pseries: Add support for new KVM hash table control call
[qemu/rayw.git] / block / qcow2-refcount.c
blob5e3f9153fb9cfaedeb07e054b7a01bb96a4fb454
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 = 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 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 = g_malloc0(blocks_clusters * s->cluster_size);
327 uint64_t *new_table = g_malloc0(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 g_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 be64_to_cpus(&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 g_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 0;
405 fail_table:
406 g_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 fprintf(stderr, "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 fprintf(stderr, "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 int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
586 int nb_clusters)
588 BDRVQcowState *s = bs->opaque;
589 uint64_t cluster_index;
590 uint64_t old_free_cluster_index;
591 int i, refcount, ret;
593 /* Check how many clusters there are free */
594 cluster_index = offset >> s->cluster_bits;
595 for(i = 0; i < nb_clusters; i++) {
596 refcount = get_refcount(bs, cluster_index++);
598 if (refcount < 0) {
599 return refcount;
600 } else if (refcount != 0) {
601 break;
605 /* And then allocate them */
606 old_free_cluster_index = s->free_cluster_index;
607 s->free_cluster_index = cluster_index + i;
609 ret = update_refcount(bs, offset, i << s->cluster_bits, 1);
610 if (ret < 0) {
611 return ret;
614 s->free_cluster_index = old_free_cluster_index;
616 return i;
619 /* only used to allocate compressed sectors. We try to allocate
620 contiguous sectors. size must be <= cluster_size */
621 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
623 BDRVQcowState *s = bs->opaque;
624 int64_t offset, cluster_offset;
625 int free_in_cluster;
627 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
628 assert(size > 0 && size <= s->cluster_size);
629 if (s->free_byte_offset == 0) {
630 offset = qcow2_alloc_clusters(bs, s->cluster_size);
631 if (offset < 0) {
632 return offset;
634 s->free_byte_offset = offset;
636 redo:
637 free_in_cluster = s->cluster_size -
638 (s->free_byte_offset & (s->cluster_size - 1));
639 if (size <= free_in_cluster) {
640 /* enough space in current cluster */
641 offset = s->free_byte_offset;
642 s->free_byte_offset += size;
643 free_in_cluster -= size;
644 if (free_in_cluster == 0)
645 s->free_byte_offset = 0;
646 if ((offset & (s->cluster_size - 1)) != 0)
647 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
648 } else {
649 offset = qcow2_alloc_clusters(bs, s->cluster_size);
650 if (offset < 0) {
651 return offset;
653 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
654 if ((cluster_offset + s->cluster_size) == offset) {
655 /* we are lucky: contiguous data */
656 offset = s->free_byte_offset;
657 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
658 s->free_byte_offset += size;
659 } else {
660 s->free_byte_offset = offset;
661 goto redo;
665 bdrv_flush(bs->file);
666 return offset;
669 void qcow2_free_clusters(BlockDriverState *bs,
670 int64_t offset, int64_t size)
672 int ret;
674 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
675 ret = update_refcount(bs, offset, size, -1);
676 if (ret < 0) {
677 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
678 /* TODO Remember the clusters to free them later and avoid leaking */
683 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
684 * normal cluster, compressed cluster, etc.)
