]> git.proxmox.com Git - mirror_qemu.git/blame - block/qcow2-refcount.c
qcow2: flush in qcow2_update_snapshot_refcount()
[mirror_qemu.git] / block / qcow2-refcount.c
CommitLineData
f7d0fe02
KW
1/*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
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:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
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.
23 */
24
25#include "qemu-common.h"
737e150e 26#include "block/block_int.h"
f7d0fe02
KW
27#include "block/qcow2.h"
28
29static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
92dcb59f 30static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
f7d0fe02
KW
31 int64_t offset, int64_t length,
32 int addend);
33
3b88e52b 34
f7d0fe02
KW
35/*********************************************************/
36/* refcount handling */
37
ed6ccf0f 38int qcow2_refcount_init(BlockDriverState *bs)
f7d0fe02
KW
39{
40 BDRVQcowState *s = bs->opaque;
41 int ret, refcount_table_size2, i;
42
f7d0fe02 43 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
7267c094 44 s->refcount_table = g_malloc(refcount_table_size2);
f7d0fe02 45 if (s->refcount_table_size > 0) {
66f82cee
KW
46 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
47 ret = bdrv_pread(bs->file, s->refcount_table_offset,
f7d0fe02
KW
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]);
53 }
54 return 0;
55 fail:
56 return -ENOMEM;
57}
58
ed6ccf0f 59void qcow2_refcount_close(BlockDriverState *bs)
f7d0fe02
KW
60{
61 BDRVQcowState *s = bs->opaque;
7267c094 62 g_free(s->refcount_table);
f7d0fe02
KW
63}
64
65
66static int load_refcount_block(BlockDriverState *bs,
29c1a730
KW
67 int64_t refcount_block_offset,
68 void **refcount_block)
f7d0fe02
KW
69{
70 BDRVQcowState *s = bs->opaque;
71 int ret;
3b88e52b 72
66f82cee 73 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
29c1a730
KW
74 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
75 refcount_block);
e14e8ba5 76
29c1a730 77 return ret;
f7d0fe02
KW
78}
79
018faafd
KW
80/*
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.
84 */
f7d0fe02
KW
85static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
86{
87 BDRVQcowState *s = bs->opaque;
88 int refcount_table_index, block_index;
89 int64_t refcount_block_offset;
018faafd 90 int ret;
29c1a730
KW
91 uint16_t *refcount_block;
92 uint16_t refcount;
f7d0fe02
KW
93
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;
29c1a730
KW
100
101 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
102 (void**) &refcount_block);
103 if (ret < 0) {
104 return ret;
f7d0fe02 105 }
29c1a730 106
f7d0fe02
KW
107 block_index = cluster_index &
108 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
29c1a730
KW
109 refcount = be16_to_cpu(refcount_block[block_index]);
110
111 ret = qcow2_cache_put(bs, s->refcount_block_cache,
112 (void**) &refcount_block);
113 if (ret < 0) {
114 return ret;
115 }
116
117 return refcount;
f7d0fe02
KW
118}
119
05121aed
KW
120/*
121 * Rounds the refcount table size up to avoid growing the table for each single
122 * refcount block that is allocated.
123 */
124static unsigned int next_refcount_table_size(BDRVQcowState *s,
125 unsigned int min_size)
126{
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));
130
131 while (min_clusters > refcount_table_clusters) {
132 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
133 }
134
135 return refcount_table_clusters << (s->cluster_bits - 3);
136}
137
92dcb59f
KW
138
139/* Checks if two offsets are described by the same refcount block */
140static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a,
141 uint64_t offset_b)
142{
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);
145
146 return (block_a == block_b);
147}
148
149/*
150 * Loads a refcount block. If it doesn't exist yet, it is allocated first
151 * (including growing the refcount table if needed).
152 *
29c1a730 153 * Returns 0 on success or -errno in error case
92dcb59f 154 */
29c1a730
KW
155static int alloc_refcount_block(BlockDriverState *bs,
156 int64_t cluster_index, uint16_t **refcount_block)
f7d0fe02
KW
157{
158 BDRVQcowState *s = bs->opaque;
92dcb59f
KW
159 unsigned int refcount_table_index;
160 int ret;
161
66f82cee 162 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
8252278a 163
92dcb59f
KW
164 /* Find the refcount block for the given cluster */
165 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
166
167 if (refcount_table_index < s->refcount_table_size) {
168
169 uint64_t refcount_block_offset =
76dc9e0c 170 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
92dcb59f
KW
171
172 /* If it's already there, we're done */
173 if (refcount_block_offset) {
29c1a730
KW
174 return load_refcount_block(bs, refcount_block_offset,
175 (void**) refcount_block);
92dcb59f
KW
176 }
177 }
178
179 /*
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.
183 *
184 * Note that allocating clusters here needs some special care:
185 *
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.
190 *
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.
