PPC: Fix rldcl
[qemu/agraf.git] / block / qcow2-refcount.c
blobb32738f8d9ce6a8f34d2550759699ac9487a48a3
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"
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 = g_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 g_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] & REFT_OFFSET_MASK;
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 ret = qcow2_cache_flush(bs, s->l2_table_cache);
205 if (ret < 0) {
206 return ret;
209 /* Allocate the refcount block itself and mark it as used */
210 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
211 if (new_block < 0) {
212 return new_block;
215 #ifdef DEBUG_ALLOC2
216 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
217 " at %" PRIx64 "\n",
218 refcount_table_index, cluster_index << s->cluster_bits, new_block);
219 #endif
221 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
222 /* Zero the new refcount block before updating it */
223 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
224 (void**) refcount_block);
225 if (ret < 0) {
226 goto fail_block;
229 memset(*refcount_block, 0, s->cluster_size);
231 /* The block describes itself, need to update the cache */
232 int block_index = (new_block >> s->cluster_bits) &
233 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
234 (*refcount_block)[block_index] = cpu_to_be16(1);
235 } else {
236 /* Described somewhere else. This can recurse at most twice before we
237 * arrive at a block that describes itself. */
238 ret = update_refcount(bs, new_block, s->cluster_size, 1);
239 if (ret < 0) {
240 goto fail_block;
243 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
244 if (ret < 0) {
245 goto fail_block;
248 /* Initialize the new refcount block only after updating its refcount,
249 * update_refcount uses the refcount cache itself */
250 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
251 (void**) refcount_block);
252 if (ret < 0) {
253 goto fail_block;
256 memset(*refcount_block, 0, s->cluster_size);
259 /* Now the new refcount block needs to be written to disk */
260 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
261 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block);
262 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
263 if (ret < 0) {
264 goto fail_block;
267 /* If the refcount table is big enough, just hook the block up there */
268 if (refcount_table_index < s->refcount_table_size) {
269 uint64_t data64 = cpu_to_be64(new_block);
270 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
271 ret = bdrv_pwrite_sync(bs->file,
272 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
273 &data64, sizeof(data64));
274 if (ret < 0) {
275 goto fail_block;
278 s->refcount_table[refcount_table_index] = new_block;
279 return 0;
282 ret = qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
283 if (ret < 0) {
284 goto fail_block;
288 * If we come here, we need to grow the refcount table. Again, a new
289 * refcount table needs some space and we can't simply allocate to avoid
290 * endless recursion.
292 * Therefore let's grab new refcount blocks at the end of the image, which
293 * will describe themselves and the new refcount table. This way we can
294 * reference them only in the new table and do the switch to the new
295 * refcount table at once without producing an inconsistent state in
296 * between.
298 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
300 /* Calculate the number of refcount blocks needed so far */
301 uint64_t refcount_block_clusters = 1 << (s->cluster_bits - REFCOUNT_SHIFT);
302 uint64_t blocks_used = (s->free_cluster_index +
303 refcount_block_clusters - 1) / refcount_block_clusters;
305 /* And now we need at least one block more for the new metadata */
306 uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
307 uint64_t last_table_size;
308 uint64_t blocks_clusters;
309 do {
310 uint64_t table_clusters =
311 size_to_clusters(s, table_size * sizeof(uint64_t));
312 blocks_clusters = 1 +
313 ((table_clusters + refcount_block_clusters - 1)
314 / refcount_block_clusters);
315 uint64_t meta_clusters = table_clusters + blocks_clusters;
317 last_table_size = table_size;
318 table_size = next_refcount_table_size(s, blocks_used +
319 ((meta_clusters + refcount_block_clusters - 1)
320 / refcount_block_clusters));
322 } while (last_table_size != table_size);
324 #ifdef DEBUG_ALLOC2
325 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
326 s->refcount_table_size, table_size);
327 #endif
329 /* Create the new refcount table and blocks */
330 uint64_t meta_offset = (blocks_used * refcount_block_clusters) *
331 s->cluster_size;
332 uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
333 uint16_t *new_blocks = g_malloc0(blocks_clusters * s->cluster_size);
334 uint64_t *new_table = g_malloc0(table_size * sizeof(uint64_t));
336 assert(meta_offset >= (s->free_cluster_index * s->cluster_size));
