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/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu-common.h"
28 #include "block/block_int.h"
29 #include "block/qcow2.h"
30 #include "qemu/range.h"
31 #include "qemu/bswap.h"
33 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, uint64_t size
);
34 static int QEMU_WARN_UNUSED_RESULT
update_refcount(BlockDriverState
*bs
,
35 int64_t offset
, int64_t length
, uint64_t addend
,
36 bool decrease
, enum qcow2_discard_type type
);
38 static uint64_t get_refcount_ro0(const void *refcount_array
, uint64_t index
);
39 static uint64_t get_refcount_ro1(const void *refcount_array
, uint64_t index
);
40 static uint64_t get_refcount_ro2(const void *refcount_array
, uint64_t index
);
41 static uint64_t get_refcount_ro3(const void *refcount_array
, uint64_t index
);
42 static uint64_t get_refcount_ro4(const void *refcount_array
, uint64_t index
);
43 static uint64_t get_refcount_ro5(const void *refcount_array
, uint64_t index
);
44 static uint64_t get_refcount_ro6(const void *refcount_array
, uint64_t index
);
46 static void set_refcount_ro0(void *refcount_array
, uint64_t index
,
48 static void set_refcount_ro1(void *refcount_array
, uint64_t index
,
50 static void set_refcount_ro2(void *refcount_array
, uint64_t index
,
52 static void set_refcount_ro3(void *refcount_array
, uint64_t index
,
54 static void set_refcount_ro4(void *refcount_array
, uint64_t index
,
56 static void set_refcount_ro5(void *refcount_array
, uint64_t index
,
58 static void set_refcount_ro6(void *refcount_array
, uint64_t index
,
62 static Qcow2GetRefcountFunc
*const get_refcount_funcs
[] = {
72 static Qcow2SetRefcountFunc
*const set_refcount_funcs
[] = {
83 /*********************************************************/
84 /* refcount handling */
86 int qcow2_refcount_init(BlockDriverState
*bs
)
88 BDRVQcow2State
*s
= bs
->opaque
;
89 unsigned int refcount_table_size2
, i
;
92 assert(s
->refcount_order
>= 0 && s
->refcount_order
<= 6);
94 s
->get_refcount
= get_refcount_funcs
[s
->refcount_order
];
95 s
->set_refcount
= set_refcount_funcs
[s
->refcount_order
];
97 assert(s
->refcount_table_size
<= INT_MAX
/ sizeof(uint64_t));
98 refcount_table_size2
= s
->refcount_table_size
* sizeof(uint64_t);
99 s
->refcount_table
= g_try_malloc(refcount_table_size2
);
101 if (s
->refcount_table_size
> 0) {
102 if (s
->refcount_table
== NULL
) {
106 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_LOAD
);
107 ret
= bdrv_pread(bs
->file
, s
->refcount_table_offset
,
108 s
->refcount_table
, refcount_table_size2
);
112 for(i
= 0; i
< s
->refcount_table_size
; i
++)
113 be64_to_cpus(&s
->refcount_table
[i
]);
120 void qcow2_refcount_close(BlockDriverState
*bs
)
122 BDRVQcow2State
*s
= bs
->opaque
;
123 g_free(s
->refcount_table
);
127 static uint64_t get_refcount_ro0(const void *refcount_array
, uint64_t index
)
129 return (((const uint8_t *)refcount_array
)[index
/ 8] >> (index
% 8)) & 0x1;
132 static void set_refcount_ro0(void *refcount_array
, uint64_t index
,
135 assert(!(value
>> 1));
136 ((uint8_t *)refcount_array
)[index
/ 8] &= ~(0x1 << (index
% 8));
137 ((uint8_t *)refcount_array
)[index
/ 8] |= value
<< (index
% 8);
140 static uint64_t get_refcount_ro1(const void *refcount_array
, uint64_t index
)
142 return (((const uint8_t *)refcount_array
)[index
/ 4] >> (2 * (index
% 4)))
146 static void set_refcount_ro1(void *refcount_array
, uint64_t index
,
149 assert(!(value
>> 2));
150 ((uint8_t *)refcount_array
)[index
/ 4] &= ~(0x3 << (2 * (index
% 4)));
151 ((uint8_t *)refcount_array
)[index
/ 4] |= value
<< (2 * (index
% 4));
154 static uint64_t get_refcount_ro2(const void *refcount_array
, uint64_t index
)
156 return (((const uint8_t *)refcount_array
)[index
/ 2] >> (4 * (index
% 2)))
160 static void set_refcount_ro2(void *refcount_array
, uint64_t index
,
163 assert(!(value
>> 4));
164 ((uint8_t *)refcount_array
)[index
/ 2] &= ~(0xf << (4 * (index
% 2)));
165 ((uint8_t *)refcount_array
)[index
/ 2] |= value
<< (4 * (index
% 2));
168 static uint64_t get_refcount_ro3(const void *refcount_array
, uint64_t index
)
170 return ((const uint8_t *)refcount_array
)[index
];
173 static void set_refcount_ro3(void *refcount_array
, uint64_t index
,
176 assert(!(value
>> 8));
177 ((uint8_t *)refcount_array
)[index
] = value
;
180 static uint64_t get_refcount_ro4(const void *refcount_array
, uint64_t index
)
182 return be16_to_cpu(((const uint16_t *)refcount_array
)[index
]);
185 static void set_refcount_ro4(void *refcount_array
, uint64_t index
,
188 assert(!(value
>> 16));
189 ((uint16_t *)refcount_array
)[index
] = cpu_to_be16(value
);
192 static uint64_t get_refcount_ro5(const void *refcount_array
, uint64_t index
)
194 return be32_to_cpu(((const uint32_t *)refcount_array
)[index
]);
197 static void set_refcount_ro5(void *refcount_array
, uint64_t index
,
200 assert(!(value
>> 32));
201 ((uint32_t *)refcount_array
)[index
] = cpu_to_be32(value
);
204 static uint64_t get_refcount_ro6(const void *refcount_array
, uint64_t index
)
206 return be64_to_cpu(((const uint64_t *)refcount_array
)[index
]);
209 static void set_refcount_ro6(void *refcount_array
, uint64_t index
,
212 ((uint64_t *)refcount_array
)[index
] = cpu_to_be64(value
);
216 static int load_refcount_block(BlockDriverState
*bs
,
217 int64_t refcount_block_offset
,
218 void **refcount_block
)
220 BDRVQcow2State
*s
= bs
->opaque
;
222 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_LOAD
);
223 return 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
,
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 DIV_ROUND_UP(table_clusters
, s
->refcount_block_size
);
491 uint64_t meta_clusters
= table_clusters
+ blocks_clusters
;
493 last_table_size
= table_size
;
494 table_size
= next_refcount_table_size(s
, blocks_used
+
495 DIV_ROUND_UP(meta_clusters
, s
->refcount_block_size
));
497 } while (last_table_size
!= table_size
);
500 fprintf(stderr
, "qcow2: Grow refcount table %" PRId32
" => %" PRId64
"\n",
501 s
->refcount_table_size
, table_size
);
504 /* Create the new refcount table and blocks */
505 uint64_t meta_offset
= (blocks_used
* s
->refcount_block_size
) *
507 uint64_t table_offset
= meta_offset
+ blocks_clusters
* s
->cluster_size
;
508 uint64_t *new_table
= g_try_new0(uint64_t, table_size
);
509 void *new_blocks
= g_try_malloc0(blocks_clusters
* s
->cluster_size
);
511 assert(table_size
> 0 && blocks_clusters
> 0);
512 if (new_table
== NULL
|| new_blocks
== NULL
) {
517 /* Fill the new refcount table */
518 memcpy(new_table
, s
->refcount_table
,
519 s
->refcount_table_size
* sizeof(uint64_t));
520 new_table
[refcount_table_index
] = new_block
;
523 for (i
= 0; i
< blocks_clusters
; i
++) {
524 new_table
[blocks_used
+ i
] = meta_offset
+ (i
* s
->cluster_size
);
527 /* Fill the refcount blocks */
528 uint64_t table_clusters
= size_to_clusters(s
, table_size
* sizeof(uint64_t));
530 for (i
= 0; i
< table_clusters
+ blocks_clusters
; i
++) {
531 s
->set_refcount(new_blocks
, block
++, 1);
534 /* Write refcount blocks to disk */
535 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS
);
536 ret
= bdrv_pwrite_sync(bs
->file
, meta_offset
, new_blocks
,
537 blocks_clusters
* s
->cluster_size
);
544 /* Write refcount table to disk */
545 for(i
= 0; i
< table_size
; i
++) {
546 cpu_to_be64s(&new_table
[i
]);
549 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE
);
550 ret
= bdrv_pwrite_sync(bs
->file
, table_offset
, new_table
,
551 table_size
* sizeof(uint64_t));
556 for(i
= 0; i
< table_size
; i
++) {
557 be64_to_cpus(&new_table
[i
]);
560 /* Hook up the new refcount table in the qcow2 header */
565 data
.d64
= cpu_to_be64(table_offset
);
566 data
.d32
= cpu_to_be32(table_clusters
);
567 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE
);
568 ret
= bdrv_pwrite_sync(bs
->file
,
569 offsetof(QCowHeader
, refcount_table_offset
),
570 &data
, sizeof(data
));
575 /* And switch it in memory */
576 uint64_t old_table_offset
= s
->refcount_table_offset
;
577 uint64_t old_table_size
= s
->refcount_table_size
;
579 g_free(s
->refcount_table
);
580 s
->refcount_table
= new_table
;
581 s
->refcount_table_size
= table_size
;
582 s
->refcount_table_offset
= table_offset
;
584 /* Free old table. */
585 qcow2_free_clusters(bs
, old_table_offset
, old_table_size
* sizeof(uint64_t),
586 QCOW2_DISCARD_OTHER
);
588 ret
= load_refcount_block(bs
, new_block
, refcount_block
);
593 /* If we were trying to do the initial refcount update for some cluster
594 * allocation, we might have used the same clusters to store newly
595 * allocated metadata. Make the caller search some new space. */
602 if (*refcount_block
!= NULL
) {
603 qcow2_cache_put(bs
