]> git.proxmox.com Git - qemu.git/blame - block/qcow2-refcount.c
qcow2: Use bdrv_(p)write_sync for metadata writes
[qemu.git] / block / qcow2-refcount.c
CommitLineData
f7d0fe02
KW
1/*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25#include "qemu-common.h"
26#include "block_int.h"
27#include "block/qcow2.h"
28
29static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
92dcb59f 30static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
f7d0fe02
KW
31 int64_t offset, int64_t length,
32 int addend);
33
3b88e52b
KW
34
35static int cache_refcount_updates = 0;
36
66f82cee 37static int write_refcount_block(BlockDriverState *bs)
3b88e52b 38{
66f82cee 39 BDRVQcowState *s = bs->opaque;
3b88e52b
KW
40 size_t size = s->cluster_size;
41
42 if (s->refcount_block_cache_offset == 0) {
43 return 0;
44 }
45
66f82cee 46 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE);
8b3b7206
KW
47 if (bdrv_pwrite_sync(bs->file, s->refcount_block_cache_offset,
48 s->refcount_block_cache, size) < 0)
3b88e52b
KW
49 {
50 return -EIO;
51 }
52
53 return 0;
54}
55
f7d0fe02
KW
56/*********************************************************/
57/* refcount handling */
58
ed6ccf0f 59int qcow2_refcount_init(BlockDriverState *bs)
f7d0fe02
KW
60{
61 BDRVQcowState *s = bs->opaque;
62 int ret, refcount_table_size2, i;
63
64 s->refcount_block_cache = qemu_malloc(s->cluster_size);
65 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
66 s->refcount_table = qemu_malloc(refcount_table_size2);
67 if (s->refcount_table_size > 0) {
66f82cee
KW
68 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
69 ret = bdrv_pread(bs->file, s->refcount_table_offset,
f7d0fe02
KW
70 s->refcount_table, refcount_table_size2);
71 if (ret != refcount_table_size2)
72 goto fail;
73 for(i = 0; i < s->refcount_table_size; i++)
74 be64_to_cpus(&s->refcount_table[i]);
75 }
76 return 0;
77 fail:
78 return -ENOMEM;
79}
80
ed6ccf0f 81void qcow2_refcount_close(BlockDriverState *bs)
f7d0fe02
KW
82{
83 BDRVQcowState *s = bs->opaque;
84 qemu_free(s->refcount_block_cache);
85 qemu_free(s->refcount_table);
86}
87
88
89static int load_refcount_block(BlockDriverState *bs,
90 int64_t refcount_block_offset)
91{
92 BDRVQcowState *s = bs->opaque;
93 int ret;
3b88e52b
KW
94
95 if (cache_refcount_updates) {
e14e8ba5
KW
96 ret = write_refcount_block(bs);
97 if (ret < 0) {
98 return ret;
99 }
3b88e52b
KW
100 }
101
66f82cee
KW
102 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
103 ret = bdrv_pread(bs->file, refcount_block_offset, s->refcount_block_cache,
f7d0fe02 104 s->cluster_size);
e14e8ba5
KW
105 if (ret < 0) {
106 return ret;
107 }
108
f7d0fe02
KW
109 s->refcount_block_cache_offset = refcount_block_offset;
110 return 0;
111}
112
018faafd
KW
113/*
114 * Returns the refcount of the cluster given by its index. Any non-negative
115 * return value is the refcount of the cluster, negative values are -errno
116 * and indicate an error.
117 */
f7d0fe02
KW
118static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
119{
120 BDRVQcowState *s = bs->opaque;
121 int refcount_table_index, block_index;
122 int64_t refcount_block_offset;
018faafd 123 int ret;
f7d0fe02
KW
124
125 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
126 if (refcount_table_index >= s->refcount_table_size)
127 return 0;
128 refcount_block_offset = s->refcount_table[refcount_table_index];
129 if (!refcount_block_offset)
130 return 0;
131 if (refcount_block_offset != s->refcount_block_cache_offset) {
132 /* better than nothing: return allocated if read error */
018faafd
KW
133 ret = load_refcount_block(bs, refcount_block_offset);
134 if (ret < 0) {
135 return ret;
136 }
f7d0fe02
KW
137 }
138 block_index = cluster_index &
139 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
140 return be16_to_cpu(s->refcount_block_cache[block_index]);
141}
142
05121aed
KW
143/*
144 * Rounds the refcount table size up to avoid growing the table for each single
145 * refcount block that is allocated.