196 *
197 * - alloc_clusters_noref and qcow2_free_clusters may load a different
198 * refcount block into the cache
199 */
200
29c1a730
KW
201 *refcount_block = NULL;
202
203 /* We write to the refcount table, so we might depend on L2 tables */
9991923b
SH
204 ret = qcow2_cache_flush(bs, s->l2_table_cache);
205 if (ret < 0) {
206 return ret;
207 }
92dcb59f
KW
208
209 /* Allocate the refcount block itself and mark it as used */
2eaa8f63
KW
210 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
211 if (new_block < 0) {
212 return new_block;
213 }
f7d0fe02 214
f7d0fe02 215#ifdef DEBUG_ALLOC2
92dcb59f
KW
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);
f7d0fe02 219#endif
92dcb59f
KW
220
221 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
25408c09 222 /* Zero the new refcount block before updating it */
29c1a730
KW
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;
227 }
228
229 memset(*refcount_block, 0, s->cluster_size);
25408c09 230
92dcb59f
KW
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);
29c1a730 234 (*refcount_block)[block_index] = cpu_to_be16(1);
92dcb59f
KW
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;
241 }
25408c09 242
9991923b
SH
243 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
244 if (ret < 0) {
245 goto fail_block;
246 }
1c4c2814 247
25408c09
KW
248 /* Initialize the new refcount block only after updating its refcount,
249 * update_refcount uses the refcount cache itself */
29c1a730
KW
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;
254 }
255
256 memset(*refcount_block, 0, s->cluster_size);
92dcb59f
KW
257 }
258
259 /* Now the new refcount block needs to be written to disk */
66f82cee 260 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
29c1a730
KW
261 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block);
262 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
92dcb59f
KW
263 if (ret < 0) {
264 goto fail_block;
265 }
266
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);
66f82cee 270 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
8b3b7206 271 ret = bdrv_pwrite_sync(bs->file,
92dcb59f
KW
272 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
273 &data64, sizeof(data64));
274 if (ret < 0) {
275 goto fail_block;
276 }
277
278 s->refcount_table[refcount_table_index] = new_block;
29c1a730
KW
279 return 0;
280 }
281
282 ret = qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
283 if (ret < 0) {
284 goto fail_block;
92dcb59f
KW
285 }
286
287 /*
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.
291 *
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.
297 */
66f82cee 298 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
8252278a 299
92dcb59f
KW
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;
304
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 {
a3548077
KW
310 uint64_t table_clusters =
311 size_to_clusters(s, table_size * sizeof(uint64_t));
92dcb59f
KW
312 blocks_clusters = 1 +
313 ((table_clusters + refcount_block_clusters - 1)
314 / refcount_block_clusters);
315 uint64_t meta_clusters = table_clusters + blocks_clusters;
316
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));
321
322 } while (last_table_size != table_size);
323
324#ifdef DEBUG_ALLOC2
325 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
326 s->refcount_table_size, table_size);
327#endif
328
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;
7267c094
AL
333 uint16_t *new_blocks = g_malloc0(blocks_clusters * s->cluster_size);
334 uint64_t *new_table = g_malloc0(table_size * sizeof(uint64_t));
92dcb59f
KW
335
336 assert(meta_offset >= (s->free_cluster_index * s->cluster_size));
337
338 /* Fill the new refcount table */
f7d0fe02 339 memcpy(new_table, s->refcount_table,
92dcb59f
KW
340 s->refcount_table_size * sizeof(uint64_t));
341 new_table[refcount_table_index] = new_block;
342
343 int i;
344 for (i = 0; i < blocks_clusters; i++) {
345 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
346 }
347
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);
353 }
354
355 /* Write refcount blocks to disk */
66f82cee 356 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
8b3b7206 357 ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
92dcb59f 358 blocks_clusters * s->cluster_size);
7267c094 359 g_free(new_blocks);
92dcb59f
KW
360 if (ret < 0) {
361 goto fail_table;
362 }
363
364 /* Write refcount table to disk */
365 for(i = 0; i < table_size; i++) {
366 cpu_to_be64s(&new_table[i]);
367 }
368
66f82cee 369 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
8b3b7206 370 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
92dcb59f
KW
371 table_size * sizeof(uint64_t));
372 if (ret < 0) {
373 goto fail_table;
374 }
375
376 for(i = 0; i < table_size; i++) {
87267753 377 be64_to_cpus(&new_table[i]);
92dcb59f 378 }
f7d0fe02 379
92dcb59f
KW
380 /* Hook up the new refcount table in the qcow2 header */
381 uint8_t data[12];
f7d0fe02 382 cpu_to_be64w((uint64_t*)data, table_offset);
