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"
30 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, uint64_t size
);
31 static int QEMU_WARN_UNUSED_RESULT
update_refcount(BlockDriverState
*bs
,
32 int64_t offset
, int64_t length
, uint64_t addend
,
33 bool decrease
, enum qcow2_discard_type type
);
35 static uint64_t get_refcount_ro0(const void *refcount_array
, uint64_t index
);
36 static uint64_t get_refcount_ro1(const void *refcount_array
, uint64_t index
);
37 static uint64_t get_refcount_ro2(const void *refcount_array
, uint64_t index
);
38 static uint64_t get_refcount_ro3(const void *refcount_array
, uint64_t index
);
39 static uint64_t get_refcount_ro4(const void *refcount_array
, uint64_t index
);
40 static uint64_t get_refcount_ro5(const void *refcount_array
, uint64_t index
);
41 static uint64_t get_refcount_ro6(const void *refcount_array
, uint64_t index
);
43 static void set_refcount_ro0(void *refcount_array
, uint64_t index
,
45 static void set_refcount_ro1(void *refcount_array
, uint64_t index
,
47 static void set_refcount_ro2(void *refcount_array
, uint64_t index
,
49 static void set_refcount_ro3(void *refcount_array
, uint64_t index
,
51 static void set_refcount_ro4(void *refcount_array
, uint64_t index
,
53 static void set_refcount_ro5(void *refcount_array
, uint64_t index
,
55 static void set_refcount_ro6(void *refcount_array
, uint64_t index
,
59 static Qcow2GetRefcountFunc
*const get_refcount_funcs
[] = {
69 static Qcow2SetRefcountFunc
*const set_refcount_funcs
[] = {
80 /*********************************************************/
81 /* refcount handling */
83 int qcow2_refcount_init(BlockDriverState
*bs
)
85 BDRVQcow2State
*s
= bs
->opaque
;
86 unsigned int refcount_table_size2
, i
;
89 assert(s
->refcount_order
>= 0 && s
->refcount_order
<= 6);
91 s
->get_refcount
= get_refcount_funcs
[s
->refcount_order
];
92 s
->set_refcount
= set_refcount_funcs
[s
->refcount_order
];
94 assert(s
->refcount_table_size
<= INT_MAX
/ sizeof(uint64_t));
95 refcount_table_size2
= s
->refcount_table_size
* sizeof(uint64_t);
96 s
->refcount_table
= g_try_malloc(refcount_table_size2
);
98 if (s
->refcount_table_size
> 0) {
99 if (s
->refcount_table
== NULL
) {
103 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_LOAD
);
104 ret
= bdrv_pread(bs
->file
->bs
, s
->refcount_table_offset
,
105 s
->refcount_table
, refcount_table_size2
);
109 for(i
= 0; i
< s
->refcount_table_size
; i
++)
110 be64_to_cpus(&s
->refcount_table
[i
]);
117 void qcow2_refcount_close(BlockDriverState
*bs
)
119 BDRVQcow2State
*s
= bs
->opaque
;
120 g_free(s
->refcount_table
);
124 static uint64_t get_refcount_ro0(const void *refcount_array
, uint64_t index
)
126 return (((const uint8_t *)refcount_array
)[index
/ 8] >> (index
% 8)) & 0x1;
129 static void set_refcount_ro0(void *refcount_array
, uint64_t index
,
132 assert(!(value
>> 1));
133 ((uint8_t *)refcount_array
)[index
/ 8] &= ~(0x1 << (index
% 8));
134 ((uint8_t *)refcount_array
)[index
/ 8] |= value
<< (index
% 8);
137 static uint64_t get_refcount_ro1(const void *refcount_array
, uint64_t index
)
139 return (((const uint8_t *)refcount_array
)[index
/ 4] >> (2 * (index
% 4)))
143 static void set_refcount_ro1(void *refcount_array
, uint64_t index
,
146 assert(!(value
>> 2));
147 ((uint8_t *)refcount_array
)[index
/ 4] &= ~(0x3 << (2 * (index
% 4)));
148 ((uint8_t *)refcount_array
)[index
/ 4] |= value
<< (2 * (index
% 4));
151 static uint64_t get_refcount_ro2(const void *refcount_array
, uint64_t index
)
153 return (((const uint8_t *)refcount_array
)[index
/ 2] >> (4 * (index
% 2)))
157 static void set_refcount_ro2(void *refcount_array
, uint64_t index
,
160 assert(!(value
>> 4));
161 ((uint8_t *)refcount_array
)[index
/ 2] &= ~(0xf << (4 * (index
% 2)));
162 ((uint8_t *)refcount_array
)[index
/ 2] |= value
<< (4 * (index
% 2));
165 static uint64_t get_refcount_ro3(const void *refcount_array
, uint64_t index
)
167 return ((const uint8_t *)refcount_array
)[index
];
170 static void set_refcount_ro3(void *refcount_array
, uint64_t index
,
173 assert(!(value
>> 8));
174 ((uint8_t *)refcount_array
)[index
] = value
;
177 static uint64_t get_refcount_ro4(const void *refcount_array
, uint64_t index
)
179 return be16_to_cpu(((const uint16_t *)refcount_array
)[index
]);
182 static void set_refcount_ro4(void *refcount_array
, uint64_t index
,
185 assert(!(value
>> 16));
186 ((uint16_t *)refcount_array
)[index
] = cpu_to_be16(value
);
189 static uint64_t get_refcount_ro5(const void *refcount_array
, uint64_t index
)
191 return be32_to_cpu(((const uint32_t *)refcount_array
)[index
]);
194 static void set_refcount_ro5(void *refcount_array
, uint64_t index
,
197 assert(!(value
>> 32));
198 ((uint32_t *)refcount_array
)[index
] = cpu_to_be32(value
);
201 static uint64_t get_refcount_ro6(const void *refcount_array
, uint64_t index
)
203 return be64_to_cpu(((const uint64_t *)refcount_array
)[index
]);
206 static void set_refcount_ro6(void *refcount_array
, uint64_t index
,
209 ((uint64_t *)refcount_array
)[index
] = cpu_to_be64(value
);
213 static int load_refcount_block(BlockDriverState
*bs
,
214 int64_t refcount_block_offset
,
215 void **refcount_block
)
217 BDRVQcow2State
*s
= bs
->opaque
;
220 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_LOAD
);
221 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, refcount_block_offset
,
228 * Retrieves the refcount of the cluster given by its index and stores it in
229 * *refcount. Returns 0 on success and -errno on failure.
231 int qcow2_get_refcount(BlockDriverState
*bs
, int64_t cluster_index
,
234 BDRVQcow2State
*s
= bs
->opaque
;
235 uint64_t refcount_table_index
, block_index
;
236 int64_t refcount_block_offset
;
238 void *refcount_block
;
240 refcount_table_index
= cluster_index
>> s
->refcount_block_bits
;
241 if (refcount_table_index
>= s
->refcount_table_size
) {
245 refcount_block_offset
=
246 s
->refcount_table
[refcount_table_index
] & REFT_OFFSET_MASK
;
247 if (!refcount_block_offset
) {
252 if (offset_into_cluster(s
, refcount_block_offset
)) {
253 qcow2_signal_corruption(bs
, true, -1, -1, "Refblock offset %#" PRIx64
254 " unaligned (reftable index: %#" PRIx64
")",
255 refcount_block_offset
, refcount_table_index
);
259 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, refcount_block_offset
,
265 block_index
= cluster_index
& (s
->refcount_block_size
- 1);
266 *refcount
= s
->get_refcount(refcount_block
, block_index
);
268 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refcount_block
);
274 * Rounds the refcount table size up to avoid growing the table for each single
275 * refcount block that is allocated.
277 static unsigned int next_refcount_table_size(BDRVQcow2State
*s
,
278 unsigned int min_size
)
280 unsigned int min_clusters
= (min_size
>> (s
->cluster_bits
- 3)) + 1;
281 unsigned int refcount_table_clusters
=
282 MAX(1, s
->refcount_table_size
>> (s
->cluster_bits
- 3));
284 while (min_clusters
> refcount_table_clusters
) {
285 refcount_table_clusters
= (refcount_table_clusters
* 3 + 1) / 2;
288 return refcount_table_clusters
<< (s
->cluster_bits
- 3);
292 /* Checks if two offsets are described by the same refcount block */
293 static int in_same_refcount_block(BDRVQcow2State
*s
, uint64_t offset_a
,
296 uint64_t block_a
= offset_a
>> (s
->cluster_bits
+ s
->refcount_block_bits
);
297 uint64_t block_b
= offset_b
>> (s
->cluster_bits
+ s
->refcount_block_bits
);
299 return (block_a
== block_b
);
303 * Loads a refcount block. If it doesn't exist yet, it is allocated first
304 * (including growing the refcount table if needed).
306 * Returns 0 on success or -errno in error case
308 static int alloc_refcount_block(BlockDriverState
*bs
,
309 int64_t cluster_index
, void **refcount_block
)
311 BDRVQcow2State
*s
= bs
->opaque
;
312 unsigned int refcount_table_index
;
315 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
317 /* Find the refcount block for the given cluster */
318 refcount_table_index
= cluster_index
>> s
->refcount_block_bits
;
320 if (refcount_table_index
< s
->refcount_table_size
) {
322 uint64_t refcount_block_offset
=
323 s
->refcount_table
[refcount_table_index
] & REFT_OFFSET_MASK
;
325 /* If it's already there, we're done */
326 if (refcount_block_offset
) {
327 if (offset_into_cluster(s
, refcount_block_offset
)) {
328 qcow2_signal_corruption(bs
, true, -1, -1, "Refblock offset %#"
329 PRIx64
" unaligned (reftable index: "
330 "%#x)", refcount_block_offset
,
331 refcount_table_index
);
335 return load_refcount_block(bs
, refcount_block_offset
,
341 * If we came here, we need to allocate something. Something is at least
342 * a cluster for the new refcount block. It may also include a new refcount
343 * table if the old refcount table is too small.
345 * Note that allocating clusters here needs some special care:
347 * - We can't use the normal qcow2_alloc_clusters(), it would try to
348 * increase the refcount and very likely we would end up with an endless
349 * recursion. Instead we must place the refcount blocks in a way that
350 * they can describe them themselves.
352 * - We need to consider that at this point we are inside update_refcounts
353 * and potentially doing an initial refcount increase. This means that
354 * some clusters have already been allocated by the caller, but their
355 * refcount isn't accurate yet. If we allocate clusters for metadata, we
356 * need to return -EAGAIN to signal the caller that it needs to restart
357 * the search for free clusters.