686 void qcow2_free_any_clusters(BlockDriverState *bs,
687 uint64_t l2_entry, int nb_clusters)
689 BDRVQcowState *s = bs->opaque;
691 switch (qcow2_get_cluster_type(l2_entry)) {
692 case QCOW2_CLUSTER_COMPRESSED:
694 int nb_csectors;
695 nb_csectors = ((l2_entry >> s->csize_shift) &
696 s->csize_mask) + 1;
697 qcow2_free_clusters(bs,
698 (l2_entry & s->cluster_offset_mask) & ~511,
699 nb_csectors * 512);
701 break;
702 case QCOW2_CLUSTER_NORMAL:
703 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
704 nb_clusters << s->cluster_bits);
705 break;
706 case QCOW2_CLUSTER_UNALLOCATED:
707 case QCOW2_CLUSTER_ZERO:
708 break;
709 default:
710 abort();
716 /*********************************************************/
717 /* snapshots and image creation */
721 /* update the refcounts of snapshots and the copied flag */
722 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
723 int64_t l1_table_offset, int l1_size, int addend)
725 BDRVQcowState *s = bs->opaque;
726 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
727 int64_t old_offset, old_l2_offset;
728 int i, j, l1_modified = 0, nb_csectors, refcount;
729 int ret;
731 l2_table = NULL;
732 l1_table = NULL;
733 l1_size2 = l1_size * sizeof(uint64_t);
735 /* WARNING: qcow2_snapshot_goto relies on this function not using the
736 * l1_table_offset when it is the current s->l1_table_offset! Be careful
737 * when changing this! */
738 if (l1_table_offset != s->l1_table_offset) {
739 if (l1_size2 != 0) {
740 l1_table = g_malloc0(align_offset(l1_size2, 512));
741 } else {
742 l1_table = NULL;
744 l1_allocated = 1;
745 if (bdrv_pread(bs->file, l1_table_offset,
746 l1_table, l1_size2) != l1_size2)
748 ret = -EIO;
749 goto fail;
752 for(i = 0;i < l1_size; i++)
753 be64_to_cpus(&l1_table[i]);
754 } else {
755 assert(l1_size == s->l1_size);
756 l1_table = s->l1_table;
757 l1_allocated = 0;
760 for(i = 0; i < l1_size; i++) {
761 l2_offset = l1_table[i];
762 if (l2_offset) {
763 old_l2_offset = l2_offset;
764 l2_offset &= L1E_OFFSET_MASK;
766 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
767 (void**) &l2_table);
768 if (ret < 0) {
769 goto fail;
772 for(j = 0; j < s->l2_size; j++) {
773 offset = be64_to_cpu(l2_table[j]);
774 if (offset != 0) {
775 old_offset = offset;
776 offset &= ~QCOW_OFLAG_COPIED;
777 if (offset & QCOW_OFLAG_COMPRESSED) {
778 nb_csectors = ((offset >> s->csize_shift) &
779 s->csize_mask) + 1;
780 if (addend != 0) {
781 int ret;
782 ret = update_refcount(bs,
783 (offset & s->cluster_offset_mask) & ~511,
784 nb_csectors * 512, addend);
785 if (ret < 0) {
786 goto fail;
789 /* TODO Flushing once for the whole function should
790 * be enough */
791 bdrv_flush(bs->file);
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 = -EIO;
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 = -EIO;
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 = 0;
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 (addend >= 0 && l1_modified) {
855 for(i = 0; i < l1_size; i++)
856 cpu_to_be64s(&l1_table[i]);
857 if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table,
858 l1_size2) < 0)
859 goto fail;
860 for(i = 0; i < l1_size; i++)
861 be64_to_cpus(&l1_table[i]);
863 if (l1_allocated)
864 g_free(l1_table);
865 return ret;
871 /*********************************************************/
872 /* refcount checking functions */
877 * Increases the refcount for a range of clusters in a given refcount table.
878 * This is used to construct a temporary refcount table out of L1 and L2 tables
879 * which can be compared the the refcount table saved in the image.
881 * Modifies the number of errors in res.
883 static void inc_refcounts(BlockDriverState *bs,
884 BdrvCheckResult *res,
885 uint16_t *refcount_table,
886 int refcount_table_size,
887 int64_t offset, int64_t size)
889 BDRVQcowState *s = bs->opaque;
890 int64_t start, last, cluster_offset;
891 int k;
893 if (size <= 0)
894 return;
896 start = offset & ~(s->cluster_size - 1);
897 last = (offset + size - 1) & ~(s->cluster_size - 1);
898 for(cluster_offset = start; cluster_offset <= last;
899 cluster_offset += s->cluster_size) {
900 k = cluster_offset >> s->cluster_bits;
901 if (k < 0) {
902 fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
903 cluster_offset);
904 res->corruptions++;
905 } else if (k >= refcount_table_size) {
906 fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after "
907 "the end of the image file, can't properly check refcounts.\n",
908 cluster_offset);
909 res->check_errors++;
910 } else {
911 if (++refcount_table[k] == 0) {
912 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
913 "\n", cluster_offset);
914 res->corruptions++;
921 * Increases the refcount in the given refcount table for the all clusters
922 * referenced in the L2 table. While doing so, performs some checks on L2
923 * entries.
925 * Returns the number of errors found by the checks or -errno if an internal
926 * error occurred.