92dcb59f 383 cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
66f82cee 384 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
8b3b7206 385 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
92dcb59f
KW
386 data, sizeof(data));
387 if (ret < 0) {
388 goto fail_table;
f2b7c8b3
KW
389 }
390
92dcb59f
KW
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;
394
7267c094 395 g_free(s->refcount_table);
f7d0fe02 396 s->refcount_table = new_table;
92dcb59f 397 s->refcount_table_size = table_size;
f7d0fe02
KW
398 s->refcount_table_offset = table_offset;
399
92dcb59f
KW
400 /* Free old table. Remember, we must not change free_cluster_index */
401 uint64_t old_free_cluster_index = s->free_cluster_index;
ed6ccf0f 402 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
92dcb59f 403 s->free_cluster_index = old_free_cluster_index;
f7d0fe02 404
29c1a730 405 ret = load_refcount_block(bs, new_block, (void**) refcount_block);
92dcb59f 406 if (ret < 0) {
29c1a730 407 return ret;
f7d0fe02
KW
408 }
409
2795ecf6 410 return 0;
f7d0fe02 411
92dcb59f 412fail_table:
7267c094 413 g_free(new_table);
92dcb59f 414fail_block:
29c1a730
KW
415 if (*refcount_block != NULL) {
416 qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
3b88e52b 417 }
29c1a730 418 return ret;
9923e05e
KW
419}
420
f7d0fe02 421/* XXX: cache several refcount block clusters ? */
db3a964f
KW
422static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
423 int64_t offset, int64_t length, int addend)
f7d0fe02
KW
424{
425 BDRVQcowState *s = bs->opaque;
426 int64_t start, last, cluster_offset;
29c1a730
KW
427 uint16_t *refcount_block = NULL;
428 int64_t old_table_index = -1;
09508d13 429 int ret;
f7d0fe02
KW
430
431#ifdef DEBUG_ALLOC2
35ee5e39 432 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
f7d0fe02
KW
433 offset, length, addend);
434#endif
7322afe7 435 if (length < 0) {
f7d0fe02 436 return -EINVAL;
7322afe7
KW
437 } else if (length == 0) {
438 return 0;
439 }
440
29c1a730
KW
441 if (addend < 0) {
442 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
443 s->l2_table_cache);
444 }
445
f7d0fe02
KW
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)
450 {
451 int block_index, refcount;
452 int64_t cluster_index = cluster_offset >> s->cluster_bits;
29c1a730
KW
453 int64_t table_index =
454 cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
f7d0fe02 455
29c1a730
KW
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;
463 }
464 }
9923e05e 465
29c1a730 466 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
ed0df867 467 if (ret < 0) {
29c1a730 468 goto fail;
f7d0fe02 469 }
f7d0fe02 470 }
29c1a730 471 old_table_index = table_index;
f7d0fe02 472
29c1a730 473 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
f7d0fe02
KW
474
475 /* we can update the count and save it */
476 block_index = cluster_index &
477 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
f7d0fe02 478
29c1a730 479 refcount = be16_to_cpu(refcount_block[block_index]);
f7d0fe02 480 refcount += addend;
09508d13
KW
481 if (refcount < 0 || refcount > 0xffff) {
482 ret = -EINVAL;
483 goto fail;
484 }
f7d0fe02
KW
485 if (refcount == 0 && cluster_index < s->free_cluster_index) {
486 s->free_cluster_index = cluster_index;
487 }
29c1a730 488 refcount_block[block_index] = cpu_to_be16(refcount);
f7d0fe02
KW
489 }
490
09508d13
KW
491 ret = 0;
492fail:
f7d0fe02 493 /* Write last changed block to disk */
29c1a730 494 if (refcount_block) {
ed0df867 495 int wret;
29c1a730
KW
496 wret = qcow2_cache_put(bs, s->refcount_block_cache,
497 (void**) &refcount_block);
ed0df867
KW
498 if (wret < 0) {
499 return ret < 0 ? ret : wret;
f7d0fe02
KW
500 }
501 }
502
09508d13
KW
503 /*
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)
506 */
507 if (ret < 0) {
508 int dummy;
509 dummy = update_refcount(bs, offset, cluster_offset - offset, -addend);
83e3f76c 510 (void)dummy;
09508d13
KW
511 }
512
513 return ret;
f7d0fe02
KW
514}
515
018faafd
KW
516/*
517 * Increases or decreases the refcount of a given cluster by one.
518 * addend must be 1 or -1.
519 *
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.
522 */
f7d0fe02
KW
523static int update_cluster_refcount(BlockDriverState *bs,
524 int64_t cluster_index,
525 int addend)
526{
527 BDRVQcowState *s = bs->opaque;
528 int ret;
529
530 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
531 if (ret < 0) {
532 return ret;
533 }
534
1c4c2814
KW
535 bdrv_flush(bs->file);
536
f7d0fe02
KW
537 return get_refcount(bs, cluster_index);
538}
539
540
541
542/*********************************************************/
543/* cluster allocation functions */
544
545
546
547/* return < 0 if error */
548static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
549{
550 BDRVQcowState *s = bs->opaque;
2eaa8f63 551 int i, nb_clusters, refcount;
f7d0fe02
KW
552
553 nb_clusters = size_to_clusters(s, size);
554retry:
555 for(i = 0; i < nb_clusters; i++) {
508e0893 556 int64_t next_cluster_index = s->free_cluster_index++;
2eaa8f63
KW
