
= offset_a
>> (2 * s
->cluster_bits
- REFCOUNT_SHIFT
);
149 uint64_t block_b
= offset_b
>> (2 * s
->cluster_bits
- REFCOUNT_SHIFT
);
151 return (block_a
== block_b
);
155 * Loads a refcount block. If it doesn't exist yet, it is allocated first
156 * (including growing the refcount table if needed).
158 * Returns 0 on success or -errno in error case
160 static int alloc_refcount_block(BlockDriverState
*bs
,
161 int64_t cluster_index
, uint16_t **refcount_block
)
163 BDRVQcowState
*s
= bs
->opaque
;
164 unsigned int refcount_table_index
;
167 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
169 /* Find the refcount block for the given cluster */
170 refcount_table_index
= cluster_index
>> (s
->cluster_bits
- REFCOUNT_SHIFT
);
172 if (refcount_table_index
< s
->refcount_table_size
) {
174 uint64_t refcount_block_offset
=
175 s
->refcount_table
[refcount_table_index
] & REFT_OFFSET_MASK
;
177 /* If it's already there, we're done */
178 if (refcount_block_offset
) {
179 return load_refcount_block(bs
, refcount_block_offset
,
180 (void**) refcount_block
);
185 * If we came here, we need to allocate something. Something is at least
186 * a cluster for the new refcount block. It may also include a new refcount
187 * table if the old refcount table is too small.
189 * Note that allocating clusters here needs some special care:
191 * - We can't use the normal qcow2_alloc_clusters(), it would try to
192 * increase the refcount and very likely we would end up with an endless
193 * recursion. Instead we must place the refcount blocks in a way that
194 * they can describe them themselves.
196 * - We need to consider that at this point we are inside update_refcounts
197 * and potentially doing an initial refcount increase. This means that
198 * some clusters have already been allocated by the caller, but their
199 * refcount isn't accurate yet. If we allocate clusters for metadata, we
200 * need to return -EAGAIN to signal the caller that it needs to restart
201 * the search for free clusters.
203 * - alloc_clusters_noref and qcow2_free_clusters may load a different
204 * refcount block into the cache
207 *refcount_block
= NULL
;
209 /* We write to the refcount table, so we might depend on L2 tables */
210 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
215 /* Allocate the refcount block itself and mark it as used */
216 int64_t new_block
= alloc_clusters_noref(bs
, s
->cluster_size
);
222 fprintf(stderr
, "qcow2: Allocate refcount block %d for %" PRIx64
224 refcount_table_index
, cluster_index
<< s
->cluster_bits
, new_block
);
227 if (in_same_refcount_block(s
, new_block
, cluster_index
<< s
->cluster_bits
)) {
228 /* Zero the new refcount block before updating it */
229 ret
= qcow2_cache_get_empty(bs
, s
->refcount_block_cache
, new_block
,
230 (void**) refcount_block
);
235 memset(*refcount_block
, 0, s
->cluster_size
);
237 /* The block describes itself, need to update the cache */
238 int block_index
= (new_block
>> s
->cluster_bits
) &
239 ((1 << (s
->cluster_bits
- REFCOUNT_SHIFT
)) - 1);
240 (*refcount_block
)[block_index
] = cpu_to_be16(1);
242 /* Described somewhere else. This can recurse at most twice before we
243 * arrive at a block that describes itself. */
244 ret
= update_refcount(bs
, new_block
, s
->cluster_size
, 1,
245 QCOW2_DISCARD_NEVER
);
250 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
255 /* Initialize the new refcount block only after updating its refcount,
256 * update_refcount uses the refcount cache itself */
257 ret
= qcow2_cache_get_empty(bs
, s
->refcount_block_cache
, new_block
,
258 (void**) refcount_block
);
263 memset(*refcount_block
, 0, s
->cluster_size
);
266 /* Now the new refcount block needs to be written to disk */
267 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE
);
268 qcow2_cache_entry_mark_dirty(s
->refcount_block_cache
, *refcount_block
);
269 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
274 /* If the refcount table is big enough, just hook the block up there */
275 if (refcount_table_index
< s
->refcount_table_size
) {
276 uint64_t data64
= cpu_to_be64(new_block
);
277 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_HOOKUP
);
278 ret
= bdrv_pwrite_sync(bs
->file
,
279 s
->refcount_table_offset
+ refcount_table_index
* sizeof(uint64_t),
280 &data64
, sizeof(data64
));
285 s
->refcount_table
[refcount_table_index
] = new_block
;
287 /* The new refcount block may be where the caller intended to put its
288 * data, so let it restart the search. */
292 ret
= qcow2_cache_put(bs
, s
->refcount_block_cache
, (void**) refcount_block
);
298 * If we come here, we need to grow the refcount table. Again, a new
299 * refcount table needs some space and we can't simply allocate to avoid
302 * Therefore let's grab new refcount blocks at the end of the image, which
303 * will describe themselves and the new refcount table. This way we can
304 * reference them only in the new table and do the switch to the new
305 * refcount table at once without producing an inconsistent state in
308 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_GROW
);
310 /* Calculate the number of refcount blocks needed so far */
311 uint64_t refcount_block_clusters
= 1 << (s
->cluster_bits
- REFCOUNT_SHIFT
);
312 uint64_t blocks_used
= DIV_ROUND_UP(cluster_index
, refcount_block_clusters
);
314 /* And now we need at least one block more for the new metadata */
315 uint64_t table_size
= next_refcount_table_size(s
, blocks_used
+ 1);
316 uint64_t last_table_size
;
317 uint64_t blocks_clusters
;
319 uint64_t table_clusters
=
320 size_to_clusters(s
, table_size
* sizeof(uint64_t));
321 blocks_clusters
= 1 +
322 ((table_clusters
+ refcount_block_clusters
- 1)
323 / refcount_block_clusters
);
324 uint64_t meta_clusters
= table_clusters
+ blocks_clusters
;
326 last_table_size
= table_size
;
327 table_size
= next_refcount_table_size(s
, blocks_used
+
328 ((meta_clusters
+ refcount_block_clusters
- 1)
329 / refcount_block_clusters
));
331 } while (last_table_size
!= table_size
);
334 fprintf(stderr
, "qcow2: Grow refcount table %" PRId32
" => %" PRId64
"\n",
335 s
->refcount_table_size
, table_size
);
338 /* Create the new refcount table and blocks */
339 uint64_t meta_offset
= (blocks_used
* refcount_block_clusters
) *
341 uint64_t table_offset
= meta_offset
+ blocks_clusters
* s
->cluster_size
;
342 uint16_t *new_blocks
= g_malloc0(blocks_clusters
* s
->cluster_size
);
343 uint64_t *new_table
