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"
28 #include "qemu/range.h"
29 #include "qemu/bswap.h"
30 #include "qemu/cutils.h"
33 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, uint64_t size
,
35 static int QEMU_WARN_UNUSED_RESULT
update_refcount(BlockDriverState
*bs
,
36 int64_t offset
, int64_t length
, uint64_t addend
,
37 bool decrease
, enum qcow2_discard_type type
);
39 static uint64_t get_refcount_ro0(const void *refcount_array
, uint64_t index
);
40 static uint64_t get_refcount_ro1(const void *refcount_array
, uint64_t index
);
41 static uint64_t get_refcount_ro2(const void *refcount_array
, uint64_t index
);
42 static uint64_t get_refcount_ro3(const void *refcount_array
, uint64_t index
);
43 static uint64_t get_refcount_ro4(const void *refcount_array
, uint64_t index
);
44 static uint64_t get_refcount_ro5(const void *refcount_array
, uint64_t index
);
45 static uint64_t get_refcount_ro6(const void *refcount_array
, uint64_t index
);
47 static void set_refcount_ro0(void *refcount_array
, uint64_t index
,
49 static void set_refcount_ro1(void *refcount_array
, uint64_t index
,
51 static void set_refcount_ro2(void *refcount_array
, uint64_t index
,
53 static void set_refcount_ro3(void *refcount_array
, uint64_t index
,
55 static void set_refcount_ro4(void *refcount_array
, uint64_t index
,
57 static void set_refcount_ro5(void *refcount_array
, uint64_t index
,
59 static void set_refcount_ro6(void *refcount_array
, uint64_t index
,
63 static Qcow2GetRefcountFunc
*const get_refcount_funcs
[] = {
73 static Qcow2SetRefcountFunc
*const set_refcount_funcs
[] = {
84 /*********************************************************/
85 /* refcount handling */
87 static void update_max_refcount_table_index(BDRVQcow2State
*s
)
89 unsigned i
= s
->refcount_table_size
- 1;
90 while (i
> 0 && (s
->refcount_table
[i
] & REFT_OFFSET_MASK
) == 0) {
93 /* Set s->max_refcount_table_index to the index of the last used entry */
94 s
->max_refcount_table_index
= i
;
97 int qcow2_refcount_init(BlockDriverState
*bs
)
99 BDRVQcow2State
*s
= bs
->opaque
;
100 unsigned int refcount_table_size2
, i
;
103 assert(s
->refcount_order
>= 0 && s
->refcount_order
<= 6);
105 s
->get_refcount
= get_refcount_funcs
[s
->refcount_order
];
106 s
->set_refcount
= set_refcount_funcs
[s
->refcount_order
];
108 assert(s
->refcount_table_size
<= INT_MAX
/ REFTABLE_ENTRY_SIZE
);
109 refcount_table_size2
= s
->refcount_table_size
* REFTABLE_ENTRY_SIZE
;
110 s
->refcount_table
= g_try_malloc(refcount_table_size2
);
112 if (s
->refcount_table_size
> 0) {
113 if (s
->refcount_table
== NULL
) {
117 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_LOAD
);
118 ret
= bdrv_pread(bs
->file
, s
->refcount_table_offset
,
119 s
->refcount_table
, refcount_table_size2
);
123 for(i
= 0; i
< s
->refcount_table_size
; i
++)
124 be64_to_cpus(&s
->refcount_table
[i
]);
125 update_max_refcount_table_index(s
);
132 void qcow2_refcount_close(BlockDriverState
*bs
)
134 BDRVQcow2State
*s
= bs
->opaque
;
135 g_free(s
->refcount_table
);
139 static uint64_t get_refcount_ro0(const void *refcount_array
, uint64_t index
)
141 return (((const uint8_t *)refcount_array
)[index
/ 8] >> (index
% 8)) & 0x1;
144 static void set_refcount_ro0(void *refcount_array
, uint64_t index
,
147 assert(!(value
>> 1));
148 ((uint8_t *)refcount_array
)[index
/ 8] &= ~(0x1 << (index
% 8));
149 ((uint8_t *)refcount_array
)[index
/ 8] |= value
<< (index
% 8);
152 static uint64_t get_refcount_ro1(const void *refcount_array
, uint64_t index
)
154 return (((const uint8_t *)refcount_array
)[index
/ 4] >> (2 * (index
% 4)))
158 static void set_refcount_ro1(void *refcount_array
, uint64_t index
,
161 assert(!(value
>> 2));
162 ((uint8_t *)refcount_array
)[index
/ 4] &= ~(0x3 << (2 * (index
% 4)));
163 ((uint8_t *)refcount_array
)[index
/ 4] |= value
<< (2 * (index
% 4));
166 static uint64_t get_refcount_ro2(const void *refcount_array
, uint64_t index
)
168 return (((const uint8_t *)refcount_array
)[index
/ 2] >> (4 * (index
% 2)))
172 static void set_refcount_ro2(void *refcount_array
, uint64_t index
,
175 assert(!(value
>> 4));
176 ((uint8_t *)refcount_array
)[index
/ 2] &= ~(0xf << (4 * (index
% 2)));
177 ((uint8_t *)refcount_array
)[index
/ 2] |= value
<< (4 * (index
% 2));
180 static uint64_t get_refcount_ro3(const void *refcount_array
, uint64_t index
)
182 return ((const uint8_t *)refcount_array
)[index
];
185 static void set_refcount_ro3(void *refcount_array
, uint64_t index
,
188 assert(!(value
>> 8));
189 ((uint8_t *)refcount_array
)[index
] = value
;
192 static uint64_t get_refcount_ro4(const void *refcount_array
, uint64_t index
)
194 return be16_to_cpu(((const uint16_t *)refcount_array
)[index
]);
197 static void set_refcount_ro4(void *refcount_array
, uint64_t index
,
200 assert(!(value
>> 16));
201 ((uint16_t *)refcount_array
)[index
] = cpu_to_be16(value
);
204 static uint64_t get_refcount_ro5(const void *refcount_array
, uint64_t index
)
206 return be32_to_cpu(((const uint32_t *)refcount_array
)[index
]);
209 static void set_refcount_ro5(void *refcount_array
, uint64_t index
,
212 assert(!(value
>> 32));
213 ((uint32_t *)refcount_array
)[index
] = cpu_to_be32(value
);
216 static uint64_t get_refcount_ro6(const void *refcount_array
, uint64_t index
)
218 return be64_to_cpu(((const uint64_t *)refcount_array
)[index
]);
221 static void set_refcount_ro6(void *refcount_array
, uint64_t index
,
224 ((uint64_t *)refcount_array
)[index
] = cpu_to_be64(value
);
228 static int load_refcount_block(BlockDriverState
*bs
,
229 int64_t refcount_block_offset
,
230 void **refcount_block
)
232 BDRVQcow2State
*s
= bs
->opaque
;
234 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_LOAD
);
235 return qcow2_cache_get(bs
, s
->refcount_block_cache
, refcount_block_offset
,
240 * Retrieves the refcount of the cluster given by its index and stores it in
241 * *refcount. Returns 0 on success and -errno on failure.
243 int qcow2_get_refcount(BlockDriverState
*bs
, int64_t cluster_index
,
246 BDRVQcow2State
*s
= bs
->opaque
;
247 uint64_t refcount_table_index
, block_index
;
248 int64_t refcount_block_offset
;
250 void *refcount_block
;
252 refcount_table_index
= cluster_index
>> s
->refcount_block_bits
;
253 if (refcount_table_index
>= s
->refcount_table_size
) {
257 refcount_block_offset
=
258 s
->refcount_table
[refcount_table_index
] & REFT_OFFSET_MASK
;
259 if (!refcount_block_offset
) {
264 if (offset_into_cluster(s
, refcount_block_offset
)) {
265 qcow2_signal_corruption(bs
, true, -1, -1, "Refblock offset %#" PRIx64
266 " unaligned (reftable index: %#" PRIx64
")",
267 refcount_block_offset
, refcount_table_index
);
271 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, refcount_block_offset
,
277 block_index
= cluster_index
& (s
->refcount_block_size
- 1);
278 *refcount
= s
->get_refcount(refcount_block
, block_index
);
280 qcow2_cache_put(s
->refcount_block_cache
, &refcount_block
);
285 /* Checks if two offsets are described by the same refcount block */
286 static int in_same_refcount_block(BDRVQcow2State
*s
, uint64_t offset_a
,
289 uint64_t block_a
= offset_a
>> (s
->cluster_bits
+ s
->refcount_block_bits
);
290 uint64_t block_b
= offset_b
>> (s
->cluster_bits
+ s
->refcount_block_bits
);
292 return (block_a
== block_b
);
296 * Loads a refcount block. If it doesn't exist yet, it is allocated first
297 * (including growing the refcount table if needed).
299 * Returns 0 on success or -errno in error case
301 static int alloc_refcount_block(BlockDriverState
*bs
,
302 int64_t cluster_index
, void **refcount_block
)
304 BDRVQcow2State
*s
= bs
->opaque
;
305 unsigned int refcount_table_index
;
308 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
310 /* Find the refcount block for the given cluster */
311 refcount_table_index
= cluster_index
>> s
->refcount_block_bits
;
313 if (refcount_table_index
< s
->refcount_table_size
) {
315 uint64_t refcount_block_offset
=
316 s
->refcount_table
[refcount_table_index
] & REFT_OFFSET_MASK
;
318 /* If it's already there, we're done */
319 if (refcount_block_offset
) {
320 if (offset_into_cluster(s
, refcount_block_offset
)) {
321 qcow2_signal_corruption(bs
, true, -1, -1, "Refblock offset %#"
322 PRIx64
" unaligned (reftable index: "
323 "%#x)", refcount_block_offset
,
324 refcount_table_index
);
328 return load_refcount_block(bs
, refcount_block_offset
,
334 * If we came here, we need to allocate something. Something is at least
335 * a cluster for the new refcount block. It may also include a new refcount
336 * table if the old refcount table is too small.
338 * Note that allocating clusters here needs some special care:
340 * - We can't use the normal qcow2_alloc_clusters(), it would try to
341 * increase the refcount and very likely we would end up with an endless
342 * recursion. Instead we must place the refcount blocks in a way that
343 * they can describe them themselves.
345 * - We need to consider that at this point we are inside update_refcounts
346 * and potentially doing an initial refcount increase. This means that
347 * some clusters have already been allocated by the caller, but their
348 * refcount isn't accurate yet. If we allocate clusters for metadata, we
349 * need to return -EAGAIN to signal the caller that it needs to restart
350 * the search for free clusters.
352 * - alloc_clusters_noref and qcow2_free_clusters may load a different
353 * refcount block into the cache
356 *refcount_block
= NULL
;
358 /* We write to the refcount table, so we might depend on L2 tables */
359 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
364 /* Allocate the refcount block itself and mark it as used */
365 int64_t new_block
= alloc_clusters_noref(bs
, s
->cluster_size
, INT64_MAX
);
370 /* The offset must fit in the offset field of the refcount table entry */
371 assert((new_block
& REFT_OFFSET_MASK
) == new_block
);
373 /* If we're allocating the block at offset 0 then something is wrong */
374 if (new_block
== 0) {
375 qcow2_signal_corruption(bs
, true, -1, -1, "Preventing invalid "
376 "allocation of refcount block at offset 0");
381 fprintf(stderr
, "qcow2: Allocate refcount block %d for %" PRIx64
383 refcount_table_index
, cluster_index
<< s
->cluster_bits
, new_block
);
386 if (in_same_refcount_block(s
, new_block
, cluster_index
<< s
->cluster_bits
)) {
387 /* Zero the new refcount block before updating it */
388 ret
= qcow2_cache_get_empty(bs
, s
->refcount_block_cache
, new_block
,
394 memset(*refcount_block
, 0, s
->cluster_size
);
396 /* The block describes itself, need to update the cache */
397 int block_index
= (new_block
>> s
->cluster_bits
) &
398 (s
->refcount_block_size
- 1);
399 s
->set_refcount(*refcount_block
, block_index
, 1);
401 /* Described somewhere else. This can recurse at most twice before we
402 * arrive at a block that describes itself. */
403 ret
= update_refcount(bs
, new_block
, s
->cluster_size
, 1, false,
404 QCOW2_DISCARD_NEVER
);
409 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
414 /* Initialize the new refcount block only after updating its refcount,
415 * update_refcount uses the refcount cache itself */
416 ret
= qcow2_cache_get_empty(bs
, s
->refcount_block_cache
, new_block
,
422 memset(*refcount_block
, 0, s
->cluster_size
);
425 /* Now the new refcount block needs to be written to disk */
426 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE
);
427 qcow2_cache_entry_mark_dirty(s
->refcount_block_cache
, *refcount_block
);
428 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
433 /* If the refcount table is big enough, just hook the block up there */
434 if (refcount_table_index
< s
->refcount_table_size
) {
435 uint64_t data64
= cpu_to_be64(new_block
);
436 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_HOOKUP
);
437 ret
= bdrv_pwrite_sync(bs
->file
, s
->refcount_table_offset
+
438 refcount_table_index
* REFTABLE_ENTRY_SIZE
,
439 &data64
, sizeof(data64
));
444 s
->refcount_table
[refcount_table_index
] = new_block
;
445 /* If there's a hole in s->refcount_table then it can happen
446 * that refcount_table_index < s->max_refcount_table_index */
447 s
->max_refcount_table_index
=
448 MAX(s
->max_refcount_table_index
, refcount_table_index
);
450 /* The new refcount block may be where the caller intended to put its
451 * data, so let it restart the search. */
455 qcow2_cache_put(s
->refcount_block_cache
, refcount_block
);
458 * If we come here, we need to grow the refcount table. Again, a new
459 * refcount table needs some space and we can't simply allocate to avoid
462 * Therefore let's grab new refcount blocks at the end of the image, which
463 * will describe themselves and the new refcount table. This way we can
464 * reference them only in the new table and do the switch to the new
465 * refcount table at once without producing an inconsistent state in
468 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_GROW
);
470 /* Calculate the number of refcount blocks needed so far; this will be the
471 * basis for calculating the index of the first cluster used for the
472 * self-describing refcount structures which we are about to create.