359 * - alloc_clusters_noref and qcow2_free_clusters may load a different
360 * refcount block into the cache
363 *refcount_block
= NULL
;
365 /* We write to the refcount table, so we might depend on L2 tables */
366 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
371 /* Allocate the refcount block itself and mark it as used */
372 int64_t new_block
= alloc_clusters_noref(bs
, s
->cluster_size
);
378 fprintf(stderr
, "qcow2: Allocate refcount block %d for %" PRIx64
380 refcount_table_index
, cluster_index
<< s
->cluster_bits
, new_block
);
383 if (in_same_refcount_block(s
, new_block
, cluster_index
<< s
->cluster_bits
)) {
384 /* Zero the new refcount block before updating it */
385 ret
= qcow2_cache_get_empty(bs
, s
->refcount_block_cache
, new_block
,
391 memset(*refcount_block
, 0, s
->cluster_size
);
393 /* The block describes itself, need to update the cache */
394 int block_index
= (new_block
>> s
->cluster_bits
) &
395 (s
->refcount_block_size
- 1);
396 s
->set_refcount(*refcount_block
, block_index
, 1);
398 /* Described somewhere else. This can recurse at most twice before we
399 * arrive at a block that describes itself. */
400 ret
= update_refcount(bs
, new_block
, s
->cluster_size
, 1, false,
401 QCOW2_DISCARD_NEVER
);
406 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
411 /* Initialize the new refcount block only after updating its refcount,
412 * update_refcount uses the refcount cache itself */
413 ret
= qcow2_cache_get_empty(bs
, s
->refcount_block_cache
, new_block
,
419 memset(*refcount_block
, 0, s
->cluster_size
);
422 /* Now the new refcount block needs to be written to disk */
423 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE
);
424 qcow2_cache_entry_mark_dirty(bs
, s
->refcount_block_cache
, *refcount_block
);
425 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
430 /* If the refcount table is big enough, just hook the block up there */
431 if (refcount_table_index
< s
->refcount_table_size
) {
432 uint64_t data64
= cpu_to_be64(new_block
);
433 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_HOOKUP
);
434 ret
= bdrv_pwrite_sync(bs
->file
->bs
,
435 s
->refcount_table_offset
+ refcount_table_index
* sizeof(uint64_t),
436 &data64
, sizeof(data64
));
441 s
->refcount_table
[refcount_table_index
] = new_block
;
443 /* The new refcount block may be where the caller intended to put its
444 * data, so let it restart the search. */
448 qcow2_cache_put(bs
, s
->refcount_block_cache
, refcount_block
);
451 * If we come here, we need to grow the refcount table. Again, a new
452 * refcount table needs some space and we can't simply allocate to avoid
455 * Therefore let's grab new refcount blocks at the end of the image, which
456 * will describe themselves and the new refcount table. This way we can
457 * reference them only in the new table and do the switch to the new
458 * refcount table at once without producing an inconsistent state in
461 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_GROW
);
463 /* Calculate the number of refcount blocks needed so far; this will be the
464 * basis for calculating the index of the first cluster used for the
465 * self-describing refcount structures which we are about to create.
467 * Because we reached this point, there cannot be any refcount entries for
468 * cluster_index or higher indices yet. However, because new_block has been
469 * allocated to describe that cluster (and it will assume this role later
470 * on), we cannot use that index; also, new_block may actually have a higher
471 * cluster index than cluster_index, so it needs to be taken into account
472 * here (and 1 needs to be added to its value because that cluster is used).
474 uint64_t blocks_used
= DIV_ROUND_UP(MAX(cluster_index
+ 1,
475 (new_block
>> s
->cluster_bits
) + 1),
476 s
->refcount_block_size
);
478 if (blocks_used
> QCOW_MAX_REFTABLE_SIZE
/ sizeof(uint64_t)) {
482 /* And now we need at least one block more for the new metadata */
483 uint64_t table_size
= next_refcount_table_size(s
, blocks_used
+ 1);
484 uint64_t last_table_size
;
485 uint64_t blocks_clusters
;
487 uint64_t table_clusters
=
488 size_to_clusters(s
, table_size
* sizeof(uint64_t));
489 blocks_clusters
= 1 +
490 ((table_clusters
+ s
->refcount_block_size
- 1)
491 / s
->refcount_block_size
);
492 uint64_t meta_clusters
= table_clusters
+ blocks_clusters
;
494 last_table_size
= table_size
;
495 table_size
= next_refcount_table_size(s
, blocks_used
+
496 ((meta_clusters
+ s
->refcount_block_size
- 1)
497 / s
->refcount_block_size
));
499 } while (last_table_size
!= table_size
);
502 fprintf(stderr
, "qcow2: Grow refcount table %" PRId32
" => %" PRId64
"\n",
503 s
->refcount_table_size
, table_size
);
506 /* Create the new refcount table and blocks */
507 uint64_t meta_offset
= (blocks_used
* s
->refcount_block_size
) *
509 uint64_t table_offset
= meta_offset
+ blocks_clusters
* s
->cluster_size
;
510 uint64_t *new_table
= g_try_new0(uint64_t, table_size
);
511 void *new_blocks
= g_try_malloc0(blocks_clusters
* s
->cluster_size
);
513 assert(table_size
> 0 && blocks_clusters
> 0);
514 if (new_table
== NULL
|| new_blocks
== NULL
) {
519 /* Fill the new refcount table */
520 memcpy(new_table
, s
->refcount_table
,
521 s
->refcount_table_size
* sizeof(uint64_t));
522 new_table
[refcount_table_index
] = new_block
;
525 for (i
= 0; i
< blocks_clusters
; i
++) {
526 new_table
[blocks_used
+ i
] = meta_offset
+ (i
* s
->cluster_size
);
529 /* Fill the refcount blocks */
530 uint64_t table_clusters
= size_to_clusters(s
, table_size
* sizeof(uint64_t));
532 for (i
= 0; i
< table_clusters
+ blocks_clusters
; i
++) {
533 s
->set_refcount(new_blocks
, block
++, 1);
536 /* Write refcount blocks to disk */
537 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS
);
538 ret
= bdrv_pwrite_sync(bs
->file
->bs
, meta_offset
, new_blocks
,
539 blocks_clusters
* s
->cluster_size
);
546 /* Write refcount table to disk */
547 for(i
= 0; i
< table_size
; i
++) {
548 cpu_to_be64s(&new_table
[i
]);
551 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE
);
552 ret
= bdrv_pwrite_sync(bs
->file
->bs
, table_offset
, new_table
,
553 table_size
* sizeof(uint64_t));
558 for(i
= 0; i
< table_size
; i
++) {
559 be64_to_cpus(&new_table
[i
]);
562 /* Hook up the new refcount table in the qcow2 header */
567 cpu_to_be64w(&data
.d64
, table_offset
);
568 cpu_to_be32w(&data
.d32
, table_clusters
);
569 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE
);
570 ret
= bdrv_pwrite_sync(bs
->file
->bs
,
571 offsetof(QCowHeader
, refcount_table_offset
),
572 &data
, sizeof(data
));
577 /* And switch it in memory */
578 uint64_t old_table_offset
= s
->refcount_table_offset
;
579 uint64_t old_table_size
= s
->refcount_table_size
;
581 g_free(s
->refcount_table
);
582 s
->refcount_table
= new_table
;
583 s
->refcount_table_size
= table_size
;
584 s
->refcount_table_offset
= table_offset
;
586 /* Free old table. */
587 qcow2_free_clusters(bs
, old_table_offset
, old_table_size
* sizeof(uint64_t),
588 QCOW2_DISCARD_OTHER
);
590 ret
= load_refcount_block(bs
, new_block
, refcount_block
);
595 /* If we were trying to do the initial refcount update for some cluster
596 * allocation, we might have used the same clusters to store newly
597 * allocated metadata. Make the caller search some new space. */
604 if (*refcount_block
!= NULL
) {
605 qcow2_cache_put(bs
, s
->refcount_block_cache
, refcount_block
);
610 void qcow2_process_discards(BlockDriverState
*bs
, int ret
)
612 BDRVQcow2State
*s
= bs
->opaque
;
613 Qcow2DiscardRegion
*d
, *next
;
615 QTAILQ_FOREACH_SAFE(d
, &s
->discards
, next
, next
) {
616 QTAILQ_REMOVE(&s
->discards
, d
, next
);
618 /* Discard is optional, ignore the return value */
620 bdrv_discard(bs
->file
->bs
,
621 d
->offset
>> BDRV_SECTOR_BITS
,
622 d
->bytes
>> BDRV_SECTOR_BITS
);
629 static void update_refcount_discard(BlockDriverState
*bs
,
630 uint64_t offset
, uint64_t length
)
632 BDRVQcow2State
*s
= bs
->opaque
;
633 Qcow2DiscardRegion
*d
, *p
, *next
;
635 QTAILQ_FOREACH(d
, &s
->discards
, next
) {
636 uint64_t new_start
= MIN(offset
, d
->offset
);
637 uint64_t new_end
= MAX(offset
+ length
, d
->offset
+ d
->bytes
);
639 if (new_end
- new_start
<= length
+ d
->bytes
) {
640 /* There can't be any overlap, areas ending up here have no
641 * references any more and therefore shouldn't get freed another
643 assert(d
->bytes
+ length
== new_end
- new_start
);
644 d
->offset
= new_start
;
645 d
->bytes
= new_end
- new_start
;
650 d
= g_malloc(sizeof(*d
));
651 *d
= (Qcow2DiscardRegion
) {
656 QTAILQ_INSERT_TAIL(&s
->discards
, d
, next
);
659 /* Merge discard requests if they are adjacent now */
660 QTAILQ_FOREACH_SAFE(p
, &s
->discards
, next
, next
) {
662 || p
->offset
> d
->offset
+ d
->bytes
663 || d
->offset
> p
->offset
+ p
->bytes
)
668 /* Still no overlap possible */
669 assert(p
->offset
== d
->offset
+ d
->bytes
670 || d
->offset
== p
->offset
+ p
->bytes
);
672 QTAILQ_REMOVE(&s
->discards
, p
, next
);
673 d
->offset
= MIN(d
->offset
, p
->offset
);
674 d
->bytes
+= p
->bytes
;
679 /* XXX: cache several refcount block clusters ? */
680 /* @addend is the absolute value of the addend; if @decrease is set, @addend
681 * will be subtracted from the current refcount, otherwise it will be added */
682 static int QEMU_WARN_UNUSED_RESULT
update_refcount(BlockDriverState
*bs
,
687 enum qcow2_discard_type type
)
689 BDRVQcow2State
*s
= bs
->opaque
;
690 int64_t start
, last
, cluster_offset
;
691 void *refcount_block
= NULL
;
692 int64_t old_table_index
= -1;
696 fprintf(stderr
, "update_refcount: offset=%" PRId64
" size=%" PRId64
697 " addend=%s%" PRIu64
"\n", offset
, length
, decrease
? "-" : "",
702 } else if (length
== 0) {
707 qcow2_cache_set_dependency(bs
, s
->refcount_block_cache
,
711 start
= start_of_cluster(s
, offset
);
712 last
= start_of_cluster(s
, offset
+ length
- 1);
713 for(cluster_offset
= start
; cluster_offset
<= last
;
714 cluster_offset
+= s
->cluster_size
)
718 int64_t cluster_index
= cluster_offset
>> s
->cluster_bits
;
719 int64_t table_index
= cluster_index
>> s
->refcount_block_bits
;
721 /* Load the refcount block and allocate it if needed */
722 if (table_index
!= old_table_index
) {
723 if (refcount_block
) {
724 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refcount_block
);
726 ret
= alloc_refcount_block(bs
, cluster_index
, &refcount_block
);
731 old_table_index
= table_index
;
733 qcow2_cache_entry_mark_dirty(bs
, s
->refcount_block_cache
,
736 /* we can update the count and save it */
737 block_index
= cluster_index
& (s
->refcount_block_size
- 1);
739 refcount
= s
->get_refcount(refcount_block
, block_index
);
740 if (decrease
? (refcount
- addend
> refcount
)
741 : (refcount
+ addend
< refcount
||
742 refcount
+ addend
> s
->refcount_max
))
752 if (refcount
== 0 && cluster_index
< s
->free_cluster_index
) {
753 s
->free_cluster_index
= cluster_index
;
755 s
->set_refcount(refcount_block
, block_index
, refcount
);
757 if (refcount
== 0 && s
->discard_passthrough
[type
]) {
758 update_refcount_discard(bs
, cluster_offset
, s
->cluster_size
);
764 if (!s
->cache_discards
) {
765 qcow2_process_discards(bs
, ret
);
768 /* Write last changed block to disk */
769 if (refcount_block
) {
770 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refcount_block
);
774 * Try do undo any updates if an error is returned (This may succeed in
775 * some cases like ENOSPC for allocating a new refcount block)
779 dummy
= update_refcount(bs
, offset
, cluster_offset
- offset
, addend
,
780 !decrease
, QCOW2_DISCARD_NEVER
);
788 * Increases or decreases the refcount of a given cluster.