, s
->refcount_block_cache
, refcount_block
);
608 void qcow2_process_discards(BlockDriverState
*bs
, int ret
)
610 BDRVQcow2State
*s
= bs
->opaque
;
611 Qcow2DiscardRegion
*d
, *next
;
613 QTAILQ_FOREACH_SAFE(d
, &s
->discards
, next
, next
) {
614 QTAILQ_REMOVE(&s
->discards
, d
, next
);
616 /* Discard is optional, ignore the return value */
618 bdrv_pdiscard(bs
->file
->bs
, d
->offset
, d
->bytes
);
625 static void update_refcount_discard(BlockDriverState
*bs
,
626 uint64_t offset
, uint64_t length
)
628 BDRVQcow2State
*s
= bs
->opaque
;
629 Qcow2DiscardRegion
*d
, *p
, *next
;
631 QTAILQ_FOREACH(d
, &s
->discards
, next
) {
632 uint64_t new_start
= MIN(offset
, d
->offset
);
633 uint64_t new_end
= MAX(offset
+ length
, d
->offset
+ d
->bytes
);
635 if (new_end
- new_start
<= length
+ d
->bytes
) {
636 /* There can't be any overlap, areas ending up here have no
637 * references any more and therefore shouldn't get freed another
639 assert(d
->bytes
+ length
== new_end
- new_start
);
640 d
->offset
= new_start
;
641 d
->bytes
= new_end
- new_start
;
646 d
= g_malloc(sizeof(*d
));
647 *d
= (Qcow2DiscardRegion
) {
652 QTAILQ_INSERT_TAIL(&s
->discards
, d
, next
);
655 /* Merge discard requests if they are adjacent now */
656 QTAILQ_FOREACH_SAFE(p
, &s
->discards
, next
, next
) {
658 || p
->offset
> d
->offset
+ d
->bytes
659 || d
->offset
> p
->offset
+ p
->bytes
)
664 /* Still no overlap possible */
665 assert(p
->offset
== d
->offset
+ d
->bytes
666 || d
->offset
== p
->offset
+ p
->bytes
);
668 QTAILQ_REMOVE(&s
->discards
, p
, next
);
669 d
->offset
= MIN(d
->offset
, p
->offset
);
670 d
->bytes
+= p
->bytes
;
675 /* XXX: cache several refcount block clusters ? */
676 /* @addend is the absolute value of the addend; if @decrease is set, @addend
677 * will be subtracted from the current refcount, otherwise it will be added */
678 static int QEMU_WARN_UNUSED_RESULT
update_refcount(BlockDriverState
*bs
,
683 enum qcow2_discard_type type
)
685 BDRVQcow2State
*s
= bs
->opaque
;
686 int64_t start
, last
, cluster_offset
;
687 void *refcount_block
= NULL
;
688 int64_t old_table_index
= -1;
692 fprintf(stderr
, "update_refcount: offset=%" PRId64
" size=%" PRId64
693 " addend=%s%" PRIu64
"\n", offset
, length
, decrease
? "-" : "",
698 } else if (length
== 0) {
703 qcow2_cache_set_dependency(bs
, s
->refcount_block_cache
,
707 start
= start_of_cluster(s
, offset
);
708 last
= start_of_cluster(s
, offset
+ length
- 1);
709 for(cluster_offset
= start
; cluster_offset
<= last
;
710 cluster_offset
+= s
->cluster_size
)
714 int64_t cluster_index
= cluster_offset
>> s
->cluster_bits
;
715 int64_t table_index
= cluster_index
>> s
->refcount_block_bits
;
717 /* Load the refcount block and allocate it if needed */
718 if (table_index
!= old_table_index
) {
719 if (refcount_block
) {
720 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refcount_block
);
722 ret
= alloc_refcount_block(bs
, cluster_index
, &refcount_block
);
727 old_table_index
= table_index
;
729 qcow2_cache_entry_mark_dirty(bs
, s
->refcount_block_cache
,
732 /* we can update the count and save it */
733 block_index
= cluster_index
& (s
->refcount_block_size
- 1);
735 refcount
= s
->get_refcount(refcount_block
, block_index
);
736 if (decrease
? (refcount
- addend
> refcount
)
737 : (refcount
+ addend
< refcount
||
738 refcount
+ addend
> s
->refcount_max
))
748 if (refcount
== 0 && cluster_index
< s
->free_cluster_index
) {
749 s
->free_cluster_index
= cluster_index
;
751 s
->set_refcount(refcount_block
, block_index
, refcount
);
753 if (refcount
== 0 && s
->discard_passthrough
[type
]) {
754 update_refcount_discard(bs
, cluster_offset
, s
->cluster_size
);
760 if (!s
->cache_discards
) {
761 qcow2_process_discards(bs
, ret
);
764 /* Write last changed block to disk */
765 if (refcount_block
) {
766 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refcount_block
);
770 * Try do undo any updates if an error is returned (This may succeed in
771 * some cases like ENOSPC for allocating a new refcount block)
775 dummy
= update_refcount(bs
, offset
, cluster_offset
- offset
, addend
,
776 !decrease
, QCOW2_DISCARD_NEVER
);
784 * Increases or decreases the refcount of a given cluster.
786 * @addend is the absolute value of the addend; if @decrease is set, @addend
787 * will be subtracted from the current refcount, otherwise it will be added.
789 * On success 0 is returned; on failure -errno is returned.
791 int qcow2_update_cluster_refcount(BlockDriverState
*bs
,
792 int64_t cluster_index
,
793 uint64_t addend
, bool decrease
,
794 enum qcow2_discard_type type
)
796 BDRVQcow2State
*s
= bs
->opaque
;
799 ret
= update_refcount(bs
, cluster_index
<< s
->cluster_bits
, 1, addend
,
810 /*********************************************************/
811 /* cluster allocation functions */
815 /* return < 0 if error */
816 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, uint64_t size
)
818 BDRVQcow2State
*s
= bs
->opaque
;
819 uint64_t i
, nb_clusters
, refcount
;
822 /* We can't allocate clusters if they may still be queued for discard. */
823 if (s
->cache_discards
) {
824 qcow2_process_discards(bs
, 0);
827 nb_clusters
= size_to_clusters(s
, size
);
829 for(i
= 0; i
< nb_clusters
; i
++) {
830 uint64_t next_cluster_index
= s
->free_cluster_index
++;
831 ret
= qcow2_get_refcount(bs
, next_cluster_index
, &refcount
);
835 } else if (refcount
!= 0) {
840 /* Make sure that all offsets in the "allocated" range are representable
842 if (s
->free_cluster_index
> 0 &&
843 s
->free_cluster_index
- 1 > (INT64_MAX
>> s
->cluster_bits
))
849 fprintf(stderr
, "alloc_clusters: size=%" PRId64
" -> %" PRId64
"\n",
851 (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
);
853 return (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
;
856 int64_t qcow2_alloc_clusters(BlockDriverState
*bs
, uint64_t size
)
861 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC
);
863 offset
= alloc_clusters_noref(bs
, size
);
868 ret
= update_refcount(bs
, offset
, size
, 1, false, QCOW2_DISCARD_NEVER
);
869 } while (ret
== -EAGAIN
);
878 int64_t qcow2_alloc_clusters_at(BlockDriverState
*bs
, uint64_t offset
,
881 BDRVQcow2State
*s
= bs
->opaque
;
882 uint64_t cluster_index
, refcount
;
886 assert(nb_clusters
>= 0);
887 if (nb_clusters
== 0) {
892 /* Check how many clusters there are free */
893 cluster_index
= offset
>> s
->cluster_bits
;
894 for(i
= 0; i
< nb_clusters
; i
++) {
895 ret
= qcow2_get_refcount(bs
, cluster_index
++, &refcount
);
898 } else if (refcount
!= 0) {
903 /* And then allocate them */
904 ret
= update_refcount(bs
, offset
, i
<< s
->cluster_bits
, 1, false,
905 QCOW2_DISCARD_NEVER
);
906 } while (ret
== -EAGAIN
);
915 /* only used to allocate compressed sectors. We try to allocate
916 contiguous sectors. size must be <= cluster_size */
917 int64_t qcow2_alloc_bytes(BlockDriverState
*bs
, int size
)
919 BDRVQcow2State
*s
= bs
->opaque
;
921 size_t free_in_cluster
;
924 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC_BYTES
);
925 assert(size
> 0 && size
<= s
->cluster_size
);
926 assert(!s
->free_byte_offset
|| offset_into_cluster(s
, s
->free_byte_offset
));
928 offset
= s
->free_byte_offset
;
932 ret
= qcow2_get_refcount(bs
, offset
>> s
->cluster_bits
, &refcount
);
937 if (refcount
== s
->refcount_max
) {
942 free_in_cluster
= s
->cluster_size
- offset_into_cluster(s
, offset
);
944 if (!offset
|| free_in_cluster
< size
) {
945 int64_t new_cluster
= alloc_clusters_noref(bs
, s
->cluster_size
);
946 if (new_cluster
< 0) {
950 if (!offset
|| ROUND_UP(offset
, s
->cluster_size
) != new_cluster
) {
951 offset
= new_cluster
;
952 free_in_cluster
= s
->cluster_size
;
954 free_in_cluster
+= s
->cluster_size
;
959 ret
= update_refcount(bs
, offset
, size
, 1, false, QCOW2_DISCARD_NEVER
);
963 } while (ret
== -EAGAIN
);
968 /* The cluster refcount was incremented; refcount blocks must be flushed
969 * before the caller's L2 table updates. */
970 qcow2_cache_set_dependency(bs
, s
->l2_table_cache
, s
->refcount_block_cache
);
972 s
->free_byte_offset
= offset
+ size
;
973 if (!offset_into_cluster(s
, s
->free_byte_offset
)) {
974 s
->free_byte_offset
= 0;
980 void qcow2_free_clusters(BlockDriverState
*bs
,
981 int64_t offset
, int64_t size
,
982 enum qcow2_discard_type type
)
986 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_FREE
);
987 ret
= update_refcount(bs
, offset
, size
, 1, true, type
);
989 fprintf(stderr
, "qcow2_free_clusters failed: %s\n", strerror(-ret
));
990 /* TODO Remember the clusters to free them later and avoid leaking */
995 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
996 * normal cluster, compressed cluster, etc.)
