]> git.proxmox.com Git - qemu.git/blob - block/qcow2-refcount.c
92519ea62a7269e4c26d86a8b8df8d01f9dad7a9
[qemu.git] / block / qcow2-refcount.c
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"
26 #include "block/block_int.h"
27 #include "block/qcow2.h"
28
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);
33
34
35 /*********************************************************/
36 /* refcount handling */
37
38 int qcow2_refcount_init(BlockDriverState *bs)
39 {
40 BDRVQcowState *s = bs->opaque;
41 int ret, refcount_table_size2, i;
42
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]);
53 }
54 return 0;
55 fail:
56 return -ENOMEM;
57 }
58
59 void qcow2_refcount_close(BlockDriverState *bs)
60 {
61 BDRVQcowState *s = bs->opaque;
62 g_free(s->refcount_table);
63 }
64
65
66 static int load_refcount_block(BlockDriverState *bs,
67 int64_t refcount_block_offset,
68 void **refcount_block)
69 {
70 BDRVQcowState *s = bs->opaque;
71 int ret;
72
73 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
74 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
75 refcount_block);
76
77 return ret;
78 }
79
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 */
85 static 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;
90 int ret;
91 uint16_t *refcount_block;
92 uint16_t refcount;
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;
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;
105 }
106
107 block_index = cluster_index &
108 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
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;
118 }
119
120 /*
121 * Rounds the refcount table size up to avoid growing the table for each single
122 * refcount block that is allocated.
123 */
124 static 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
138
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)
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 *
153 * Returns 0 on success or -errno in error case
154 */
155 static int alloc_refcount_block(BlockDriverState *bs,
156 int64_t cluster_index, uint16_t **refcount_block)
157 {
158 BDRVQcowState *s = bs->opaque;
159 unsigned int refcount_table_index;
160 int ret;
161
162 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
163
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 =
170 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
171
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);
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
201 *refcount_block = NULL;
202
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;
207 }
208
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;
213 }
214
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
220
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;
227 }
228
229 memset(*refcount_block, 0, s->cluster_size);
230
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;
241 }
242
243 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
244 if (ret < 0) {
245 goto fail_block;
246 }
247
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;
254 }
255
256 memset(*refcount_block, 0, s->cluster_size);
257 }
258
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;
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);
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;
276 }
277
278 s->refcount_table[refcount_table_index] = new_block;
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;
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 */
298 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
299
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 {
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;
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;
333 uint16_t *new_blocks = g_malloc0(blocks_clusters * s->cluster_size);
334 uint64_t *new_table = g_malloc0(table_size * sizeof(uint64_t));
335
336 assert(meta_offset >= (s->free_cluster_index * s->cluster_size));
337
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;
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 */
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;
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
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;
374 }
375
376 for(i = 0; i < table_size; i++) {
377 be64_to_cpus(&new_table[i]);
378 }
379
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;
389 }
390
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
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;
399
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;
404
405 ret = load_refcount_block(bs, new_block, (void**) refcount_block);
406 if (ret < 0) {
407 return ret;
408 }
409
410 return 0;
411
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);
417 }
418 return ret;
419 }
420
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)