338 /* Fill the new refcount table */
339 memcpy(new_table, s->refcount_table,
340 s->refcount_table_size * sizeof(uint64_t));
341 new_table[refcount_table_index] = new_block;
343 int i;
344 for (i = 0; i < blocks_clusters; i++) {
345 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
348 /* Fill the refcount blocks */
349 uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
350 int block = 0;
351 for (i = 0; i < table_clusters + blocks_clusters; i++) {
352 new_blocks[block++] = cpu_to_be16(1);
355 /* Write refcount blocks to disk */
356 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
357 ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
358 blocks_clusters * s->cluster_size);
359 g_free(new_blocks);
360 if (ret < 0) {
361 goto fail_table;
364 /* Write refcount table to disk */
365 for(i = 0; i < table_size; i++) {
366 cpu_to_be64s(&new_table[i]);
369 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
370 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
371 table_size * sizeof(uint64_t));
372 if (ret < 0) {
373 goto fail_table;
376 for(i = 0; i < table_size; i++) {
377 be64_to_cpus(&new_table[i]);
380 /* Hook up the new refcount table in the qcow2 header */
381 uint8_t data[12];
382 cpu_to_be64w((uint64_t*)data, table_offset);
383 cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
384 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
385 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
386 data, sizeof(data));
387 if (ret < 0) {
388 goto fail_table;
391 /* And switch it in memory */
392 uint64_t old_table_offset = s->refcount_table_offset;
393 uint64_t old_table_size = s->refcount_table_size;
395 g_free(s->refcount_table);
396 s->refcount_table = new_table;
397 s->refcount_table_size = table_size;
398 s->refcount_table_offset = table_offset;
400 /* Free old table. Remember, we must not change free_cluster_index */
401 uint64_t old_free_cluster_index = s->free_cluster_index;
402 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
403 s->free_cluster_index = old_free_cluster_index;
405 ret = load_refcount_block(bs, new_block, (void**) refcount_block);
406 if (ret < 0) {
407 return ret;
410 return 0;
412 fail_table:
413 g_free(new_table);
414 fail_block:
415 if (*refcount_block != NULL) {
416 qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
418 return ret;
421 /* XXX: cache several refcount block clusters ? */
422 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
423 int64_t offset, int64_t length, int addend)
425 BDRVQcowState *s = bs->opaque;
426 int64_t start, last, cluster_offset;
427 uint16_t *refcount_block = NULL;
428 int64_t old_table_index = -1;
429 int ret;
431 #ifdef DEBUG_ALLOC2
432 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
433 offset, length, addend);
434 #endif
435 if (length < 0) {
436 return -EINVAL;
437 } else if (length == 0) {
438 return 0;
441 if (addend < 0) {
442 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
443 s->l2_table_cache);
446 start = offset & ~(s->cluster_size - 1);
447 last = (offset + length - 1) & ~(s->cluster_size - 1);
448 for(cluster_offset = start; cluster_offset <= last;
449 cluster_offset += s->cluster_size)
451 int block_index, refcount;
452 int64_t cluster_index = cluster_offset >> s->cluster_bits;
453 int64_t table_index =
454 cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
456 /* Load the refcount block and allocate it if needed */
457 if (table_index != old_table_index) {
458 if (refcount_block) {
459 ret = qcow2_cache_put(bs, s->refcount_block_cache,
460 (void**) &refcount_block);
461 if (ret < 0) {
462 goto fail;
466 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
467 if (ret < 0) {
468 goto fail;
471 old_table_index = table_index;
473 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
475 /* we can update the count and save it */
476 block_index = cluster_index &
477 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
479 refcount = be16_to_cpu(refcount_block[block_index]);
480 refcount += addend;
481 if (refcount < 0 || refcount > 0xffff) {
482 ret = -EINVAL;
483 goto fail;
485 if (refcount == 0 && cluster_index < s->free_cluster_index) {
486 s->free_cluster_index = cluster_index;
488 refcount_block[block_index] = cpu_to_be16(refcount);
491 ret = 0;
492 fail:
493 /* Write last changed block to disk */
494 if (refcount_block) {
495 int wret;
496 wret = qcow2_cache_put(bs, s->refcount_block_cache,
497 (void**) &refcount_block);
498 if (wret < 0) {
499 return ret < 0 ? ret : wret;
504 * Try do undo any updates if an error is returned (This may succeed in
505 * some cases like ENOSPC for allocating a new refcount block)
507 if (ret < 0) {
508 int dummy;
509 dummy = update_refcount(bs, offset, cluster_offset - offset, -addend);
510 (void)dummy;
513 return ret;
517 * Increases or decreases the refcount of a given cluster by one.
