ivshmem: Improve MSI irqfd error handling
[qemu/ar7.git] / block / qcow2-refcount.c
blob92701ab7af936a7ef5da7131c993c599380303c0
1 /*
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
22 * THE SOFTWARE.
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu-common.h"
28 #include "block/block_int.h"
29 #include "block/qcow2.h"
30 #include "qemu/range.h"
31 #include "qemu/bswap.h"
32 #include "qemu/cutils.h"
34 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,
48 uint64_t value);
49 static void set_refcount_ro1(void *refcount_array, uint64_t index,
50 uint64_t value);
51 static void set_refcount_ro2(void *refcount_array, uint64_t index,
52 uint64_t value);
53 static void set_refcount_ro3(void *refcount_array, uint64_t index,
54 uint64_t value);
55 static void set_refcount_ro4(void *refcount_array, uint64_t index,
56 uint64_t value);
57 static void set_refcount_ro5(void *refcount_array, uint64_t index,
58 uint64_t value);
59 static void set_refcount_ro6(void *refcount_array, uint64_t index,
60 uint64_t value);
63 static Qcow2GetRefcountFunc *const get_refcount_funcs[] = {
64 &get_refcount_ro0,
65 &get_refcount_ro1,
66 &get_refcount_ro2,
67 &get_refcount_ro3,
68 &get_refcount_ro4,
69 &get_refcount_ro5,
70 &get_refcount_ro6
73 static Qcow2SetRefcountFunc *const set_refcount_funcs[] = {
74 &set_refcount_ro0,
75 &set_refcount_ro1,
76 &set_refcount_ro2,
77 &set_refcount_ro3,
78 &set_refcount_ro4,
79 &set_refcount_ro5,
80 &set_refcount_ro6
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) {
91 i--;
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;
101 int ret;
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 / sizeof(uint64_t));
109 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
110 s->refcount_table = g_try_malloc(refcount_table_size2);
112 if (s->refcount_table_size > 0) {
113 if (s->refcount_table == NULL) {
114 ret = -ENOMEM;
115 goto fail;
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);
120 if (ret < 0) {
121 goto fail;
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);
127 return 0;
128 fail:
129 return ret;
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,
145 uint64_t value)
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)))
155 & 0x3;
158 static void set_refcount_ro1(void *refcount_array, uint64_t index,
159 uint64_t value)
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)))
169 & 0xf;
172 static void set_refcount_ro2(void *refcount_array, uint64_t index,
173 uint64_t value)
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,
186 uint64_t value)
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,
198 uint64_t value)
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,
210 uint64_t value)
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,
222 uint64_t value)
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,
236 refcount_block);
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,
244 uint64_t *refcount)
246 BDRVQcow2State *s = bs->opaque;
247 uint64_t refcount_table_index, block_index;
248 int64_t refcount_block_offset;
249 int ret;
250 void *refcount_block;
252 refcount_table_index = cluster_index >> s->refcount_block_bits;
253 if (refcount_table_index >= s->refcount_table_size) {
254 *refcount = 0;
255 return 0;
257 refcount_block_offset =
258 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
259 if (!refcount_block_offset) {
260 *refcount = 0;
261 return 0;
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);
268 return -EIO;
271 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
272 &refcount_block);
273 if (ret < 0) {
274 return ret;
277 block_index = cluster_index & (s->refcount_block_size - 1);
278 *refcount = s->get_refcount(refcount_block, block_index);
280 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
282 return 0;
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,
287 uint64_t offset_b)
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;
306 int64_t ret;
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);
325 return -EIO;
328 return load_refcount_block(bs, refcount_block_offset,
329 refcount_block);
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);
360 if (ret < 0) {
361 return ret;
364 /* Allocate the refcount block itself and mark it as used */
365 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
366 if (new_block < 0) {
367 return new_block;
370 /* If we're allocating the block at offset 0 then something is wrong */
371 if (new_block == 0) {
372 qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid "
373 "allocation of refcount block at offset 0");
374 return -EIO;
377 #ifdef DEBUG_ALLOC2
378 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
379 " at %" PRIx64 "\n",
380 refcount_table_index, cluster_index << s->cluster_bits, new_block);
381 #endif
383 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
384 /* Zero the new refcount block before updating it */
385 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
386 refcount_block);
387 if (ret < 0) {
388 goto fail;
391 memset(*refcount_block, 0, s->cluster_size);
393 /* The block describes itself, need to update the cache */
394 int block_index = (new_block >> s->cluster_bits) &
395 (s->refcount_block_size - 1);
396 s->set_refcount(*refcount_block, block_index, 1);
397 } else {
398 /* Described somewhere else. This can recurse at most twice before we
399 * arrive at a block that describes itself. */
400 ret = update_refcount(bs, new_block, s->cluster_size, 1, false,
401 QCOW2_DISCARD_NEVER);
402 if (ret < 0) {
403 goto fail;
406 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
407 if (ret < 0) {
408 goto fail;
411 /* Initialize the new refcount block only after updating its refcount,
412 * update_refcount uses the refcount cache itself */
413 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
414 refcount_block);
415 if (ret < 0) {
416 goto fail;
419 memset(*refcount_block, 0, s->cluster_size);
422 /* Now the new refcount block needs to be written to disk */
423 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
424 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache, *refcount_block);
425 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
426 if (ret < 0) {
427 goto fail;
430 /* If the refcount table is big enough, just hook the block up there */
431 if (refcount_table_index < s->refcount_table_size) {
432 uint64_t data64 = cpu_to_be64(new_block);
433 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
434 ret = bdrv_pwrite_sync(bs->file,
435 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
436 &data64, sizeof(data64));
437 if (ret < 0) {
438 goto fail;
441 s->refcount_table[refcount_table_index] = new_block;
442 /* If there's a hole in s->refcount_table then it can happen
443 * that refcount_table_index < s->max_refcount_table_index */
444 s->max_refcount_table_index =
445 MAX(s->max_refcount_table_index, refcount_table_index);
447 /* The new refcount block may be where the caller intended to put its
448 * data, so let it restart the search. */
449 return -EAGAIN;
452 qcow2_cache_put(bs, s->refcount_block_cache, refcount_block);
455 * If we come here, we need to grow the refcount table. Again, a new
456 * refcount table needs some space and we can't simply allocate to avoid
457 * endless recursion.
459 * Therefore let's grab new refcount blocks at the end of the image, which
460 * will describe themselves and the new refcount table. This way we can
461 * reference them only in the new table and do the switch to the new
462 * refcount table at once without producing an inconsistent state in
463 * between.
465 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
467 /* Calculate the number of refcount blocks needed so far; this will be the
468 * basis for calculating the index of the first cluster used for the
469 * self-describing refcount structures which we are about to create.
471 * Because we reached this point, there cannot be any refcount entries for
472 * cluster_index or higher indices yet. However, because new_block has been
473 * allocated to describe that cluster (and it will assume this role later
474 * on), we cannot use that index; also, new_block may actually have a higher
475 * cluster index than cluster_index, so it needs to be taken into account
476 * here (and 1 needs to be added to its value because that cluster is used).
478 uint64_t blocks_used = DIV_ROUND_UP(MAX(cluster_index + 1,
479 (new_block >> s->cluster_bits) + 1),
480 s->refcount_block_size);
482 /* Create the new refcount table and blocks */
483 uint64_t meta_offset = (blocks_used * s->refcount_block_size) *
484 s->cluster_size;
486 ret = qcow2_refcount_area(bs, meta_offset, 0, false,
487 refcount_table_index, new_block);
488 if (ret < 0) {
489 return ret;
492 ret = load_refcount_block(bs, new_block, refcount_block);
493 if (ret < 0) {
494 return ret;
497 /* If we were trying to do the initial refcount update for some cluster
498 * allocation, we might have used the same clusters to store newly
499 * allocated metadata. Make the caller search some new space. */
500 return -EAGAIN;
502 fail:
503 if (*refcount_block != NULL) {
504 qcow2_cache_put(bs, s->refcount_block_cache, refcount_block);
506 return ret;
510 * Starting at @start_offset, this function creates new self-covering refcount
511 * structures: A new refcount table and refcount blocks which cover all of
512 * themselves, and a number of @additional_clusters beyond their end.
513 * @start_offset must be at the end of the image file, that is, there must be
514 * only empty space beyond it.
515 * If @exact_size is false, the refcount table will have 50 % more entries than
516 * necessary so it will not need to grow again soon.
517 * If @new_refblock_offset is not zero, it contains the offset of a refcount
518 * block that should be entered into the new refcount table at index
519 * @new_refblock_index.
521 * Returns: The offset after the new refcount structures (i.e. where the
522 * @additional_clusters may be placed) on success, -errno on error.