790 * @addend is the absolute value of the addend; if @decrease is set, @addend
791 * will be subtracted from the current refcount, otherwise it will be added.
793 * On success 0 is returned; on failure -errno is returned.
795 int qcow2_update_cluster_refcount(BlockDriverState
*bs
,
796 int64_t cluster_index
,
797 uint64_t addend
, bool decrease
,
798 enum qcow2_discard_type type
)
800 BDRVQcow2State
*s
= bs
->opaque
;
803 ret
= update_refcount(bs
, cluster_index
<< s
->cluster_bits
, 1, addend
,
814 /*********************************************************/
815 /* cluster allocation functions */
819 /* return < 0 if error */
820 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, uint64_t size
)
822 BDRVQcow2State
*s
= bs
->opaque
;
823 uint64_t i
, nb_clusters
, refcount
;
826 /* We can't allocate clusters if they may still be queued for discard. */
827 if (s
->cache_discards
) {
828 qcow2_process_discards(bs
, 0);
831 nb_clusters
= size_to_clusters(s
, size
);
833 for(i
= 0; i
< nb_clusters
; i
++) {
834 uint64_t next_cluster_index
= s
->free_cluster_index
++;
835 ret
= qcow2_get_refcount(bs
, next_cluster_index
, &refcount
);
839 } else if (refcount
!= 0) {
844 /* Make sure that all offsets in the "allocated" range are representable
846 if (s
->free_cluster_index
> 0 &&
847 s
->free_cluster_index
- 1 > (INT64_MAX
>> s
->cluster_bits
))
853 fprintf(stderr
, "alloc_clusters: size=%" PRId64
" -> %" PRId64
"\n",
855 (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
);
857 return (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
;
860 int64_t qcow2_alloc_clusters(BlockDriverState
*bs
, uint64_t size
)
865 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC
);
867 offset
= alloc_clusters_noref(bs
, size
);
872 ret
= update_refcount(bs
, offset
, size
, 1, false, QCOW2_DISCARD_NEVER
);
873 } while (ret
== -EAGAIN
);
882 int64_t qcow2_alloc_clusters_at(BlockDriverState
*bs
, uint64_t offset
,
885 BDRVQcow2State
*s
= bs
->opaque
;
886 uint64_t cluster_index
, refcount
;
890 assert(nb_clusters
>= 0);
891 if (nb_clusters
== 0) {
896 /* Check how many clusters there are free */
897 cluster_index
= offset
>> s
->cluster_bits
;
898 for(i
= 0; i
< nb_clusters
; i
++) {
899 ret
= qcow2_get_refcount(bs
, cluster_index
++, &refcount
);
902 } else if (refcount
!= 0) {
907 /* And then allocate them */
908 ret
= update_refcount(bs
, offset
, i
<< s
->cluster_bits
, 1, false,
909 QCOW2_DISCARD_NEVER
);
910 } while (ret
== -EAGAIN
);
919 /* only used to allocate compressed sectors. We try to allocate
920 contiguous sectors. size must be <= cluster_size */
921 int64_t qcow2_alloc_bytes(BlockDriverState
*bs
, int size
)
923 BDRVQcow2State
*s
= bs
->opaque
;
925 size_t free_in_cluster
;
928 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC_BYTES
);
929 assert(size
> 0 && size
<= s
->cluster_size
);
930 assert(!s
->free_byte_offset
|| offset_into_cluster(s
, s
->free_byte_offset
));
932 offset
= s
->free_byte_offset
;
936 ret
= qcow2_get_refcount(bs
, offset
>> s
->cluster_bits
, &refcount
);
941 if (refcount
== s
->refcount_max
) {
946 free_in_cluster
= s
->cluster_size
- offset_into_cluster(s
, offset
);
948 if (!offset
|| free_in_cluster
< size
) {
949 int64_t new_cluster
= alloc_clusters_noref(bs
, s
->cluster_size
);
950 if (new_cluster
< 0) {
954 if (!offset
|| ROUND_UP(offset
, s
->cluster_size
) != new_cluster
) {
955 offset
= new_cluster
;
956 free_in_cluster
= s
->cluster_size
;
958 free_in_cluster
+= s
->cluster_size
;
963 ret
= update_refcount(bs
, offset
, size
, 1, false, QCOW2_DISCARD_NEVER
);
967 } while (ret
== -EAGAIN
);
972 /* The cluster refcount was incremented; refcount blocks must be flushed
973 * before the caller's L2 table updates. */
974 qcow2_cache_set_dependency(bs
, s
->l2_table_cache
, s
->refcount_block_cache
);
976 s
->free_byte_offset
= offset
+ size
;
977 if (!offset_into_cluster(s
, s
->free_byte_offset
)) {
978 s
->free_byte_offset
= 0;
984 void qcow2_free_clusters(BlockDriverState
*bs
,
985 int64_t offset
, int64_t size
,
986 enum qcow2_discard_type type
)
990 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_FREE
);
991 ret
= update_refcount(bs
, offset
, size
, 1, true, type
);
993 fprintf(stderr
, "qcow2_free_clusters failed: %s\n", strerror(-ret
));
994 /* TODO Remember the clusters to free them later and avoid leaking */
999 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
1000 * normal cluster, compressed cluster, etc.)
1002 void qcow2_free_any_clusters(BlockDriverState
*bs
, uint64_t l2_entry
,
1003 int nb_clusters
, enum qcow2_discard_type type
)
1005 BDRVQcow2State
*s
= bs
->opaque
;
1007 switch (qcow2_get_cluster_type(l2_entry
)) {
1008 case QCOW2_CLUSTER_COMPRESSED
:
1011 nb_csectors
= ((l2_entry
>> s
->csize_shift
) &
1013 qcow2_free_clusters(bs
,
1014 (l2_entry
& s
->cluster_offset_mask
) & ~511,
1015 nb_csectors
* 512, type
);
1018 case QCOW2_CLUSTER_NORMAL
:
1019 case QCOW2_CLUSTER_ZERO
:
1020 if (l2_entry
& L2E_OFFSET_MASK
) {
1021 if (offset_into_cluster(s
, l2_entry
& L2E_OFFSET_MASK
)) {
1022 qcow2_signal_corruption(bs
, false, -1, -1,
1023 "Cannot free unaligned cluster %#llx",
1024 l2_entry
& L2E_OFFSET_MASK
);
1026 qcow2_free_clusters(bs
, l2_entry
& L2E_OFFSET_MASK
,
1027 nb_clusters
<< s
->cluster_bits
, type
);
1031 case QCOW2_CLUSTER_UNALLOCATED
:
1040 /*********************************************************/
1041 /* snapshots and image creation */
1045 /* update the refcounts of snapshots and the copied flag */
1046 int qcow2_update_snapshot_refcount(BlockDriverState
*bs
,
1047 int64_t l1_table_offset
, int l1_size
, int addend
)
1049 BDRVQcow2State
*s
= bs
->opaque
;
1050 uint64_t *l1_table
, *l2_table
, l2_offset
, offset
, l1_size2
, refcount
;
1051 bool l1_allocated
= false;
1052 int64_t old_offset
, old_l2_offset
;
1053 int i
, j
, l1_modified
= 0, nb_csectors
;
1056 assert(addend
>= -1 && addend
<= 1);
1060 l1_size2
= l1_size
* sizeof(uint64_t);
1062 s
->cache_discards
= true;
1064 /* WARNING: qcow2_snapshot_goto relies on this function not using the
1065 * l1_table_offset when it is the current s->l1_table_offset! Be careful
1066 * when changing this! */
1067 if (l1_table_offset
!= s
->l1_table_offset
) {
1068 l1_table
= g_try_malloc0(align_offset(l1_size2
, 512));
1069 if (l1_size2
&& l1_table
== NULL
) {
1073 l1_allocated
= true;
1075 ret
= bdrv_pread(bs
->file
->bs
, l1_table_offset
, l1_table
, l1_size2
);
1080 for(i
= 0;i
< l1_size
; i
++)
1081 be64_to_cpus(&l1_table
[i
]);
1083 assert(l1_size
== s
->l1_size
);
1084 l1_table
= s
->l1_table
;
1085 l1_allocated
= false;
1088 for(i
= 0; i
< l1_size
; i
++) {
1089 l2_offset
= l1_table
[i
];
1091 old_l2_offset
= l2_offset
;
1092 l2_offset
&= L1E_OFFSET_MASK
;
1094 if (offset_into_cluster(s
, l2_offset
)) {
1095 qcow2_signal_corruption(bs
, true, -1, -1, "L2 table offset %#"
1096 PRIx64
" unaligned (L1 index: %#x)",
1102 ret
= qcow2_cache_get(bs
, s
->l2_table_cache
, l2_offset
,
1103 (void**) &l2_table
);
1108 for(j
= 0; j
< s
->l2_size
; j
++) {
1109 uint64_t cluster_index
;
1111 offset
= be64_to_cpu(l2_table
[j
]);
1112 old_offset
= offset
;
1113 offset
&= ~QCOW_OFLAG_COPIED
;
1115 switch (qcow2_get_cluster_type(offset
)) {
1116 case QCOW2_CLUSTER_COMPRESSED
:
1117 nb_csectors
= ((offset
>> s
->csize_shift
) &
1120 ret
= update_refcount(bs
,
1121 (offset
& s
->cluster_offset_mask
) & ~511,
1122 nb_csectors
* 512, abs(addend
), addend
< 0,
1123 QCOW2_DISCARD_SNAPSHOT
);
1128 /* compressed clusters are never modified */
1132 case QCOW2_CLUSTER_NORMAL
:
1133 case QCOW2_CLUSTER_ZERO
:
1134 if (offset_into_cluster(s
, offset
& L2E_OFFSET_MASK
)) {
1135 qcow2_signal_corruption(bs
, true, -1, -1, "Data "
1136 "cluster offset %#llx "
1137 "unaligned (L2 offset: %#"
1138 PRIx64
", L2 index: %#x)",
1139 offset
& L2E_OFFSET_MASK
,
1145 cluster_index
= (offset
& L2E_OFFSET_MASK
) >> s
->cluster_bits
;
1146 if (!cluster_index
) {
1152 ret
= qcow2_update_cluster_refcount(bs
,
1153 cluster_index
, abs(addend
), addend
< 0,
1154 QCOW2_DISCARD_SNAPSHOT
);
1160 ret
= qcow2_get_refcount(bs
, cluster_index
, &refcount
);
1166 case QCOW2_CLUSTER_UNALLOCATED
:
1174 if (refcount
== 1) {
1175 offset
|= QCOW_OFLAG_COPIED
;
1177 if (offset
!= old_offset
) {
1179 qcow2_cache_set_dependency(bs
, s
->l2_table_cache
,
1180 s
->refcount_block_cache
);
1182 l2_table
[j
] = cpu_to_be64(offset
);
1183 qcow2_cache_entry_mark_dirty(bs
, s
->l2_table_cache
,
1188 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &l2_table
);
1191 ret
= qcow2_update_cluster_refcount(bs
, l2_offset
>>
1193 abs(addend
), addend
< 0,
1194 QCOW2_DISCARD_SNAPSHOT
);
1199 ret
= qcow2_get_refcount(bs
, l2_offset
>> s
->cluster_bits
,
1203 } else if (refcount
== 1) {
1204 l2_offset
|= QCOW_OFLAG_COPIED
;
1206 if (l2_offset
!= old_l2_offset
) {
1207 l1_table
[i
] = l2_offset
;
1213 ret
= bdrv_flush(bs
);
1216 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
1219 s
->cache_discards
= false;
1220 qcow2_process_discards(bs
, ret
);
1222 /* Update L1 only if it isn't deleted anyway (addend = -1) */
1223 if (ret
== 0 && addend
>= 0 && l1_modified
) {
1224 for (i
= 0; i
< l1_size
; i
++) {
1225 cpu_to_be64s(&l1_table
[i
]);
1228 ret
= bdrv_pwrite_sync(bs
->file
->bs
, l1_table_offset
,
1229 l1_table
, l1_size2
);
1231 for (i
= 0; i
< l1_size
; i
++) {
1232 be64_to_cpus(&l1_table
[i
]);
1243 /*********************************************************/
1244 /* refcount checking functions */
1247 static uint64_t refcount_array_byte_size(BDRVQcow2State
*s
, uint64_t entries
)
1249 /* This assertion holds because there is no way we can address more than
1250 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because
1251 * offsets have to be representable in bytes); due to every cluster
1252 * corresponding to one refcount entry, we are well below that limit */
1253 assert(entries
< (UINT64_C(1) << (64 - 9)));
1255 /* Thanks to the assertion this will not overflow, because
1256 * s->refcount_order < 7.
