Merge remote-tracking branch 'awilliam/ipxe' into staging
[qemu.git] / block / qcow2-refcount.c
blob915d85acbfdf988cc79600a4d6319ab030648de8
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_int.h"
27 #include "block/qcow2.h"
29 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
30 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
31 int64_t offset, int64_t length,
32 int addend);
35 /*********************************************************/
36 /* refcount handling */
38 int qcow2_refcount_init(BlockDriverState *bs)
40 BDRVQcowState *s = bs->opaque;
41 int ret, refcount_table_size2, i;
43 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
44 s->refcount_table = qemu_malloc(refcount_table_size2);
45 if (s->refcount_table_size > 0) {
46 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
47 ret = bdrv_pread(bs->file, s->refcount_table_offset,
48 s->refcount_table, refcount_table_size2);
49 if (ret != refcount_table_size2)
50 goto fail;
51 for(i = 0; i < s->refcount_table_size; i++)
52 be64_to_cpus(&s->refcount_table[i]);
54 return 0;
55 fail:
56 return -ENOMEM;
59 void qcow2_refcount_close(BlockDriverState *bs)
61 BDRVQcowState *s = bs->opaque;
62 qemu_free(s->refcount_table);
66 static int load_refcount_block(BlockDriverState *bs,
67 int64_t refcount_block_offset,
68 void **refcount_block)
70 BDRVQcowState *s = bs->opaque;
71 int ret;
73 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
74 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
75 refcount_block);
77 return ret;
81 * Returns the refcount of the cluster given by its index. Any non-negative
82 * return value is the refcount of the cluster, negative values are -errno
83 * and indicate an error.
85 static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
87 BDRVQcowState *s = bs->opaque;
88 int refcount_table_index, block_index;
89 int64_t refcount_block_offset;
90 int ret;
91 uint16_t *refcount_block;
92 uint16_t refcount;
94 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
95 if (refcount_table_index >= s->refcount_table_size)
96 return 0;
97 refcount_block_offset = s->refcount_table[refcount_table_index];
98 if (!refcount_block_offset)
99 return 0;
101 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
102 (void**) &refcount_block);
103 if (ret < 0) {
104 return ret;
107 block_index = cluster_index &
108 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
109 refcount = be16_to_cpu(refcount_block[block_index]);
111 ret = qcow2_cache_put(bs, s->refcount_block_cache,
112 (void**) &refcount_block);
113 if (ret < 0) {
114 return ret;
117 return refcount;
121 * Rounds the refcount table size up to avoid growing the table for each single
122 * refcount block that is allocated.
124 static unsigned int next_refcount_table_size(BDRVQcowState *s,
125 unsigned int min_size)
127 unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1;
128 unsigned int refcount_table_clusters =
129 MAX(1, s->refcount_table_size >> (s->cluster_bits - 3));
131 while (min_clusters > refcount_table_clusters) {
132 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
135 return refcount_table_clusters << (s->cluster_bits - 3);
139 /* Checks if two offsets are described by the same refcount block */
140 static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a,
141 uint64_t offset_b)
143 uint64_t block_a = offset_a >> (2 * s->cluster_bits - REFCOUNT_SHIFT);
144 uint64_t block_b = offset_b >> (2 * s->cluster_bits - REFCOUNT_SHIFT);
146 return (block_a == block_b);
150 * Loads a refcount block. If it doesn't exist yet, it is allocated first
151 * (including growing the refcount table if needed).
153 * Returns 0 on success or -errno in error case
155 static int alloc_refcount_block(BlockDriverState *bs,
156 int64_t cluster_index, uint16_t **refcount_block)
158 BDRVQcowState *s = bs->opaque;
159 unsigned int refcount_table_index;
160 int ret;
162 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
164 /* Find the refcount block for the given cluster */
165 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
167 if (refcount_table_index < s->refcount_table_size) {
169 uint64_t refcount_block_offset =
170 s->refcount_table[refcount_table_index];
172 /* If it's already there, we're done */
173 if (refcount_block_offset) {
174 return load_refcount_block(bs, refcount_block_offset,
175 (void**) refcount_block);
180 * If we came here, we need to allocate something. Something is at least
181 * a cluster for the new refcount block. It may also include a new refcount
182 * table if the old refcount table is too small.