524 int64_t qcow2_refcount_area(BlockDriverState *bs, uint64_t start_offset,
525 uint64_t additional_clusters, bool exact_size,
526 int new_refblock_index,
527 uint64_t new_refblock_offset)
529 BDRVQcow2State *s = bs->opaque;
530 uint64_t total_refblock_count_u64, additional_refblock_count;
531 int total_refblock_count, table_size, area_reftable_index, table_clusters;
532 int i;
533 uint64_t table_offset, block_offset, end_offset;
534 int ret;
535 uint64_t *new_table;
537 assert(!(start_offset % s->cluster_size));
539 qcow2_refcount_metadata_size(start_offset / s->cluster_size +
540 additional_clusters,
541 s->cluster_size, s->refcount_order,
542 !exact_size, &total_refblock_count_u64);
543 if (total_refblock_count_u64 > QCOW_MAX_REFTABLE_SIZE) {
544 return -EFBIG;
546 total_refblock_count = total_refblock_count_u64;
548 /* Index in the refcount table of the first refcount block to cover the area
549 * of refcount structures we are about to create; we know that
550 * @total_refblock_count can cover @start_offset, so this will definitely
551 * fit into an int. */
552 area_reftable_index = (start_offset / s->cluster_size) /
553 s->refcount_block_size;
555 if (exact_size) {
556 table_size = total_refblock_count;
557 } else {
558 table_size = total_refblock_count +
559 DIV_ROUND_UP(total_refblock_count, 2);
561 /* The qcow2 file can only store the reftable size in number of clusters */
562 table_size = ROUND_UP(table_size, s->cluster_size / sizeof(uint64_t));
563 table_clusters = (table_size * sizeof(uint64_t)) / s->cluster_size;
565 if (table_size > QCOW_MAX_REFTABLE_SIZE) {
566 return -EFBIG;
569 new_table = g_try_new0(uint64_t, table_size);
571 assert(table_size > 0);
572 if (new_table == NULL) {
573 ret = -ENOMEM;
574 goto fail;
577 /* Fill the new refcount table */
578 if (table_size > s->max_refcount_table_index) {
579 /* We're actually growing the reftable */
580 memcpy(new_table, s->refcount_table,
581 (s->max_refcount_table_index + 1) * sizeof(uint64_t));
582 } else {
583 /* Improbable case: We're shrinking the reftable. However, the caller
584 * has assured us that there is only empty space beyond @start_offset,
585 * so we can simply drop all of the refblocks that won't fit into the
586 * new reftable. */
587 memcpy(new_table, s->refcount_table, table_size * sizeof(uint64_t));
590 if (new_refblock_offset) {
591 assert(new_refblock_index < total_refblock_count);
592 new_table[new_refblock_index] = new_refblock_offset;
595 /* Count how many new refblocks we have to create */
596 additional_refblock_count = 0;
597 for (i = area_reftable_index; i < total_refblock_count; i++) {
598 if (!new_table[i]) {
599 additional_refblock_count++;
603 table_offset = start_offset + additional_refblock_count * s->cluster_size;
604 end_offset = table_offset + table_clusters * s->cluster_size;
606 /* Fill the refcount blocks, and create new ones, if necessary */
607 block_offset = start_offset;
608 for (i = area_reftable_index; i < total_refblock_count; i++) {
609 void *refblock_data;
610 uint64_t first_offset_covered;
612 /* Reuse an existing refblock if possible, create a new one otherwise */
613 if (new_table[i]) {
614 ret = qcow2_cache_get(bs, s->refcount_block_cache, new_table[i],
615 &refblock_data);
616 if (ret < 0) {
617 goto fail;
619 } else {
620 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache,
621 block_offset, &refblock_data);
622 if (ret < 0) {
623 goto fail;
625 memset(refblock_data, 0, s->cluster_size);
626 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache,
627 refblock_data);
629 new_table[i] = block_offset;
630 block_offset += s->cluster_size;
633 /* First host offset covered by this refblock */
634 first_offset_covered = (uint64_t)i * s->refcount_block_size *
635 s->cluster_size;
636 if (first_offset_covered < end_offset) {
637 int j, end_index;
639 /* Set the refcount of all of the new refcount structures to 1 */
641 if (first_offset_covered < start_offset) {
642 assert(i == area_reftable_index);
643 j = (start_offset - first_offset_covered) / s->cluster_size;
644 assert(j < s->refcount_block_size);
645 } else {
646 j = 0;
649 end_index = MIN((end_offset - first_offset_covered) /
650 s->cluster_size,
651 s->refcount_block_size);
653 for (; j < end_index; j++) {
654 /* The caller guaranteed us this space would be empty */
655 assert(s->get_refcount(refblock_data, j) == 0);
656 s->set_refcount(refblock_data, j, 1);
659 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache,
660 refblock_data);
663 qcow2_cache_put(bs, s->refcount_block_cache, &refblock_data);
666 assert(block_offset == table_offset);
668 /* Write refcount blocks to disk */
669 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
670 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
671 if (ret < 0) {
672 goto fail;
675 /* Write refcount table to disk */
676 for (i = 0; i < total_refblock_count; i++) {
677 cpu_to_be64s(&new_table[i]);
680 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
681 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
682 table_size * sizeof(uint64_t));
683 if (ret < 0) {
684 goto fail;
687 for (i = 0; i < total_refblock_count; i++) {
688 be64_to_cpus(&new_table[i]);
691 /* Hook up the new refcount table in the qcow2 header */
692 struct QEMU_PACKED {
693 uint64_t d64;
694 uint32_t d32;
695 } data;
696 data.d64 = cpu_to_be64(table_offset);
697 data.d32 = cpu_to_be32(table_clusters);
698 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
699 ret = bdrv_pwrite_sync(bs->file,
700 offsetof(QCowHeader, refcount_table_offset),
701 &data, sizeof(data));
702 if (ret < 0) {
703 goto fail;
706 /* And switch it in memory */
707 uint64_t old_table_offset = s->refcount_table_offset;
708 uint64_t old_table_size = s->refcount_table_size;
710 g_free(s->refcount_table);
711 s->refcount_table = new_table;
712 s->refcount_table_size = table_size;
713 s->refcount_table_offset = table_offset;
714 update_max_refcount_table_index(s);
716 /* Free old table. */
717 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t),
718 QCOW2_DISCARD_OTHER);
720 return end_offset;
722 fail:
723 g_free(new_table);
724 return ret;
727 void qcow2_process_discards(BlockDriverState *bs, int ret)
729 BDRVQcow2State *s = bs->opaque;
730 Qcow2DiscardRegion *d, *next;
732 QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) {
733 QTAILQ_REMOVE(&s->discards, d, next);
735 /* Discard is optional, ignore the return value */
736 if (ret >= 0) {
737 bdrv_pdiscard(bs->file->bs, d->offset, d->bytes);
740 g_free(d);
744 static void update_refcount_discard(BlockDriverState *bs,
745 uint64_t offset, uint64_t length)
747 BDRVQcow2State *s = bs->opaque;
748 Qcow2DiscardRegion *d, *p, *next;
750 QTAILQ_FOREACH(d, &s->discards, next) {
751 uint64_t new_start = MIN(offset, d->offset);
752 uint64_t new_end = MAX(offset + length, d->offset + d->bytes);
754 if (new_end - new_start <= length + d->bytes) {
755 /* There can't be any overlap, areas ending up here have no
756 * references any more and therefore shouldn't get freed another
757 * time. */
758 assert(d->bytes + length == new_end - new_start);
759 d->offset = new_start;
760 d->bytes = new_end - new_start;
761 goto found;
765 d = g_malloc(sizeof(*d));
766 *d = (Qcow2DiscardRegion) {
767 .bs = bs,
768 .offset = offset,
769 .bytes = length,
771 QTAILQ_INSERT_TAIL(&s->discards, d, next);
773 found:
774 /* Merge discard requests if they are adjacent now */
775 QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) {
776 if (p == d
777 || p->offset > d->offset + d->bytes
778 || d->offset > p->offset + p->bytes)
780 continue;
783 /* Still no overlap possible */
784 assert(p->offset == d->offset + d->bytes
785 || d->offset == p->offset + p->bytes);
787 QTAILQ_REMOVE(&s->discards, p, next);
788 d->offset = MIN(d->offset, p->offset);
789 d->bytes += p->bytes;
790 g_free(p);
794 /* XXX: cache several refcount block clusters ? */
795 /* @addend is the absolute value of the addend; if @decrease is set, @addend
796 * will be subtracted from the current refcount, otherwise it will be added */
797 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
798 int64_t offset,
799 int64_t length,
800 uint64_t addend,
801 bool decrease,
802 enum qcow2_discard_type type)
804 BDRVQcow2State *s = bs->opaque;
805 int64_t start, last, cluster_offset;
806 void *refcount_block = NULL;
807 int64_t old_table_index = -1;
808 int ret;
810 #ifdef DEBUG_ALLOC2
811 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64
812 " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "",
813 addend);
814 #endif
815 if (length < 0) {
816 return -EINVAL;
817 } else if (length == 0) {
818 return 0;
821 if (decrease) {
822 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
823 s->l2_table_cache);
826 start = start_of_cluster(s, offset);
827 last = start_of_cluster(s, offset + length - 1);
828 for(cluster_offset = start; cluster_offset <= last;
829 cluster_offset += s->cluster_size)
831 int block_index;
832 uint64_t refcount;
833 int64_t cluster_index = cluster_offset >> s->cluster_bits;
834 int64_t table_index = cluster_index >> s->refcount_block_bits;
836 /* Load the refcount block and allocate it if needed */
837 if (table_index != old_table_index) {
838 if (refcount_block) {
839 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
841 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
842 if (ret < 0) {
843 goto fail;
846 old_table_index = table_index;
848 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache,
849 refcount_block);
851 /* we can update the count and save it */
852 block_index = cluster_index & (s->refcount_block_size - 1);
854 refcount = s->get_refcount(refcount_block, block_index);
855 if (decrease ? (refcount - addend > refcount)
856 : (refcount + addend < refcount ||
857 refcount + addend > s->refcount_max))
859 ret = -EINVAL;
860 goto fail;
862 if (decrease) {
863 refcount -= addend;
864 } else {
865 refcount += addend;
867 if (refcount == 0 && cluster_index < s->free_cluster_index) {
868 s->free_cluster_index = cluster_index;
870 s->set_refcount(refcount_block, block_index, refcount);
872 if (refcount == 0) {
873 void *table;
875 table = qcow2_cache_is_table_offset(bs, s->refcount_block_cache,
876 offset);
877 if (table != NULL) {
878 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
879 qcow2_cache_discard(bs, s->refcount_block_cache, table);
882 table = qcow2_cache_is_table_offset(bs, s->l2_table_cache, offset);
883 if (table != NULL) {
884 qcow2_cache_discard(bs, s->l2_table_cache, table);
887 if (s->discard_passthrough[type]) {
888 update_refcount_discard(bs, cluster_offset, s->cluster_size);
893 ret = 0;
894 fail:
895 if (!s->cache_discards) {
896 qcow2_process_discards(bs, ret);
899 /* Write last changed block to disk */
900 if (refcount_block) {
901 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
905 * Try do undo any updates if an error is returned (This may succeed in
906 * some cases like ENOSPC for allocating a new refcount block)
908 if (ret < 0) {
909 int dummy;
910 dummy = update_refcount(bs, offset, cluster_offset - offset, addend,
911 !decrease, QCOW2_DISCARD_NEVER);
912 (void)dummy;
915 return ret;
919 * Increases or decreases the refcount of a given cluster.
921 * @addend is the absolute value of the addend; if @decrease is set, @addend
922 * will be subtracted from the current refcount, otherwise it will be added.
924 * On success 0 is returned; on failure -errno is returned.
926 int qcow2_update_cluster_refcount(BlockDriverState *bs,
927 int64_t cluster_index,
928 uint64_t addend, bool decrease,
929 enum qcow2_discard_type type)
931 BDRVQcow2State *s = bs->opaque;
932 int ret;
934 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend,
935 decrease, type);
936 if (ret < 0) {
937 return ret;
940 return 0;
945 /*********************************************************/
946 /* cluster allocation functions */
950 /* return < 0 if error */
951 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size)
953 BDRVQcow2State *s = bs->opaque;
954 uint64_t i, nb_clusters, refcount;
955 int ret;
957 /* We can't allocate clusters if they may still be queued for discard. */
958 if (s->cache_discards) {
959 qcow2_process_discards(bs, 0);
962 nb_clusters = size_to_clusters(s, size);
963 retry:
964 for(i = 0; i < nb_clusters; i++) {
965 uint64_t next_cluster_index = s->free_cluster_index++;
966 ret = qcow2_get_refcount(bs, next_cluster_index, &refcount);
968 if (ret < 0) {
969 return ret;
970 } else if (refcount != 0) {
971 goto retry;
975 /* Make sure that all offsets in the "allocated" range are representable
976 * in an int64_t */
977 if (s->free_cluster_index > 0 &&
978 s->free_cluster_index - 1 > (INT64_MAX >> s->cluster_bits))
980 return -EFBIG;
983 #ifdef DEBUG_ALLOC2
984 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
985 size,
986 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
987 #endif
988 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
991 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size)
993 int64_t offset;
994 int ret;
996 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
997 do {
998 offset = alloc_clusters_noref(bs, size);
999 if (offset < 0) {
1000 return offset;
1003 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
1004 } while (ret == -EAGAIN);
1006 if (ret < 0) {
1007 return ret;
1010 return offset;
1013 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
1014 int64_t nb_clusters)
1016 BDRVQcow2State *s = bs->opaque;
1017 uint64_t cluster_index, refcount;
1018 uint64_t i;
1019 int ret;
1021 assert(nb_clusters >= 0);
1022 if (nb_clusters == 0) {
1023 return 0;
1026 do {
1027 /* Check how many clusters there are free */
1028 cluster_index = offset >> s->cluster_bits;
1029 for(i = 0; i < nb_clusters; i++) {
1030 ret = qcow2_get_refcount(bs, cluster_index++, &refcount);
1031 if (ret < 0) {
1032 return ret;
1033 } else if (refcount != 0) {
1034 break;
1038 /* And then allocate them */
1039 ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false,
1040 QCOW2_DISCARD_NEVER);
1041 } while (ret == -EAGAIN);
1043 if (ret < 0) {
1044 return ret;
1047 return i;
1050 /* only used to allocate compressed sectors. We try to allocate
1051 contiguous sectors. size must be <= cluster_size */
1052 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
1054 BDRVQcow2State *s = bs->opaque;
1055 int64_t offset;
1056 size_t free_in_cluster;
1057 int ret;
1059 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
1060 assert(size > 0 && size <= s->cluster_size);
1061 assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset));
1063 offset = s->free_byte_offset;
1065 if (offset) {
1066 uint64_t refcount;
1067 ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount);
1068 if (ret < 0) {
1069 return ret;
1072 if (refcount == s->refcount_max) {
1073 offset = 0;
1077 free_in_cluster = s->cluster_size - offset_into_cluster(s, offset);
1078 do {
1079 if (!offset || free_in_cluster < size) {
1080 int64_t new_cluster = alloc_clusters_noref(bs, s->cluster_size);
1081 if (new_cluster < 0) {
1082 return new_cluster;
1085 if (new_cluster == 0) {
1086 qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid "
1087 "allocation of compressed cluster "
1088 "at offset 0");
1089 return -EIO;
1092 if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) {
1093 offset = new_cluster;
1094 free_in_cluster = s->cluster_size;
1095 } else {
1096 free_in_cluster += s->cluster_size;
1100 assert(offset);
1101 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
1102 if (ret < 0) {
1103 offset = 0;
1105 } while (ret == -EAGAIN);
1106 if (ret < 0) {
1107 return ret;
1110 /* The cluster refcount was incremented; refcount blocks must be flushed
1111 * before the caller's L2 table updates. */
1112 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
1114 s->free_byte_offset = offset + size;
1115 if (!offset_into_cluster(s, s->free_byte_offset)) {
1116 s->free_byte_offset = 0;
1119 return offset;
1122 void qcow2_free_clusters(BlockDriverState *bs,
1123 int64_t offset, int64_t size,
1124 enum qcow2_discard_type type)
1126 int ret;
1128 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
1129 ret = update_refcount(bs, offset, size, 1, true, type);
1130 if (ret < 0) {
1131 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
1132 /* TODO Remember the clusters to free them later and avoid leaking */
1137 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
1138 * normal cluster, compressed cluster, etc.)