1257 * (note: x << s->refcount_order == x * s->refcount_bits) */
1258 return DIV_ROUND_UP(entries
<< s
->refcount_order
, 8);
1262 * Reallocates *array so that it can hold new_size entries. *size must contain
1263 * the current number of entries in *array. If the reallocation fails, *array
1264 * and *size will not be modified and -errno will be returned. If the
1265 * reallocation is successful, *array will be set to the new buffer, *size
1266 * will be set to new_size and 0 will be returned. The size of the reallocated
1267 * refcount array buffer will be aligned to a cluster boundary, and the newly
1268 * allocated area will be zeroed.
1270 static int realloc_refcount_array(BDRVQcow2State
*s
, void **array
,
1271 int64_t *size
, int64_t new_size
)
1273 int64_t old_byte_size
, new_byte_size
;
1276 /* Round to clusters so the array can be directly written to disk */
1277 old_byte_size
= size_to_clusters(s
, refcount_array_byte_size(s
, *size
))
1279 new_byte_size
= size_to_clusters(s
, refcount_array_byte_size(s
, new_size
))
1282 if (new_byte_size
== old_byte_size
) {
1287 assert(new_byte_size
> 0);
1289 if (new_byte_size
> SIZE_MAX
) {
1293 new_ptr
= g_try_realloc(*array
, new_byte_size
);
1298 if (new_byte_size
> old_byte_size
) {
1299 memset((char *)new_ptr
+ old_byte_size
, 0,
1300 new_byte_size
- old_byte_size
);
1310 * Increases the refcount for a range of clusters in a given refcount table.
1311 * This is used to construct a temporary refcount table out of L1 and L2 tables
1312 * which can be compared to the refcount table saved in the image.
1314 * Modifies the number of errors in res.
1316 static int inc_refcounts(BlockDriverState
*bs
,
1317 BdrvCheckResult
*res
,
1318 void **refcount_table
,
1319 int64_t *refcount_table_size
,
1320 int64_t offset
, int64_t size
)
1322 BDRVQcow2State
*s
= bs
->opaque
;
1323 uint64_t start
, last
, cluster_offset
, k
, refcount
;
1330 start
= start_of_cluster(s
, offset
);
1331 last
= start_of_cluster(s
, offset
+ size
- 1);
1332 for(cluster_offset
= start
; cluster_offset
<= last
;
1333 cluster_offset
+= s
->cluster_size
) {
1334 k
= cluster_offset
>> s
->cluster_bits
;
1335 if (k
>= *refcount_table_size
) {
1336 ret
= realloc_refcount_array(s
, refcount_table
,
1337 refcount_table_size
, k
+ 1);
1339 res
->check_errors
++;
1344 refcount
= s
->get_refcount(*refcount_table
, k
);
1345 if (refcount
== s
->refcount_max
) {
1346 fprintf(stderr
, "ERROR: overflow cluster offset=0x%" PRIx64
1347 "\n", cluster_offset
);
1348 fprintf(stderr
, "Use qemu-img amend to increase the refcount entry "
1349 "width or qemu-img convert to create a clean copy if the "
1350 "image cannot be opened for writing\n");
1354 s
->set_refcount(*refcount_table
, k
, refcount
+ 1);
1360 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
1362 CHECK_FRAG_INFO
= 0x2, /* update BlockFragInfo counters */
1366 * Increases the refcount in the given refcount table for the all clusters
1367 * referenced in the L2 table. While doing so, performs some checks on L2
1370 * Returns the number of errors found by the checks or -errno if an internal
1373 static int check_refcounts_l2(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1374 void **refcount_table
,
1375 int64_t *refcount_table_size
, int64_t l2_offset
,
1378 BDRVQcow2State
*s
= bs
->opaque
;
1379 uint64_t *l2_table
, l2_entry
;
1380 uint64_t next_contiguous_offset
= 0;
1381 int i
, l2_size
, nb_csectors
, ret
;
1383 /* Read L2 table from disk */
1384 l2_size
= s
->l2_size
* sizeof(uint64_t);
1385 l2_table
= g_malloc(l2_size
);
1387 ret
= bdrv_pread(bs
->file
->bs
, l2_offset
, l2_table
, l2_size
);
1389 fprintf(stderr
, "ERROR: I/O error in check_refcounts_l2\n");
1390 res
->check_errors
++;
1394 /* Do the actual checks */
1395 for(i
= 0; i
< s
->l2_size
; i
++) {
1396 l2_entry
= be64_to_cpu(l2_table
[i
]);
1398 switch (qcow2_get_cluster_type(l2_entry
)) {
1399 case QCOW2_CLUSTER_COMPRESSED
:
1400 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1401 if (l2_entry
& QCOW_OFLAG_COPIED
) {
1402 fprintf(stderr
, "ERROR: cluster %" PRId64
": "
1403 "copied flag must never be set for compressed "
1404 "clusters\n", l2_entry
>> s
->cluster_bits
);
1405 l2_entry
&= ~QCOW_OFLAG_COPIED
;
1409 /* Mark cluster as used */
1410 nb_csectors
= ((l2_entry
>> s
->csize_shift
) &
1412 l2_entry
&= s
->cluster_offset_mask
;
1413 ret
= inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1414 l2_entry
& ~511, nb_csectors
* 512);
1419 if (flags
& CHECK_FRAG_INFO
) {
1420 res
->bfi
.allocated_clusters
++;
1421 res
->bfi
.compressed_clusters
++;
1423 /* Compressed clusters are fragmented by nature. Since they
1424 * take up sub-sector space but we only have sector granularity
1425 * I/O we need to re-read the same sectors even for adjacent
1426 * compressed clusters.
1428 res
->bfi
.fragmented_clusters
++;
1432 case QCOW2_CLUSTER_ZERO
:
1433 if ((l2_entry
& L2E_OFFSET_MASK
) == 0) {
1438 case QCOW2_CLUSTER_NORMAL
:
1440 uint64_t offset
= l2_entry
& L2E_OFFSET_MASK
;
1442 if (flags
& CHECK_FRAG_INFO
) {
1443 res
->bfi
.allocated_clusters
++;
1444 if (next_contiguous_offset
&&
1445 offset
!= next_contiguous_offset
) {
1446 res
->bfi
.fragmented_clusters
++;
1448 next_contiguous_offset
= offset
+ s
->cluster_size
;
1451 /* Mark cluster as used */
1452 ret
= inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1453 offset
, s
->cluster_size
);
1458 /* Correct offsets are cluster aligned */
1459 if (offset_into_cluster(s
, offset
)) {
1460 fprintf(stderr
, "ERROR offset=%" PRIx64
": Cluster is not "
1461 "properly aligned; L2 entry corrupted.\n", offset
);
1467 case QCOW2_CLUSTER_UNALLOCATED
:
1484 * Increases the refcount for the L1 table, its L2 tables and all referenced
1485 * clusters in the given refcount table. While doing so, performs some checks
1486 * on L1 and L2 entries.
1488 * Returns the number of errors found by the checks or -errno if an internal
1491 static int check_refcounts_l1(BlockDriverState
*bs
,
1492 BdrvCheckResult
*res
,
1493 void **refcount_table
,
1494 int64_t *refcount_table_size
,
1495 int64_t l1_table_offset
, int l1_size
,
1498 BDRVQcow2State
*s
= bs
->opaque
;
1499 uint64_t *l1_table
= NULL
, l2_offset
, l1_size2
;
1502 l1_size2
= l1_size
* sizeof(uint64_t);
1504 /* Mark L1 table as used */
1505 ret
= inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1506 l1_table_offset
, l1_size2
);
1511 /* Read L1 table entries from disk */
1513 l1_table
= g_try_malloc(l1_size2
);
1514 if (l1_table
== NULL
) {
1516 res
->check_errors
++;
1519 ret
= bdrv_pread(bs
->file
->bs
, l1_table_offset
, l1_table
, l1_size2
);
1521 fprintf(stderr
, "ERROR: I/O error in check_refcounts_l1\n");
1522 res
->check_errors
++;
1525 for(i
= 0;i
< l1_size
; i
++)
1526 be64_to_cpus(&l1_table
[i
]);
1529 /* Do the actual checks */
1530 for(i
= 0; i
< l1_size
; i
++) {
1531 l2_offset
= l1_table
[i
];
1533 /* Mark L2 table as used */
1534 l2_offset
&= L1E_OFFSET_MASK
;
1535 ret
= inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1536 l2_offset
, s
->cluster_size
);
1541 /* L2 tables are cluster aligned */
1542 if (offset_into_cluster(s
, l2_offset
)) {
1543 fprintf(stderr
, "ERROR l2_offset=%" PRIx64
": Table is not "
1544 "cluster aligned; L1 entry corrupted\n", l2_offset
);
1548 /* Process and check L2 entries */
1549 ret
= check_refcounts_l2(bs
, res
, refcount_table
,
1550 refcount_table_size
, l2_offset
, flags
);
1565 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1567 * This function does not print an error message nor does it increment
1568 * check_errors if qcow2_get_refcount fails (this is because such an error will
1569 * have been already detected and sufficiently signaled by the calling function
1570 * (qcow2_check_refcounts) by the time this function is called).