184 * Note that allocating clusters here needs some special care:
186 * - We can't use the normal qcow2_alloc_clusters(), it would try to
187 * increase the refcount and very likely we would end up with an endless
188 * recursion. Instead we must place the refcount blocks in a way that
189 * they can describe them themselves.
191 * - We need to consider that at this point we are inside update_refcounts
192 * and doing the initial refcount increase. This means that some clusters
193 * have already been allocated by the caller, but their refcount isn't
194 * accurate yet. free_cluster_index tells us where this allocation ends
195 * as long as we don't overwrite it by freeing clusters.
197 * - alloc_clusters_noref and qcow2_free_clusters may load a different
198 * refcount block into the cache
201 *refcount_block = NULL;
203 /* We write to the refcount table, so we might depend on L2 tables */
204 qcow2_cache_flush(bs, s->l2_table_cache);
206 /* Allocate the refcount block itself and mark it as used */
207 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
208 if (new_block < 0) {
209 return new_block;
212 #ifdef DEBUG_ALLOC2
213 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
214 " at %" PRIx64 "\n",
215 refcount_table_index, cluster_index << s->cluster_bits, new_block);
216 #endif
218 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
219 /* Zero the new refcount block before updating it */
220 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
221 (void**) refcount_block);
222 if (ret < 0) {
223 goto fail_block;
226 memset(*refcount_block, 0, s->cluster_size);
228 /* The block describes itself, need to update the cache */
229 int block_index = (new_block >> s->cluster_bits) &
230 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
231 (*refcount_block)[block_index] = cpu_to_be16(1);
232 } else {
233 /* Described somewhere else. This can recurse at most twice before we
234 * arrive at a block that describes itself. */
235 ret = update_refcount(bs, new_block, s->cluster_size, 1);
236 if (ret < 0) {
237 goto fail_block;
240 bdrv_flush(bs->file);
242 /* Initialize the new refcount block only after updating its refcount,
243 * update_refcount uses the refcount cache itself */
244 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
245 (void**) refcount_block);
246 if (ret < 0) {
247 goto fail_block;
250 memset(*refcount_block, 0, s->cluster_size);
253 /* Now the new refcount block needs to be written to disk */
254 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
255 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block);
256 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
257 if (ret < 0) {
258 goto fail_block;
261 /* If the refcount table is big enough, just hook the block up there */
262 if (refcount_table_index < s->refcount_table_size) {
263 uint64_t data64 = cpu_to_be64(new_block);
264 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
265 ret = bdrv_pwrite_sync(bs->file,
266 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
267 &data64, sizeof(data64));
268 if (ret < 0) {
269 goto fail_block;
272 s->refcount_table[refcount_table_index] = new_block;
273 return 0;
276 ret = qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
277 if (ret < 0) {
278 goto fail_block;
282 * If we come here, we need to grow the refcount table. Again, a new
283 * refcount table needs some space and we can't simply allocate to avoid
284 * endless recursion.
286 * Therefore let's grab new refcount blocks at the end of the image, which
287 * will describe themselves and the new refcount table. This way we can
288 * reference them only in the new table and do the switch to the new
289 * refcount table at once without producing an inconsistent state in
290 * between.