998 void qcow2_free_any_clusters(BlockDriverState
*bs
, uint64_t l2_entry
,
999 int nb_clusters
, enum qcow2_discard_type type
)
1001 BDRVQcow2State
*s
= bs
->opaque
;
1003 switch (qcow2_get_cluster_type(l2_entry
)) {
1004 case QCOW2_CLUSTER_COMPRESSED
:
1007 nb_csectors
= ((l2_entry
>> s
->csize_shift
) &
1009 qcow2_free_clusters(bs
,
1010 (l2_entry
& s
->cluster_offset_mask
) & ~511,
1011 nb_csectors
* 512, type
);
1014 case QCOW2_CLUSTER_NORMAL
:
1015 case QCOW2_CLUSTER_ZERO
:
1016 if (l2_entry
& L2E_OFFSET_MASK
) {
1017 if (offset_into_cluster(s
, l2_entry
& L2E_OFFSET_MASK
)) {
1018 qcow2_signal_corruption(bs
, false, -1, -1,
1019 "Cannot free unaligned cluster %#llx",
1020 l2_entry
& L2E_OFFSET_MASK
);
1022 qcow2_free_clusters(bs
, l2_entry
& L2E_OFFSET_MASK
,
1023 nb_clusters
<< s
->cluster_bits
, type
);
1027 case QCOW2_CLUSTER_UNALLOCATED
:
1036 /*********************************************************/
1037 /* snapshots and image creation */
1041 /* update the refcounts of snapshots and the copied flag */
1042 int qcow2_update_snapshot_refcount(BlockDriverState
*bs
,
1043 int64_t l1_table_offset
, int l1_size
, int addend
)
1045 BDRVQcow2State
*s
= bs
->opaque
;
1046 uint64_t *l1_table
, *l2_table
, l2_offset
, offset
, l1_size2
, refcount
;
1047 bool l1_allocated
= false;
1048 int64_t old_offset
, old_l2_offset
;
1049 int i
, j
, l1_modified
= 0, nb_csectors
;
1052 assert(addend
>= -1 && addend
<= 1);
1056 l1_size2
= l1_size
* sizeof(uint64_t);
1058 s
->cache_discards
= true;
1060 /* WARNING: qcow2_snapshot_goto relies on this function not using the
1061 * l1_table_offset when it is the current s->l1_table_offset! Be careful
1062 * when changing this! */
1063 if (l1_table_offset
!= s
->l1_table_offset
) {
1064 l1_table
= g_try_malloc0(align_offset(l1_size2
, 512));
1065 if (l1_size2
&& l1_table
== NULL
) {
1069 l1_allocated
= true;
1071 ret
= bdrv_pread(bs
->file
, l1_table_offset
, l1_table
, l1_size2
);
1076 for(i
= 0;i
< l1_size
; i
++)
1077 be64_to_cpus(&l1_table
[i
]);
1079 assert(l1_size
== s
->l1_size
);
1080 l1_table
= s
->l1_table
;
1081 l1_allocated
= false;
1084 for(i
= 0; i
< l1_size
; i
++) {
1085 l2_offset
= l1_table
[i
];
1087 old_l2_offset
= l2_offset
;
1088 l2_offset
&= L1E_OFFSET_MASK
;
1090 if (offset_into_cluster(s
, l2_offset
)) {
1091 qcow2_signal_corruption(bs
, true, -1, -1, "L2 table offset %#"
1092 PRIx64
" unaligned (L1 index: %#x)",
1098 ret
= qcow2_cache_get(bs
, s
->l2_table_cache
, l2_offset
,
1099 (void**) &l2_table
);
1104 for(j
= 0; j
< s
->l2_size
; j
++) {
1105 uint64_t cluster_index
;
1107 offset
= be64_to_cpu(l2_table
[j
]);
1108 old_offset
= offset
;
1109 offset
&= ~QCOW_OFLAG_COPIED
;
1111 switch (qcow2_get_cluster_type(offset
)) {
1112 case QCOW2_CLUSTER_COMPRESSED
:
1113 nb_csectors
= ((offset
>> s
->csize_shift
) &
1116 ret
= update_refcount(bs
,
1117 (offset
& s
->cluster_offset_mask
) & ~511,
1118 nb_csectors
* 512, abs(addend
), addend
< 0,
1119 QCOW2_DISCARD_SNAPSHOT
);
1124 /* compressed clusters are never modified */
1128 case QCOW2_CLUSTER_NORMAL
:
1129 case QCOW2_CLUSTER_ZERO
:
1130 if (offset_into_cluster(s
, offset
& L2E_OFFSET_MASK
)) {
1131 qcow2_signal_corruption(bs
, true, -1, -1, "Data "
1132 "cluster offset %#llx "
1133 "unaligned (L2 offset: %#"
1134 PRIx64
", L2 index: %#x)",
1135 offset
& L2E_OFFSET_MASK
,
1141 cluster_index
= (offset
& L2E_OFFSET_MASK
) >> s
->cluster_bits
;
1142 if (!cluster_index
) {
1148 ret
= qcow2_update_cluster_refcount(bs
,
1149 cluster_index
, abs(addend
), addend
< 0,
1150 QCOW2_DISCARD_SNAPSHOT
);
1156 ret
= qcow2_get_refcount(bs
, cluster_index
, &refcount
);
1162 case QCOW2_CLUSTER_UNALLOCATED
:
1170 if (refcount
== 1) {
1171 offset
|= QCOW_OFLAG_COPIED
;
1173 if (offset
!= old_offset
) {
1175 qcow2_cache_set_dependency(bs
, s
->l2_table_cache
,
1176 s
->refcount_block_cache
);
1178 l2_table
[j
] = cpu_to_be64(offset
);
1179 qcow2_cache_entry_mark_dirty(bs
, s
->l2_table_cache
,
1184 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &l2_table
);
1187 ret
= qcow2_update_cluster_refcount(bs
, l2_offset
>>
1189 abs(addend
), addend
< 0,
1190 QCOW2_DISCARD_SNAPSHOT
);
1195 ret
= qcow2_get_refcount(bs
, l2_offset
>> s
->cluster_bits
,
1199 } else if (refcount
== 1) {
1200 l2_offset
|= QCOW_OFLAG_COPIED
;
1202 if (l2_offset
!= old_l2_offset
) {
1203 l1_table
[i
] = l2_offset
;
1209 ret
= bdrv_flush(bs
);
1212 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
1215 s
->cache_discards
= false;
1216 qcow2_process_discards(bs
, ret
);
1218 /* Update L1 only if it isn't deleted anyway (addend = -1) */
1219 if (ret
== 0 && addend
>= 0 && l1_modified
) {
1220 for (i
= 0; i
< l1_size
; i
++) {
1221 cpu_to_be64s(&l1_table
[i
]);
1224 ret
= bdrv_pwrite_sync(bs
->file
, l1_table_offset
,
1225 l1_table
, l1_size2
);
1227 for (i
= 0; i
< l1_size
; i
++) {
1228 be64_to_cpus(&l1_table
[i
]);
1239 /*********************************************************/
1240 /* refcount checking functions */
1243 static uint64_t refcount_array_byte_size(BDRVQcow2State
*s
, uint64_t entries
)
1245 /* This assertion holds because there is no way we can address more than
1246 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because
1247 * offsets have to be representable in bytes); due to every cluster
1248 * corresponding to one refcount entry, we are well below that limit */
1249 assert(entries
< (UINT64_C(1) << (64 - 9)));
1251 /* Thanks to the assertion this will not overflow, because
1252 * s->refcount_order < 7.
1253 * (note: x << s->refcount_order == x * s->refcount_bits) */
1254 return DIV_ROUND_UP(entries
<< s
->refcount_order
, 8);
1258 * Reallocates *array so that it can hold new_size entries. *size must contain
1259 * the current number of entries in *array. If the reallocation fails, *array
1260 * and *size will not be modified and -errno will be returned. If the
1261 * reallocation is successful, *array will be set to the new buffer, *size
1262 * will be set to new_size and 0 will be returned. The size of the reallocated
1263 * refcount array buffer will be aligned to a cluster boundary, and the newly
1264 * allocated area will be zeroed.
1266 static int realloc_refcount_array(BDRVQcow2State
*s
, void **array
,
1267 int64_t *size
, int64_t new_size
)
1269 int64_t old_byte_size
, new_byte_size
;
1272 /* Round to clusters so the array can be directly written to disk */
1273 old_byte_size
= size_to_clusters(s
, refcount_array_byte_size(s
, *size
))
1275 new_byte_size
= size_to_clusters(s
, refcount_array_byte_size(s
, new_size
))
1278 if (new_byte_size
== old_byte_size
) {
1283 assert(new_byte_size
> 0);
1285 if (new_byte_size
> SIZE_MAX
) {
1289 new_ptr
= g_try_realloc(*array
, new_byte_size
);
1294 if (new_byte_size
> old_byte_size
) {
1295 memset((char *)new_ptr
+ old_byte_size
, 0,
1296 new_byte_size
- old_byte_size
);
1306 * Increases the refcount for a range of clusters in a given refcount table.
1307 * This is used to construct a temporary refcount table out of L1 and L2 tables
1308 * which can be compared to the refcount table saved in the image.
1310 * Modifies the number of errors in res.
