]> git.proxmox.com Git - qemu.git/blob - block/qcow2-refcount.c
qcow2: support compressed clusters in BlockFragInfo
[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 qcow2_cache_flush(bs, s->l2_table_cache);
205
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;
210 }
211
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
217
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;
224 }
225
226 memset(*refcount_block, 0, s->cluster_size);
227
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;
238 }
239
240 bdrv_flush(bs->file);
241
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;
248 }
249
250 memset(*refcount_block, 0, s->cluster_size);
251 }
252
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;
259 }
260
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;
270 }
271
272 s->refcount_table[refcount_table_index] = new_block;
273 return 0;
274 }
275
276 ret = qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
277 if (ret < 0) {
278 goto fail_block;
279 }
280
281 /*
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.
285 *
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.
291 */
292 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
293
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;
298
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 =
305 size_to_clusters(s, table_size * sizeof(uint64_t));
306 blocks_clusters = 1 +
307 ((table_clusters + refcount_block_clusters - 1)
308 / refcount_block_clusters);
309 uint64_t meta_clusters = table_clusters + blocks_clusters;
310
311 last_table_size = table_size;
312 table_size = next_refcount_table_size(s, blocks_used +
313 ((meta_clusters + refcount_block_clusters - 1)
314 / refcount_block_clusters));
315
316 } while (last_table_size != table_size);
317
318 #ifdef DEBUG_ALLOC2
319 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
320 s->refcount_table_size, table_size);
321 #endif
322
323 /* Create the new refcount table and blocks */
324 uint64_t meta_offset = (blocks_used * refcount_block_clusters) *
325 s->cluster_size;
326 uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
327 uint16_t *new_blocks = g_malloc0(blocks_clusters * s->cluster_size);
328 uint64_t *new_table = g_malloc0(table_size * sizeof(uint64_t));
329
330 assert(meta_offset >= (s->free_cluster_index * s->cluster_size));
331
332 /* Fill the new refcount table */
333 memcpy(new_table, s->refcount_table,
334 s->refcount_table_size * sizeof(uint64_t));
335 new_table[refcount_table_index] = new_block;
336
337 int i;
338 for (i = 0; i < blocks_clusters; i++) {
339 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
340 }
341
342 /* Fill the refcount blocks */
343 uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
344 int block = 0;
345 for (i = 0; i < table_clusters + blocks_clusters; i++) {
346 new_blocks[block++] = cpu_to_be16(1);
347 }
348
349 /* Write refcount blocks to disk */
350 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
351 ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
352 blocks_clusters * s->cluster_size);
353 g_free(new_blocks);
354 if (ret < 0) {
355 goto fail_table;
356 }
357
358 /* Write refcount table to disk */
359 for(i = 0; i < table_size; i++) {
360 cpu_to_be64s(&new_table[i]);
361 }
362
363 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
364 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
365 table_size * sizeof(uint64_t));
366 if (ret < 0) {
367 goto fail_table;
368 }
369
370 for(i = 0; i < table_size; i++) {
371 be64_to_cpus(&new_table[i]);
372 }
373
374 /* Hook up the new refcount table in the qcow2 header */
375 uint8_t data[12];
376 cpu_to_be64w((uint64_t*)data, table_offset);
377 cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
378 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
379 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
380 data, sizeof(data));
381 if (ret < 0) {
382 goto fail_table;
383 }
384
385 /* And switch it in memory */
386 uint64_t old_table_offset = s->refcount_table_offset;
387 uint64_t old_table_size = s->refcount_table_size;
388
389 g_free(s->refcount_table);
390 s->refcount_table = new_table;
391 s->refcount_table_size = table_size;
392 s->refcount_table_offset = table_offset;
393