928 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
929 uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
930 int check_copied)
932 BDRVQcowState *s = bs->opaque;
933 uint64_t *l2_table, l2_entry;
934 int i, l2_size, nb_csectors, refcount;
936 /* Read L2 table from disk */
937 l2_size = s->l2_size * sizeof(uint64_t);
938 l2_table = g_malloc(l2_size);
940 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
941 goto fail;
943 /* Do the actual checks */
944 for(i = 0; i < s->l2_size; i++) {
945 l2_entry = be64_to_cpu(l2_table[i]);
947 switch (qcow2_get_cluster_type(l2_entry)) {
948 case QCOW2_CLUSTER_COMPRESSED:
949 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
950 if (l2_entry & QCOW_OFLAG_COPIED) {
951 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
952 "copied flag must never be set for compressed "
953 "clusters\n", l2_entry >> s->cluster_bits);
954 l2_entry &= ~QCOW_OFLAG_COPIED;
955 res->corruptions++;
958 /* Mark cluster as used */
959 nb_csectors = ((l2_entry >> s->csize_shift) &
960 s->csize_mask) + 1;
961 l2_entry &= s->cluster_offset_mask;
962 inc_refcounts(bs, res, refcount_table, refcount_table_size,
963 l2_entry & ~511, nb_csectors * 512);
964 break;
966 case QCOW2_CLUSTER_ZERO:
967 if ((l2_entry & L2E_OFFSET_MASK) == 0) {
968 break;
970 /* fall through */
972 case QCOW2_CLUSTER_NORMAL:
974 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
975 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
977 if (check_copied) {
978 refcount = get_refcount(bs, offset >> s->cluster_bits);
979 if (refcount < 0) {
980 fprintf(stderr, "Can't get refcount for offset %"
981 PRIx64 ": %s\n", l2_entry, strerror(-refcount));
982 goto fail;
984 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
985 fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
986 PRIx64 " refcount=%d\n", l2_entry, refcount);
987 res->corruptions++;
991 /* Mark cluster as used */
992 inc_refcounts(bs, res, refcount_table,refcount_table_size,
993 offset, s->cluster_size);
995 /* Correct offsets are cluster aligned */
996 if (offset & (s->cluster_size - 1)) {
997 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
998 "properly aligned; L2 entry corrupted.\n", offset);
999 res->corruptions++;
1001 break;
1004 case QCOW2_CLUSTER_UNALLOCATED:
1005 break;
1007 default:
1008 abort();
1012 g_free(l2_table);
1013 return 0;
1015 fail:
1016 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1017 g_free(l2_table);
1018 return -EIO;
1022 * Increases the refcount for the L1 table, its L2 tables and all referenced
1023 * clusters in the given refcount table. While doing so, performs some checks
1024 * on L1 and L2 entries.
1026 * Returns the number of errors found by the checks or -errno if an internal
1027 * error occurred.
1029 static int check_refcounts_l1(BlockDriverState *bs,
1030 BdrvCheckResult *res,
1031 uint16_t *refcount_table,
1032 int refcount_table_size,
1033 int64_t l1_table_offset, int l1_size,
1034 int check_copied)
1036 BDRVQcowState *s = bs->opaque;
1037 uint64_t *l1_table, l2_offset, l1_size2;
1038 int i, refcount, ret;
1040 l1_size2 = l1_size * sizeof(uint64_t);
1042 /* Mark L1 table as used */
1043 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1044 l1_table_offset, l1_size2);
1046 /* Read L1 table entries from disk */
1047 if (l1_size2 == 0) {
1048 l1_table = NULL;
1049 } else {
1050 l1_table = g_malloc(l1_size2);
1051 if (bdrv_pread(bs->file, l1_table_offset,
1052 l1_table, l1_size2) != l1_size2)
1053 goto fail;
1054 for(i = 0;i < l1_size; i++)
1055 be64_to_cpus(&l1_table[i]);
1058 /* Do the actual checks */
1059 for(i = 0; i < l1_size; i++) {
1060 l2_offset = l1_table[i];
1061 if (l2_offset) {
1062 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
1063 if (check_copied) {
1064 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
1065 >> s->cluster_bits);
1066 if (refcount < 0) {
1067 fprintf(stderr, "Can't get refcount for l2_offset %"
1068 PRIx64 ": %s\n", l2_offset, strerror(-refcount));
1069 goto fail;
1071 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
1072 fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
1073 " refcount=%d\n", l2_offset, refcount);
1074 res->corruptions++;
1078 /* Mark L2 table as used */
1079 l2_offset &= L1E_OFFSET_MASK;
1080 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1081 l2_offset, s->cluster_size);
1083 /* L2 tables are cluster aligned */
1084 if (l2_offset & (s->cluster_size - 1)) {
1085 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1086 "cluster aligned; L1 entry corrupted\n", l2_offset);
1087 res->corruptions++;
1090 /* Process and check L2 entries */
1091 ret = check_refcounts_l2(bs, res, refcount_table,
1092 refcount_table_size, l2_offset, check_copied);
1093 if (ret < 0) {
1094 goto fail;
1098 g_free(l1_table);
1099 return 0;
1101 fail:
1102 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1103 res->check_errors++;
1104 g_free(l1_table);
1105 return -EIO;
1109 * Checks an image for refcount consistency.