474 * Because we reached this point, there cannot be any refcount entries for
475 * cluster_index or higher indices yet. However, because new_block has been
476 * allocated to describe that cluster (and it will assume this role later
477 * on), we cannot use that index; also, new_block may actually have a higher
478 * cluster index than cluster_index, so it needs to be taken into account
479 * here (and 1 needs to be added to its value because that cluster is used).
481 uint64_t blocks_used
= DIV_ROUND_UP(MAX(cluster_index
+ 1,
482 (new_block
>> s
->cluster_bits
) + 1),
483 s
->refcount_block_size
);
485 /* Create the new refcount table and blocks */
486 uint64_t meta_offset
= (blocks_used
* s
->refcount_block_size
) *
489 ret
= qcow2_refcount_area(bs
, meta_offset
, 0, false,
490 refcount_table_index
, new_block
);
495 ret
= load_refcount_block(bs
, new_block
, refcount_block
);
500 /* If we were trying to do the initial refcount update for some cluster
501 * allocation, we might have used the same clusters to store newly
502 * allocated metadata. Make the caller search some new space. */
506 if (*refcount_block
!= NULL
) {
507 qcow2_cache_put(s
->refcount_block_cache
, refcount_block
);
513 * Starting at @start_offset, this function creates new self-covering refcount
514 * structures: A new refcount table and refcount blocks which cover all of
515 * themselves, and a number of @additional_clusters beyond their end.
516 * @start_offset must be at the end of the image file, that is, there must be
517 * only empty space beyond it.
518 * If @exact_size is false, the refcount table will have 50 % more entries than
519 * necessary so it will not need to grow again soon.
520 * If @new_refblock_offset is not zero, it contains the offset of a refcount
521 * block that should be entered into the new refcount table at index
522 * @new_refblock_index.
524 * Returns: The offset after the new refcount structures (i.e. where the
525 * @additional_clusters may be placed) on success, -errno on error.
527 int64_t qcow2_refcount_area(BlockDriverState
*bs
, uint64_t start_offset
,
528 uint64_t additional_clusters
, bool exact_size
,
529 int new_refblock_index
,
530 uint64_t new_refblock_offset
)
532 BDRVQcow2State
*s
= bs
->opaque
;
533 uint64_t total_refblock_count_u64
, additional_refblock_count
;
534 int total_refblock_count
, table_size
, area_reftable_index
, table_clusters
;
536 uint64_t table_offset
, block_offset
, end_offset
;
540 assert(!(start_offset
% s
->cluster_size
));
542 qcow2_refcount_metadata_size(start_offset
/ s
->cluster_size
+
544 s
->cluster_size
, s
->refcount_order
,
545 !exact_size
, &total_refblock_count_u64
);
546 if (total_refblock_count_u64
> QCOW_MAX_REFTABLE_SIZE
) {
549 total_refblock_count
= total_refblock_count_u64
;
551 /* Index in the refcount table of the first refcount block to cover the area
552 * of refcount structures we are about to create; we know that
553 * @total_refblock_count can cover @start_offset, so this will definitely
554 * fit into an int. */
555 area_reftable_index
= (start_offset
/ s
->cluster_size
) /
556 s
->refcount_block_size
;
559 table_size
= total_refblock_count
;
561 table_size
= total_refblock_count
+
562 DIV_ROUND_UP(total_refblock_count
, 2);
564 /* The qcow2 file can only store the reftable size in number of clusters */
565 table_size
= ROUND_UP(table_size
, s
->cluster_size
/ REFTABLE_ENTRY_SIZE
);
566 table_clusters
= (table_size
* REFTABLE_ENTRY_SIZE
) / s
->cluster_size
;
568 if (table_size
> QCOW_MAX_REFTABLE_SIZE
) {
572 new_table
= g_try_new0(uint64_t, table_size
);
574 assert(table_size
> 0);
575 if (new_table
== NULL
) {
580 /* Fill the new refcount table */
581 if (table_size
> s
->max_refcount_table_index
) {
582 /* We're actually growing the reftable */
583 memcpy(new_table
, s
->refcount_table
,
584 (s
->max_refcount_table_index
+ 1) * REFTABLE_ENTRY_SIZE
);
586 /* Improbable case: We're shrinking the reftable. However, the caller
587 * has assured us that there is only empty space beyond @start_offset,
588 * so we can simply drop all of the refblocks that won't fit into the
590 memcpy(new_table
, s
->refcount_table
, table_size
* REFTABLE_ENTRY_SIZE
);
593 if (new_refblock_offset
) {
594 assert(new_refblock_index
< total_refblock_count
);
595 new_table
[new_refblock_index
] = new_refblock_offset
;
598 /* Count how many new refblocks we have to create */
599 additional_refblock_count
= 0;
600 for (i
= area_reftable_index
; i
< total_refblock_count
; i
++) {
602 additional_refblock_count
++;
606 table_offset
= start_offset
+ additional_refblock_count
* s
->cluster_size
;
607 end_offset
= table_offset
+ table_clusters
* s
->cluster_size
;
609 /* Fill the refcount blocks, and create new ones, if necessary */
610 block_offset
= start_offset
;
611 for (i
= area_reftable_index
; i
< total_refblock_count
; i
++) {
613 uint64_t first_offset_covered
;
615 /* Reuse an existing refblock if possible, create a new one otherwise */
617 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, new_table
[i
],
623 ret
= qcow2_cache_get_empty(bs
, s
->refcount_block_cache
,
624 block_offset
, &refblock_data
);
628 memset(refblock_data
, 0, s
->cluster_size
);
629 qcow2_cache_entry_mark_dirty(s
->refcount_block_cache
,
632 new_table
[i
] = block_offset
;
633 block_offset
+= s
->cluster_size
;
636 /* First host offset covered by this refblock */
637 first_offset_covered
= (uint64_t)i
* s
->refcount_block_size
*
639 if (first_offset_covered
< end_offset
) {
642 /* Set the refcount of all of the new refcount structures to 1 */
644 if (first_offset_covered
< start_offset
) {
645 assert(i
== area_reftable_index
);
646 j
= (start_offset
- first_offset_covered
) / s
->cluster_size
;
647 assert(j
< s
->refcount_block_size
);
652 end_index
= MIN((end_offset
- first_offset_covered
) /
654 s
->refcount_block_size
);
656 for (; j
< end_index
; j
++) {
657 /* The caller guaranteed us this space would be empty */
658 assert(s
->get_refcount(refblock_data
, j
) == 0);
659 s
->set_refcount(refblock_data
, j
, 1);
662 qcow2_cache_entry_mark_dirty(s
->refcount_block_cache
,
666 qcow2_cache_put(s
->refcount_block_cache
, &refblock_data
);
669 assert(block_offset
== table_offset
);
671 /* Write refcount blocks to disk */
672 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS
);
673 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
678 /* Write refcount table to disk */
679 for (i
= 0; i
< total_refblock_count
; i
++) {
680 cpu_to_be64s(&new_table
[i
]);
683 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE
);
684 ret
= bdrv_pwrite_sync(bs
->file
, table_offset
, new_table
,
685 table_size
* REFTABLE_ENTRY_SIZE
);
690 for (i
= 0; i
< total_refblock_count
; i
++) {
691 be64_to_cpus(&new_table
[i
]);
694 /* Hook up the new refcount table in the qcow2 header */
699 data
.d64
= cpu_to_be64(table_offset
);
700 data
.d32
= cpu_to_be32(table_clusters
);
701 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE
);
702 ret
= bdrv_pwrite_sync(bs
->file
,
703 offsetof(QCowHeader
, refcount_table_offset
),
704 &data
, sizeof(data
));
709 /* And switch it in memory */
710 uint64_t old_table_offset
= s
->refcount_table_offset
;
711 uint64_t old_table_size
= s
->refcount_table_size
;
713 g_free(s
->refcount_table
);
714 s
->refcount_table
= new_table
;
715 s
->refcount_table_size
= table_size
;
716 s
->refcount_table_offset
= table_offset
;
717 update_max_refcount_table_index(s
);
719 /* Free old table. */
720 qcow2_free_clusters(bs
, old_table_offset
,
721 old_table_size
* REFTABLE_ENTRY_SIZE
,
722 QCOW2_DISCARD_OTHER
);
731 void qcow2_process_discards(BlockDriverState
*bs
, int ret
)
733 BDRVQcow2State
*s
= bs
->opaque
;
734 Qcow2DiscardRegion
*d
, *next
;
736 QTAILQ_FOREACH_SAFE(d
, &s
->discards
, next
, next
) {
737 QTAILQ_REMOVE(&s
->discards
, d
, next
);
739 /* Discard is optional, ignore the return value */
741 int r2
= bdrv_pdiscard(bs
->file
, d
->offset
, d
->bytes
);
743 trace_qcow2_process_discards_failed_region(d
->offset
, d
->bytes
,
752 static void update_refcount_discard(BlockDriverState
*bs
,
753 uint64_t offset
, uint64_t length
)
755 BDRVQcow2State
*s
= bs
->opaque
;
756 Qcow2DiscardRegion
*d
, *p
, *next
;
758 QTAILQ_FOREACH(d
, &s
->discards
, next
) {
759 uint64_t new_start
= MIN(offset
, d
->offset
);
760 uint64_t new_end
= MAX(offset
+ length
, d
->offset
+ d
->bytes
);
762 if (new_end
- new_start
<= length
+ d
->bytes
) {
763 /* There can't be any overlap, areas ending up here have no
764 * references any more and therefore shouldn't get freed another
766 assert(d
->bytes
+ length
== new_end
- new_start
);
767 d
->offset
= new_start
;
768 d
->bytes
= new_end
- new_start
;
773 d
= g_malloc(sizeof(*d
));
774 *d
= (Qcow2DiscardRegion
) {
779 QTAILQ_INSERT_TAIL(&s
->discards
, d
, next
);
782 /* Merge discard requests if they are adjacent now */
783 QTAILQ_FOREACH_SAFE(p
, &s
->discards
, next
, next
) {
785 || p
->offset
> d
->offset
+ d
->bytes
786 || d
->offset
> p
->offset
+ p
->bytes
)
791 /* Still no overlap possible */
792 assert(p
->offset
== d
->offset
+ d
->bytes
793 || d
->offset
== p
->offset
+ p
->bytes
);
795 QTAILQ_REMOVE(&s
->discards
, p
, next
);
796 d
->offset
= MIN(d
->offset
, p
->offset
);
797 d
->bytes
+= p
->bytes
;
802 /* XXX: cache several refcount block clusters ? */
803 /* @addend is the absolute value of the addend; if @decrease is set, @addend
804 * will be subtracted from the current refcount, otherwise it will be added */
805 static int QEMU_WARN_UNUSED_RESULT
update_refcount(BlockDriverState
*bs
,
810 enum qcow2_discard_type type
)
812 BDRVQcow2State
*s
= bs
->opaque
;
813 int64_t start
, last
, cluster_offset
;
814 void *refcount_block
= NULL
;
815 int64_t old_table_index
= -1;
819 fprintf(stderr
, "update_refcount: offset=%" PRId64
" size=%" PRId64
820 " addend=%s%" PRIu64
"\n", offset
, length
, decrease
? "-" : "",
825 } else if (length
== 0) {
830 qcow2_cache_set_dependency(bs
, s
->refcount_block_cache
,
834 start
= start_of_cluster(s
, offset
);
835 last
= start_of_cluster(s
, offset
+ length
- 1);
836 for(cluster_offset
= start
; cluster_offset
<= last
;
837 cluster_offset
+= s
->cluster_size
)
841 int64_t cluster_index
= cluster_offset
>> s
->cluster_bits
;
842 int64_t table_index
= cluster_index
>> s
->refcount_block_bits
;
844 /* Load the refcount block and allocate it if needed */
845 if (table_index
!= old_table_index
) {
846 if (refcount_block
) {
847 qcow2_cache_put(s
->refcount_block_cache
, &refcount_block
);
849 ret
= alloc_refcount_block(bs
, cluster_index
, &refcount_block
);
850 /* If the caller needs to restart the search for free clusters,
851 * try the same ones first to see if they're still free. */
852 if (ret
== -EAGAIN
) {
853 if (s
->free_cluster_index
> (start
>> s
->cluster_bits
)) {
854 s
->free_cluster_index
= (start
>> s
->cluster_bits
);
861 old_table_index
= table_index
;
863 qcow2_cache_entry_mark_dirty(s
->refcount_block_cache
, refcount_block
);
865 /* we can update the count and save it */
866 block_index
= cluster_index
& (s
->refcount_block_size
- 1);
868 refcount
= s
->get_refcount(refcount_block
, block_index
);
869 if (decrease
? (refcount
- addend
> refcount
)
870 : (refcount
+ addend
< refcount
||
871 refcount
+ addend
> s
->refcount_max
))
881 if (refcount
== 0 && cluster_index
< s
->free_cluster_index
) {
882 s
->free_cluster_index
= cluster_index
;
884 s
->set_refcount(refcount_block
, block_index
, refcount
);
889 table
= qcow2_cache_is_table_offset(s
->refcount_block_cache
,
892 qcow2_cache_put(s
->refcount_block_cache
, &refcount_block
);
893 old_table_index
= -1;
894 qcow2_cache_discard(s
->refcount_block_cache
, table
);
897 table
= qcow2_cache_is_table_offset(s
->l2_table_cache
, offset
);
899 qcow2_cache_discard(s
->l2_table_cache
, table
);
902 if (s
->discard_passthrough
[type
]) {
903 update_refcount_discard(bs
, cluster_offset
, s
->cluster_size
);
910 if (!s
->cache_discards
) {
911 qcow2_process_discards(bs
, ret
);
914 /* Write last changed block to disk */
915 if (refcount_block
) {
916 qcow2_cache_put(s
->refcount_block_cache
, &refcount_block
);
920 * Try do undo any updates if an error is returned (This may succeed in
921 * some cases like ENOSPC for allocating a new refcount block)
925 dummy
= update_refcount(bs
, offset
, cluster_offset
- offset
, addend
,
926 !decrease
, QCOW2_DISCARD_NEVER
);
934 * Increases or decreases the refcount of a given cluster.