= g_malloc0(table_size
* sizeof(uint64_t));
345 /* Fill the new refcount table */
346 memcpy(new_table
, s
->refcount_table
,
347 s
->refcount_table_size
* sizeof(uint64_t));
348 new_table
[refcount_table_index
] = new_block
;
351 for (i
= 0; i
< blocks_clusters
; i
++) {
352 new_table
[blocks_used
+ i
] = meta_offset
+ (i
* s
->cluster_size
);
355 /* Fill the refcount blocks */
356 uint64_t table_clusters
= size_to_clusters(s
, table_size
* sizeof(uint64_t));
358 for (i
= 0; i
< table_clusters
+ blocks_clusters
; i
++) {
359 new_blocks
[block
++] = cpu_to_be16(1);
362 /* Write refcount blocks to disk */
363 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS
);
364 ret
= bdrv_pwrite_sync(bs
->file
, meta_offset
, new_blocks
,
365 blocks_clusters
* s
->cluster_size
);
371 /* Write refcount table to disk */
372 for(i
= 0; i
< table_size
; i
++) {
373 cpu_to_be64s(&new_table
[i
]);
376 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE
);
377 ret
= bdrv_pwrite_sync(bs
->file
, table_offset
, new_table
,
378 table_size
* sizeof(uint64_t));
383 for(i
= 0; i
< table_size
; i
++) {
384 be64_to_cpus(&new_table
[i
]);
387 /* Hook up the new refcount table in the qcow2 header */
389 cpu_to_be64w((uint64_t*)data
, table_offset
);
390 cpu_to_be32w((uint32_t*)(data
+ 8), table_clusters
);
391 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE
);
392 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, refcount_table_offset
),
398 /* And switch it in memory */
399 uint64_t old_table_offset
= s
->refcount_table_offset
;
400 uint64_t old_table_size
= s
->refcount_table_size
;
402 g_free(s
->refcount_table
);
403 s
->refcount_table
= new_table
;
404 s
->refcount_table_size
= table_size
;
405 s
->refcount_table_offset
= table_offset
;
407 /* Free old table. */
408 qcow2_free_clusters(bs
, old_table_offset
, old_table_size
* sizeof(uint64_t),
409 QCOW2_DISCARD_OTHER
);
411 ret
= load_refcount_block(bs
, new_block
, (void**) refcount_block
);
416 /* If we were trying to do the initial refcount update for some cluster
417 * allocation, we might have used the same clusters to store newly
418 * allocated metadata. Make the caller search some new space. */
424 if (*refcount_block
!= NULL
) {
425 qcow2_cache_put(bs
, s
->refcount_block_cache
, (void**) refcount_block
);
430 void qcow2_process_discards(BlockDriverState
*bs
, int ret
)
432 BDRVQcowState
*s
= bs
->opaque
;
433 Qcow2DiscardRegion
*d
, *next
;
435 QTAILQ_FOREACH_SAFE(d
, &s
->discards
, next
, next
) {
436 QTAILQ_REMOVE(&s
->discards
, d
, next
);
438 /* Discard is optional, ignore the return value */
440 bdrv_discard(bs
->file
,
441 d
->offset
>> BDRV_SECTOR_BITS
,
442 d
->bytes
>> BDRV_SECTOR_BITS
);
449 static void update_refcount_discard(BlockDriverState
*bs
,
450 uint64_t offset
, uint64_t length
)
452 BDRVQcowState
*s
= bs
->opaque
;
453 Qcow2DiscardRegion
*d
, *p
, *next
;
455 QTAILQ_FOREACH(d
, &s
->discards
, next
) {
456 uint64_t new_start
= MIN(offset
, d
->offset
);
457 uint64_t new_end
= MAX(offset
+ length
, d
->offset
+ d
->bytes
);
459 if (new_end
- new_start
<= length
+ d
->bytes
) {
460 /* There can't be any overlap, areas ending up here have no
461 * references any more and therefore shouldn't get freed another
463 assert(d
->bytes
+ length
== new_end
- new_start
);
464 d
->offset
= new_start
;
465 d
->bytes
= new_end
- new_start
;
470 d
= g_malloc(sizeof(*d
));
471 *d
= (Qcow2DiscardRegion
) {
476 QTAILQ_INSERT_TAIL(&s
->discards
, d
, next
);
479 /* Merge discard requests if they are adjacent now */
480 QTAILQ_FOREACH_SAFE(p
, &s
->discards
, next
, next
) {
482 || p
->offset
> d
->offset
+ d
->bytes
483 || d
->offset
> p
->offset
+ p
->bytes
)
488 /* Still no overlap possible */
489 assert(p
->offset
== d
->offset
+ d
->bytes
490 || d
->offset
== p
->offset
+ p
->bytes
);
492 QTAILQ_REMOVE(&s
->discards
, p
, next
);
493 d
->offset
= MIN(d
->offset
, p
->offset
);
494 d
->bytes
+= p
->bytes
;
498 /* XXX: cache several refcount block clusters ? */
499 static int QEMU_WARN_UNUSED_RESULT
update_refcount(BlockDriverState
*bs
,
500 int64_t offset
, int64_t length
, int addend
, enum qcow2_discard_type type
)
502 BDRVQcowState
*s
= bs
->opaque
;
503 int64_t start
, last
, cluster_offset
;
504 uint16_t *refcount_block
= NULL
;
505 int64_t old_table_index
= -1;
509 fprintf(stderr
, "update_refcount: offset=%" PRId64
" size=%" PRId64
" addend=%d\n",
510 offset
, length
, addend
);
514 } else if (length
== 0) {
519 qcow2_cache_set_dependency(bs
, s
->refcount_block_cache
,
523 start
= start_of_cluster(s
, offset
);
524 last
= start_of_cluster(s
, offset
+ length
- 1);
525 for(cluster_offset
= start
; cluster_offset
<= last
;
526 cluster_offset
+= s
->cluster_size
)
528 int block_index
, refcount
;
529 int64_t cluster_index
= cluster_offset
>> s
->cluster_bits
;
530 int64_t table_index
=
531 cluster_index
>> (s
->cluster_bits
- REFCOUNT_SHIFT
);
533 /* Load the refcount block and allocate it if needed */
534 if (table_index
!= old_table_index
) {
535 if (refcount_block
) {
536 ret
= qcow2_cache_put(bs
, s
->refcount_block_cache
,
537 (void**) &refcount_block
);
543 ret
= alloc_refcount_block(bs
, cluster_index
, &refcount_block
);
548 old_table_index
= table_index
;
550 qcow2_cache_entry_mark_dirty(s
->refcount_block_cache
, refcount_block
);
552 /* we can update the count and save it */
553 block_index
= cluster_index
&
554 ((1 << (s
->cluster_bits
- REFCOUNT_SHIFT
)) - 1);
556 refcount
= be16_to_cpu(refcount_block
[block_index
]);
558 if (refcount
< 0 || refcount
> 0xffff) {
562 if (refcount
== 0 && cluster_index
< s
->free_cluster_index
) {
563 s
->free_cluster_index
= cluster_index
;
565 refcount_block
[block_index
] = cpu_to_be16(refcount
);
567 if (refcount
== 0 && s
->discard_passthrough
[type
]) {
568 update_refcount_discard(bs
, cluster_offset
, s
->cluster_size
);
574 if (!s
->cache_discards
) {
575 qcow2_process_discards(bs
, ret
);
578 /* Write last changed block to disk */
579 if (refcount_block
) {
581 wret
= qcow2_cache_put(bs
, s
->refcount_block_cache
,
582 (void**) &refcount_block
);
584 return ret
< 0 ? ret
: wret
;
589 * Try do undo any updates if an error is returned (This may succeed in
590 * some cases like ENOSPC for allocating a new refcount block)
594 dummy
= update_refcount(bs
, offset
, cluster_offset
- offset
, -addend
,
595 QCOW2_DISCARD_NEVER
);
603 * Increases or decreases the refcount of a given cluster by one.
