]> git.proxmox.com Git - qemu.git/blob - block/qcow2-refcount.c
qcow2: set L2 cache dependency in qcow2_alloc_bytes()
[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 bdrv_flush(bs->file);
536
537 return get_refcount(bs, cluster_index);
538 }
539
540
541
542 /*********************************************************/
543 /* cluster allocation functions */
544
545
546
547 /* return < 0 if error */
548 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
549 {
550 BDRVQcowState *s = bs->opaque;
551 int i, nb_clusters, refcount;
552
553 nb_clusters = size_to_clusters(s, size);
554 retry:
555 for(i = 0; i < nb_clusters; i++) {
556 int64_t next_cluster_index = s->free_cluster_index++;
557 refcount = get_refcount(bs, next_cluster_index);
558
559 if (refcount < 0) {
560 return refcount;
561 } else if (refcount != 0) {
562 goto retry;
563 }
564 }
565 #ifdef DEBUG_ALLOC2
566 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
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
573 int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
574 {
575 int64_t offset;
576 int ret;
577
578 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
579 offset = alloc_clusters_noref(bs, size);
580 if (offset < 0) {
581 return offset;
582 }
583
584 ret = update_refcount(bs, offset, size, 1);
585 if (ret < 0) {
586 return ret;
587 }
588
589 return offset;
590 }
591
592 int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
593 int nb_clusters)
594 {
595 BDRVQcowState *s = bs->opaque;
596 uint64_t cluster_index;
597 uint64_t old_free_cluster_index;
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 */
613 old_free_cluster_index = s->free_cluster_index;
614 s->free_cluster_index = cluster_index + i;
615
616 ret = update_refcount(bs, offset, i << s->cluster_bits, 1);
617 if (ret < 0) {
618 return ret;
619 }
620
621 s->free_cluster_index = old_free_cluster_index;
622
623 return i;
624 }
625
626 /* only used to allocate compressed sectors. We try to allocate
627 contiguous sectors. size must be <= cluster_size */
628 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
629 {
630 BDRVQcowState *s = bs->opaque;
631 int64_t offset, cluster_offset;
632 int free_in_cluster;
633
634 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
635 assert(size > 0 && size <= s->cluster_size);
636 if (s->free_byte_offset == 0) {
637 offset = qcow2_alloc_clusters(bs, s->cluster_size);
638 if (offset < 0) {
639 return offset;
640 }
641 s->free_byte_offset = offset;
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 {
656 offset = qcow2_alloc_clusters(bs, s->cluster_size);
657 if (offset < 0) {
658 return offset;
659 }
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 }
671
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);
677 return offset;
678 }
679
680 void qcow2_free_clusters(BlockDriverState *bs,
681 int64_t offset, int64_t size)
682 {
683 int ret;
684
685 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
686 ret = update_refcount(bs, offset, size, -1);
687 if (ret < 0) {
688 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
689 /* TODO Remember the clusters to free them later and avoid leaking */
690 }
691 }
692
693 /*
694 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
695 * normal cluster, compressed cluster, etc.)
696 */
697 void qcow2_free_any_clusters(BlockDriverState *bs,
698 uint64_t l2_entry, int nb_clusters)
699 {
700 BDRVQcowState *s = bs->opaque;
701
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:
718 case QCOW2_CLUSTER_ZERO:
719 break;
720 default:
721 abort();
722 }
723 }
724
725
726
727 /*********************************************************/
728 /* snapshots and image creation */
729
730
731
732 /* update the refcounts of snapshots and the copied flag */
733 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
734 int64_t l1_table_offset, int l1_size, int addend)
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;
739 int i, j, l1_modified = 0, nb_csectors, refcount;
740 int ret;
741
742 l2_table = NULL;
743 l1_table = NULL;
744 l1_size2 = l1_size * sizeof(uint64_t);
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! */
749 if (l1_table_offset != s->l1_table_offset) {
750 l1_table = g_malloc0(align_offset(l1_size2, 512));
751 l1_allocated = 1;
752 if (bdrv_pread(bs->file, l1_table_offset,
753 l1_table, l1_size2) != l1_size2)
754 {
755 ret = -EIO;
756 goto fail;
757 }
758
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
767 for(i = 0; i < l1_size; i++) {
768 l2_offset = l1_table[i];
769 if (l2_offset) {
770 old_l2_offset = l2_offset;
771 l2_offset &= L1E_OFFSET_MASK;
772
773 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
774 (void**) &l2_table);
775 if (ret < 0) {
776 goto fail;
777 }
778
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;
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 }
795
796 /* TODO Flushing once for the whole function should
797 * be enough */
798 bdrv_flush(bs->file);
799 }
800 /* compressed clusters are never modified */
801 refcount = 2;
802 } else {
803 uint64_t cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
804 if (addend != 0) {
805 refcount = update_cluster_refcount(bs, cluster_index, addend);
806 } else {
807 refcount = get_refcount(bs, cluster_index);
808 }
809
810 if (refcount < 0) {
811 ret = -EIO;
812 goto fail;
813 }
814 }
815
816 if (refcount == 1) {
817 offset |= QCOW_OFLAG_COPIED;
818 }
819 if (offset != old_offset) {