1312 static int inc_refcounts(BlockDriverState
*bs
,
1313 BdrvCheckResult
*res
,
1314 void **refcount_table
,
1315 int64_t *refcount_table_size
,
1316 int64_t offset
, int64_t size
)
1318 BDRVQcow2State
*s
= bs
->opaque
;
1319 uint64_t start
, last
, cluster_offset
, k
, refcount
;
1326 start
= start_of_cluster(s
, offset
);
1327 last
= start_of_cluster(s
, offset
+ size
- 1);
1328 for(cluster_offset
= start
; cluster_offset
<= last
;
1329 cluster_offset
+= s
->cluster_size
) {
1330 k
= cluster_offset
>> s
->cluster_bits
;
1331 if (k
>= *refcount_table_size
) {
1332 ret
= realloc_refcount_array(s
, refcount_table
,
1333 refcount_table_size
, k
+ 1);
1335 res
->check_errors
++;
1340 refcount
= s
->get_refcount(*refcount_table
, k
);
1341 if (refcount
== s
->refcount_max
) {
1342 fprintf(stderr
, "ERROR: overflow cluster offset=0x%" PRIx64
1343 "\n", cluster_offset
);
1344 fprintf(stderr
, "Use qemu-img amend to increase the refcount entry "
1345 "width or qemu-img convert to create a clean copy if the "
1346 "image cannot be opened for writing\n");
1350 s
->set_refcount(*refcount_table
, k
, refcount
+ 1);
1356 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
1358 CHECK_FRAG_INFO
= 0x2, /* update BlockFragInfo counters */
1362 * Increases the refcount in the given refcount table for the all clusters
1363 * referenced in the L2 table. While doing so, performs some checks on L2
1366 * Returns the number of errors found by the checks or -errno if an internal
1369 static int check_refcounts_l2(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1370 void **refcount_table
,
1371 int64_t *refcount_table_size
, int64_t l2_offset
,
1374 BDRVQcow2State
*s
= bs
->opaque
;
1375 uint64_t *l2_table
, l2_entry
;
1376 uint64_t next_contiguous_offset
= 0;
1377 int i
, l2_size
, nb_csectors
, ret
;
1379 /* Read L2 table from disk */
1380 l2_size
= s
->l2_size
* sizeof(uint64_t);
1381 l2_table
= g_malloc(l2_size
);
1383 ret
= bdrv_pread(bs
->file
, l2_offset
, l2_table
, l2_size
);
1385 fprintf(stderr
, "ERROR: I/O error in check_refcounts_l2\n");
1386 res
->check_errors
++;
1390 /* Do the actual checks */
1391 for(i
= 0; i
< s
->l2_size
; i
++) {
1392 l2_entry
= be64_to_cpu(l2_table
[i
]);
1394 switch (qcow2_get_cluster_type(l2_entry
)) {
1395 case QCOW2_CLUSTER_COMPRESSED
:
1396 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1397 if (l2_entry
& QCOW_OFLAG_COPIED
) {
1398 fprintf(stderr
, "ERROR: cluster %" PRId64
": "
1399 "copied flag must never be set for compressed "
1400 "clusters\n", l2_entry
>> s
->cluster_bits
);
1401 l2_entry
&= ~QCOW_OFLAG_COPIED
;
1405 /* Mark cluster as used */
1406 nb_csectors
= ((l2_entry
>> s
->csize_shift
) &
1408 l2_entry
&= s
->cluster_offset_mask
;
1409 ret
= inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1410 l2_entry
& ~511, nb_csectors
* 512);
1415 if (flags
& CHECK_FRAG_INFO
) {
1416 res
->bfi
.allocated_clusters
++;
1417 res
->bfi
.compressed_clusters
++;
1419 /* Compressed clusters are fragmented by nature. Since they
1420 * take up sub-sector space but we only have sector granularity
1421 * I/O we need to re-read the same sectors even for adjacent
1422 * compressed clusters.
1424 res
->bfi
.fragmented_clusters
++;
1428 case QCOW2_CLUSTER_ZERO
:
1429 if ((l2_entry
& L2E_OFFSET_MASK
) == 0) {
1434 case QCOW2_CLUSTER_NORMAL
:
1436 uint64_t offset
= l2_entry
& L2E_OFFSET_MASK
;
1438 if (flags
& CHECK_FRAG_INFO
) {
1439 res
->bfi
.allocated_clusters
++;
1440 if (next_contiguous_offset
&&
1441 offset
!= next_contiguous_offset
) {
1442 res
->bfi
.fragmented_clusters
++;
1444 next_contiguous_offset
= offset
+ s
->cluster_size
;
1447 /* Mark cluster as used */
1448 ret
= inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1449 offset
, s
->cluster_size
);
1454 /* Correct offsets are cluster aligned */
1455 if (offset_into_cluster(s
, offset
)) {
1456 fprintf(stderr
, "ERROR offset=%" PRIx64
": Cluster is not "
1457 "properly aligned; L2 entry corrupted.\n", offset
);
1463 case QCOW2_CLUSTER_UNALLOCATED
:
1480 * Increases the refcount for the L1 table, its L2 tables and all referenced
1481 * clusters in the given refcount table. While doing so, performs some checks
1482 * on L1 and L2 entries.
1484 * Returns the number of errors found by the checks or -errno if an internal
1487 static int check_refcounts_l1(BlockDriverState
*bs
,
1488 BdrvCheckResult
*res
,
1489 void **refcount_table
,
1490 int64_t *refcount_table_size
,
1491 int64_t l1_table_offset
, int l1_size
,
1494 BDRVQcow2State
*s
= bs
->opaque
;
1495 uint64_t *l1_table
= NULL
, l2_offset
, l1_size2
;
1498 l1_size2
= l1_size
* sizeof(uint64_t);
1500 /* Mark L1 table as used */
1501 ret
= inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1502 l1_table_offset
, l1_size2
);
1507 /* Read L1 table entries from disk */
1509 l1_table
= g_try_malloc(l1_size2
);
1510 if (l1_table
== NULL
) {
1512 res
->check_errors
++;
1515 ret
= bdrv_pread(bs
->file
, l1_table_offset
, l1_table
, l1_size2
);
1517 fprintf(stderr
, "ERROR: I/O error in check_refcounts_l1\n");
1518 res
->check_errors
++;
1521 for(i
= 0;i
< l1_size
; i
++)
1522 be64_to_cpus(&l1_table
[i
]);
1525 /* Do the actual checks */
1526 for(i
= 0; i
< l1_size
; i
++) {
1527 l2_offset
= l1_table
[i
];
1529 /* Mark L2 table as used */
1530 l2_offset
&= L1E_OFFSET_MASK
;
1531 ret
= inc_refcounts(bs
, res
, refcount_table
, refcount_table_size
,
1532 l2_offset
, s
->cluster_size
);
1537 /* L2 tables are cluster aligned */
1538 if (offset_into_cluster(s
, l2_offset
)) {
1539 fprintf(stderr
, "ERROR l2_offset=%" PRIx64
": Table is not "
1540 "cluster aligned; L1 entry corrupted\n", l2_offset
);
1544 /* Process and check L2 entries */
1545 ret
= check_refcounts_l2(bs
, res
, refcount_table
,
1546 refcount_table_size
, l2_offset
, flags
);
1561 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1563 * This function does not print an error message nor does it increment
1564 * check_errors if qcow2_get_refcount fails (this is because such an error will
1565 * have been already detected and sufficiently signaled by the calling function
1566 * (qcow2_check_refcounts) by the time this function is called).
1568 static int check_oflag_copied(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1571 BDRVQcow2State
*s
= bs
->opaque
;
1572 uint64_t *l2_table
= qemu_blockalign(bs
, s
->cluster_size
);
1577 for (i
= 0; i
< s
->l1_size
; i
++) {
1578 uint64_t l1_entry
= s
->l1_table
[i
];
1579 uint64_t l2_offset
= l1_entry
& L1E_OFFSET_MASK
;
1580 bool l2_dirty
= false;
1586 ret
= qcow2_get_refcount(bs
, l2_offset
>> s
->cluster_bits
,
1589 /* don't print message nor increment check_errors */
1592 if ((refcount
== 1) != ((l1_entry
& QCOW_OFLAG_COPIED
) != 0)) {
1593 fprintf(stderr
, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1594 "l1_entry=%" PRIx64
" refcount=%" PRIu64
"\n",
1595 fix
& BDRV_FIX_ERRORS
? "Repairing" :
1597 i
, l1_entry
, refcount
);
1598 if (fix
& BDRV_FIX_ERRORS
) {
1599 s
->l1_table
[i
] = refcount
== 1
1600 ? l1_entry
| QCOW_OFLAG_COPIED
1601 : l1_entry
& ~QCOW_OFLAG_COPIED
;
1602 ret
= qcow2_write_l1_entry(bs
, i
);
1604 res
->check_errors
++;
1607 res
->corruptions_fixed
++;
1613 ret
= bdrv_pread(bs
->file
, l2_offset
, l2_table
,
1614 s
->l2_size
* sizeof(uint64_t));
1616 fprintf(stderr
, "ERROR: Could not read L2 table: %s\n",
1618 res
->check_errors
++;
1622 for (j
= 0; j
< s
->l2_size
; j
++) {
1623 uint64_t l2_entry
= be64_to_cpu(l2_table
[j
]);
1624 uint64_t data_offset
= l2_entry
& L2E_OFFSET_MASK
;
1625 int cluster_type
= qcow2_get_cluster_type(l2_entry
);
1627 if ((cluster_type
== QCOW2_CLUSTER_NORMAL
) ||
1628 ((cluster_type
== QCOW2_CLUSTER_ZERO
) && (data_offset
!= 0))) {
1629 ret
= qcow2_get_refcount(bs
,
1630 data_offset
>> s
->cluster_bits
,
1633 /* don't print message nor increment check_errors */
1636 if ((refcount
== 1) != ((l2_entry
& QCOW_OFLAG_COPIED
) != 0)) {
1637 fprintf(stderr
, "%s OFLAG_COPIED data cluster: "
1638 "l2_entry=%" PRIx64
" refcount=%" PRIu64
"\n",
1639 fix
& BDRV_FIX_ERRORS
? "Repairing" :
1641 l2_entry
, refcount
);
1642 if (fix
& BDRV_FIX_ERRORS
) {
1643 l2_table
[j
] = cpu_to_be64(refcount
== 1
1644 ? l2_entry
| QCOW_OFLAG_COPIED
1645 : l2_entry
& ~QCOW_OFLAG_COPIED
);
1647 res
->corruptions_fixed
++;
1656 ret
= qcow2_pre_write_overlap_check(bs
, QCOW2_OL_ACTIVE_L2
,
1657 l2_offset
, s
->cluster_size
);
1659 fprintf(stderr
, "ERROR: Could not write L2 table; metadata "
1660 "overlap check failed: %s\n", strerror(-ret
));
1661 res
->check_errors
++;
1665 ret
= bdrv_pwrite(bs
->file
, l2_offset
, l2_table
,
1668 fprintf(stderr
, "ERROR: Could not write L2 table: %s\n",
1670 res
->check_errors
++;
1679 qemu_vfree(l2_table
);
1684 * Checks consistency of refblocks and accounts for each refblock in
1687 static int check_refblocks(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1688 BdrvCheckMode fix
, bool *rebuild
,
1689 void **refcount_table
, int64_t *nb_clusters
)
1691 BDRVQcow2State
*s
= bs
->opaque
;
1695 for(i
= 0; i
< s
->refcount_table_size
; i
++) {
1696 uint64_t offset
, cluster
;
1697 offset
= s
->refcount_table
[i
];
1698 cluster
= offset
>> s
->cluster_bits
;
1700 /* Refcount blocks are cluster aligned */
1701 if (offset_into_cluster(s
, offset
)) {
1702 fprintf(stderr
, "ERROR refcount block %" PRId64
" is not "
1703 "cluster aligned; refcount table entry corrupted\n", i
);
1709 if (cluster
>= *nb_clusters
) {
1710 fprintf(stderr
, "%s refcount block %" PRId64
" is outside image\n",
1711 fix
& BDRV_FIX_ERRORS
? "Repairing" : "ERROR", i
);
1713 if (fix
& BDRV_FIX_ERRORS
) {
1714 int64_t new_nb_clusters
;
1716 if (offset
> INT64_MAX
- s
->cluster_size
) {
1721 ret
= bdrv_truncate(bs
->file
->bs
, offset
+ s
->cluster_size
);
1725 size
= bdrv_getlength(bs
->file
->bs
);
1731 new_nb_clusters
= size_to_clusters(s
, size
);
1732 assert(new_nb_clusters
>= *nb_clusters
);
1734 ret
= realloc_refcount_array(s
, refcount_table
,
1735 nb_clusters
, new_nb_clusters
);
1737 res
->check_errors
++;
1741 if (cluster
>= *nb_clusters
) {
1746 res
->corruptions_fixed
++;
1747 ret
= inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1748 offset
, s
->cluster_size
);
1752 /* No need to check whether the refcount is now greater than 1:
1753 * This area was just allocated and zeroed, so it can only be
1754 * exactly 1 after inc_refcounts() */
1760 fprintf(stderr
, "ERROR could not resize image: %s\n",
1769 ret
= inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1770 offset
, s
->cluster_size
);
1774 if (s
->get_refcount(*refcount_table
, cluster
) != 1) {
1775 fprintf(stderr
, "ERROR refcount block %" PRId64
1776 " refcount=%" PRIu64
"\n", i
,
1777 s
->get_refcount(*refcount_table
, cluster
));
1788 * Calculates an in-memory refcount table.