518 * addend must be 1 or -1.
520 * If the return value is non-negative, it is the new refcount of the cluster.
521 * If it is negative, it is -errno and indicates an error.
523 static int update_cluster_refcount(BlockDriverState *bs,
524 int64_t cluster_index,
525 int addend)
527 BDRVQcowState *s = bs->opaque;
528 int ret;
530 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
531 if (ret < 0) {
532 return ret;
535 return get_refcount(bs, cluster_index);
540 /*********************************************************/
541 /* cluster allocation functions */
545 /* return < 0 if error */
546 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
548 BDRVQcowState *s = bs->opaque;
549 int i, nb_clusters, refcount;
551 nb_clusters = size_to_clusters(s, size);
552 retry:
553 for(i = 0; i < nb_clusters; i++) {
554 int64_t next_cluster_index = s->free_cluster_index++;
555 refcount = get_refcount(bs, next_cluster_index);
557 if (refcount < 0) {
558 return refcount;
559 } else if (refcount != 0) {
560 goto retry;
563 #ifdef DEBUG_ALLOC2
564 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
565 size,
566 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
567 #endif
568 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
571 int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
573 int64_t offset;
574 int ret;
576 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
577 offset = alloc_clusters_noref(bs, size);
578 if (offset < 0) {
579 return offset;
582 ret = update_refcount(bs, offset, size, 1);
583 if (ret < 0) {
584 return ret;
587 return offset;
590 int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
591 int nb_clusters)
593 BDRVQcowState *s = bs->opaque;
594 uint64_t cluster_index;
595 uint64_t old_free_cluster_index;
596 int i, refcount, ret;
598 /* Check how many clusters there are free */
599 cluster_index = offset >> s->cluster_bits;
600 for(i = 0; i < nb_clusters; i++) {
601 refcount = get_refcount(bs, cluster_index++);
603 if (refcount < 0) {
604 return refcount;
605 } else if (refcount != 0) {
606 break;
610 /* And then allocate them */
611 old_free_cluster_index = s->free_cluster_index;
612 s->free_cluster_index = cluster_index + i;
614 ret = update_refcount(bs, offset, i << s->cluster_bits, 1);
615 if (ret < 0) {
616 return ret;
619 s->free_cluster_index = old_free_cluster_index;
621 return i;
624 /* only used to allocate compressed sectors. We try to allocate
625 contiguous sectors. size must be <= cluster_size */
626 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
628 BDRVQcowState *s = bs->opaque;
629 int64_t offset, cluster_offset;
630 int free_in_cluster;
632 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
633 assert(size > 0 && size <= s->cluster_size);
634 if (s->free_byte_offset == 0) {
635 offset = qcow2_alloc_clusters(bs, s->cluster_size);
636 if (offset < 0) {
637 return offset;
639 s->free_byte_offset = offset;
641 redo:
642 free_in_cluster = s->cluster_size -
643 (s->free_byte_offset & (s->cluster_size - 1));
644 if (size <= free_in_cluster) {
645 /* enough space in current cluster */
646 offset = s->free_byte_offset;
647 s->free_byte_offset += size;
648 free_in_cluster -= size;
649 if (free_in_cluster == 0)
650 s->free_byte_offset = 0;
651 if ((offset & (s->cluster_size - 1)) != 0)
652 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
653 } else {
654 offset = qcow2_alloc_clusters(bs, s->cluster_size);
655 if (offset < 0) {
656 return offset;
658 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
659 if ((cluster_offset + s->cluster_size) == offset) {
660 /* we are lucky: contiguous data */
661 offset = s->free_byte_offset;
662 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
663 s->free_byte_offset += size;
664 } else {
665 s->free_byte_offset = offset;
666 goto redo;
670 /* The cluster refcount was incremented, either by qcow2_alloc_clusters()
671 * or explicitly by update_cluster_refcount(). Refcount blocks must be
672 * flushed before the caller's L2 table updates.