1140 void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
1141 int nb_clusters, enum qcow2_discard_type type)
1143 BDRVQcow2State *s = bs->opaque;
1145 switch (qcow2_get_cluster_type(l2_entry)) {
1146 case QCOW2_CLUSTER_COMPRESSED:
1148 int nb_csectors;
1149 nb_csectors = ((l2_entry >> s->csize_shift) &
1150 s->csize_mask) + 1;
1151 qcow2_free_clusters(bs,
1152 (l2_entry & s->cluster_offset_mask) & ~511,
1153 nb_csectors * 512, type);
1155 break;
1156 case QCOW2_CLUSTER_NORMAL:
1157 case QCOW2_CLUSTER_ZERO_ALLOC:
1158 if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) {
1159 qcow2_signal_corruption(bs, false, -1, -1,
1160 "Cannot free unaligned cluster %#llx",
1161 l2_entry & L2E_OFFSET_MASK);
1162 } else {
1163 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
1164 nb_clusters << s->cluster_bits, type);
1166 break;
1167 case QCOW2_CLUSTER_ZERO_PLAIN:
1168 case QCOW2_CLUSTER_UNALLOCATED:
1169 break;
1170 default:
1171 abort();
1177 /*********************************************************/
1178 /* snapshots and image creation */
1182 /* update the refcounts of snapshots and the copied flag */
1183 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
1184 int64_t l1_table_offset, int l1_size, int addend)
1186 BDRVQcow2State *s = bs->opaque;
1187 uint64_t *l1_table, *l2_table, l2_offset, entry, l1_size2, refcount;
1188 bool l1_allocated = false;
1189 int64_t old_entry, old_l2_offset;
1190 int i, j, l1_modified = 0, nb_csectors;
1191 int ret;
1193 assert(addend >= -1 && addend <= 1);
1195 l2_table = NULL;
1196 l1_table = NULL;
1197 l1_size2 = l1_size * sizeof(uint64_t);
1199 s->cache_discards = true;
1201 /* WARNING: qcow2_snapshot_goto relies on this function not using the
1202 * l1_table_offset when it is the current s->l1_table_offset! Be careful
1203 * when changing this! */
1204 if (l1_table_offset != s->l1_table_offset) {
1205 l1_table = g_try_malloc0(align_offset(l1_size2, 512));
1206 if (l1_size2 && l1_table == NULL) {
1207 ret = -ENOMEM;
1208 goto fail;
1210 l1_allocated = true;
1212 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
1213 if (ret < 0) {
1214 goto fail;
1217 for (i = 0; i < l1_size; i++) {
1218 be64_to_cpus(&l1_table[i]);
1220 } else {
1221 assert(l1_size == s->l1_size);
1222 l1_table = s->l1_table;
1223 l1_allocated = false;
1226 for (i = 0; i < l1_size; i++) {
1227 l2_offset = l1_table[i];
1228 if (l2_offset) {
1229 old_l2_offset = l2_offset;
1230 l2_offset &= L1E_OFFSET_MASK;
1232 if (offset_into_cluster(s, l2_offset)) {
1233 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1234 PRIx64 " unaligned (L1 index: %#x)",
1235 l2_offset, i);
1236 ret = -EIO;
1237 goto fail;
1240 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1241 (void**) &l2_table);
1242 if (ret < 0) {
1243 goto fail;
1246 for (j = 0; j < s->l2_size; j++) {
1247 uint64_t cluster_index;
1248 uint64_t offset;
1250 entry = be64_to_cpu(l2_table[j]);
1251 old_entry = entry;
1252 entry &= ~QCOW_OFLAG_COPIED;
1253 offset = entry & L2E_OFFSET_MASK;
1255 switch (qcow2_get_cluster_type(entry)) {
1256 case QCOW2_CLUSTER_COMPRESSED:
1257 nb_csectors = ((entry >> s->csize_shift) &
1258 s->csize_mask) + 1;
1259 if (addend != 0) {
1260 ret = update_refcount(bs,
1261 (entry & s->cluster_offset_mask) & ~511,
1262 nb_csectors * 512, abs(addend), addend < 0,
1263 QCOW2_DISCARD_SNAPSHOT);
1264 if (ret < 0) {
1265 goto fail;
1268 /* compressed clusters are never modified */
1269 refcount = 2;
1270 break;
1272 case QCOW2_CLUSTER_NORMAL:
1273 case QCOW2_CLUSTER_ZERO_ALLOC:
1274 if (offset_into_cluster(s, offset)) {
1275 qcow2_signal_corruption(bs, true, -1, -1, "Cluster "
1276 "allocation offset %#" PRIx64
1277 " unaligned (L2 offset: %#"
1278 PRIx64 ", L2 index: %#x)",
1279 offset, l2_offset, j);
1280 ret = -EIO;
1281 goto fail;
1284 cluster_index = offset >> s->cluster_bits;
1285 assert(cluster_index);
1286 if (addend != 0) {
1287 ret = qcow2_update_cluster_refcount(bs,
1288 cluster_index, abs(addend), addend < 0,
1289 QCOW2_DISCARD_SNAPSHOT);
1290 if (ret < 0) {
1291 goto fail;
1295 ret = qcow2_get_refcount(bs, cluster_index, &refcount);
1296 if (ret < 0) {
1297 goto fail;
1299 break;
1301 case QCOW2_CLUSTER_ZERO_PLAIN:
1302 case QCOW2_CLUSTER_UNALLOCATED:
1303 refcount = 0;
1304 break;
1306 default:
1307 abort();
1310 if (refcount == 1) {
1311 entry |= QCOW_OFLAG_COPIED;
1313 if (entry != old_entry) {
1314 if (addend > 0) {
1315 qcow2_cache_set_dependency(bs, s->l2_table_cache,
1316 s->refcount_block_cache);
1318 l2_table[j] = cpu_to_be64(entry);
1319 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache,
1320 l2_table);
1324 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1326 if (addend != 0) {
1327 ret = qcow2_update_cluster_refcount(bs, l2_offset >>
1328 s->cluster_bits,
1329 abs(addend), addend < 0,
1330 QCOW2_DISCARD_SNAPSHOT);
1331 if (ret < 0) {
1332 goto fail;
1335 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1336 &refcount);
1337 if (ret < 0) {
1338 goto fail;
1339 } else if (refcount == 1) {
1340 l2_offset |= QCOW_OFLAG_COPIED;
1342 if (l2_offset != old_l2_offset) {
1343 l1_table[i] = l2_offset;
1344 l1_modified = 1;
1349 ret = bdrv_flush(bs);
1350 fail:
1351 if (l2_table) {
1352 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1355 s->cache_discards = false;
1356 qcow2_process_discards(bs, ret);
1358 /* Update L1 only if it isn't deleted anyway (addend = -1) */
1359 if (ret == 0 && addend >= 0 && l1_modified) {
1360 for (i = 0; i < l1_size; i++) {
1361 cpu_to_be64s(&l1_table[i]);
1364 ret = bdrv_pwrite_sync(bs->file, l1_table_offset,
1365 l1_table, l1_size2);
1367 for (i = 0; i < l1_size; i++) {
1368 be64_to_cpus(&l1_table[i]);
1371 if (l1_allocated)
1372 g_free(l1_table);
1373 return ret;
1379 /*********************************************************/
1380 /* refcount checking functions */
1383 static uint64_t refcount_array_byte_size(BDRVQcow2State *s, uint64_t entries)
1385 /* This assertion holds because there is no way we can address more than
1386 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because
1387 * offsets have to be representable in bytes); due to every cluster
1388 * corresponding to one refcount entry, we are well below that limit */
1389 assert(entries < (UINT64_C(1) << (64 - 9)));
1391 /* Thanks to the assertion this will not overflow, because
1392 * s->refcount_order < 7.
1393 * (note: x << s->refcount_order == x * s->refcount_bits) */
1394 return DIV_ROUND_UP(entries << s->refcount_order, 8);
1398 * Reallocates *array so that it can hold new_size entries. *size must contain
1399 * the current number of entries in *array. If the reallocation fails, *array
1400 * and *size will not be modified and -errno will be returned. If the
1401 * reallocation is successful, *array will be set to the new buffer, *size
1402 * will be set to new_size and 0 will be returned. The size of the reallocated
1403 * refcount array buffer will be aligned to a cluster boundary, and the newly
1404 * allocated area will be zeroed.
1406 static int realloc_refcount_array(BDRVQcow2State *s, void **array,
1407 int64_t *size, int64_t new_size)
1409 int64_t old_byte_size, new_byte_size;
1410 void *new_ptr;
1412 /* Round to clusters so the array can be directly written to disk */
1413 old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size))
1414 * s->cluster_size;
1415 new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size))
1416 * s->cluster_size;
1418 if (new_byte_size == old_byte_size) {
1419 *size = new_size;
1420 return 0;
1423 assert(new_byte_size > 0);
1425 if (new_byte_size > SIZE_MAX) {
1426 return -ENOMEM;
1429 new_ptr = g_try_realloc(*array, new_byte_size);
1430 if (!new_ptr) {
1431 return -ENOMEM;
1434 if (new_byte_size > old_byte_size) {
1435 memset((char *)new_ptr + old_byte_size, 0,
1436 new_byte_size - old_byte_size);
1439 *array = new_ptr;
1440 *size = new_size;
1442 return 0;
1446 * Increases the refcount for a range of clusters in a given refcount table.
1447 * This is used to construct a temporary refcount table out of L1 and L2 tables
1448 * which can be compared to the refcount table saved in the image.
1450 * Modifies the number of errors in res.