1790 static int calculate_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1791 BdrvCheckMode fix
, bool *rebuild
,
1792 void **refcount_table
, int64_t *nb_clusters
)
1794 BDRVQcow2State
*s
= bs
->opaque
;
1799 if (!*refcount_table
) {
1800 int64_t old_size
= 0;
1801 ret
= realloc_refcount_array(s
, refcount_table
,
1802 &old_size
, *nb_clusters
);
1804 res
->check_errors
++;
1810 ret
= inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1811 0, s
->cluster_size
);
1816 /* current L1 table */
1817 ret
= check_refcounts_l1(bs
, res
, refcount_table
, nb_clusters
,
1818 s
->l1_table_offset
, s
->l1_size
, CHECK_FRAG_INFO
);
1824 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
1825 sn
= s
->snapshots
+ i
;
1826 ret
= check_refcounts_l1(bs
, res
, refcount_table
, nb_clusters
,
1827 sn
->l1_table_offset
, sn
->l1_size
, 0);
1832 ret
= inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1833 s
->snapshots_offset
, s
->snapshots_size
);
1839 ret
= inc_refcounts(bs
, res
, refcount_table
, nb_clusters
,
1840 s
->refcount_table_offset
,
1841 s
->refcount_table_size
* sizeof(uint64_t));
1846 return check_refblocks(bs
, res
, fix
, rebuild
, refcount_table
, nb_clusters
);
1850 * Compares the actual reference count for each cluster in the image against the
1851 * refcount as reported by the refcount structures on-disk.
1853 static void compare_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1854 BdrvCheckMode fix
, bool *rebuild
,
1855 int64_t *highest_cluster
,
1856 void *refcount_table
, int64_t nb_clusters
)
1858 BDRVQcow2State
*s
= bs
->opaque
;
1860 uint64_t refcount1
, refcount2
;
1863 for (i
= 0, *highest_cluster
= 0; i
< nb_clusters
; i
++) {
1864 ret
= qcow2_get_refcount(bs
, i
, &refcount1
);
1866 fprintf(stderr
, "Can't get refcount for cluster %" PRId64
": %s\n",
1868 res
->check_errors
++;
1872 refcount2
= s
->get_refcount(refcount_table
, i
);
1874 if (refcount1
> 0 || refcount2
> 0) {
1875 *highest_cluster
= i
;
1878 if (refcount1
!= refcount2
) {
1879 /* Check if we're allowed to fix the mismatch */
1880 int *num_fixed
= NULL
;
1881 if (refcount1
== 0) {
1883 } else if (refcount1
> refcount2
&& (fix
& BDRV_FIX_LEAKS
)) {
1884 num_fixed
= &res
->leaks_fixed
;
1885 } else if (refcount1
< refcount2
&& (fix
& BDRV_FIX_ERRORS
)) {
1886 num_fixed
= &res
->corruptions_fixed
;
1889 fprintf(stderr
, "%s cluster %" PRId64
" refcount=%" PRIu64
1890 " reference=%" PRIu64
"\n",
1891 num_fixed
!= NULL
? "Repairing" :
1892 refcount1
< refcount2
? "ERROR" :
1894 i
, refcount1
, refcount2
);
1897 ret
= update_refcount(bs
, i
<< s
->cluster_bits
, 1,
1898 refcount_diff(refcount1
, refcount2
),
1899 refcount1
> refcount2
,
1900 QCOW2_DISCARD_ALWAYS
);
1907 /* And if we couldn't, print an error */
1908 if (refcount1
< refcount2
) {
1918 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to
1919 * the on-disk refcount structures.
1921 * On input, *first_free_cluster tells where to start looking, and need not
1922 * actually be a free cluster; the returned offset will not be before that
1923 * cluster. On output, *first_free_cluster points to the first gap found, even
1924 * if that gap was too small to be used as the returned offset.
1926 * Note that *first_free_cluster is a cluster index whereas the return value is
1929 static int64_t alloc_clusters_imrt(BlockDriverState
*bs
,
1931 void **refcount_table
,
1932 int64_t *imrt_nb_clusters
,
1933 int64_t *first_free_cluster
)
1935 BDRVQcow2State
*s
= bs
->opaque
;
1936 int64_t cluster
= *first_free_cluster
, i
;
1937 bool first_gap
= true;
1938 int contiguous_free_clusters
;
1941 /* Starting at *first_free_cluster, find a range of at least cluster_count
1942 * continuously free clusters */
1943 for (contiguous_free_clusters
= 0;
1944 cluster
< *imrt_nb_clusters
&&
1945 contiguous_free_clusters
< cluster_count
;
1948 if (!s
->get_refcount(*refcount_table
, cluster
)) {
1949 contiguous_free_clusters
++;
1951 /* If this is the first free cluster found, update
1952 * *first_free_cluster accordingly */
1953 *first_free_cluster
= cluster
;
1956 } else if (contiguous_free_clusters
) {
1957 contiguous_free_clusters
= 0;
1961 /* If contiguous_free_clusters is greater than zero, it contains the number
1962 * of continuously free clusters until the current cluster; the first free
1963 * cluster in the current "gap" is therefore
1964 * cluster - contiguous_free_clusters */
1966 /* If no such range could be found, grow the in-memory refcount table
1967 * accordingly to append free clusters at the end of the image */
1968 if (contiguous_free_clusters
< cluster_count
) {
1969 /* contiguous_free_clusters clusters are already empty at the image end;
1970 * we need cluster_count clusters; therefore, we have to allocate
1971 * cluster_count - contiguous_free_clusters new clusters at the end of
1972 * the image (which is the current value of cluster; note that cluster
1973 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond
1975 ret
= realloc_refcount_array(s
, refcount_table
, imrt_nb_clusters
,
1976 cluster
+ cluster_count
1977 - contiguous_free_clusters
);
1983 /* Go back to the first free cluster */
1984 cluster
-= contiguous_free_clusters
;
1985 for (i
= 0; i
< cluster_count
; i
++) {
1986 s
->set_refcount(*refcount_table
, cluster
+ i
, 1);
1989 return cluster
<< s
->cluster_bits
;
1993 * Creates a new refcount structure based solely on the in-memory information
1994 * given through *refcount_table. All necessary allocations will be reflected
1997 * On success, the old refcount structure is leaked (it will be covered by the
1998 * new refcount structure).
