qapi: Support downstream events and commands
[qemu.git] / block / qcow2-refcount.c
blobf47260b808faae26fe8c7b29fe71d7657a0ed98f
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-common.h"
26 #include "block/block_int.h"
27 #include "block/qcow2.h"
28 #include "qemu/range.h"
30 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size);
31 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
32 int64_t offset, int64_t length, uint64_t addend,
33 bool decrease, enum qcow2_discard_type type);
35 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index);
36 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index);
37 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index);
38 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index);
39 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index);
40 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index);
41 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index);
43 static void set_refcount_ro0(void *refcount_array, uint64_t index,
44 uint64_t value);
45 static void set_refcount_ro1(void *refcount_array, uint64_t index,
46 uint64_t value);
47 static void set_refcount_ro2(void *refcount_array, uint64_t index,
48 uint64_t value);
49 static void set_refcount_ro3(void *refcount_array, uint64_t index,
50 uint64_t value);
51 static void set_refcount_ro4(void *refcount_array, uint64_t index,
52 uint64_t value);
53 static void set_refcount_ro5(void *refcount_array, uint64_t index,
54 uint64_t value);
55 static void set_refcount_ro6(void *refcount_array, uint64_t index,
56 uint64_t value);
59 static Qcow2GetRefcountFunc *const get_refcount_funcs[] = {
60 &get_refcount_ro0,
61 &get_refcount_ro1,
62 &get_refcount_ro2,
63 &get_refcount_ro3,
64 &get_refcount_ro4,
65 &get_refcount_ro5,
66 &get_refcount_ro6
69 static Qcow2SetRefcountFunc *const set_refcount_funcs[] = {
70 &set_refcount_ro0,
71 &set_refcount_ro1,
72 &set_refcount_ro2,
73 &set_refcount_ro3,
74 &set_refcount_ro4,
75 &set_refcount_ro5,
76 &set_refcount_ro6
80 /*********************************************************/
81 /* refcount handling */
83 int qcow2_refcount_init(BlockDriverState *bs)
85 BDRVQcowState *s = bs->opaque;
86 unsigned int refcount_table_size2, i;
87 int ret;
89 assert(s->refcount_order >= 0 && s->refcount_order <= 6);
91 s->get_refcount = get_refcount_funcs[s->refcount_order];
92 s->set_refcount = set_refcount_funcs[s->refcount_order];
94 assert(s->refcount_table_size <= INT_MAX / sizeof(uint64_t));
95 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
96 s->refcount_table = g_try_malloc(refcount_table_size2);
98 if (s->refcount_table_size > 0) {
99 if (s->refcount_table == NULL) {
100 ret = -ENOMEM;
101 goto fail;
103 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
104 ret = bdrv_pread(bs->file, s->refcount_table_offset,
105 s->refcount_table, refcount_table_size2);
106 if (ret < 0) {
107 goto fail;
109 for(i = 0; i < s->refcount_table_size; i++)
110 be64_to_cpus(&s->refcount_table[i]);
112 return 0;
113 fail:
114 return ret;
117 void qcow2_refcount_close(BlockDriverState *bs)
119 BDRVQcowState *s = bs->opaque;
120 g_free(s->refcount_table);
124 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index)
126 return (((const uint8_t *)refcount_array)[index / 8] >> (index % 8)) & 0x1;
129 static void set_refcount_ro0(void *refcount_array, uint64_t index,
130 uint64_t value)
132 assert(!(value >> 1));
133 ((uint8_t *)refcount_array)[index / 8] &= ~(0x1 << (index % 8));
134 ((uint8_t *)refcount_array)[index / 8] |= value << (index % 8);
137 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index)
139 return (((const uint8_t *)refcount_array)[index / 4] >> (2 * (index % 4)))
140 & 0x3;
143 static void set_refcount_ro1(void *refcount_array, uint64_t index,
144 uint64_t value)
146 assert(!(value >> 2));
147 ((uint8_t *)refcount_array)[index / 4] &= ~(0x3 << (2 * (index % 4)));
148 ((uint8_t *)refcount_array)[index / 4] |= value << (2 * (index % 4));
151 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index)
153 return (((const uint8_t *)refcount_array)[index / 2] >> (4 * (index % 2)))
154 & 0xf;
157 static void set_refcount_ro2(void *refcount_array, uint64_t index,
158 uint64_t value)
160 assert(!(value >> 4));
161 ((uint8_t *)refcount_array)[index / 2] &= ~(0xf << (4 * (index % 2)));
162 ((uint8_t *)refcount_array)[index / 2] |= value << (4 * (index % 2));
165 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index)
167 return ((const uint8_t *)refcount_array)[index];
170 static void set_refcount_ro3(void *refcount_array, uint64_t index,
171 uint64_t value)
173 assert(!(value >> 8));
174 ((uint8_t *)refcount_array)[index] = value;
177 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index)
179 return be16_to_cpu(((const uint16_t *)refcount_array)[index]);
182 static void set_refcount_ro4(void *refcount_array, uint64_t index,
183 uint64_t value)
185 assert(!(value >> 16));
186 ((uint16_t *)refcount_array)[index] = cpu_to_be16(value);
189 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index)
191 return be32_to_cpu(((const uint32_t *)refcount_array)[index]);
194 static void set_refcount_ro5(void *refcount_array, uint64_t index,
195 uint64_t value)
197 assert(!(value >> 32));
198 ((uint32_t *)refcount_array)[index] = cpu_to_be32(value);
201 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index)
203 return be64_to_cpu(((const uint64_t *)refcount_array)[index]);
206 static void set_refcount_ro6(void *refcount_array, uint64_t index,
207 uint64_t value)
209 ((uint64_t *)refcount_array)[index] = cpu_to_be64(value);
213 static int load_refcount_block(BlockDriverState *bs,
214 int64_t refcount_block_offset,
215 void **refcount_block)
217 BDRVQcowState *s = bs->opaque;
218 int ret;
220 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
221 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
222 refcount_block);
224 return ret;
228 * Retrieves the refcount of the cluster given by its index and stores it in
229 * *refcount. Returns 0 on success and -errno on failure.
231 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index,
232 uint64_t *refcount)
234 BDRVQcowState *s = bs->opaque;
235 uint64_t refcount_table_index, block_index;
236 int64_t refcount_block_offset;
237 int ret;
238 void *refcount_block;
240 refcount_table_index = cluster_index >> s->refcount_block_bits;
241 if (refcount_table_index >= s->refcount_table_size) {
242 *refcount = 0;
243 return 0;
245 refcount_block_offset =
246 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
247 if (!refcount_block_offset) {
248 *refcount = 0;
249 return 0;
252 if (offset_into_cluster(s, refcount_block_offset)) {
253 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" PRIx64
254 " unaligned (reftable index: %#" PRIx64 ")",
255 refcount_block_offset, refcount_table_index);
256 return -EIO;
259 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
260 &refcount_block);
261 if (ret < 0) {
262 return ret;
265 block_index = cluster_index & (s->refcount_block_size - 1);
266 *refcount = s->get_refcount(refcount_block, block_index);
268 ret = qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
269 if (ret < 0) {
270 return ret;
273 return 0;
277 * Rounds the refcount table size up to avoid growing the table for each single
278 * refcount block that is allocated.
280 static unsigned int next_refcount_table_size(BDRVQcowState *s,
281 unsigned int min_size)
283 unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1;
284 unsigned int refcount_table_clusters =
285 MAX(1, s->refcount_table_size >> (s->cluster_bits - 3));
287 while (min_clusters > refcount_table_clusters) {
288 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
291 return refcount_table_clusters << (s->cluster_bits - 3);
295 /* Checks if two offsets are described by the same refcount block */
296 static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a,
297 uint64_t offset_b)
299 uint64_t block_a = offset_a >> (s->cluster_bits + s->refcount_block_bits);
300 uint64_t block_b = offset_b >> (s->cluster_bits + s->refcount_block_bits);
302 return (block_a == block_b);
306 * Loads a refcount block. If it doesn't exist yet, it is allocated first
307 * (including growing the refcount table if needed).
309 * Returns 0 on success or -errno in error case
311 static int alloc_refcount_block(BlockDriverState *bs,
312 int64_t cluster_index, void **refcount_block)
314 BDRVQcowState *s = bs->opaque;
315 unsigned int refcount_table_index;
316 int ret;
318 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
320 /* Find the refcount block for the given cluster */
321 refcount_table_index = cluster_index >> s->refcount_block_bits;
323 if (refcount_table_index < s->refcount_table_size) {
325 uint64_t refcount_block_offset =
326 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
328 /* If it's already there, we're done */
329 if (refcount_block_offset) {
330 if (offset_into_cluster(s, refcount_block_offset)) {
331 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
332 PRIx64 " unaligned (reftable index: "
333 "%#x)", refcount_block_offset,
334 refcount_table_index);
335 return -EIO;
338 return load_refcount_block(bs, refcount_block_offset,
339 refcount_block);
344 * If we came here, we need to allocate something. Something is at least
345 * a cluster for the new refcount block. It may also include a new refcount
346 * table if the old refcount table is too small.