1572 static int check_oflag_copied(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1575 BDRVQcow2State
*s
= bs
->opaque
;
1576 uint64_t *l2_table
= qemu_blockalign(bs
, s
->cluster_size
);
1581 for (i
= 0; i
< s
->l1_size
; i
++) {
1582 uint64_t l1_entry
= s
->l1_table
[i
];
1583 uint64_t l2_offset
= l1_entry
& L1E_OFFSET_MASK
;
1584 bool l2_dirty
= false;
1590 ret
= qcow2_get_refcount(bs
, l2_offset
>> s
->cluster_bits
,
1593 /* don't print message nor increment check_errors */
1596 if ((refcount
== 1) != ((l1_entry
& QCOW_OFLAG_COPIED
) != 0)) {
1597 fprintf(stderr
, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1598 "l1_entry=%" PRIx64
" refcount=%" PRIu64
"\n",
1599 fix
& BDRV_FIX_ERRORS
? "Repairing" :
1601 i
, l1_entry
, refcount
);
1602 if (fix
& BDRV_FIX_ERRORS
) {
1603 s
->l1_table
[i
] = refcount
== 1
1604 ? l1_entry
| QCOW_OFLAG_COPIED
1605 : l1_entry
& ~QCOW_OFLAG_COPIED
;
1606 ret
= qcow2_write_l1_entry(bs
, i
);
1608 res
->check_errors
++;
1611 res
->corruptions_fixed
++;
1617 ret
= bdrv_pread(bs
->file
->bs
, l2_offset
, l2_table
,
1618 s
->l2_size
* sizeof(uint64_t));
1620 fprintf(stderr
, "ERROR: Could not read L2 table: %s\n",
1622 res
->check_errors
++;
1626 for (j
= 0; j
< s
->l2_size
; j
++) {
1627 uint64_t l2_entry
= be64_to_cpu(l2_table
[j
]);
1628 uint64_t data_offset
= l2_entry
& L2E_OFFSET_MASK
;
1629 int cluster_type
= qcow2_get_cluster_type(l2_entry
);
1631 if ((cluster_type
== QCOW2_CLUSTER_NORMAL
) ||
1632 ((cluster_type
== QCOW2_CLUSTER_ZERO
) && (data_offset
!= 0))) {
1633 ret
= qcow2_get_refcount(bs
,
1634 data_offset
>> s
->cluster_bits
,
1637 /* don't print message nor increment check_errors */
1640 if ((refcount
== 1) != ((l2_entry
& QCOW_OFLAG_COPIED
) != 0)) {
1641 fprintf(stderr
, "%s OFLAG_COPIED data cluster: "
1642 "l2_entry=%" PRIx64
" refcount=%" PRIu64
"\n",
1643 fix
& BDRV_FIX_ERRORS
? "Repairing" :
1645 l2_entry
, refcount
);
1646 if (fix
& BDRV_FIX_ERRORS
) {
1647 l2_table
[j
] = cpu_to_be64(refcount
== 1
1648 ? l2_entry
| QCOW_OFLAG_COPIED
1649 : l2_entry
& ~QCOW_OFLAG_COPIED
);
1651 res
->corruptions_fixed
++;
1660 ret
= qcow2_pre_write_overlap_check(bs
, QCOW2_OL_ACTIVE_L2
,
1661 l2_offset
, s
->cluster_size
);
1663 fprintf(stderr
, "ERROR: Could not write L2 table; metadata "
1664 "overlap check failed: %s\n", strerror(-ret
));
1665 res
->check_errors
++;
1669 ret
= bdrv_pwrite(bs
->file
->bs
, l2_offset
, l2_table
,
1672 fprintf(stderr
, "ERROR: Could not write L2 table: %s\n",
1674 res
->check_errors
++;
1683 qemu_vfree(l2_table
);
1688 * Checks consistency of refblocks and accounts for each refblock in
1691 static int check_refblocks(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1692 BdrvCheckMode fix
, bool *rebuild
,
1693 void **refcount_table
, int64_t *nb_clusters
)
1695 BDRVQcow2State
*s
= bs
->opaque
;
1699 for(i
= 0; i
< s
->refcount_table_size
; i
++) {
1700 uint64_t offset
, cluster
;
1701 offset
= s
->refcount_table
[i
];
1702 cluster
= offset
>> s
->cluster_bits
;
1704 /* Refcount blocks are cluster aligned */
1705 if (offset_into_cluster(s
, offset
)) {
1706 fprintf(stderr
, "ERROR refcount block %" PRId64
" is not "
1707 "cluster aligned; refcount table entry corrupted\n", i
);
1713 if (cluster
>= *nb_clusters
) {
1714 fprintf(stderr
, "%s refcount block %" PRId64
" is outside image\n",
1715 fix
& BDRV_FIX_ERRORS
? "Repairing" : "ERROR", i
);
1717 if (fix
& BDRV_FIX_ERRORS
) {
1718 int64_t new_nb_clusters
;
1720 if (offset
> INT64_MAX
- s
->cluster_size
) {
1725 ret
= bdrv_truncate(bs
->file
->bs
, offset
+ s
->cluster_size
);
1729 size
= bdrv_getlength(bs
->file
->bs
);
1735 new_nb_clusters
= size_to_clusters(s
, size
);
1736 assert(new_nb_clusters
>= *nb_clusters
);
1738 ret
= realloc_refcount_array(s
, refcount_table
,
1739 nb_clusters
, new_nb_clusters
);
1741 res
->check_errors
++;
1745 if (cluster
>= *nb_clusters
) {
1750 res
->corruptions_fixed
++;
1751 ret
= inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1752 offset
, s
->cluster_size
);
1756 /* No need to check whether the refcount is now greater than 1:
1757 * This area was just allocated and zeroed, so it can only be
1758 * exactly 1 after inc_refcounts() */
1764 fprintf(stderr
, "ERROR could not resize image: %s\n",
1773 ret
= inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1774 offset
, s
->cluster_size
);
1778 if (s
->get_refcount(*refcount_table
, cluster
) != 1) {
1779 fprintf(stderr
, "ERROR refcount block %" PRId64
1780 " refcount=%" PRIu64
"\n", i
,
1781 s
->get_refcount(*refcount_table
, cluster
));
1792 * Calculates an in-memory refcount table.
1794 static int calculate_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1795 BdrvCheckMode fix
, bool *rebuild
,
1796 void **refcount_table
, int64_t *nb_clusters
)
1798 BDRVQcow2State
*s
= bs
->opaque
;
1803 if (!*refcount_table
) {
1804 int64_t old_size
= 0;
1805 ret
= realloc_refcount_array(s
, refcount_table
,
1806 &old_size
, *nb_clusters
);
1808 res
->check_errors
++;
1814 ret
= inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1815 0, s
->cluster_size
);
1820 /* current L1 table */
1821 ret
= check_refcounts_l1(bs
, res
, refcount_table
, nb_clusters
,
1822 s
->l1_table_offset
, s
->l1_size
, CHECK_FRAG_INFO
);
1828 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
1829 sn
= s
->snapshots
+ i
;
1830 ret
= check_refcounts_l1(bs
, res
, refcount_table
, nb_clusters
,
1831 sn
->l1_table_offset
, sn
->l1_size
, 0);
1836 ret
= inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1837 s
->snapshots_offset
, s
->snapshots_size
);
1843 ret
= inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1844 s
->refcount_table_offset
,
1845 s
->refcount_table_size
* sizeof(uint64_t));
1850 return check_refblocks(bs
, res
, fix
, rebuild
, refcount_table
, nb_clusters
);
1854 * Compares the actual reference count for each cluster in the image against the
1855 * refcount as reported by the refcount structures on-disk.
1857 static void compare_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1858 BdrvCheckMode fix
, bool *rebuild
,
1859 int64_t *highest_cluster
,
1860 void *refcount_table
, int64_t nb_clusters
)
1862 BDRVQcow2State
*s
= bs
->opaque
;
1864 uint64_t refcount1
, refcount2
;
1867 for (i
= 0, *highest_cluster
= 0; i
< nb_clusters
; i
++) {
1868 ret
= qcow2_get_refcount(bs
, i
, &refcount1
);
1870 fprintf(stderr
, "Can't get refcount for cluster %" PRId64
": %s\n",
1872 res
->check_errors
++;
1876 refcount2
= s
->get_refcount(refcount_table
, i
);
1878 if (refcount1
> 0 || refcount2
> 0) {
1879 *highest_cluster
= i
;
1882 if (refcount1
!= refcount2
) {
1883 /* Check if we're allowed to fix the mismatch */
1884 int *num_fixed
= NULL
;
1885 if (refcount1
== 0) {
1887 } else if (refcount1
> refcount2
&& (fix
& BDRV_FIX_LEAKS
)) {
1888 num_fixed
= &res
->leaks_fixed
;
1889 } else if (refcount1
< refcount2
&& (fix
& BDRV_FIX_ERRORS
)) {
1890 num_fixed
= &res
->corruptions_fixed
;
1893 fprintf(stderr
, "%s cluster %" PRId64
" refcount=%" PRIu64
1894 " reference=%" PRIu64
"\n",
1895 num_fixed
!= NULL
? "Repairing" :
1896 refcount1
< refcount2
? "ERROR" :
1898 i
, refcount1
, refcount2
);
1901 ret
= update_refcount(bs
, i
<< s
->cluster_bits
, 1,
1902 refcount_diff(refcount1
, refcount2
),
1903 refcount1
> refcount2
,
1904 QCOW2_DISCARD_ALWAYS
);
1911 /* And if we couldn't, print an error */
1912 if (refcount1
< refcount2
) {
1922 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to
1923 * the on-disk refcount structures.
1925 * On input, *first_free_cluster tells where to start looking, and need not
1926 * actually be a free cluster; the returned offset will not be before that
1927 * cluster. On output, *first_free_cluster points to the first gap found, even
1928 * if that gap was too small to be used as the returned offset.