394 /* Free old table. Remember, we must not change free_cluster_index */
395 uint64_t old_free_cluster_index = s->free_cluster_index;
396 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
397 s->free_cluster_index = old_free_cluster_index;
398
399 ret = load_refcount_block(bs, new_block, (void**) refcount_block);
400 if (ret < 0) {
401 return ret;
402 }
403
404 return 0;
405
406 fail_table:
407 g_free(new_table);
408 fail_block:
409 if (*refcount_block != NULL) {
410 qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
411 }
412 return ret;
413 }
414
415 /* XXX: cache several refcount block clusters ? */
416 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
417 int64_t offset, int64_t length, int addend)
418 {
419 BDRVQcowState *s = bs->opaque;
420 int64_t start, last, cluster_offset;
421 uint16_t *refcount_block = NULL;
422 int64_t old_table_index = -1;
423 int ret;
424
425 #ifdef DEBUG_ALLOC2
426 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
427 offset, length, addend);
428 #endif
429 if (length < 0) {
430 return -EINVAL;
431 } else if (length == 0) {
432 return 0;
433 }
434
435 if (addend < 0) {
436 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
437 s->l2_table_cache);
438 }
439
440 start = offset & ~(s->cluster_size - 1);
441 last = (offset + length - 1) & ~(s->cluster_size - 1);
442 for(cluster_offset = start; cluster_offset <= last;
443 cluster_offset += s->cluster_size)
444 {
445 int block_index, refcount;
446 int64_t cluster_index = cluster_offset >> s->cluster_bits;
447 int64_t table_index =
448 cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
449
450 /* Load the refcount block and allocate it if needed */
451 if (table_index != old_table_index) {
452 if (refcount_block) {
453 ret = qcow2_cache_put(bs, s->refcount_block_cache,
454 (void**) &refcount_block);
455 if (ret < 0) {
456 goto fail;
457 }
458 }
459
460 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
461 if (ret < 0) {
462 goto fail;
463 }
464 }
465 old_table_index = table_index;
466
467 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
468
469 /* we can update the count and save it */
470 block_index = cluster_index &
471 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
472
473 refcount = be16_to_cpu(refcount_block[block_index]);
474 refcount += addend;
475 if (refcount < 0 || refcount > 0xffff) {
476 ret = -EINVAL;
477 goto fail;
478 }
479 if (refcount == 0 && cluster_index < s->free_cluster_index) {
480 s->free_cluster_index = cluster_index;
481 }
482 refcount_block[block_index] = cpu_to_be16(refcount);
483 }
484
485 ret = 0;
486 fail:
487 /* Write last changed block to disk */
488 if (refcount_block) {
489 int wret;
490 wret = qcow2_cache_put(bs, s->refcount_block_cache,
491 (void**) &refcount_block);
492 if (wret < 0) {
493 return ret < 0 ? ret : wret;
494 }
495 }
496
497 /*
498 * Try do undo any updates if an error is returned (This may succeed in
499 * some cases like ENOSPC for allocating a new refcount block)
500 */
501 if (ret < 0) {
502 int dummy;
503 dummy = update_refcount(bs, offset, cluster_offset - offset, -addend);
504 (void)dummy;
505 }
506
507 return ret;
508 }
509
510 /*
511 * Increases or decreases the refcount of a given cluster by one.
512 * addend must be 1 or -1.
513 *
514 * If the return value is non-negative, it is the new refcount of the cluster.
515 * If it is negative, it is -errno and indicates an error.
516 */
517 static int update_cluster_refcount(BlockDriverState *bs,
518 int64_t cluster_index,
519 int addend)
520 {
521 BDRVQcowState *s = bs->opaque;
522 int ret;
523
524 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
525 if (ret < 0) {
526 return ret;
527 }
528
529 bdrv_flush(bs->file);
530
531 return get_refcount(bs, cluster_index);
532 }
533
534
535
536 /*********************************************************/
537 /* cluster allocation functions */
538
539
540
541 /* return < 0 if error */
542 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
543 {
544 BDRVQcowState *s = bs->opaque;
545 int i, nb_clusters, refcount;
546
547 nb_clusters = size_to_clusters(s, size);
548 retry:
549 for(i = 0; i < nb_clusters; i++) {
550 int64_t next_cluster_index = s->free_cluster_index++;
551 refcount = get_refcount(bs, next_cluster_index);