1452 int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res,
1453 void **refcount_table,
1454 int64_t *refcount_table_size,
1455 int64_t offset, int64_t size)
1457 BDRVQcow2State *s = bs->opaque;
1458 uint64_t start, last, cluster_offset, k, refcount;
1459 int ret;
1461 if (size <= 0) {
1462 return 0;
1465 start = start_of_cluster(s, offset);
1466 last = start_of_cluster(s, offset + size - 1);
1467 for(cluster_offset = start; cluster_offset <= last;
1468 cluster_offset += s->cluster_size) {
1469 k = cluster_offset >> s->cluster_bits;
1470 if (k >= *refcount_table_size) {
1471 ret = realloc_refcount_array(s, refcount_table,
1472 refcount_table_size, k + 1);
1473 if (ret < 0) {
1474 res->check_errors++;
1475 return ret;
1479 refcount = s->get_refcount(*refcount_table, k);
1480 if (refcount == s->refcount_max) {
1481 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
1482 "\n", cluster_offset);
1483 fprintf(stderr, "Use qemu-img amend to increase the refcount entry "
1484 "width or qemu-img convert to create a clean copy if the "
1485 "image cannot be opened for writing\n");
1486 res->corruptions++;
1487 continue;
1489 s->set_refcount(*refcount_table, k, refcount + 1);
1492 return 0;
1495 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
1496 enum {
1497 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
1501 * Increases the refcount in the given refcount table for the all clusters
1502 * referenced in the L2 table. While doing so, performs some checks on L2
1503 * entries.
1505 * Returns the number of errors found by the checks or -errno if an internal
1506 * error occurred.
1508 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
1509 void **refcount_table,
1510 int64_t *refcount_table_size, int64_t l2_offset,
1511 int flags, BdrvCheckMode fix)
1513 BDRVQcow2State *s = bs->opaque;
1514 uint64_t *l2_table, l2_entry;
1515 uint64_t next_contiguous_offset = 0;
1516 int i, l2_size, nb_csectors, ret;
1518 /* Read L2 table from disk */
1519 l2_size = s->l2_size * sizeof(uint64_t);
1520 l2_table = g_malloc(l2_size);
1522 ret = bdrv_pread(bs->file, l2_offset, l2_table, l2_size);
1523 if (ret < 0) {
1524 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1525 res->check_errors++;
1526 goto fail;
1529 /* Do the actual checks */
1530 for(i = 0; i < s->l2_size; i++) {
1531 l2_entry = be64_to_cpu(l2_table[i]);
1533 switch (qcow2_get_cluster_type(l2_entry)) {
1534 case QCOW2_CLUSTER_COMPRESSED:
1535 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1536 if (l2_entry & QCOW_OFLAG_COPIED) {
1537 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
1538 "copied flag must never be set for compressed "
1539 "clusters\n", l2_entry >> s->cluster_bits);
1540 l2_entry &= ~QCOW_OFLAG_COPIED;
1541 res->corruptions++;
1544 /* Mark cluster as used */
1545 nb_csectors = ((l2_entry >> s->csize_shift) &
1546 s->csize_mask) + 1;
1547 l2_entry &= s->cluster_offset_mask;
1548 ret = qcow2_inc_refcounts_imrt(bs, res,
1549 refcount_table, refcount_table_size,
1550 l2_entry & ~511, nb_csectors * 512);
1551 if (ret < 0) {
1552 goto fail;
1555 if (flags & CHECK_FRAG_INFO) {
1556 res->bfi.allocated_clusters++;
1557 res->bfi.compressed_clusters++;
1559 /* Compressed clusters are fragmented by nature. Since they
1560 * take up sub-sector space but we only have sector granularity
1561 * I/O we need to re-read the same sectors even for adjacent
1562 * compressed clusters.
1564 res->bfi.fragmented_clusters++;
1566 break;
1568 case QCOW2_CLUSTER_ZERO_ALLOC:
1569 case QCOW2_CLUSTER_NORMAL:
1571 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
1573 if (flags & CHECK_FRAG_INFO) {
1574 res->bfi.allocated_clusters++;
1575 if (next_contiguous_offset &&
1576 offset != next_contiguous_offset) {
1577 res->bfi.fragmented_clusters++;
1579 next_contiguous_offset = offset + s->cluster_size;
1582 /* Correct offsets are cluster aligned */
1583 if (offset_into_cluster(s, offset)) {
1584 if (qcow2_get_cluster_type(l2_entry) ==
1585 QCOW2_CLUSTER_ZERO_ALLOC)
1587 fprintf(stderr, "%s offset=%" PRIx64 ": Preallocated zero "
1588 "cluster is not properly aligned; L2 entry "
1589 "corrupted.\n",
1590 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR",
1591 offset);
1592 if (fix & BDRV_FIX_ERRORS) {
1593 uint64_t l2e_offset =
1594 l2_offset + (uint64_t)i * sizeof(uint64_t);
1596 l2_entry = QCOW_OFLAG_ZERO;
1597 l2_table[i] = cpu_to_be64(l2_entry);
1598 ret = qcow2_pre_write_overlap_check(bs,
1599 QCOW2_OL_ACTIVE_L2 | QCOW2_OL_INACTIVE_L2,
1600 l2e_offset, sizeof(uint64_t));
1601 if (ret < 0) {
1602 fprintf(stderr, "ERROR: Overlap check failed\n");
1603 res->check_errors++;
1604 /* Something is seriously wrong, so abort checking
1605 * this L2 table */
1606 goto fail;
1609 ret = bdrv_pwrite_sync(bs->file, l2e_offset,
1610 &l2_table[i], sizeof(uint64_t));
1611 if (ret < 0) {
1612 fprintf(stderr, "ERROR: Failed to overwrite L2 "
1613 "table entry: %s\n", strerror(-ret));
1614 res->check_errors++;
1615 /* Do not abort, continue checking the rest of this
1616 * L2 table's entries */
1617 } else {
1618 res->corruptions_fixed++;
1619 /* Skip marking the cluster as used
1620 * (it is unused now) */
1621 continue;
1623 } else {
1624 res->corruptions++;
1626 } else {
1627 fprintf(stderr, "ERROR offset=%" PRIx64 ": Data cluster is "
1628 "not properly aligned; L2 entry corrupted.\n", offset);
1629 res->corruptions++;
1633 /* Mark cluster as used */
1634 ret = qcow2_inc_refcounts_imrt(bs, res,
1635 refcount_table, refcount_table_size,
1636 offset, s->cluster_size);
1637 if (ret < 0) {
1638 goto fail;
1640 break;
1643 case QCOW2_CLUSTER_ZERO_PLAIN:
1644 case QCOW2_CLUSTER_UNALLOCATED:
1645 break;
1647 default:
1648 abort();
1652 g_free(l2_table);
1653 return 0;
1655 fail:
1656 g_free(l2_table);
1657 return ret;
1661 * Increases the refcount for the L1 table, its L2 tables and all referenced
1662 * clusters in the given refcount table. While doing so, performs some checks
1663 * on L1 and L2 entries.
1665 * Returns the number of errors found by the checks or -errno if an internal
1666 * error occurred.
1668 static int check_refcounts_l1(BlockDriverState *bs,
1669 BdrvCheckResult *res,
1670 void **refcount_table,
1671 int64_t *refcount_table_size,
1672 int64_t l1_table_offset, int l1_size,
1673 int flags, BdrvCheckMode fix)
1675 BDRVQcow2State *s = bs->opaque;
1676 uint64_t *l1_table = NULL, l2_offset, l1_size2;
1677 int i, ret;
1679 l1_size2 = l1_size * sizeof(uint64_t);
1681 /* Mark L1 table as used */
1682 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, refcount_table_size,
1683 l1_table_offset, l1_size2);
1684 if (ret < 0) {
1685 goto fail;
1688 /* Read L1 table entries from disk */
1689 if (l1_size2 > 0) {
1690 l1_table = g_try_malloc(l1_size2);
1691 if (l1_table == NULL) {
1692 ret = -ENOMEM;
1693 res->check_errors++;
1694 goto fail;
1696 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
1697 if (ret < 0) {
1698 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1699 res->check_errors++;
1700 goto fail;
1702 for(i = 0;i < l1_size; i++)
1703 be64_to_cpus(&l1_table[i]);
1706 /* Do the actual checks */
1707 for(i = 0; i < l1_size; i++) {
1708 l2_offset = l1_table[i];
1709 if (l2_offset) {
1710 /* Mark L2 table as used */
1711 l2_offset &= L1E_OFFSET_MASK;
1712 ret = qcow2_inc_refcounts_imrt(bs, res,
1713 refcount_table, refcount_table_size,
1714 l2_offset, s->cluster_size);
1715 if (ret < 0) {
1716 goto fail;
1719 /* L2 tables are cluster aligned */
1720 if (offset_into_cluster(s, l2_offset)) {
1721 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1722 "cluster aligned; L1 entry corrupted\n", l2_offset);
1723 res->corruptions++;
1726 /* Process and check L2 entries */
1727 ret = check_refcounts_l2(bs, res, refcount_table,
1728 refcount_table_size, l2_offset, flags,
1729 fix);
1730 if (ret < 0) {
1731 goto fail;
1735 g_free(l1_table);
1736 return 0;
1738 fail:
1739 g_free(l1_table);
1740 return ret;
1744 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1746 * This function does not print an error message nor does it increment
1747 * check_errors if qcow2_get_refcount fails (this is because such an error will
1748 * have been already detected and sufficiently signaled by the calling function
1749 * (qcow2_check_refcounts) by the time this function is called).
1751 static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
1752 BdrvCheckMode fix)
1754 BDRVQcow2State *s = bs->opaque;
1755 uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size);
1756 int ret;
1757 uint64_t refcount;
1758 int i, j;
1760 for (i = 0; i < s->l1_size; i++) {
1761 uint64_t l1_entry = s->l1_table[i];
1762 uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK;
1763 bool l2_dirty = false;
1765 if (!l2_offset) {
1766 continue;
1769 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1770 &refcount);
1771 if (ret < 0) {
1772 /* don't print message nor increment check_errors */
1773 continue;
1775 if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) {
1776 fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1777 "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
1778 fix & BDRV_FIX_ERRORS ? "Repairing" :
1779 "ERROR",
1780 i, l1_entry, refcount);
1781 if (fix & BDRV_FIX_ERRORS) {
1782 s->l1_table[i] = refcount == 1
1783 ? l1_entry | QCOW_OFLAG_COPIED
1784 : l1_entry & ~QCOW_OFLAG_COPIED;
1785 ret = qcow2_write_l1_entry(bs, i);
1786 if (ret < 0) {
1787 res->check_errors++;
1788 goto fail;
1790 res->corruptions_fixed++;
1791 } else {
1792 res->corruptions++;
1796 ret = bdrv_pread(bs->file, l2_offset, l2_table,
1797 s->l2_size * sizeof(uint64_t));
1798 if (ret < 0) {
1799 fprintf(stderr, "ERROR: Could not read L2 table: %s\n",
1800 strerror(-ret));
1801 res->check_errors++;
1802 goto fail;
1805 for (j = 0; j < s->l2_size; j++) {
1806 uint64_t l2_entry = be64_to_cpu(l2_table[j]);
1807 uint64_t data_offset = l2_entry & L2E_OFFSET_MASK;
1808 QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry);
1810 if (cluster_type == QCOW2_CLUSTER_NORMAL ||
1811 cluster_type == QCOW2_CLUSTER_ZERO_ALLOC) {
1812 ret = qcow2_get_refcount(bs,
1813 data_offset >> s->cluster_bits,
1814 &refcount);
1815 if (ret < 0) {
1816 /* don't print message nor increment check_errors */
1817 continue;
1819 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
1820 fprintf(stderr, "%s OFLAG_COPIED data cluster: "
1821 "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
1822 fix & BDRV_FIX_ERRORS ? "Repairing" :
1823 "ERROR",
1824 l2_entry, refcount);
1825 if (fix & BDRV_FIX_ERRORS) {
1826 l2_table[j] = cpu_to_be64(refcount == 1
1827 ? l2_entry | QCOW_OFLAG_COPIED
1828 : l2_entry & ~QCOW_OFLAG_COPIED);
1829 l2_dirty = true;
1830 res->corruptions_fixed++;
1831 } else {
1832 res->corruptions++;
1838 if (l2_dirty) {
1839 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
1840 l2_offset, s->cluster_size);
1841 if (ret < 0) {
1842 fprintf(stderr, "ERROR: Could not write L2 table; metadata "
1843 "overlap check failed: %s\n", strerror(-ret));
1844 res->check_errors++;
1845 goto fail;
1848 ret = bdrv_pwrite(bs->file, l2_offset, l2_table,
1849 s->cluster_size);
1850 if (ret < 0) {
1851 fprintf(stderr, "ERROR: Could not write L2 table: %s\n",
1852 strerror(-ret));
1853 res->check_errors++;
1854 goto fail;
1859 ret = 0;
1861 fail:
1862 qemu_vfree(l2_table);
1863 return ret;
1867 * Checks consistency of refblocks and accounts for each refblock in
1868 * *refcount_table.