936 * @addend is the absolute value of the addend; if @decrease is set, @addend
937 * will be subtracted from the current refcount, otherwise it will be added.
939 * On success 0 is returned; on failure -errno is returned.
941 int qcow2_update_cluster_refcount(BlockDriverState
*bs
,
942 int64_t cluster_index
,
943 uint64_t addend
, bool decrease
,
944 enum qcow2_discard_type type
)
946 BDRVQcow2State
*s
= bs
->opaque
;
949 ret
= update_refcount(bs
, cluster_index
<< s
->cluster_bits
, 1, addend
,
960 /*********************************************************/
961 /* cluster allocation functions */
965 /* return < 0 if error */
966 static int64_t alloc_clusters_noref(BlockDriverState
*bs
, uint64_t size
,
969 BDRVQcow2State
*s
= bs
->opaque
;
970 uint64_t i
, nb_clusters
, refcount
;
973 /* We can't allocate clusters if they may still be queued for discard. */
974 if (s
->cache_discards
) {
975 qcow2_process_discards(bs
, 0);
978 nb_clusters
= size_to_clusters(s
, size
);
980 for(i
= 0; i
< nb_clusters
; i
++) {
981 uint64_t next_cluster_index
= s
->free_cluster_index
++;
982 ret
= qcow2_get_refcount(bs
, next_cluster_index
, &refcount
);
986 } else if (refcount
!= 0) {
991 /* Make sure that all offsets in the "allocated" range are representable
992 * in the requested max */
993 if (s
->free_cluster_index
> 0 &&
994 s
->free_cluster_index
- 1 > (max
>> s
->cluster_bits
))
1000 fprintf(stderr
, "alloc_clusters: size=%" PRId64
" -> %" PRId64
"\n",
1002 (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
);
1004 return (s
->free_cluster_index
- nb_clusters
) << s
->cluster_bits
;
1007 int64_t qcow2_alloc_clusters(BlockDriverState
*bs
, uint64_t size
)
1012 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC
);
1014 offset
= alloc_clusters_noref(bs
, size
, QCOW_MAX_CLUSTER_OFFSET
);
1019 ret
= update_refcount(bs
, offset
, size
, 1, false, QCOW2_DISCARD_NEVER
);
1020 } while (ret
== -EAGAIN
);
1029 int64_t qcow2_alloc_clusters_at(BlockDriverState
*bs
, uint64_t offset
,
1030 int64_t nb_clusters
)
1032 BDRVQcow2State
*s
= bs
->opaque
;
1033 uint64_t cluster_index
, refcount
;
1037 assert(nb_clusters
>= 0);
1038 if (nb_clusters
== 0) {
1043 /* Check how many clusters there are free */
1044 cluster_index
= offset
>> s
->cluster_bits
;
1045 for(i
= 0; i
< nb_clusters
; i
++) {
1046 ret
= qcow2_get_refcount(bs
, cluster_index
++, &refcount
);
1049 } else if (refcount
!= 0) {
1054 /* And then allocate them */
1055 ret
= update_refcount(bs
, offset
, i
<< s
->cluster_bits
, 1, false,
1056 QCOW2_DISCARD_NEVER
);
1057 } while (ret
== -EAGAIN
);
1066 /* only used to allocate compressed sectors. We try to allocate
1067 contiguous sectors. size must be <= cluster_size */
1068 int64_t qcow2_alloc_bytes(BlockDriverState
*bs
, int size
)
1070 BDRVQcow2State
*s
= bs
->opaque
;
1072 size_t free_in_cluster
;
1075 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_ALLOC_BYTES
);
1076 assert(size
> 0 && size
<= s
->cluster_size
);
1077 assert(!s
->free_byte_offset
|| offset_into_cluster(s
, s
->free_byte_offset
));
1079 offset
= s
->free_byte_offset
;
1083 ret
= qcow2_get_refcount(bs
, offset
>> s
->cluster_bits
, &refcount
);
1088 if (refcount
== s
->refcount_max
) {
1093 free_in_cluster
= s
->cluster_size
- offset_into_cluster(s
, offset
);
1095 if (!offset
|| free_in_cluster
< size
) {
1096 int64_t new_cluster
;
1098 new_cluster
= alloc_clusters_noref(bs
, s
->cluster_size
,
1099 MIN(s
->cluster_offset_mask
,
1100 QCOW_MAX_CLUSTER_OFFSET
));
1101 if (new_cluster
< 0) {
1105 if (new_cluster
== 0) {
1106 qcow2_signal_corruption(bs
, true, -1, -1, "Preventing invalid "
1107 "allocation of compressed cluster "
1112 if (!offset
|| ROUND_UP(offset
, s
->cluster_size
) != new_cluster
) {
1113 offset
= new_cluster
;
1114 free_in_cluster
= s
->cluster_size
;
1116 free_in_cluster
+= s
->cluster_size
;
1121 ret
= update_refcount(bs
, offset
, size
, 1, false, QCOW2_DISCARD_NEVER
);
1125 } while (ret
== -EAGAIN
);
1130 /* The cluster refcount was incremented; refcount blocks must be flushed
1131 * before the caller's L2 table updates. */
1132 qcow2_cache_set_dependency(bs
, s
->l2_table_cache
, s
->refcount_block_cache
);
1134 s
->free_byte_offset
= offset
+ size
;
1135 if (!offset_into_cluster(s
, s
->free_byte_offset
)) {
1136 s
->free_byte_offset
= 0;
1142 void qcow2_free_clusters(BlockDriverState
*bs
,
1143 int64_t offset
, int64_t size
,
1144 enum qcow2_discard_type type
)
1148 BLKDBG_EVENT(bs
->file
, BLKDBG_CLUSTER_FREE
);
1149 ret
= update_refcount(bs
, offset
, size
, 1, true, type
);
1151 fprintf(stderr
, "qcow2_free_clusters failed: %s\n", strerror(-ret
));
1152 /* TODO Remember the clusters to free them later and avoid leaking */
1157 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
1158 * normal cluster, compressed cluster, etc.)
1160 void qcow2_free_any_cluster(BlockDriverState
*bs
, uint64_t l2_entry
,
1161 enum qcow2_discard_type type
)
1163 BDRVQcow2State
*s
= bs
->opaque
;
1164 QCow2ClusterType ctype
= qcow2_get_cluster_type(bs
, l2_entry
);
1166 if (has_data_file(bs
)) {
1167 if (s
->discard_passthrough
[type
] &&
1168 (ctype
== QCOW2_CLUSTER_NORMAL
||
1169 ctype
== QCOW2_CLUSTER_ZERO_ALLOC
))
1171 bdrv_pdiscard(s
->data_file
, l2_entry
& L2E_OFFSET_MASK
,
1178 case QCOW2_CLUSTER_COMPRESSED
:
1183 qcow2_parse_compressed_l2_entry(bs
, l2_entry
, &coffset
, &csize
);
1184 qcow2_free_clusters(bs
, coffset
, csize
, type
);
1187 case QCOW2_CLUSTER_NORMAL
:
1188 case QCOW2_CLUSTER_ZERO_ALLOC
:
1189 if (offset_into_cluster(s
, l2_entry
& L2E_OFFSET_MASK
)) {
1190 qcow2_signal_corruption(bs
, false, -1, -1,
1191 "Cannot free unaligned cluster %#llx",
1192 l2_entry
& L2E_OFFSET_MASK
);
1194 qcow2_free_clusters(bs
, l2_entry
& L2E_OFFSET_MASK
,
1195 s
->cluster_size
, type
);
1198 case QCOW2_CLUSTER_ZERO_PLAIN
:
1199 case QCOW2_CLUSTER_UNALLOCATED
:
1206 int coroutine_fn
qcow2_write_caches(BlockDriverState
*bs
)
1208 BDRVQcow2State
*s
= bs
->opaque
;
1211 ret
= qcow2_cache_write(bs
, s
->l2_table_cache
);
1216 if (qcow2_need_accurate_refcounts(s
)) {
1217 ret
= qcow2_cache_write(bs
, s
->refcount_block_cache
);
1226 int coroutine_fn
qcow2_flush_caches(BlockDriverState
*bs
)
1228 int ret
= qcow2_write_caches(bs
);
1233 return bdrv_flush(bs
->file
->bs
);
1236 /*********************************************************/
1237 /* snapshots and image creation */
1241 /* update the refcounts of snapshots and the copied flag */
1242 int qcow2_update_snapshot_refcount(BlockDriverState
*bs
,
1243 int64_t l1_table_offset
, int l1_size
, int addend
)
1245 BDRVQcow2State
*s
= bs
->opaque
;
1246 uint64_t *l1_table
, *l2_slice
, l2_offset
, entry
, l1_size2
, refcount
;
1247 bool l1_allocated
= false;
1248 int64_t old_entry
, old_l2_offset
;
1249 unsigned slice
, slice_size2
, n_slices
;
1250 int i
, j
, l1_modified
= 0;
1253 assert(addend
>= -1 && addend
<= 1);
1257 l1_size2
= l1_size
* L1E_SIZE
;
1258 slice_size2
= s
->l2_slice_size
* l2_entry_size(s
);
1259 n_slices
= s
->cluster_size
/ slice_size2
;
1261 s
->cache_discards
= true;
1263 /* WARNING: qcow2_snapshot_goto relies on this function not using the
1264 * l1_table_offset when it is the current s->l1_table_offset! Be careful
1265 * when changing this! */
1266 if (l1_table_offset
!= s
->l1_table_offset
) {
1267 l1_table
= g_try_malloc0(l1_size2
);
1268 if (l1_size2
&& l1_table
== NULL
) {
1272 l1_allocated
= true;
1274 ret
= bdrv_pread(bs
->file
, l1_table_offset
, l1_table
, l1_size2
);
1279 for (i
= 0; i
< l1_size
; i
++) {
1280 be64_to_cpus(&l1_table
[i
]);
1283 assert(l1_size
== s
->l1_size
);
1284 l1_table
= s
->l1_table
;
1285 l1_allocated
= false;
1288 for (i
= 0; i
< l1_size
; i
++) {
1289 l2_offset
= l1_table
[i
];
1291 old_l2_offset
= l2_offset
;
1292 l2_offset
&= L1E_OFFSET_MASK
;
1294 if (offset_into_cluster(s
, l2_offset
)) {
1295 qcow2_signal_corruption(bs
, true, -1, -1, "L2 table offset %#"
1296 PRIx64
" unaligned (L1 index: %#x)",
1302 for (slice
= 0; slice
< n_slices
; slice
++) {
1303 ret
= qcow2_cache_get(bs
, s
->l2_table_cache
,
1304 l2_offset
+ slice
* slice_size2
,
1305 (void **) &l2_slice
);
1310 for (j
= 0; j
< s
->l2_slice_size
; j
++) {
1311 uint64_t cluster_index
;
1314 entry
= get_l2_entry(s
, l2_slice
, j
);
1316 entry
&= ~QCOW_OFLAG_COPIED
;
1317 offset
= entry
& L2E_OFFSET_MASK
;
1319 switch (qcow2_get_cluster_type(bs
, entry
)) {
1320 case QCOW2_CLUSTER_COMPRESSED
:
1325 qcow2_parse_compressed_l2_entry(bs
, entry
,
1327 ret
= update_refcount(
1329 abs(addend
), addend
< 0,
1330 QCOW2_DISCARD_SNAPSHOT
);
1335 /* compressed clusters are never modified */
1339 case QCOW2_CLUSTER_NORMAL
:
1340 case QCOW2_CLUSTER_ZERO_ALLOC
:
1341 if (offset_into_cluster(s
, offset
)) {
1342 /* Here l2_index means table (not slice) index */
1343 int l2_index
= slice
* s
->l2_slice_size
+ j
;
1344 qcow2_signal_corruption(
1345 bs
, true, -1, -1, "Cluster "
1346 "allocation offset %#" PRIx64
1347 " unaligned (L2 offset: %#"
1348 PRIx64
", L2 index: %#x)",
1349 offset
, l2_offset
, l2_index
);
1354 cluster_index
= offset
>> s
->cluster_bits
;
1355 assert(cluster_index
);
1357 ret
= qcow2_update_cluster_refcount(
1358 bs
, cluster_index
, abs(addend
), addend
< 0,
1359 QCOW2_DISCARD_SNAPSHOT
);
1365 ret
= qcow2_get_refcount(bs
, cluster_index
, &refcount
);
1371 case QCOW2_CLUSTER_ZERO_PLAIN
:
1372 case QCOW2_CLUSTER_UNALLOCATED
:
1380 if (refcount
== 1) {
1381 entry
|= QCOW_OFLAG_COPIED
;
1383 if (entry
!= old_entry
) {
1385 qcow2_cache_set_dependency(bs
, s
->l2_table_cache
,
1386 s
->refcount_block_cache
);
1388 set_l2_entry(s
, l2_slice
, j
, entry
);
1389 qcow2_cache_entry_mark_dirty(s
->l2_table_cache
,
1394 qcow2_cache_put(s
->l2_table_cache
, (void **) &l2_slice
);
1398 ret
= qcow2_update_cluster_refcount(bs
, l2_offset
>>
1400 abs(addend
), addend
< 0,
1401 QCOW2_DISCARD_SNAPSHOT
);
1406 ret
= qcow2_get_refcount(bs
, l2_offset
>> s
->cluster_bits
,
1410 } else if (refcount
== 1) {
1411 l2_offset
|= QCOW_OFLAG_COPIED
;
1413 if (l2_offset
!= old_l2_offset
) {
1414 l1_table
[i
] = l2_offset
;
1420 ret
= bdrv_flush(bs
);
1423 qcow2_cache_put(s
->l2_table_cache
, (void **) &l2_slice
);
1426 s
->cache_discards
= false;
1427 qcow2_process_discards(bs
, ret
);
1429 /* Update L1 only if it isn't deleted anyway (addend = -1) */
1430 if (ret
== 0 && addend
>= 0 && l1_modified
) {
1431 for (i
= 0; i
< l1_size
; i
++) {
1432 cpu_to_be64s(&l1_table
[i
]);
1435 ret
= bdrv_pwrite_sync(bs
->file
, l1_table_offset
,
1436 l1_table
, l1_size2
);
1438 for (i
= 0; i
< l1_size
; i
++) {
1439 be64_to_cpus(&l1_table
[i
]);
1450 /*********************************************************/
1451 /* refcount checking functions */
1454 static uint64_t refcount_array_byte_size(BDRVQcow2State
*s
, uint64_t entries
)
1456 /* This assertion holds because there is no way we can address more than
1457 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because
1458 * offsets have to be representable in bytes); due to every cluster
1459 * corresponding to one refcount entry, we are well below that limit */
1460 assert(entries
< (UINT64_C(1) << (64 - 9)));
1462 /* Thanks to the assertion this will not overflow, because
1463 * s->refcount_order < 7.