146 */
147static unsigned int next_refcount_table_size(BDRVQcowState *s,
148 unsigned int min_size)
149{
150 unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1;
151 unsigned int refcount_table_clusters =
152 MAX(1, s->refcount_table_size >> (s->cluster_bits - 3));
153
154 while (min_clusters > refcount_table_clusters) {
155 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
156 }
157
158 return refcount_table_clusters << (s->cluster_bits - 3);
159}
160
92dcb59f
KW
161
162/* Checks if two offsets are described by the same refcount block */
163static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a,
164 uint64_t offset_b)
165{
166 uint64_t block_a = offset_a >> (2 * s->cluster_bits - REFCOUNT_SHIFT);
167 uint64_t block_b = offset_b >> (2 * s->cluster_bits - REFCOUNT_SHIFT);
168
169 return (block_a == block_b);
170}
171
172/*
173 * Loads a refcount block. If it doesn't exist yet, it is allocated first
174 * (including growing the refcount table if needed).
175 *
176 * Returns the offset of the refcount block on success or -errno in error case
177 */
178static int64_t alloc_refcount_block(BlockDriverState *bs, int64_t cluster_index)
f7d0fe02
KW
179{
180 BDRVQcowState *s = bs->opaque;
92dcb59f
KW
181 unsigned int refcount_table_index;
182 int ret;
183
66f82cee 184 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
8252278a 185
92dcb59f
KW
186 /* Find the refcount block for the given cluster */
187 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
188
189 if (refcount_table_index < s->refcount_table_size) {
190
191 uint64_t refcount_block_offset =
192 s->refcount_table[refcount_table_index];
193
194 /* If it's already there, we're done */
195 if (refcount_block_offset) {
196 if (refcount_block_offset != s->refcount_block_cache_offset) {
197 ret = load_refcount_block(bs, refcount_block_offset);
198 if (ret < 0) {
199 return ret;
200 }
201 }
202 return refcount_block_offset;
203 }
204 }
205
206 /*
207 * If we came here, we need to allocate something. Something is at least
208 * a cluster for the new refcount block. It may also include a new refcount
209 * table if the old refcount table is too small.
210 *
211 * Note that allocating clusters here needs some special care:
212 *
213 * - We can't use the normal qcow2_alloc_clusters(), it would try to
214 * increase the refcount and very likely we would end up with an endless
215 * recursion. Instead we must place the refcount blocks in a way that
216 * they can describe them themselves.
217 *
218 * - We need to consider that at this point we are inside update_refcounts
219 * and doing the initial refcount increase. This means that some clusters
220 * have already been allocated by the caller, but their refcount isn't
221 * accurate yet. free_cluster_index tells us where this allocation ends
222 * as long as we don't overwrite it by freeing clusters.
223 *
224 * - alloc_clusters_noref and qcow2_free_clusters may load a different
225 * refcount block into the cache
226 */
227
228 if (cache_refcount_updates) {
66f82cee 229 ret = write_refcount_block(bs);
92dcb59f
KW
230 if (ret < 0) {
231 return ret;
232 }
233 }
234
235 /* Allocate the refcount block itself and mark it as used */
2eaa8f63
KW
236 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
237 if (new_block < 0) {
238 return new_block;
239 }
f7d0fe02 240
f7d0fe02 241#ifdef DEBUG_ALLOC2
92dcb59f
KW
242 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
243 " at %" PRIx64 "\n",
244 refcount_table_index, cluster_index << s->cluster_bits, new_block);
f7d0fe02 245#endif
92dcb59f
KW
246
247 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
25408c09
KW
248 /* Zero the new refcount block before updating it */
249 memset(s->refcount_block_cache, 0, s->cluster_size);
250 s->refcount_block_cache_offset = new_block;
251
92dcb59f
KW
252 /* The block describes itself, need to update the cache */
253 int block_index = (new_block >> s->cluster_bits) &