557 refcount = get_refcount(bs, next_cluster_index);
558
559 if (refcount < 0) {
560 return refcount;
561 } else if (refcount != 0) {
f7d0fe02 562 goto retry;
2eaa8f63 563 }
f7d0fe02
KW
564 }
565#ifdef DEBUG_ALLOC2
35ee5e39 566 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
f7d0fe02
KW
567 size,
568 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
569#endif
570 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
571}
572
ed6ccf0f 573int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
f7d0fe02
KW
574{
575 int64_t offset;
db3a964f 576 int ret;
f7d0fe02 577
66f82cee 578 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
f7d0fe02 579 offset = alloc_clusters_noref(bs, size);
2eaa8f63
KW
580 if (offset < 0) {
581 return offset;
582 }
583
db3a964f
KW
584 ret = update_refcount(bs, offset, size, 1);
585 if (ret < 0) {
586 return ret;
587 }
1c4c2814 588
f7d0fe02
KW
589 return offset;
590}
591
256900b1
KW
592int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
593 int nb_clusters)
594{
595 BDRVQcowState *s = bs->opaque;
596 uint64_t cluster_index;
f24423bd 597 uint64_t old_free_cluster_index;
256900b1
KW
598 int i, refcount, ret;
599
600 /* Check how many clusters there are free */
601 cluster_index = offset >> s->cluster_bits;
602 for(i = 0; i < nb_clusters; i++) {
603 refcount = get_refcount(bs, cluster_index++);
604
605 if (refcount < 0) {
606 return refcount;
607 } else if (refcount != 0) {
608 break;
609 }
610 }
611
612 /* And then allocate them */
f24423bd
KW
613 old_free_cluster_index = s->free_cluster_index;
614 s->free_cluster_index = cluster_index + i;
615
256900b1
KW
616 ret = update_refcount(bs, offset, i << s->cluster_bits, 1);
617 if (ret < 0) {
618 return ret;
619 }
620
f24423bd
KW
621 s->free_cluster_index = old_free_cluster_index;
622
256900b1
KW
623 return i;
624}
625
f7d0fe02
KW
626/* only used to allocate compressed sectors. We try to allocate
627 contiguous sectors. size must be <= cluster_size */
ed6ccf0f 628int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
f7d0fe02
KW
629{
630 BDRVQcowState *s = bs->opaque;
631 int64_t offset, cluster_offset;
632 int free_in_cluster;
633
66f82cee 634 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
f7d0fe02
KW
635 assert(size > 0 && size <= s->cluster_size);
636 if (s->free_byte_offset == 0) {
206e6d85
SH
637 offset = qcow2_alloc_clusters(bs, s->cluster_size);
638 if (offset < 0) {
639 return offset;
5d757b56 640 }
206e6d85 641 s->free_byte_offset = offset;
f7d0fe02
KW
642 }
643 redo:
644 free_in_cluster = s->cluster_size -
645 (s->free_byte_offset & (s->cluster_size - 1));
646 if (size <= free_in_cluster) {
647 /* enough space in current cluster */
648 offset = s->free_byte_offset;
649 s->free_byte_offset += size;
650 free_in_cluster -= size;
651 if (free_in_cluster == 0)
652 s->free_byte_offset = 0;
653 if ((offset & (s->cluster_size - 1)) != 0)
654 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
655 } else {
ed6ccf0f 656 offset = qcow2_alloc_clusters(bs, s->cluster_size);
5d757b56
KW
657 if (offset < 0) {
658 return offset;
659 }
f7d0fe02
KW
660 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
661 if ((cluster_offset + s->cluster_size) == offset) {
662 /* we are lucky: contiguous data */
663 offset = s->free_byte_offset;
664 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
665 s->free_byte_offset += size;
666 } else {
667 s->free_byte_offset = offset;
668 goto redo;
669 }
670 }
29216ed1 671
c1f5bafd
SH
672 /* The cluster refcount was incremented, either by qcow2_alloc_clusters()
673 * or explicitly by update_cluster_refcount(). Refcount blocks must be
674 * flushed before the caller's L2 table updates.
675 */
676 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
f7d0fe02
KW
677 return offset;
678}
679
ed6ccf0f 680void qcow2_free_clusters(BlockDriverState *bs,
f7d0fe02
KW
681 int64_t offset, int64_t size)
682{
db3a964f
KW
683 int ret;
684
66f82cee 685 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
db3a964f
KW
686 ret = update_refcount(bs, offset, size, -1);
687 if (ret < 0) {
688 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
003fad6e 689 /* TODO Remember the clusters to free them later and avoid leaking */
db3a964f 690 }
f7d0fe02
KW
691}
692
45aba42f 693/*
c7a4c37a
KW
694 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
695 * normal cluster, compressed cluster, etc.)
45aba42f 696 */
ed6ccf0f 697void qcow2_free_any_clusters(BlockDriverState *bs,
c7a4c37a 698 uint64_t l2_entry, int nb_clusters)
45aba42f
KW
699{
700 BDRVQcowState *s = bs->opaque;
701
c7a4c37a
KW
702 switch (qcow2_get_cluster_type(l2_entry)) {
703 case QCOW2_CLUSTER_COMPRESSED:
704 {
705 int nb_csectors;
706 nb_csectors = ((l2_entry >> s->csize_shift) &
707 s->csize_mask) + 1;
708 qcow2_free_clusters(bs,
709 (l2_entry & s->cluster_offset_mask) & ~511,
710 nb_csectors * 512);
711 }
712 break;
713 case QCOW2_CLUSTER_NORMAL:
714 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
715 nb_clusters << s->cluster_bits);