604 * addend must be 1 or -1.
606 * If the return value is non-negative, it is the new refcount of the cluster.
607 * If it is negative, it is -errno and indicates an error.
609 int qcow2_update_cluster_refcount(BlockDriverState
*bs
,
610 int64_t cluster_index
,
612 enum qcow2_discard_type type
)
614 BDRVQcowState
*s
= bs
->opaque
;
617 ret
= update_refcount(bs
, cluster_index
<< s
->cluster_bits
, 1, addend
,
623 return get_refcount(bs
, cluster_index
);
628 /*********************************************************/
629 /* cluster allocation functions */
633 /* return < 0 if error */
634 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, int64_t size
)
636 BDRVQcowState
*s
= bs
->opaque
;
637 int i
, nb_clusters
, refcount
;
639 nb_clusters
= size_to_clusters(s
, size
);
641 for(i
= 0; i
< nb_clusters
; i
++) {
642 int64_t next_cluster_index
= s
->free_cluster_index
++;
643 refcount
= get_refcount(bs
, next_cluster_index
);
647 } else if (refcount
!= 0) {
652 fprintf(stderr
, "alloc_clusters: size=%" PRId64
" -> %" PRId64
"\n",
654 (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
);
656 return (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
;
659 int64_t qcow2_alloc_clusters(BlockDriverState
*bs
, int64_t size
)
664 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC
);
666 offset
= alloc_clusters_noref(bs
, size
);
671 ret
= update_refcount(bs
, offset
, size
, 1, QCOW2_DISCARD_NEVER
);
672 } while (ret
== -EAGAIN
);
681 int qcow2_alloc_clusters_at(BlockDriverState
*bs
, uint64_t offset
,
684 BDRVQcowState
*s
= bs
->opaque
;
685 uint64_t cluster_index
;
689 assert(nb_clusters
>= 0);
690 if (nb_clusters
== 0) {
695 /* Check how many clusters there are free */
696 cluster_index
= offset
>> s
->cluster_bits
;
697 for(i
= 0; i
< nb_clusters
; i
++) {
698 refcount
= get_refcount(bs
, cluster_index
++);
702 } else if (refcount
!= 0) {
707 /* And then allocate them */
708 ret
= update_refcount(bs
, offset
, i
<< s
->cluster_bits
, 1,
709 QCOW2_DISCARD_NEVER
);
710 } while (ret
== -EAGAIN
);
719 /* only used to allocate compressed sectors. We try to allocate
720 contiguous sectors. size must be <= cluster_size */
721 int64_t qcow2_alloc_bytes(BlockDriverState
*bs
, int size
)
723 BDRVQcowState
*s
= bs
->opaque
;
724 int64_t offset
, cluster_offset
;
727 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC_BYTES
);
728 assert(size
> 0 && size
<= s
->cluster_size
);
729 if (s
->free_byte_offset
== 0) {
730 offset
= qcow2_alloc_clusters(bs
, s
->cluster_size
);
734 s
->free_byte_offset
= offset
;
737 free_in_cluster
= s
->cluster_size
-
738 offset_into_cluster(s
, s
->free_byte_offset
);
739 if (size
<= free_in_cluster
) {
740 /* enough space in current cluster */
741 offset
= s
->free_byte_offset
;
742 s
->free_byte_offset
+= size
;
743 free_in_cluster
-= size
;
744 if (free_in_cluster
== 0)
745 s
->free_byte_offset
= 0;
746 if (offset_into_cluster(s
, offset
) != 0)
747 qcow2_update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, 1,
748 QCOW2_DISCARD_NEVER
);
750 offset
= qcow2_alloc_clusters(bs
, s
->cluster_size
);
754 cluster_offset
= start_of_cluster(s
, s
->free_byte_offset
);
755 if ((cluster_offset
+ s
->cluster_size
) == offset
) {
756 /* we are lucky: contiguous data */
757 offset
= s
->free_byte_offset
;
758 qcow2_update_cluster_refcount(bs
, offset
>> s
->cluster_bits
, 1,
759 QCOW2_DISCARD_NEVER
);
760 s
->free_byte_offset
+= size
;
762 s
->free_byte_offset
= offset
;
767 /* The cluster refcount was incremented, either by qcow2_alloc_clusters()
768 * or explicitly by qcow2_update_cluster_refcount(). Refcount blocks must
769 * be flushed before the caller's L2 table updates.
771 qcow2_cache_set_dependency(bs
, s
->l2_table_cache
, s
->refcount_block_cache
);
775 void qcow2_free_clusters(BlockDriverState
*bs
,
776 int64_t offset
, int64_t size
,
777 enum qcow2_discard_type type
)
781 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_FREE
);
782 ret
= update_refcount(bs
, offset
, size
, -1, type
);
784 fprintf(stderr
, "qcow2_free_clusters failed: %s\n", strerror(-ret
));
785 /* TODO Remember the clusters to free them later and avoid leaking */
790 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
791 * normal cluster, compressed cluster, etc.)