254 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
255 s->refcount_block_cache[block_index] = cpu_to_be16(1);
256 } else {
257 /* Described somewhere else. This can recurse at most twice before we
258 * arrive at a block that describes itself. */
259 ret = update_refcount(bs, new_block, s->cluster_size, 1);
260 if (ret < 0) {
261 goto fail_block;
262 }
25408c09
KW
263
264 /* Initialize the new refcount block only after updating its refcount,
265 * update_refcount uses the refcount cache itself */
266 memset(s->refcount_block_cache, 0, s->cluster_size);
267 s->refcount_block_cache_offset = new_block;
92dcb59f
KW
268 }
269
270 /* Now the new refcount block needs to be written to disk */
66f82cee 271 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
8b3b7206 272 ret = bdrv_pwrite_sync(bs->file, new_block, s->refcount_block_cache,
92dcb59f
KW
273 s->cluster_size);
274 if (ret < 0) {
275 goto fail_block;
276 }
277
278 /* If the refcount table is big enough, just hook the block up there */
279 if (refcount_table_index < s->refcount_table_size) {
280 uint64_t data64 = cpu_to_be64(new_block);
66f82cee 281 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
8b3b7206 282 ret = bdrv_pwrite_sync(bs->file,
92dcb59f
KW
283 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
284 &data64, sizeof(data64));
285 if (ret < 0) {
286 goto fail_block;
287 }
288
289 s->refcount_table[refcount_table_index] = new_block;
290 return new_block;
291 }
292
293 /*
294 * If we come here, we need to grow the refcount table. Again, a new
295 * refcount table needs some space and we can't simply allocate to avoid
296 * endless recursion.
297 *
298 * Therefore let's grab new refcount blocks at the end of the image, which
299 * will describe themselves and the new refcount table. This way we can
300 * reference them only in the new table and do the switch to the new
301 * refcount table at once without producing an inconsistent state in
302 * between.
303 */
66f82cee 304 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
8252278a 305
92dcb59f
KW
306 /* Calculate the number of refcount blocks needed so far */
307 uint64_t refcount_block_clusters = 1 << (s->cluster_bits - REFCOUNT_SHIFT);
308 uint64_t blocks_used = (s->free_cluster_index +
309 refcount_block_clusters - 1) / refcount_block_clusters;
310
311 /* And now we need at least one block more for the new metadata */
312 uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
313 uint64_t last_table_size;
314 uint64_t blocks_clusters;
315 do {
316 uint64_t table_clusters = size_to_clusters(s, table_size);
317 blocks_clusters = 1 +
318 ((table_clusters + refcount_block_clusters - 1)
319 / refcount_block_clusters);
320 uint64_t meta_clusters = table_clusters + blocks_clusters;
321
322 last_table_size = table_size;
323 table_size = next_refcount_table_size(s, blocks_used +
324 ((meta_clusters + refcount_block_clusters - 1)
325 / refcount_block_clusters));
326
327 } while (last_table_size != table_size);
328
329#ifdef DEBUG_ALLOC2
330 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
331 s->refcount_table_size, table_size);
332#endif
333
334 /* Create the new refcount table and blocks */
335 uint64_t meta_offset = (blocks_used * refcount_block_clusters) *
336 s->cluster_size;
337 uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
338 uint16_t *new_blocks = qemu_mallocz(blocks_clusters * s->cluster_size);
339 uint64_t *new_table = qemu_mallocz(table_size * sizeof(uint64_t));
340
341 assert(meta_offset >= (s->free_cluster_index * s->cluster_size));
342
343 /* Fill the new refcount table */
f7d0fe02 344 memcpy(new_table, s->refcount_table,
92dcb59f
KW
345 s->refcount_table_size * sizeof(uint64_t));
346 new_table[refcount_table_index] = new_block;
347
348 int i;
349 for (i = 0; i < blocks_clusters; i++) {
350 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
351 }
352
353 /* Fill the refcount blocks */
354 uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