2000 static int rebuild_refcount_structure(BlockDriverState
*bs
,
2001 BdrvCheckResult
*res
,
2002 void **refcount_table
,
2003 int64_t *nb_clusters
)
2005 BDRVQcow2State
*s
= bs
->opaque
;
2006 int64_t first_free_cluster
= 0, reftable_offset
= -1, cluster
= 0;
2007 int64_t refblock_offset
, refblock_start
, refblock_index
;
2008 uint32_t reftable_size
= 0;
2009 uint64_t *on_disk_reftable
= NULL
;
2010 void *on_disk_refblock
;
2013 uint64_t reftable_offset
;
2014 uint32_t reftable_clusters
;
2015 } QEMU_PACKED reftable_offset_and_clusters
;
2017 qcow2_cache_empty(bs
, s
->refcount_block_cache
);
2020 for (; cluster
< *nb_clusters
; cluster
++) {
2021 if (!s
->get_refcount(*refcount_table
, cluster
)) {
2025 refblock_index
= cluster
>> s
->refcount_block_bits
;
2026 refblock_start
= refblock_index
<< s
->refcount_block_bits
;
2028 /* Don't allocate a cluster in a refblock already written to disk */
2029 if (first_free_cluster
< refblock_start
) {
2030 first_free_cluster
= refblock_start
;
2032 refblock_offset
= alloc_clusters_imrt(bs
, 1, refcount_table
,
2033 nb_clusters
, &first_free_cluster
);
2034 if (refblock_offset
< 0) {
2035 fprintf(stderr
, "ERROR allocating refblock: %s\n",
2036 strerror(-refblock_offset
));
2037 res
->check_errors
++;
2038 ret
= refblock_offset
;
2042 if (reftable_size
<= refblock_index
) {
2043 uint32_t old_reftable_size
= reftable_size
;
2044 uint64_t *new_on_disk_reftable
;
2046 reftable_size
= ROUND_UP((refblock_index
+ 1) * sizeof(uint64_t),
2047 s
->cluster_size
) / sizeof(uint64_t);
2048 new_on_disk_reftable
= g_try_realloc(on_disk_reftable
,
2051 if (!new_on_disk_reftable
) {
2052 res
->check_errors
++;
2056 on_disk_reftable
= new_on_disk_reftable
;
2058 memset(on_disk_reftable
+ old_reftable_size
, 0,
2059 (reftable_size
- old_reftable_size
) * sizeof(uint64_t));
2061 /* The offset we have for the reftable is now no longer valid;
2062 * this will leak that range, but we can easily fix that by running
2063 * a leak-fixing check after this rebuild operation */
2064 reftable_offset
= -1;
2066 on_disk_reftable
[refblock_index
] = refblock_offset
;
2068 /* If this is apparently the last refblock (for now), try to squeeze the
2070 if (refblock_index
== (*nb_clusters
- 1) >> s
->refcount_block_bits
&&
2071 reftable_offset
< 0)
2073 uint64_t reftable_clusters
= size_to_clusters(s
, reftable_size
*
2075 reftable_offset
= alloc_clusters_imrt(bs
, reftable_clusters
,
2076 refcount_table
, nb_clusters
,
2077 &first_free_cluster
);
2078 if (reftable_offset
< 0) {
2079 fprintf(stderr
, "ERROR allocating reftable: %s\n",
2080 strerror(-reftable_offset
));
2081 res
->check_errors
++;
2082 ret
= reftable_offset
;
2087 ret
= qcow2_pre_write_overlap_check(bs
, 0, refblock_offset
,
2090 fprintf(stderr
, "ERROR writing refblock: %s\n", strerror(-ret
));
2094 /* The size of *refcount_table is always cluster-aligned, therefore the
2095 * write operation will not overflow */
2096 on_disk_refblock
= (void *)((char *) *refcount_table
+
2097 refblock_index
* s
->cluster_size
);
2099 ret
= bdrv_write(bs
->file
, refblock_offset
/ BDRV_SECTOR_SIZE
,
2100 on_disk_refblock
, s
->cluster_sectors
);
2102 fprintf(stderr
, "ERROR writing refblock: %s\n", strerror(-ret
));
2106 /* Go to the end of this refblock */
2107 cluster
= refblock_start
+ s
->refcount_block_size
- 1;
2110 if (reftable_offset
< 0) {
2111 uint64_t post_refblock_start
, reftable_clusters
;
2113 post_refblock_start
= ROUND_UP(*nb_clusters
, s
->refcount_block_size
);
2114 reftable_clusters
= size_to_clusters(s
,
2115 reftable_size
* sizeof(uint64_t));
2116 /* Not pretty but simple */
2117 if (first_free_cluster
< post_refblock_start
) {
2118 first_free_cluster
= post_refblock_start
;
2120 reftable_offset
= alloc_clusters_imrt(bs
, reftable_clusters
,
2121 refcount_table
, nb_clusters
,
2122 &first_free_cluster
);
2123 if (reftable_offset
< 0) {
2124 fprintf(stderr
, "ERROR allocating reftable: %s\n",
2125 strerror(-reftable_offset
));
2126 res
->check_errors
++;
2127 ret
= reftable_offset
;
2131 goto write_refblocks
;
2134 assert(on_disk_reftable
);
2136 for (refblock_index
= 0; refblock_index
< reftable_size
; refblock_index
++) {
2137 cpu_to_be64s(&on_disk_reftable
[refblock_index
]);
2140 ret
= qcow2_pre_write_overlap_check(bs
, 0, reftable_offset
,
2141 reftable_size
* sizeof(uint64_t));
2143 fprintf(stderr
, "ERROR writing reftable: %s\n", strerror(-ret
));
2147 assert(reftable_size
< INT_MAX
/ sizeof(uint64_t));
2148 ret
= bdrv_pwrite(bs
->file
, reftable_offset
, on_disk_reftable
,
2149 reftable_size
* sizeof(uint64_t));
2151 fprintf(stderr
, "ERROR writing reftable: %s\n", strerror(-ret
));
2155 /* Enter new reftable into the image header */
2156 reftable_offset_and_clusters
.reftable_offset
= cpu_to_be64(reftable_offset
);
2157 reftable_offset_and_clusters
.reftable_clusters
=
2158 cpu_to_be32(size_to_clusters(s
, reftable_size
* sizeof(uint64_t)));
2159 ret
= bdrv_pwrite_sync(bs
->file
,
2160 offsetof(QCowHeader
, refcount_table_offset
),
2161 &reftable_offset_and_clusters
,
2162 sizeof(reftable_offset_and_clusters
));
2164 fprintf(stderr
, "ERROR setting reftable: %s\n", strerror(-ret
));
2168 for (refblock_index
= 0; refblock_index
< reftable_size
; refblock_index
++) {
2169 be64_to_cpus(&on_disk_reftable
[refblock_index
]);
2171 s
->refcount_table
= on_disk_reftable
;
2172 s
->refcount_table_offset
= reftable_offset
;
2173 s
->refcount_table_size
= reftable_size
;
2178 g_free(on_disk_reftable
);
2183 * Checks an image for refcount consistency.
2185 * Returns 0 if no errors are found, the number of errors in case the image is
2186 * detected as corrupted, and -errno when an internal error occurred.
2188 int qcow2_check_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
2191 BDRVQcow2State
*s
= bs
->opaque
;
2192 BdrvCheckResult pre_compare_res
;
2193 int64_t size
, highest_cluster
, nb_clusters
;
2194 void *refcount_table
= NULL
;
2195 bool rebuild
= false;
2198 size
= bdrv_getlength(bs
->file
->bs
);
2200 res
->check_errors
++;
2204 nb_clusters
= size_to_clusters(s
, size
);
2205 if (nb_clusters
> INT_MAX
) {
2206 res
->check_errors
++;
2210 res
->bfi
.total_clusters
=
2211 size_to_clusters(s
, bs
->total_sectors
* BDRV_SECTOR_SIZE
);
2213 ret
= calculate_refcounts(bs
, res
, fix
, &rebuild
, &refcount_table
,
2219 /* In case we don't need to rebuild the refcount structure (but want to fix
2220 * something), this function is immediately called again, in which case the
2221 * result should be ignored */
2222 pre_compare_res
= *res
;
2223 compare_refcounts(bs
, res
, 0, &rebuild
, &highest_cluster
, refcount_table
,
2226 if (rebuild
&& (fix
& BDRV_FIX_ERRORS
)) {
2227 BdrvCheckResult old_res
= *res
;
2228 int fresh_leaks
= 0;
2230 fprintf(stderr
, "Rebuilding refcount structure\n");
2231 ret
= rebuild_refcount_structure(bs
, res
, &refcount_table
,
2237 res
->corruptions
= 0;
2240 /* Because the old reftable has been exchanged for a new one the
2241 * references have to be recalculated */
2243 memset(refcount_table
, 0, refcount_array_byte_size(s
, nb_clusters
));
2244 ret
= calculate_refcounts(bs
, res
, 0, &rebuild
, &refcount_table
,
2250 if (fix
& BDRV_FIX_LEAKS
) {
2251 /* The old refcount structures are now leaked, fix it; the result
2252 * can be ignored, aside from leaks which were introduced by
2253 * rebuild_refcount_structure() that could not be fixed */
2254 BdrvCheckResult saved_res
= *res
;
2255 *res
= (BdrvCheckResult
){ 0 };
2257 compare_refcounts(bs
, res
, BDRV_FIX_LEAKS
, &rebuild
,
2258 &highest_cluster
, refcount_table
, nb_clusters
);
2260 fprintf(stderr
, "ERROR rebuilt refcount structure is still "
2264 /* Any leaks accounted for here were introduced by
2265 * rebuild_refcount_structure() because that function has created a
2266 * new refcount structure from scratch */
2267 fresh_leaks
= res
->leaks
;
2271 if (res
->corruptions
< old_res
.corruptions
) {
2272 res
->corruptions_fixed
+= old_res
.corruptions
- res
->corruptions
;
2274 if (res
->leaks
< old_res
.leaks
) {
2275 res
->leaks_fixed
+= old_res
.leaks
- res
->leaks
;
2277 res
->leaks
+= fresh_leaks
;
2280 fprintf(stderr
, "ERROR need to rebuild refcount structures\n");
2281 res
->check_errors
++;
2286 if (res
->leaks
|| res
->corruptions
) {
2287 *res
= pre_compare_res
;
2288 compare_refcounts(bs
, res
, fix
, &rebuild
, &highest_cluster
,
2289 refcount_table
, nb_clusters
);
2293 /* check OFLAG_COPIED */
2294 ret
= check_oflag_copied(bs
, res
, fix
);
2299 res
->image_end_offset
= (highest_cluster
+ 1) * s
->cluster_size
;
2303 g_free(refcount_table
);
2308 #define overlaps_with(ofs, sz) \
2309 ranges_overlap(offset, size, ofs, sz)
2312 * Checks if the given offset into the image file is actually free to use by
2313 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
2314 * i.e. a sanity check without relying on the refcount tables.
2316 * The ign parameter specifies what checks not to perform (being a bitmask of
2317 * QCow2MetadataOverlap values), i.e., what sections to ignore.