1870 static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res,
1871 BdrvCheckMode fix, bool *rebuild,
1872 void **refcount_table, int64_t *nb_clusters)
1874 BDRVQcow2State *s = bs->opaque;
1875 int64_t i, size;
1876 int ret;
1878 for(i = 0; i < s->refcount_table_size; i++) {
1879 uint64_t offset, cluster;
1880 offset = s->refcount_table[i];
1881 cluster = offset >> s->cluster_bits;
1883 /* Refcount blocks are cluster aligned */
1884 if (offset_into_cluster(s, offset)) {
1885 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1886 "cluster aligned; refcount table entry corrupted\n", i);
1887 res->corruptions++;
1888 *rebuild = true;
1889 continue;
1892 if (cluster >= *nb_clusters) {
1893 fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n",
1894 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
1896 if (fix & BDRV_FIX_ERRORS) {
1897 int64_t new_nb_clusters;
1898 Error *local_err = NULL;
1900 if (offset > INT64_MAX - s->cluster_size) {
1901 ret = -EINVAL;
1902 goto resize_fail;
1905 ret = bdrv_truncate(bs->file, offset + s->cluster_size,
1906 PREALLOC_MODE_OFF, &local_err);
1907 if (ret < 0) {
1908 error_report_err(local_err);
1909 goto resize_fail;
1911 size = bdrv_getlength(bs->file->bs);
1912 if (size < 0) {
1913 ret = size;
1914 goto resize_fail;
1917 new_nb_clusters = size_to_clusters(s, size);
1918 assert(new_nb_clusters >= *nb_clusters);
1920 ret = realloc_refcount_array(s, refcount_table,
1921 nb_clusters, new_nb_clusters);
1922 if (ret < 0) {
1923 res->check_errors++;
1924 return ret;
1927 if (cluster >= *nb_clusters) {
1928 ret = -EINVAL;
1929 goto resize_fail;
1932 res->corruptions_fixed++;
1933 ret = qcow2_inc_refcounts_imrt(bs, res,
1934 refcount_table, nb_clusters,
1935 offset, s->cluster_size);
1936 if (ret < 0) {
1937 return ret;
1939 /* No need to check whether the refcount is now greater than 1:
1940 * This area was just allocated and zeroed, so it can only be
1941 * exactly 1 after qcow2_inc_refcounts_imrt() */
1942 continue;
1944 resize_fail:
1945 res->corruptions++;
1946 *rebuild = true;
1947 fprintf(stderr, "ERROR could not resize image: %s\n",
1948 strerror(-ret));
1949 } else {
1950 res->corruptions++;
1952 continue;
1955 if (offset != 0) {
1956 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
1957 offset, s->cluster_size);
1958 if (ret < 0) {
1959 return ret;
1961 if (s->get_refcount(*refcount_table, cluster) != 1) {
1962 fprintf(stderr, "ERROR refcount block %" PRId64
1963 " refcount=%" PRIu64 "\n", i,
1964 s->get_refcount(*refcount_table, cluster));
1965 res->corruptions++;
1966 *rebuild = true;
1971 return 0;
1975 * Calculates an in-memory refcount table.
1977 static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1978 BdrvCheckMode fix, bool *rebuild,
1979 void **refcount_table, int64_t *nb_clusters)
1981 BDRVQcow2State *s = bs->opaque;
1982 int64_t i;
1983 QCowSnapshot *sn;
1984 int ret;
1986 if (!*refcount_table) {
1987 int64_t old_size = 0;
1988 ret = realloc_refcount_array(s, refcount_table,
1989 &old_size, *nb_clusters);
1990 if (ret < 0) {
1991 res->check_errors++;
1992 return ret;
1996 /* header */
1997 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
1998 0, s->cluster_size);
1999 if (ret < 0) {
2000 return ret;
2003 /* current L1 table */
2004 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
2005 s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO,
2006 fix);
2007 if (ret < 0) {
2008 return ret;
2011 /* snapshots */
2012 for (i = 0; i < s->nb_snapshots; i++) {
2013 sn = s->snapshots + i;
2014 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
2015 sn->l1_table_offset, sn->l1_size, 0, fix);
2016 if (ret < 0) {
2017 return ret;
2020 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2021 s->snapshots_offset, s->snapshots_size);
2022 if (ret < 0) {
2023 return ret;
2026 /* refcount data */
2027 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2028 s->refcount_table_offset,
2029 s->refcount_table_size * sizeof(uint64_t));
2030 if (ret < 0) {
2031 return ret;
2034 /* encryption */
2035 if (s->crypto_header.length) {
2036 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2037 s->crypto_header.offset,
2038 s->crypto_header.length);
2039 if (ret < 0) {
2040 return ret;
2044 /* bitmaps */
2045 ret = qcow2_check_bitmaps_refcounts(bs, res, refcount_table, nb_clusters);
2046 if (ret < 0) {
2047 return ret;
2050 return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters);
2054 * Compares the actual reference count for each cluster in the image against the
2055 * refcount as reported by the refcount structures on-disk.
2057 static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2058 BdrvCheckMode fix, bool *rebuild,
2059 int64_t *highest_cluster,
2060 void *refcount_table, int64_t nb_clusters)
2062 BDRVQcow2State *s = bs->opaque;
2063 int64_t i;
2064 uint64_t refcount1, refcount2;
2065 int ret;
2067 for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) {
2068 ret = qcow2_get_refcount(bs, i, &refcount1);
2069 if (ret < 0) {
2070 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
2071 i, strerror(-ret));
2072 res->check_errors++;
2073 continue;
2076 refcount2 = s->get_refcount(refcount_table, i);
2078 if (refcount1 > 0 || refcount2 > 0) {
2079 *highest_cluster = i;
2082 if (refcount1 != refcount2) {
2083 /* Check if we're allowed to fix the mismatch */
2084 int *num_fixed = NULL;
2085 if (refcount1 == 0) {
2086 *rebuild = true;
2087 } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
2088 num_fixed = &res->leaks_fixed;
2089 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
2090 num_fixed = &res->corruptions_fixed;
2093 fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64
2094 " reference=%" PRIu64 "\n",
2095 num_fixed != NULL ? "Repairing" :
2096 refcount1 < refcount2 ? "ERROR" :
2097 "Leaked",
2098 i, refcount1, refcount2);
2100 if (num_fixed) {
2101 ret = update_refcount(bs, i << s->cluster_bits, 1,
2102 refcount_diff(refcount1, refcount2),
2103 refcount1 > refcount2,
2104 QCOW2_DISCARD_ALWAYS);
2105 if (ret >= 0) {
2106 (*num_fixed)++;
2107 continue;
2111 /* And if we couldn't, print an error */
2112 if (refcount1 < refcount2) {
2113 res->corruptions++;
2114 } else {
2115 res->leaks++;
2122 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to
2123 * the on-disk refcount structures.
2125 * On input, *first_free_cluster tells where to start looking, and need not
2126 * actually be a free cluster; the returned offset will not be before that
2127 * cluster. On output, *first_free_cluster points to the first gap found, even
2128 * if that gap was too small to be used as the returned offset.
2130 * Note that *first_free_cluster is a cluster index whereas the return value is
2131 * an offset.
2133 static int64_t alloc_clusters_imrt(BlockDriverState *bs,
2134 int cluster_count,
2135 void **refcount_table,
2136 int64_t *imrt_nb_clusters,
2137 int64_t *first_free_cluster)
2139 BDRVQcow2State *s = bs->opaque;
2140 int64_t cluster = *first_free_cluster, i;
2141 bool first_gap = true;
2142 int contiguous_free_clusters;
2143 int ret;
2145 /* Starting at *first_free_cluster, find a range of at least cluster_count
2146 * continuously free clusters */
2147 for (contiguous_free_clusters = 0;
2148 cluster < *imrt_nb_clusters &&
2149 contiguous_free_clusters < cluster_count;
2150 cluster++)
2152 if (!s->get_refcount(*refcount_table, cluster)) {
2153 contiguous_free_clusters++;
2154 if (first_gap) {
2155 /* If this is the first free cluster found, update
2156 * *first_free_cluster accordingly */
2157 *first_free_cluster = cluster;
2158 first_gap = false;
2160 } else if (contiguous_free_clusters) {
2161 contiguous_free_clusters = 0;
2165 /* If contiguous_free_clusters is greater than zero, it contains the number
2166 * of continuously free clusters until the current cluster; the first free
2167 * cluster in the current "gap" is therefore
2168 * cluster - contiguous_free_clusters */
2170 /* If no such range could be found, grow the in-memory refcount table
2171 * accordingly to append free clusters at the end of the image */
2172 if (contiguous_free_clusters < cluster_count) {
2173 /* contiguous_free_clusters clusters are already empty at the image end;
2174 * we need cluster_count clusters; therefore, we have to allocate
2175 * cluster_count - contiguous_free_clusters new clusters at the end of
2176 * the image (which is the current value of cluster; note that cluster
2177 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond
2178 * the image end) */
2179 ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters,
2180 cluster + cluster_count
2181 - contiguous_free_clusters);
2182 if (ret < 0) {
2183 return ret;
2187 /* Go back to the first free cluster */
2188 cluster -= contiguous_free_clusters;
2189 for (i = 0; i < cluster_count; i++) {
2190 s->set_refcount(*refcount_table, cluster + i, 1);
2193 return cluster << s->cluster_bits;
2197 * Creates a new refcount structure based solely on the in-memory information
2198 * given through *refcount_table. All necessary allocations will be reflected
2199 * in that array.
2201 * On success, the old refcount structure is leaked (it will be covered by the
2202 * new refcount structure).