348 * Note that allocating clusters here needs some special care:
350 * - We can't use the normal qcow2_alloc_clusters(), it would try to
351 * increase the refcount and very likely we would end up with an endless
352 * recursion. Instead we must place the refcount blocks in a way that
353 * they can describe them themselves.
355 * - We need to consider that at this point we are inside update_refcounts
356 * and potentially doing an initial refcount increase. This means that
357 * some clusters have already been allocated by the caller, but their
358 * refcount isn't accurate yet. If we allocate clusters for metadata, we
359 * need to return -EAGAIN to signal the caller that it needs to restart
360 * the search for free clusters.
362 * - alloc_clusters_noref and qcow2_free_clusters may load a different
363 * refcount block into the cache
366 *refcount_block = NULL;
368 /* We write to the refcount table, so we might depend on L2 tables */
369 ret = qcow2_cache_flush(bs, s->l2_table_cache);
370 if (ret < 0) {
371 return ret;
374 /* Allocate the refcount block itself and mark it as used */
375 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
376 if (new_block < 0) {
377 return new_block;
380 #ifdef DEBUG_ALLOC2
381 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
382 " at %" PRIx64 "\n",
383 refcount_table_index, cluster_index << s->cluster_bits, new_block);
384 #endif
386 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
387 /* Zero the new refcount block before updating it */
388 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
389 refcount_block);
390 if (ret < 0) {
391 goto fail_block;
394 memset(*refcount_block, 0, s->cluster_size);
396 /* The block describes itself, need to update the cache */
397 int block_index = (new_block >> s->cluster_bits) &
398 (s->refcount_block_size - 1);
399 s->set_refcount(*refcount_block, block_index, 1);
400 } else {
401 /* Described somewhere else. This can recurse at most twice before we
402 * arrive at a block that describes itself. */
403 ret = update_refcount(bs, new_block, s->cluster_size, 1, false,
404 QCOW2_DISCARD_NEVER);
405 if (ret < 0) {
406 goto fail_block;
409 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
410 if (ret < 0) {
411 goto fail_block;
414 /* Initialize the new refcount block only after updating its refcount,
415 * update_refcount uses the refcount cache itself */
416 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
417 refcount_block);
418 if (ret < 0) {
419 goto fail_block;
422 memset(*refcount_block, 0, s->cluster_size);
425 /* Now the new refcount block needs to be written to disk */
426 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
427 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block);
428 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
429 if (ret < 0) {
430 goto fail_block;
433 /* If the refcount table is big enough, just hook the block up there */
434 if (refcount_table_index < s->refcount_table_size) {
435 uint64_t data64 = cpu_to_be64(new_block);
436 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
437 ret = bdrv_pwrite_sync(bs->file,
438 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
439 &data64, sizeof(data64));
440 if (ret < 0) {
441 goto fail_block;
444 s->refcount_table[refcount_table_index] = new_block;
446 /* The new refcount block may be where the caller intended to put its
447 * data, so let it restart the search. */
448 return -EAGAIN;
451 ret = qcow2_cache_put(bs, s->refcount_block_cache, refcount_block);
452 if (ret < 0) {
453 goto fail_block;
457 * If we come here, we need to grow the refcount table. Again, a new
458 * refcount table needs some space and we can't simply allocate to avoid
459 * endless recursion.
461 * Therefore let's grab new refcount blocks at the end of the image, which
462 * will describe themselves and the new refcount table. This way we can
463 * reference them only in the new table and do the switch to the new
464 * refcount table at once without producing an inconsistent state in
465 * between.
467 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
469 /* Calculate the number of refcount blocks needed so far; this will be the
470 * basis for calculating the index of the first cluster used for the
471 * self-describing refcount structures which we are about to create.
473 * Because we reached this point, there cannot be any refcount entries for
474 * cluster_index or higher indices yet. However, because new_block has been
475 * allocated to describe that cluster (and it will assume this role later
476 * on), we cannot use that index; also, new_block may actually have a higher
477 * cluster index than cluster_index, so it needs to be taken into account
478 * here (and 1 needs to be added to its value because that cluster is used).
480 uint64_t blocks_used = DIV_ROUND_UP(MAX(cluster_index + 1,
481 (new_block >> s->cluster_bits) + 1),
482 s->refcount_block_size);
484 if (blocks_used > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) {
485 return -EFBIG;
488 /* And now we need at least one block more for the new metadata */
489 uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
490 uint64_t last_table_size;
491 uint64_t blocks_clusters;
492 do {
493 uint64_t table_clusters =
494 size_to_clusters(s, table_size * sizeof(uint64_t));
495 blocks_clusters = 1 +
496 ((table_clusters + s->refcount_block_size - 1)
497 / s->refcount_block_size);
498 uint64_t meta_clusters = table_clusters + blocks_clusters;
500 last_table_size = table_size;
501 table_size = next_refcount_table_size(s, blocks_used +
502 ((meta_clusters + s->refcount_block_size - 1)
503 / s->refcount_block_size));
505 } while (last_table_size != table_size);
507 #ifdef DEBUG_ALLOC2
508 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
509 s->refcount_table_size, table_size);
510 #endif
512 /* Create the new refcount table and blocks */
513 uint64_t meta_offset = (blocks_used * s->refcount_block_size) *
514 s->cluster_size;
515 uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
516 uint64_t *new_table = g_try_new0(uint64_t, table_size);
517 void *new_blocks = g_try_malloc0(blocks_clusters * s->cluster_size);
519 assert(table_size > 0 && blocks_clusters > 0);
520 if (new_table == NULL || new_blocks == NULL) {
521 ret = -ENOMEM;
522 goto fail_table;
525 /* Fill the new refcount table */
526 memcpy(new_table, s->refcount_table,
527 s->refcount_table_size * sizeof(uint64_t));
528 new_table[refcount_table_index] = new_block;
530 int i;
531 for (i = 0; i < blocks_clusters; i++) {
532 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
535 /* Fill the refcount blocks */
536 uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
537 int block = 0;
538 for (i = 0; i < table_clusters + blocks_clusters; i++) {
539 s->set_refcount(new_blocks, block++, 1);
542 /* Write refcount blocks to disk */
543 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
544 ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
545 blocks_clusters * s->cluster_size);
546 g_free(new_blocks);
547 new_blocks = NULL;
548 if (ret < 0) {
549 goto fail_table;
552 /* Write refcount table to disk */
553 for(i = 0; i < table_size; i++) {
554 cpu_to_be64s(&new_table[i]);
557 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
558 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
559 table_size * sizeof(uint64_t));
560 if (ret < 0) {
561 goto fail_table;
564 for(i = 0; i < table_size; i++) {
565 be64_to_cpus(&new_table[i]);
568 /* Hook up the new refcount table in the qcow2 header */
569 uint8_t data[12];
570 cpu_to_be64w((uint64_t*)data, table_offset);
571 cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
572 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
573 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
574 data, sizeof(data));
575 if (ret < 0) {
576 goto fail_table;
579 /* And switch it in memory */
580 uint64_t old_table_offset = s->refcount_table_offset;
581 uint64_t old_table_size = s->refcount_table_size;
583 g_free(s->refcount_table);
584 s->refcount_table = new_table;
585 s->refcount_table_size = table_size;
586 s->refcount_table_offset = table_offset;
588 /* Free old table. */
589 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t),
590 QCOW2_DISCARD_OTHER);
592 ret = load_refcount_block(bs, new_block, refcount_block);
593 if (ret < 0) {
594 return ret;
597 /* If we were trying to do the initial refcount update for some cluster
598 * allocation, we might have used the same clusters to store newly
599 * allocated metadata. Make the caller search some new space. */
600 return -EAGAIN;
602 fail_table:
603 g_free(new_blocks);
604 g_free(new_table);
605 fail_block:
606 if (*refcount_block != NULL) {
607 qcow2_cache_put(bs, s->refcount_block_cache, refcount_block);
609 return ret;
612 void qcow2_process_discards(BlockDriverState *bs, int ret)
614 BDRVQcowState *s = bs->opaque;
615 Qcow2DiscardRegion *d, *next;
617 QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) {
618 QTAILQ_REMOVE(&s->discards, d, next);
620 /* Discard is optional, ignore the return value */
621 if (ret >= 0) {
622 bdrv_discard(bs->file,