552
553 if (refcount < 0) {
554 return refcount;
555 } else if (refcount != 0) {
556 goto retry;
557 }
558 }
559 #ifdef DEBUG_ALLOC2
560 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
561 size,
562 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
563 #endif
564 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
565 }
566
567 int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
568 {
569 int64_t offset;
570 int ret;
571
572 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
573 offset = alloc_clusters_noref(bs, size);
574 if (offset < 0) {
575 return offset;
576 }
577
578 ret = update_refcount(bs, offset, size, 1);
579 if (ret < 0) {
580 return ret;
581 }
582
583 return offset;
584 }
585
586 int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
587 int nb_clusters)
588 {
589 BDRVQcowState *s = bs->opaque;
590 uint64_t cluster_index;
591 uint64_t old_free_cluster_index;
592 int i, refcount, ret;
593
594 /* Check how many clusters there are free */
595 cluster_index = offset >> s->cluster_bits;
596 for(i = 0; i < nb_clusters; i++) {
597 refcount = get_refcount(bs, cluster_index++);
598
599 if (refcount < 0) {
600 return refcount;
601 } else if (refcount != 0) {
602 break;
603 }
604 }
605
606 /* And then allocate them */
607 old_free_cluster_index = s->free_cluster_index;
608 s->free_cluster_index = cluster_index + i;
609
610 ret = update_refcount(bs, offset, i << s->cluster_bits, 1);
611 if (ret < 0) {
612 return ret;
613 }
614
615 s->free_cluster_index = old_free_cluster_index;
616
617 return i;
618 }
619
620 /* only used to allocate compressed sectors. We try to allocate
621 contiguous sectors. size must be <= cluster_size */
622 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
623 {
624 BDRVQcowState *s = bs->opaque;
625 int64_t offset, cluster_offset;
626 int free_in_cluster;
627
628 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
629 assert(size > 0 && size <= s->cluster_size);
630 if (s->free_byte_offset == 0) {
631 offset = qcow2_alloc_clusters(bs, s->cluster_size);
632 if (offset < 0) {
633 return offset;
634 }
635 s->free_byte_offset = offset;
636 }
637 redo:
638 free_in_cluster = s->cluster_size -
639 (s->free_byte_offset & (s->cluster_size - 1));
640 if (size <= free_in_cluster) {
641 /* enough space in current cluster */
642 offset = s->free_byte_offset;
643 s->free_byte_offset += size;
644 free_in_cluster -= size;
645 if (free_in_cluster == 0)
646 s->free_byte_offset = 0;
647 if ((offset & (s->cluster_size - 1)) != 0)
648 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
649 } else {
650 offset = qcow2_alloc_clusters(bs, s->cluster_size);
651 if (offset < 0) {
652 return offset;
653 }
654 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
655 if ((cluster_offset + s->cluster_size) == offset) {
656 /* we are lucky: contiguous data */
657 offset = s->free_byte_offset;
658 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
659 s->free_byte_offset += size;
660 } else {
661 s->free_byte_offset = offset;
662 goto redo;
663 }
664 }
665
666 bdrv_flush(bs->file);
667 return offset;
668 }
669
670 void qcow2_free_clusters(BlockDriverState *bs,
671 int64_t offset, int64_t size)
672 {
673 int ret;
674
675 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
676 ret = update_refcount(bs, offset, size, -1);
677 if (ret < 0) {
678 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
679 /* TODO Remember the clusters to free them later and avoid leaking */
680 }
681 }
682
683 /*
684 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
685 * normal cluster, compressed cluster, etc.)
686 */
687 void qcow2_free_any_clusters(BlockDriverState *bs,
688 uint64_t l2_entry, int nb_clusters)
689 {
690 BDRVQcowState *s = bs->opaque;
691
692 switch (qcow2_get_cluster_type(l2_entry)) {
693 case QCOW2_CLUSTER_COMPRESSED:
694 {
695 int nb_csectors;
696 nb_csectors = ((l2_entry >> s->csize_shift) &
697 s->csize_mask) + 1;
698 qcow2_free_clusters(bs,
699 (l2_entry & s->cluster_offset_mask) & ~511,
700 nb_csectors * 512);
701 }
702 break;
703 case QCOW2_CLUSTER_NORMAL:
704 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
705 nb_clusters << s->cluster_bits);
706 break;
707 case QCOW2_CLUSTER_UNALLOCATED:
708 case QCOW2_CLUSTER_ZERO:
709 break;
710 default:
711 abort();
712 }
713 }
714
715
716
717 /*********************************************************/
718 /* snapshots and image creation */
719
720
721
722 /* update the refcounts of snapshots and the copied flag */
723 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
724 int64_t l1_table_offset, int l1_size, int addend)
725 {
726 BDRVQcowState *s = bs->opaque;
727 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
728 int64_t old_offset, old_l2_offset;
729 int i, j, l1_modified = 0, nb_csectors, refcount;
730 int ret;
731
732 l2_table = NULL;
733 l1_table = NULL;
734 l1_size2 = l1_size * sizeof(uint64_t);
735
736 /* WARNING: qcow2_snapshot_goto relies on this function not using the
737 * l1_table_offset when it is the current s->l1_table_offset! Be careful
738 * when changing this! */
739 if (l1_table_offset != s->l1_table_offset) {
740 l1_table = g_malloc0(align_offset(l1_size2, 512));
741 l1_allocated = 1;
742 if (bdrv_pread(bs->file, l1_table_offset,
743 l1_table, l1_size2) != l1_size2)
744 {
745 ret = -EIO;
746 goto fail;
747 }
748
749 for(i = 0;i < l1_size; i++)
750 be64_to_cpus(&l1_table[i]);
751 } else {
752 assert(l1_size == s->l1_size);
753 l1_table = s->l1_table;
754 l1_allocated = 0;
755 }
756
757 for(i = 0; i < l1_size; i++) {
758 l2_offset = l1_table[i];
759 if (l2_offset) {
760 old_l2_offset = l2_offset;
761 l2_offset &= L1E_OFFSET_MASK;
762
763 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
764 (void**) &l2_table);
765 if (ret < 0) {
766 goto fail;
767 }
768
769 for(j = 0; j < s->l2_size; j++) {
770 offset = be64_to_cpu(l2_table[j]);
771 if (offset != 0) {
772 old_offset = offset;
773 offset &= ~QCOW_OFLAG_COPIED;
774 if (offset & QCOW_OFLAG_COMPRESSED) {
775 nb_csectors = ((offset >> s->csize_shift) &
776 s->csize_mask) + 1;
777 if (addend != 0) {
778 int ret;
779 ret = update_refcount(bs,
780 (offset & s->cluster_offset_mask) & ~511,
781 nb_csectors * 512, addend);
782 if (ret < 0) {
783 goto fail;
784 }
785
786 /* TODO Flushing once for the whole function should
787 * be enough */
788 bdrv_flush(bs->file);
789 }
790 /* compressed clusters are never modified */
791 refcount = 2;
792 } else {
793 uint64_t cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
794 if (addend != 0) {
795 refcount = update_cluster_refcount(bs, cluster_index, addend);
796 } else {
797 refcount = get_refcount(bs, cluster_index);
798 }
799
800 if (refcount < 0) {
801 ret = -EIO;
802 goto fail;
803 }
804 }
805
806 if (refcount == 1) {
807 offset |= QCOW_OFLAG_COPIED;
808 }
809 if (offset != old_offset) {
810 if (addend > 0) {
811 qcow2_cache_set_dependency(bs, s->l2_table_cache,
812 s->refcount_block_cache);
813 }
814 l2_table[j] = cpu_to_be64(offset);
815 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
816 }
817 }
818 }
819
820 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
821 if (ret < 0) {
822 goto fail;
823 }
824
825
826 if (addend != 0) {
827 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
828 } else {
829 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
830 }
831 if (refcount < 0) {
832 ret = -EIO;
833 goto fail;
834 } else if (refcount == 1) {
835 l2_offset |= QCOW_OFLAG_COPIED;
836 }
837 if (l2_offset != old_l2_offset) {
838 l1_table[i] = l2_offset;
839 l1_modified = 1;
840 }
841 }
842 }
843
844 ret = 0;
845 fail:
846 if (l2_table) {
847 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
848 }
849
850 /* Update L1 only if it isn't deleted anyway (addend = -1) */
851 if (addend >= 0 && l1_modified) {
852 for(i = 0; i < l1_size; i++)
853 cpu_to_be64s(&l1_table[i]);
854 if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table,
855 l1_size2) < 0)
856 goto fail;
857 for(i = 0; i < l1_size; i++)
858 be64_to_cpus(&l1_table[i]);
859 }
860 if (l1_allocated)
861 g_free(l1_table);
862 return ret;
863 }
864
865
866
867
868 /*********************************************************/
869 /* refcount checking functions */
870
871
872
873 /*
874 * Increases the refcount for a range of clusters in a given refcount table.
875 * This is used to construct a temporary refcount table out of L1 and L2 tables
876 * which can be compared the the refcount table saved in the image.