674 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
675 return offset;
678 void qcow2_free_clusters(BlockDriverState *bs,
679 int64_t offset, int64_t size)
681 int ret;
683 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
684 ret = update_refcount(bs, offset, size, -1);
685 if (ret < 0) {
686 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
687 /* TODO Remember the clusters to free them later and avoid leaking */
692 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
693 * normal cluster, compressed cluster, etc.)
695 void qcow2_free_any_clusters(BlockDriverState *bs,
696 uint64_t l2_entry, int nb_clusters)
698 BDRVQcowState *s = bs->opaque;
700 switch (qcow2_get_cluster_type(l2_entry)) {
701 case QCOW2_CLUSTER_COMPRESSED:
703 int nb_csectors;
704 nb_csectors = ((l2_entry >> s->csize_shift) &
705 s->csize_mask) + 1;
706 qcow2_free_clusters(bs,
707 (l2_entry & s->cluster_offset_mask) & ~511,
708 nb_csectors * 512);
710 break;
711 case QCOW2_CLUSTER_NORMAL:
712 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
713 nb_clusters << s->cluster_bits);
714 break;
715 case QCOW2_CLUSTER_UNALLOCATED:
716 case QCOW2_CLUSTER_ZERO:
717 break;
718 default:
719 abort();
725 /*********************************************************/
726 /* snapshots and image creation */
730 /* update the refcounts of snapshots and the copied flag */
731 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
732 int64_t l1_table_offset, int l1_size, int addend)
734 BDRVQcowState *s = bs->opaque;
735 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
736 int64_t old_offset, old_l2_offset;
737 int i, j, l1_modified = 0, nb_csectors, refcount;
738 int ret;
740 l2_table = NULL;
741 l1_table = NULL;
742 l1_size2 = l1_size * sizeof(uint64_t);
744 /* WARNING: qcow2_snapshot_goto relies on this function not using the
745 * l1_table_offset when it is the current s->l1_table_offset! Be careful
746 * when changing this! */
747 if (l1_table_offset != s->l1_table_offset) {
748 l1_table = g_malloc0(align_offset(l1_size2, 512));
749 l1_allocated = 1;
751 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
752 if (ret < 0) {
753 goto fail;
756 for(i = 0;i < l1_size; i++)
757 be64_to_cpus(&l1_table[i]);
758 } else {
759 assert(l1_size == s->l1_size);
760 l1_table = s->l1_table;
761 l1_allocated = 0;
764 for(i = 0; i < l1_size; i++) {
765 l2_offset = l1_table[i];
766 if (l2_offset) {
767 old_l2_offset = l2_offset;
768 l2_offset &= L1E_OFFSET_MASK;
770 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
771 (void**) &l2_table);
772 if (ret < 0) {
773 goto fail;
776 for(j = 0; j < s->l2_size; j++) {
777 offset = be64_to_cpu(l2_table[j]);
778 if (offset != 0) {
779 old_offset = offset;
780 offset &= ~QCOW_OFLAG_COPIED;
781 if (offset & QCOW_OFLAG_COMPRESSED) {
782 nb_csectors = ((offset >> s->csize_shift) &
783 s->csize_mask) + 1;
784 if (addend != 0) {
785 int ret;
786 ret = update_refcount(bs,
787 (offset & s->cluster_offset_mask) & ~511,
788 nb_csectors * 512, addend);
789 if (ret < 0) {
790 goto fail;
793 /* compressed clusters are never modified */
794 refcount = 2;
795 } else {
796 uint64_t cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
797 if (addend != 0) {
798 refcount = update_cluster_refcount(bs, cluster_index, addend);
799 } else {
800 refcount = get_refcount(bs, cluster_index);
803 if (refcount < 0) {
804 ret = refcount;
805 goto fail;
809 if (refcount == 1) {
810 offset |= QCOW_OFLAG_COPIED;
812 if (offset != old_offset) {
813 if (addend > 0) {
814 qcow2_cache_set_dependency(bs, s->l2_table_cache,
815 s->refcount_block_cache);
817 l2_table[j] = cpu_to_be64(offset);
818 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
823 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
824 if (ret < 0) {
825 goto fail;
829 if (addend != 0) {
830 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
831 } else {
832 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
834 if (refcount < 0) {
835 ret = refcount;
836 goto fail;
837 } else if (refcount == 1) {
838 l2_offset |= QCOW_OFLAG_COPIED;
840 if (l2_offset != old_l2_offset) {
841 l1_table[i] = l2_offset;
842 l1_modified = 1;
847 ret = bdrv_flush(bs);
848 fail:
849 if (l2_table) {
850 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
853 /* Update L1 only if it isn't deleted anyway (addend = -1) */
854 if (ret == 0 && addend >= 0 && l1_modified) {
855 for (i = 0; i < l1_size; i++) {
856 cpu_to_be64s(&l1_table[i]);
859 ret = bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table, l1_size2);
861 for (i = 0; i < l1_size; i++) {
862 be64_to_cpus(&l1_table[i]);
865 if (l1_allocated)
866 g_free(l1_table);
867 return ret;
873 /*********************************************************/
874 /* refcount checking functions */
879 * Increases the refcount for a range of clusters in a given refcount table.