292 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
294 /* Calculate the number of refcount blocks needed so far */
295 uint64_t refcount_block_clusters = 1 << (s->cluster_bits - REFCOUNT_SHIFT);
296 uint64_t blocks_used = (s->free_cluster_index +
297 refcount_block_clusters - 1) / refcount_block_clusters;
299 /* And now we need at least one block more for the new metadata */
300 uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
301 uint64_t last_table_size;
302 uint64_t blocks_clusters;
303 do {
304 uint64_t table_clusters = size_to_clusters(s, table_size);
305 blocks_clusters = 1 +
306 ((table_clusters + refcount_block_clusters - 1)
307 / refcount_block_clusters);
308 uint64_t meta_clusters = table_clusters + blocks_clusters;
310 last_table_size = table_size;
311 table_size = next_refcount_table_size(s, blocks_used +
312 ((meta_clusters + refcount_block_clusters - 1)
313 / refcount_block_clusters));
315 } while (last_table_size != table_size);
317 #ifdef DEBUG_ALLOC2
318 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
319 s->refcount_table_size, table_size);
320 #endif
322 /* Create the new refcount table and blocks */
323 uint64_t meta_offset = (blocks_used * refcount_block_clusters) *
324 s->cluster_size;
325 uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
326 uint16_t *new_blocks = qemu_mallocz(blocks_clusters * s->cluster_size);
327 uint64_t *new_table = qemu_mallocz(table_size * sizeof(uint64_t));
329 assert(meta_offset >= (s->free_cluster_index * s->cluster_size));
331 /* Fill the new refcount table */
332 memcpy(new_table, s->refcount_table,
333 s->refcount_table_size * sizeof(uint64_t));
334 new_table[refcount_table_index] = new_block;
336 int i;
337 for (i = 0; i < blocks_clusters; i++) {
338 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
341 /* Fill the refcount blocks */
342 uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
343 int block = 0;
344 for (i = 0; i < table_clusters + blocks_clusters; i++) {
345 new_blocks[block++] = cpu_to_be16(1);
348 /* Write refcount blocks to disk */
349 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
350 ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
351 blocks_clusters * s->cluster_size);
352 qemu_free(new_blocks);
353 if (ret < 0) {
354 goto fail_table;
357 /* Write refcount table to disk */
358 for(i = 0; i < table_size; i++) {
359 cpu_to_be64s(&new_table[i]);
362 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
363 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
364 table_size * sizeof(uint64_t));
365 if (ret < 0) {
366 goto fail_table;
369 for(i = 0; i < table_size; i++) {
370 cpu_to_be64s(&new_table[i]);
373 /* Hook up the new refcount table in the qcow2 header */
374 uint8_t data[12];
375 cpu_to_be64w((uint64_t*)data, table_offset);
376 cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
377 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
378 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
379 data, sizeof(data));
380 if (ret < 0) {
381 goto fail_table;
384 /* And switch it in memory */
385 uint64_t old_table_offset = s->refcount_table_offset;
386 uint64_t old_table_size = s->refcount_table_size;
388 qemu_free(s->refcount_table);
389 s->refcount_table = new_table;
390 s->refcount_table_size = table_size;
391 s->refcount_table_offset = table_offset;
393 /* Free old table. Remember, we must not change free_cluster_index */
394 uint64_t old_free_cluster_index = s->free_cluster_index;
395 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
396 s->free_cluster_index = old_free_cluster_index;
398 ret = load_refcount_block(bs, new_block, (void**) refcount_block);
399 if (ret < 0) {
400 return ret;
403 return new_block;
405 fail_table:
406 qemu_free(new_table);
407 fail_block:
408 if (*refcount_block != NULL) {
409 qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
411 return ret;
414 /* XXX: cache several refcount block clusters ? */
415 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