424 {
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;
430
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;
439 }
440
441 if (addend < 0) {
442 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
443 s->l2_table_cache);
444 }
445
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;
453 int64_t table_index =
454 cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
455
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 }
465
466 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
467 if (ret < 0) {
468 goto fail;
469 }
470 }
471 old_table_index = table_index;
472
473 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
474
475 /* we can update the count and save it */
476 block_index = cluster_index &
477 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
478
479 refcount = be16_to_cpu(refcount_block[block_index]);
480 refcount += addend;
481 if (refcount < 0 || refcount > 0xffff) {
482 ret = -EINVAL;
483 goto fail;
484 }
485 if (refcount == 0 && cluster_index < s->free_cluster_index) {
486 s->free_cluster_index = cluster_index;
487 }
488 refcount_block[block_index] = cpu_to_be16(refcount);
489 }
490
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;
500 }
501 }
502
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);
510 (void)dummy;
511 }
512
513 return ret;
514 }
515
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 */
523 static 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
535 return get_refcount(bs, cluster_index);
536 }
537
538
539
540 /*********************************************************/
541 /* cluster allocation functions */
542
543
544
545 /* return < 0 if error */
546 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
547 {
548 BDRVQcowState *s = bs->opaque;
549 int i, nb_clusters, refcount;
550
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);
556
557 if (refcount < 0) {
558 return refcount;
559 } else if (refcount != 0) {
560 goto retry;
561 }
562 }
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;
569 }
570
571 int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
572 {
573 int64_t offset;
574 int ret;
575
576 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
577 offset = alloc_clusters_noref(bs, size);
578 if (offset < 0) {
579 return offset;
580 }
581
582 ret = update_refcount(bs, offset, size, 1);
583 if (ret < 0) {
584 return ret;
585 }
586
587 return offset;
588 }
589
590 int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
591 int nb_clusters)
592 {
593 BDRVQcowState *s = bs->opaque;
594 uint64_t cluster_index;
595 uint64_t old_free_cluster_index;
596 int i, refcount, ret;
597
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++);
602
603 if (refcount < 0) {
604 return refcount;
605 } else if (refcount != 0) {
606 break;
607 }
608 }
609
610 /* And then allocate them */
611 old_free_cluster_index = s->free_cluster_index;
612 s->free_cluster_index = cluster_index + i;
613
614 ret = update_refcount(bs, offset, i << s->cluster_bits, 1);
615 if (ret < 0) {
616 return ret;
617 }
618
619 s->free_cluster_index = old_free_cluster_index;
620
621 return i;
622 }
623
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)
627 {
628 BDRVQcowState *s = bs->opaque;
629 int64_t offset, cluster_offset;
630 int free_in_cluster;
631
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;
638 }
639 s->free_byte_offset = offset;
640 }
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;
657 }
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;
667 }
668 }
669
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.
673 */
674 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
675 return offset;
676 }
677
678 void qcow2_free_clusters(BlockDriverState *bs,
679 int64_t offset, int64_t size)
680 {
681 int ret;
682
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 */
688 }
689 }
690
691 /*
692 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
693 * normal cluster, compressed cluster, etc.)
694 */
695 void qcow2_free_any_clusters(BlockDriverState *bs,
696 uint64_t l2_entry, int nb_clusters)
697 {
698 BDRVQcowState *s = bs->opaque;
699
700 switch (qcow2_get_cluster_type(l2_entry)) {
701 case QCOW2_CLUSTER_COMPRESSED:
702 {
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);
709 }
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();
720 }
721 }
722
723
724
725 /*********************************************************/
726 /* snapshots and image creation */
727
728
729
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)
733 {
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;
739
740 l2_table = NULL;
741 l1_table = NULL;
742 l1_size2 = l1_size * sizeof(uint64_t);
743
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;
750 if (bdrv_pread(bs->file, l1_table_offset,