2320 * - 0 if writing to this offset will not affect the mentioned metadata
2321 * - a positive QCow2MetadataOverlap value indicating one overlapping section
2322 * - a negative value (-errno) indicating an error while performing a check,
2323 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
2325 int qcow2_check_metadata_overlap(BlockDriverState
*bs
, int ign
, int64_t offset
,
2328 BDRVQcow2State
*s
= bs
->opaque
;
2329 int chk
= s
->overlap_check
& ~ign
;
2336 if (chk
& QCOW2_OL_MAIN_HEADER
) {
2337 if (offset
< s
->cluster_size
) {
2338 return QCOW2_OL_MAIN_HEADER
;
2342 /* align range to test to cluster boundaries */
2343 size
= align_offset(offset_into_cluster(s
, offset
) + size
, s
->cluster_size
);
2344 offset
= start_of_cluster(s
, offset
);
2346 if ((chk
& QCOW2_OL_ACTIVE_L1
) && s
->l1_size
) {
2347 if (overlaps_with(s
->l1_table_offset
, s
->l1_size
* sizeof(uint64_t))) {
2348 return QCOW2_OL_ACTIVE_L1
;
2352 if ((chk
& QCOW2_OL_REFCOUNT_TABLE
) && s
->refcount_table_size
) {
2353 if (overlaps_with(s
->refcount_table_offset
,
2354 s
->refcount_table_size
* sizeof(uint64_t))) {
2355 return QCOW2_OL_REFCOUNT_TABLE
;
2359 if ((chk
& QCOW2_OL_SNAPSHOT_TABLE
) && s
->snapshots_size
) {
2360 if (overlaps_with(s
->snapshots_offset
, s
->snapshots_size
)) {
2361 return QCOW2_OL_SNAPSHOT_TABLE
;
2365 if ((chk
& QCOW2_OL_INACTIVE_L1
) && s
->snapshots
) {
2366 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
2367 if (s
->snapshots
[i
].l1_size
&&
2368 overlaps_with(s
->snapshots
[i
].l1_table_offset
,
2369 s
->snapshots
[i
].l1_size
* sizeof(uint64_t))) {
2370 return QCOW2_OL_INACTIVE_L1
;
2375 if ((chk
& QCOW2_OL_ACTIVE_L2
) && s
->l1_table
) {
2376 for (i
= 0; i
< s
->l1_size
; i
++) {
2377 if ((s
->l1_table
[i
] & L1E_OFFSET_MASK
) &&
2378 overlaps_with(s
->l1_table
[i
] & L1E_OFFSET_MASK
,
2380 return QCOW2_OL_ACTIVE_L2
;
2385 if ((chk
& QCOW2_OL_REFCOUNT_BLOCK
) && s
->refcount_table
) {
2386 for (i
= 0; i
< s
->refcount_table_size
; i
++) {
2387 if ((s
->refcount_table
[i
] & REFT_OFFSET_MASK
) &&
2388 overlaps_with(s
->refcount_table
[i
] & REFT_OFFSET_MASK
,
2390 return QCOW2_OL_REFCOUNT_BLOCK
;
2395 if ((chk
& QCOW2_OL_INACTIVE_L2
) && s
->snapshots
) {
2396 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
2397 uint64_t l1_ofs
= s
->snapshots
[i
].l1_table_offset
;
2398 uint32_t l1_sz
= s
->snapshots
[i
].l1_size
;
2399 uint64_t l1_sz2
= l1_sz
* sizeof(uint64_t);
2400 uint64_t *l1
= g_try_malloc(l1_sz2
);
2403 if (l1_sz2
&& l1
== NULL
) {
2407 ret
= bdrv_pread(bs
->file
, l1_ofs
, l1
, l1_sz2
);
2413 for (j
= 0; j
< l1_sz
; j
++) {
2414 uint64_t l2_ofs
= be64_to_cpu(l1
[j
]) & L1E_OFFSET_MASK
;
2415 if (l2_ofs
&& overlaps_with(l2_ofs
, s
->cluster_size
)) {
2417 return QCOW2_OL_INACTIVE_L2
;
2428 static const char *metadata_ol_names
[] = {
2429 [QCOW2_OL_MAIN_HEADER_BITNR
] = "qcow2_header",
2430 [QCOW2_OL_ACTIVE_L1_BITNR
] = "active L1 table",
2431 [QCOW2_OL_ACTIVE_L2_BITNR
] = "active L2 table",
2432 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = "refcount table",
2433 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = "refcount block",
2434 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = "snapshot table",
2435 [QCOW2_OL_INACTIVE_L1_BITNR
] = "inactive L1 table",
2436 [QCOW2_OL_INACTIVE_L2_BITNR
] = "inactive L2 table",
2440 * First performs a check for metadata overlaps (through
2441 * qcow2_check_metadata_overlap); if that fails with a negative value (error
2442 * while performing a check), that value is returned. If an impending overlap
2443 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
2444 * and -EIO returned.
2446 * Returns 0 if there were neither overlaps nor errors while checking for
2447 * overlaps; or a negative value (-errno) on error.
2449 int qcow2_pre_write_overlap_check(BlockDriverState
*bs
, int ign
, int64_t offset
,
2452 int ret
= qcow2_check_metadata_overlap(bs
, ign
, offset
, size
);
2456 } else if (ret
> 0) {
2457 int metadata_ol_bitnr
= ctz32(ret
);
2458 assert(metadata_ol_bitnr
< QCOW2_OL_MAX_BITNR
);
2460 qcow2_signal_corruption(bs
, true, offset
, size
, "Preventing invalid "
2461 "write on metadata (overlaps with %s)",
2462 metadata_ol_names
[metadata_ol_bitnr
]);
2469 /* A pointer to a function of this type is given to walk_over_reftable(). That
2470 * function will create refblocks and pass them to a RefblockFinishOp once they
2471 * are completed (@refblock). @refblock_empty is set if the refblock is
2474 * Along with the refblock, a corresponding reftable entry is passed, in the
2475 * reftable @reftable (which may be reallocated) at @reftable_index.
2477 * @allocated should be set to true if a new cluster has been allocated.
2479 typedef int (RefblockFinishOp
)(BlockDriverState
*bs
, uint64_t **reftable
,
2480 uint64_t reftable_index
, uint64_t *reftable_size
,
2481 void *refblock
, bool refblock_empty
,
2482 bool *allocated
, Error
**errp
);
2485 * This "operation" for walk_over_reftable() allocates the refblock on disk (if
2486 * it is not empty) and inserts its offset into the new reftable. The size of
2487 * this new reftable is increased as required.
2489 static int alloc_refblock(BlockDriverState
*bs
, uint64_t **reftable
,
2490 uint64_t reftable_index
, uint64_t *reftable_size
,
2491 void *refblock
, bool refblock_empty
, bool *allocated
,
2494 BDRVQcow2State
*s
= bs
->opaque
;
2497 if (!refblock_empty
&& reftable_index
>= *reftable_size
) {
2498 uint64_t *new_reftable
;
2499 uint64_t new_reftable_size
;
2501 new_reftable_size
= ROUND_UP(reftable_index
+ 1,
2502 s
->cluster_size
/ sizeof(uint64_t));
2503 if (new_reftable_size
> QCOW_MAX_REFTABLE_SIZE
/ sizeof(uint64_t)) {
2505 "This operation would make the refcount table grow "
2506 "beyond the maximum size supported by QEMU, aborting");
2510 new_reftable
= g_try_realloc(*reftable
, new_reftable_size
*
2512 if (!new_reftable
) {
2513 error_setg(errp
, "Failed to increase reftable buffer size");
2517 memset(new_reftable
+ *reftable_size
, 0,
2518 (new_reftable_size
- *reftable_size
) * sizeof(uint64_t));
2520 *reftable
= new_reftable
;
2521 *reftable_size
= new_reftable_size
;
2524 if (!refblock_empty
&& !(*reftable
)[reftable_index
]) {
2525 offset
= qcow2_alloc_clusters(bs
, s
->cluster_size
);
2527 error_setg_errno(errp
, -offset
, "Failed to allocate refblock");
2530 (*reftable
)[reftable_index
] = offset
;
2538 * This "operation" for walk_over_reftable() writes the refblock to disk at the
2539 * offset specified by the new reftable's entry. It does not modify the new
2540 * reftable or change any refcounts.
2542 static int flush_refblock(BlockDriverState
*bs
, uint64_t **reftable
,
2543 uint64_t reftable_index
, uint64_t *reftable_size
,
2544 void *refblock
, bool refblock_empty
, bool *allocated
,
2547 BDRVQcow2State
*s
= bs
->opaque
;
2551 if (reftable_index
< *reftable_size
&& (*reftable
)[reftable_index
]) {
2552 offset
= (*reftable
)[reftable_index
];
2554 ret
= qcow2_pre_write_overlap_check(bs
, 0, offset
, s
->cluster_size
);
2556 error_setg_errno(errp
, -ret
, "Overlap check failed");
2560 ret
= bdrv_pwrite(bs
->file
, offset
, refblock
, s
->cluster_size
);
2562 error_setg_errno(errp
, -ret
, "Failed to write refblock");
2566 assert(refblock_empty
);
2573 * This function walks over the existing reftable and every referenced refblock;
2574 * if @new_set_refcount is non-NULL, it is called for every refcount entry to
2575 * create an equal new entry in the passed @new_refblock. Once that
2576 * @new_refblock is completely filled, @operation will be called.
2578 * @status_cb and @cb_opaque are used for the amend operation's status callback.
2579 * @index is the index of the walk_over_reftable() calls and @total is the total
2580 * number of walk_over_reftable() calls per amend operation. Both are used for
2581 * calculating the parameters for the status callback.
2583 * @allocated is set to true if a new cluster has been allocated.