1464 * (note: x << s->refcount_order == x * s->refcount_bits) */
1465 return DIV_ROUND_UP(entries
<< s
->refcount_order
, 8);
1469 * Reallocates *array so that it can hold new_size entries. *size must contain
1470 * the current number of entries in *array. If the reallocation fails, *array
1471 * and *size will not be modified and -errno will be returned. If the
1472 * reallocation is successful, *array will be set to the new buffer, *size
1473 * will be set to new_size and 0 will be returned. The size of the reallocated
1474 * refcount array buffer will be aligned to a cluster boundary, and the newly
1475 * allocated area will be zeroed.
1477 static int realloc_refcount_array(BDRVQcow2State
*s
, void **array
,
1478 int64_t *size
, int64_t new_size
)
1480 int64_t old_byte_size
, new_byte_size
;
1483 /* Round to clusters so the array can be directly written to disk */
1484 old_byte_size
= size_to_clusters(s
, refcount_array_byte_size(s
, *size
))
1486 new_byte_size
= size_to_clusters(s
, refcount_array_byte_size(s
, new_size
))
1489 if (new_byte_size
== old_byte_size
) {
1494 assert(new_byte_size
> 0);
1496 if (new_byte_size
> SIZE_MAX
) {
1500 new_ptr
= g_try_realloc(*array
, new_byte_size
);
1505 if (new_byte_size
> old_byte_size
) {
1506 memset((char *)new_ptr
+ old_byte_size
, 0,
1507 new_byte_size
- old_byte_size
);
1517 * Increases the refcount for a range of clusters in a given refcount table.
1518 * This is used to construct a temporary refcount table out of L1 and L2 tables
1519 * which can be compared to the refcount table saved in the image.
1521 * Modifies the number of errors in res.
1523 int qcow2_inc_refcounts_imrt(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1524 void **refcount_table
,
1525 int64_t *refcount_table_size
,
1526 int64_t offset
, int64_t size
)
1528 BDRVQcow2State
*s
= bs
->opaque
;
1529 uint64_t start
, last
, cluster_offset
, k
, refcount
;
1537 file_len
= bdrv_getlength(bs
->file
->bs
);
1543 * Last cluster of qcow2 image may be semi-allocated, so it may be OK to
1544 * reference some space after file end but it should be less than one
1547 if (offset
+ size
- file_len
>= s
->cluster_size
) {
1548 fprintf(stderr
, "ERROR: counting reference for region exceeding the "
1549 "end of the file by one cluster or more: offset 0x%" PRIx64
1550 " size 0x%" PRIx64
"\n", offset
, size
);
1555 start
= start_of_cluster(s
, offset
);
1556 last
= start_of_cluster(s
, offset
+ size
- 1);
1557 for(cluster_offset
= start
; cluster_offset
<= last
;
1558 cluster_offset
+= s
->cluster_size
) {
1559 k
= cluster_offset
>> s
->cluster_bits
;
1560 if (k
>= *refcount_table_size
) {
1561 ret
= realloc_refcount_array(s
, refcount_table
,
1562 refcount_table_size
, k
+ 1);
1564 res
->check_errors
++;
1569 refcount
= s
->get_refcount(*refcount_table
, k
);
1570 if (refcount
== s
->refcount_max
) {
1571 fprintf(stderr
, "ERROR: overflow cluster offset=0x%" PRIx64
1572 "\n", cluster_offset
);
1573 fprintf(stderr
, "Use qemu-img amend to increase the refcount entry "
1574 "width or qemu-img convert to create a clean copy if the "
1575 "image cannot be opened for writing\n");
1579 s
->set_refcount(*refcount_table
, k
, refcount
+ 1);
1585 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
1587 CHECK_FRAG_INFO
= 0x2, /* update BlockFragInfo counters */
1591 * Fix L2 entry by making it QCOW2_CLUSTER_ZERO_PLAIN (or making all its present
1592 * subclusters QCOW2_SUBCLUSTER_ZERO_PLAIN).
1594 * This function decrements res->corruptions on success, so the caller is
1595 * responsible to increment res->corruptions prior to the call.
1597 * On failure in-memory @l2_table may be modified.
1599 static int fix_l2_entry_by_zero(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1601 uint64_t *l2_table
, int l2_index
, bool active
,
1602 bool *metadata_overlap
)
1604 BDRVQcow2State
*s
= bs
->opaque
;
1606 int idx
= l2_index
* (l2_entry_size(s
) / sizeof(uint64_t));
1607 uint64_t l2e_offset
= l2_offset
+ (uint64_t)l2_index
* l2_entry_size(s
);
1608 int ign
= active
? QCOW2_OL_ACTIVE_L2
: QCOW2_OL_INACTIVE_L2
;
1610 if (has_subclusters(s
)) {
1611 uint64_t l2_bitmap
= get_l2_bitmap(s
, l2_table
, l2_index
);
1613 /* Allocated subclusters become zero */
1614 l2_bitmap
|= l2_bitmap
<< 32;
1615 l2_bitmap
&= QCOW_L2_BITMAP_ALL_ZEROES
;
1617 set_l2_bitmap(s
, l2_table
, l2_index
, l2_bitmap
);
1618 set_l2_entry(s
, l2_table
, l2_index
, 0);
1620 set_l2_entry(s
, l2_table
, l2_index
, QCOW_OFLAG_ZERO
);
1623 ret
= qcow2_pre_write_overlap_check(bs
, ign
, l2e_offset
, l2_entry_size(s
),
1625 if (metadata_overlap
) {
1626 *metadata_overlap
= ret
< 0;
1629 fprintf(stderr
, "ERROR: Overlap check failed\n");
1633 ret
= bdrv_pwrite_sync(bs
->file
, l2e_offset
, &l2_table
[idx
],
1636 fprintf(stderr
, "ERROR: Failed to overwrite L2 "
1637 "table entry: %s\n", strerror(-ret
));
1642 res
->corruptions_fixed
++;
1646 res
->check_errors
++;
1651 * Increases the refcount in the given refcount table for the all clusters
1652 * referenced in the L2 table. While doing so, performs some checks on L2
1655 * Returns the number of errors found by the checks or -errno if an internal
1658 static int check_refcounts_l2(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1659 void **refcount_table
,
1660 int64_t *refcount_table_size
, int64_t l2_offset
,
1661 int flags
, BdrvCheckMode fix
, bool active
)
1663 BDRVQcow2State
*s
= bs
->opaque
;
1664 uint64_t l2_entry
, l2_bitmap
;
1665 uint64_t next_contiguous_offset
= 0;
1667 size_t l2_size_bytes
= s
->l2_size
* l2_entry_size(s
);
1668 g_autofree
uint64_t *l2_table
= g_malloc(l2_size_bytes
);
1669 bool metadata_overlap
;
1671 /* Read L2 table from disk */
1672 ret
= bdrv_pread(bs
->file
, l2_offset
, l2_table
, l2_size_bytes
);
1674 fprintf(stderr
, "ERROR: I/O error in check_refcounts_l2\n");
1675 res
->check_errors
++;
1679 /* Do the actual checks */
1680 for (i
= 0; i
< s
->l2_size
; i
++) {
1683 QCow2ClusterType type
;
1685 l2_entry
= get_l2_entry(s
, l2_table
, i
);
1686 l2_bitmap
= get_l2_bitmap(s
, l2_table
, i
);
1687 type
= qcow2_get_cluster_type(bs
, l2_entry
);
1689 if (type
!= QCOW2_CLUSTER_COMPRESSED
) {
1690 /* Check reserved bits of Standard Cluster Descriptor */
1691 if (l2_entry
& L2E_STD_RESERVED_MASK
) {
1692 fprintf(stderr
, "ERROR found l2 entry with reserved bits set: "
1693 "%" PRIx64
"\n", l2_entry
);
1699 case QCOW2_CLUSTER_COMPRESSED
:
1700 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1701 if (l2_entry
& QCOW_OFLAG_COPIED
) {
1702 fprintf(stderr
, "ERROR: coffset=0x%" PRIx64
": "
1703 "copied flag must never be set for compressed "
1704 "clusters\n", l2_entry
& s
->cluster_offset_mask
);
1705 l2_entry
&= ~QCOW_OFLAG_COPIED
;
1709 if (has_data_file(bs
)) {
1710 fprintf(stderr
, "ERROR compressed cluster %d with data file, "
1711 "entry=0x%" PRIx64
"\n", i
, l2_entry
);
1717 fprintf(stderr
, "ERROR compressed cluster %d with non-zero "
1718 "subcluster allocation bitmap, entry=0x%" PRIx64
"\n",
1724 /* Mark cluster as used */
1725 qcow2_parse_compressed_l2_entry(bs
, l2_entry
, &coffset
, &csize
);
1726 ret
= qcow2_inc_refcounts_imrt(
1727 bs
, res
, refcount_table
, refcount_table_size
, coffset
, csize
);
1732 if (flags
& CHECK_FRAG_INFO
) {
1733 res
->bfi
.allocated_clusters
++;
1734 res
->bfi
.compressed_clusters
++;
1737 * Compressed clusters are fragmented by nature. Since they
1738 * take up sub-sector space but we only have sector granularity
1739 * I/O we need to re-read the same sectors even for adjacent
1740 * compressed clusters.
1742 res
->bfi
.fragmented_clusters
++;
1746 case QCOW2_CLUSTER_ZERO_ALLOC
:
1747 case QCOW2_CLUSTER_NORMAL
:
1749 uint64_t offset
= l2_entry
& L2E_OFFSET_MASK
;
1751 if ((l2_bitmap
>> 32) & l2_bitmap
) {
1753 fprintf(stderr
, "ERROR offset=%" PRIx64
": Allocated "
1754 "cluster has corrupted subcluster allocation bitmap\n",
1758 /* Correct offsets are cluster aligned */
1759 if (offset_into_cluster(s
, offset
)) {
1763 if (has_subclusters(s
)) {
1764 contains_data
= (l2_bitmap
& QCOW_L2_BITMAP_ALL_ALLOC
);
1766 contains_data
= !(l2_entry
& QCOW_OFLAG_ZERO
);
1769 if (!contains_data
) {
1770 fprintf(stderr
, "%s offset=%" PRIx64
": Preallocated "
1771 "cluster is not properly aligned; L2 entry "
1773 fix
& BDRV_FIX_ERRORS
? "Repairing" : "ERROR",
1775 if (fix
& BDRV_FIX_ERRORS
) {
1776 ret
= fix_l2_entry_by_zero(bs
, res
, l2_offset
,
1777 l2_table
, i
, active
,
1779 if (metadata_overlap
) {
1781 * Something is seriously wrong, so abort checking
1789 * Skip marking the cluster as used
1790 * (it is unused now).
1797 * Do not abort, continue checking the rest of this
1798 * L2 table's entries.
1802 fprintf(stderr
, "ERROR offset=%" PRIx64
": Data cluster is "
1803 "not properly aligned; L2 entry corrupted.\n", offset
);
1807 if (flags
& CHECK_FRAG_INFO
) {
1808 res
->bfi
.allocated_clusters
++;
1809 if (next_contiguous_offset
&&
1810 offset
!= next_contiguous_offset
) {
1811 res
->bfi
.fragmented_clusters
++;
1813 next_contiguous_offset
= offset
+ s
->cluster_size
;
1816 /* Mark cluster as used */
1817 if (!has_data_file(bs
)) {
1818 ret
= qcow2_inc_refcounts_imrt(bs
, res
, refcount_table
,
1819 refcount_table_size
,
1820 offset
, s
->cluster_size
);
1828 case QCOW2_CLUSTER_ZERO_PLAIN
:
1829 /* Impossible when image has subclusters */
1833 case QCOW2_CLUSTER_UNALLOCATED
:
1834 if (l2_bitmap
& QCOW_L2_BITMAP_ALL_ALLOC
) {
1836 fprintf(stderr
, "ERROR: Unallocated "
1837 "cluster has non-zero subcluster allocation map\n");
1850 * Increases the refcount for the L1 table, its L2 tables and all referenced
1851 * clusters in the given refcount table. While doing so, performs some checks
1852 * on L1 and L2 entries.
1854 * Returns the number of errors found by the checks or -errno if an internal
1857 static int check_refcounts_l1(BlockDriverState
*bs
,
1858 BdrvCheckResult
*res
,
1859 void **refcount_table
,
1860 int64_t *refcount_table_size
,
1861 int64_t l1_table_offset
, int l1_size
,
1862 int flags
, BdrvCheckMode fix
, bool active
)
1864 BDRVQcow2State
*s
= bs
->opaque
;
1865 size_t l1_size_bytes
= l1_size
* L1E_SIZE
;
1866 g_autofree
uint64_t *l1_table
= NULL
;
1874 /* Mark L1 table as used */
1875 ret
= qcow2_inc_refcounts_imrt(bs
, res
, refcount_table
, refcount_table_size
,
1876 l1_table_offset
, l1_size_bytes
);
1881 l1_table
= g_try_malloc(l1_size_bytes
);
1882 if (l1_table
== NULL
) {
1883 res
->check_errors
++;
1887 /* Read L1 table entries from disk */
1888 ret
= bdrv_pread(bs
->file
, l1_table_offset
, l1_table
, l1_size_bytes
);
1890 fprintf(stderr
, "ERROR: I/O error in check_refcounts_l1\n");
1891 res
->check_errors
++;
1895 for (i
= 0; i
< l1_size
; i
++) {
1896 be64_to_cpus(&l1_table
[i
]);
1899 /* Do the actual checks */
1900 for (i
= 0; i
< l1_size
; i
++) {
1905 if (l1_table
[i
] & L1E_RESERVED_MASK
) {
1906 fprintf(stderr
, "ERROR found L1 entry with reserved bits set: "
1907 "%" PRIx64
"\n", l1_table
[i
]);
1911 l2_offset
= l1_table
[i
] & L1E_OFFSET_MASK
;
1913 /* Mark L2 table as used */
1914 ret
= qcow2_inc_refcounts_imrt(bs
, res
,
1915 refcount_table
, refcount_table_size
,
1916 l2_offset
, s
->cluster_size
);
1921 /* L2 tables are cluster aligned */
1922 if (offset_into_cluster(s
, l2_offset
)) {
1923 fprintf(stderr
, "ERROR l2_offset=%" PRIx64
": Table is not "
1924 "cluster aligned; L1 entry corrupted\n", l2_offset
);
1928 /* Process and check L2 entries */
1929 ret
= check_refcounts_l2(bs
, res
, refcount_table
,
1930 refcount_table_size
, l2_offset
, flags
,
1941 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1943 * This function does not print an error message nor does it increment
1944 * check_errors if qcow2_get_refcount fails (this is because such an error will
1945 * have been already detected and sufficiently signaled by the calling function
1946 * (qcow2_check_refcounts) by the time this function is called).