751 l1_table, l1_size2) != l1_size2)
752 {
753 ret = -EIO;
754 goto fail;
755 }
756
757 for(i = 0;i < l1_size; i++)
758 be64_to_cpus(&l1_table[i]);
759 } else {
760 assert(l1_size == s->l1_size);
761 l1_table = s->l1_table;
762 l1_allocated = 0;
763 }
764
765 for(i = 0; i < l1_size; i++) {
766 l2_offset = l1_table[i];
767 if (l2_offset) {
768 old_l2_offset = l2_offset;
769 l2_offset &= L1E_OFFSET_MASK;
770
771 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
772 (void**) &l2_table);
773 if (ret < 0) {
774 goto fail;
775 }
776
777 for(j = 0; j < s->l2_size; j++) {
778 offset = be64_to_cpu(l2_table[j]);
779 if (offset != 0) {
780 old_offset = offset;
781 offset &= ~QCOW_OFLAG_COPIED;
782 if (offset & QCOW_OFLAG_COMPRESSED) {
783 nb_csectors = ((offset >> s->csize_shift) &
784 s->csize_mask) + 1;
785 if (addend != 0) {
786 int ret;
787 ret = update_refcount(bs,
788 (offset & s->cluster_offset_mask) & ~511,
789 nb_csectors * 512, addend);
790 if (ret < 0) {
791 goto fail;
792 }
793
794 /* TODO Flushing once for the whole function should
795 * be enough */
796 bdrv_flush(bs->file);
797 }
798 /* compressed clusters are never modified */
799 refcount = 2;
800 } else {
801 uint64_t cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
802 if (addend != 0) {
803 refcount = update_cluster_refcount(bs, cluster_index, addend);
804 } else {
805 refcount = get_refcount(bs, cluster_index);
806 }
807
808 if (refcount < 0) {
809 ret = -EIO;
810 goto fail;
811 }
812 }
813
814 if (refcount == 1) {
815 offset |= QCOW_OFLAG_COPIED;
816 }
817 if (offset != old_offset) {
818 if (addend > 0) {
819 qcow2_cache_set_dependency(bs, s->l2_table_cache,
820 s->refcount_block_cache);
821 }
822 l2_table[j] = cpu_to_be64(offset);
823 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
824 }
825 }
826 }
827
828 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
829 if (ret < 0) {
830 goto fail;
831 }
832
833
834 if (addend != 0) {
835 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
836 } else {
837 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
838 }
839 if (refcount < 0) {
840 ret = -EIO;
841 goto fail;
842 } else if (refcount == 1) {
843 l2_offset |= QCOW_OFLAG_COPIED;
844 }
845 if (l2_offset != old_l2_offset) {
846 l1_table[i] = l2_offset;
847 l1_modified = 1;
848 }
849 }
850 }
851
852 ret = bdrv_flush(bs);
853 fail:
854 if (l2_table) {
855 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
856 }
857
858 /* Update L1 only if it isn't deleted anyway (addend = -1) */
859 if (addend >= 0 && l1_modified) {
860 for(i = 0; i < l1_size; i++)
861 cpu_to_be64s(&l1_table[i]);
862 if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table,
863 l1_size2) < 0)
864 goto fail;
865 for(i = 0; i < l1_size; i++)
866 be64_to_cpus(&l1_table[i]);
867 }
868 if (l1_allocated)
869 g_free(l1_table);
870 return ret;
871 }
872
873
874
875
876 /*********************************************************/
877 /* refcount checking functions */
878
879
880
881 /*
882 * Increases the refcount for a range of clusters in a given refcount table.
883 * This is used to construct a temporary refcount table out of L1 and L2 tables
884 * which can be compared the the refcount table saved in the image.
885 *
886 * Modifies the number of errors in res.
887 */
888 static void inc_refcounts(BlockDriverState *bs,
889 BdrvCheckResult *res,
890 uint16_t *refcount_table,
891 int refcount_table_size,
892 int64_t offset, int64_t size)
893 {
894 BDRVQcowState *s = bs->opaque;
895 int64_t start, last, cluster_offset;
896 int k;
897
898 if (size <= 0)
899 return;
900
901 start = offset & ~(s->cluster_size - 1);
902 last = (offset + size - 1) & ~(s->cluster_size - 1);
903 for(cluster_offset = start; cluster_offset <= last;
904 cluster_offset += s->cluster_size) {
905 k = cluster_offset >> s->cluster_bits;
906 if (k < 0) {
907 fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
908 cluster_offset);
909 res->corruptions++;
910 } else if (k >= refcount_table_size) {
911 fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after "
912 "the end of the image file, can't properly check refcounts.\n",
913 cluster_offset);
914 res->check_errors++;
915 } else {
916 if (++refcount_table[k] == 0) {
917 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
918 "\n", cluster_offset);
919 res->corruptions++;
920 }
921 }
922 }
923 }
924
925 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
926 enum {
927 CHECK_OFLAG_COPIED = 0x1, /* check QCOW_OFLAG_COPIED matches refcount */
928 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
929 };
930
931 /*
932 * Increases the refcount in the given refcount table for the all clusters
933 * referenced in the L2 table. While doing so, performs some checks on L2
934 * entries.