2204 static int rebuild_refcount_structure(BlockDriverState *bs,
2205 BdrvCheckResult *res,
2206 void **refcount_table,
2207 int64_t *nb_clusters)
2209 BDRVQcow2State *s = bs->opaque;
2210 int64_t first_free_cluster = 0, reftable_offset = -1, cluster = 0;
2211 int64_t refblock_offset, refblock_start, refblock_index;
2212 uint32_t reftable_size = 0;
2213 uint64_t *on_disk_reftable = NULL;
2214 void *on_disk_refblock;
2215 int ret = 0;
2216 struct {
2217 uint64_t reftable_offset;
2218 uint32_t reftable_clusters;
2219 } QEMU_PACKED reftable_offset_and_clusters;
2221 qcow2_cache_empty(bs, s->refcount_block_cache);
2223 write_refblocks:
2224 for (; cluster < *nb_clusters; cluster++) {
2225 if (!s->get_refcount(*refcount_table, cluster)) {
2226 continue;
2229 refblock_index = cluster >> s->refcount_block_bits;
2230 refblock_start = refblock_index << s->refcount_block_bits;
2232 /* Don't allocate a cluster in a refblock already written to disk */
2233 if (first_free_cluster < refblock_start) {
2234 first_free_cluster = refblock_start;
2236 refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table,
2237 nb_clusters, &first_free_cluster);
2238 if (refblock_offset < 0) {
2239 fprintf(stderr, "ERROR allocating refblock: %s\n",
2240 strerror(-refblock_offset));
2241 res->check_errors++;
2242 ret = refblock_offset;
2243 goto fail;
2246 if (reftable_size <= refblock_index) {
2247 uint32_t old_reftable_size = reftable_size;
2248 uint64_t *new_on_disk_reftable;
2250 reftable_size = ROUND_UP((refblock_index + 1) * sizeof(uint64_t),
2251 s->cluster_size) / sizeof(uint64_t);
2252 new_on_disk_reftable = g_try_realloc(on_disk_reftable,
2253 reftable_size *
2254 sizeof(uint64_t));
2255 if (!new_on_disk_reftable) {
2256 res->check_errors++;
2257 ret = -ENOMEM;
2258 goto fail;
2260 on_disk_reftable = new_on_disk_reftable;
2262 memset(on_disk_reftable + old_reftable_size, 0,
2263 (reftable_size - old_reftable_size) * sizeof(uint64_t));
2265 /* The offset we have for the reftable is now no longer valid;
2266 * this will leak that range, but we can easily fix that by running
2267 * a leak-fixing check after this rebuild operation */
2268 reftable_offset = -1;
2269 } else {
2270 assert(on_disk_reftable);
2272 on_disk_reftable[refblock_index] = refblock_offset;
2274 /* If this is apparently the last refblock (for now), try to squeeze the
2275 * reftable in */
2276 if (refblock_index == (*nb_clusters - 1) >> s->refcount_block_bits &&
2277 reftable_offset < 0)
2279 uint64_t reftable_clusters = size_to_clusters(s, reftable_size *
2280 sizeof(uint64_t));
2281 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2282 refcount_table, nb_clusters,
2283 &first_free_cluster);
2284 if (reftable_offset < 0) {
2285 fprintf(stderr, "ERROR allocating reftable: %s\n",
2286 strerror(-reftable_offset));
2287 res->check_errors++;
2288 ret = reftable_offset;
2289 goto fail;
2293 ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset,
2294 s->cluster_size);
2295 if (ret < 0) {
2296 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2297 goto fail;
2300 /* The size of *refcount_table is always cluster-aligned, therefore the
2301 * write operation will not overflow */
2302 on_disk_refblock = (void *)((char *) *refcount_table +
2303 refblock_index * s->cluster_size);
2305 ret = bdrv_write(bs->file, refblock_offset / BDRV_SECTOR_SIZE,
2306 on_disk_refblock, s->cluster_sectors);
2307 if (ret < 0) {
2308 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2309 goto fail;
2312 /* Go to the end of this refblock */
2313 cluster = refblock_start + s->refcount_block_size - 1;
2316 if (reftable_offset < 0) {
2317 uint64_t post_refblock_start, reftable_clusters;
2319 post_refblock_start = ROUND_UP(*nb_clusters, s->refcount_block_size);
2320 reftable_clusters = size_to_clusters(s,
2321 reftable_size * sizeof(uint64_t));
2322 /* Not pretty but simple */
2323 if (first_free_cluster < post_refblock_start) {
2324 first_free_cluster = post_refblock_start;
2326 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2327 refcount_table, nb_clusters,
2328 &first_free_cluster);
2329 if (reftable_offset < 0) {
2330 fprintf(stderr, "ERROR allocating reftable: %s\n",
2331 strerror(-reftable_offset));
2332 res->check_errors++;
2333 ret = reftable_offset;
2334 goto fail;
2337 goto write_refblocks;
2340 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2341 cpu_to_be64s(&on_disk_reftable[refblock_index]);
2344 ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset,
2345 reftable_size * sizeof(uint64_t));
2346 if (ret < 0) {
2347 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2348 goto fail;
2351 assert(reftable_size < INT_MAX / sizeof(uint64_t));
2352 ret = bdrv_pwrite(bs->file, reftable_offset, on_disk_reftable,
2353 reftable_size * sizeof(uint64_t));
2354 if (ret < 0) {
2355 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2356 goto fail;
2359 /* Enter new reftable into the image header */
2360 reftable_offset_and_clusters.reftable_offset = cpu_to_be64(reftable_offset);
2361 reftable_offset_and_clusters.reftable_clusters =
2362 cpu_to_be32(size_to_clusters(s, reftable_size * sizeof(uint64_t)));
2363 ret = bdrv_pwrite_sync(bs->file,
2364 offsetof(QCowHeader, refcount_table_offset),
2365 &reftable_offset_and_clusters,
2366 sizeof(reftable_offset_and_clusters));
2367 if (ret < 0) {
2368 fprintf(stderr, "ERROR setting reftable: %s\n", strerror(-ret));
2369 goto fail;
2372 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2373 be64_to_cpus(&on_disk_reftable[refblock_index]);
2375 s->refcount_table = on_disk_reftable;
2376 s->refcount_table_offset = reftable_offset;
2377 s->refcount_table_size = reftable_size;
2378 update_max_refcount_table_index(s);
2380 return 0;
2382 fail:
2383 g_free(on_disk_reftable);
2384 return ret;
2388 * Checks an image for refcount consistency.
2390 * Returns 0 if no errors are found, the number of errors in case the image is
2391 * detected as corrupted, and -errno when an internal error occurred.
2393 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2394 BdrvCheckMode fix)
2396 BDRVQcow2State *s = bs->opaque;
2397 BdrvCheckResult pre_compare_res;
2398 int64_t size, highest_cluster, nb_clusters;
2399 void *refcount_table = NULL;
2400 bool rebuild = false;
2401 int ret;
2403 size = bdrv_getlength(bs->file->bs);
2404 if (size < 0) {
2405 res->check_errors++;
2406 return size;
2409 nb_clusters = size_to_clusters(s, size);
2410 if (nb_clusters > INT_MAX) {
2411 res->check_errors++;
2412 return -EFBIG;
2415 res->bfi.total_clusters =
2416 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE);
2418 ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table,
2419 &nb_clusters);
2420 if (ret < 0) {
2421 goto fail;
2424 /* In case we don't need to rebuild the refcount structure (but want to fix
2425 * something), this function is immediately called again, in which case the
2426 * result should be ignored */
2427 pre_compare_res = *res;
2428 compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table,
2429 nb_clusters);
2431 if (rebuild && (fix & BDRV_FIX_ERRORS)) {
2432 BdrvCheckResult old_res = *res;
2433 int fresh_leaks = 0;
2435 fprintf(stderr, "Rebuilding refcount structure\n");
2436 ret = rebuild_refcount_structure(bs, res, &refcount_table,
2437 &nb_clusters);
2438 if (ret < 0) {
2439 goto fail;
2442 res->corruptions = 0;
2443 res->leaks = 0;
2445 /* Because the old reftable has been exchanged for a new one the
2446 * references have to be recalculated */
2447 rebuild = false;
2448 memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters));
2449 ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table,
2450 &nb_clusters);
2451 if (ret < 0) {
2452 goto fail;
2455 if (fix & BDRV_FIX_LEAKS) {
2456 /* The old refcount structures are now leaked, fix it; the result
2457 * can be ignored, aside from leaks which were introduced by
2458 * rebuild_refcount_structure() that could not be fixed */
2459 BdrvCheckResult saved_res = *res;
2460 *res = (BdrvCheckResult){ 0 };
2462 compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild,
2463 &highest_cluster, refcount_table, nb_clusters);
2464 if (rebuild) {
2465 fprintf(stderr, "ERROR rebuilt refcount structure is still "
2466 "broken\n");
2469 /* Any leaks accounted for here were introduced by
2470 * rebuild_refcount_structure() because that function has created a
2471 * new refcount structure from scratch */
2472 fresh_leaks = res->leaks;
2473 *res = saved_res;
2476 if (res->corruptions < old_res.corruptions) {
2477 res->corruptions_fixed += old_res.corruptions - res->corruptions;
2479 if (res->leaks < old_res.leaks) {
2480 res->leaks_fixed += old_res.leaks - res->leaks;
2482 res->leaks += fresh_leaks;
2483 } else if (fix) {
2484 if (rebuild) {
2485 fprintf(stderr, "ERROR need to rebuild refcount structures\n");
2486 res->check_errors++;
2487 ret = -EIO;
2488 goto fail;
2491 if (res->leaks || res->corruptions) {
2492 *res = pre_compare_res;
2493 compare_refcounts(bs, res, fix, &rebuild, &highest_cluster,
2494 refcount_table, nb_clusters);
2498 /* check OFLAG_COPIED */
2499 ret = check_oflag_copied(bs, res, fix);
2500 if (ret < 0) {
2501 goto fail;
2504 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
2505 ret = 0;
2507 fail:
2508 g_free(refcount_table);
2510 return ret;
2513 #define overlaps_with(ofs, sz) \
2514 ranges_overlap(offset, size, ofs, sz)
2517 * Checks if the given offset into the image file is actually free to use by
2518 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
2519 * i.e. a sanity check without relying on the refcount tables.
2521 * The ign parameter specifies what checks not to perform (being a bitmask of
2522 * QCow2MetadataOverlap values), i.e., what sections to ignore.
2524 * Returns:
2525 * - 0 if writing to this offset will not affect the mentioned metadata
2526 * - a positive QCow2MetadataOverlap value indicating one overlapping section
2527 * - a negative value (-errno) indicating an error while performing a check,
2528 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
2530 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset,
2531 int64_t size)
2533 BDRVQcow2State *s = bs->opaque;
2534 int chk = s->overlap_check & ~ign;
2535 int i, j;
2537 if (!size) {
2538 return 0;
2541 if (chk & QCOW2_OL_MAIN_HEADER) {
2542 if (offset < s->cluster_size) {
2543 return QCOW2_OL_MAIN_HEADER;
2547 /* align range to test to cluster boundaries */
2548 size = align_offset(offset_into_cluster(s, offset) + size, s->cluster_size);
2549 offset = start_of_cluster(s, offset);
2551 if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) {
2552 if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) {
2553 return QCOW2_OL_ACTIVE_L1;
2557 if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) {
2558 if (overlaps_with(s->refcount_table_offset,
2559 s->refcount_table_size * sizeof(uint64_t))) {
2560 return QCOW2_OL_REFCOUNT_TABLE;
2564 if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) {
2565 if (overlaps_with(s->snapshots_offset, s->snapshots_size)) {
2566 return QCOW2_OL_SNAPSHOT_TABLE;
2570 if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) {
2571 for (i = 0; i < s->nb_snapshots; i++) {
2572 if (s->snapshots[i].l1_size &&
2573 overlaps_with(s->snapshots[i].l1_table_offset,
2574 s->snapshots[i].l1_size * sizeof(uint64_t))) {
2575 return QCOW2_OL_INACTIVE_L1;
2580 if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) {
2581 for (i = 0; i < s->l1_size; i++) {
2582 if ((s->l1_table[i] & L1E_OFFSET_MASK) &&
2583 overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK,
2584 s->cluster_size)) {
2585 return QCOW2_OL_ACTIVE_L2;
2590 if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) {
2591 unsigned last_entry = s->max_refcount_table_index;
2592 assert(last_entry < s->refcount_table_size);
2593 assert(last_entry + 1 == s->refcount_table_size ||
2594 (s->refcount_table[last_entry + 1] & REFT_OFFSET_MASK) == 0);
2595 for (i = 0; i <= last_entry; i++) {
2596 if ((s->refcount_table[i] & REFT_OFFSET_MASK) &&
2597 overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK,
2598 s->cluster_size)) {
2599 return QCOW2_OL_REFCOUNT_BLOCK;
2604 if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) {
2605 for (i = 0; i < s->nb_snapshots; i++) {
2606 uint64_t l1_ofs = s->snapshots[i].l1_table_offset;
2607 uint32_t l1_sz = s->snapshots[i].l1_size;
2608 uint64_t l1_sz2 = l1_sz * sizeof(uint64_t);
2609 uint64_t *l1 = g_try_malloc(l1_sz2);
2610 int ret;
2612 if (l1_sz2 && l1 == NULL) {
2613 return -ENOMEM;
2616 ret = bdrv_pread(bs->file, l1_ofs, l1, l1_sz2);
2617 if (ret < 0) {
2618 g_free(l1);
2619 return ret;
2622 for (j = 0; j < l1_sz; j++) {
2623 uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK;
2624 if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) {
2625 g_free(l1);
2626 return QCOW2_OL_INACTIVE_L2;
2630 g_free(l1);
2634 return 0;
2637 static const char *metadata_ol_names[] = {
2638 [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header",
2639 [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table",
2640 [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table",
2641 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table",
2642 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block",
2643 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table",
2644 [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table",
2645 [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table",
2649 * First performs a check for metadata overlaps (through
2650 * qcow2_check_metadata_overlap); if that fails with a negative value (error
2651 * while performing a check), that value is returned. If an impending overlap
2652 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
2653 * and -EIO returned.