716 break;
717 case QCOW2_CLUSTER_UNALLOCATED:
6377af48 718 case QCOW2_CLUSTER_ZERO:
c7a4c37a
KW
719 break;
720 default:
721 abort();
45aba42f 722 }
45aba42f
KW
723}
724
f7d0fe02
KW
725
726
727/*********************************************************/
728/* snapshots and image creation */
729
730
731
f7d0fe02 732/* update the refcounts of snapshots and the copied flag */
ed6ccf0f
KW
733int qcow2_update_snapshot_refcount(BlockDriverState *bs,
734 int64_t l1_table_offset, int l1_size, int addend)
f7d0fe02
KW
735{
736 BDRVQcowState *s = bs->opaque;
737 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
738 int64_t old_offset, old_l2_offset;
93913dfd 739 int i, j, l1_modified = 0, nb_csectors, refcount;
29c1a730 740 int ret;
f7d0fe02
KW
741
742 l2_table = NULL;
743 l1_table = NULL;
744 l1_size2 = l1_size * sizeof(uint64_t);
43a0cac4
KW
745
746 /* WARNING: qcow2_snapshot_goto relies on this function not using the
747 * l1_table_offset when it is the current s->l1_table_offset! Be careful
748 * when changing this! */
f7d0fe02 749 if (l1_table_offset != s->l1_table_offset) {
6528499f 750 l1_table = g_malloc0(align_offset(l1_size2, 512));
f7d0fe02 751 l1_allocated = 1;
66f82cee 752 if (bdrv_pread(bs->file, l1_table_offset,
f7d0fe02 753 l1_table, l1_size2) != l1_size2)
93913dfd
KW
754 {
755 ret = -EIO;
f7d0fe02 756 goto fail;
93913dfd
KW
757 }
758
f7d0fe02
KW
759 for(i = 0;i < l1_size; i++)
760 be64_to_cpus(&l1_table[i]);
761 } else {
762 assert(l1_size == s->l1_size);
763 l1_table = s->l1_table;
764 l1_allocated = 0;
765 }
766
f7d0fe02
KW
767 for(i = 0; i < l1_size; i++) {
768 l2_offset = l1_table[i];
769 if (l2_offset) {
770 old_l2_offset = l2_offset;
8e37f681 771 l2_offset &= L1E_OFFSET_MASK;
29c1a730
KW
772
773 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
774 (void**) &l2_table);
775 if (ret < 0) {
f7d0fe02 776 goto fail;
29c1a730
KW
777 }
778
f7d0fe02
KW
779 for(j = 0; j < s->l2_size; j++) {
780 offset = be64_to_cpu(l2_table[j]);
781 if (offset != 0) {
782 old_offset = offset;
783 offset &= ~QCOW_OFLAG_COPIED;
784 if (offset & QCOW_OFLAG_COMPRESSED) {
785 nb_csectors = ((offset >> s->csize_shift) &
786 s->csize_mask) + 1;
db3a964f
KW
787 if (addend != 0) {
788 int ret;
789 ret = update_refcount(bs,
790 (offset & s->cluster_offset_mask) & ~511,
791 nb_csectors * 512, addend);
792 if (ret < 0) {
793 goto fail;
794 }
1c4c2814
KW
795
796 /* TODO Flushing once for the whole function should
797 * be enough */
798 bdrv_flush(bs->file);
db3a964f 799 }
f7d0fe02
KW
800 /* compressed clusters are never modified */
801 refcount = 2;
802 } else {
8e37f681 803 uint64_t cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
f7d0fe02 804 if (addend != 0) {
8e37f681 805 refcount = update_cluster_refcount(bs, cluster_index, addend);
f7d0fe02 806 } else {
8e37f681 807 refcount = get_refcount(bs, cluster_index);
f7d0fe02 808 }
018faafd
KW
809
810 if (refcount < 0) {
93913dfd 811 ret = -EIO;
018faafd
KW
812 goto fail;
813 }
f7d0fe02
KW
814 }
815
816 if (refcount == 1) {
817 offset |= QCOW_OFLAG_COPIED;
818 }
819 if (offset != old_offset) {
29c1a730
KW
820 if (addend > 0) {
821 qcow2_cache_set_dependency(bs, s->l2_table_cache,
822 s->refcount_block_cache);
823 }
f7d0fe02 824 l2_table[j] = cpu_to_be64(offset);
29c1a730 825 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
f7d0fe02
KW
826 }
827 }
828 }
29c1a730
KW
829
830 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
831 if (ret < 0) {
832 goto fail;
f7d0fe02
KW
833 }
834
29c1a730 835
f7d0fe02
KW
836 if (addend != 0) {
837 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
838 } else {
839 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
840 }
018faafd 841 if (refcount < 0) {
93913dfd 842 ret = -EIO;
018faafd
KW
843 goto fail;
844 } else if (refcount == 1) {
f7d0fe02
KW
845 l2_offset |= QCOW_OFLAG_COPIED;
846 }
847 if (l2_offset != old_l2_offset) {
848 l1_table[i] = l2_offset;
849 l1_modified = 1;
850 }
851 }
852 }
93913dfd 853
2154f24e 854 ret = bdrv_flush(bs);
93913dfd
KW
855fail:
856 if (l2_table) {
857 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
858 }
859
43a0cac4
KW
860 /* Update L1 only if it isn't deleted anyway (addend = -1) */
861 if (addend >= 0 && l1_modified) {
f7d0fe02
KW
862 for(i = 0; i < l1_size; i++)
863 cpu_to_be64s(&l1_table[i]);
8b3b7206
KW
864 if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table,
865 l1_size2) < 0)
f7d0fe02
KW
866 goto fail;
867 for(i = 0; i < l1_size; i++)
868 be64_to_cpus(&l1_table[i]);
869 }
870 if (l1_allocated)
7267c094 871 g_free(l1_table);
93913dfd 872 return ret;
f7d0fe02
KW
873}
874
875
876
877
878/*********************************************************/
879/* refcount checking functions */
880
881
882
883/*
884 * Increases the refcount for a range of clusters in a given refcount table.
885 * This is used to construct a temporary refcount table out of L1 and L2 tables
886 * which can be compared the the refcount table saved in the image.