623 d->offset >> BDRV_SECTOR_BITS,
624 d->bytes >> BDRV_SECTOR_BITS);
627 g_free(d);
631 static void update_refcount_discard(BlockDriverState *bs,
632 uint64_t offset, uint64_t length)
634 BDRVQcowState *s = bs->opaque;
635 Qcow2DiscardRegion *d, *p, *next;
637 QTAILQ_FOREACH(d, &s->discards, next) {
638 uint64_t new_start = MIN(offset, d->offset);
639 uint64_t new_end = MAX(offset + length, d->offset + d->bytes);
641 if (new_end - new_start <= length + d->bytes) {
642 /* There can't be any overlap, areas ending up here have no
643 * references any more and therefore shouldn't get freed another
644 * time. */
645 assert(d->bytes + length == new_end - new_start);
646 d->offset = new_start;
647 d->bytes = new_end - new_start;
648 goto found;
652 d = g_malloc(sizeof(*d));
653 *d = (Qcow2DiscardRegion) {
654 .bs = bs,
655 .offset = offset,
656 .bytes = length,
658 QTAILQ_INSERT_TAIL(&s->discards, d, next);
660 found:
661 /* Merge discard requests if they are adjacent now */
662 QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) {
663 if (p == d
664 || p->offset > d->offset + d->bytes
665 || d->offset > p->offset + p->bytes)
667 continue;
670 /* Still no overlap possible */
671 assert(p->offset == d->offset + d->bytes
672 || d->offset == p->offset + p->bytes);
674 QTAILQ_REMOVE(&s->discards, p, next);
675 d->offset = MIN(d->offset, p->offset);
676 d->bytes += p->bytes;
677 g_free(p);
681 /* XXX: cache several refcount block clusters ? */
682 /* @addend is the absolute value of the addend; if @decrease is set, @addend
683 * will be subtracted from the current refcount, otherwise it will be added */
684 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
685 int64_t offset,
686 int64_t length,
687 uint64_t addend,
688 bool decrease,
689 enum qcow2_discard_type type)
691 BDRVQcowState *s = bs->opaque;
692 int64_t start, last, cluster_offset;
693 void *refcount_block = NULL;
694 int64_t old_table_index = -1;
695 int ret;
697 #ifdef DEBUG_ALLOC2
698 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64
699 " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "",
700 addend);
701 #endif
702 if (length < 0) {
703 return -EINVAL;
704 } else if (length == 0) {
705 return 0;
708 if (decrease) {
709 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
710 s->l2_table_cache);
713 start = start_of_cluster(s, offset);
714 last = start_of_cluster(s, offset + length - 1);
715 for(cluster_offset = start; cluster_offset <= last;
716 cluster_offset += s->cluster_size)
718 int block_index;
719 uint64_t refcount;
720 int64_t cluster_index = cluster_offset >> s->cluster_bits;
721 int64_t table_index = cluster_index >> s->refcount_block_bits;
723 /* Load the refcount block and allocate it if needed */
724 if (table_index != old_table_index) {
725 if (refcount_block) {
726 ret = qcow2_cache_put(bs, s->refcount_block_cache,
727 &refcount_block);
728 if (ret < 0) {
729 goto fail;
733 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
734 if (ret < 0) {
735 goto fail;
738 old_table_index = table_index;
740 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
742 /* we can update the count and save it */
743 block_index = cluster_index & (s->refcount_block_size - 1);
745 refcount = s->get_refcount(refcount_block, block_index);
746 if (decrease ? (refcount - addend > refcount)
747 : (refcount + addend < refcount ||
748 refcount + addend > s->refcount_max))
750 ret = -EINVAL;
751 goto fail;
753 if (decrease) {
754 refcount -= addend;
755 } else {
756 refcount += addend;
758 if (refcount == 0 && cluster_index < s->free_cluster_index) {
759 s->free_cluster_index = cluster_index;
761 s->set_refcount(refcount_block, block_index, refcount);
763 if (refcount == 0 && s->discard_passthrough[type]) {
764 update_refcount_discard(bs, cluster_offset, s->cluster_size);
768 ret = 0;
769 fail:
770 if (!s->cache_discards) {
771 qcow2_process_discards(bs, ret);
774 /* Write last changed block to disk */
775 if (refcount_block) {
776 int wret;
777 wret = qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
778 if (wret < 0) {
779 return ret < 0 ? ret : wret;
784 * Try do undo any updates if an error is returned (This may succeed in
785 * some cases like ENOSPC for allocating a new refcount block)
787 if (ret < 0) {
788 int dummy;
789 dummy = update_refcount(bs, offset, cluster_offset - offset, addend,
790 !decrease, QCOW2_DISCARD_NEVER);
791 (void)dummy;
794 return ret;
798 * Increases or decreases the refcount of a given cluster.
800 * @addend is the absolute value of the addend; if @decrease is set, @addend
801 * will be subtracted from the current refcount, otherwise it will be added.
803 * On success 0 is returned; on failure -errno is returned.
805 int qcow2_update_cluster_refcount(BlockDriverState *bs,
806 int64_t cluster_index,
807 uint64_t addend, bool decrease,
808 enum qcow2_discard_type type)
810 BDRVQcowState *s = bs->opaque;
811 int ret;
813 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend,
814 decrease, type);
815 if (ret < 0) {
816 return ret;
819 return 0;
824 /*********************************************************/
825 /* cluster allocation functions */
829 /* return < 0 if error */
830 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size)
832 BDRVQcowState *s = bs->opaque;
833 uint64_t i, nb_clusters, refcount;
834 int ret;
836 nb_clusters = size_to_clusters(s, size);
837 retry:
838 for(i = 0; i < nb_clusters; i++) {
839 uint64_t next_cluster_index = s->free_cluster_index++;
840 ret = qcow2_get_refcount(bs, next_cluster_index, &refcount);
842 if (ret < 0) {
843 return ret;
844 } else if (refcount != 0) {
845 goto retry;
849 /* Make sure that all offsets in the "allocated" range are representable
850 * in an int64_t */
851 if (s->free_cluster_index > 0 &&
852 s->free_cluster_index - 1 > (INT64_MAX >> s->cluster_bits))
854 return -EFBIG;
857 #ifdef DEBUG_ALLOC2
858 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
859 size,
860 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
861 #endif
862 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
865 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size)
867 int64_t offset;
868 int ret;
870 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
871 do {
872 offset = alloc_clusters_noref(bs, size);
873 if (offset < 0) {
874 return offset;
877 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
878 } while (ret == -EAGAIN);
880 if (ret < 0) {
881 return ret;
884 return offset;
887 int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
888 int nb_clusters)
890 BDRVQcowState *s = bs->opaque;
891 uint64_t cluster_index, refcount;
892 uint64_t i;
893 int ret;
895 assert(nb_clusters >= 0);
896 if (nb_clusters == 0) {
897 return 0;
900 do {
901 /* Check how many clusters there are free */
902 cluster_index = offset >> s->cluster_bits;
903 for(i = 0; i < nb_clusters; i++) {
904 ret = qcow2_get_refcount(bs, cluster_index++, &refcount);
905 if (ret < 0) {
906 return ret;
907 } else if (refcount != 0) {
908 break;
912 /* And then allocate them */
913 ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false,
914 QCOW2_DISCARD_NEVER);
915 } while (ret == -EAGAIN);
917 if (ret < 0) {
918 return ret;
921 return i;
924 /* only used to allocate compressed sectors. We try to allocate
925 contiguous sectors. size must be <= cluster_size */
926 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
928 BDRVQcowState *s = bs->opaque;
929 int64_t offset;
930 size_t free_in_cluster;
931 int ret;
933 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
934 assert(size > 0 && size <= s->cluster_size);
935 assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset));
937 offset = s->free_byte_offset;
939 if (offset) {
940 uint64_t refcount;
941 ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount);
942 if (ret < 0) {
943 return ret;
946 if (refcount == s->refcount_max) {
947 offset = 0;
951 free_in_cluster = s->cluster_size - offset_into_cluster(s, offset);
952 if (!offset || free_in_cluster < size) {
953 int64_t new_cluster = alloc_clusters_noref(bs, s->cluster_size);
954 if (new_cluster < 0) {
955 return new_cluster;
958 if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) {
959 offset = new_cluster;
963 assert(offset);
964 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
965 if (ret < 0) {
966 return ret;
969 /* The cluster refcount was incremented; refcount blocks must be flushed
970 * before the caller's L2 table updates. */
971 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
973 s->free_byte_offset = offset + size;
974 if (!offset_into_cluster(s, s->free_byte_offset)) {
975 s->free_byte_offset = 0;
978 return offset;
981 void qcow2_free_clusters(BlockDriverState *bs,
982 int64_t offset, int64_t size,
983 enum qcow2_discard_type type)
985 int ret;
987 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
988 ret = update_refcount(bs, offset, size, 1, true, type);
989 if (ret < 0) {
990 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
991 /* TODO Remember the clusters to free them later and avoid leaking */
996 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
997 * normal cluster, compressed cluster, etc.)