416 int64_t offset, int64_t length, int addend)
418 BDRVQcowState *s = bs->opaque;
419 int64_t start, last, cluster_offset;
420 uint16_t *refcount_block = NULL;
421 int64_t old_table_index = -1;
422 int ret;
424 #ifdef DEBUG_ALLOC2
425 printf("update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
426 offset, length, addend);
427 #endif
428 if (length < 0) {
429 return -EINVAL;
430 } else if (length == 0) {
431 return 0;
434 if (addend < 0) {
435 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
436 s->l2_table_cache);
439 start = offset & ~(s->cluster_size - 1);
440 last = (offset + length - 1) & ~(s->cluster_size - 1);
441 for(cluster_offset = start; cluster_offset <= last;
442 cluster_offset += s->cluster_size)
444 int block_index, refcount;
445 int64_t cluster_index = cluster_offset >> s->cluster_bits;
446 int64_t table_index =
447 cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
449 /* Load the refcount block and allocate it if needed */
450 if (table_index != old_table_index) {
451 if (refcount_block) {
452 ret = qcow2_cache_put(bs, s->refcount_block_cache,
453 (void**) &refcount_block);
454 if (ret < 0) {
455 goto fail;
459 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
460 if (ret < 0) {
461 goto fail;
464 old_table_index = table_index;
466 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
468 /* we can update the count and save it */
469 block_index = cluster_index &
470 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
472 refcount = be16_to_cpu(refcount_block[block_index]);
473 refcount += addend;
474 if (refcount < 0 || refcount > 0xffff) {
475 ret = -EINVAL;
476 goto fail;
478 if (refcount == 0 && cluster_index < s->free_cluster_index) {
479 s->free_cluster_index = cluster_index;
481 refcount_block[block_index] = cpu_to_be16(refcount);
484 ret = 0;
485 fail:
486 /* Write last changed block to disk */
487 if (refcount_block) {
488 int wret;
489 wret = qcow2_cache_put(bs, s->refcount_block_cache,
490 (void**) &refcount_block);
491 if (wret < 0) {
492 return ret < 0 ? ret : wret;
497 * Try do undo any updates if an error is returned (This may succeed in
498 * some cases like ENOSPC for allocating a new refcount block)
500 if (ret < 0) {
501 int dummy;
502 dummy = update_refcount(bs, offset, cluster_offset - offset, -addend);
503 (void)dummy;
506 return ret;
510 * Increases or decreases the refcount of a given cluster by one.
511 * addend must be 1 or -1.
513 * If the return value is non-negative, it is the new refcount of the cluster.
514 * If it is negative, it is -errno and indicates an error.
516 static int update_cluster_refcount(BlockDriverState *bs,
517 int64_t cluster_index,
518 int addend)
520 BDRVQcowState *s = bs->opaque;
521 int ret;
523 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
524 if (ret < 0) {
525 return ret;
528 bdrv_flush(bs->file);
530 return get_refcount(bs, cluster_index);
535 /*********************************************************/
536 /* cluster allocation functions */
540 /* return < 0 if error */
541 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
543 BDRVQcowState *s = bs->opaque;
544 int i, nb_clusters, refcount;
546 nb_clusters = size_to_clusters(s, size);
547 retry:
548 for(i = 0; i < nb_clusters; i++) {
549 int64_t next_cluster_index = s->free_cluster_index++;
550 refcount = get_refcount(bs, next_cluster_index);
552 if (refcount < 0) {
553 return refcount;
554 } else if (refcount != 0) {
555 goto retry;
558 #ifdef DEBUG_ALLOC2
559 printf("alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
560 size,
561 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
562 #endif
563 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
566 int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
568 int64_t offset;
569 int ret;
571 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
572 offset = alloc_clusters_noref(bs, size);
573 if (offset < 0) {
574 return offset;
577 ret = update_refcount(bs, offset, size, 1);
578 if (ret < 0) {
579 return ret;