1948 static int check_oflag_copied(BlockDriverState
*bs
, BdrvCheckResult
*res
,
1951 BDRVQcow2State
*s
= bs
->opaque
;
1952 uint64_t *l2_table
= qemu_blockalign(bs
, s
->cluster_size
);
1958 if (fix
& BDRV_FIX_ERRORS
) {
1961 } else if (fix
& BDRV_FIX_LEAKS
) {
1962 /* Repair only if that seems safe: This function is always
1963 * called after the refcounts have been fixed, so the refcount
1964 * is accurate if that repair was successful */
1965 repair
= !res
->check_errors
&& !res
->corruptions
&& !res
->leaks
;
1970 for (i
= 0; i
< s
->l1_size
; i
++) {
1971 uint64_t l1_entry
= s
->l1_table
[i
];
1972 uint64_t l2_offset
= l1_entry
& L1E_OFFSET_MASK
;
1979 ret
= qcow2_get_refcount(bs
, l2_offset
>> s
->cluster_bits
,
1982 /* don't print message nor increment check_errors */
1985 if ((refcount
== 1) != ((l1_entry
& QCOW_OFLAG_COPIED
) != 0)) {
1987 fprintf(stderr
, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1988 "l1_entry=%" PRIx64
" refcount=%" PRIu64
"\n",
1989 repair
? "Repairing" : "ERROR", i
, l1_entry
, refcount
);
1991 s
->l1_table
[i
] = refcount
== 1
1992 ? l1_entry
| QCOW_OFLAG_COPIED
1993 : l1_entry
& ~QCOW_OFLAG_COPIED
;
1994 ret
= qcow2_write_l1_entry(bs
, i
);
1996 res
->check_errors
++;
2000 res
->corruptions_fixed
++;
2004 ret
= bdrv_pread(bs
->file
, l2_offset
, l2_table
,
2005 s
->l2_size
* l2_entry_size(s
));
2007 fprintf(stderr
, "ERROR: Could not read L2 table: %s\n",
2009 res
->check_errors
++;
2013 for (j
= 0; j
< s
->l2_size
; j
++) {
2014 uint64_t l2_entry
= get_l2_entry(s
, l2_table
, j
);
2015 uint64_t data_offset
= l2_entry
& L2E_OFFSET_MASK
;
2016 QCow2ClusterType cluster_type
= qcow2_get_cluster_type(bs
, l2_entry
);
2018 if (cluster_type
== QCOW2_CLUSTER_NORMAL
||
2019 cluster_type
== QCOW2_CLUSTER_ZERO_ALLOC
) {
2020 if (has_data_file(bs
)) {
2023 ret
= qcow2_get_refcount(bs
,
2024 data_offset
>> s
->cluster_bits
,
2027 /* don't print message nor increment check_errors */
2031 if ((refcount
== 1) != ((l2_entry
& QCOW_OFLAG_COPIED
) != 0)) {
2033 fprintf(stderr
, "%s OFLAG_COPIED data cluster: "
2034 "l2_entry=%" PRIx64
" refcount=%" PRIu64
"\n",
2035 repair
? "Repairing" : "ERROR", l2_entry
, refcount
);
2037 set_l2_entry(s
, l2_table
, j
,
2039 l2_entry
| QCOW_OFLAG_COPIED
:
2040 l2_entry
& ~QCOW_OFLAG_COPIED
);
2048 ret
= qcow2_pre_write_overlap_check(bs
, QCOW2_OL_ACTIVE_L2
,
2049 l2_offset
, s
->cluster_size
,
2052 fprintf(stderr
, "ERROR: Could not write L2 table; metadata "
2053 "overlap check failed: %s\n", strerror(-ret
));
2054 res
->check_errors
++;
2058 ret
= bdrv_pwrite(bs
->file
, l2_offset
, l2_table
,
2061 fprintf(stderr
, "ERROR: Could not write L2 table: %s\n",
2063 res
->check_errors
++;
2066 res
->corruptions
-= l2_dirty
;
2067 res
->corruptions_fixed
+= l2_dirty
;
2074 qemu_vfree(l2_table
);
2079 * Checks consistency of refblocks and accounts for each refblock in
2082 static int check_refblocks(BlockDriverState
*bs
, BdrvCheckResult
*res
,
2083 BdrvCheckMode fix
, bool *rebuild
,
2084 void **refcount_table
, int64_t *nb_clusters
)
2086 BDRVQcow2State
*s
= bs
->opaque
;
2090 for(i
= 0; i
< s
->refcount_table_size
; i
++) {
2091 uint64_t offset
, cluster
;
2092 offset
= s
->refcount_table
[i
] & REFT_OFFSET_MASK
;
2093 cluster
= offset
>> s
->cluster_bits
;
2095 if (s
->refcount_table
[i
] & REFT_RESERVED_MASK
) {
2096 fprintf(stderr
, "ERROR refcount table entry %" PRId64
" has "
2097 "reserved bits set\n", i
);
2103 /* Refcount blocks are cluster aligned */
2104 if (offset_into_cluster(s
, offset
)) {
2105 fprintf(stderr
, "ERROR refcount block %" PRId64
" is not "
2106 "cluster aligned; refcount table entry corrupted\n", i
);
2112 if (cluster
>= *nb_clusters
) {
2114 fprintf(stderr
, "%s refcount block %" PRId64
" is outside image\n",
2115 fix
& BDRV_FIX_ERRORS
? "Repairing" : "ERROR", i
);
2117 if (fix
& BDRV_FIX_ERRORS
) {
2118 int64_t new_nb_clusters
;
2119 Error
*local_err
= NULL
;
2121 if (offset
> INT64_MAX
- s
->cluster_size
) {
2126 ret
= bdrv_truncate(bs
->file
, offset
+ s
->cluster_size
, false,
2127 PREALLOC_MODE_OFF
, 0, &local_err
);
2129 error_report_err(local_err
);
2132 size
= bdrv_getlength(bs
->file
->bs
);
2138 new_nb_clusters
= size_to_clusters(s
, size
);
2139 assert(new_nb_clusters
>= *nb_clusters
);
2141 ret
= realloc_refcount_array(s
, refcount_table
,
2142 nb_clusters
, new_nb_clusters
);
2144 res
->check_errors
++;
2148 if (cluster
>= *nb_clusters
) {
2154 res
->corruptions_fixed
++;
2155 ret
= qcow2_inc_refcounts_imrt(bs
, res
,
2156 refcount_table
, nb_clusters
,
2157 offset
, s
->cluster_size
);
2161 /* No need to check whether the refcount is now greater than 1:
2162 * This area was just allocated and zeroed, so it can only be
2163 * exactly 1 after qcow2_inc_refcounts_imrt() */
2168 fprintf(stderr
, "ERROR could not resize image: %s\n",
2175 ret
= qcow2_inc_refcounts_imrt(bs
, res
, refcount_table
, nb_clusters
,
2176 offset
, s
->cluster_size
);
2180 if (s
->get_refcount(*refcount_table
, cluster
) != 1) {
2181 fprintf(stderr
, "ERROR refcount block %" PRId64
2182 " refcount=%" PRIu64
"\n", i
,
2183 s
->get_refcount(*refcount_table
, cluster
));
2194 * Calculates an in-memory refcount table.
2196 static int calculate_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
2197 BdrvCheckMode fix
, bool *rebuild
,
2198 void **refcount_table
, int64_t *nb_clusters
)
2200 BDRVQcow2State
*s
= bs
->opaque
;
2205 if (!*refcount_table
) {
2206 int64_t old_size
= 0;
2207 ret
= realloc_refcount_array(s
, refcount_table
,
2208 &old_size
, *nb_clusters
);
2210 res
->check_errors
++;
2216 ret
= qcow2_inc_refcounts_imrt(bs
, res
, refcount_table
, nb_clusters
,
2217 0, s
->cluster_size
);
2222 /* current L1 table */
2223 ret
= check_refcounts_l1(bs
, res
, refcount_table
, nb_clusters
,
2224 s
->l1_table_offset
, s
->l1_size
, CHECK_FRAG_INFO
,
2231 if (has_data_file(bs
) && s
->nb_snapshots
) {
2232 fprintf(stderr
, "ERROR %d snapshots in image with data file\n",
2237 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
2238 sn
= s
->snapshots
+ i
;
2239 if (offset_into_cluster(s
, sn
->l1_table_offset
)) {
2240 fprintf(stderr
, "ERROR snapshot %s (%s) l1_offset=%#" PRIx64
": "
2241 "L1 table is not cluster aligned; snapshot table entry "
2242 "corrupted\n", sn
->id_str
, sn
->name
, sn
->l1_table_offset
);
2246 if (sn
->l1_size
> QCOW_MAX_L1_SIZE
/ L1E_SIZE
) {
2247 fprintf(stderr
, "ERROR snapshot %s (%s) l1_size=%#" PRIx32
": "
2248 "L1 table is too large; snapshot table entry corrupted\n",
2249 sn
->id_str
, sn
->name
, sn
->l1_size
);
2253 ret
= check_refcounts_l1(bs
, res
, refcount_table
, nb_clusters
,
2254 sn
->l1_table_offset
, sn
->l1_size
, 0, fix
,
2260 ret
= qcow2_inc_refcounts_imrt(bs
, res
, refcount_table
, nb_clusters
,
2261 s
->snapshots_offset
, s
->snapshots_size
);
2267 ret
= qcow2_inc_refcounts_imrt(bs
, res
, refcount_table
, nb_clusters
,
2268 s
->refcount_table_offset
,
2269 s
->refcount_table_size
*
2270 REFTABLE_ENTRY_SIZE
);
2276 if (s
->crypto_header
.length
) {
2277 ret
= qcow2_inc_refcounts_imrt(bs
, res
, refcount_table
, nb_clusters
,
2278 s
->crypto_header
.offset
,
2279 s
->crypto_header
.length
);
2286 ret
= qcow2_check_bitmaps_refcounts(bs
, res
, refcount_table
, nb_clusters
);
2291 return check_refblocks(bs
, res
, fix
, rebuild
, refcount_table
, nb_clusters
);
2295 * Compares the actual reference count for each cluster in the image against the
2296 * refcount as reported by the refcount structures on-disk.
2298 static void compare_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
2299 BdrvCheckMode fix
, bool *rebuild
,
2300 int64_t *highest_cluster
,
2301 void *refcount_table
, int64_t nb_clusters
)
2303 BDRVQcow2State
*s
= bs
->opaque
;
2305 uint64_t refcount1
, refcount2
;
2308 for (i
= 0, *highest_cluster
= 0; i
< nb_clusters
; i
++) {
2309 ret
= qcow2_get_refcount(bs
, i
, &refcount1
);
2311 fprintf(stderr
, "Can't get refcount for cluster %" PRId64
": %s\n",
2313 res
->check_errors
++;
2317 refcount2
= s
->get_refcount(refcount_table
, i
);
2319 if (refcount1
> 0 || refcount2
> 0) {
2320 *highest_cluster
= i
;
2323 if (refcount1
!= refcount2
) {
2324 /* Check if we're allowed to fix the mismatch */
2325 int *num_fixed
= NULL
;
2326 if (refcount1
== 0) {
2328 } else if (refcount1
> refcount2
&& (fix
& BDRV_FIX_LEAKS
)) {
2329 num_fixed
= &res
->leaks_fixed
;
2330 } else if (refcount1
< refcount2
&& (fix
& BDRV_FIX_ERRORS
)) {
2331 num_fixed
= &res
->corruptions_fixed
;
2334 fprintf(stderr
, "%s cluster %" PRId64
" refcount=%" PRIu64
2335 " reference=%" PRIu64
"\n",
2336 num_fixed
!= NULL
? "Repairing" :
2337 refcount1
< refcount2
? "ERROR" :
2339 i
, refcount1
, refcount2
);
2342 ret
= update_refcount(bs
, i
<< s
->cluster_bits
, 1,
2343 refcount_diff(refcount1
, refcount2
),
2344 refcount1
> refcount2
,
2345 QCOW2_DISCARD_ALWAYS
);
2352 /* And if we couldn't, print an error */
2353 if (refcount1
< refcount2
) {
2363 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to
2364 * the on-disk refcount structures.
2366 * On input, *first_free_cluster tells where to start looking, and need not
2367 * actually be a free cluster; the returned offset will not be before that
2368 * cluster. On output, *first_free_cluster points to the first gap found, even
2369 * if that gap was too small to be used as the returned offset.