887 *
9ac228e0 888 * Modifies the number of errors in res.
f7d0fe02 889 */
9ac228e0
KW
890static void inc_refcounts(BlockDriverState *bs,
891 BdrvCheckResult *res,
f7d0fe02
KW
892 uint16_t *refcount_table,
893 int refcount_table_size,
894 int64_t offset, int64_t size)
895{
896 BDRVQcowState *s = bs->opaque;
897 int64_t start, last, cluster_offset;
898 int k;
f7d0fe02
KW
899
900 if (size <= 0)
9ac228e0 901 return;
f7d0fe02
KW
902
903 start = offset & ~(s->cluster_size - 1);
904 last = (offset + size - 1) & ~(s->cluster_size - 1);
905 for(cluster_offset = start; cluster_offset <= last;
906 cluster_offset += s->cluster_size) {
907 k = cluster_offset >> s->cluster_bits;
9ac228e0 908 if (k < 0) {
f7d0fe02
KW
909 fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
910 cluster_offset);
9ac228e0
KW
911 res->corruptions++;
912 } else if (k >= refcount_table_size) {
913 fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after "
914 "the end of the image file, can't properly check refcounts.\n",
915 cluster_offset);
916 res->check_errors++;
f7d0fe02
KW
917 } else {
918 if (++refcount_table[k] == 0) {
919 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
920 "\n", cluster_offset);
9ac228e0 921 res->corruptions++;
f7d0fe02
KW
922 }
923 }
924 }
f7d0fe02
KW
925}
926
801f7044
SH
927/* Flags for check_refcounts_l1() and check_refcounts_l2() */
928enum {
929 CHECK_OFLAG_COPIED = 0x1, /* check QCOW_OFLAG_COPIED matches refcount */
fba31bae 930 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
801f7044
SH
931};
932
f7d0fe02
KW
933/*
934 * Increases the refcount in the given refcount table for the all clusters
935 * referenced in the L2 table. While doing so, performs some checks on L2
936 * entries.
937 *
938 * Returns the number of errors found by the checks or -errno if an internal
939 * error occurred.
940 */
9ac228e0 941static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
f7d0fe02 942 uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
801f7044 943 int flags)
f7d0fe02
KW
944{
945 BDRVQcowState *s = bs->opaque;
afdf0abe 946 uint64_t *l2_table, l2_entry;
fba31bae 947 uint64_t next_contiguous_offset = 0;
f7d0fe02 948 int i, l2_size, nb_csectors, refcount;
f7d0fe02
KW
949
950 /* Read L2 table from disk */
951 l2_size = s->l2_size * sizeof(uint64_t);
7267c094 952 l2_table = g_malloc(l2_size);
f7d0fe02 953
66f82cee 954 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
f7d0fe02
KW
955 goto fail;
956
957 /* Do the actual checks */
958 for(i = 0; i < s->l2_size; i++) {
afdf0abe
KW
959 l2_entry = be64_to_cpu(l2_table[i]);
960
961 switch (qcow2_get_cluster_type(l2_entry)) {
962 case QCOW2_CLUSTER_COMPRESSED:
963 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
964 if (l2_entry & QCOW_OFLAG_COPIED) {
965 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
966 "copied flag must never be set for compressed "
967 "clusters\n", l2_entry >> s->cluster_bits);
968 l2_entry &= ~QCOW_OFLAG_COPIED;
969 res->corruptions++;
970 }
f7d0fe02 971
afdf0abe
KW
972 /* Mark cluster as used */
973 nb_csectors = ((l2_entry >> s->csize_shift) &
974 s->csize_mask) + 1;
975 l2_entry &= s->cluster_offset_mask;
976 inc_refcounts(bs, res, refcount_table, refcount_table_size,
977 l2_entry & ~511, nb_csectors * 512);
fba31bae
SH
978
979 if (flags & CHECK_FRAG_INFO) {
980 res->bfi.allocated_clusters++;
4db35162 981 res->bfi.compressed_clusters++;
fba31bae
SH
982
983 /* Compressed clusters are fragmented by nature. Since they
984 * take up sub-sector space but we only have sector granularity
985 * I/O we need to re-read the same sectors even for adjacent
986 * compressed clusters.
987 */
988 res->bfi.fragmented_clusters++;
989 }
afdf0abe 990 break;
f7d0fe02 991
6377af48
KW
992 case QCOW2_CLUSTER_ZERO:
993 if ((l2_entry & L2E_OFFSET_MASK) == 0) {
994 break;
995 }
996 /* fall through */
997
afdf0abe
KW
998 case QCOW2_CLUSTER_NORMAL:
999 {
1000 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
1001 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
f7d0fe02 1002
801f7044 1003 if (flags & CHECK_OFLAG_COPIED) {
afdf0abe
KW
1004 refcount = get_refcount(bs, offset >> s->cluster_bits);
1005 if (refcount < 0) {
1006 fprintf(stderr, "Can't get refcount for offset %"
1007 PRIx64 ": %s\n", l2_entry, strerror(-refcount));
1008 goto fail;
1009 }
1010 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
1011 fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
1012 PRIx64 " refcount=%d\n", l2_entry, refcount);
9ac228e0 1013 res->corruptions++;
f7d0fe02
KW
1014 }
1015 }
afdf0abe 1016
fba31bae
SH
1017 if (flags & CHECK_FRAG_INFO) {
1018 res->bfi.allocated_clusters++;
1019 if (next_contiguous_offset &&
1020 offset != next_contiguous_offset) {
1021 res->bfi.fragmented_clusters++;
1022 }
1023 next_contiguous_offset = offset + s->cluster_size;
1024 }
1025
afdf0abe
KW
1026 /* Mark cluster as used */
1027 inc_refcounts(bs, res, refcount_table,refcount_table_size,
1028 offset, s->cluster_size);
1029
1030 /* Correct offsets are cluster aligned */
1031 if (offset & (s->cluster_size - 1)) {
1032 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
1033 "properly aligned; L2 entry corrupted.\n", offset);
1034 res->corruptions++;
1035 }
1036 break;
1037 }
1038
1039 case QCOW2_CLUSTER_UNALLOCATED:
1040 break;
1041
1042 default:
1043 abort();
f7d0fe02
KW
1044 }
1045 }
1046
7267c094 1047 g_free(l2_table);
9ac228e0 1048 return 0;
f7d0fe02
KW
1049
1050fail:
9ac228e0 1051 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
7267c094 1052 g_free(l2_table);
f7d0fe02
KW
1053 return -EIO;
1054}
1055
1056/*
1057 * Increases the refcount for the L1 table, its L2 tables and all referenced
1058 * clusters in the given refcount table. While doing so, performs some checks
1059 * on L1 and L2 entries.