793 void qcow2_free_any_clusters(BlockDriverState
*bs
, uint64_t l2_entry
,
794 int nb_clusters
, enum qcow2_discard_type type
)
796 BDRVQcowState
*s
= bs
->opaque
;
798 switch (qcow2_get_cluster_type(l2_entry
)) {
799 case QCOW2_CLUSTER_COMPRESSED
:
802 nb_csectors
= ((l2_entry
>> s
->csize_shift
) &
804 qcow2_free_clusters(bs
,
805 (l2_entry
& s
->cluster_offset_mask
) & ~511,
806 nb_csectors
* 512, type
);
809 case QCOW2_CLUSTER_NORMAL
:
810 case QCOW2_CLUSTER_ZERO
:
811 if (l2_entry
& L2E_OFFSET_MASK
) {
812 qcow2_free_clusters(bs
, l2_entry
& L2E_OFFSET_MASK
,
813 nb_clusters
<< s
->cluster_bits
, type
);
816 case QCOW2_CLUSTER_UNALLOCATED
:
825 /*********************************************************/
826 /* snapshots and image creation */
830 /* update the refcounts of snapshots and the copied flag */
831 int qcow2_update_snapshot_refcount(BlockDriverState
*bs
,
832 int64_t l1_table_offset
, int l1_size
, int addend
)
834 BDRVQcowState
*s
= bs
->opaque
;
835 uint64_t *l1_table
, *l2_table
, l2_offset
, offset
, l1_size2
, l1_allocated
;
836 int64_t old_offset
, old_l2_offset
;
837 int i
, j
, l1_modified
= 0, nb_csectors
, refcount
;
842 l1_size2
= l1_size
* sizeof(uint64_t);
844 s
->cache_discards
= true;
846 /* WARNING: qcow2_snapshot_goto relies on this function not using the
847 * l1_table_offset when it is the current s->l1_table_offset! Be careful
848 * when changing this! */
849 if (l1_table_offset
!= s
->l1_table_offset
) {
850 l1_table
= g_malloc0(align_offset(l1_size2
, 512));
853 ret
= bdrv_pread(bs
->file
, l1_table_offset
, l1_table
, l1_size2
);
858 for(i
= 0;i
< l1_size
; i
++)
859 be64_to_cpus(&l1_table
[i
]);
861 assert(l1_size
== s
->l1_size
);
862 l1_table
= s
->l1_table
;
866 for(i
= 0; i
< l1_size
; i
++) {
867 l2_offset
= l1_table
[i
];
869 old_l2_offset
= l2_offset
;
870 l2_offset
&= L1E_OFFSET_MASK
;
872 ret
= qcow2_cache_get(bs
, s
->l2_table_cache
, l2_offset
,
878 for(j
= 0; j
< s
->l2_size
; j
++) {
879 uint64_t cluster_index
;
881 offset
= be64_to_cpu(l2_table
[j
]);
883 offset
&= ~QCOW_OFLAG_COPIED
;
885 switch (qcow2_get_cluster_type(offset
)) {
886 case QCOW2_CLUSTER_COMPRESSED
:
887 nb_csectors
= ((offset
>> s
->csize_shift
) &
890 ret
= update_refcount(bs
,
891 (offset
& s
->cluster_offset_mask
) & ~511,
892 nb_csectors
* 512, addend
,
893 QCOW2_DISCARD_SNAPSHOT
);
898 /* compressed clusters are never modified */
902 case QCOW2_CLUSTER_NORMAL
:
903 case QCOW2_CLUSTER_ZERO
:
904 cluster_index
= (offset
& L2E_OFFSET_MASK
) >> s
->cluster_bits
;
905 if (!cluster_index
) {
911 refcount
= qcow2_update_cluster_refcount(bs
,
912 cluster_index
, addend
,
913 QCOW2_DISCARD_SNAPSHOT
);
915 refcount
= get_refcount(bs
, cluster_index
);
924 case QCOW2_CLUSTER_UNALLOCATED
:
933 offset
|= QCOW_OFLAG_COPIED
;
935 if (offset
!= old_offset
) {
937 qcow2_cache_set_dependency(bs
, s
->l2_table_cache
,
938 s
->refcount_block_cache
);
940 l2_table
[j
] = cpu_to_be64(offset
);
941 qcow2_cache_entry_mark_dirty(s
->l2_table_cache
, l2_table
);
945 ret
= qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
952 refcount
= qcow2_update_cluster_refcount(bs
, l2_offset
>>
953 s
->cluster_bits
, addend
, QCOW2_DISCARD_SNAPSHOT
);
955 refcount
= get_refcount(bs
, l2_offset
>> s
->cluster_bits
);
960 } else if (refcount
== 1) {
961 l2_offset
|= QCOW_OFLAG_COPIED
;
963 if (l2_offset
!= old_l2_offset
) {
964 l1_table
[i
] = l2_offset
;
970 ret
= bdrv_flush(bs
);
973 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
976 s
->cache_discards
= false;
977 qcow2_process_discards(bs
, ret
);
979 /* Update L1 only if it isn't deleted anyway (addend = -1) */
980 if (ret
== 0 && addend
>= 0 && l1_modified
) {
981 for (i
= 0; i
< l1_size
; i
++) {
982 cpu_to_be64s(&l1_table
[i
]);
985 ret
= bdrv_pwrite_sync(bs
->file
, l1_table_offset
, l1_table
, l1_size2
);
987 for (i
= 0; i
< l1_size
; i
++) {
988 be64_to_cpus(&l1_table
[i
]);
999 /*********************************************************/
1000 /* refcount checking functions */
1005 * Increases the refcount for a range of clusters in a given refcount table.
1006 * This is used to construct a temporary refcount table out of L1 and L2 tables
1007 * which can be compared the the refcount table saved in the image.
1009 * Modifies the number of errors in res.