2371 * Note that *first_free_cluster is a cluster index whereas the return value is
2374 static int64_t alloc_clusters_imrt(BlockDriverState
*bs
,
2376 void **refcount_table
,
2377 int64_t *imrt_nb_clusters
,
2378 int64_t *first_free_cluster
)
2380 BDRVQcow2State
*s
= bs
->opaque
;
2381 int64_t cluster
= *first_free_cluster
, i
;
2382 bool first_gap
= true;
2383 int contiguous_free_clusters
;
2386 /* Starting at *first_free_cluster, find a range of at least cluster_count
2387 * continuously free clusters */
2388 for (contiguous_free_clusters
= 0;
2389 cluster
< *imrt_nb_clusters
&&
2390 contiguous_free_clusters
< cluster_count
;
2393 if (!s
->get_refcount(*refcount_table
, cluster
)) {
2394 contiguous_free_clusters
++;
2396 /* If this is the first free cluster found, update
2397 * *first_free_cluster accordingly */
2398 *first_free_cluster
= cluster
;
2401 } else if (contiguous_free_clusters
) {
2402 contiguous_free_clusters
= 0;
2406 /* If contiguous_free_clusters is greater than zero, it contains the number
2407 * of continuously free clusters until the current cluster; the first free
2408 * cluster in the current "gap" is therefore
2409 * cluster - contiguous_free_clusters */
2411 /* If no such range could be found, grow the in-memory refcount table
2412 * accordingly to append free clusters at the end of the image */
2413 if (contiguous_free_clusters
< cluster_count
) {
2414 /* contiguous_free_clusters clusters are already empty at the image end;
2415 * we need cluster_count clusters; therefore, we have to allocate
2416 * cluster_count - contiguous_free_clusters new clusters at the end of
2417 * the image (which is the current value of cluster; note that cluster
2418 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond
2420 ret
= realloc_refcount_array(s
, refcount_table
, imrt_nb_clusters
,
2421 cluster
+ cluster_count
2422 - contiguous_free_clusters
);
2428 /* Go back to the first free cluster */
2429 cluster
-= contiguous_free_clusters
;
2430 for (i
= 0; i
< cluster_count
; i
++) {
2431 s
->set_refcount(*refcount_table
, cluster
+ i
, 1);
2434 return cluster
<< s
->cluster_bits
;
2438 * Creates a new refcount structure based solely on the in-memory information
2439 * given through *refcount_table. All necessary allocations will be reflected
2442 * On success, the old refcount structure is leaked (it will be covered by the
2443 * new refcount structure).
2445 static int rebuild_refcount_structure(BlockDriverState
*bs
,
2446 BdrvCheckResult
*res
,
2447 void **refcount_table
,
2448 int64_t *nb_clusters
)
2450 BDRVQcow2State
*s
= bs
->opaque
;
2451 int64_t first_free_cluster
= 0, reftable_offset
= -1, cluster
= 0;
2452 int64_t refblock_offset
, refblock_start
, refblock_index
;
2453 uint32_t reftable_size
= 0;
2454 uint64_t *on_disk_reftable
= NULL
;
2455 void *on_disk_refblock
;
2458 uint64_t reftable_offset
;
2459 uint32_t reftable_clusters
;
2460 } QEMU_PACKED reftable_offset_and_clusters
;
2462 qcow2_cache_empty(bs
, s
->refcount_block_cache
);
2465 for (; cluster
< *nb_clusters
; cluster
++) {
2466 if (!s
->get_refcount(*refcount_table
, cluster
)) {
2470 refblock_index
= cluster
>> s
->refcount_block_bits
;
2471 refblock_start
= refblock_index
<< s
->refcount_block_bits
;
2473 /* Don't allocate a cluster in a refblock already written to disk */
2474 if (first_free_cluster
< refblock_start
) {
2475 first_free_cluster
= refblock_start
;
2477 refblock_offset
= alloc_clusters_imrt(bs
, 1, refcount_table
,
2478 nb_clusters
, &first_free_cluster
);
2479 if (refblock_offset
< 0) {
2480 fprintf(stderr
, "ERROR allocating refblock: %s\n",
2481 strerror(-refblock_offset
));
2482 res
->check_errors
++;
2483 ret
= refblock_offset
;
2487 if (reftable_size
<= refblock_index
) {
2488 uint32_t old_reftable_size
= reftable_size
;
2489 uint64_t *new_on_disk_reftable
;
2491 reftable_size
= ROUND_UP((refblock_index
+ 1) * REFTABLE_ENTRY_SIZE
,
2492 s
->cluster_size
) / REFTABLE_ENTRY_SIZE
;
2493 new_on_disk_reftable
= g_try_realloc(on_disk_reftable
,
2495 REFTABLE_ENTRY_SIZE
);
2496 if (!new_on_disk_reftable
) {
2497 res
->check_errors
++;
2501 on_disk_reftable
= new_on_disk_reftable
;
2503 memset(on_disk_reftable
+ old_reftable_size
, 0,
2504 (reftable_size
- old_reftable_size
) * REFTABLE_ENTRY_SIZE
);
2506 /* The offset we have for the reftable is now no longer valid;
2507 * this will leak that range, but we can easily fix that by running
2508 * a leak-fixing check after this rebuild operation */
2509 reftable_offset
= -1;
2511 assert(on_disk_reftable
);
2513 on_disk_reftable
[refblock_index
] = refblock_offset
;
2515 /* If this is apparently the last refblock (for now), try to squeeze the
2517 if (refblock_index
== (*nb_clusters
- 1) >> s
->refcount_block_bits
&&
2518 reftable_offset
< 0)
2520 uint64_t reftable_clusters
= size_to_clusters(s
, reftable_size
*
2521 REFTABLE_ENTRY_SIZE
);
2522 reftable_offset
= alloc_clusters_imrt(bs
, reftable_clusters
,
2523 refcount_table
, nb_clusters
,
2524 &first_free_cluster
);
2525 if (reftable_offset
< 0) {
2526 fprintf(stderr
, "ERROR allocating reftable: %s\n",
2527 strerror(-reftable_offset
));
2528 res
->check_errors
++;
2529 ret
= reftable_offset
;
2534 ret
= qcow2_pre_write_overlap_check(bs
, 0, refblock_offset
,
2535 s
->cluster_size
, false);
2537 fprintf(stderr
, "ERROR writing refblock: %s\n", strerror(-ret
));
2541 /* The size of *refcount_table is always cluster-aligned, therefore the
2542 * write operation will not overflow */
2543 on_disk_refblock
= (void *)((char *) *refcount_table
+
2544 refblock_index
* s
->cluster_size
);
2546 ret
= bdrv_pwrite(bs
->file
, refblock_offset
, on_disk_refblock
,
2549 fprintf(stderr
, "ERROR writing refblock: %s\n", strerror(-ret
));
2553 /* Go to the end of this refblock */
2554 cluster
= refblock_start
+ s
->refcount_block_size
- 1;
2557 if (reftable_offset
< 0) {
2558 uint64_t post_refblock_start
, reftable_clusters
;
2560 post_refblock_start
= ROUND_UP(*nb_clusters
, s
->refcount_block_size
);
2562 size_to_clusters(s
, reftable_size
* REFTABLE_ENTRY_SIZE
);
2563 /* Not pretty but simple */
2564 if (first_free_cluster
< post_refblock_start
) {
2565 first_free_cluster
= post_refblock_start
;
2567 reftable_offset
= alloc_clusters_imrt(bs
, reftable_clusters
,
2568 refcount_table
, nb_clusters
,
2569 &first_free_cluster
);
2570 if (reftable_offset
< 0) {
2571 fprintf(stderr
, "ERROR allocating reftable: %s\n",
2572 strerror(-reftable_offset
));
2573 res
->check_errors
++;
2574 ret
= reftable_offset
;
2578 goto write_refblocks
;
2581 for (refblock_index
= 0; refblock_index
< reftable_size
; refblock_index
++) {
2582 cpu_to_be64s(&on_disk_reftable
[refblock_index
]);
2585 ret
= qcow2_pre_write_overlap_check(bs
, 0, reftable_offset
,
2586 reftable_size
* REFTABLE_ENTRY_SIZE
,
2589 fprintf(stderr
, "ERROR writing reftable: %s\n", strerror(-ret
));
2593 assert(reftable_size
< INT_MAX
/ REFTABLE_ENTRY_SIZE
);
2594 ret
= bdrv_pwrite(bs
->file
, reftable_offset
, on_disk_reftable
,
2595 reftable_size
* REFTABLE_ENTRY_SIZE
);
2597 fprintf(stderr
, "ERROR writing reftable: %s\n", strerror(-ret
));
2601 /* Enter new reftable into the image header */
2602 reftable_offset_and_clusters
.reftable_offset
= cpu_to_be64(reftable_offset
);
2603 reftable_offset_and_clusters
.reftable_clusters
=
2604 cpu_to_be32(size_to_clusters(s
, reftable_size
* REFTABLE_ENTRY_SIZE
));
2605 ret
= bdrv_pwrite_sync(bs
->file
,
2606 offsetof(QCowHeader
, refcount_table_offset
),
2607 &reftable_offset_and_clusters
,
2608 sizeof(reftable_offset_and_clusters
));
2610 fprintf(stderr
, "ERROR setting reftable: %s\n", strerror(-ret
));
2614 for (refblock_index
= 0; refblock_index
< reftable_size
; refblock_index
++) {
2615 be64_to_cpus(&on_disk_reftable
[refblock_index
]);
2617 s
->refcount_table
= on_disk_reftable
;
2618 s
->refcount_table_offset
= reftable_offset
;
2619 s
->refcount_table_size
= reftable_size
;
2620 update_max_refcount_table_index(s
);
2625 g_free(on_disk_reftable
);
2630 * Checks an image for refcount consistency.
2632 * Returns 0 if no errors are found, the number of errors in case the image is
2633 * detected as corrupted, and -errno when an internal error occurred.
2635 int qcow2_check_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
2638 BDRVQcow2State
*s
= bs
->opaque
;
2639 BdrvCheckResult pre_compare_res
;
2640 int64_t size
, highest_cluster
, nb_clusters
;
2641 void *refcount_table
= NULL
;
2642 bool rebuild
= false;
2645 size
= bdrv_getlength(bs
->file
->bs
);
2647 res
->check_errors
++;
2651 nb_clusters
= size_to_clusters(s
, size
);
2652 if (nb_clusters
> INT_MAX
) {
2653 res
->check_errors
++;
2657 res
->bfi
.total_clusters
=
2658 size_to_clusters(s
, bs
->total_sectors
* BDRV_SECTOR_SIZE
);
2660 ret
= calculate_refcounts(bs
, res
, fix
, &rebuild
, &refcount_table
,
2666 /* In case we don't need to rebuild the refcount structure (but want to fix
2667 * something), this function is immediately called again, in which case the
2668 * result should be ignored */
2669 pre_compare_res
= *res
;
2670 compare_refcounts(bs
, res
, 0, &rebuild
, &highest_cluster
, refcount_table
,
2673 if (rebuild
&& (fix
& BDRV_FIX_ERRORS
)) {
2674 BdrvCheckResult old_res
= *res
;
2675 int fresh_leaks
= 0;
2677 fprintf(stderr
, "Rebuilding refcount structure\n");
2678 ret
= rebuild_refcount_structure(bs
, res
, &refcount_table
,
2684 res
->corruptions
= 0;
2687 /* Because the old reftable has been exchanged for a new one the
2688 * references have to be recalculated */
2690 memset(refcount_table
, 0, refcount_array_byte_size(s
, nb_clusters
));
2691 ret
= calculate_refcounts(bs
, res
, 0, &rebuild
, &refcount_table
,
2697 if (fix
& BDRV_FIX_LEAKS
) {
2698 /* The old refcount structures are now leaked, fix it; the result
2699 * can be ignored, aside from leaks which were introduced by
2700 * rebuild_refcount_structure() that could not be fixed */
2701 BdrvCheckResult saved_res
= *res
;
2702 *res
= (BdrvCheckResult
){ 0 };
2704 compare_refcounts(bs
, res
, BDRV_FIX_LEAKS
, &rebuild
,
2705 &highest_cluster
, refcount_table
, nb_clusters
);
2707 fprintf(stderr
, "ERROR rebuilt refcount structure is still "
2711 /* Any leaks accounted for here were introduced by
2712 * rebuild_refcount_structure() because that function has created a
2713 * new refcount structure from scratch */
2714 fresh_leaks
= res
->leaks
;
2718 if (res
->corruptions
< old_res
.corruptions
) {
2719 res
->corruptions_fixed
+= old_res
.corruptions
- res
->corruptions
;
2721 if (res
->leaks
< old_res
.leaks
) {
2722 res
->leaks_fixed
+= old_res
.leaks
- res
->leaks
;
2724 res
->leaks
+= fresh_leaks
;
2727 fprintf(stderr
, "ERROR need to rebuild refcount structures\n");
2728 res
->check_errors
++;
2733 if (res
->leaks
|| res
->corruptions
) {
2734 *res
= pre_compare_res
;
2735 compare_refcounts(bs
, res
, fix
, &rebuild
, &highest_cluster
,
2736 refcount_table
, nb_clusters
);
2740 /* check OFLAG_COPIED */
2741 ret
= check_oflag_copied(bs
, res
, fix
);
2746 res
->image_end_offset
= (highest_cluster
+ 1) * s
->cluster_size
;
2750 g_free(refcount_table
);
2755 #define overlaps_with(ofs, sz) \
2756 ranges_overlap(offset, size, ofs, sz)
2759 * Checks if the given offset into the image file is actually free to use by
2760 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
2761 * i.e. a sanity check without relying on the refcount tables.
2763 * The ign parameter specifies what checks not to perform (being a bitmask of
2764 * QCow2MetadataOverlap values), i.e., what sections to ignore.