877 *
878 * Modifies the number of errors in res.
879 */
880 static void inc_refcounts(BlockDriverState *bs,
881 BdrvCheckResult *res,
882 uint16_t *refcount_table,
883 int refcount_table_size,
884 int64_t offset, int64_t size)
885 {
886 BDRVQcowState *s = bs->opaque;
887 int64_t start, last, cluster_offset;
888 int k;
889
890 if (size <= 0)
891 return;
892
893 start = offset & ~(s->cluster_size - 1);
894 last = (offset + size - 1) & ~(s->cluster_size - 1);
895 for(cluster_offset = start; cluster_offset <= last;
896 cluster_offset += s->cluster_size) {
897 k = cluster_offset >> s->cluster_bits;
898 if (k < 0) {
899 fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
900 cluster_offset);
901 res->corruptions++;
902 } else if (k >= refcount_table_size) {
903 fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after "
904 "the end of the image file, can't properly check refcounts.\n",
905 cluster_offset);
906 res->check_errors++;
907 } else {
908 if (++refcount_table[k] == 0) {
909 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
910 "\n", cluster_offset);
911 res->corruptions++;
912 }
913 }
914 }
915 }
916
917 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
918 enum {
919 CHECK_OFLAG_COPIED = 0x1, /* check QCOW_OFLAG_COPIED matches refcount */
920 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
921 };
922
923 /*
924 * Increases the refcount in the given refcount table for the all clusters
925 * referenced in the L2 table. While doing so, performs some checks on L2
926 * entries.
927 *
928 * Returns the number of errors found by the checks or -errno if an internal
929 * error occurred.
930 */
931 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
932 uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
933 int flags)
934 {
935 BDRVQcowState *s = bs->opaque;
936 uint64_t *l2_table, l2_entry;
937 uint64_t next_contiguous_offset = 0;
938 int i, l2_size, nb_csectors, refcount;
939
940 /* Read L2 table from disk */
941 l2_size = s->l2_size * sizeof(uint64_t);
942 l2_table = g_malloc(l2_size);
943
944 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
945 goto fail;
946
947 /* Do the actual checks */
948 for(i = 0; i < s->l2_size; i++) {
949 l2_entry = be64_to_cpu(l2_table[i]);
950
951 switch (qcow2_get_cluster_type(l2_entry)) {
952 case QCOW2_CLUSTER_COMPRESSED:
953 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
954 if (l2_entry & QCOW_OFLAG_COPIED) {
955 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
956 "copied flag must never be set for compressed "
957 "clusters\n", l2_entry >> s->cluster_bits);
958 l2_entry &= ~QCOW_OFLAG_COPIED;
959 res->corruptions++;
960 }
961
962 /* Mark cluster as used */
963 nb_csectors = ((l2_entry >> s->csize_shift) &
964 s->csize_mask) + 1;
965 l2_entry &= s->cluster_offset_mask;
966 inc_refcounts(bs, res, refcount_table, refcount_table_size,
967 l2_entry & ~511, nb_csectors * 512);
968
969 if (flags & CHECK_FRAG_INFO) {
970 res->bfi.allocated_clusters++;
971 res->bfi.compressed_clusters++;
972
973 /* Compressed clusters are fragmented by nature. Since they
974 * take up sub-sector space but we only have sector granularity
975 * I/O we need to re-read the same sectors even for adjacent
976 * compressed clusters.