999 void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
1000 int nb_clusters, enum qcow2_discard_type type)
1002 BDRVQcowState *s = bs->opaque;
1004 switch (qcow2_get_cluster_type(l2_entry)) {
1005 case QCOW2_CLUSTER_COMPRESSED:
1007 int nb_csectors;
1008 nb_csectors = ((l2_entry >> s->csize_shift) &
1009 s->csize_mask) + 1;
1010 qcow2_free_clusters(bs,
1011 (l2_entry & s->cluster_offset_mask) & ~511,
1012 nb_csectors * 512, type);
1014 break;
1015 case QCOW2_CLUSTER_NORMAL:
1016 case QCOW2_CLUSTER_ZERO:
1017 if (l2_entry & L2E_OFFSET_MASK) {
1018 if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) {
1019 qcow2_signal_corruption(bs, false, -1, -1,
1020 "Cannot free unaligned cluster %#llx",
1021 l2_entry & L2E_OFFSET_MASK);
1022 } else {
1023 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
1024 nb_clusters << s->cluster_bits, type);
1027 break;
1028 case QCOW2_CLUSTER_UNALLOCATED:
1029 break;
1030 default:
1031 abort();
1037 /*********************************************************/
1038 /* snapshots and image creation */
1042 /* update the refcounts of snapshots and the copied flag */
1043 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
1044 int64_t l1_table_offset, int l1_size, int addend)
1046 BDRVQcowState *s = bs->opaque;
1047 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, refcount;
1048 bool l1_allocated = false;
1049 int64_t old_offset, old_l2_offset;
1050 int i, j, l1_modified = 0, nb_csectors;
1051 int ret;
1053 assert(addend >= -1 && addend <= 1);
1055 l2_table = NULL;
1056 l1_table = NULL;
1057 l1_size2 = l1_size * sizeof(uint64_t);
1059 s->cache_discards = true;
1061 /* WARNING: qcow2_snapshot_goto relies on this function not using the
1062 * l1_table_offset when it is the current s->l1_table_offset! Be careful
1063 * when changing this! */
1064 if (l1_table_offset != s->l1_table_offset) {
1065 l1_table = g_try_malloc0(align_offset(l1_size2, 512));
1066 if (l1_size2 && l1_table == NULL) {
1067 ret = -ENOMEM;
1068 goto fail;
1070 l1_allocated = true;
1072 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
1073 if (ret < 0) {
1074 goto fail;
1077 for(i = 0;i < l1_size; i++)
1078 be64_to_cpus(&l1_table[i]);
1079 } else {
1080 assert(l1_size == s->l1_size);
1081 l1_table = s->l1_table;
1082 l1_allocated = false;
1085 for(i = 0; i < l1_size; i++) {
1086 l2_offset = l1_table[i];
1087 if (l2_offset) {
1088 old_l2_offset = l2_offset;
1089 l2_offset &= L1E_OFFSET_MASK;
1091 if (offset_into_cluster(s, l2_offset)) {
1092 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1093 PRIx64 " unaligned (L1 index: %#x)",
1094 l2_offset, i);
1095 ret = -EIO;
1096 goto fail;
1099 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1100 (void**) &l2_table);
1101 if (ret < 0) {
1102 goto fail;
1105 for(j = 0; j < s->l2_size; j++) {
1106 uint64_t cluster_index;
1108 offset = be64_to_cpu(l2_table[j]);
1109 old_offset = offset;
1110 offset &= ~QCOW_OFLAG_COPIED;
1112 switch (qcow2_get_cluster_type(offset)) {
1113 case QCOW2_CLUSTER_COMPRESSED:
1114 nb_csectors = ((offset >> s->csize_shift) &
1115 s->csize_mask) + 1;
1116 if (addend != 0) {
1117 ret = update_refcount(bs,
1118 (offset & s->cluster_offset_mask) & ~511,
1119 nb_csectors * 512, abs(addend), addend < 0,
1120 QCOW2_DISCARD_SNAPSHOT);
1121 if (ret < 0) {
1122 goto fail;
1125 /* compressed clusters are never modified */
1126 refcount = 2;
1127 break;
1129 case QCOW2_CLUSTER_NORMAL:
1130 case QCOW2_CLUSTER_ZERO:
1131 if (offset_into_cluster(s, offset & L2E_OFFSET_MASK)) {
1132 qcow2_signal_corruption(bs, true, -1, -1, "Data "
1133 "cluster offset %#llx "
1134 "unaligned (L2 offset: %#"
1135 PRIx64 ", L2 index: %#x)",
1136 offset & L2E_OFFSET_MASK,
1137 l2_offset, j);
1138 ret = -EIO;
1139 goto fail;
1142 cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
1143 if (!cluster_index) {
1144 /* unallocated */
1145 refcount = 0;
1146 break;
1148 if (addend != 0) {
1149 ret = qcow2_update_cluster_refcount(bs,
1150 cluster_index, abs(addend), addend < 0,
1151 QCOW2_DISCARD_SNAPSHOT);
1152 if (ret < 0) {
1153 goto fail;
1157 ret = qcow2_get_refcount(bs, cluster_index, &refcount);
1158 if (ret < 0) {
1159 goto fail;
1161 break;
1163 case QCOW2_CLUSTER_UNALLOCATED:
1164 refcount = 0;
1165 break;
1167 default:
1168 abort();
1171 if (refcount == 1) {
1172 offset |= QCOW_OFLAG_COPIED;
1174 if (offset != old_offset) {
1175 if (addend > 0) {
1176 qcow2_cache_set_dependency(bs, s->l2_table_cache,
1177 s->refcount_block_cache);
1179 l2_table[j] = cpu_to_be64(offset);
1180 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
1184 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1185 if (ret < 0) {
1186 goto fail;
1190 if (addend != 0) {
1191 ret = qcow2_update_cluster_refcount(bs, l2_offset >>
1192 s->cluster_bits,
1193 abs(addend), addend < 0,
1194 QCOW2_DISCARD_SNAPSHOT);
1195 if (ret < 0) {
1196 goto fail;
1199 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1200 &refcount);
1201 if (ret < 0) {
1202 goto fail;
1203 } else if (refcount == 1) {
1204 l2_offset |= QCOW_OFLAG_COPIED;
1206 if (l2_offset != old_l2_offset) {
1207 l1_table[i] = l2_offset;
1208 l1_modified = 1;
1213 ret = bdrv_flush(bs);
1214 fail:
1215 if (l2_table) {
1216 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1219 s->cache_discards = false;
1220 qcow2_process_discards(bs, ret);
1222 /* Update L1 only if it isn't deleted anyway (addend = -1) */
1223 if (ret == 0 && addend >= 0 && l1_modified) {
1224 for (i = 0; i < l1_size; i++) {
1225 cpu_to_be64s(&l1_table[i]);
1228 ret = bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table, l1_size2);
1230 for (i = 0; i < l1_size; i++) {
1231 be64_to_cpus(&l1_table[i]);
1234 if (l1_allocated)
1235 g_free(l1_table);
1236 return ret;
1242 /*********************************************************/
1243 /* refcount checking functions */
1246 static size_t refcount_array_byte_size(BDRVQcowState *s, uint64_t entries)
1248 /* This assertion holds because there is no way we can address more than
1249 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because
1250 * offsets have to be representable in bytes); due to every cluster
1251 * corresponding to one refcount entry, we are well below that limit */
1252 assert(entries < (UINT64_C(1) << (64 - 9)));
1254 /* Thanks to the assertion this will not overflow, because
1255 * s->refcount_order < 7.
1256 * (note: x << s->refcount_order == x * s->refcount_bits) */
1257 return DIV_ROUND_UP(entries << s->refcount_order, 8);
1261 * Reallocates *array so that it can hold new_size entries. *size must contain
1262 * the current number of entries in *array. If the reallocation fails, *array
1263 * and *size will not be modified and -errno will be returned. If the
1264 * reallocation is successful, *array will be set to the new buffer, *size
1265 * will be set to new_size and 0 will be returned. The size of the reallocated
1266 * refcount array buffer will be aligned to a cluster boundary, and the newly
1267 * allocated area will be zeroed.
1269 static int realloc_refcount_array(BDRVQcowState *s, void **array,
1270 int64_t *size, int64_t new_size)
1272 size_t old_byte_size, new_byte_size;
1273 void *new_ptr;
1275 /* Round to clusters so the array can be directly written to disk */
1276 old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size))
1277 * s->cluster_size;
1278 new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size))
1279 * s->cluster_size;
1281 if (new_byte_size == old_byte_size) {
1282 *size = new_size;
1283 return 0;
1286 assert(new_byte_size > 0);
1288 new_ptr = g_try_realloc(*array, new_byte_size);
1289 if (!new_ptr) {
1290 return -ENOMEM;
1293 if (new_byte_size > old_byte_size) {
1294 memset((void *)((uintptr_t)new_ptr + old_byte_size), 0,
1295 new_byte_size - old_byte_size);
1298 *array = new_ptr;
1299 *size = new_size;
1301 return 0;
1305 * Increases the refcount for a range of clusters in a given refcount table.