355 int block = 0;
356 for (i = 0; i < table_clusters + blocks_clusters; i++) {
357 new_blocks[block++] = cpu_to_be16(1);
358 }
359
360 /* Write refcount blocks to disk */
66f82cee 361 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
8b3b7206 362 ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
92dcb59f
KW
363 blocks_clusters * s->cluster_size);
364 qemu_free(new_blocks);
365 if (ret < 0) {
366 goto fail_table;
367 }
368
369 /* Write refcount table to disk */
370 for(i = 0; i < table_size; i++) {
371 cpu_to_be64s(&new_table[i]);
372 }
373
66f82cee 374 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
8b3b7206 375 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
92dcb59f
KW
376 table_size * sizeof(uint64_t));
377 if (ret < 0) {
378 goto fail_table;
379 }
380
381 for(i = 0; i < table_size; i++) {
f7d0fe02 382 cpu_to_be64s(&new_table[i]);
92dcb59f 383 }
f7d0fe02 384
92dcb59f
KW
385 /* Hook up the new refcount table in the qcow2 header */
386 uint8_t data[12];
f7d0fe02 387 cpu_to_be64w((uint64_t*)data, table_offset);
92dcb59f 388 cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
66f82cee 389 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
8b3b7206 390 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
92dcb59f
KW
391 data, sizeof(data));
392 if (ret < 0) {
393 goto fail_table;
f2b7c8b3
KW
394 }
395
92dcb59f
KW
396 /* And switch it in memory */
397 uint64_t old_table_offset = s->refcount_table_offset;
398 uint64_t old_table_size = s->refcount_table_size;
399
f7d0fe02 400 qemu_free(s->refcount_table);
f7d0fe02 401 s->refcount_table = new_table;
92dcb59f 402 s->refcount_table_size = table_size;
f7d0fe02
KW
403 s->refcount_table_offset = table_offset;
404
92dcb59f
KW
405 /* Free old table. Remember, we must not change free_cluster_index */
406 uint64_t old_free_cluster_index = s->free_cluster_index;
ed6ccf0f 407 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
92dcb59f 408 s->free_cluster_index = old_free_cluster_index;
f7d0fe02 409
92dcb59f
KW
410 ret = load_refcount_block(bs, new_block);
411 if (ret < 0) {
412 goto fail_block;
f7d0fe02
KW
413 }
414
92dcb59f 415 return new_block;
f7d0fe02 416
92dcb59f
KW
417fail_table:
418 qemu_free(new_table);
419fail_block:
420 s->refcount_block_cache_offset = 0;
421 return ret;
f7d0fe02
KW
422}
423
9923e05e 424#define REFCOUNTS_PER_SECTOR (512 >> REFCOUNT_SHIFT)
66f82cee 425static int write_refcount_block_entries(BlockDriverState *bs,
9923e05e
KW
426 int64_t refcount_block_offset, int first_index, int last_index)
427{
66f82cee 428 BDRVQcowState *s = bs->opaque;
9923e05e 429 size_t size;
ed0df867 430 int ret;
9923e05e 431
3b88e52b
KW
432 if (cache_refcount_updates) {
433 return 0;
434 }
435
86fa8da8
KW
436 if (first_index < 0) {
437 return 0;
438 }
439
9923e05e
KW
440 first_index &= ~(REFCOUNTS_PER_SECTOR - 1);
441 last_index = (last_index + REFCOUNTS_PER_SECTOR)
442 & ~(REFCOUNTS_PER_SECTOR - 1);
443
444 size = (last_index - first_index) << REFCOUNT_SHIFT;
ed0df867 445
66f82cee 446 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE_PART);
8b3b7206 447 ret = bdrv_pwrite_sync(bs->file,
9923e05e 448 refcount_block_offset + (first_index << REFCOUNT_SHIFT),
ed0df867
KW
449 &s->refcount_block_cache[first_index], size);
450 if (ret < 0) {
451 return ret;
9923e05e
KW
452 }
453
454 return 0;
455}
456
f7d0fe02 457/* XXX: cache several refcount block clusters ? */
db3a964f
KW
458static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
459 int64_t offset, int64_t length, int addend)
f7d0fe02
KW
460{
461 BDRVQcowState *s = bs->opaque;
462 int64_t start, last, cluster_offset;
463 int64_t refcount_block_offset = 0;
464 int64_t table_index = -1, old_table_index;
465 int first_index = -1, last_index = -1;
09508d13 466 int ret;
f7d0fe02
KW
467
468#ifdef DEBUG_ALLOC2
0bf9e31a 469 printf("update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
f7d0fe02
KW
470 offset, length, addend);
471#endif