935 *
936 * Returns the number of errors found by the checks or -errno if an internal
937 * error occurred.
938 */
939 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
940 uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
941 int flags)
942 {
943 BDRVQcowState *s = bs->opaque;
944 uint64_t *l2_table, l2_entry;
945 uint64_t next_contiguous_offset = 0;
946 int i, l2_size, nb_csectors, refcount;
947
948 /* Read L2 table from disk */
949 l2_size = s->l2_size * sizeof(uint64_t);
950 l2_table = g_malloc(l2_size);
951
952 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
953 goto fail;
954
955 /* Do the actual checks */
956 for(i = 0; i < s->l2_size; i++) {
957 l2_entry = be64_to_cpu(l2_table[i]);
958
959 switch (qcow2_get_cluster_type(l2_entry)) {
960 case QCOW2_CLUSTER_COMPRESSED:
961 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
962 if (l2_entry & QCOW_OFLAG_COPIED) {
963 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
964 "copied flag must never be set for compressed "
965 "clusters\n", l2_entry >> s->cluster_bits);
966 l2_entry &= ~QCOW_OFLAG_COPIED;
967 res->corruptions++;
968 }
969
970 /* Mark cluster as used */
971 nb_csectors = ((l2_entry >> s->csize_shift) &
972 s->csize_mask) + 1;
973 l2_entry &= s->cluster_offset_mask;
974 inc_refcounts(bs, res, refcount_table, refcount_table_size,
975 l2_entry & ~511, nb_csectors * 512);
976
977 if (flags & CHECK_FRAG_INFO) {
978 res->bfi.allocated_clusters++;
979 res->bfi.compressed_clusters++;
980
981 /* Compressed clusters are fragmented by nature. Since they
982 * take up sub-sector space but we only have sector granularity
983 * I/O we need to re-read the same sectors even for adjacent
984 * compressed clusters.
985 */
986 res->bfi.fragmented_clusters++;
987 }
988 break;
989
990 case QCOW2_CLUSTER_ZERO:
991 if ((l2_entry & L2E_OFFSET_MASK) == 0) {
992 break;
993 }
994 /* fall through */
995
996 case QCOW2_CLUSTER_NORMAL:
997 {
998 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
999 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
1000
1001 if (flags & CHECK_OFLAG_COPIED) {
1002 refcount = get_refcount(bs, offset >> s->cluster_bits);
1003 if (refcount < 0) {
1004 fprintf(stderr, "Can't get refcount for offset %"
1005 PRIx64 ": %s\n", l2_entry, strerror(-refcount));
1006 goto fail;
1007 }
1008 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
1009 fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
1010 PRIx64 " refcount=%d\n", l2_entry, refcount);
1011 res->corruptions++;
1012 }
1013 }
1014
1015 if (flags & CHECK_FRAG_INFO) {
1016 res->bfi.allocated_clusters++;
1017 if (next_contiguous_offset &&
1018 offset != next_contiguous_offset) {
1019 res->bfi.fragmented_clusters++;
1020 }
1021 next_contiguous_offset = offset + s->cluster_size;
1022 }
1023
1024 /* Mark cluster as used */
1025 inc_refcounts(bs, res, refcount_table,refcount_table_size,
1026 offset, s->cluster_size);
1027
1028 /* Correct offsets are cluster aligned */
1029 if (offset & (s->cluster_size - 1)) {
1030 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
1031 "properly aligned; L2 entry corrupted.\n", offset);
1032 res->corruptions++;
1033 }
1034 break;
1035 }
1036
1037 case QCOW2_CLUSTER_UNALLOCATED:
1038 break;
1039
1040 default:
1041 abort();
1042 }
1043 }
1044
1045 g_free(l2_table);
1046 return 0;
1047
1048 fail:
1049 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1050 g_free(l2_table);
1051 return -EIO;
1052 }
1053
1054 /*
1055 * Increases the refcount for the L1 table, its L2 tables and all referenced
1056 * clusters in the given refcount table. While doing so, performs some checks
1057 * on L1 and L2 entries.