1306 * This is used to construct a temporary refcount table out of L1 and L2 tables
1307 * which can be compared the the refcount table saved in the image.
1309 * Modifies the number of errors in res.
1311 static int inc_refcounts(BlockDriverState *bs,
1312 BdrvCheckResult *res,
1313 void **refcount_table,
1314 int64_t *refcount_table_size,
1315 int64_t offset, int64_t size)
1317 BDRVQcowState *s = bs->opaque;
1318 uint64_t start, last, cluster_offset, k, refcount;
1319 int ret;
1321 if (size <= 0) {
1322 return 0;
1325 start = start_of_cluster(s, offset);
1326 last = start_of_cluster(s, offset + size - 1);
1327 for(cluster_offset = start; cluster_offset <= last;
1328 cluster_offset += s->cluster_size) {
1329 k = cluster_offset >> s->cluster_bits;
1330 if (k >= *refcount_table_size) {
1331 ret = realloc_refcount_array(s, refcount_table,
1332 refcount_table_size, k + 1);
1333 if (ret < 0) {
1334 res->check_errors++;
1335 return ret;
1339 refcount = s->get_refcount(*refcount_table, k);
1340 if (refcount == s->refcount_max) {
1341 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
1342 "\n", cluster_offset);
1343 res->corruptions++;
1344 continue;
1346 s->set_refcount(*refcount_table, k, refcount + 1);
1349 return 0;
1352 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
1353 enum {
1354 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
1358 * Increases the refcount in the given refcount table for the all clusters
1359 * referenced in the L2 table. While doing so, performs some checks on L2
1360 * entries.
1362 * Returns the number of errors found by the checks or -errno if an internal
1363 * error occurred.
1365 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
1366 void **refcount_table,
1367 int64_t *refcount_table_size, int64_t l2_offset,
1368 int flags)
1370 BDRVQcowState *s = bs->opaque;
1371 uint64_t *l2_table, l2_entry;
1372 uint64_t next_contiguous_offset = 0;
1373 int i, l2_size, nb_csectors, ret;
1375 /* Read L2 table from disk */
1376 l2_size = s->l2_size * sizeof(uint64_t);
1377 l2_table = g_malloc(l2_size);
1379 ret = bdrv_pread(bs->file, l2_offset, l2_table, l2_size);
1380 if (ret < 0) {
1381 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1382 res->check_errors++;
1383 goto fail;
1386 /* Do the actual checks */
1387 for(i = 0; i < s->l2_size; i++) {
1388 l2_entry = be64_to_cpu(l2_table[i]);
1390 switch (qcow2_get_cluster_type(l2_entry)) {
1391 case QCOW2_CLUSTER_COMPRESSED:
1392 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1393 if (l2_entry & QCOW_OFLAG_COPIED) {
1394 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
1395 "copied flag must never be set for compressed "
1396 "clusters\n", l2_entry >> s->cluster_bits);
1397 l2_entry &= ~QCOW_OFLAG_COPIED;
1398 res->corruptions++;
1401 /* Mark cluster as used */
1402 nb_csectors = ((l2_entry >> s->csize_shift) &
1403 s->csize_mask) + 1;
1404 l2_entry &= s->cluster_offset_mask;
1405 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1406 l2_entry & ~511, nb_csectors * 512);
1407 if (ret < 0) {
1408 goto fail;
1411 if (flags & CHECK_FRAG_INFO) {
1412 res->bfi.allocated_clusters++;
1413 res->bfi.compressed_clusters++;
1415 /* Compressed clusters are fragmented by nature. Since they
1416 * take up sub-sector space but we only have sector granularity
1417 * I/O we need to re-read the same sectors even for adjacent
1418 * compressed clusters.
1420 res->bfi.fragmented_clusters++;
1422 break;
1424 case QCOW2_CLUSTER_ZERO:
1425 if ((l2_entry & L2E_OFFSET_MASK) == 0) {
1426 break;
1428 /* fall through */
1430 case QCOW2_CLUSTER_NORMAL:
1432 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
1434 if (flags & CHECK_FRAG_INFO) {
1435 res->bfi.allocated_clusters++;
1436 if (next_contiguous_offset &&
1437 offset != next_contiguous_offset) {
1438 res->bfi.fragmented_clusters++;
1440 next_contiguous_offset = offset + s->cluster_size;
1443 /* Mark cluster as used */
1444 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1445 offset, s->cluster_size);
1446 if (ret < 0) {
1447 goto fail;
1450 /* Correct offsets are cluster aligned */
1451 if (offset_into_cluster(s, offset)) {
1452 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
1453 "properly aligned; L2 entry corrupted.\n", offset);
1454 res->corruptions++;
1456 break;
1459 case QCOW2_CLUSTER_UNALLOCATED:
1460 break;
1462 default:
1463 abort();
1467 g_free(l2_table);
1468 return 0;
1470 fail:
1471 g_free(l2_table);
1472 return ret;
1476 * Increases the refcount for the L1 table, its L2 tables and all referenced
1477 * clusters in the given refcount table. While doing so, performs some checks
1478 * on L1 and L2 entries.
1480 * Returns the number of errors found by the checks or -errno if an internal
1481 * error occurred.
1483 static int check_refcounts_l1(BlockDriverState *bs,
1484 BdrvCheckResult *res,
1485 void **refcount_table,
1486 int64_t *refcount_table_size,
1487 int64_t l1_table_offset, int l1_size,
1488 int flags)
1490 BDRVQcowState *s = bs->opaque;
1491 uint64_t *l1_table = NULL, l2_offset, l1_size2;
1492 int i, ret;
1494 l1_size2 = l1_size * sizeof(uint64_t);
1496 /* Mark L1 table as used */
1497 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1498 l1_table_offset, l1_size2);
1499 if (ret < 0) {
1500 goto fail;
1503 /* Read L1 table entries from disk */
1504 if (l1_size2 > 0) {
1505 l1_table = g_try_malloc(l1_size2);
1506 if (l1_table == NULL) {
1507 ret = -ENOMEM;
1508 res->check_errors++;
1509 goto fail;
1511 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
1512 if (ret < 0) {
1513 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1514 res->check_errors++;
1515 goto fail;
1517 for(i = 0;i < l1_size; i++)
1518 be64_to_cpus(&l1_table[i]);
1521 /* Do the actual checks */
1522 for(i = 0; i < l1_size; i++) {
1523 l2_offset = l1_table[i];
1524 if (l2_offset) {
1525 /* Mark L2 table as used */
1526 l2_offset &= L1E_OFFSET_MASK;
1527 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1528 l2_offset, s->cluster_size);
1529 if (ret < 0) {
1530 goto fail;
1533 /* L2 tables are cluster aligned */
1534 if (offset_into_cluster(s, l2_offset)) {
1535 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1536 "cluster aligned; L1 entry corrupted\n", l2_offset);
1537 res->corruptions++;
1540 /* Process and check L2 entries */
1541 ret = check_refcounts_l2(bs, res, refcount_table,
1542 refcount_table_size, l2_offset, flags);
1543 if (ret < 0) {
1544 goto fail;
1548 g_free(l1_table);
1549 return 0;
1551 fail:
1552 g_free(l1_table);
1553 return ret;
1557 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1559 * This function does not print an error message nor does it increment
1560 * check_errors if qcow2_get_refcount fails (this is because such an error will
1561 * have been already detected and sufficiently signaled by the calling function
1562 * (qcow2_check_refcounts) by the time this function is called).