7322afe7 472 if (length < 0) {
f7d0fe02 473 return -EINVAL;
7322afe7
KW
474 } else if (length == 0) {
475 return 0;
476 }
477
f7d0fe02
KW
478 start = offset & ~(s->cluster_size - 1);
479 last = (offset + length - 1) & ~(s->cluster_size - 1);
480 for(cluster_offset = start; cluster_offset <= last;
481 cluster_offset += s->cluster_size)
482 {
483 int block_index, refcount;
484 int64_t cluster_index = cluster_offset >> s->cluster_bits;
09508d13 485 int64_t new_block;
f7d0fe02
KW
486
487 /* Only write refcount block to disk when we are done with it */
488 old_table_index = table_index;
489 table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
490 if ((old_table_index >= 0) && (table_index != old_table_index)) {
9923e05e 491
ed0df867
KW
492 ret = write_refcount_block_entries(bs, refcount_block_offset,
493 first_index, last_index);
494 if (ret < 0) {
495 return ret;
f7d0fe02
KW
496 }
497
498 first_index = -1;
499 last_index = -1;
500 }
501
502 /* Load the refcount block and allocate it if needed */
09508d13
KW
503 new_block = alloc_refcount_block(bs, cluster_index);
504 if (new_block < 0) {
505 ret = new_block;
506 goto fail;
f7d0fe02 507 }
09508d13 508 refcount_block_offset = new_block;
f7d0fe02
KW
509
510 /* we can update the count and save it */
511 block_index = cluster_index &
512 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
513 if (first_index == -1 || block_index < first_index) {
514 first_index = block_index;
515 }
516 if (block_index > last_index) {
517 last_index = block_index;
518 }
519
520 refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
521 refcount += addend;
09508d13
KW
522 if (refcount < 0 || refcount > 0xffff) {
523 ret = -EINVAL;
524 goto fail;
525 }
f7d0fe02
KW
526 if (refcount == 0 && cluster_index < s->free_cluster_index) {
527 s->free_cluster_index = cluster_index;
528 }
529 s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
530 }
531
09508d13
KW
532 ret = 0;
533fail:
534
f7d0fe02
KW
535 /* Write last changed block to disk */
536 if (refcount_block_offset != 0) {
ed0df867
KW
537 int wret;
538 wret = write_refcount_block_entries(bs, refcount_block_offset,
539 first_index, last_index);
540 if (wret < 0) {
541 return ret < 0 ? ret : wret;
f7d0fe02
KW
542 }
543 }
544
09508d13
KW
545 /*
546 * Try do undo any updates if an error is returned (This may succeed in
547 * some cases like ENOSPC for allocating a new refcount block)
548 */
549 if (ret < 0) {
550 int dummy;
551 dummy = update_refcount(bs, offset, cluster_offset - offset, -addend);
552 }
553
554 return ret;
f7d0fe02
KW
555}
556
018faafd
KW
557/*
558 * Increases or decreases the refcount of a given cluster by one.
559 * addend must be 1 or -1.
560 *
561 * If the return value is non-negative, it is the new refcount of the cluster.
562 * If it is negative, it is -errno and indicates an error.
563 */
f7d0fe02
KW
564static int update_cluster_refcount(BlockDriverState *bs,
565 int64_t cluster_index,
566 int addend)
567{
568 BDRVQcowState *s = bs->opaque;
569 int ret;
570
571 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
572 if (ret < 0) {
573 return ret;
574 }
575
576 return get_refcount(bs, cluster_index);
577}
578
579
580
581/*********************************************************/
582/* cluster allocation functions */
583
584
585
586/* return < 0 if error */
587static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
588{
589 BDRVQcowState *s = bs->opaque;
2eaa8f63 590 int i, nb_clusters, refcount;
f7d0fe02
KW
591
592 nb_clusters = size_to_clusters(s, size);
593retry:
594 for(i = 0; i < nb_clusters; i++) {
508e0893 595 int64_t next_cluster_index = s->free_cluster_index++;
2eaa8f63
KW
596 refcount = get_refcount(bs, next_cluster_index);
597
598 if (refcount < 0) {
599 return refcount;
600 } else if (refcount != 0) {
f7d0fe02 601 goto retry;
2eaa8f63 602 }
f7d0fe02
KW
603 }
604#ifdef DEBUG_ALLOC2
0bf9e31a 605 printf("alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
f7d0fe02
KW