880 * This is used to construct a temporary refcount table out of L1 and L2 tables
881 * which can be compared the the refcount table saved in the image.
883 * Modifies the number of errors in res.
885 static void inc_refcounts(BlockDriverState *bs,
886 BdrvCheckResult *res,
887 uint16_t *refcount_table,
888 int refcount_table_size,
889 int64_t offset, int64_t size)
891 BDRVQcowState *s = bs->opaque;
892 int64_t start, last, cluster_offset;
893 int k;
895 if (size <= 0)
896 return;
898 start = offset & ~(s->cluster_size - 1);
899 last = (offset + size - 1) & ~(s->cluster_size - 1);
900 for(cluster_offset = start; cluster_offset <= last;
901 cluster_offset += s->cluster_size) {
902 k = cluster_offset >> s->cluster_bits;
903 if (k < 0) {
904 fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
905 cluster_offset);
906 res->corruptions++;
907 } else if (k >= refcount_table_size) {
908 fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after "
909 "the end of the image file, can't properly check refcounts.\n",
910 cluster_offset);
911 res->check_errors++;
912 } else {
913 if (++refcount_table[k] == 0) {
914 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
915 "\n", cluster_offset);
916 res->corruptions++;
922 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
923 enum {
924 CHECK_OFLAG_COPIED = 0x1, /* check QCOW_OFLAG_COPIED matches refcount */
925 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
929 * Increases the refcount in the given refcount table for the all clusters
930 * referenced in the L2 table. While doing so, performs some checks on L2
931 * entries.
933 * Returns the number of errors found by the checks or -errno if an internal
934 * error occurred.
936 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
937 uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
938 int flags)
940 BDRVQcowState *s = bs->opaque;
941 uint64_t *l2_table, l2_entry;
942 uint64_t next_contiguous_offset = 0;
943 int i, l2_size, nb_csectors, refcount;
945 /* Read L2 table from disk */
946 l2_size = s->l2_size * sizeof(uint64_t);
947 l2_table = g_malloc(l2_size);
949 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
950 goto fail;
952 /* Do the actual checks */
953 for(i = 0; i < s->l2_size; i++) {
954 l2_entry = be64_to_cpu(l2_table[i]);
956 switch (qcow2_get_cluster_type(l2_entry)) {
957 case QCOW2_CLUSTER_COMPRESSED:
958 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
959 if (l2_entry & QCOW_OFLAG_COPIED) {
960 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
961 "copied flag must never be set for compressed "
962 "clusters\n", l2_entry >> s->cluster_bits);
963 l2_entry &= ~QCOW_OFLAG_COPIED;
964 res->corruptions++;
967 /* Mark cluster as used */
968 nb_csectors = ((l2_entry >> s->csize_shift) &
969 s->csize_mask) + 1;
970 l2_entry &= s->cluster_offset_mask;
971 inc_refcounts(bs, res, refcount_table, refcount_table_size,
972 l2_entry & ~511, nb_csectors * 512);
974 if (flags & CHECK_FRAG_INFO) {
975 res->bfi.allocated_clusters++;
976 res->bfi.compressed_clusters++;
978 /* Compressed clusters are fragmented by nature. Since they
979 * take up sub-sector space but we only have sector granularity
980 * I/O we need to re-read the same sectors even for adjacent
981 * compressed clusters.