1564 static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
1565 BdrvCheckMode fix)
1567 BDRVQcowState *s = bs->opaque;
1568 uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size);
1569 int ret;
1570 uint64_t refcount;
1571 int i, j;
1573 for (i = 0; i < s->l1_size; i++) {
1574 uint64_t l1_entry = s->l1_table[i];
1575 uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK;
1576 bool l2_dirty = false;
1578 if (!l2_offset) {
1579 continue;
1582 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1583 &refcount);
1584 if (ret < 0) {
1585 /* don't print message nor increment check_errors */
1586 continue;
1588 if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) {
1589 fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1590 "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
1591 fix & BDRV_FIX_ERRORS ? "Repairing" :
1592 "ERROR",
1593 i, l1_entry, refcount);
1594 if (fix & BDRV_FIX_ERRORS) {
1595 s->l1_table[i] = refcount == 1
1596 ? l1_entry | QCOW_OFLAG_COPIED
1597 : l1_entry & ~QCOW_OFLAG_COPIED;
1598 ret = qcow2_write_l1_entry(bs, i);
1599 if (ret < 0) {
1600 res->check_errors++;
1601 goto fail;
1603 res->corruptions_fixed++;
1604 } else {
1605 res->corruptions++;
1609 ret = bdrv_pread(bs->file, l2_offset, l2_table,
1610 s->l2_size * sizeof(uint64_t));
1611 if (ret < 0) {
1612 fprintf(stderr, "ERROR: Could not read L2 table: %s\n",
1613 strerror(-ret));
1614 res->check_errors++;
1615 goto fail;
1618 for (j = 0; j < s->l2_size; j++) {
1619 uint64_t l2_entry = be64_to_cpu(l2_table[j]);
1620 uint64_t data_offset = l2_entry & L2E_OFFSET_MASK;
1621 int cluster_type = qcow2_get_cluster_type(l2_entry);
1623 if ((cluster_type == QCOW2_CLUSTER_NORMAL) ||
1624 ((cluster_type == QCOW2_CLUSTER_ZERO) && (data_offset != 0))) {
1625 ret = qcow2_get_refcount(bs,
1626 data_offset >> s->cluster_bits,
1627 &refcount);
1628 if (ret < 0) {
1629 /* don't print message nor increment check_errors */
1630 continue;
1632 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
1633 fprintf(stderr, "%s OFLAG_COPIED data cluster: "
1634 "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
1635 fix & BDRV_FIX_ERRORS ? "Repairing" :
1636 "ERROR",
1637 l2_entry, refcount);
1638 if (fix & BDRV_FIX_ERRORS) {
1639 l2_table[j] = cpu_to_be64(refcount == 1
1640 ? l2_entry | QCOW_OFLAG_COPIED
1641 : l2_entry & ~QCOW_OFLAG_COPIED);
1642 l2_dirty = true;
1643 res->corruptions_fixed++;
1644 } else {
1645 res->corruptions++;
1651 if (l2_dirty) {
1652 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
1653 l2_offset, s->cluster_size);
1654 if (ret < 0) {
1655 fprintf(stderr, "ERROR: Could not write L2 table; metadata "
1656 "overlap check failed: %s\n", strerror(-ret));
1657 res->check_errors++;
1658 goto fail;
1661 ret = bdrv_pwrite(bs->file, l2_offset, l2_table, s->cluster_size);
1662 if (ret < 0) {
1663 fprintf(stderr, "ERROR: Could not write L2 table: %s\n",
1664 strerror(-ret));
1665 res->check_errors++;
1666 goto fail;
1671 ret = 0;
1673 fail:
1674 qemu_vfree(l2_table);
1675 return ret;
1679 * Checks consistency of refblocks and accounts for each refblock in
1680 * *refcount_table.
1682 static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res,
1683 BdrvCheckMode fix, bool *rebuild,
1684 void **refcount_table, int64_t *nb_clusters)
1686 BDRVQcowState *s = bs->opaque;
1687 int64_t i, size;
1688 int ret;
1690 for(i = 0; i < s->refcount_table_size; i++) {
1691 uint64_t offset, cluster;
1692 offset = s->refcount_table[i];
1693 cluster = offset >> s->cluster_bits;
1695 /* Refcount blocks are cluster aligned */
1696 if (offset_into_cluster(s, offset)) {
1697 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1698 "cluster aligned; refcount table entry corrupted\n", i);
1699 res->corruptions++;
1700 *rebuild = true;
1701 continue;
1704 if (cluster >= *nb_clusters) {
1705 fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n",
1706 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
1708 if (fix & BDRV_FIX_ERRORS) {
1709 int64_t new_nb_clusters;
1711 if (offset > INT64_MAX - s->cluster_size) {
1712 ret = -EINVAL;
1713 goto resize_fail;
1716 ret = bdrv_truncate(bs->file, offset + s->cluster_size);
1717 if (ret < 0) {
1718 goto resize_fail;
1720 size = bdrv_getlength(bs->file);
1721 if (size < 0) {
1722 ret = size;
1723 goto resize_fail;
1726 new_nb_clusters = size_to_clusters(s, size);
1727 assert(new_nb_clusters >= *nb_clusters);
1729 ret = realloc_refcount_array(s, refcount_table,
1730 nb_clusters, new_nb_clusters);
1731 if (ret < 0) {
1732 res->check_errors++;
1733 return ret;
1736 if (cluster >= *nb_clusters) {
1737 ret = -EINVAL;
1738 goto resize_fail;
1741 res->corruptions_fixed++;
1742 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1743 offset, s->cluster_size);
1744 if (ret < 0) {
1745 return ret;
1747 /* No need to check whether the refcount is now greater than 1:
1748 * This area was just allocated and zeroed, so it can only be
1749 * exactly 1 after inc_refcounts() */
1750 continue;
1752 resize_fail:
1753 res->corruptions++;
1754 *rebuild = true;
1755 fprintf(stderr, "ERROR could not resize image: %s\n",
1756 strerror(-ret));
1757 } else {
1758 res->corruptions++;
1760 continue;
1763 if (offset != 0) {
1764 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1765 offset, s->cluster_size);
1766 if (ret < 0) {
1767 return ret;
1769 if (s->get_refcount(*refcount_table, cluster) != 1) {
1770 fprintf(stderr, "ERROR refcount block %" PRId64
1771 " refcount=%" PRIu64 "\n", i,
1772 s->get_refcount(*refcount_table, cluster));
1773 res->corruptions++;
1774 *rebuild = true;
1779 return 0;
1783 * Calculates an in-memory refcount table.
1785 static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1786 BdrvCheckMode fix, bool *rebuild,
1787 void **refcount_table, int64_t *nb_clusters)
1789 BDRVQcowState *s = bs->opaque;
1790 int64_t i;
1791 QCowSnapshot *sn;
1792 int ret;
1794 if (!*refcount_table) {
1795 int64_t old_size = 0;
1796 ret = realloc_refcount_array(s, refcount_table,
1797 &old_size, *nb_clusters);
1798 if (ret < 0) {
1799 res->check_errors++;
1800 return ret;
1804 /* header */
1805 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1806 0, s->cluster_size);
1807 if (ret < 0) {
1808 return ret;
1811 /* current L1 table */
1812 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1813 s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO);
1814 if (ret < 0) {
1815 return ret;
1818 /* snapshots */
1819 for (i = 0; i < s->nb_snapshots; i++) {
1820 sn = s->snapshots + i;
1821 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1822 sn->l1_table_offset, sn->l1_size, 0);
1823 if (ret < 0) {
1824 return ret;
1827 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1828 s->snapshots_offset, s->snapshots_size);
1829 if (ret < 0) {
1830 return ret;
1833 /* refcount data */
1834 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1835 s->refcount_table_offset,
1836 s->refcount_table_size * sizeof(uint64_t));
1837 if (ret < 0) {
1838 return ret;
1841 return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters);
1845 * Compares the actual reference count for each cluster in the image against the
1846 * refcount as reported by the refcount structures on-disk.
1848 static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1849 BdrvCheckMode fix, bool *rebuild,
1850 int64_t *highest_cluster,
1851 void *refcount_table, int64_t nb_clusters)
1853 BDRVQcowState *s = bs->opaque;
1854 int64_t i;
1855 uint64_t refcount1, refcount2;
1856 int ret;
1858 for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) {
1859 ret = qcow2_get_refcount(bs, i, &refcount1);
1860 if (ret < 0) {
1861 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
1862 i, strerror(-ret));
1863 res->check_errors++;
1864 continue;
1867 refcount2 = s->get_refcount(refcount_table, i);
1869 if (refcount1 > 0 || refcount2 > 0) {
1870 *highest_cluster = i;
1873 if (refcount1 != refcount2) {
1874 /* Check if we're allowed to fix the mismatch */
1875 int *num_fixed = NULL;
1876 if (refcount1 == 0) {
1877 *rebuild = true;
1878 } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
1879 num_fixed = &res->leaks_fixed;
1880 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
1881 num_fixed = &res->corruptions_fixed;
1884 fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64
1885 " reference=%" PRIu64 "\n",
1886 num_fixed != NULL ? "Repairing" :
1887 refcount1 < refcount2 ? "ERROR" :
1888 "Leaked",
1889 i, refcount1, refcount2);
1891 if (num_fixed) {
1892 ret = update_refcount(bs, i << s->cluster_bits, 1,
1893 refcount_diff(refcount1, refcount2),
1894 refcount1 > refcount2,
1895 QCOW2_DISCARD_ALWAYS);
1896 if (ret >= 0) {
1897 (*num_fixed)++;
1898 continue;
1902 /* And if we couldn't, print an error */
1903 if (refcount1 < refcount2) {
1904 res->corruptions++;
1905 } else {
1906 res->leaks++;
1913 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to
1914 * the on-disk refcount structures.