2585 static int walk_over_reftable(BlockDriverState
*bs
, uint64_t **new_reftable
,
2586 uint64_t *new_reftable_index
,
2587 uint64_t *new_reftable_size
,
2588 void *new_refblock
, int new_refblock_size
,
2589 int new_refcount_bits
,
2590 RefblockFinishOp
*operation
, bool *allocated
,
2591 Qcow2SetRefcountFunc
*new_set_refcount
,
2592 BlockDriverAmendStatusCB
*status_cb
,
2593 void *cb_opaque
, int index
, int total
,
2596 BDRVQcow2State
*s
= bs
->opaque
;
2597 uint64_t reftable_index
;
2598 bool new_refblock_empty
= true;
2600 int new_refblock_index
= 0;
2603 for (reftable_index
= 0; reftable_index
< s
->refcount_table_size
;
2606 uint64_t refblock_offset
= s
->refcount_table
[reftable_index
]
2609 status_cb(bs
, (uint64_t)index
* s
->refcount_table_size
+ reftable_index
,
2610 (uint64_t)total
* s
->refcount_table_size
, cb_opaque
);
2612 if (refblock_offset
) {
2615 if (offset_into_cluster(s
, refblock_offset
)) {
2616 qcow2_signal_corruption(bs
, true, -1, -1, "Refblock offset %#"
2617 PRIx64
" unaligned (reftable index: %#"
2618 PRIx64
")", refblock_offset
,
2621 "Image is corrupt (unaligned refblock offset)");
2625 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, refblock_offset
,
2628 error_setg_errno(errp
, -ret
, "Failed to retrieve refblock");
2632 for (refblock_index
= 0; refblock_index
< s
->refcount_block_size
;
2637 if (new_refblock_index
>= new_refblock_size
) {
2638 /* new_refblock is now complete */
2639 ret
= operation(bs
, new_reftable
, *new_reftable_index
,
2640 new_reftable_size
, new_refblock
,
2641 new_refblock_empty
, allocated
, errp
);
2643 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refblock
);
2647 (*new_reftable_index
)++;
2648 new_refblock_index
= 0;
2649 new_refblock_empty
= true;
2652 refcount
= s
->get_refcount(refblock
, refblock_index
);
2653 if (new_refcount_bits
< 64 && refcount
>> new_refcount_bits
) {
2656 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refblock
);
2658 offset
= ((reftable_index
<< s
->refcount_block_bits
)
2659 + refblock_index
) << s
->cluster_bits
;
2661 error_setg(errp
, "Cannot decrease refcount entry width to "
2662 "%i bits: Cluster at offset %#" PRIx64
" has a "
2663 "refcount of %" PRIu64
, new_refcount_bits
,
2668 if (new_set_refcount
) {
2669 new_set_refcount(new_refblock
, new_refblock_index
++,
2672 new_refblock_index
++;
2674 new_refblock_empty
= new_refblock_empty
&& refcount
== 0;
2677 qcow2_cache_put(bs
, s
->refcount_block_cache
, &refblock
);
2679 /* No refblock means every refcount is 0 */
2680 for (refblock_index
= 0; refblock_index
< s
->refcount_block_size
;
2683 if (new_refblock_index
>= new_refblock_size
) {
2684 /* new_refblock is now complete */
2685 ret
= operation(bs
, new_reftable
, *new_reftable_index
,
2686 new_reftable_size
, new_refblock
,
2687 new_refblock_empty
, allocated
, errp
);
2692 (*new_reftable_index
)++;
2693 new_refblock_index
= 0;
2694 new_refblock_empty
= true;
2697 if (new_set_refcount
) {
2698 new_set_refcount(new_refblock
, new_refblock_index
++, 0);
2700 new_refblock_index
++;
2706 if (new_refblock_index
> 0) {
2707 /* Complete the potentially existing partially filled final refblock */
2708 if (new_set_refcount
) {
2709 for (; new_refblock_index
< new_refblock_size
;
2710 new_refblock_index
++)
2712 new_set_refcount(new_refblock
, new_refblock_index
, 0);
2716 ret
= operation(bs
, new_reftable
, *new_reftable_index
,
2717 new_reftable_size
, new_refblock
, new_refblock_empty
,
2723 (*new_reftable_index
)++;
2726 status_cb(bs
, (uint64_t)(index
+ 1) * s
->refcount_table_size
,
2727 (uint64_t)total
* s
->refcount_table_size
, cb_opaque
);
2732 int qcow2_change_refcount_order(BlockDriverState
*bs
, int refcount_order
,
2733 BlockDriverAmendStatusCB
*status_cb
,
2734 void *cb_opaque
, Error
**errp
)
2736 BDRVQcow2State
*s
= bs
->opaque
;
2737 Qcow2GetRefcountFunc
*new_get_refcount
;
2738 Qcow2SetRefcountFunc
*new_set_refcount
;
2739 void *new_refblock
= qemu_blockalign(bs
->file
->bs
, s
->cluster_size
);
2740 uint64_t *new_reftable
= NULL
, new_reftable_size
= 0;
2741 uint64_t *old_reftable
, old_reftable_size
, old_reftable_offset
;
2742 uint64_t new_reftable_index
= 0;
2744 int64_t new_reftable_offset
= 0, allocated_reftable_size
= 0;
2745 int new_refblock_size
, new_refcount_bits
= 1 << refcount_order
;
2746 int old_refcount_order
;
2749 bool new_allocation
;
2751 assert(s
->qcow_version
>= 3);
2752 assert(refcount_order
>= 0 && refcount_order
<= 6);
2754 /* see qcow2_open() */
2755 new_refblock_size
= 1 << (s
->cluster_bits
- (refcount_order
- 3));
2757 new_get_refcount
= get_refcount_funcs
[refcount_order
];
2758 new_set_refcount
= set_refcount_funcs
[refcount_order
];
2764 new_allocation
= false;
2766 /* At least we have to do this walk and the one which writes the
2767 * refblocks; also, at least we have to do this loop here at least
2768 * twice (normally), first to do the allocations, and second to
2769 * determine that everything is correctly allocated, this then makes
2770 * three walks in total */
2771 total_walks
= MAX(walk_index
+ 2, 3);
2773 /* First, allocate the structures so they are present in the refcount
2775 ret
= walk_over_reftable(bs
, &new_reftable
, &new_reftable_index
,
2776 &new_reftable_size
, NULL
, new_refblock_size
,
2777 new_refcount_bits
, &alloc_refblock
,
2778 &new_allocation
, NULL
, status_cb
, cb_opaque
,
2779 walk_index
++, total_walks
, errp
);
2784 new_reftable_index
= 0;
2786 if (new_allocation
) {
2787 if (new_reftable_offset
) {
2788 qcow2_free_clusters(bs
, new_reftable_offset
,
2789 allocated_reftable_size
* sizeof(uint64_t),
2790 QCOW2_DISCARD_NEVER
);
2793 new_reftable_offset
= qcow2_alloc_clusters(bs
, new_reftable_size
*
2795 if (new_reftable_offset
< 0) {
2796 error_setg_errno(errp
, -new_reftable_offset
,
2797 "Failed to allocate the new reftable");
2798 ret
= new_reftable_offset
;
2801 allocated_reftable_size
= new_reftable_size
;
2803 } while (new_allocation
);
2805 /* Second, write the new refblocks */
2806 ret
= walk_over_reftable(bs
, &new_reftable
, &new_reftable_index
,
2807 &new_reftable_size
, new_refblock
,
2808 new_refblock_size
, new_refcount_bits
,
2809 &flush_refblock
, &new_allocation
, new_set_refcount
,
2810 status_cb
, cb_opaque
, walk_index
, walk_index
+ 1,
2815 assert(!new_allocation
);
2818 /* Write the new reftable */
2819 ret
= qcow2_pre_write_overlap_check(bs
, 0, new_reftable_offset
,
2820 new_reftable_size
* sizeof(uint64_t));
2822 error_setg_errno(errp
, -ret
, "Overlap check failed");
2826 for (i
= 0; i
< new_reftable_size
; i
++) {
2827 cpu_to_be64s(&new_reftable
[i
]);
2830 ret
= bdrv_pwrite(bs
->file
, new_reftable_offset
, new_reftable
,
2831 new_reftable_size
* sizeof(uint64_t));
2833 for (i
= 0; i
< new_reftable_size
; i
++) {
2834 be64_to_cpus(&new_reftable
[i
]);
2838 error_setg_errno(errp
, -ret
, "Failed to write the new reftable");
2843 /* Empty the refcount cache */
2844 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2846 error_setg_errno(errp
, -ret
, "Failed to flush the refblock cache");
2850 /* Update the image header to point to the new reftable; this only updates
2851 * the fields which are relevant to qcow2_update_header(); other fields
2852 * such as s->refcount_table or s->refcount_bits stay stale for now
2853 * (because we have to restore everything if qcow2_update_header() fails) */
2854 old_refcount_order
= s
->refcount_order
;
2855 old_reftable_size
= s
->refcount_table_size
;
2856 old_reftable_offset
= s
->refcount_table_offset
;
2858 s
->refcount_order
= refcount_order
;
2859 s
->refcount_table_size
= new_reftable_size
;
2860 s
->refcount_table_offset
= new_reftable_offset
;
2862 ret
= qcow2_update_header(bs
);
2864 s
->refcount_order
= old_refcount_order
;
2865 s
->refcount_table_size
= old_reftable_size
;
2866 s
->refcount_table_offset
= old_reftable_offset
;
2867 error_setg_errno(errp
, -ret
, "Failed to update the qcow2 header");
2871 /* Now update the rest of the in-memory information */
2872 old_reftable
= s
->refcount_table
;
2873 s
->refcount_table
= new_reftable
;
2875 s
->refcount_bits
= 1 << refcount_order
;
2876 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
2877 s
->refcount_max
+= s
->refcount_max
- 1;
2879 s
->refcount_block_bits
= s
->cluster_bits
- (refcount_order
- 3);
2880 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
2882 s
->get_refcount
= new_get_refcount
;
2883 s
->set_refcount
= new_set_refcount
;
2885 /* For cleaning up all old refblocks and the old reftable below the "done"
2887 new_reftable
= old_reftable
;
2888 new_reftable_size
= old_reftable_size
;
2889 new_reftable_offset
= old_reftable_offset
;
2893 /* On success, new_reftable actually points to the old reftable (and
2894 * new_reftable_size is the old reftable's size); but that is just
2896 for (i
= 0; i
< new_reftable_size
; i
++) {
2897 uint64_t offset
= new_reftable
[i
] & REFT_OFFSET_MASK
;
2899 qcow2_free_clusters(bs
, offset
, s
->cluster_size
,
2900 QCOW2_DISCARD_OTHER
);
2903 g_free(new_reftable
);
2905 if (new_reftable_offset
> 0) {
2906 qcow2_free_clusters(bs
, new_reftable_offset
,
2907 new_reftable_size
* sizeof(uint64_t),
2908 QCOW2_DISCARD_OTHER
);
2912 qemu_vfree(new_refblock
);