582 return offset;
585 /* only used to allocate compressed sectors. We try to allocate
586 contiguous sectors. size must be <= cluster_size */
587 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
589 BDRVQcowState *s = bs->opaque;
590 int64_t offset, cluster_offset;
591 int free_in_cluster;
593 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
594 assert(size > 0 && size <= s->cluster_size);
595 if (s->free_byte_offset == 0) {
596 s->free_byte_offset = qcow2_alloc_clusters(bs, s->cluster_size);
597 if (s->free_byte_offset < 0) {
598 return s->free_byte_offset;
601 redo:
602 free_in_cluster = s->cluster_size -
603 (s->free_byte_offset & (s->cluster_size - 1));
604 if (size <= free_in_cluster) {
605 /* enough space in current cluster */
606 offset = s->free_byte_offset;
607 s->free_byte_offset += size;
608 free_in_cluster -= size;
609 if (free_in_cluster == 0)
610 s->free_byte_offset = 0;
611 if ((offset & (s->cluster_size - 1)) != 0)
612 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
613 } else {
614 offset = qcow2_alloc_clusters(bs, s->cluster_size);
615 if (offset < 0) {
616 return offset;
618 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
619 if ((cluster_offset + s->cluster_size) == offset) {
620 /* we are lucky: contiguous data */
621 offset = s->free_byte_offset;
622 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
623 s->free_byte_offset += size;
624 } else {
625 s->free_byte_offset = offset;
626 goto redo;
630 bdrv_flush(bs->file);
631 return offset;
634 void qcow2_free_clusters(BlockDriverState *bs,
635 int64_t offset, int64_t size)
637 int ret;
639 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
640 ret = update_refcount(bs, offset, size, -1);
641 if (ret < 0) {
642 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
643 /* TODO Remember the clusters to free them later and avoid leaking */
648 * free_any_clusters
650 * free clusters according to its type: compressed or not
654 void qcow2_free_any_clusters(BlockDriverState *bs,
655 uint64_t cluster_offset, int nb_clusters)
657 BDRVQcowState *s = bs->opaque;
659 /* free the cluster */
661 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
662 int nb_csectors;
663 nb_csectors = ((cluster_offset >> s->csize_shift) &
664 s->csize_mask) + 1;
665 qcow2_free_clusters(bs,
666 (cluster_offset & s->cluster_offset_mask) & ~511,
667 nb_csectors * 512);
668 return;
671 qcow2_free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
673 return;
678 /*********************************************************/
679 /* snapshots and image creation */
683 void qcow2_create_refcount_update(QCowCreateState *s, int64_t offset,
684 int64_t size)
686 int refcount;
687 int64_t start, last, cluster_offset;
688 uint16_t *p;
690 start = offset & ~(s->cluster_size - 1);
691 last = (offset + size - 1) & ~(s->cluster_size - 1);
692 for(cluster_offset = start; cluster_offset <= last;
693 cluster_offset += s->cluster_size) {
694 p = &s->refcount_block[cluster_offset >> s->cluster_bits];
695 refcount = be16_to_cpu(*p);
696 refcount++;
697 *p = cpu_to_be16(refcount);
701 /* update the refcounts of snapshots and the copied flag */
702 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
703 int64_t l1_table_offset, int l1_size, int addend)
705 BDRVQcowState *s = bs->opaque;
706 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
707 int64_t old_offset, old_l2_offset;
708 int i, j, l1_modified, nb_csectors, refcount;
709 int ret;
711 l2_table = NULL;
712 l1_table = NULL;
713 l1_size2 = l1_size * sizeof(uint64_t);
714 if (l1_table_offset != s->l1_table_offset) {
715 if (l1_size2 != 0) {
716 l1_table = qemu_mallocz(align_offset(l1_size2, 512));
717 } else {
718 l1_table = NULL;
720 l1_allocated = 1;
721 if (bdrv_pread(bs->file, l1_table_offset,
722 l1_table, l1_size2) != l1_size2)
723 goto fail;
724 for(i = 0;i < l1_size; i++)
725 be64_to_cpus(&l1_table[i]);
726 } else {
727 assert(l1_size == s->l1_size);
728 l1_table = s->l1_table;
729 l1_allocated = 0;