606 size,
607 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
608#endif
609 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
610}
611
ed6ccf0f 612int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
f7d0fe02
KW
613{
614 int64_t offset;
db3a964f 615 int ret;
f7d0fe02 616
66f82cee 617 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
f7d0fe02 618 offset = alloc_clusters_noref(bs, size);
2eaa8f63
KW
619 if (offset < 0) {
620 return offset;
621 }
622
db3a964f
KW
623 ret = update_refcount(bs, offset, size, 1);
624 if (ret < 0) {
625 return ret;
626 }
f7d0fe02
KW
627 return offset;
628}
629
630/* only used to allocate compressed sectors. We try to allocate
631 contiguous sectors. size must be <= cluster_size */
ed6ccf0f 632int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
f7d0fe02
KW
633{
634 BDRVQcowState *s = bs->opaque;
635 int64_t offset, cluster_offset;
636 int free_in_cluster;
637
66f82cee 638 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
f7d0fe02
KW
639 assert(size > 0 && size <= s->cluster_size);
640 if (s->free_byte_offset == 0) {
ed6ccf0f 641 s->free_byte_offset = qcow2_alloc_clusters(bs, s->cluster_size);
5d757b56
KW
642 if (s->free_byte_offset < 0) {
643 return s->free_byte_offset;
644 }
f7d0fe02
KW
645 }
646 redo:
647 free_in_cluster = s->cluster_size -
648 (s->free_byte_offset & (s->cluster_size - 1));
649 if (size <= free_in_cluster) {
650 /* enough space in current cluster */
651 offset = s->free_byte_offset;
652 s->free_byte_offset += size;
653 free_in_cluster -= size;
654 if (free_in_cluster == 0)
655 s->free_byte_offset = 0;
656 if ((offset & (s->cluster_size - 1)) != 0)
657 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
658 } else {
ed6ccf0f 659 offset = qcow2_alloc_clusters(bs, s->cluster_size);
5d757b56
KW
660 if (offset < 0) {
661 return offset;
662 }
f7d0fe02
KW
663 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
664 if ((cluster_offset + s->cluster_size) == offset) {
665 /* we are lucky: contiguous data */
666 offset = s->free_byte_offset;
667 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
668 s->free_byte_offset += size;
669 } else {
670 s->free_byte_offset = offset;
671 goto redo;
672 }
673 }
674 return offset;
675}
676
ed6ccf0f 677void qcow2_free_clusters(BlockDriverState *bs,
f7d0fe02
KW
678 int64_t offset, int64_t size)
679{
db3a964f
KW
680 int ret;
681
66f82cee 682 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
db3a964f
KW
683 ret = update_refcount(bs, offset, size, -1);
684 if (ret < 0) {
685 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
003fad6e 686 /* TODO Remember the clusters to free them later and avoid leaking */
db3a964f 687 }
f7d0fe02
KW
688}
689
45aba42f
KW
690/*
691 * free_any_clusters
692 *
693 * free clusters according to its type: compressed or not
694 *
695 */
696
ed6ccf0f 697void qcow2_free_any_clusters(BlockDriverState *bs,
45aba42f
KW
698 uint64_t cluster_offset, int nb_clusters)
699{
700 BDRVQcowState *s = bs->opaque;
701
702 /* free the cluster */
703
704 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
705 int nb_csectors;
706 nb_csectors = ((cluster_offset >> s->csize_shift) &
707 s->csize_mask) + 1;
ed6ccf0f
KW
708 qcow2_free_clusters(bs,
709 (cluster_offset & s->cluster_offset_mask) & ~511,
710 nb_csectors * 512);
45aba42f
KW
711 return;
712 }
713
ed6ccf0f 714 qcow2_free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
45aba42f
KW
715
716 return;
717}
718
f7d0fe02
KW
719
720
721/*********************************************************/
722/* snapshots and image creation */
723
724
725
ed6ccf0f
KW
726void qcow2_create_refcount_update(QCowCreateState *s, int64_t offset,
727 int64_t size)
f7d0fe02
KW
728{
729 int refcount;
730 int64_t start, last, cluster_offset;
731 uint16_t *p;
732
733 start = offset & ~(s->cluster_size - 1);
734 last = (offset + size - 1) & ~(s->cluster_size - 1);
735 for(cluster_offset = start; cluster_offset <= last;