983 res->bfi.fragmented_clusters++;
985 break;
987 case QCOW2_CLUSTER_ZERO:
988 if ((l2_entry & L2E_OFFSET_MASK) == 0) {
989 break;
991 /* fall through */
993 case QCOW2_CLUSTER_NORMAL:
995 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
996 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
998 if (flags & CHECK_OFLAG_COPIED) {
999 refcount = get_refcount(bs, offset >> s->cluster_bits);
1000 if (refcount < 0) {
1001 fprintf(stderr, "Can't get refcount for offset %"
1002 PRIx64 ": %s\n", l2_entry, strerror(-refcount));
1003 goto fail;
1005 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
1006 fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
1007 PRIx64 " refcount=%d\n", l2_entry, refcount);
1008 res->corruptions++;
1012 if (flags & CHECK_FRAG_INFO) {
1013 res->bfi.allocated_clusters++;
1014 if (next_contiguous_offset &&
1015 offset != next_contiguous_offset) {
1016 res->bfi.fragmented_clusters++;
1018 next_contiguous_offset = offset + s->cluster_size;
1021 /* Mark cluster as used */
1022 inc_refcounts(bs, res, refcount_table,refcount_table_size,
1023 offset, s->cluster_size);
1025 /* Correct offsets are cluster aligned */
1026 if (offset & (s->cluster_size - 1)) {
1027 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
1028 "properly aligned; L2 entry corrupted.\n", offset);
1029 res->corruptions++;
1031 break;
1034 case QCOW2_CLUSTER_UNALLOCATED:
1035 break;
1037 default:
1038 abort();
1042 g_free(l2_table);
1043 return 0;
1045 fail:
1046 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1047 g_free(l2_table);
1048 return -EIO;
1052 * Increases the refcount for the L1 table, its L2 tables and all referenced
1053 * clusters in the given refcount table. While doing so, performs some checks
1054 * on L1 and L2 entries.
1056 * Returns the number of errors found by the checks or -errno if an internal
1057 * error occurred.
1059 static int check_refcounts_l1(BlockDriverState *bs,
1060 BdrvCheckResult *res,
1061 uint16_t *refcount_table,
1062 int refcount_table_size,
1063 int64_t l1_table_offset, int l1_size,
1064 int flags)
1066 BDRVQcowState *s = bs->opaque;
1067 uint64_t *l1_table, l2_offset, l1_size2;
1068 int i, refcount, ret;
1070 l1_size2 = l1_size * sizeof(uint64_t);
1072 /* Mark L1 table as used */
1073 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1074 l1_table_offset, l1_size2);
1076 /* Read L1 table entries from disk */
1077 if (l1_size2 == 0) {
1078 l1_table = NULL;
1079 } else {
1080 l1_table = g_malloc(l1_size2);
1081 if (bdrv_pread(bs->file, l1_table_offset,
1082 l1_table, l1_size2) != l1_size2)
1083 goto fail;
1084 for(i = 0;i < l1_size; i++)
1085 be64_to_cpus(&l1_table[i]);
1088 /* Do the actual checks */
1089 for(i = 0; i < l1_size; i++) {
1090 l2_offset = l1_table[i];
1091 if (l2_offset) {
1092 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
1093 if (flags & CHECK_OFLAG_COPIED) {
1094 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
1095 >> s->cluster_bits);
1096 if (refcount < 0) {
1097 fprintf(stderr, "Can't get refcount for l2_offset %"
1098 PRIx64 ": %s\n", l2_offset, strerror(-refcount));
1099 goto fail;
1101 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
1102 fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
1103 " refcount=%d\n", l2_offset, refcount);
1104 res->corruptions++;
1108 /* Mark L2 table as used */
1109 l2_offset &= L1E_OFFSET_MASK;
1110 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1111 l2_offset, s->cluster_size);
1113 /* L2 tables are cluster aligned */
1114 if (l2_offset & (s->cluster_size - 1)) {
1115 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1116 "cluster aligned; L1 entry corrupted\n", l2_offset);
1117 res->corruptions++;
1120 /* Process and check L2 entries */
1121 ret = check_refcounts_l2(bs, res, refcount_table,
1122 refcount_table_size, l2_offset, flags);
1123 if (ret < 0) {
1124 goto fail;
1128 g_free(l1_table);
1129 return 0;
1131 fail:
1132 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1133 res->check_errors++;
1134 g_free(l1_table);
1135 return -EIO;
1139 * Checks an image for refcount consistency.