1111 * Returns 0 if no errors are found, the number of errors in case the image is
1112 * detected as corrupted, and -errno when an internal error occurred.
1114 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1115 BdrvCheckMode fix)
1117 BDRVQcowState *s = bs->opaque;
1118 int64_t size, i;
1119 int nb_clusters, refcount1, refcount2;
1120 QCowSnapshot *sn;
1121 uint16_t *refcount_table;
1122 int ret;
1124 size = bdrv_getlength(bs->file);
1125 nb_clusters = size_to_clusters(s, size);
1126 refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t));
1128 /* header */
1129 inc_refcounts(bs, res, refcount_table, nb_clusters,
1130 0, s->cluster_size);
1132 /* current L1 table */
1133 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1134 s->l1_table_offset, s->l1_size, 1);
1135 if (ret < 0) {
1136 goto fail;
1139 /* snapshots */
1140 for(i = 0; i < s->nb_snapshots; i++) {
1141 sn = s->snapshots + i;
1142 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1143 sn->l1_table_offset, sn->l1_size, 0);
1144 if (ret < 0) {
1145 goto fail;
1148 inc_refcounts(bs, res, refcount_table, nb_clusters,
1149 s->snapshots_offset, s->snapshots_size);
1151 /* refcount data */
1152 inc_refcounts(bs, res, refcount_table, nb_clusters,
1153 s->refcount_table_offset,
1154 s->refcount_table_size * sizeof(uint64_t));
1156 for(i = 0; i < s->refcount_table_size; i++) {
1157 uint64_t offset, cluster;
1158 offset = s->refcount_table[i];
1159 cluster = offset >> s->cluster_bits;
1161 /* Refcount blocks are cluster aligned */
1162 if (offset & (s->cluster_size - 1)) {
1163 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1164 "cluster aligned; refcount table entry corrupted\n", i);
1165 res->corruptions++;
1166 continue;
1169 if (cluster >= nb_clusters) {
1170 fprintf(stderr, "ERROR refcount block %" PRId64
1171 " is outside image\n", i);
1172 res->corruptions++;
1173 continue;
1176 if (offset != 0) {
1177 inc_refcounts(bs, res, refcount_table, nb_clusters,
1178 offset, s->cluster_size);
1179 if (refcount_table[cluster] != 1) {
1180 fprintf(stderr, "ERROR refcount block %" PRId64
1181 " refcount=%d\n",
1182 i, refcount_table[cluster]);
1183 res->corruptions++;
1188 /* compare ref counts */
1189 for(i = 0; i < nb_clusters; i++) {
1190 refcount1 = get_refcount(bs, i);
1191 if (refcount1 < 0) {
1192 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
1193 i, strerror(-refcount1));
1194 res->check_errors++;
1195 continue;
1198 refcount2 = refcount_table[i];
1199 if (refcount1 != refcount2) {
1201 /* Check if we're allowed to fix the mismatch */
1202 int *num_fixed = NULL;
1203 if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
1204 num_fixed = &res->leaks_fixed;
1205 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
1206 num_fixed = &res->corruptions_fixed;
1209 fprintf(stderr, "%s cluster %" PRId64 " refcount=%d reference=%d\n",
1210 num_fixed != NULL ? "Repairing" :
1211 refcount1 < refcount2 ? "ERROR" :
1212 "Leaked",
1213 i, refcount1, refcount2);
1215 if (num_fixed) {
1216 ret = update_refcount(bs, i << s->cluster_bits, 1,
1217 refcount2 - refcount1);
1218 if (ret >= 0) {
1219 (*num_fixed)++;
1220 continue;
1224 /* And if we couldn't, print an error */
1225 if (refcount1 < refcount2) {
1226 res->corruptions++;
1227 } else {
1228 res->leaks++;
1233 ret = 0;
1235 fail:
1236 g_free(refcount_table);
1238 return ret;