1930 * Note that *first_free_cluster is a cluster index whereas the return value is
1933 static int64_t alloc_clusters_imrt(BlockDriverState
*bs
,
1935 void **refcount_table
,
1936 int64_t *imrt_nb_clusters
,
1937 int64_t *first_free_cluster
)
1939 BDRVQcow2State
*s
= bs
->opaque
;
1940 int64_t cluster
= *first_free_cluster
, i
;
1941 bool first_gap
= true;
1942 int contiguous_free_clusters
;
1945 /* Starting at *first_free_cluster, find a range of at least cluster_count
1946 * continuously free clusters */
1947 for (contiguous_free_clusters
= 0;
1948 cluster
< *imrt_nb_clusters
&&
1949 contiguous_free_clusters
< cluster_count
;
1952 if (!s
->get_refcount(*refcount_table
, cluster
)) {
1953 contiguous_free_clusters
++;
1955 /* If this is the first free cluster found, update
1956 * *first_free_cluster accordingly */
1957 *first_free_cluster
= cluster
;
1960 } else if (contiguous_free_clusters
) {
1961 contiguous_free_clusters
= 0;
1965 /* If contiguous_free_clusters is greater than zero, it contains the number
1966 * of continuously free clusters until the current cluster; the first free
1967 * cluster in the current "gap" is therefore
1968 * cluster - contiguous_free_clusters */
1970 /* If no such range could be found, grow the in-memory refcount table
1971 * accordingly to append free clusters at the end of the image */
1972 if (contiguous_free_clusters
< cluster_count
) {
1973 /* contiguous_free_clusters clusters are already empty at the image end;
1974 * we need cluster_count clusters; therefore, we have to allocate
1975 * cluster_count - contiguous_free_clusters new clusters at the end of
1976 * the image (which is the current value of cluster; note that cluster
1977 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond
1979 ret
= realloc_refcount_array(s
, refcount_table
, imrt_nb_clusters
,
1980 cluster
+ cluster_count
1981 - contiguous_free_clusters
);
1987 /* Go back to the first free cluster */
1988 cluster
-= contiguous_free_clusters
;
1989 for (i
= 0; i
< cluster_count
; i
++) {
1990 s
->set_refcount(*refcount_table
, cluster
+ i
, 1);
1993 return cluster
<< s
->cluster_bits
;
1997 * Creates a new refcount structure based solely on the in-memory information
1998 * given through *refcount_table. All necessary allocations will be reflected
2001 * On success, the old refcount structure is leaked (it will be covered by the
2002 * new refcount structure).
2004 static int rebuild_refcount_structure(BlockDriverState
*bs
,
2005 BdrvCheckResult
*res
,
2006 void **refcount_table
,
2007 int64_t *nb_clusters
)
2009 BDRVQcow2State
*s
= bs
->opaque
;
2010 int64_t first_free_cluster
= 0, reftable_offset
= -1, cluster
= 0;
2011 int64_t refblock_offset
, refblock_start
, refblock_index
;
2012 uint32_t reftable_size
= 0;
2013 uint64_t *on_disk_reftable
= NULL
;
2014 void *on_disk_refblock
;
2017 uint64_t reftable_offset
;
2018 uint32_t reftable_clusters
;
2019 } QEMU_PACKED reftable_offset_and_clusters
;
2021 qcow2_cache_empty(bs
, s
->refcount_block_cache
);
2024 for (; cluster
< *nb_clusters
; cluster
++) {
2025 if (!s
->get_refcount(*refcount_table
, cluster
)) {
2029 refblock_index
= cluster
>> s
->refcount_block_bits
;
2030 refblock_start
= refblock_index
<< s
->refcount_block_bits
;
2032 /* Don't allocate a cluster in a refblock already written to disk */
2033 if (first_free_cluster
< refblock_start
) {
2034 first_free_cluster
= refblock_start
;
2036 refblock_offset
= alloc_clusters_imrt(bs
, 1, refcount_table
,
2037 nb_clusters
, &first_free_cluster
);
2038 if (refblock_offset
< 0) {
2039 fprintf(stderr
, "ERROR allocating refblock: %s\n",
2040 strerror(-refblock_offset
));
2041 res
->check_errors
++;
2042 ret
= refblock_offset
;
2046 if (reftable_size
<= refblock_index
) {
2047 uint32_t old_reftable_size
= reftable_size
;
2048 uint64_t *new_on_disk_reftable
;
2050 reftable_size
= ROUND_UP((refblock_index
+ 1) * sizeof(uint64_t),
2051 s
->cluster_size
) / sizeof(uint64_t);
2052 new_on_disk_reftable
= g_try_realloc(on_disk_reftable
,
2055 if (!new_on_disk_reftable
) {
2056 res
->check_errors
++;
2060 on_disk_reftable
= new_on_disk_reftable
;
2062 memset(on_disk_reftable
+ old_reftable_size
, 0,
2063 (reftable_size
- old_reftable_size
) * sizeof(uint64_t));
2065 /* The offset we have for the reftable is now no longer valid;
2066 * this will leak that range, but we can easily fix that by running
2067 * a leak-fixing check after this rebuild operation */
2068 reftable_offset
= -1;
2070 on_disk_reftable
[refblock_index
] = refblock_offset
;
2072 /* If this is apparently the last refblock (for now), try to squeeze the
2074 if (refblock_index
== (*nb_clusters
- 1) >> s
->refcount_block_bits
&&
2075 reftable_offset
< 0)
2077 uint64_t reftable_clusters
= size_to_clusters(s
, reftable_size
*
2079 reftable_offset
= alloc_clusters_imrt(bs
, reftable_clusters
,
2080 refcount_table
, nb_clusters
,
2081 &first_free_cluster
);
2082 if (reftable_offset
< 0) {
2083 fprintf(stderr
, "ERROR allocating reftable: %s\n",
2084 strerror(-reftable_offset
));
2085 res
->check_errors
++;
2086 ret
= reftable_offset
;
2091 ret
= qcow2_pre_write_overlap_check(bs
, 0, refblock_offset
,
2094 fprintf(stderr
, "ERROR writing refblock: %s\n", strerror(-ret
));
2098 /* The size of *refcount_table is always cluster-aligned, therefore the
2099 * write operation will not overflow */
2100 on_disk_refblock
= (void *)((char *) *refcount_table
+
2101 refblock_index
* s
->cluster_size
);
2103 ret
= bdrv_write(bs
->file
->bs
, refblock_offset
/ BDRV_SECTOR_SIZE
,
2104 on_disk_refblock
, s
->cluster_sectors
);
2106 fprintf(stderr
, "ERROR writing refblock: %s\n", strerror(-ret
));
2110 /* Go to the end of this refblock */
2111 cluster
= refblock_start
+ s
->refcount_block_size
- 1;
2114 if (reftable_offset
< 0) {
2115 uint64_t post_refblock_start
, reftable_clusters
;
2117 post_refblock_start
= ROUND_UP(*nb_clusters
, s
->refcount_block_size
);
2118 reftable_clusters
= size_to_clusters(s
,
2119 reftable_size
* sizeof(uint64_t));
2120 /* Not pretty but simple */
2121 if (first_free_cluster
< post_refblock_start
) {
2122 first_free_cluster
= post_refblock_start
;
2124 reftable_offset
= alloc_clusters_imrt(bs
, reftable_clusters
,
2125 refcount_table
, nb_clusters
,
2126 &first_free_cluster
);
2127 if (reftable_offset
< 0) {
2128 fprintf(stderr
, "ERROR allocating reftable: %s\n",
2129 strerror(-reftable_offset
));
2130 res
->check_errors
++;
2131 ret
= reftable_offset
;
2135 goto write_refblocks
;
2138 assert(on_disk_reftable
);
2140 for (refblock_index
= 0; refblock_index
< reftable_size
; refblock_index
++) {
2141 cpu_to_be64s(&on_disk_reftable
[refblock_index
]);
2144 ret
= qcow2_pre_write_overlap_check(bs
, 0, reftable_offset
,
2145 reftable_size
* sizeof(uint64_t));
2147 fprintf(stderr
, "ERROR writing reftable: %s\n", strerror(-ret
));
2151 assert(reftable_size
< INT_MAX
/ sizeof(uint64_t));
2152 ret
= bdrv_pwrite(bs
->file
->bs
, reftable_offset
, on_disk_reftable
,
2153 reftable_size
* sizeof(uint64_t));
2155 fprintf(stderr
, "ERROR writing reftable: %s\n", strerror(-ret
));
2159 /* Enter new reftable into the image header */
2160 cpu_to_be64w(&reftable_offset_and_clusters
.reftable_offset
,
2162 cpu_to_be32w(&reftable_offset_and_clusters
.reftable_clusters
,
2163 size_to_clusters(s
, reftable_size
* sizeof(uint64_t)));
2164 ret
= bdrv_pwrite_sync(bs
->file
->bs
, offsetof(QCowHeader
,
2165 refcount_table_offset
),
2166 &reftable_offset_and_clusters
,
2167 sizeof(reftable_offset_and_clusters
));
2169 fprintf(stderr
, "ERROR setting reftable: %s\n", strerror(-ret
));
2173 for (refblock_index
= 0; refblock_index
< reftable_size
; refblock_index
++) {
2174 be64_to_cpus(&on_disk_reftable
[refblock_index
]);
2176 s
->refcount_table
= on_disk_reftable
;
2177 s
->refcount_table_offset
= reftable_offset
;
2178 s
->refcount_table_size
= reftable_size
;
2183 g_free(on_disk_reftable
);
2188 * Checks an image for refcount consistency.
2190 * Returns 0 if no errors are found, the number of errors in case the image is
2191 * detected as corrupted, and -errno when an internal error occurred.
2193 int qcow2_check_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
2196 BDRVQcow2State
*s
= bs
->opaque
;
2197 BdrvCheckResult pre_compare_res
;
2198 int64_t size
, highest_cluster
, nb_clusters
;
2199 void *refcount_table
= NULL
;
2200 bool rebuild
= false;
2203 size
= bdrv_getlength(bs
->file
->bs
);
2205 res
->check_errors
++;
2209 nb_clusters
= size_to_clusters(s
, size
);
2210 if (nb_clusters
> INT_MAX
) {
2211 res
->check_errors
++;
2215 res
->bfi
.total_clusters
=
2216 size_to_clusters(s
, bs
->total_sectors
* BDRV_SECTOR_SIZE
);
2218 ret
= calculate_refcounts(bs
, res
, fix
, &rebuild
, &refcount_table
,
2224 /* In case we don't need to rebuild the refcount structure (but want to fix
2225 * something), this function is immediately called again, in which case the
2226 * result should be ignored */
2227 pre_compare_res
= *res
;
2228 compare_refcounts(bs
, res
, 0, &rebuild
, &highest_cluster
, refcount_table
,
2231 if (rebuild
&& (fix
& BDRV_FIX_ERRORS
)) {
2232 BdrvCheckResult old_res
= *res
;
2233 int fresh_leaks
= 0;
2235 fprintf(stderr
, "Rebuilding refcount structure\n");
2236 ret
= rebuild_refcount_structure(bs
, res
, &refcount_table
,
2242 res
->corruptions
= 0;
2245 /* Because the old reftable has been exchanged for a new one the
2246 * references have to be recalculated */
2248 memset(refcount_table
, 0, refcount_array_byte_size(s
, nb_clusters
));
2249 ret
= calculate_refcounts(bs
, res
, 0, &rebuild
, &refcount_table
,
2255 if (fix
& BDRV_FIX_LEAKS
) {
2256 /* The old refcount structures are now leaked, fix it; the result
2257 * can be ignored, aside from leaks which were introduced by
2258 * rebuild_refcount_structure() that could not be fixed */
2259 BdrvCheckResult saved_res
= *res
;
2260 *res
= (BdrvCheckResult
){ 0 };
2262 compare_refcounts(bs
, res
, BDRV_FIX_LEAKS
, &rebuild
,
2263 &highest_cluster
, refcount_table
, nb_clusters
);
2265 fprintf(stderr
, "ERROR rebuilt refcount structure is still "
2269 /* Any leaks accounted for here were introduced by
2270 * rebuild_refcount_structure() because that function has created a
2271 * new refcount structure from scratch */
2272 fresh_leaks
= res
->leaks
;
2276 if (res
->corruptions
< old_res
.corruptions
) {
2277 res
->corruptions_fixed
+= old_res
.corruptions
- res
->corruptions
;
2279 if (res
->leaks
< old_res
.leaks
) {
2280 res
->leaks_fixed
+= old_res
.leaks
- res
->leaks
;
2282 res
->leaks
+= fresh_leaks
;
2285 fprintf(stderr
, "ERROR need to rebuild refcount structures\n");
2286 res
->check_errors
++;
2291 if (res
->leaks
|| res
->corruptions
) {
2292 *res
= pre_compare_res
;
2293 compare_refcounts(bs
, res
, fix
, &rebuild
, &highest_cluster
,
2294 refcount_table
, nb_clusters
);
2298 /* check OFLAG_COPIED */
2299 ret
= check_oflag_copied(bs
, res
, fix
);
2304 res
->image_end_offset
= (highest_cluster
+ 1) * s
->cluster_size
;
2308 g_free(refcount_table
);
2313 #define overlaps_with(ofs, sz) \
2314 ranges_overlap(offset, size, ofs, sz)
2317 * Checks if the given offset into the image file is actually free to use by
2318 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
2319 * i.e. a sanity check without relying on the refcount tables.