1058 *
1059 * Returns the number of errors found by the checks or -errno if an internal
1060 * error occurred.
1061 */
1062 static int check_refcounts_l1(BlockDriverState *bs,
1063 BdrvCheckResult *res,
1064 uint16_t *refcount_table,
1065 int refcount_table_size,
1066 int64_t l1_table_offset, int l1_size,
1067 int flags)
1068 {
1069 BDRVQcowState *s = bs->opaque;
1070 uint64_t *l1_table, l2_offset, l1_size2;
1071 int i, refcount, ret;
1072
1073 l1_size2 = l1_size * sizeof(uint64_t);
1074
1075 /* Mark L1 table as used */
1076 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1077 l1_table_offset, l1_size2);
1078
1079 /* Read L1 table entries from disk */
1080 if (l1_size2 == 0) {
1081 l1_table = NULL;
1082 } else {
1083 l1_table = g_malloc(l1_size2);
1084 if (bdrv_pread(bs->file, l1_table_offset,
1085 l1_table, l1_size2) != l1_size2)
1086 goto fail;
1087 for(i = 0;i < l1_size; i++)
1088 be64_to_cpus(&l1_table[i]);
1089 }
1090
1091 /* Do the actual checks */
1092 for(i = 0; i < l1_size; i++) {
1093 l2_offset = l1_table[i];
1094 if (l2_offset) {
1095 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
1096 if (flags & CHECK_OFLAG_COPIED) {
1097 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
1098 >> s->cluster_bits);
1099 if (refcount < 0) {
1100 fprintf(stderr, "Can't get refcount for l2_offset %"
1101 PRIx64 ": %s\n", l2_offset, strerror(-refcount));
1102 goto fail;
1103 }
1104 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
1105 fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
1106 " refcount=%d\n", l2_offset, refcount);
1107 res->corruptions++;
1108 }
1109 }
1110
1111 /* Mark L2 table as used */
1112 l2_offset &= L1E_OFFSET_MASK;
1113 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1114 l2_offset, s->cluster_size);
1115
1116 /* L2 tables are cluster aligned */
1117 if (l2_offset & (s->cluster_size - 1)) {
1118 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1119 "cluster aligned; L1 entry corrupted\n", l2_offset);
1120 res->corruptions++;
1121 }
1122
1123 /* Process and check L2 entries */
1124 ret = check_refcounts_l2(bs, res, refcount_table,
1125 refcount_table_size, l2_offset, flags);
1126 if (ret < 0) {
1127 goto fail;
1128 }
1129 }
1130 }
1131 g_free(l1_table);
1132 return 0;
1133
1134 fail:
1135 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1136 res->check_errors++;
1137 g_free(l1_table);
1138 return -EIO;
1139 }
1140
1141 /*
1142 * Checks an image for refcount consistency.
1143 *
1144 * Returns 0 if no errors are found, the number of errors in case the image is
1145 * detected as corrupted, and -errno when an internal error occurred.