820 if (addend > 0) {
821 qcow2_cache_set_dependency(bs, s->l2_table_cache,
822 s->refcount_block_cache);
823 }
824 l2_table[j] = cpu_to_be64(offset);
825 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
826 }
827 }
828 }
829
830 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
831 if (ret < 0) {
832 goto fail;
833 }
834
835
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 }
841 if (refcount < 0) {
842 ret = -EIO;
843 goto fail;
844 } else if (refcount == 1) {
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 }
853
854 ret = 0;
855 fail:
856 if (l2_table) {
857 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
858 }
859
860 /* Update L1 only if it isn't deleted anyway (addend = -1) */
861 if (addend >= 0 && l1_modified) {
862 for(i = 0; i < l1_size; i++)
863 cpu_to_be64s(&l1_table[i]);
864 if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table,
865 l1_size2) < 0)
866 goto fail;
867 for(i = 0; i < l1_size; i++)
868 be64_to_cpus(&l1_table[i]);
869 }
870 if (l1_allocated)
871 g_free(l1_table);
872 return ret;
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 *
888 * Modifies the number of errors in res.
889 */
890 static void inc_refcounts(BlockDriverState *bs,
891 BdrvCheckResult *res,
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;
899
900 if (size <= 0)
901 return;
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;
908 if (k < 0) {
909 fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
910 cluster_offset);
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++;
917 } else {
918 if (++refcount_table[k] == 0) {
919 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
920 "\n", cluster_offset);
921 res->corruptions++;
922 }
923 }
924 }
925 }
926
927 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
928 enum {
929 CHECK_OFLAG_COPIED = 0x1, /* check QCOW_OFLAG_COPIED matches refcount */
930 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
931 };
932
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 */
941 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
942 uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
943 int flags)
944 {
945 BDRVQcowState *s = bs->opaque;
946 uint64_t *l2_table, l2_entry;
947 uint64_t next_contiguous_offset = 0;
948 int i, l2_size, nb_csectors, refcount;
949
950 /* Read L2 table from disk */
951 l2_size = s->l2_size * sizeof(uint64_t);
952 l2_table = g_malloc(l2_size);
953
954 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
955 goto fail;
956
957 /* Do the actual checks */
958 for(i = 0; i < s->l2_size; i++) {
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 }
971
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);
978
979 if (flags & CHECK_FRAG_INFO) {
980 res->bfi.allocated_clusters++;
981 res->bfi.compressed_clusters++;
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 }
990 break;
991
992 case QCOW2_CLUSTER_ZERO:
993 if ((l2_entry & L2E_OFFSET_MASK) == 0) {
994 break;
995 }
996 /* fall through */
997
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;
1002
1003 if (flags & CHECK_OFLAG_COPIED) {
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);
1013 res->corruptions++;
1014 }
1015 }
1016
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
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();
1044 }
1045 }
1046
1047 g_free(l2_table);
1048 return 0;
1049
1050 fail:
1051 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1052 g_free(l2_table);
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 */
1064 static int check_refcounts_l1(BlockDriverState *bs,
1065 BdrvCheckResult *res,
1066 uint16_t *refcount_table,
1067 int refcount_table_size,
1068 int64_t l1_table_offset, int l1_size,
1069 int flags)
1070 {
1071 BDRVQcowState *s = bs->opaque;
1072 uint64_t *l1_table, l2_offset, l1_size2;
1073 int i, refcount, ret;
1074
1075 l1_size2 = l1_size * sizeof(uint64_t);
1076
1077 /* Mark L1 table as used */
1078 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1079 l1_table_offset, l1_size2);
1080
1081 /* Read L1 table entries from disk */
1082 if (l1_size2 == 0) {
1083 l1_table = NULL;
1084 } else {
1085 l1_table = g_malloc(l1_size2);
1086 if (bdrv_pread(bs->file, l1_table_offset,
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 }
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 */
1098 if (flags & CHECK_OFLAG_COPIED) {
1099 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
1100 >> s->cluster_bits);
1101 if (refcount < 0) {
1102 fprintf(stderr, "Can't get refcount for l2_offset %"
1103 PRIx64 ": %s\n", l2_offset, strerror(-refcount));
1104 goto fail;
1105 }
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);
1109 res->corruptions++;
1110 }
1111 }
1112
1113 /* Mark L2 table as used */
1114 l2_offset &= L1E_OFFSET_MASK;
1115 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1116 l2_offset, s->cluster_size);
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);
1122 res->corruptions++;
1123 }
1124
1125 /* Process and check L2 entries */
1126 ret = check_refcounts_l2(bs, res, refcount_table,
1127 refcount_table_size, l2_offset, flags);
1128 if (ret < 0) {
1129 goto fail;
1130 }
1131 }
1132 }
1133 g_free(l1_table);
1134 return 0;
1135
1136 fail:
1137 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1138 res->check_errors++;
1139 g_free(l1_table);
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
1147 * detected as corrupted, and -errno when an internal error occurred.