1916 * On input, *first_free_cluster tells where to start looking, and need not
1917 * actually be a free cluster; the returned offset will not be before that
1918 * cluster. On output, *first_free_cluster points to the first gap found, even
1919 * if that gap was too small to be used as the returned offset.
1921 * Note that *first_free_cluster is a cluster index whereas the return value is
1922 * an offset.
1924 static int64_t alloc_clusters_imrt(BlockDriverState *bs,
1925 int cluster_count,
1926 void **refcount_table,
1927 int64_t *imrt_nb_clusters,
1928 int64_t *first_free_cluster)
1930 BDRVQcowState *s = bs->opaque;
1931 int64_t cluster = *first_free_cluster, i;
1932 bool first_gap = true;
1933 int contiguous_free_clusters;
1934 int ret;
1936 /* Starting at *first_free_cluster, find a range of at least cluster_count
1937 * continuously free clusters */
1938 for (contiguous_free_clusters = 0;
1939 cluster < *imrt_nb_clusters &&
1940 contiguous_free_clusters < cluster_count;
1941 cluster++)
1943 if (!s->get_refcount(*refcount_table, cluster)) {
1944 contiguous_free_clusters++;
1945 if (first_gap) {
1946 /* If this is the first free cluster found, update
1947 * *first_free_cluster accordingly */
1948 *first_free_cluster = cluster;
1949 first_gap = false;
1951 } else if (contiguous_free_clusters) {
1952 contiguous_free_clusters = 0;
1956 /* If contiguous_free_clusters is greater than zero, it contains the number
1957 * of continuously free clusters until the current cluster; the first free
1958 * cluster in the current "gap" is therefore
1959 * cluster - contiguous_free_clusters */
1961 /* If no such range could be found, grow the in-memory refcount table
1962 * accordingly to append free clusters at the end of the image */
1963 if (contiguous_free_clusters < cluster_count) {
1964 /* contiguous_free_clusters clusters are already empty at the image end;
1965 * we need cluster_count clusters; therefore, we have to allocate
1966 * cluster_count - contiguous_free_clusters new clusters at the end of
1967 * the image (which is the current value of cluster; note that cluster
1968 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond
1969 * the image end) */
1970 ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters,
1971 cluster + cluster_count
1972 - contiguous_free_clusters);
1973 if (ret < 0) {
1974 return ret;
1978 /* Go back to the first free cluster */
1979 cluster -= contiguous_free_clusters;
1980 for (i = 0; i < cluster_count; i++) {
1981 s->set_refcount(*refcount_table, cluster + i, 1);
1984 return cluster << s->cluster_bits;
1988 * Creates a new refcount structure based solely on the in-memory information
1989 * given through *refcount_table. All necessary allocations will be reflected
1990 * in that array.
1992 * On success, the old refcount structure is leaked (it will be covered by the
1993 * new refcount structure).
1995 static int rebuild_refcount_structure(BlockDriverState *bs,
1996 BdrvCheckResult *res,
1997 void **refcount_table,
1998 int64_t *nb_clusters)
2000 BDRVQcowState *s = bs->opaque;
2001 int64_t first_free_cluster = 0, reftable_offset = -1, cluster = 0;
2002 int64_t refblock_offset, refblock_start, refblock_index;
2003 uint32_t reftable_size = 0;
2004 uint64_t *on_disk_reftable = NULL;
2005 void *on_disk_refblock;
2006 int ret = 0;
2007 struct {
2008 uint64_t reftable_offset;
2009 uint32_t reftable_clusters;
2010 } QEMU_PACKED reftable_offset_and_clusters;
2012 qcow2_cache_empty(bs, s->refcount_block_cache);
2014 write_refblocks:
2015 for (; cluster < *nb_clusters; cluster++) {
2016 if (!s->get_refcount(*refcount_table, cluster)) {
2017 continue;
2020 refblock_index = cluster >> s->refcount_block_bits;
2021 refblock_start = refblock_index << s->refcount_block_bits;
2023 /* Don't allocate a cluster in a refblock already written to disk */
2024 if (first_free_cluster < refblock_start) {
2025 first_free_cluster = refblock_start;
2027 refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table,
2028 nb_clusters, &first_free_cluster);
2029 if (refblock_offset < 0) {
2030 fprintf(stderr, "ERROR allocating refblock: %s\n",
2031 strerror(-refblock_offset));
2032 res->check_errors++;
2033 ret = refblock_offset;
2034 goto fail;
2037 if (reftable_size <= refblock_index) {
2038 uint32_t old_reftable_size = reftable_size;
2039 uint64_t *new_on_disk_reftable;
2041 reftable_size = ROUND_UP((refblock_index + 1) * sizeof(uint64_t),
2042 s->cluster_size) / sizeof(uint64_t);
2043 new_on_disk_reftable = g_try_realloc(on_disk_reftable,
2044 reftable_size *
2045 sizeof(uint64_t));
2046 if (!new_on_disk_reftable) {
2047 res->check_errors++;
2048 ret = -ENOMEM;
2049 goto fail;
2051 on_disk_reftable = new_on_disk_reftable;
2053 memset(on_disk_reftable + old_reftable_size, 0,
2054 (reftable_size - old_reftable_size) * sizeof(uint64_t));
2056 /* The offset we have for the reftable is now no longer valid;
2057 * this will leak that range, but we can easily fix that by running
2058 * a leak-fixing check after this rebuild operation */
2059 reftable_offset = -1;
2061 on_disk_reftable[refblock_index] = refblock_offset;
2063 /* If this is apparently the last refblock (for now), try to squeeze the
2064 * reftable in */
2065 if (refblock_index == (*nb_clusters - 1) >> s->refcount_block_bits &&
2066 reftable_offset < 0)
2068 uint64_t reftable_clusters = size_to_clusters(s, reftable_size *
2069 sizeof(uint64_t));
2070 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2071 refcount_table, nb_clusters,
2072 &first_free_cluster);
2073 if (reftable_offset < 0) {
2074 fprintf(stderr, "ERROR allocating reftable: %s\n",
2075 strerror(-reftable_offset));
2076 res->check_errors++;
2077 ret = reftable_offset;
2078 goto fail;
2082 ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset,
2083 s->cluster_size);
2084 if (ret < 0) {
2085 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2086 goto fail;
2089 /* The size of *refcount_table is always cluster-aligned, therefore the
2090 * write operation will not overflow */
2091 on_disk_refblock = (void *)((char *) *refcount_table +
2092 refblock_index * s->cluster_size);
2094 ret = bdrv_write(bs->file, refblock_offset / BDRV_SECTOR_SIZE,
2095 on_disk_refblock, s->cluster_sectors);
2096 if (ret < 0) {
2097 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2098 goto fail;
2101 /* Go to the end of this refblock */
2102 cluster = refblock_start + s->refcount_block_size - 1;
2105 if (reftable_offset < 0) {
2106 uint64_t post_refblock_start, reftable_clusters;
2108 post_refblock_start = ROUND_UP(*nb_clusters, s->refcount_block_size);
2109 reftable_clusters = size_to_clusters(s,
2110 reftable_size * sizeof(uint64_t));
2111 /* Not pretty but simple */
2112 if (first_free_cluster < post_refblock_start) {
2113 first_free_cluster = post_refblock_start;
2115 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2116 refcount_table, nb_clusters,
2117 &first_free_cluster);
2118 if (reftable_offset < 0) {
2119 fprintf(stderr, "ERROR allocating reftable: %s\n",
2120 strerror(-reftable_offset));
2121 res->check_errors++;
2122 ret = reftable_offset;
2123 goto fail;
2126 goto write_refblocks;
2129 assert(on_disk_reftable);
2131 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2132 cpu_to_be64s(&on_disk_reftable[refblock_index]);
2135 ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset,
2136 reftable_size * sizeof(uint64_t));
2137 if (ret < 0) {
2138 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2139 goto fail;
2142 assert(reftable_size < INT_MAX / sizeof(uint64_t));
2143 ret = bdrv_pwrite(bs->file, reftable_offset, on_disk_reftable,
2144 reftable_size * sizeof(uint64_t));
2145 if (ret < 0) {
2146 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2147 goto fail;
2150 /* Enter new reftable into the image header */
2151 cpu_to_be64w(&reftable_offset_and_clusters.reftable_offset,
2152 reftable_offset);
2153 cpu_to_be32w(&reftable_offset_and_clusters.reftable_clusters,
2154 size_to_clusters(s, reftable_size * sizeof(uint64_t)));
2155 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader,
2156 refcount_table_offset),
2157 &reftable_offset_and_clusters,
2158 sizeof(reftable_offset_and_clusters));
2159 if (ret < 0) {
2160 fprintf(stderr, "ERROR setting reftable: %s\n", strerror(-ret));
2161 goto fail;
2164 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2165 be64_to_cpus(&on_disk_reftable[refblock_index]);
2167 s->refcount_table = on_disk_reftable;
2168 s->refcount_table_offset = reftable_offset;
2169 s->refcount_table_size = reftable_size;
2171 return 0;
2173 fail:
2174 g_free(on_disk_reftable);
2175 return ret;
2179 * Checks an image for refcount consistency.