2655 * Returns 0 if there were neither overlaps nor errors while checking for
2656 * overlaps; or a negative value (-errno) on error.
2658 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset,
2659 int64_t size)
2661 int ret = qcow2_check_metadata_overlap(bs, ign, offset, size);
2663 if (ret < 0) {
2664 return ret;
2665 } else if (ret > 0) {
2666 int metadata_ol_bitnr = ctz32(ret);
2667 assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR);
2669 qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid "
2670 "write on metadata (overlaps with %s)",
2671 metadata_ol_names[metadata_ol_bitnr]);
2672 return -EIO;
2675 return 0;
2678 /* A pointer to a function of this type is given to walk_over_reftable(). That
2679 * function will create refblocks and pass them to a RefblockFinishOp once they
2680 * are completed (@refblock). @refblock_empty is set if the refblock is
2681 * completely empty.
2683 * Along with the refblock, a corresponding reftable entry is passed, in the
2684 * reftable @reftable (which may be reallocated) at @reftable_index.
2686 * @allocated should be set to true if a new cluster has been allocated.
2688 typedef int (RefblockFinishOp)(BlockDriverState *bs, uint64_t **reftable,
2689 uint64_t reftable_index, uint64_t *reftable_size,
2690 void *refblock, bool refblock_empty,
2691 bool *allocated, Error **errp);
2694 * This "operation" for walk_over_reftable() allocates the refblock on disk (if
2695 * it is not empty) and inserts its offset into the new reftable. The size of
2696 * this new reftable is increased as required.
2698 static int alloc_refblock(BlockDriverState *bs, uint64_t **reftable,
2699 uint64_t reftable_index, uint64_t *reftable_size,
2700 void *refblock, bool refblock_empty, bool *allocated,
2701 Error **errp)
2703 BDRVQcow2State *s = bs->opaque;
2704 int64_t offset;
2706 if (!refblock_empty && reftable_index >= *reftable_size) {
2707 uint64_t *new_reftable;
2708 uint64_t new_reftable_size;
2710 new_reftable_size = ROUND_UP(reftable_index + 1,
2711 s->cluster_size / sizeof(uint64_t));
2712 if (new_reftable_size > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) {
2713 error_setg(errp,
2714 "This operation would make the refcount table grow "
2715 "beyond the maximum size supported by QEMU, aborting");
2716 return -ENOTSUP;
2719 new_reftable = g_try_realloc(*reftable, new_reftable_size *
2720 sizeof(uint64_t));
2721 if (!new_reftable) {
2722 error_setg(errp, "Failed to increase reftable buffer size");
2723 return -ENOMEM;
2726 memset(new_reftable + *reftable_size, 0,
2727 (new_reftable_size - *reftable_size) * sizeof(uint64_t));
2729 *reftable = new_reftable;
2730 *reftable_size = new_reftable_size;
2733 if (!refblock_empty && !(*reftable)[reftable_index]) {
2734 offset = qcow2_alloc_clusters(bs, s->cluster_size);
2735 if (offset < 0) {
2736 error_setg_errno(errp, -offset, "Failed to allocate refblock");
2737 return offset;
2739 (*reftable)[reftable_index] = offset;
2740 *allocated = true;
2743 return 0;
2747 * This "operation" for walk_over_reftable() writes the refblock to disk at the
2748 * offset specified by the new reftable's entry. It does not modify the new
2749 * reftable or change any refcounts.
2751 static int flush_refblock(BlockDriverState *bs, uint64_t **reftable,
2752 uint64_t reftable_index, uint64_t *reftable_size,
2753 void *refblock, bool refblock_empty, bool *allocated,
2754 Error **errp)
2756 BDRVQcow2State *s = bs->opaque;
2757 int64_t offset;
2758 int ret;
2760 if (reftable_index < *reftable_size && (*reftable)[reftable_index]) {
2761 offset = (*reftable)[reftable_index];
2763 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
2764 if (ret < 0) {
2765 error_setg_errno(errp, -ret, "Overlap check failed");
2766 return ret;
2769 ret = bdrv_pwrite(bs->file, offset, refblock, s->cluster_size);
2770 if (ret < 0) {
2771 error_setg_errno(errp, -ret, "Failed to write refblock");
2772 return ret;
2774 } else {
2775 assert(refblock_empty);
2778 return 0;
2782 * This function walks over the existing reftable and every referenced refblock;
2783 * if @new_set_refcount is non-NULL, it is called for every refcount entry to
2784 * create an equal new entry in the passed @new_refblock. Once that
2785 * @new_refblock is completely filled, @operation will be called.
2787 * @status_cb and @cb_opaque are used for the amend operation's status callback.
2788 * @index is the index of the walk_over_reftable() calls and @total is the total
2789 * number of walk_over_reftable() calls per amend operation. Both are used for
2790 * calculating the parameters for the status callback.
2792 * @allocated is set to true if a new cluster has been allocated.
2794 static int walk_over_reftable(BlockDriverState *bs, uint64_t **new_reftable,
2795 uint64_t *new_reftable_index,
2796 uint64_t *new_reftable_size,
2797 void *new_refblock, int new_refblock_size,
2798 int new_refcount_bits,
2799 RefblockFinishOp *operation, bool *allocated,
2800 Qcow2SetRefcountFunc *new_set_refcount,
2801 BlockDriverAmendStatusCB *status_cb,
2802 void *cb_opaque, int index, int total,
2803 Error **errp)
2805 BDRVQcow2State *s = bs->opaque;
2806 uint64_t reftable_index;
2807 bool new_refblock_empty = true;
2808 int refblock_index;
2809 int new_refblock_index = 0;
2810 int ret;
2812 for (reftable_index = 0; reftable_index < s->refcount_table_size;
2813 reftable_index++)
2815 uint64_t refblock_offset = s->refcount_table[reftable_index]
2816 & REFT_OFFSET_MASK;
2818 status_cb(bs, (uint64_t)index * s->refcount_table_size + reftable_index,
2819 (uint64_t)total * s->refcount_table_size, cb_opaque);
2821 if (refblock_offset) {
2822 void *refblock;
2824 if (offset_into_cluster(s, refblock_offset)) {
2825 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
2826 PRIx64 " unaligned (reftable index: %#"
2827 PRIx64 ")", refblock_offset,
2828 reftable_index);
2829 error_setg(errp,
2830 "Image is corrupt (unaligned refblock offset)");
2831 return -EIO;
2834 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offset,
2835 &refblock);
2836 if (ret < 0) {
2837 error_setg_errno(errp, -ret, "Failed to retrieve refblock");
2838 return ret;
2841 for (refblock_index = 0; refblock_index < s->refcount_block_size;
2842 refblock_index++)
2844 uint64_t refcount;
2846 if (new_refblock_index >= new_refblock_size) {
2847 /* new_refblock is now complete */
2848 ret = operation(bs, new_reftable, *new_reftable_index,
2849 new_reftable_size, new_refblock,
2850 new_refblock_empty, allocated, errp);
2851 if (ret < 0) {
2852 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2853 return ret;
2856 (*new_reftable_index)++;
2857 new_refblock_index = 0;
2858 new_refblock_empty = true;
2861 refcount = s->get_refcount(refblock, refblock_index);
2862 if (new_refcount_bits < 64 && refcount >> new_refcount_bits) {
2863 uint64_t offset;
2865 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2867 offset = ((reftable_index << s->refcount_block_bits)
2868 + refblock_index) << s->cluster_bits;
2870 error_setg(errp, "Cannot decrease refcount entry width to "
2871 "%i bits: Cluster at offset %#" PRIx64 " has a "
2872 "refcount of %" PRIu64, new_refcount_bits,
2873 offset, refcount);
2874 return -EINVAL;
2877 if (new_set_refcount) {
2878 new_set_refcount(new_refblock, new_refblock_index++,
2879 refcount);
2880 } else {
2881 new_refblock_index++;
2883 new_refblock_empty = new_refblock_empty && refcount == 0;
2886 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2887 } else {
2888 /* No refblock means every refcount is 0 */
2889 for (refblock_index = 0; refblock_index < s->refcount_block_size;
2890 refblock_index++)
2892 if (new_refblock_index >= new_refblock_size) {
2893 /* new_refblock is now complete */
2894 ret = operation(bs, new_reftable, *new_reftable_index,
2895 new_reftable_size, new_refblock,
2896 new_refblock_empty, allocated, errp);
2897 if (ret < 0) {
2898 return ret;
2901 (*new_reftable_index)++;
2902 new_refblock_index = 0;
2903 new_refblock_empty = true;
2906 if (new_set_refcount) {
2907 new_set_refcount(new_refblock, new_refblock_index++, 0);
2908 } else {
2909 new_refblock_index++;
2915 if (new_refblock_index > 0) {
2916 /* Complete the potentially existing partially filled final refblock */
2917 if (new_set_refcount) {
2918 for (; new_refblock_index < new_refblock_size;
2919 new_refblock_index++)
2921 new_set_refcount(new_refblock, new_refblock_index, 0);
2925 ret = operation(bs, new_reftable, *new_reftable_index,
2926 new_reftable_size, new_refblock, new_refblock_empty,
2927 allocated, errp);
2928 if (ret < 0) {
2929 return ret;
2932 (*new_reftable_index)++;
2935 status_cb(bs, (uint64_t)(index + 1) * s->refcount_table_size,
2936 (uint64_t)total * s->refcount_table_size, cb_opaque);
2938 return 0;
2941 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order,
2942 BlockDriverAmendStatusCB *status_cb,
2943 void *cb_opaque, Error **errp)
2945 BDRVQcow2State *s = bs->opaque;
2946 Qcow2GetRefcountFunc *new_get_refcount;
2947 Qcow2SetRefcountFunc *new_set_refcount;
2948 void *new_refblock = qemu_blockalign(bs->file->bs, s->cluster_size);
2949 uint64_t *new_reftable = NULL, new_reftable_size = 0;
2950 uint64_t *old_reftable, old_reftable_size, old_reftable_offset;
2951 uint64_t new_reftable_index = 0;
2952 uint64_t i;
2953 int64_t new_reftable_offset = 0, allocated_reftable_size = 0;
2954 int new_refblock_size, new_refcount_bits = 1 << refcount_order;
2955 int old_refcount_order;
2956 int walk_index = 0;
2957 int ret;
2958 bool new_allocation;
2960 assert(s->qcow_version >= 3);
2961 assert(refcount_order >= 0 && refcount_order <= 6);
2963 /* see qcow2_open() */
2964 new_refblock_size = 1 << (s->cluster_bits - (refcount_order - 3));