736 cluster_offset += s->cluster_size) {
737 p = &s->refcount_block[cluster_offset >> s->cluster_bits];
738 refcount = be16_to_cpu(*p);
739 refcount++;
740 *p = cpu_to_be16(refcount);
741 }
742}
743
744/* update the refcounts of snapshots and the copied flag */
ed6ccf0f
KW
745int qcow2_update_snapshot_refcount(BlockDriverState *bs,
746 int64_t l1_table_offset, int l1_size, int addend)
f7d0fe02
KW
747{
748 BDRVQcowState *s = bs->opaque;
749 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
750 int64_t old_offset, old_l2_offset;
751 int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
752
ed6ccf0f 753 qcow2_l2_cache_reset(bs);
3b88e52b 754 cache_refcount_updates = 1;
f7d0fe02
KW
755
756 l2_table = NULL;
757 l1_table = NULL;
758 l1_size2 = l1_size * sizeof(uint64_t);
f7d0fe02 759 if (l1_table_offset != s->l1_table_offset) {
702ef63f
KW
760 if (l1_size2 != 0) {
761 l1_table = qemu_mallocz(align_offset(l1_size2, 512));
762 } else {
763 l1_table = NULL;
764 }
f7d0fe02 765 l1_allocated = 1;
66f82cee 766 if (bdrv_pread(bs->file, l1_table_offset,
f7d0fe02
KW
767 l1_table, l1_size2) != l1_size2)
768 goto fail;
769 for(i = 0;i < l1_size; i++)
770 be64_to_cpus(&l1_table[i]);
771 } else {
772 assert(l1_size == s->l1_size);
773 l1_table = s->l1_table;
774 l1_allocated = 0;
775 }
776
777 l2_size = s->l2_size * sizeof(uint64_t);
778 l2_table = qemu_malloc(l2_size);
779 l1_modified = 0;
780 for(i = 0; i < l1_size; i++) {
781 l2_offset = l1_table[i];
782 if (l2_offset) {
783 old_l2_offset = l2_offset;
784 l2_offset &= ~QCOW_OFLAG_COPIED;
785 l2_modified = 0;
66f82cee 786 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
f7d0fe02
KW
787 goto fail;
788 for(j = 0; j < s->l2_size; j++) {
789 offset = be64_to_cpu(l2_table[j]);
790 if (offset != 0) {
791 old_offset = offset;
792 offset &= ~QCOW_OFLAG_COPIED;
793 if (offset & QCOW_OFLAG_COMPRESSED) {
794 nb_csectors = ((offset >> s->csize_shift) &
795 s->csize_mask) + 1;
db3a964f
KW
796 if (addend != 0) {
797 int ret;
798 ret = update_refcount(bs,
799 (offset & s->cluster_offset_mask) & ~511,
800 nb_csectors * 512, addend);
801 if (ret < 0) {
802 goto fail;
803 }
804 }
f7d0fe02
KW
805 /* compressed clusters are never modified */
806 refcount = 2;
807 } else {
808 if (addend != 0) {
809 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
810 } else {
811 refcount = get_refcount(bs, offset >> s->cluster_bits);
812 }
018faafd
KW
813
814 if (refcount < 0) {
815 goto fail;
816 }
f7d0fe02
KW
817 }
818
819 if (refcount == 1) {
820 offset |= QCOW_OFLAG_COPIED;
821 }
822 if (offset != old_offset) {
823 l2_table[j] = cpu_to_be64(offset);
824 l2_modified = 1;
825 }
826 }
827 }
828 if (l2_modified) {
8b3b7206
KW
829 if (bdrv_pwrite_sync(bs->file,
830 l2_offset, l2_table, l2_size) < 0)
f7d0fe02
KW
831 goto fail;
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 }
018faafd
KW
839 if (refcount < 0) {
840 goto fail;
841 } else if (refcount == 1) {
f7d0fe02
KW
842 l2_offset |= QCOW_OFLAG_COPIED;
843 }
844 if (l2_offset != old_l2_offset) {
845 l1_table[i] = l2_offset;
846 l1_modified = 1;
847 }
848 }
849 }
850 if (l1_modified) {
851 for(i = 0; i < l1_size; i++)
852 cpu_to_be64s(&l1_table[i]);
8b3b7206
KW
853 if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table,
854 l1_size2) < 0)
f7d0fe02
KW
855 goto fail;
856 for(i = 0; i < l1_size; i++)
857 be64_to_cpus(&l1_table[i]);
858 }
859 if (l1_allocated)
860 qemu_free(l1_table);
861 qemu_free(l2_table);
3b88e52b 862 cache_refcount_updates = 0;
66f82cee 863 write_refcount_block(bs);
f7d0fe02
KW
864 return 0;
865 fail:
866 if (l1_allocated)
867 qemu_free(l1_table);
868 qemu_free(l2_table);
3b88e52b 869 cache_refcount_updates = 0;
66f82cee 870 write_refcount_block(bs);
f7d0fe02
KW
871 return -EIO;
872}
873
874
875
876
877/*********************************************************/
878/* refcount checking functions */
879
880
881
882/*
883 * Increases the refcount for a range of clusters in a given refcount table.