1011 static void inc_refcounts(BlockDriverState
*bs
,
1012 BdrvCheckResult
*res
,
1013 uint16_t *refcount_table
,
1014 int refcount_table_size
,
1015 int64_t offset
, int64_t size
)
1017 BDRVQcowState
*s
= bs
->opaque
;
1018 int64_t start
, last
, cluster_offset
;
1024 start
= start_of_cluster(s
, offset
);
1025 last
= start_of_cluster(s
, offset
+ size
- 1);
1026 for(cluster_offset
= start
; cluster_offset
<= last
;
1027 cluster_offset
+= s
->cluster_size
) {
1028 k
= cluster_offset
>> s
->cluster_bits
;
1030 fprintf(stderr
, "ERROR: invalid cluster offset=0x%" PRIx64
"\n",
1033 } else if (k
>= refcount_table_size
) {
1034 fprintf(stderr
, "Warning: cluster offset=0x%" PRIx64
" is after "
1035 "the end of the image file, can't properly check refcounts.\n",
1037 res
->check_errors
++;
1039 if (++refcount_table
[k
] == 0) {
1040 fprintf(stderr
, "ERROR: overflow cluster offset=0x%" PRIx64
1041 "\n", cluster_offset
);
1048 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
1050 CHECK_FRAG_INFO
= 0x2, /* update BlockFragInfo counters */
1054 * Increases the refcount in the given refcount table for the all clusters
1055 * referenced in the L2 table. While doing so, performs some checks on L2
1058 * Returns the number of errors found by the checks or -errno if an internal
1061 static int check_refcounts_l2(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1062 uint16_t *refcount_table
, int refcount_table_size
, int64_t l2_offset
,
1065 BDRVQcowState
*s
= bs
->opaque
;
1066 uint64_t *l2_table
, l2_entry
;
1067 uint64_t next_contiguous_offset
= 0;
1068 int i
, l2_size
, nb_csectors
;
1070 /* Read L2 table from disk */
1071 l2_size
= s
->l2_size
* sizeof(uint64_t);
1072 l2_table
= g_malloc(l2_size
);
1074 if (bdrv_pread(bs
->file
, l2_offset
, l2_table
, l2_size
) != l2_size
)
1077 /* Do the actual checks */
1078 for(i
= 0; i
< s
->l2_size
; i
++) {
1079 l2_entry
= be64_to_cpu(l2_table
[i
]);
1081 switch (qcow2_get_cluster_type(l2_entry
)) {
1082 case QCOW2_CLUSTER_COMPRESSED
:
1083 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1084 if (l2_entry
& QCOW_OFLAG_COPIED
) {
1085 fprintf(stderr
, "ERROR: cluster %" PRId64
": "
1086 "copied flag must never be set for compressed "
1087 "clusters\n", l2_entry
>> s
->cluster_bits
);
1088 l2_entry
&= ~QCOW_OFLAG_COPIED
;
1092 /* Mark cluster as used */
1093 nb_csectors
= ((l2_entry
>> s
->csize_shift
) &
1095 l2_entry
&= s
->cluster_offset_mask
;
1096 inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1097 l2_entry
& ~511, nb_csectors
* 512);
1099 if (flags
& CHECK_FRAG_INFO
) {
1100 res
->bfi
.allocated_clusters
++;
1101 res
->bfi
.compressed_clusters
++;
1103 /* Compressed clusters are fragmented by nature. Since they
1104 * take up sub-sector space but we only have sector granularity
1105 * I/O we need to re-read the same sectors even for adjacent
1106 * compressed clusters.
1108 res
->bfi
.fragmented_clusters
++;
1112 case QCOW2_CLUSTER_ZERO
:
1113 if ((l2_entry
& L2E_OFFSET_MASK
) == 0) {
1118 case QCOW2_CLUSTER_NORMAL
:
1120 uint64_t offset
= l2_entry
& L2E_OFFSET_MASK
;
1122 if (flags
& CHECK_FRAG_INFO
) {
1123 res
->bfi
.allocated_clusters
++;
1124 if (next_contiguous_offset
&&
1125 offset
!= next_contiguous_offset
) {
1126 res
->bfi
.fragmented_clusters
++;
1128 next_contiguous_offset
= offset
+ s
->cluster_size
;
1131 /* Mark cluster as used */
1132 inc_refcounts(bs
, res
, refcount_table
,refcount_table_size
,
1133 offset
, s
->cluster_size
);
1135 /* Correct offsets are cluster aligned */
1136 if (offset_into_cluster(s
, offset
)) {
1137 fprintf(stderr
, "ERROR offset=%" PRIx64
": Cluster is not "
1138 "properly aligned; L2 entry corrupted.\n", offset
);
1144 case QCOW2_CLUSTER_UNALLOCATED
:
1156 fprintf(stderr
, "ERROR: I/O error in check_refcounts_l2\n");
1162 * Increases the refcount for the L1 table, its L2 tables and all referenced
1163 * clusters in the given refcount table. While doing so, performs some checks
1164 * on L1 and L2 entries.
1166 * Returns the number of errors found by the checks or -errno if an internal
1169 static int check_refcounts_l1(BlockDriverState
*bs
,
1170 BdrvCheckResult
*res
,
1171 uint16_t *refcount_table
,
1172 int refcount_table_size
,
1173 int64_t l1_table_offset
, int l1_size
,
1176 BDRVQcowState
*s
= bs
->opaque
;
1177 uint64_t *l1_table
, l2_offset
, l1_size2
;
1180 l1_size2
= l1_size
* sizeof(uint64_t);
1182 /* Mark L1 table as used */
1183 inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1184 l1_table_offset
, l1_size2
);
1186 /* Read L1 table entries from disk */
1187 if (l1_size2
== 0) {
1190 l1_table
= g_malloc(l1_size2
);
1191 if (bdrv_pread(bs
->file
, l1_table_offset
,
1192 l1_table
, l1_size2
) != l1_size2
)
1194 for(i
= 0;i
< l1_size
; i
++)
1195 be64_to_cpus(&l1_table
[i
]);
1198 /* Do the actual checks */
1199 for(i
= 0; i
< l1_size
; i
++) {
1200 l2_offset
= l1_table
[i
];
1202 /* Mark L2 table as used */
1203 l2_offset
&= L1E_OFFSET_MASK
;
1204 inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1205 l2_offset
, s
->cluster_size
);
1207 /* L2 tables are cluster aligned */
1208 if (offset_into_cluster(s
, l2_offset
)) {
1209 fprintf(stderr
, "ERROR l2_offset=%" PRIx64
": Table is not "
1210 "cluster aligned; L1 entry corrupted\n", l2_offset
);
1214 /* Process and check L2 entries */
1215 ret
= check_refcounts_l2(bs
, res
, refcount_table
,
1216 refcount_table_size
, l2_offset
, flags
);
1226 fprintf(stderr
, "ERROR: I/O error in check_refcounts_l1\n");
1227 res
->check_errors
++;
1233 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1235 * This function does not print an error message nor does it increment
1236 * check_errors if get_refcount fails (this is because such an error will have
1237 * been already detected and sufficiently signaled by the calling function
1238 * (qcow2_check_refcounts) by the time this function is called).