1060 *
1061 * Returns the number of errors found by the checks or -errno if an internal
1062 * error occurred.
1063 */
1064static int check_refcounts_l1(BlockDriverState *bs,
9ac228e0 1065 BdrvCheckResult *res,
f7d0fe02
KW
1066 uint16_t *refcount_table,
1067 int refcount_table_size,
1068 int64_t l1_table_offset, int l1_size,
801f7044 1069 int flags)
f7d0fe02
KW
1070{
1071 BDRVQcowState *s = bs->opaque;
1072 uint64_t *l1_table, l2_offset, l1_size2;
1073 int i, refcount, ret;
f7d0fe02
KW
1074
1075 l1_size2 = l1_size * sizeof(uint64_t);
1076
1077 /* Mark L1 table as used */
9ac228e0
KW
1078 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1079 l1_table_offset, l1_size2);
f7d0fe02
KW
1080
1081 /* Read L1 table entries from disk */
702ef63f
KW
1082 if (l1_size2 == 0) {
1083 l1_table = NULL;
1084 } else {
7267c094 1085 l1_table = g_malloc(l1_size2);
66f82cee 1086 if (bdrv_pread(bs->file, l1_table_offset,
702ef63f
KW
1087 l1_table, l1_size2) != l1_size2)
1088 goto fail;
1089 for(i = 0;i < l1_size; i++)
1090 be64_to_cpus(&l1_table[i]);
1091 }
f7d0fe02
KW
1092
1093 /* Do the actual checks */
1094 for(i = 0; i < l1_size; i++) {
1095 l2_offset = l1_table[i];
1096 if (l2_offset) {
1097 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
801f7044 1098 if (flags & CHECK_OFLAG_COPIED) {
f7d0fe02
KW
1099 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
1100 >> s->cluster_bits);
018faafd
KW
1101 if (refcount < 0) {
1102 fprintf(stderr, "Can't get refcount for l2_offset %"
1103 PRIx64 ": %s\n", l2_offset, strerror(-refcount));
9ac228e0 1104 goto fail;
018faafd 1105 }
f7d0fe02
KW
1106 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
1107 fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
1108 " refcount=%d\n", l2_offset, refcount);
9ac228e0 1109 res->corruptions++;
f7d0fe02
KW
1110 }
1111 }
1112
1113 /* Mark L2 table as used */
afdf0abe 1114 l2_offset &= L1E_OFFSET_MASK;
9ac228e0
KW
1115 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1116 l2_offset, s->cluster_size);
f7d0fe02
KW
1117
1118 /* L2 tables are cluster aligned */
1119 if (l2_offset & (s->cluster_size - 1)) {
1120 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1121 "cluster aligned; L1 entry corrupted\n", l2_offset);
9ac228e0 1122 res->corruptions++;
f7d0fe02
KW
1123 }
1124
1125 /* Process and check L2 entries */
9ac228e0 1126 ret = check_refcounts_l2(bs, res, refcount_table,
801f7044 1127 refcount_table_size, l2_offset, flags);
f7d0fe02
KW
1128 if (ret < 0) {
1129 goto fail;
1130 }
f7d0fe02
KW
1131 }
1132 }
7267c094 1133 g_free(l1_table);
9ac228e0 1134 return 0;
f7d0fe02
KW
1135
1136fail:
1137 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
9ac228e0 1138 res->check_errors++;
7267c094 1139 g_free(l1_table);
f7d0fe02
KW
1140 return -EIO;
1141}
1142
1143/*
1144 * Checks an image for refcount consistency.