1148 */
1149 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1150 BdrvCheckMode fix)
1151 {
1152 BDRVQcowState *s = bs->opaque;
1153 int64_t size, i, highest_cluster;
1154 int nb_clusters, refcount1, refcount2;
1155 QCowSnapshot *sn;
1156 uint16_t *refcount_table;
1157 int ret;
1158
1159 size = bdrv_getlength(bs->file);
1160 nb_clusters = size_to_clusters(s, size);
1161 res->bfi.total_clusters = nb_clusters;
1162 refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t));
1163
1164 /* header */
1165 inc_refcounts(bs, res, refcount_table, nb_clusters,
1166 0, s->cluster_size);
1167
1168 /* current L1 table */
1169 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1170 s->l1_table_offset, s->l1_size,
1171 CHECK_OFLAG_COPIED | CHECK_FRAG_INFO);
1172 if (ret < 0) {
1173 goto fail;
1174 }
1175
1176 /* snapshots */
1177 for(i = 0; i < s->nb_snapshots; i++) {
1178 sn = s->snapshots + i;
1179 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1180 sn->l1_table_offset, sn->l1_size, 0);
1181 if (ret < 0) {
1182 goto fail;
1183 }
1184 }
1185 inc_refcounts(bs, res, refcount_table, nb_clusters,
1186 s->snapshots_offset, s->snapshots_size);
1187
1188 /* refcount data */
1189 inc_refcounts(bs, res, refcount_table, nb_clusters,
1190 s->refcount_table_offset,
1191 s->refcount_table_size * sizeof(uint64_t));
1192
1193 for(i = 0; i < s->refcount_table_size; i++) {
1194 uint64_t offset, cluster;
1195 offset = s->refcount_table[i];
1196 cluster = offset >> s->cluster_bits;
1197
1198 /* Refcount blocks are cluster aligned */
1199 if (offset & (s->cluster_size - 1)) {
1200 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1201 "cluster aligned; refcount table entry corrupted\n", i);
1202 res->corruptions++;
1203 continue;
1204 }
1205
1206 if (cluster >= nb_clusters) {
1207 fprintf(stderr, "ERROR refcount block %" PRId64
1208 " is outside image\n", i);
1209 res->corruptions++;
1210 continue;
1211 }
1212
1213 if (offset != 0) {
1214 inc_refcounts(bs, res, refcount_table, nb_clusters,
1215 offset, s->cluster_size);
1216 if (refcount_table[cluster] != 1) {
1217 fprintf(stderr, "ERROR refcount block %" PRId64
1218 " refcount=%d\n",
1219 i, refcount_table[cluster]);
1220 res->corruptions++;
1221 }
1222 }
1223 }
1224
1225 /* compare ref counts */
1226 for (i = 0, highest_cluster = 0; i < nb_clusters; i++) {
1227 refcount1 = get_refcount(bs, i);
1228 if (refcount1 < 0) {
1229 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
1230 i, strerror(-refcount1));
1231 res->check_errors++;
1232 continue;
1233 }
1234
1235 refcount2 = refcount_table[i];
1236
1237 if (refcount1 > 0 || refcount2 > 0) {
1238 highest_cluster = i;
1239 }
1240
1241 if (refcount1 != refcount2) {
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",
1255 i, refcount1, refcount2);
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 */
1267 if (refcount1 < refcount2) {
1268 res->corruptions++;
1269 } else {
1270 res->leaks++;
1271 }
1272 }
1273 }
1274
1275 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
1276 ret = 0;
1277
1278 fail:
1279 g_free(refcount_table);
1280
1281 return ret;
1282 }
1283