732 l1_modified = 0;
733 for(i = 0; i < l1_size; i++) {
734 l2_offset = l1_table[i];
735 if (l2_offset) {
736 old_l2_offset = l2_offset;
737 l2_offset &= ~QCOW_OFLAG_COPIED;
739 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
740 (void**) &l2_table);
741 if (ret < 0) {
742 goto fail;
745 for(j = 0; j < s->l2_size; j++) {
746 offset = be64_to_cpu(l2_table[j]);
747 if (offset != 0) {
748 old_offset = offset;
749 offset &= ~QCOW_OFLAG_COPIED;
750 if (offset & QCOW_OFLAG_COMPRESSED) {
751 nb_csectors = ((offset >> s->csize_shift) &
752 s->csize_mask) + 1;
753 if (addend != 0) {
754 int ret;
755 ret = update_refcount(bs,
756 (offset & s->cluster_offset_mask) & ~511,
757 nb_csectors * 512, addend);
758 if (ret < 0) {
759 goto fail;
762 /* TODO Flushing once for the whole function should
763 * be enough */
764 bdrv_flush(bs->file);
766 /* compressed clusters are never modified */
767 refcount = 2;
768 } else {
769 if (addend != 0) {
770 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
771 } else {
772 refcount = get_refcount(bs, offset >> s->cluster_bits);
775 if (refcount < 0) {
776 goto fail;
780 if (refcount == 1) {
781 offset |= QCOW_OFLAG_COPIED;
783 if (offset != old_offset) {
784 if (addend > 0) {
785 qcow2_cache_set_dependency(bs, s->l2_table_cache,
786 s->refcount_block_cache);
788 l2_table[j] = cpu_to_be64(offset);
789 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
794 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
795 if (ret < 0) {
796 goto fail;
800 if (addend != 0) {
801 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
802 } else {
803 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
805 if (refcount < 0) {
806 goto fail;
807 } else if (refcount == 1) {
808 l2_offset |= QCOW_OFLAG_COPIED;
810 if (l2_offset != old_l2_offset) {
811 l1_table[i] = l2_offset;
812 l1_modified = 1;
816 if (l1_modified) {
817 for(i = 0; i < l1_size; i++)
818 cpu_to_be64s(&l1_table[i]);
819 if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table,
820 l1_size2) < 0)
821 goto fail;
822 for(i = 0; i < l1_size; i++)
823 be64_to_cpus(&l1_table[i]);
825 if (l1_allocated)
826 qemu_free(l1_table);
827 return 0;
828 fail:
829 if (l2_table) {
830 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
833 if (l1_allocated)
834 qemu_free(l1_table);
835 return -EIO;
841 /*********************************************************/
842 /* refcount checking functions */
847 * Increases the refcount for a range of clusters in a given refcount table.
848 * This is used to construct a temporary refcount table out of L1 and L2 tables
849 * which can be compared the the refcount table saved in the image.
851 * Modifies the number of errors in res.
853 static void inc_refcounts(BlockDriverState *bs,
854 BdrvCheckResult *res,
855 uint16_t *refcount_table,
856 int refcount_table_size,
857 int64_t offset, int64_t size)
859 BDRVQcowState *s = bs->opaque;
860 int64_t start, last, cluster_offset;
861 int k;
863 if (size <= 0)
864 return;
866 start = offset & ~(s->cluster_size - 1);
867 last = (offset + size - 1) & ~(s->cluster_size - 1);
868 for(cluster_offset = start; cluster_offset <= last;
869 cluster_offset += s->cluster_size) {
870 k = cluster_offset >> s->cluster_bits;
871 if (k < 0) {
872 fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
873 cluster_offset);
874 res->corruptions++;
875 } else if (k >= refcount_table_size) {
876 fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after "
877 "the end of the image file, can't properly check refcounts.\n",
878 cluster_offset);
879 res->check_errors++;
880 } else {
881 if (++refcount_table[k] == 0) {
882 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
883 "\n", cluster_offset);
884 res->corruptions++;
891 * Increases the refcount in the given refcount table for the all clusters
892 * referenced in the L2 table. While doing so, performs some checks on L2
893 * entries.
895 * Returns the number of errors found by the checks or -errno if an internal
896 * error occurred.