1146 */
1147 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1148 BdrvCheckMode fix)
1149 {
1150 BDRVQcowState *s = bs->opaque;
1151 int64_t size, i, highest_cluster;
1152 int nb_clusters, refcount1, refcount2;
1153 QCowSnapshot *sn;
1154 uint16_t *refcount_table;
1155 int ret;
1156
1157 size = bdrv_getlength(bs->file);
1158 nb_clusters = size_to_clusters(s, size);
1159 res->bfi.total_clusters = nb_clusters;
1160 refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t));
1161
1162 /* header */
1163 inc_refcounts(bs, res, refcount_table, nb_clusters,
1164 0, s->cluster_size);
1165
1166 /* current L1 table */
1167 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1168 s->l1_table_offset, s->l1_size,
1169 CHECK_OFLAG_COPIED | CHECK_FRAG_INFO);
1170 if (ret < 0) {
1171 goto fail;
1172 }
1173
1174 /* snapshots */
1175 for(i = 0; i < s->nb_snapshots; i++) {
1176 sn = s->snapshots + i;
1177 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1178 sn->l1_table_offset, sn->l1_size, 0);
1179 if (ret < 0) {
1180 goto fail;
1181 }
1182 }
1183 inc_refcounts(bs, res, refcount_table, nb_clusters,
1184 s->snapshots_offset, s->snapshots_size);
1185
1186 /* refcount data */
1187 inc_refcounts(bs, res, refcount_table, nb_clusters,
1188 s->refcount_table_offset,
1189 s->refcount_table_size * sizeof(uint64_t));
1190
1191 for(i = 0; i < s->refcount_table_size; i++) {
1192 uint64_t offset, cluster;
1193 offset = s->refcount_table[i];
1194 cluster = offset >> s->cluster_bits;
1195
1196 /* Refcount blocks are cluster aligned */
1197 if (offset & (s->cluster_size - 1)) {
1198 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1199 "cluster aligned; refcount table entry corrupted\n", i);
1200 res->corruptions++;
1201 continue;
1202 }
1203
1204 if (cluster >= nb_clusters) {
1205 fprintf(stderr, "ERROR refcount block %" PRId64
1206 " is outside image\n", i);
1207 res->corruptions++;
1208 continue;
1209 }
1210
1211 if (offset != 0) {
1212 inc_refcounts(bs, res, refcount_table, nb_clusters,
1213 offset, s->cluster_size);
1214 if (refcount_table[cluster] != 1) {
1215 fprintf(stderr, "ERROR refcount block %" PRId64
1216 " refcount=%d\n",
1217 i, refcount_table[cluster]);
1218 res->corruptions++;
1219 }
1220 }
1221 }
1222
1223 /* compare ref counts */
1224 for (i = 0, highest_cluster = 0; i < nb_clusters; i++) {
1225 refcount1 = get_refcount(bs, i);
1226 if (refcount1 < 0) {
1227 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
1228 i, strerror(-refcount1));
1229 res->check_errors++;
1230 continue;
1231 }
1232
1233 refcount2 = refcount_table[i];
1234
1235 if (refcount1 > 0 || refcount2 > 0) {
1236 highest_cluster = i;
1237 }
1238
1239 if (refcount1 != refcount2) {
1240
1241 /* Check if we're allowed to fix the mismatch */
1242 int *num_fixed = NULL;
1243 if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
1244 num_fixed = &res->leaks_fixed;
1245 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
1246 num_fixed = &res->corruptions_fixed;
1247 }
1248
1249 fprintf(stderr, "%s cluster %" PRId64 " refcount=%d reference=%d\n",
1250 num_fixed != NULL ? "Repairing" :
1251 refcount1 < refcount2 ? "ERROR" :
1252 "Leaked",
1253 i, refcount1, refcount2);
1254
1255 if (num_fixed) {
1256 ret = update_refcount(bs, i << s->cluster_bits, 1,
1257 refcount2 - refcount1);
1258 if (ret >= 0) {
1259 (*num_fixed)++;
1260 continue;
1261 }
1262 }
1263
1264 /* And if we couldn't, print an error */
1265 if (refcount1 < refcount2) {
1266 res->corruptions++;
1267 } else {
1268 res->leaks++;
1269 }
1270 }
1271 }
1272
1273 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
1274 ret = 0;
1275
1276 fail:
1277 g_free(refcount_table);
1278
1279 return ret;
1280 }
1281