2767 * - 0 if writing to this offset will not affect the mentioned metadata
2768 * - a positive QCow2MetadataOverlap value indicating one overlapping section
2769 * - a negative value (-errno) indicating an error while performing a check,
2770 * e.g. when bdrv_pread failed on QCOW2_OL_INACTIVE_L2
2772 int qcow2_check_metadata_overlap(BlockDriverState
*bs
, int ign
, int64_t offset
,
2775 BDRVQcow2State
*s
= bs
->opaque
;
2776 int chk
= s
->overlap_check
& ~ign
;
2783 if (chk
& QCOW2_OL_MAIN_HEADER
) {
2784 if (offset
< s
->cluster_size
) {
2785 return QCOW2_OL_MAIN_HEADER
;
2789 /* align range to test to cluster boundaries */
2790 size
= ROUND_UP(offset_into_cluster(s
, offset
) + size
, s
->cluster_size
);
2791 offset
= start_of_cluster(s
, offset
);
2793 if ((chk
& QCOW2_OL_ACTIVE_L1
) && s
->l1_size
) {
2794 if (overlaps_with(s
->l1_table_offset
, s
->l1_size
* L1E_SIZE
)) {
2795 return QCOW2_OL_ACTIVE_L1
;
2799 if ((chk
& QCOW2_OL_REFCOUNT_TABLE
) && s
->refcount_table_size
) {
2800 if (overlaps_with(s
->refcount_table_offset
,
2801 s
->refcount_table_size
* REFTABLE_ENTRY_SIZE
)) {
2802 return QCOW2_OL_REFCOUNT_TABLE
;
2806 if ((chk
& QCOW2_OL_SNAPSHOT_TABLE
) && s
->snapshots_size
) {
2807 if (overlaps_with(s
->snapshots_offset
, s
->snapshots_size
)) {
2808 return QCOW2_OL_SNAPSHOT_TABLE
;
2812 if ((chk
& QCOW2_OL_INACTIVE_L1
) && s
->snapshots
) {
2813 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
2814 if (s
->snapshots
[i
].l1_size
&&
2815 overlaps_with(s
->snapshots
[i
].l1_table_offset
,
2816 s
->snapshots
[i
].l1_size
* L1E_SIZE
)) {
2817 return QCOW2_OL_INACTIVE_L1
;
2822 if ((chk
& QCOW2_OL_ACTIVE_L2
) && s
->l1_table
) {
2823 for (i
= 0; i
< s
->l1_size
; i
++) {
2824 if ((s
->l1_table
[i
] & L1E_OFFSET_MASK
) &&
2825 overlaps_with(s
->l1_table
[i
] & L1E_OFFSET_MASK
,
2827 return QCOW2_OL_ACTIVE_L2
;
2832 if ((chk
& QCOW2_OL_REFCOUNT_BLOCK
) && s
->refcount_table
) {
2833 unsigned last_entry
= s
->max_refcount_table_index
;
2834 assert(last_entry
< s
->refcount_table_size
);
2835 assert(last_entry
+ 1 == s
->refcount_table_size
||
2836 (s
->refcount_table
[last_entry
+ 1] & REFT_OFFSET_MASK
) == 0);
2837 for (i
= 0; i
<= last_entry
; i
++) {
2838 if ((s
->refcount_table
[i
] & REFT_OFFSET_MASK
) &&
2839 overlaps_with(s
->refcount_table
[i
] & REFT_OFFSET_MASK
,
2841 return QCOW2_OL_REFCOUNT_BLOCK
;
2846 if ((chk
& QCOW2_OL_INACTIVE_L2
) && s
->snapshots
) {
2847 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
2848 uint64_t l1_ofs
= s
->snapshots
[i
].l1_table_offset
;
2849 uint32_t l1_sz
= s
->snapshots
[i
].l1_size
;
2850 uint64_t l1_sz2
= l1_sz
* L1E_SIZE
;
2854 ret
= qcow2_validate_table(bs
, l1_ofs
, l1_sz
, L1E_SIZE
,
2855 QCOW_MAX_L1_SIZE
, "", NULL
);
2860 l1
= g_try_malloc(l1_sz2
);
2862 if (l1_sz2
&& l1
== NULL
) {
2866 ret
= bdrv_pread(bs
->file
, l1_ofs
, l1
, l1_sz2
);
2872 for (j
= 0; j
< l1_sz
; j
++) {
2873 uint64_t l2_ofs
= be64_to_cpu(l1
[j
]) & L1E_OFFSET_MASK
;
2874 if (l2_ofs
&& overlaps_with(l2_ofs
, s
->cluster_size
)) {
2876 return QCOW2_OL_INACTIVE_L2
;
2884 if ((chk
& QCOW2_OL_BITMAP_DIRECTORY
) &&
2885 (s
->autoclear_features
& QCOW2_AUTOCLEAR_BITMAPS
))
2887 if (overlaps_with(s
->bitmap_directory_offset
,
2888 s
->bitmap_directory_size
))
2890 return QCOW2_OL_BITMAP_DIRECTORY
;
2897 static const char *metadata_ol_names
[] = {
2898 [QCOW2_OL_MAIN_HEADER_BITNR
] = "qcow2_header",
2899 [QCOW2_OL_ACTIVE_L1_BITNR
] = "active L1 table",
2900 [QCOW2_OL_ACTIVE_L2_BITNR
] = "active L2 table",
2901 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = "refcount table",
2902 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = "refcount block",
2903 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = "snapshot table",
2904 [QCOW2_OL_INACTIVE_L1_BITNR
] = "inactive L1 table",
2905 [QCOW2_OL_INACTIVE_L2_BITNR
] = "inactive L2 table",
2906 [QCOW2_OL_BITMAP_DIRECTORY_BITNR
] = "bitmap directory",
2908 QEMU_BUILD_BUG_ON(QCOW2_OL_MAX_BITNR
!= ARRAY_SIZE(metadata_ol_names
));
2911 * First performs a check for metadata overlaps (through
2912 * qcow2_check_metadata_overlap); if that fails with a negative value (error
2913 * while performing a check), that value is returned. If an impending overlap
2914 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
2915 * and -EIO returned.
2917 * Returns 0 if there were neither overlaps nor errors while checking for
2918 * overlaps; or a negative value (-errno) on error.
2920 int qcow2_pre_write_overlap_check(BlockDriverState
*bs
, int ign
, int64_t offset
,
2921 int64_t size
, bool data_file
)
2925 if (data_file
&& has_data_file(bs
)) {
2929 ret
= qcow2_check_metadata_overlap(bs
, ign
, offset
, size
);
2932 } else if (ret
> 0) {
2933 int metadata_ol_bitnr
= ctz32(ret
);
2934 assert(metadata_ol_bitnr
< QCOW2_OL_MAX_BITNR
);
2936 qcow2_signal_corruption(bs
, true, offset
, size
, "Preventing invalid "
2937 "write on metadata (overlaps with %s)",
2938 metadata_ol_names
[metadata_ol_bitnr
]);
2945 /* A pointer to a function of this type is given to walk_over_reftable(). That
2946 * function will create refblocks and pass them to a RefblockFinishOp once they
2947 * are completed (@refblock). @refblock_empty is set if the refblock is
2950 * Along with the refblock, a corresponding reftable entry is passed, in the
2951 * reftable @reftable (which may be reallocated) at @reftable_index.
2953 * @allocated should be set to true if a new cluster has been allocated.
2955 typedef int (RefblockFinishOp
)(BlockDriverState
*bs
, uint64_t **reftable
,
2956 uint64_t reftable_index
, uint64_t *reftable_size
,
2957 void *refblock
, bool refblock_empty
,
2958 bool *allocated
, Error
**errp
);
2961 * This "operation" for walk_over_reftable() allocates the refblock on disk (if
2962 * it is not empty) and inserts its offset into the new reftable. The size of
2963 * this new reftable is increased as required.
2965 static int alloc_refblock(BlockDriverState
*bs
, uint64_t **reftable
,
2966 uint64_t reftable_index
, uint64_t *reftable_size
,
2967 void *refblock
, bool refblock_empty
, bool *allocated
,
2970 BDRVQcow2State
*s
= bs
->opaque
;
2973 if (!refblock_empty
&& reftable_index
>= *reftable_size
) {
2974 uint64_t *new_reftable
;
2975 uint64_t new_reftable_size
;
2977 new_reftable_size
= ROUND_UP(reftable_index
+ 1,
2978 s
->cluster_size
/ REFTABLE_ENTRY_SIZE
);
2979 if (new_reftable_size
> QCOW_MAX_REFTABLE_SIZE
/ REFTABLE_ENTRY_SIZE
) {
2981 "This operation would make the refcount table grow "
2982 "beyond the maximum size supported by QEMU, aborting");
2986 new_reftable
= g_try_realloc(*reftable
, new_reftable_size
*
2987 REFTABLE_ENTRY_SIZE
);
2988 if (!new_reftable
) {
2989 error_setg(errp
, "Failed to increase reftable buffer size");
2993 memset(new_reftable
+ *reftable_size
, 0,
2994 (new_reftable_size
- *reftable_size
) * REFTABLE_ENTRY_SIZE
);
2996 *reftable
= new_reftable
;
2997 *reftable_size
= new_reftable_size
;
3000 if (!refblock_empty
&& !(*reftable
)[reftable_index
]) {
3001 offset
= qcow2_alloc_clusters(bs
, s
->cluster_size
);
3003 error_setg_errno(errp
, -offset
, "Failed to allocate refblock");
3006 (*reftable
)[reftable_index
] = offset
;
3014 * This "operation" for walk_over_reftable() writes the refblock to disk at the
3015 * offset specified by the new reftable's entry. It does not modify the new
3016 * reftable or change any refcounts.
3018 static int flush_refblock(BlockDriverState
*bs
, uint64_t **reftable
,
3019 uint64_t reftable_index
, uint64_t *reftable_size
,
3020 void *refblock
, bool refblock_empty
, bool *allocated
,
3023 BDRVQcow2State
*s
= bs
->opaque
;
3027 if (reftable_index
< *reftable_size
&& (*reftable
)[reftable_index
]) {
3028 offset
= (*reftable
)[reftable_index
];
3030 ret
= qcow2_pre_write_overlap_check(bs
, 0, offset
, s
->cluster_size
,
3033 error_setg_errno(errp
, -ret
, "Overlap check failed");
3037 ret
= bdrv_pwrite(bs
->file
, offset
, refblock
, s
->cluster_size
);
3039 error_setg_errno(errp
, -ret
, "Failed to write refblock");
3043 assert(refblock_empty
);
3050 * This function walks over the existing reftable and every referenced refblock;
3051 * if @new_set_refcount is non-NULL, it is called for every refcount entry to
3052 * create an equal new entry in the passed @new_refblock. Once that
3053 * @new_refblock is completely filled, @operation will be called.
3055 * @status_cb and @cb_opaque are used for the amend operation's status callback.
3056 * @index is the index of the walk_over_reftable() calls and @total is the total
3057 * number of walk_over_reftable() calls per amend operation. Both are used for
3058 * calculating the parameters for the status callback.
3060 * @allocated is set to true if a new cluster has been allocated.