898 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
899 uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
900 int check_copied)
902 BDRVQcowState *s = bs->opaque;
903 uint64_t *l2_table, offset;
904 int i, l2_size, nb_csectors, refcount;
906 /* Read L2 table from disk */
907 l2_size = s->l2_size * sizeof(uint64_t);
908 l2_table = qemu_malloc(l2_size);
910 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
911 goto fail;
913 /* Do the actual checks */
914 for(i = 0; i < s->l2_size; i++) {
915 offset = be64_to_cpu(l2_table[i]);
916 if (offset != 0) {
917 if (offset & QCOW_OFLAG_COMPRESSED) {
918 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
919 if (offset & QCOW_OFLAG_COPIED) {
920 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
921 "copied flag must never be set for compressed "
922 "clusters\n", offset >> s->cluster_bits);
923 offset &= ~QCOW_OFLAG_COPIED;
924 res->corruptions++;
927 /* Mark cluster as used */
928 nb_csectors = ((offset >> s->csize_shift) &
929 s->csize_mask) + 1;
930 offset &= s->cluster_offset_mask;
931 inc_refcounts(bs, res, refcount_table, refcount_table_size,
932 offset & ~511, nb_csectors * 512);
933 } else {
934 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
935 if (check_copied) {
936 uint64_t entry = offset;
937 offset &= ~QCOW_OFLAG_COPIED;
938 refcount = get_refcount(bs, offset >> s->cluster_bits);
939 if (refcount < 0) {
940 fprintf(stderr, "Can't get refcount for offset %"
941 PRIx64 ": %s\n", entry, strerror(-refcount));
942 goto fail;
944 if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) {
945 fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
946 PRIx64 " refcount=%d\n", entry, refcount);
947 res->corruptions++;
951 /* Mark cluster as used */
952 offset &= ~QCOW_OFLAG_COPIED;
953 inc_refcounts(bs, res, refcount_table,refcount_table_size,
954 offset, s->cluster_size);
956 /* Correct offsets are cluster aligned */
957 if (offset & (s->cluster_size - 1)) {
958 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
959 "properly aligned; L2 entry corrupted.\n", offset);
960 res->corruptions++;
966 qemu_free(l2_table);
967 return 0;
969 fail:
970 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
971 qemu_free(l2_table);
972 return -EIO;
976 * Increases the refcount for the L1 table, its L2 tables and all referenced
977 * clusters in the given refcount table. While doing so, performs some checks
978 * on L1 and L2 entries.
980 * Returns the number of errors found by the checks or -errno if an internal
981 * error occurred.
983 static int check_refcounts_l1(BlockDriverState *bs,
984 BdrvCheckResult *res,
985 uint16_t *refcount_table,
986 int refcount_table_size,
987 int64_t l1_table_offset, int l1_size,
988 int check_copied)
990 BDRVQcowState *s = bs->opaque;
991 uint64_t *l1_table, l2_offset, l1_size2;
992 int i, refcount, ret;
994 l1_size2 = l1_size * sizeof(uint64_t);
996 /* Mark L1 table as used */
997 inc_refcounts(bs, res, refcount_table, refcount_table_size,
998 l1_table_offset, l1_size2);
1000 /* Read L1 table entries from disk */
1001 if (l1_size2 == 0) {
1002 l1_table = NULL;
1003 } else {
1004 l1_table = qemu_malloc(l1_size2);
1005 if (bdrv_pread(bs->file, l1_table_offset,
1006 l1_table, l1_size2) != l1_size2)
1007 goto fail;
1008 for(i = 0;i < l1_size; i++)
1009 be64_to_cpus(&l1_table[i]);
1012 /* Do the actual checks */
1013 for(i = 0; i < l1_size; i++) {
1014 l2_offset = l1_table[i];
1015 if (l2_offset) {
1016 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
1017 if (check_copied) {
1018 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
1019 >> s->cluster_bits);
1020 if (refcount < 0) {
1021 fprintf(stderr, "Can't get refcount for l2_offset %"
1022 PRIx64 ": %s\n", l2_offset, strerror(-refcount));
1023 goto fail;
1025 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
1026 fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
1027 " refcount=%d\n", l2_offset, refcount);
1028 res->corruptions++;
1032 /* Mark L2 table as used */
1033 l2_offset &= ~QCOW_OFLAG_COPIED;
1034 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1035 l2_offset, s->cluster_size);
1037 /* L2 tables are cluster aligned */
1038 if (l2_offset & (s->cluster_size - 1)) {
1039 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1040 "cluster aligned; L1 entry corrupted\n", l2_offset);
1041 res->corruptions++;
1044 /* Process and check L2 entries */
1045 ret = check_refcounts_l2(bs, res, refcount_table,
1046 refcount_table_size, l2_offset, check_copied);
1047 if (ret < 0) {
1048 goto fail;
1052 qemu_free(l1_table);
1053 return 0;
1055 fail:
1056 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1057 res->check_errors++;
1058 qemu_free(l1_table);
1059 return -EIO;
1063 * Checks an image for refcount consistency.