977 */
978 res->bfi.fragmented_clusters++;
979 }
980 break;
981
982 case QCOW2_CLUSTER_ZERO:
983 if ((l2_entry & L2E_OFFSET_MASK) == 0) {
984 break;
985 }
986 /* fall through */
987
988 case QCOW2_CLUSTER_NORMAL:
989 {
990 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
991 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
992
993 if (flags & CHECK_OFLAG_COPIED) {
994 refcount = get_refcount(bs, offset >> s->cluster_bits);
995 if (refcount < 0) {
996 fprintf(stderr, "Can't get refcount for offset %"
997 PRIx64 ": %s\n", l2_entry, strerror(-refcount));
998 goto fail;
999 }
1000 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
1001 fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
1002 PRIx64 " refcount=%d\n", l2_entry, refcount);
1003 res->corruptions++;
1004 }
1005 }
1006
1007 if (flags & CHECK_FRAG_INFO) {
1008 res->bfi.allocated_clusters++;
1009 if (next_contiguous_offset &&
1010 offset != next_contiguous_offset) {
1011 res->bfi.fragmented_clusters++;
1012 }
1013 next_contiguous_offset = offset + s->cluster_size;
1014 }
1015
1016 /* Mark cluster as used */
1017 inc_refcounts(bs, res, refcount_table,refcount_table_size,
1018 offset, s->cluster_size);
1019
1020 /* Correct offsets are cluster aligned */
1021 if (offset & (s->cluster_size - 1)) {
1022 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
1023 "properly aligned; L2 entry corrupted.\n", offset);
1024 res->corruptions++;
1025 }
1026 break;
1027 }
1028
1029 case QCOW2_CLUSTER_UNALLOCATED:
1030 break;
1031
1032 default:
1033 abort();
1034 }
1035 }
1036
1037 g_free(l2_table);
1038 return 0;
1039
1040 fail:
1041 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1042 g_free(l2_table);
1043 return -EIO;
1044 }
1045
1046 /*
1047 * Increases the refcount for the L1 table, its L2 tables and all referenced
1048 * clusters in the given refcount table. While doing so, performs some checks
1049 * on L1 and L2 entries.
1050 *
1051 * Returns the number of errors found by the checks or -errno if an internal
1052 * error occurred.
1053 */
1054 static int check_refcounts_l1(BlockDriverState *bs,
1055 BdrvCheckResult *res,
1056 uint16_t *refcount_table,
1057 int refcount_table_size,
1058 int64_t l1_table_offset, int l1_size,
1059 int flags)
1060 {
1061 BDRVQcowState *s = bs->opaque;
1062 uint64_t *l1_table, l2_offset, l1_size2;
1063 int i, refcount, ret;
1064
1065 l1_size2 = l1_size * sizeof(uint64_t);
1066
1067 /* Mark L1 table as used */
1068 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1069 l1_table_offset, l1_size2);
1070
1071 /* Read L1 table entries from disk */
1072 if (l1_size2 == 0) {
1073 l1_table = NULL;
1074 } else {
1075 l1_table = g_malloc(l1_size2);
1076 if (bdrv_pread(bs->file, l1_table_offset,
1077 l1_table, l1_size2) != l1_size2)
1078 goto fail;
1079 for(i = 0;i < l1_size; i++)
1080 be64_to_cpus(&l1_table[i]);
1081 }
1082
1083 /* Do the actual checks */
1084 for(i = 0; i < l1_size; i++) {
1085 l2_offset = l1_table[i];
1086 if (l2_offset) {
1087 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
1088 if (flags & CHECK_OFLAG_COPIED) {
1089 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
1090 >> s->cluster_bits);
1091 if (refcount < 0) {
1092 fprintf(stderr, "Can't get refcount for l2_offset %"
1093 PRIx64 ": %s\n", l2_offset, strerror(-refcount));
1094 goto fail;
1095 }
1096 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
1097 fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
1098 " refcount=%d\n", l2_offset, refcount);
1099 res->corruptions++;
1100 }
1101 }
1102
1103 /* Mark L2 table as used */
1104 l2_offset &= L1E_OFFSET_MASK;
1105 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1106 l2_offset, s->cluster_size);
1107
1108 /* L2 tables are cluster aligned */
1109 if (l2_offset & (s->cluster_size - 1)) {
1110 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1111 "cluster aligned; L1 entry corrupted\n", l2_offset);
1112 res->corruptions++;
1113 }
1114
1115 /* Process and check L2 entries */
1116 ret = check_refcounts_l2(bs, res, refcount_table,
1117 refcount_table_size, l2_offset, flags);
1118 if (ret < 0) {
1119 goto fail;
1120 }
1121 }
1122 }
1123 g_free(l1_table);
1124 return 0;
1125
1126 fail:
1127 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1128 res->check_errors++;
1129 g_free(l1_table);
1130 return -EIO;
1131 }
1132
1133 /*
1134 * Checks an image for refcount consistency.
1135 *
1136 * Returns 0 if no errors are found, the number of errors in case the image is
1137 * detected as corrupted, and -errno when an internal error occurred.