2966 new_get_refcount = get_refcount_funcs[refcount_order];
2967 new_set_refcount = set_refcount_funcs[refcount_order];
2970 do {
2971 int total_walks;
2973 new_allocation = false;
2975 /* At least we have to do this walk and the one which writes the
2976 * refblocks; also, at least we have to do this loop here at least
2977 * twice (normally), first to do the allocations, and second to
2978 * determine that everything is correctly allocated, this then makes
2979 * three walks in total */
2980 total_walks = MAX(walk_index + 2, 3);
2982 /* First, allocate the structures so they are present in the refcount
2983 * structures */
2984 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
2985 &new_reftable_size, NULL, new_refblock_size,
2986 new_refcount_bits, &alloc_refblock,
2987 &new_allocation, NULL, status_cb, cb_opaque,
2988 walk_index++, total_walks, errp);
2989 if (ret < 0) {
2990 goto done;
2993 new_reftable_index = 0;
2995 if (new_allocation) {
2996 if (new_reftable_offset) {
2997 qcow2_free_clusters(bs, new_reftable_offset,
2998 allocated_reftable_size * sizeof(uint64_t),
2999 QCOW2_DISCARD_NEVER);
3002 new_reftable_offset = qcow2_alloc_clusters(bs, new_reftable_size *
3003 sizeof(uint64_t));
3004 if (new_reftable_offset < 0) {
3005 error_setg_errno(errp, -new_reftable_offset,
3006 "Failed to allocate the new reftable");
3007 ret = new_reftable_offset;
3008 goto done;
3010 allocated_reftable_size = new_reftable_size;
3012 } while (new_allocation);
3014 /* Second, write the new refblocks */
3015 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
3016 &new_reftable_size, new_refblock,
3017 new_refblock_size, new_refcount_bits,
3018 &flush_refblock, &new_allocation, new_set_refcount,
3019 status_cb, cb_opaque, walk_index, walk_index + 1,
3020 errp);
3021 if (ret < 0) {
3022 goto done;
3024 assert(!new_allocation);
3027 /* Write the new reftable */
3028 ret = qcow2_pre_write_overlap_check(bs, 0, new_reftable_offset,
3029 new_reftable_size * sizeof(uint64_t));
3030 if (ret < 0) {
3031 error_setg_errno(errp, -ret, "Overlap check failed");
3032 goto done;
3035 for (i = 0; i < new_reftable_size; i++) {
3036 cpu_to_be64s(&new_reftable[i]);
3039 ret = bdrv_pwrite(bs->file, new_reftable_offset, new_reftable,
3040 new_reftable_size * sizeof(uint64_t));
3042 for (i = 0; i < new_reftable_size; i++) {
3043 be64_to_cpus(&new_reftable[i]);
3046 if (ret < 0) {
3047 error_setg_errno(errp, -ret, "Failed to write the new reftable");
3048 goto done;
3052 /* Empty the refcount cache */
3053 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
3054 if (ret < 0) {
3055 error_setg_errno(errp, -ret, "Failed to flush the refblock cache");
3056 goto done;
3059 /* Update the image header to point to the new reftable; this only updates
3060 * the fields which are relevant to qcow2_update_header(); other fields
3061 * such as s->refcount_table or s->refcount_bits stay stale for now
3062 * (because we have to restore everything if qcow2_update_header() fails) */
3063 old_refcount_order = s->refcount_order;
3064 old_reftable_size = s->refcount_table_size;
3065 old_reftable_offset = s->refcount_table_offset;
3067 s->refcount_order = refcount_order;
3068 s->refcount_table_size = new_reftable_size;
3069 s->refcount_table_offset = new_reftable_offset;
3071 ret = qcow2_update_header(bs);
3072 if (ret < 0) {
3073 s->refcount_order = old_refcount_order;
3074 s->refcount_table_size = old_reftable_size;
3075 s->refcount_table_offset = old_reftable_offset;
3076 error_setg_errno(errp, -ret, "Failed to update the qcow2 header");
3077 goto done;
3080 /* Now update the rest of the in-memory information */
3081 old_reftable = s->refcount_table;
3082 s->refcount_table = new_reftable;
3083 update_max_refcount_table_index(s);
3085 s->refcount_bits = 1 << refcount_order;
3086 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
3087 s->refcount_max += s->refcount_max - 1;
3089 s->refcount_block_bits = s->cluster_bits - (refcount_order - 3);
3090 s->refcount_block_size = 1 << s->refcount_block_bits;
3092 s->get_refcount = new_get_refcount;
3093 s->set_refcount = new_set_refcount;
3095 /* For cleaning up all old refblocks and the old reftable below the "done"
3096 * label */
3097 new_reftable = old_reftable;
3098 new_reftable_size = old_reftable_size;
3099 new_reftable_offset = old_reftable_offset;
3101 done:
3102 if (new_reftable) {
3103 /* On success, new_reftable actually points to the old reftable (and
3104 * new_reftable_size is the old reftable's size); but that is just
3105 * fine */
3106 for (i = 0; i < new_reftable_size; i++) {
3107 uint64_t offset = new_reftable[i] & REFT_OFFSET_MASK;
3108 if (offset) {
3109 qcow2_free_clusters(bs, offset, s->cluster_size,
3110 QCOW2_DISCARD_OTHER);
3113 g_free(new_reftable);
3115 if (new_reftable_offset > 0) {
3116 qcow2_free_clusters(bs, new_reftable_offset,
3117 new_reftable_size * sizeof(uint64_t),
3118 QCOW2_DISCARD_OTHER);
3122 qemu_vfree(new_refblock);
3123 return ret;
3126 static int64_t get_refblock_offset(BlockDriverState *bs, uint64_t offset)
3128 BDRVQcow2State *s = bs->opaque;
3129 uint32_t index = offset_to_reftable_index(s, offset);
3130 int64_t covering_refblock_offset = 0;
3132 if (index < s->refcount_table_size) {
3133 covering_refblock_offset = s->refcount_table[index] & REFT_OFFSET_MASK;
3135 if (!covering_refblock_offset) {
3136 qcow2_signal_corruption(bs, true, -1, -1, "Refblock at %#" PRIx64 " is "
3137 "not covered by the refcount structures",
3138 offset);
3139 return -EIO;
3142 return covering_refblock_offset;
3145 static int qcow2_discard_refcount_block(BlockDriverState *bs,
3146 uint64_t discard_block_offs)
3148 BDRVQcow2State *s = bs->opaque;
3149 int64_t refblock_offs;
3150 uint64_t cluster_index = discard_block_offs >> s->cluster_bits;
3151 uint32_t block_index = cluster_index & (s->refcount_block_size - 1);
3152 void *refblock;
3153 int ret;
3155 refblock_offs = get_refblock_offset(bs, discard_block_offs);
3156 if (refblock_offs < 0) {
3157 return refblock_offs;
3160 assert(discard_block_offs != 0);
3162 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs,
3163 &refblock);
3164 if (ret < 0) {
3165 return ret;
3168 if (s->get_refcount(refblock, block_index) != 1) {
3169 qcow2_signal_corruption(bs, true, -1, -1, "Invalid refcount:"
3170 " refblock offset %#" PRIx64
3171 ", reftable index %u"
3172 ", block offset %#" PRIx64
3173 ", refcount %#" PRIx64,
3174 refblock_offs,
3175 offset_to_reftable_index(s, discard_block_offs),
3176 discard_block_offs,
3177 s->get_refcount(refblock, block_index));
3178 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
3179 return -EINVAL;
3181 s->set_refcount(refblock, block_index, 0);
3183 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache, refblock);
3185 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
3187 if (cluster_index < s->free_cluster_index) {
3188 s->free_cluster_index = cluster_index;
3191 refblock = qcow2_cache_is_table_offset(bs, s->refcount_block_cache,
3192 discard_block_offs);
3193 if (refblock) {
3194 /* discard refblock from the cache if refblock is cached */
3195 qcow2_cache_discard(bs, s->refcount_block_cache, refblock);
3197 update_refcount_discard(bs, discard_block_offs, s->cluster_size);
3199 return 0;
3202 int qcow2_shrink_reftable(BlockDriverState *bs)
3204 BDRVQcow2State *s = bs->opaque;
3205 uint64_t *reftable_tmp =
3206 g_malloc(s->refcount_table_size * sizeof(uint64_t));
3207 int i, ret;
3209 for (i = 0; i < s->refcount_table_size; i++) {
3210 int64_t refblock_offs = s->refcount_table[i] & REFT_OFFSET_MASK;
3211 void *refblock;
3212 bool unused_block;
3214 if (refblock_offs == 0) {
3215 reftable_tmp[i] = 0;
3216 continue;
3218 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs,
3219 &refblock);
3220 if (ret < 0) {
3221 goto out;
3224 /* the refblock has own reference */
3225 if (i == offset_to_reftable_index(s, refblock_offs)) {
3226 uint64_t block_index = (refblock_offs >> s->cluster_bits) &
3227 (s->refcount_block_size - 1);
3228 uint64_t refcount = s->get_refcount(refblock, block_index);
3230 s->set_refcount(refblock, block_index, 0);
3232 unused_block = buffer_is_zero(refblock, s->cluster_size);
3234 s->set_refcount(refblock, block_index, refcount);
3235 } else {
3236 unused_block = buffer_is_zero(refblock, s->cluster_size);
3238 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
3240 reftable_tmp[i] = unused_block ? 0 : cpu_to_be64(s->refcount_table[i]);
3243 ret = bdrv_pwrite_sync(bs->file, s->refcount_table_offset, reftable_tmp,
3244 s->refcount_table_size * sizeof(uint64_t));
3246 * If the write in the reftable failed the image may contain a partially
3247 * overwritten reftable. In this case it would be better to clear the
3248 * reftable in memory to avoid possible image corruption.
3250 for (i = 0; i < s->refcount_table_size; i++) {
3251 if (s->refcount_table[i] && !reftable_tmp[i]) {
3252 if (ret == 0) {
3253 ret = qcow2_discard_refcount_block(bs, s->refcount_table[i] &
3254 REFT_OFFSET_MASK);
3256 s->refcount_table[i] = 0;
3260 if (!s->cache_discards) {
3261 qcow2_process_discards(bs, ret);
3264 out:
3265 g_free(reftable_tmp);
3266 return ret;
3269 int64_t qcow2_get_last_cluster(BlockDriverState *bs, int64_t size)
3271 BDRVQcow2State *s = bs->opaque;
3272 int64_t i;
3274 for (i = size_to_clusters(s, size) - 1; i >= 0; i--) {
3275 uint64_t refcount;
3276 int ret = qcow2_get_refcount(bs, i, &refcount);
3277 if (ret < 0) {
3278 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
3279 i, strerror(-ret));
3280 return ret;
3282 if (refcount > 0) {
3283 return i;
3286 qcow2_signal_corruption(bs, true, -1, -1,
3287 "There are no references in the refcount table.");
3288 return -EIO;