1240 static int check_oflag_copied(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1243 BDRVQcowState
*s
= bs
->opaque
;
1244 uint64_t *l2_table
= qemu_blockalign(bs
, s
->cluster_size
);
1249 for (i
= 0; i
< s
->l1_size
; i
++) {
1250 uint64_t l1_entry
= s
->l1_table
[i
];
1251 uint64_t l2_offset
= l1_entry
& L1E_OFFSET_MASK
;
1252 bool l2_dirty
= false;
1258 refcount
= get_refcount(bs
, l2_offset
>> s
->cluster_bits
);
1260 /* don't print message nor increment check_errors */
1263 if ((refcount
== 1) != ((l1_entry
& QCOW_OFLAG_COPIED
) != 0)) {
1264 fprintf(stderr
, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1265 "l1_entry=%" PRIx64
" refcount=%d\n",
1266 fix
& BDRV_FIX_ERRORS
? "Repairing" :
1268 i
, l1_entry
, refcount
);
1269 if (fix
& BDRV_FIX_ERRORS
) {
1270 s
->l1_table
[i
] = refcount
== 1
1271 ? l1_entry
| QCOW_OFLAG_COPIED
1272 : l1_entry
& ~QCOW_OFLAG_COPIED
;
1273 ret
= qcow2_write_l1_entry(bs
, i
);
1275 res
->check_errors
++;
1278 res
->corruptions_fixed
++;
1284 ret
= bdrv_pread(bs
->file
, l2_offset
, l2_table
,
1285 s
->l2_size
* sizeof(uint64_t));
1287 fprintf(stderr
, "ERROR: Could not read L2 table: %s\n",
1289 res
->check_errors
++;
1293 for (j
= 0; j
< s
->l2_size
; j
++) {
1294 uint64_t l2_entry
= be64_to_cpu(l2_table
[j
]);
1295 uint64_t data_offset
= l2_entry
& L2E_OFFSET_MASK
;
1296 int cluster_type
= qcow2_get_cluster_type(l2_entry
);
1298 if ((cluster_type
== QCOW2_CLUSTER_NORMAL
) ||
1299 ((cluster_type
== QCOW2_CLUSTER_ZERO
) && (data_offset
!= 0))) {
1300 refcount
= get_refcount(bs
, data_offset
>> s
->cluster_bits
);
1302 /* don't print message nor increment check_errors */
1305 if ((refcount
== 1) != ((l2_entry
& QCOW_OFLAG_COPIED
) != 0)) {
1306 fprintf(stderr
, "%s OFLAG_COPIED data cluster: "
1307 "l2_entry=%" PRIx64
" refcount=%d\n",
1308 fix
& BDRV_FIX_ERRORS
? "Repairing" :
1310 l2_entry
, refcount
);
1311 if (fix
& BDRV_FIX_ERRORS
) {
1312 l2_table
[j
] = cpu_to_be64(refcount
== 1
1313 ? l2_entry
| QCOW_OFLAG_COPIED
1314 : l2_entry
& ~QCOW_OFLAG_COPIED
);
1316 res
->corruptions_fixed
++;
1325 ret
= qcow2_pre_write_overlap_check(bs
, QCOW2_OL_ACTIVE_L2
,
1326 l2_offset
, s
->cluster_size
);
1328 fprintf(stderr
, "ERROR: Could not write L2 table; metadata "
1329 "overlap check failed: %s\n", strerror(-ret
));
1330 res
->check_errors
++;
1334 ret
= bdrv_pwrite(bs
->file
, l2_offset
, l2_table
, s
->cluster_size
);
1336 fprintf(stderr
, "ERROR: Could not write L2 table: %s\n",
1338 res
->check_errors
++;
1347 qemu_vfree(l2_table
);
1352 * Writes one sector of the refcount table to the disk
1354 #define RT_ENTRIES_PER_SECTOR (512 / sizeof(uint64_t))
1355 static int write_reftable_entry(BlockDriverState
*bs
, int rt_index
)
1357 BDRVQcowState
*s
= bs
->opaque
;
1358 uint64_t buf
[RT_ENTRIES_PER_SECTOR
];
1362 rt_start_index
= rt_index
& ~(RT_ENTRIES_PER_SECTOR
- 1);
1363 for (i
= 0; i
< RT_ENTRIES_PER_SECTOR
; i
++) {
1364 buf
[i
] = cpu_to_be64(s
->refcount_table
[rt_start_index
+ i
]);
1367 ret
= qcow2_pre_write_overlap_check(bs
, QCOW2_OL_REFCOUNT_TABLE
,
1368 s
->refcount_table_offset
+ rt_start_index
* sizeof(uint64_t),
1374 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
1375 ret
= bdrv_pwrite_sync(bs
->file
, s
->refcount_table_offset
+
1376 rt_start_index
* sizeof(uint64_t), buf
, sizeof(buf
));
1385 * Allocates a new cluster for the given refcount block (represented by its
1386 * offset in the image file) and copies the current content there. This function
1387 * does _not_ decrement the reference count for the currently occupied cluster.
1389 * This function prints an informative message to stderr on error (and returns
1390 * -errno); on success, the offset of the newly allocated cluster is returned.
1392 static int64_t realloc_refcount_block(BlockDriverState
*bs
, int reftable_index
,
1395 BDRVQcowState
*s
= bs
->opaque
;
1396 int64_t new_offset
= 0;
1397 void *refcount_block
= NULL
;
1400 /* allocate new refcount block */
1401 new_offset
= qcow2_alloc_clusters(bs
, s
->cluster_size
);
1402 if (new_offset
< 0) {
1403 fprintf(stderr
, "Could not allocate new cluster: %s\n",
1404 strerror(-new_offset
));
1409 /* fetch current refcount block content */
1410 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, offset
, &refcount_block
);
1412 fprintf(stderr
, "Could not fetch refcount block: %s\n", strerror(-ret
));
1413 goto fail_free_cluster
;
1416 /* new block has not yet been entered into refcount table, therefore it is
1417 * no refcount block yet (regarding this check) */
1418 ret
= qcow2_pre_write_overlap_check(bs
, 0, new_offset
, s
->cluster_size
);
1420 fprintf(stderr
, "Could not write refcount block; metadata overlap "
1421 "check failed: %s\n", strerror(-ret
));
1422 /* the image will be marked corrupt, so don't even attempt on freeing
1427 /* write to new block */
1428 ret
= bdrv_write(bs
->file
, new_offset
/ BDRV_SECTOR_SIZE
, refcount_block
,
1429 s
->cluster_sectors
);
1431 fprintf(stderr
, "Could not write refcount block: %s\n", strerror(-ret
));
1432 goto fail_free_cluster
;
1435 /* update refcount table */
1436 assert(!offset_into_cluster(s
, new_offset
));
1437 s
->refcount_table
[reftable_index
] = new_offset
;
1438 ret
= write_reftable_entry(bs
, reftable_index
);
1440 fprintf(stderr
, "Could not update refcount table: %s\n",
1442 goto fail_free_cluster
;
1448 qcow2_free_clusters(bs
, new_offset
, s
->cluster_size
, QCOW2_DISCARD_OTHER
);
1451 if (refcount_block
) {
1452 /* This should never fail, as it would only do so if the given refcount
1453 * block cannot be found in the cache. As this is impossible as long as
1454 * there are no bugs, assert the success. */
1455 int tmp
= qcow2_cache_put(bs
, s
->refcount_block_cache
, &refcount_block
);
1467 * Checks an image for refcount consistency.