1138 */
1139 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1140 BdrvCheckMode fix)
1141 {
1142 BDRVQcowState *s = bs->opaque;
1143 int64_t size, i, highest_cluster;
1144 int nb_clusters, refcount1, refcount2;
1145 QCowSnapshot *sn;
1146 uint16_t *refcount_table;
1147 int ret;
1148
1149 size = bdrv_getlength(bs->file);
1150 nb_clusters = size_to_clusters(s, size);
1151 res->bfi.total_clusters = nb_clusters;
1152 refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t));
1153
1154 /* header */
1155 inc_refcounts(bs, res, refcount_table, nb_clusters,
1156 0, s->cluster_size);
1157
1158 /* current L1 table */
1159 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1160 s->l1_table_offset, s->l1_size,
1161 CHECK_OFLAG_COPIED | CHECK_FRAG_INFO);
1162 if (ret < 0) {
1163 goto fail;
1164 }
1165
1166 /* snapshots */
1167 for(i = 0; i < s->nb_snapshots; i++) {
1168 sn = s->snapshots + i;
1169 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1170 sn->l1_table_offset, sn->l1_size, 0);
1171 if (ret < 0) {
1172 goto fail;
1173 }
1174 }
1175 inc_refcounts(bs, res, refcount_table, nb_clusters,
1176 s->snapshots_offset, s->snapshots_size);
1177
1178 /* refcount data */
1179 inc_refcounts(bs, res, refcount_table, nb_clusters,
1180 s->refcount_table_offset,
1181 s->refcount_table_size * sizeof(uint64_t));
1182
1183 for(i = 0; i < s->refcount_table_size; i++) {
1184 uint64_t offset, cluster;
1185 offset = s->refcount_table[i];
1186 cluster = offset >> s->cluster_bits;
1187
1188 /* Refcount blocks are cluster aligned */
1189 if (offset & (s->cluster_size - 1)) {
1190 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1191 "cluster aligned; refcount table entry corrupted\n", i);
1192 res->corruptions++;
1193 continue;
1194 }
1195
1196 if (cluster >= nb_clusters) {
1197 fprintf(stderr, "ERROR refcount block %" PRId64
1198 " is outside image\n", i);
1199 res->corruptions++;
1200 continue;
1201 }
1202
1203 if (offset != 0) {
1204 inc_refcounts(bs, res, refcount_table, nb_clusters,
1205 offset, s->cluster_size);
1206 if (refcount_table[cluster] != 1) {
1207 fprintf(stderr, "ERROR refcount block %" PRId64
1208 " refcount=%d\n",
1209 i, refcount_table[cluster]);
1210 res->corruptions++;
1211 }
1212 }
1213 }
1214
1215 /* compare ref counts */
1216 for (i = 0, highest_cluster = 0; i < nb_clusters; i++) {
1217 refcount1 = get_refcount(bs, i);
1218 if (refcount1 < 0) {
1219 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
1220 i, strerror(-refcount1));
1221 res->check_errors++;
1222 continue;
1223 }
1224
1225 refcount2 = refcount_table[i];
1226
1227 if (refcount1 > 0 || refcount2 > 0) {
1228 highest_cluster = i;
1229 }
1230
1231 if (refcount1 != refcount2) {
1232
1233 /* Check if we're allowed to fix the mismatch */
1234 int *num_fixed = NULL;
1235 if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
1236 num_fixed = &res->leaks_fixed;
1237 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
1238 num_fixed = &res->corruptions_fixed;
1239 }
1240
1241 fprintf(stderr, "%s cluster %" PRId64 " refcount=%d reference=%d\n",
1242 num_fixed != NULL ? "Repairing" :
1243 refcount1 < refcount2 ? "ERROR" :
1244 "Leaked",
1245 i, refcount1, refcount2);
1246
1247 if (num_fixed) {
1248 ret = update_refcount(bs, i << s->cluster_bits, 1,
1249 refcount2 - refcount1);
1250 if (ret >= 0) {
1251 (*num_fixed)++;
1252 continue;
1253 }
1254 }
1255
1256 /* And if we couldn't, print an error */
1257 if (refcount1 < refcount2) {
1258 res->corruptions++;
1259 } else {
1260 res->leaks++;
1261 }
1262 }
1263 }
1264
1265 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
1266 ret = 0;
1267
1268 fail:
1269 g_free(refcount_table);
1270
1271 return ret;
1272 }
1273