1065 * Returns 0 if no errors are found, the number of errors in case the image is
1066 * detected as corrupted, and -errno when an internal error occured.
1068 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res)
1070 BDRVQcowState *s = bs->opaque;
1071 int64_t size;
1072 int nb_clusters, refcount1, refcount2, i;
1073 QCowSnapshot *sn;
1074 uint16_t *refcount_table;
1075 int ret;
1077 size = bdrv_getlength(bs->file);
1078 nb_clusters = size_to_clusters(s, size);
1079 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
1081 /* header */
1082 inc_refcounts(bs, res, refcount_table, nb_clusters,
1083 0, s->cluster_size);
1085 /* current L1 table */
1086 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1087 s->l1_table_offset, s->l1_size, 1);
1088 if (ret < 0) {
1089 return ret;
1092 /* snapshots */
1093 for(i = 0; i < s->nb_snapshots; i++) {
1094 sn = s->snapshots + i;
1095 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1096 sn->l1_table_offset, sn->l1_size, 0);
1097 if (ret < 0) {
1098 return ret;
1101 inc_refcounts(bs, res, refcount_table, nb_clusters,
1102 s->snapshots_offset, s->snapshots_size);
1104 /* refcount data */
1105 inc_refcounts(bs, res, refcount_table, nb_clusters,
1106 s->refcount_table_offset,
1107 s->refcount_table_size * sizeof(uint64_t));
1109 for(i = 0; i < s->refcount_table_size; i++) {
1110 uint64_t offset, cluster;
1111 offset = s->refcount_table[i];
1112 cluster = offset >> s->cluster_bits;
1114 /* Refcount blocks are cluster aligned */
1115 if (offset & (s->cluster_size - 1)) {
1116 fprintf(stderr, "ERROR refcount block %d is not "
1117 "cluster aligned; refcount table entry corrupted\n", i);
1118 res->corruptions++;
1119 continue;
1122 if (cluster >= nb_clusters) {
1123 fprintf(stderr, "ERROR refcount block %d is outside image\n", i);
1124 res->corruptions++;
1125 continue;
1128 if (offset != 0) {
1129 inc_refcounts(bs, res, refcount_table, nb_clusters,
1130 offset, s->cluster_size);
1131 if (refcount_table[cluster] != 1) {
1132 fprintf(stderr, "ERROR refcount block %d refcount=%d\n",
1133 i, refcount_table[cluster]);
1134 res->corruptions++;
1139 /* compare ref counts */
1140 for(i = 0; i < nb_clusters; i++) {
1141 refcount1 = get_refcount(bs, i);
1142 if (refcount1 < 0) {
1143 fprintf(stderr, "Can't get refcount for cluster %d: %s\n",
1144 i, strerror(-refcount1));
1145 res->check_errors++;
1146 continue;
1149 refcount2 = refcount_table[i];
1150 if (refcount1 != refcount2) {
1151 fprintf(stderr, "%s cluster %d refcount=%d reference=%d\n",
1152 refcount1 < refcount2 ? "ERROR" : "Leaked",
1153 i, refcount1, refcount2);
1154 if (refcount1 < refcount2) {
1155 res->corruptions++;
1156 } else {
1157 res->leaks++;
1162 qemu_free(refcount_table);
1164 return 0;