2321 * The ign parameter specifies what checks not to perform (being a bitmask of
2322 * QCow2MetadataOverlap values), i.e., what sections to ignore.
2325 * - 0 if writing to this offset will not affect the mentioned metadata
2326 * - a positive QCow2MetadataOverlap value indicating one overlapping section
2327 * - a negative value (-errno) indicating an error while performing a check,
2328 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
2330 int qcow2_check_metadata_overlap(BlockDriverState
*bs
, int ign
, int64_t offset
,
2333 BDRVQcow2State
*s
= bs
->opaque
;
2334 int chk
= s
->overlap_check
& ~ign
;
2341 if (chk
& QCOW2_OL_MAIN_HEADER
) {
2342 if (offset
< s
->cluster_size
) {
2343 return QCOW2_OL_MAIN_HEADER
;
2347 /* align range to test to cluster boundaries */
2348 size
= align_offset(offset_into_cluster(s
, offset
) + size
, s
->cluster_size
);
2349 offset
= start_of_cluster(s
, offset
);
2351 if ((chk
& QCOW2_OL_ACTIVE_L1
) && s
->l1_size
) {
2352 if (overlaps_with(s
->l1_table_offset
, s
->l1_size
* sizeof(uint64_t))) {
2353 return QCOW2_OL_ACTIVE_L1
;
2357 if ((chk
& QCOW2_OL_REFCOUNT_TABLE
) && s
->refcount_table_size
) {
2358 if (overlaps_with(s
->refcount_table_offset
,
2359 s
->refcount_table_size
* sizeof(uint64_t))) {
2360 return QCOW2_OL_REFCOUNT_TABLE
;
2364 if ((chk
& QCOW2_OL_SNAPSHOT_TABLE
) && s
->snapshots_size
) {
2365 if (overlaps_with(s
->snapshots_offset
, s
->snapshots_size
)) {
2366 return QCOW2_OL_SNAPSHOT_TABLE
;
2370 if ((chk
& QCOW2_OL_INACTIVE_L1
) && s
->snapshots
) {
2371 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
2372 if (s
->snapshots
[i
].l1_size
&&
2373 overlaps_with(s
->snapshots
[i
].l1_table_offset
,
2374 s
->snapshots
[i
].l1_size
* sizeof(uint64_t))) {
2375 return QCOW2_OL_INACTIVE_L1
;
2380 if ((chk
& QCOW2_OL_ACTIVE_L2
) && s
->l1_table
) {
2381 for (i
= 0; i
< s
->l1_size
; i
++) {
2382 if ((s
->l1_table
[i
] & L1E_OFFSET_MASK
) &&
2383 overlaps_with(s
->l1_table
[i
] & L1E_OFFSET_MASK
,
2385 return QCOW2_OL_ACTIVE_L2
;
2390 if ((chk
& QCOW2_OL_REFCOUNT_BLOCK
) && s
->refcount_table
) {
2391 for (i
= 0; i
< s
->refcount_table_size
; i
++) {
2392 if ((s
->refcount_table
[i
] & REFT_OFFSET_MASK
) &&
2393 overlaps_with(s
->refcount_table
[i
] & REFT_OFFSET_MASK
,
2395 return QCOW2_OL_REFCOUNT_BLOCK
;
2400 if ((chk
& QCOW2_OL_INACTIVE_L2
) && s
->snapshots
) {
2401 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
2402 uint64_t l1_ofs
= s
->snapshots
[i
].l1_table_offset
;
2403 uint32_t l1_sz
= s
->snapshots
[i
].l1_size
;
2404 uint64_t l1_sz2
= l1_sz
* sizeof(uint64_t);
2405 uint64_t *l1
= g_try_malloc(l1_sz2
);
2408 if (l1_sz2
&& l1
== NULL
) {
2412 ret
= bdrv_pread(bs
->file
->bs
, l1_ofs
, l1
, l1_sz2
);
2418 for (j
= 0; j
< l1_sz
; j
++) {
2419 uint64_t l2_ofs
= be64_to_cpu(l1
[j
]) & L1E_OFFSET_MASK
;
2420 if (l2_ofs
&& overlaps_with(l2_ofs
, s
->cluster_size
)) {
2422 return QCOW2_OL_INACTIVE_L2
;
2433 static const char *metadata_ol_names
[] = {
2434 [QCOW2_OL_MAIN_HEADER_BITNR
] = "qcow2_header",
2435 [QCOW2_OL_ACTIVE_L1_BITNR
] = "active L1 table",
2436 [QCOW2_OL_ACTIVE_L2_BITNR
] = "active L2 table",
2437 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = "refcount table",
2438 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = "refcount block",
2439 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = "snapshot table",
2440 [QCOW2_OL_INACTIVE_L1_BITNR
] = "inactive L1 table",
2441 [QCOW2_OL_INACTIVE_L2_BITNR
] = "inactive L2 table",
2445 * First performs a check for metadata overlaps (through
2446 * qcow2_check_metadata_overlap); if that fails with a negative value (error
2447 * while performing a check), that value is returned. If an impending overlap
2448 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
2449 * and -EIO returned.
2451 * Returns 0 if there were neither overlaps nor errors while checking for
2452 * overlaps; or a negative value (-errno) on error.
2454 int qcow2_pre_write_overlap_check(BlockDriverState
*bs
, int ign
, int64_t offset
,
2457 int ret
= qcow2_check_metadata_overlap(bs
, ign
, offset
, size
);
2461 } else if (ret
> 0) {
2462 int metadata_ol_bitnr
= ctz32(ret
);
2463 assert(metadata_ol_bitnr
< QCOW2_OL_MAX_BITNR
);
2465 qcow2_signal_corruption(bs
, true, offset
, size
, "Preventing invalid "
2466 "write on metadata (overlaps with %s)",
2467 metadata_ol_names
[metadata_ol_bitnr
]);
2474 /* A pointer to a function of this type is given to walk_over_reftable(). That
2475 * function will create refblocks and pass them to a RefblockFinishOp once they
2476 * are completed (@refblock). @refblock_empty is set if the refblock is
2479 * Along with the refblock, a corresponding reftable entry is passed, in the
2480 * reftable @reftable (which may be reallocated) at @reftable_index.
2482 * @allocated should be set to true if a new cluster has been allocated.
2484 typedef int (RefblockFinishOp
)(BlockDriverState
*bs
, uint64_t **reftable
,
2485 uint64_t reftable_index
, uint64_t *reftable_size
,
2486 void *refblock
, bool refblock_empty
,
2487 bool *allocated
, Error
**errp
);
2490 * This "operation" for walk_over_reftable() allocates the refblock on disk (if
2491 * it is not empty) and inserts its offset into the new reftable. The size of
2492 * this new reftable is increased as required.
2494 static int alloc_refblock(BlockDriverState
*bs
, uint64_t **reftable
,
2495 uint64_t reftable_index
, uint64_t *reftable_size
,
2496 void *refblock
, bool refblock_empty
, bool *allocated
,
2499 BDRVQcow2State
*s
= bs
->opaque
;
2502 if (!refblock_empty
&& reftable_index
>= *reftable_size
) {
2503 uint64_t *new_reftable
;
2504 uint64_t new_reftable_size
;
2506 new_reftable_size
= ROUND_UP(reftable_index
+ 1,
2507 s
->cluster_size
/ sizeof(uint64_t));
2508 if (new_reftable_size
> QCOW_MAX_REFTABLE_SIZE
/ sizeof(uint64_t)) {
2510 "This operation would make the refcount table grow "
2511 "beyond the maximum size supported by QEMU, aborting");
2515 new_reftable
= g_try_realloc(*reftable
, new_reftable_size
*
2517 if (!new_reftable
) {
2518 error_setg(errp
, "Failed to increase reftable buffer size");
2522 memset(new_reftable
+ *reftable_size
, 0,
2523 (new_reftable_size
- *reftable_size
) * sizeof(uint64_t));
2525 *reftable
= new_reftable
;
2526 *reftable_size
= new_reftable_size
;
2529 if (!refblock_empty
&& !(*reftable
)[reftable_index
]) {
2530 offset
= qcow2_alloc_clusters(bs
, s
->cluster_size
);
2532 error_setg_errno(errp
, -offset
, "Failed to allocate refblock");
2535 (*reftable
)[reftable_index
] = offset
;
2543 * This "operation" for walk_over_reftable() writes the refblock to disk at the
2544 * offset specified by the new reftable's entry. It does not modify the new
2545 * reftable or change any refcounts.
2547 static int flush_refblock(BlockDriverState
*bs
, uint64_t **reftable
,
2548 uint64_t reftable_index
, uint64_t *reftable_size
,
2549 void *refblock
, bool refblock_empty
, bool *allocated
,
2552 BDRVQcow2State
*s
= bs
->opaque
;
2556 if (reftable_index
< *reftable_size
&& (*reftable
)[reftable_index
]) {
2557 offset
= (*reftable
)[reftable_index
];
2559 ret
= qcow2_pre_write_overlap_check(bs
, 0, offset
, s
->cluster_size
);
2561 error_setg_errno(errp
, -ret
, "Overlap check failed");
2565 ret
= bdrv_pwrite(bs
->file
->bs
, offset
, refblock
, s
->cluster_size
);
2567 error_setg_errno(errp
, -ret
, "Failed to write refblock");
2571 assert(refblock_empty
);
2578 * This function walks over the existing reftable and every referenced refblock;
2579 * if @new_set_refcount is non-NULL, it is called for every refcount entry to
2580 * create an equal new entry in the passed @new_refblock. Once that
2581 * @new_refblock is completely filled, @operation will be called.
2583 * @status_cb and @cb_opaque are used for the amend operation's status callback.