1469 * Returns 0 if no errors are found, the number of errors in case the image is
1470 * detected as corrupted, and -errno when an internal error occurred.
1472 int qcow2_check_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1475 BDRVQcowState
*s
= bs
->opaque
;
1476 int64_t size
, i
, highest_cluster
;
1477 int nb_clusters
, refcount1
, refcount2
;
1479 uint16_t *refcount_table
;
1482 size
= bdrv_getlength(bs
->file
);
1483 nb_clusters
= size_to_clusters(s
, size
);
1484 refcount_table
= g_malloc0(nb_clusters
* sizeof(uint16_t));
1486 res
->bfi
.total_clusters
=
1487 size_to_clusters(s
, bs
->total_sectors
* BDRV_SECTOR_SIZE
);
1490 inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1491 0, s
->cluster_size
);
1493 /* current L1 table */
1494 ret
= check_refcounts_l1(bs
, res
, refcount_table
, nb_clusters
,
1495 s
->l1_table_offset
, s
->l1_size
, CHECK_FRAG_INFO
);
1501 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
1502 sn
= s
->snapshots
+ i
;
1503 ret
= check_refcounts_l1(bs
, res
, refcount_table
, nb_clusters
,
1504 sn
->l1_table_offset
, sn
->l1_size
, 0);
1509 inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1510 s
->snapshots_offset
, s
->snapshots_size
);
1513 inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1514 s
->refcount_table_offset
,
1515 s
->refcount_table_size
* sizeof(uint64_t));
1517 for(i
= 0; i
< s
->refcount_table_size
; i
++) {
1518 uint64_t offset
, cluster
;
1519 offset
= s
->refcount_table
[i
];
1520 cluster
= offset
>> s
->cluster_bits
;
1522 /* Refcount blocks are cluster aligned */
1523 if (offset_into_cluster(s
, offset
)) {
1524 fprintf(stderr
, "ERROR refcount block %" PRId64
" is not "
1525 "cluster aligned; refcount table entry corrupted\n", i
);
1530 if (cluster
>= nb_clusters
) {
1531 fprintf(stderr
, "ERROR refcount block %" PRId64
1532 " is outside image\n", i
);
1538 inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1539 offset
, s
->cluster_size
);
1540 if (refcount_table
[cluster
] != 1) {
1541 fprintf(stderr
, "%s refcount block %" PRId64
1543 fix
& BDRV_FIX_ERRORS
? "Repairing" :
1545 i
, refcount_table
[cluster
]);
1547 if (fix
& BDRV_FIX_ERRORS
) {
1550 new_offset
= realloc_refcount_block(bs
, i
, offset
);
1551 if (new_offset
< 0) {
1556 /* update refcounts */
1557 if ((new_offset
>> s
->cluster_bits
) >= nb_clusters
) {
1558 /* increase refcount_table size if necessary */
1559 int old_nb_clusters
= nb_clusters
;
1560 nb_clusters
= (new_offset
>> s
->cluster_bits
) + 1;
1561 refcount_table
= g_realloc(refcount_table
,
1562 nb_clusters
* sizeof(uint16_t));
1563 memset(&refcount_table
[old_nb_clusters
], 0, (nb_clusters
1564 - old_nb_clusters
) * sizeof(uint16_t));
1566 refcount_table
[cluster
]--;
1567 inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1568 new_offset
, s
->cluster_size
);
1570 res
->corruptions_fixed
++;
1578 /* compare ref counts */
1579 for (i
= 0, highest_cluster
= 0; i
< nb_clusters
; i
++) {
1580 refcount1
= get_refcount(bs
, i
);
1581 if (refcount1
< 0) {
1582 fprintf(stderr
, "Can't get refcount for cluster %" PRId64
": %s\n",
1583 i
, strerror(-refcount1
));
1584 res
->check_errors
++;
1588 refcount2
= refcount_table
[i
];
1590 if (refcount1
> 0 || refcount2
> 0) {
1591 highest_cluster
= i
;
1594 if (refcount1
!= refcount2
) {
1596 /* Check if we're allowed to fix the mismatch */
1597 int *num_fixed
= NULL
;
1598 if (refcount1
> refcount2
&& (fix
& BDRV_FIX_LEAKS
)) {
1599 num_fixed
= &res
->leaks_fixed
;
1600 } else if (refcount1
< refcount2
&& (fix
& BDRV_FIX_ERRORS
)) {
1601 num_fixed
= &res
->corruptions_fixed
;
1604 fprintf(stderr
, "%s cluster %" PRId64
" refcount=%d reference=%d\n",
1605 num_fixed
!= NULL
? "Repairing" :
1606 refcount1
< refcount2
? "ERROR" :
1608 i
, refcount1
, refcount2
);
1611 ret
= update_refcount(bs
, i
<< s
->cluster_bits
, 1,
1612 refcount2
- refcount1
,
1613 QCOW2_DISCARD_ALWAYS
);
1620 /* And if we couldn't, print an error */
1621 if (refcount1
< refcount2
) {
1629 /* check OFLAG_COPIED */
1630 ret
= check_oflag_copied(bs
, res
, fix
);
1635 res
->image_end_offset
= (highest_cluster
+ 1) * s
->cluster_size
;
1639 g_free(refcount_table
);
1644 #define overlaps_with(ofs, sz) \
1645 ranges_overlap(offset, size, ofs, sz)
1648 * Checks if the given offset into the image file is actually free to use by
1649 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
1650 * i.e. a sanity check without relying on the refcount tables.