2181 * Returns 0 if no errors are found, the number of errors in case the image is
2182 * detected as corrupted, and -errno when an internal error occurred.
2184 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2185 BdrvCheckMode fix)
2187 BDRVQcowState *s = bs->opaque;
2188 BdrvCheckResult pre_compare_res;
2189 int64_t size, highest_cluster, nb_clusters;
2190 void *refcount_table = NULL;
2191 bool rebuild = false;
2192 int ret;
2194 size = bdrv_getlength(bs->file);
2195 if (size < 0) {
2196 res->check_errors++;
2197 return size;
2200 nb_clusters = size_to_clusters(s, size);
2201 if (nb_clusters > INT_MAX) {
2202 res->check_errors++;
2203 return -EFBIG;
2206 res->bfi.total_clusters =
2207 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE);
2209 ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table,
2210 &nb_clusters);
2211 if (ret < 0) {
2212 goto fail;
2215 /* In case we don't need to rebuild the refcount structure (but want to fix
2216 * something), this function is immediately called again, in which case the
2217 * result should be ignored */
2218 pre_compare_res = *res;
2219 compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table,
2220 nb_clusters);
2222 if (rebuild && (fix & BDRV_FIX_ERRORS)) {
2223 BdrvCheckResult old_res = *res;
2224 int fresh_leaks = 0;
2226 fprintf(stderr, "Rebuilding refcount structure\n");
2227 ret = rebuild_refcount_structure(bs, res, &refcount_table,
2228 &nb_clusters);
2229 if (ret < 0) {
2230 goto fail;
2233 res->corruptions = 0;
2234 res->leaks = 0;
2236 /* Because the old reftable has been exchanged for a new one the
2237 * references have to be recalculated */
2238 rebuild = false;
2239 memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters));
2240 ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table,
2241 &nb_clusters);
2242 if (ret < 0) {
2243 goto fail;
2246 if (fix & BDRV_FIX_LEAKS) {
2247 /* The old refcount structures are now leaked, fix it; the result
2248 * can be ignored, aside from leaks which were introduced by
2249 * rebuild_refcount_structure() that could not be fixed */
2250 BdrvCheckResult saved_res = *res;
2251 *res = (BdrvCheckResult){ 0 };
2253 compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild,
2254 &highest_cluster, refcount_table, nb_clusters);
2255 if (rebuild) {
2256 fprintf(stderr, "ERROR rebuilt refcount structure is still "
2257 "broken\n");
2260 /* Any leaks accounted for here were introduced by
2261 * rebuild_refcount_structure() because that function has created a
2262 * new refcount structure from scratch */
2263 fresh_leaks = res->leaks;
2264 *res = saved_res;
2267 if (res->corruptions < old_res.corruptions) {
2268 res->corruptions_fixed += old_res.corruptions - res->corruptions;
2270 if (res->leaks < old_res.leaks) {
2271 res->leaks_fixed += old_res.leaks - res->leaks;
2273 res->leaks += fresh_leaks;
2274 } else if (fix) {
2275 if (rebuild) {
2276 fprintf(stderr, "ERROR need to rebuild refcount structures\n");
2277 res->check_errors++;
2278 ret = -EIO;
2279 goto fail;
2282 if (res->leaks || res->corruptions) {
2283 *res = pre_compare_res;
2284 compare_refcounts(bs, res, fix, &rebuild, &highest_cluster,
2285 refcount_table, nb_clusters);
2289 /* check OFLAG_COPIED */
2290 ret = check_oflag_copied(bs, res, fix);
2291 if (ret < 0) {
2292 goto fail;
2295 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
2296 ret = 0;
2298 fail:
2299 g_free(refcount_table);
2301 return ret;
2304 #define overlaps_with(ofs, sz) \
2305 ranges_overlap(offset, size, ofs, sz)
2308 * Checks if the given offset into the image file is actually free to use by
2309 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
2310 * i.e. a sanity check without relying on the refcount tables.
2312 * The ign parameter specifies what checks not to perform (being a bitmask of
2313 * QCow2MetadataOverlap values), i.e., what sections to ignore.
2315 * Returns:
2316 * - 0 if writing to this offset will not affect the mentioned metadata
2317 * - a positive QCow2MetadataOverlap value indicating one overlapping section
2318 * - a negative value (-errno) indicating an error while performing a check,
2319 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
2321 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset,
2322 int64_t size)
2324 BDRVQcowState *s = bs->opaque;
2325 int chk = s->overlap_check & ~ign;
2326 int i, j;
2328 if (!size) {
2329 return 0;
2332 if (chk & QCOW2_OL_MAIN_HEADER) {
2333 if (offset < s->cluster_size) {
2334 return QCOW2_OL_MAIN_HEADER;
2338 /* align range to test to cluster boundaries */
2339 size = align_offset(offset_into_cluster(s, offset) + size, s->cluster_size);
2340 offset = start_of_cluster(s, offset);
2342 if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) {
2343 if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) {
2344 return QCOW2_OL_ACTIVE_L1;
2348 if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) {
2349 if (overlaps_with(s->refcount_table_offset,
2350 s->refcount_table_size * sizeof(uint64_t))) {
2351 return QCOW2_OL_REFCOUNT_TABLE;
2355 if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) {
2356 if (overlaps_with(s->snapshots_offset, s->snapshots_size)) {
2357 return QCOW2_OL_SNAPSHOT_TABLE;
2361 if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) {
2362 for (i = 0; i < s->nb_snapshots; i++) {
2363 if (s->snapshots[i].l1_size &&
2364 overlaps_with(s->snapshots[i].l1_table_offset,
2365 s->snapshots[i].l1_size * sizeof(uint64_t))) {
2366 return QCOW2_OL_INACTIVE_L1;
2371 if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) {
2372 for (i = 0; i < s->l1_size; i++) {
2373 if ((s->l1_table[i] & L1E_OFFSET_MASK) &&
2374 overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK,
2375 s->cluster_size)) {
2376 return QCOW2_OL_ACTIVE_L2;
2381 if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) {
2382 for (i = 0; i < s->refcount_table_size; i++) {
2383 if ((s->refcount_table[i] & REFT_OFFSET_MASK) &&
2384 overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK,
2385 s->cluster_size)) {
2386 return QCOW2_OL_REFCOUNT_BLOCK;
2391 if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) {
2392 for (i = 0; i < s->nb_snapshots; i++) {
2393 uint64_t l1_ofs = s->snapshots[i].l1_table_offset;
2394 uint32_t l1_sz = s->snapshots[i].l1_size;
2395 uint64_t l1_sz2 = l1_sz * sizeof(uint64_t);
2396 uint64_t *l1 = g_try_malloc(l1_sz2);
2397 int ret;
2399 if (l1_sz2 && l1 == NULL) {
2400 return -ENOMEM;
2403 ret = bdrv_pread(bs->file, l1_ofs, l1, l1_sz2);
2404 if (ret < 0) {
2405 g_free(l1);
2406 return ret;
2409 for (j = 0; j < l1_sz; j++) {
2410 uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK;
2411 if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) {
2412 g_free(l1);
2413 return QCOW2_OL_INACTIVE_L2;
2417 g_free(l1);
2421 return 0;
2424 static const char *metadata_ol_names[] = {
2425 [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header",
2426 [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table",
2427 [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table",
2428 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table",
2429 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block",
2430 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table",
2431 [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table",
2432 [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table",
2436 * First performs a check for metadata overlaps (through
2437 * qcow2_check_metadata_overlap); if that fails with a negative value (error
2438 * while performing a check), that value is returned. If an impending overlap
2439 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
2440 * and -EIO returned.
2442 * Returns 0 if there were neither overlaps nor errors while checking for
2443 * overlaps; or a negative value (-errno) on error.
2445 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset,
2446 int64_t size)
2448 int ret = qcow2_check_metadata_overlap(bs, ign, offset, size);
2450 if (ret < 0) {
2451 return ret;
2452 } else if (ret > 0) {
2453 int metadata_ol_bitnr = ctz32(ret);
2454 assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR);
2456 qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid "
2457 "write on metadata (overlaps with %s)",
2458 metadata_ol_names[metadata_ol_bitnr]);
2459 return -EIO;
2462 return 0;