1141 * Returns 0 if no errors are found, the number of errors in case the image is
1142 * detected as corrupted, and -errno when an internal error occurred.
1144 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1145 BdrvCheckMode fix)
1147 BDRVQcowState *s = bs->opaque;
1148 int64_t size, i, highest_cluster;
1149 int nb_clusters, refcount1, refcount2;
1150 QCowSnapshot *sn;
1151 uint16_t *refcount_table;
1152 int ret;
1154 size = bdrv_getlength(bs->file);
1155 nb_clusters = size_to_clusters(s, size);
1156 refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t));
1158 res->bfi.total_clusters =
1159 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE);
1161 /* header */
1162 inc_refcounts(bs, res, refcount_table, nb_clusters,
1163 0, s->cluster_size);
1165 /* current L1 table */
1166 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1167 s->l1_table_offset, s->l1_size,
1168 CHECK_OFLAG_COPIED | CHECK_FRAG_INFO);
1169 if (ret < 0) {
1170 goto fail;
1173 /* snapshots */
1174 for(i = 0; i < s->nb_snapshots; i++) {
1175 sn = s->snapshots + i;
1176 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1177 sn->l1_table_offset, sn->l1_size, 0);
1178 if (ret < 0) {
1179 goto fail;
1182 inc_refcounts(bs, res, refcount_table, nb_clusters,
1183 s->snapshots_offset, s->snapshots_size);
1185 /* refcount data */
1186 inc_refcounts(bs, res, refcount_table, nb_clusters,
1187 s->refcount_table_offset,
1188 s->refcount_table_size * sizeof(uint64_t));
1190 for(i = 0; i < s->refcount_table_size; i++) {
1191 uint64_t offset, cluster;
1192 offset = s->refcount_table[i];
1193 cluster = offset >> s->cluster_bits;
1195 /* Refcount blocks are cluster aligned */
1196 if (offset & (s->cluster_size - 1)) {
1197 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1198 "cluster aligned; refcount table entry corrupted\n", i);
1199 res->corruptions++;
1200 continue;
1203 if (cluster >= nb_clusters) {
1204 fprintf(stderr, "ERROR refcount block %" PRId64
1205 " is outside image\n", i);
1206 res->corruptions++;
1207 continue;
1210 if (offset != 0) {
1211 inc_refcounts(bs, res, refcount_table, nb_clusters,
1212 offset, s->cluster_size);
1213 if (refcount_table[cluster] != 1) {
1214 fprintf(stderr, "ERROR refcount block %" PRId64
1215 " refcount=%d\n",
1216 i, refcount_table[cluster]);
1217 res->corruptions++;
1222 /* compare ref counts */
1223 for (i = 0, highest_cluster = 0; i < nb_clusters; i++) {
1224 refcount1 = get_refcount(bs, i);
1225 if (refcount1 < 0) {
1226 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
1227 i, strerror(-refcount1));
1228 res->check_errors++;
1229 continue;
1232 refcount2 = refcount_table[i];
1234 if (refcount1 > 0 || refcount2 > 0) {
1235 highest_cluster = i;
1238 if (refcount1 != refcount2) {
1240 /* Check if we're allowed to fix the mismatch */
1241 int *num_fixed = NULL;
1242 if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
1243 num_fixed = &res->leaks_fixed;
1244 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
1245 num_fixed = &res->corruptions_fixed;
1248 fprintf(stderr, "%s cluster %" PRId64 " refcount=%d reference=%d\n",
1249 num_fixed != NULL ? "Repairing" :
1250 refcount1 < refcount2 ? "ERROR" :
1251 "Leaked",
1252 i, refcount1, refcount2);
1254 if (num_fixed) {
1255 ret = update_refcount(bs, i << s->cluster_bits, 1,
1256 refcount2 - refcount1);
1257 if (ret >= 0) {
1258 (*num_fixed)++;
1259 continue;
1263 /* And if we couldn't, print an error */
1264 if (refcount1 < refcount2) {
1265 res->corruptions++;
1266 } else {
1267 res->leaks++;
1272 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
1273 ret = 0;
1275 fail:
1276 g_free(refcount_table);
1278 return ret;