1145 *
1146 * Returns 0 if no errors are found, the number of errors in case the image is
a1c7273b 1147 * detected as corrupted, and -errno when an internal error occurred.
f7d0fe02 1148 */
166acf54
KW
1149int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1150 BdrvCheckMode fix)
f7d0fe02
KW
1151{
1152 BDRVQcowState *s = bs->opaque;
c6bb9ad1 1153 int64_t size, i, highest_cluster;
166acf54 1154 int nb_clusters, refcount1, refcount2;
f7d0fe02
KW
1155 QCowSnapshot *sn;
1156 uint16_t *refcount_table;
9ac228e0 1157 int ret;
f7d0fe02 1158
66f82cee 1159 size = bdrv_getlength(bs->file);
f7d0fe02 1160 nb_clusters = size_to_clusters(s, size);
fba31bae 1161 res->bfi.total_clusters = nb_clusters;
7267c094 1162 refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t));
f7d0fe02
KW
1163
1164 /* header */
9ac228e0
KW
1165 inc_refcounts(bs, res, refcount_table, nb_clusters,
1166 0, s->cluster_size);
f7d0fe02
KW
1167
1168 /* current L1 table */
9ac228e0 1169 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
801f7044 1170 s->l1_table_offset, s->l1_size,
fba31bae 1171 CHECK_OFLAG_COPIED | CHECK_FRAG_INFO);
f7d0fe02 1172 if (ret < 0) {
80fa3341 1173 goto fail;
f7d0fe02 1174 }
f7d0fe02
KW
1175
1176 /* snapshots */
1177 for(i = 0; i < s->nb_snapshots; i++) {
1178 sn = s->snapshots + i;
9ac228e0
KW
1179 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1180 sn->l1_table_offset, sn->l1_size, 0);
1181 if (ret < 0) {
80fa3341 1182 goto fail;
9ac228e0 1183 }
f7d0fe02 1184 }
9ac228e0
KW
1185 inc_refcounts(bs, res, refcount_table, nb_clusters,
1186 s->snapshots_offset, s->snapshots_size);
f7d0fe02
KW
1187
1188 /* refcount data */
9ac228e0
KW
1189 inc_refcounts(bs, res, refcount_table, nb_clusters,
1190 s->refcount_table_offset,
1191 s->refcount_table_size * sizeof(uint64_t));
1192
f7d0fe02 1193 for(i = 0; i < s->refcount_table_size; i++) {
6882c8fa 1194 uint64_t offset, cluster;
f7d0fe02 1195 offset = s->refcount_table[i];
6882c8fa 1196 cluster = offset >> s->cluster_bits;
746c3cb5
KW
1197
1198 /* Refcount blocks are cluster aligned */
1199 if (offset & (s->cluster_size - 1)) {
166acf54 1200 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
746c3cb5 1201 "cluster aligned; refcount table entry corrupted\n", i);
9ac228e0 1202 res->corruptions++;
6882c8fa
KW
1203 continue;
1204 }
1205
1206 if (cluster >= nb_clusters) {
166acf54
KW
1207 fprintf(stderr, "ERROR refcount block %" PRId64
1208 " is outside image\n", i);
9ac228e0 1209 res->corruptions++;
6882c8fa 1210 continue;
746c3cb5
KW
1211 }
1212
f7d0fe02 1213 if (offset != 0) {
9ac228e0
KW
1214 inc_refcounts(bs, res, refcount_table, nb_clusters,
1215 offset, s->cluster_size);
6882c8fa 1216 if (refcount_table[cluster] != 1) {
166acf54
KW
1217 fprintf(stderr, "ERROR refcount block %" PRId64
1218 " refcount=%d\n",
6882c8fa 1219 i, refcount_table[cluster]);
9ac228e0 1220 res->corruptions++;
746c3cb5 1221 }
f7d0fe02
KW
1222 }
1223 }
1224
1225 /* compare ref counts */
c6bb9ad1 1226 for (i = 0, highest_cluster = 0; i < nb_clusters; i++) {
f7d0fe02 1227 refcount1 = get_refcount(bs, i);
018faafd 1228 if (refcount1 < 0) {
166acf54 1229 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
018faafd 1230 i, strerror(-refcount1));
9ac228e0 1231 res->check_errors++;
f74550fd 1232 continue;
018faafd
KW
1233 }
1234
f7d0fe02 1235 refcount2 = refcount_table[i];
c6bb9ad1
FS
1236
1237 if (refcount1 > 0 || refcount2 > 0) {
1238 highest_cluster = i;
1239 }
1240
f7d0fe02 1241 if (refcount1 != refcount2) {
166acf54
KW
1242
1243 /* Check if we're allowed to fix the mismatch */
1244 int *num_fixed = NULL;
1245 if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
1246 num_fixed = &res->leaks_fixed;
1247 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
1248 num_fixed = &res->corruptions_fixed;
1249 }
1250
1251 fprintf(stderr, "%s cluster %" PRId64 " refcount=%d reference=%d\n",
1252 num_fixed != NULL ? "Repairing" :
1253 refcount1 < refcount2 ? "ERROR" :
1254 "Leaked",
f7d0fe02 1255 i, refcount1, refcount2);
166acf54
KW
1256
1257 if (num_fixed) {
1258 ret = update_refcount(bs, i << s->cluster_bits, 1,
1259 refcount2 - refcount1);
1260 if (ret >= 0) {
1261 (*num_fixed)++;
1262 continue;
1263 }
1264 }
1265
1266 /* And if we couldn't, print an error */
9ac228e0
KW
1267 if (refcount1 < refcount2) {
1268 res->corruptions++;
1269 } else {
1270 res->leaks++;
1271 }
f7d0fe02
KW
1272 }
1273 }
1274
c6bb9ad1 1275 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
80fa3341
KW
1276 ret = 0;
1277
1278fail:
7267c094 1279 g_free(refcount_table);
f7d0fe02 1280
80fa3341 1281 return ret;
f7d0fe02
KW
1282}
1283