2584 * @index is the index of the walk_over_reftable() calls and @total is the total
2585 * number of walk_over_reftable() calls per amend operation. Both are used for
2586 * calculating the parameters for the status callback.
2588 * @allocated is set to true if a new cluster has been allocated.
2590 static int walk_over_reftable(BlockDriverState
*bs
, uint64_t **new_reftable
,
2591 uint64_t *new_reftable_index
,
2592 uint64_t *new_reftable_size
,
2593 void *new_refblock
, int new_refblock_size
,
2594 int new_refcount_bits
,
2595 RefblockFinishOp
*operation
, bool *allocated
,
2596 Qcow2SetRefcountFunc
*new_set_refcount
,
2597 BlockDriverAmendStatusCB
*status_cb
,
2598 void *cb_opaque
, int index
, int total
,
2601 BDRVQcow2State
*s
= bs
->opaque
;
2602 uint64_t reftable_index
;
2603 bool new_refblock_empty
= true;
2605 int new_refblock_index
= 0;
2608 for (reftable_index
= 0; reftable_index
< s
->refcount_table_size
;
2611 uint64_t refblock_offset
= s
->refcount_table
[reftable_index
]
2614 status_cb(bs
, (uint64_t)index
* s
->refcount_table_size
+ reftable_index
,
2615 (uint64_t)total
* s
->refcount_table_size
, cb_opaque
);
2617 if (refblock_offset
) {
2620 if (offset_into_cluster(s
, refblock_offset
)) {
2621 qcow2_signal_corruption(bs
, true, -1, -1, "Refblock offset %#"
2622 PRIx64
" unaligned (reftable index: %#"
2623 PRIx64
")", refblock_offset
,
2626 "Image is corrupt (unaligned refblock offset)");
2630 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, refblock_offset
,
2633 error_setg_errno(errp
, -ret
, "Failed to retrieve refblock");
2637 for (refblock_index
= 0; refblock_index
< s
->refcount_block_size
;
2642 if (new_refblock_index
>= new_refblock_size
) {
2643 /* new_refblock is now complete */
2644 ret
= operation(bs
, new_reftable
, *new_reftable_index
,
2645 new_reftable_size
, new_refblock
,
2646 new_refblock_empty
, allocated
, errp
);
2648 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refblock
);
2652 (*new_reftable_index
)++;
2653 new_refblock_index
= 0;
2654 new_refblock_empty
= true;
2657 refcount
= s
->get_refcount(refblock
, refblock_index
);
2658 if (new_refcount_bits
< 64 && refcount
>> new_refcount_bits
) {
2661 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refblock
);
2663 offset
= ((reftable_index
<< s
->refcount_block_bits
)
2664 + refblock_index
) << s
->cluster_bits
;
2666 error_setg(errp
, "Cannot decrease refcount entry width to "
2667 "%i bits: Cluster at offset %#" PRIx64
" has a "
2668 "refcount of %" PRIu64
, new_refcount_bits
,
2673 if (new_set_refcount
) {
2674 new_set_refcount(new_refblock
, new_refblock_index
++,
2677 new_refblock_index
++;
2679 new_refblock_empty
= new_refblock_empty
&& refcount
== 0;
2682 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refblock
);
2684 /* No refblock means every refcount is 0 */
2685 for (refblock_index
= 0; refblock_index
< s
->refcount_block_size
;
2688 if (new_refblock_index
>= new_refblock_size
) {
2689 /* new_refblock is now complete */
2690 ret
= operation(bs
, new_reftable
, *new_reftable_index
,
2691 new_reftable_size
, new_refblock
,
2692 new_refblock_empty
, allocated
, errp
);
2697 (*new_reftable_index
)++;
2698 new_refblock_index
= 0;
2699 new_refblock_empty
= true;
2702 if (new_set_refcount
) {
2703 new_set_refcount(new_refblock
, new_refblock_index
++, 0);
2705 new_refblock_index
++;
2711 if (new_refblock_index
> 0) {
2712 /* Complete the potentially existing partially filled final refblock */
2713 if (new_set_refcount
) {
2714 for (; new_refblock_index
< new_refblock_size
;
2715 new_refblock_index
++)
2717 new_set_refcount(new_refblock
, new_refblock_index
, 0);
2721 ret
= operation(bs
, new_reftable
, *new_reftable_index
,
2722 new_reftable_size
, new_refblock
, new_refblock_empty
,
2728 (*new_reftable_index
)++;
2731 status_cb(bs
, (uint64_t)(index
+ 1) * s
->refcount_table_size
,
2732 (uint64_t)total
* s
->refcount_table_size
, cb_opaque
);
2737 int qcow2_change_refcount_order(BlockDriverState
*bs
, int refcount_order
,
2738 BlockDriverAmendStatusCB
*status_cb
,
2739 void *cb_opaque
, Error
**errp
)
2741 BDRVQcow2State
*s
= bs
->opaque
;
2742 Qcow2GetRefcountFunc
*new_get_refcount
;
2743 Qcow2SetRefcountFunc
*new_set_refcount
;
2744 void *new_refblock
= qemu_blockalign(bs
->file
->bs
, s
->cluster_size
);
2745 uint64_t *new_reftable
= NULL
, new_reftable_size
= 0;
2746 uint64_t *old_reftable
, old_reftable_size
, old_reftable_offset
;
2747 uint64_t new_reftable_index
= 0;
2749 int64_t new_reftable_offset
= 0, allocated_reftable_size
= 0;
2750 int new_refblock_size
, new_refcount_bits
= 1 << refcount_order
;
2751 int old_refcount_order
;
2754 bool new_allocation
;
2756 assert(s
->qcow_version
>= 3);
2757 assert(refcount_order
>= 0 && refcount_order
<= 6);
2759 /* see qcow2_open() */
2760 new_refblock_size
= 1 << (s
->cluster_bits
- (refcount_order
- 3));
2762 new_get_refcount
= get_refcount_funcs
[refcount_order
];
2763 new_set_refcount
= set_refcount_funcs
[refcount_order
];
2769 new_allocation
= false;
2771 /* At least we have to do this walk and the one which writes the
2772 * refblocks; also, at least we have to do this loop here at least
2773 * twice (normally), first to do the allocations, and second to
2774 * determine that everything is correctly allocated, this then makes
2775 * three walks in total */
2776 total_walks
= MAX(walk_index
+ 2, 3);
2778 /* First, allocate the structures so they are present in the refcount
2780 ret
= walk_over_reftable(bs
, &new_reftable
, &new_reftable_index
,
2781 &new_reftable_size
, NULL
, new_refblock_size
,
2782 new_refcount_bits
, &alloc_refblock
,
2783 &new_allocation
, NULL
, status_cb
, cb_opaque
,
2784 walk_index
++, total_walks
, errp
);
2789 new_reftable_index
= 0;
2791 if (new_allocation
) {
2792 if (new_reftable_offset
) {
2793 qcow2_free_clusters(bs
, new_reftable_offset
,
2794 allocated_reftable_size
* sizeof(uint64_t),
2795 QCOW2_DISCARD_NEVER
);
2798 new_reftable_offset
= qcow2_alloc_clusters(bs
, new_reftable_size
*
2800 if (new_reftable_offset
< 0) {
2801 error_setg_errno(errp
, -new_reftable_offset
,
2802 "Failed to allocate the new reftable");
2803 ret
= new_reftable_offset
;
2806 allocated_reftable_size
= new_reftable_size
;
2808 } while (new_allocation
);
2810 /* Second, write the new refblocks */
2811 ret
= walk_over_reftable(bs
, &new_reftable
, &new_reftable_index
,
2812 &new_reftable_size
, new_refblock
,
2813 new_refblock_size
, new_refcount_bits
,
2814 &flush_refblock
, &new_allocation
, new_set_refcount
,
2815 status_cb
, cb_opaque
, walk_index
, walk_index
+ 1,
2820 assert(!new_allocation
);
2823 /* Write the new reftable */
2824 ret
= qcow2_pre_write_overlap_check(bs
, 0, new_reftable_offset
,
2825 new_reftable_size
* sizeof(uint64_t));
2827 error_setg_errno(errp
, -ret
, "Overlap check failed");
2831 for (i
= 0; i
< new_reftable_size
; i
++) {
2832 cpu_to_be64s(&new_reftable
[i
]);
2835 ret
= bdrv_pwrite(bs
->file
->bs
, new_reftable_offset
, new_reftable
,
2836 new_reftable_size
* sizeof(uint64_t));
2838 for (i
= 0; i
< new_reftable_size
; i
++) {
2839 be64_to_cpus(&new_reftable
[i
]);
2843 error_setg_errno(errp
, -ret
, "Failed to write the new reftable");
2848 /* Empty the refcount cache */
2849 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2851 error_setg_errno(errp
, -ret
, "Failed to flush the refblock cache");
2855 /* Update the image header to point to the new reftable; this only updates
2856 * the fields which are relevant to qcow2_update_header(); other fields
2857 * such as s->refcount_table or s->refcount_bits stay stale for now
2858 * (because we have to restore everything if qcow2_update_header() fails) */
2859 old_refcount_order
= s
->refcount_order
;
2860 old_reftable_size
= s
->refcount_table_size
;
2861 old_reftable_offset
= s
->refcount_table_offset
;
2863 s
->refcount_order
= refcount_order
;
2864 s
->refcount_table_size
= new_reftable_size
;
2865 s
->refcount_table_offset
= new_reftable_offset
;
2867 ret
= qcow2_update_header(bs
);
2869 s
->refcount_order
= old_refcount_order
;
2870 s
->refcount_table_size
= old_reftable_size
;
2871 s
->refcount_table_offset
= old_reftable_offset
;
2872 error_setg_errno(errp
, -ret
, "Failed to update the qcow2 header");
2876 /* Now update the rest of the in-memory information */
2877 old_reftable
= s
->refcount_table
;
2878 s
->refcount_table
= new_reftable
;
2880 s
->refcount_bits
= 1 << refcount_order
;
2881 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
2882 s
->refcount_max
+= s
->refcount_max
- 1;
2884 s
->refcount_block_bits
= s
->cluster_bits
- (refcount_order
- 3);
2885 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
2887 s
->get_refcount
= new_get_refcount
;
2888 s
->set_refcount
= new_set_refcount
;
2890 /* For cleaning up all old refblocks and the old reftable below the "done"
2892 new_reftable
= old_reftable
;
2893 new_reftable_size
= old_reftable_size
;
2894 new_reftable_offset
= old_reftable_offset
;
2898 /* On success, new_reftable actually points to the old reftable (and
2899 * new_reftable_size is the old reftable's size); but that is just
2901 for (i
= 0; i
< new_reftable_size
; i
++) {
2902 uint64_t offset
= new_reftable
[i
] & REFT_OFFSET_MASK
;
2904 qcow2_free_clusters(bs
, offset
, s
->cluster_size
,
2905 QCOW2_DISCARD_OTHER
);
2908 g_free(new_reftable
);
2910 if (new_reftable_offset
> 0) {
2911 qcow2_free_clusters(bs
, new_reftable_offset
,
2912 new_reftable_size
* sizeof(uint64_t),
2913 QCOW2_DISCARD_OTHER
);
2917 qemu_vfree(new_refblock
);