1652 * The ign parameter specifies what checks not to perform (being a bitmask of
1653 * QCow2MetadataOverlap values), i.e., what sections to ignore.
1656 * - 0 if writing to this offset will not affect the mentioned metadata
1657 * - a positive QCow2MetadataOverlap value indicating one overlapping section
1658 * - a negative value (-errno) indicating an error while performing a check,
1659 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
1661 int qcow2_check_metadata_overlap(BlockDriverState
*bs
, int ign
, int64_t offset
,
1664 BDRVQcowState
*s
= bs
->opaque
;
1665 int chk
= s
->overlap_check
& ~ign
;
1672 if (chk
& QCOW2_OL_MAIN_HEADER
) {
1673 if (offset
< s
->cluster_size
) {
1674 return QCOW2_OL_MAIN_HEADER
;
1678 /* align range to test to cluster boundaries */
1679 size
= align_offset(offset_into_cluster(s
, offset
) + size
, s
->cluster_size
);
1680 offset
= start_of_cluster(s
, offset
);
1682 if ((chk
& QCOW2_OL_ACTIVE_L1
) && s
->l1_size
) {
1683 if (overlaps_with(s
->l1_table_offset
, s
->l1_size
* sizeof(uint64_t))) {
1684 return QCOW2_OL_ACTIVE_L1
;
1688 if ((chk
& QCOW2_OL_REFCOUNT_TABLE
) && s
->refcount_table_size
) {
1689 if (overlaps_with(s
->refcount_table_offset
,
1690 s
->refcount_table_size
* sizeof(uint64_t))) {
1691 return QCOW2_OL_REFCOUNT_TABLE
;
1695 if ((chk
& QCOW2_OL_SNAPSHOT_TABLE
) && s
->snapshots_size
) {
1696 if (overlaps_with(s
->snapshots_offset
, s
->snapshots_size
)) {
1697 return QCOW2_OL_SNAPSHOT_TABLE
;
1701 if ((chk
& QCOW2_OL_INACTIVE_L1
) && s
->snapshots
) {
1702 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
1703 if (s
->snapshots
[i
].l1_size
&&
1704 overlaps_with(s
->snapshots
[i
].l1_table_offset
,
1705 s
->snapshots
[i
].l1_size
* sizeof(uint64_t))) {
1706 return QCOW2_OL_INACTIVE_L1
;
1711 if ((chk
& QCOW2_OL_ACTIVE_L2
) && s
->l1_table
) {
1712 for (i
= 0; i
< s
->l1_size
; i
++) {
1713 if ((s
->l1_table
[i
] & L1E_OFFSET_MASK
) &&
1714 overlaps_with(s
->l1_table
[i
] & L1E_OFFSET_MASK
,
1716 return QCOW2_OL_ACTIVE_L2
;
1721 if ((chk
& QCOW2_OL_REFCOUNT_BLOCK
) && s
->refcount_table
) {
1722 for (i
= 0; i
< s
->refcount_table_size
; i
++) {
1723 if ((s
->refcount_table
[i
] & REFT_OFFSET_MASK
) &&
1724 overlaps_with(s
->refcount_table
[i
] & REFT_OFFSET_MASK
,
1726 return QCOW2_OL_REFCOUNT_BLOCK
;
1731 if ((chk
& QCOW2_OL_INACTIVE_L2
) && s
->snapshots
) {
1732 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
1733 uint64_t l1_ofs
= s
->snapshots
[i
].l1_table_offset
;
1734 uint32_t l1_sz
= s
->snapshots
[i
].l1_size
;
1735 uint64_t l1_sz2
= l1_sz
* sizeof(uint64_t);
1736 uint64_t *l1
= g_malloc(l1_sz2
);
1739 ret
= bdrv_pread(bs
->file
, l1_ofs
, l1
, l1_sz2
);
1745 for (j
= 0; j
< l1_sz
; j
++) {
1746 uint64_t l2_ofs
= be64_to_cpu(l1
[j
]) & L1E_OFFSET_MASK
;
1747 if (l2_ofs
&& overlaps_with(l2_ofs
, s
->cluster_size
)) {
1749 return QCOW2_OL_INACTIVE_L2
;
1760 static const char *metadata_ol_names
[] = {
1761 [QCOW2_OL_MAIN_HEADER_BITNR
] = "qcow2_header",
1762 [QCOW2_OL_ACTIVE_L1_BITNR
] = "active L1 table",
1763 [QCOW2_OL_ACTIVE_L2_BITNR
] = "active L2 table",
1764 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = "refcount table",
1765 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = "refcount block",
1766 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = "snapshot table",
1767 [QCOW2_OL_INACTIVE_L1_BITNR
] = "inactive L1 table",
1768 [QCOW2_OL_INACTIVE_L2_BITNR
] = "inactive L2 table",
1772 * First performs a check for metadata overlaps (through
1773 * qcow2_check_metadata_overlap); if that fails with a negative value (error
1774 * while performing a check), that value is returned. If an impending overlap
1775 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
1776 * and -EIO returned.
1778 * Returns 0 if there were neither overlaps nor errors while checking for
1779 * overlaps; or a negative value (-errno) on error.
1781 int qcow2_pre_write_overlap_check(BlockDriverState
*bs
, int ign
, int64_t offset
,
1784 int ret
= qcow2_check_metadata_overlap(bs
, ign
, offset
, size
);
1788 } else if (ret
> 0) {
1789 int metadata_ol_bitnr
= ffs(ret
) - 1;
1793 assert(metadata_ol_bitnr
< QCOW2_OL_MAX_BITNR
);
1795 fprintf(stderr
, "qcow2: Preventing invalid write on metadata (overlaps "
1796 "with %s); image marked as corrupt.\n",
1797 metadata_ol_names
[metadata_ol_bitnr
]);
1798 message
= g_strdup_printf("Prevented %s overwrite",
1799 metadata_ol_names
[metadata_ol_bitnr
]);
1800 data
= qobject_from_jsonf("{ 'device': %s, 'msg': %s, 'offset': %"
1801 PRId64
", 'size': %" PRId64
" }", bs
->device_name
, message
,
1803 monitor_protocol_event(QEVENT_BLOCK_IMAGE_CORRUPTED
, data
);
1805 qobject_decref(data
);
1807 qcow2_mark_corrupt(bs
);
1808 bs
->drv
= NULL
; /* make BDS unusable */