3062 static int walk_over_reftable(BlockDriverState
*bs
, uint64_t **new_reftable
,
3063 uint64_t *new_reftable_index
,
3064 uint64_t *new_reftable_size
,
3065 void *new_refblock
, int new_refblock_size
,
3066 int new_refcount_bits
,
3067 RefblockFinishOp
*operation
, bool *allocated
,
3068 Qcow2SetRefcountFunc
*new_set_refcount
,
3069 BlockDriverAmendStatusCB
*status_cb
,
3070 void *cb_opaque
, int index
, int total
,
3073 BDRVQcow2State
*s
= bs
->opaque
;
3074 uint64_t reftable_index
;
3075 bool new_refblock_empty
= true;
3077 int new_refblock_index
= 0;
3080 for (reftable_index
= 0; reftable_index
< s
->refcount_table_size
;
3083 uint64_t refblock_offset
= s
->refcount_table
[reftable_index
]
3086 status_cb(bs
, (uint64_t)index
* s
->refcount_table_size
+ reftable_index
,
3087 (uint64_t)total
* s
->refcount_table_size
, cb_opaque
);
3089 if (refblock_offset
) {
3092 if (offset_into_cluster(s
, refblock_offset
)) {
3093 qcow2_signal_corruption(bs
, true, -1, -1, "Refblock offset %#"
3094 PRIx64
" unaligned (reftable index: %#"
3095 PRIx64
")", refblock_offset
,
3098 "Image is corrupt (unaligned refblock offset)");
3102 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, refblock_offset
,
3105 error_setg_errno(errp
, -ret
, "Failed to retrieve refblock");
3109 for (refblock_index
= 0; refblock_index
< s
->refcount_block_size
;
3114 if (new_refblock_index
>= new_refblock_size
) {
3115 /* new_refblock is now complete */
3116 ret
= operation(bs
, new_reftable
, *new_reftable_index
,
3117 new_reftable_size
, new_refblock
,
3118 new_refblock_empty
, allocated
, errp
);
3120 qcow2_cache_put(s
->refcount_block_cache
, &refblock
);
3124 (*new_reftable_index
)++;
3125 new_refblock_index
= 0;
3126 new_refblock_empty
= true;
3129 refcount
= s
->get_refcount(refblock
, refblock_index
);
3130 if (new_refcount_bits
< 64 && refcount
>> new_refcount_bits
) {
3133 qcow2_cache_put(s
->refcount_block_cache
, &refblock
);
3135 offset
= ((reftable_index
<< s
->refcount_block_bits
)
3136 + refblock_index
) << s
->cluster_bits
;
3138 error_setg(errp
, "Cannot decrease refcount entry width to "
3139 "%i bits: Cluster at offset %#" PRIx64
" has a "
3140 "refcount of %" PRIu64
, new_refcount_bits
,
3145 if (new_set_refcount
) {
3146 new_set_refcount(new_refblock
, new_refblock_index
++,
3149 new_refblock_index
++;
3151 new_refblock_empty
= new_refblock_empty
&& refcount
== 0;
3154 qcow2_cache_put(s
->refcount_block_cache
, &refblock
);
3156 /* No refblock means every refcount is 0 */
3157 for (refblock_index
= 0; refblock_index
< s
->refcount_block_size
;
3160 if (new_refblock_index
>= new_refblock_size
) {
3161 /* new_refblock is now complete */
3162 ret
= operation(bs
, new_reftable
, *new_reftable_index
,
3163 new_reftable_size
, new_refblock
,
3164 new_refblock_empty
, allocated
, errp
);
3169 (*new_reftable_index
)++;
3170 new_refblock_index
= 0;
3171 new_refblock_empty
= true;
3174 if (new_set_refcount
) {
3175 new_set_refcount(new_refblock
, new_refblock_index
++, 0);
3177 new_refblock_index
++;
3183 if (new_refblock_index
> 0) {
3184 /* Complete the potentially existing partially filled final refblock */
3185 if (new_set_refcount
) {
3186 for (; new_refblock_index
< new_refblock_size
;
3187 new_refblock_index
++)
3189 new_set_refcount(new_refblock
, new_refblock_index
, 0);
3193 ret
= operation(bs
, new_reftable
, *new_reftable_index
,
3194 new_reftable_size
, new_refblock
, new_refblock_empty
,
3200 (*new_reftable_index
)++;
3203 status_cb(bs
, (uint64_t)(index
+ 1) * s
->refcount_table_size
,
3204 (uint64_t)total
* s
->refcount_table_size
, cb_opaque
);
3209 int qcow2_change_refcount_order(BlockDriverState
*bs
, int refcount_order
,
3210 BlockDriverAmendStatusCB
*status_cb
,
3211 void *cb_opaque
, Error
**errp
)
3213 BDRVQcow2State
*s
= bs
->opaque
;
3214 Qcow2GetRefcountFunc
*new_get_refcount
;
3215 Qcow2SetRefcountFunc
*new_set_refcount
;
3216 void *new_refblock
= qemu_blockalign(bs
->file
->bs
, s
->cluster_size
);
3217 uint64_t *new_reftable
= NULL
, new_reftable_size
= 0;
3218 uint64_t *old_reftable
, old_reftable_size
, old_reftable_offset
;
3219 uint64_t new_reftable_index
= 0;
3221 int64_t new_reftable_offset
= 0, allocated_reftable_size
= 0;
3222 int new_refblock_size
, new_refcount_bits
= 1 << refcount_order
;
3223 int old_refcount_order
;
3226 bool new_allocation
;
3228 assert(s
->qcow_version
>= 3);
3229 assert(refcount_order
>= 0 && refcount_order
<= 6);
3231 /* see qcow2_open() */
3232 new_refblock_size
= 1 << (s
->cluster_bits
- (refcount_order
- 3));
3234 new_get_refcount
= get_refcount_funcs
[refcount_order
];
3235 new_set_refcount
= set_refcount_funcs
[refcount_order
];
3241 new_allocation
= false;
3243 /* At least we have to do this walk and the one which writes the
3244 * refblocks; also, at least we have to do this loop here at least
3245 * twice (normally), first to do the allocations, and second to
3246 * determine that everything is correctly allocated, this then makes
3247 * three walks in total */
3248 total_walks
= MAX(walk_index
+ 2, 3);
3250 /* First, allocate the structures so they are present in the refcount
3252 ret
= walk_over_reftable(bs
, &new_reftable
, &new_reftable_index
,
3253 &new_reftable_size
, NULL
, new_refblock_size
,
3254 new_refcount_bits
, &alloc_refblock
,
3255 &new_allocation
, NULL
, status_cb
, cb_opaque
,
3256 walk_index
++, total_walks
, errp
);
3261 new_reftable_index
= 0;
3263 if (new_allocation
) {
3264 if (new_reftable_offset
) {
3265 qcow2_free_clusters(
3266 bs
, new_reftable_offset
,
3267 allocated_reftable_size
* REFTABLE_ENTRY_SIZE
,
3268 QCOW2_DISCARD_NEVER
);
3271 new_reftable_offset
= qcow2_alloc_clusters(bs
, new_reftable_size
*
3272 REFTABLE_ENTRY_SIZE
);
3273 if (new_reftable_offset
< 0) {
3274 error_setg_errno(errp
, -new_reftable_offset
,
3275 "Failed to allocate the new reftable");
3276 ret
= new_reftable_offset
;
3279 allocated_reftable_size
= new_reftable_size
;
3281 } while (new_allocation
);
3283 /* Second, write the new refblocks */
3284 ret
= walk_over_reftable(bs
, &new_reftable
, &new_reftable_index
,
3285 &new_reftable_size
, new_refblock
,
3286 new_refblock_size
, new_refcount_bits
,
3287 &flush_refblock
, &new_allocation
, new_set_refcount
,
3288 status_cb
, cb_opaque
, walk_index
, walk_index
+ 1,
3293 assert(!new_allocation
);
3296 /* Write the new reftable */
3297 ret
= qcow2_pre_write_overlap_check(bs
, 0, new_reftable_offset
,
3298 new_reftable_size
* REFTABLE_ENTRY_SIZE
,
3301 error_setg_errno(errp
, -ret
, "Overlap check failed");
3305 for (i
= 0; i
< new_reftable_size
; i
++) {
3306 cpu_to_be64s(&new_reftable
[i
]);
3309 ret
= bdrv_pwrite(bs
->file
, new_reftable_offset
, new_reftable
,
3310 new_reftable_size
* REFTABLE_ENTRY_SIZE
);
3312 for (i
= 0; i
< new_reftable_size
; i
++) {
3313 be64_to_cpus(&new_reftable
[i
]);
3317 error_setg_errno(errp
, -ret
, "Failed to write the new reftable");
3322 /* Empty the refcount cache */
3323 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
3325 error_setg_errno(errp
, -ret
, "Failed to flush the refblock cache");
3329 /* Update the image header to point to the new reftable; this only updates
3330 * the fields which are relevant to qcow2_update_header(); other fields
3331 * such as s->refcount_table or s->refcount_bits stay stale for now
3332 * (because we have to restore everything if qcow2_update_header() fails) */
3333 old_refcount_order
= s
->refcount_order
;
3334 old_reftable_size
= s
->refcount_table_size
;
3335 old_reftable_offset
= s
->refcount_table_offset
;
3337 s
->refcount_order
= refcount_order
;
3338 s
->refcount_table_size
= new_reftable_size
;
3339 s
->refcount_table_offset
= new_reftable_offset
;
3341 ret
= qcow2_update_header(bs
);
3343 s
->refcount_order
= old_refcount_order
;
3344 s
->refcount_table_size
= old_reftable_size
;
3345 s
->refcount_table_offset
= old_reftable_offset
;
3346 error_setg_errno(errp
, -ret
, "Failed to update the qcow2 header");
3350 /* Now update the rest of the in-memory information */
3351 old_reftable
= s
->refcount_table
;
3352 s
->refcount_table
= new_reftable
;
3353 update_max_refcount_table_index(s
);
3355 s
->refcount_bits
= 1 << refcount_order
;
3356 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
3357 s
->refcount_max
+= s
->refcount_max
- 1;
3359 s
->refcount_block_bits
= s
->cluster_bits
- (refcount_order
- 3);
3360 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
3362 s
->get_refcount
= new_get_refcount
;
3363 s
->set_refcount
= new_set_refcount
;
3365 /* For cleaning up all old refblocks and the old reftable below the "done"
3367 new_reftable
= old_reftable
;
3368 new_reftable_size
= old_reftable_size
;
3369 new_reftable_offset
= old_reftable_offset
;
3373 /* On success, new_reftable actually points to the old reftable (and
3374 * new_reftable_size is the old reftable's size); but that is just
3376 for (i
= 0; i
< new_reftable_size
; i
++) {
3377 uint64_t offset
= new_reftable
[i
] & REFT_OFFSET_MASK
;
3379 qcow2_free_clusters(bs
, offset
, s
->cluster_size
,
3380 QCOW2_DISCARD_OTHER
);
3383 g_free(new_reftable
);
3385 if (new_reftable_offset
> 0) {
3386 qcow2_free_clusters(bs
, new_reftable_offset
,
3387 new_reftable_size
* REFTABLE_ENTRY_SIZE
,
3388 QCOW2_DISCARD_OTHER
);
3392 qemu_vfree(new_refblock
);
3396 static int64_t get_refblock_offset(BlockDriverState
*bs
, uint64_t offset
)
3398 BDRVQcow2State
*s
= bs
->opaque
;
3399 uint32_t index
= offset_to_reftable_index(s
, offset
);
3400 int64_t covering_refblock_offset
= 0;
3402 if (index
< s
->refcount_table_size
) {
3403 covering_refblock_offset
= s
->refcount_table
[index
] & REFT_OFFSET_MASK
;
3405 if (!covering_refblock_offset
) {
3406 qcow2_signal_corruption(bs
, true, -1, -1, "Refblock at %#" PRIx64
" is "
3407 "not covered by the refcount structures",
3412 return covering_refblock_offset
;
3415 static int qcow2_discard_refcount_block(BlockDriverState
*bs
,
3416 uint64_t discard_block_offs
)
3418 BDRVQcow2State
*s
= bs
->opaque
;
3419 int64_t refblock_offs
;
3420 uint64_t cluster_index
= discard_block_offs
>> s
->cluster_bits
;
3421 uint32_t block_index
= cluster_index
& (s
->refcount_block_size
- 1);
3425 refblock_offs
= get_refblock_offset(bs
, discard_block_offs
);
3426 if (refblock_offs
< 0) {
3427 return refblock_offs
;
3430 assert(discard_block_offs
!= 0);
3432 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, refblock_offs
,
3438 if (s
->get_refcount(refblock
, block_index
) != 1) {
3439 qcow2_signal_corruption(bs
, true, -1, -1, "Invalid refcount:"
3440 " refblock offset %#" PRIx64
3441 ", reftable index %u"
3442 ", block offset %#" PRIx64
3443 ", refcount %#" PRIx64
,
3445 offset_to_reftable_index(s
, discard_block_offs
),
3447 s
->get_refcount(refblock
, block_index
));
3448 qcow2_cache_put(s
->refcount_block_cache
, &refblock
);
3451 s
->set_refcount(refblock
, block_index
, 0);
3453 qcow2_cache_entry_mark_dirty(s
->refcount_block_cache
, refblock
);
3455 qcow2_cache_put(s
->refcount_block_cache
, &refblock
);
3457 if (cluster_index
< s
->free_cluster_index
) {
3458 s
->free_cluster_index
= cluster_index
;
3461 refblock
= qcow2_cache_is_table_offset(s
->refcount_block_cache
,
3462 discard_block_offs
);
3464 /* discard refblock from the cache if refblock is cached */
3465 qcow2_cache_discard(s
->refcount_block_cache
, refblock
);
3467 update_refcount_discard(bs
, discard_block_offs
, s
->cluster_size
);
3472 int qcow2_shrink_reftable(BlockDriverState
*bs
)
3474 BDRVQcow2State
*s
= bs
->opaque
;
3475 uint64_t *reftable_tmp
=
3476 g_malloc(s
->refcount_table_size
* REFTABLE_ENTRY_SIZE
);
3479 for (i
= 0; i
< s
->refcount_table_size
; i
++) {
3480 int64_t refblock_offs
= s
->refcount_table
[i
] & REFT_OFFSET_MASK
;
3484 if (refblock_offs
== 0) {
3485 reftable_tmp
[i
] = 0;
3488 ret
= qcow2_cache_get(bs
, s
->refcount_block_cache
, refblock_offs
,
3494 /* the refblock has own reference */
3495 if (i
== offset_to_reftable_index(s
, refblock_offs
)) {
3496 uint64_t block_index
= (refblock_offs
>> s
->cluster_bits
) &
3497 (s
->refcount_block_size
- 1);
3498 uint64_t refcount
= s
->get_refcount(refblock
, block_index
);
3500 s
->set_refcount(refblock
, block_index
, 0);
3502 unused_block
= buffer_is_zero(refblock
, s
->cluster_size
);
3504 s
->set_refcount(refblock
, block_index
, refcount
);
3506 unused_block
= buffer_is_zero(refblock
, s
->cluster_size
);
3508 qcow2_cache_put(s
->refcount_block_cache
, &refblock
);
3510 reftable_tmp
[i
] = unused_block
? 0 : cpu_to_be64(s
->refcount_table
[i
]);
3513 ret
= bdrv_pwrite_sync(bs
->file
, s
->refcount_table_offset
, reftable_tmp
,
3514 s
->refcount_table_size
* REFTABLE_ENTRY_SIZE
);
3516 * If the write in the reftable failed the image may contain a partially
3517 * overwritten reftable. In this case it would be better to clear the
3518 * reftable in memory to avoid possible image corruption.
3520 for (i
= 0; i
< s
->refcount_table_size
; i
++) {
3521 if (s
->refcount_table
[i
] && !reftable_tmp
[i
]) {
3523 ret
= qcow2_discard_refcount_block(bs
, s
->refcount_table
[i
] &
3526 s
->refcount_table
[i
] = 0;
3530 if (!s
->cache_discards
) {
3531 qcow2_process_discards(bs
, ret
);
3535 g_free(reftable_tmp
);
3539 int64_t qcow2_get_last_cluster(BlockDriverState
*bs
, int64_t size
)
3541 BDRVQcow2State
*s
= bs
->opaque
;
3544 for (i
= size_to_clusters(s
, size
) - 1; i
>= 0; i
--) {
3546 int ret
= qcow2_get_refcount(bs
, i
, &refcount
);
3548 fprintf(stderr
, "Can't get refcount for cluster %" PRId64
": %s\n",
3556 qcow2_signal_corruption(bs
, true, -1, -1,
3557 "There are no references in the refcount table.");
3561 int qcow2_detect_metadata_preallocation(BlockDriverState
*bs
)
3563 BDRVQcow2State
*s
= bs
->opaque
;
3564 int64_t i
, end_cluster
, cluster_count
= 0, threshold
;
3565 int64_t file_length
, real_allocation
, real_clusters
;
3567 qemu_co_mutex_assert_locked(&s
->lock
);
3569 file_length
= bdrv_getlength(bs
->file
->bs
);
3570 if (file_length
< 0) {
3574 real_allocation
= bdrv_get_allocated_file_size(bs
->file
->bs
);
3575 if (real_allocation
< 0) {
3576 return real_allocation
;
3579 real_clusters
= real_allocation
/ s
->cluster_size
;
3580 threshold
= MAX(real_clusters
* 10 / 9, real_clusters
+ 2);
3582 end_cluster
= size_to_clusters(s
, file_length
);
3583 for (i
= 0; i
< end_cluster
&& cluster_count
< threshold
; i
++) {
3585 int ret
= qcow2_get_refcount(bs
, i
, &refcount
);
3589 cluster_count
+= !!refcount
;
3592 return cluster_count
>= threshold
;