884 * This is used to construct a temporary refcount table out of L1 and L2 tables
885 * which can be compared the the refcount table saved in the image.
886 *
887 * Returns the number of errors in the image that were found
888 */
889static int inc_refcounts(BlockDriverState *bs,
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 int errors = 0;
898
899 if (size <= 0)
900 return 0;
901
902 start = offset & ~(s->cluster_size - 1);
903 last = (offset + size - 1) & ~(s->cluster_size - 1);
904 for(cluster_offset = start; cluster_offset <= last;
905 cluster_offset += s->cluster_size) {
906 k = cluster_offset >> s->cluster_bits;
907 if (k < 0 || k >= refcount_table_size) {
908 fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
909 cluster_offset);
910 errors++;
911 } else {
912 if (++refcount_table[k] == 0) {
913 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
914 "\n", cluster_offset);
915 errors++;
916 }
917 }
918 }
919
920 return errors;
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 */
931static int check_refcounts_l2(BlockDriverState *bs,
932 uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
933 int check_copied)
934{
935 BDRVQcowState *s = bs->opaque;
936 uint64_t *l2_table, offset;
937 int i, l2_size, nb_csectors, refcount;
938 int errors = 0;
939
940 /* Read L2 table from disk */
941 l2_size = s->l2_size * sizeof(uint64_t);
942 l2_table = qemu_malloc(l2_size);
943
66f82cee 944 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
f7d0fe02
KW
945 goto fail;
946
947 /* Do the actual checks */
948 for(i = 0; i < s->l2_size; i++) {
949 offset = be64_to_cpu(l2_table[i]);
950 if (offset != 0) {
951 if (offset & QCOW_OFLAG_COMPRESSED) {
952 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
953 if (offset & QCOW_OFLAG_COPIED) {
954 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
955 "copied flag must never be set for compressed "
956 "clusters\n", offset >> s->cluster_bits);
957 offset &= ~QCOW_OFLAG_COPIED;
958 errors++;
959 }
960
961 /* Mark cluster as used */
962 nb_csectors = ((offset >> s->csize_shift) &
963 s->csize_mask) + 1;
964 offset &= s->cluster_offset_mask;
965 errors += inc_refcounts(bs, refcount_table,
966 refcount_table_size,
967 offset & ~511, nb_csectors * 512);
968 } else {
969 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
970 if (check_copied) {
971 uint64_t entry = offset;
972 offset &= ~QCOW_OFLAG_COPIED;
973 refcount = get_refcount(bs, offset >> s->cluster_bits);
018faafd
KW
974 if (refcount < 0) {
975 fprintf(stderr, "Can't get refcount for offset %"
976 PRIx64 ": %s\n", entry, strerror(-refcount));
977 }
f7d0fe02
KW
978 if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) {
979 fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
980 PRIx64 " refcount=%d\n", entry, refcount);
981 errors++;
982 }
983 }
984
985 /* Mark cluster as used */
986 offset &= ~QCOW_OFLAG_COPIED;
987 errors += inc_refcounts(bs, refcount_table,
988 refcount_table_size,
989 offset, s->cluster_size);
990
991 /* Correct offsets are cluster aligned */
992 if (offset & (s->cluster_size - 1)) {
993 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
994 "properly aligned; L2 entry corrupted.\n", offset);
995 errors++;
996 }
997 }
998 }
999 }
1000
1001 qemu_free(l2_table);
1002 return errors;
1003
1004fail:
1005 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1006 qemu_free(l2_table);
1007 return -EIO;
1008}
1009
1010/*
1011 * Increases the refcount for the L1 table, its L2 tables and all referenced
1012 * clusters in the given refcount table. While doing so, performs some checks
1013 * on L1 and L2 entries.
1014 *
1015 * Returns the number of errors found by the checks or -errno if an internal
1016 * error occurred.
1017 */
1018static int check_refcounts_l1(BlockDriverState *bs,
1019 uint16_t *refcount_table,
1020 int refcount_table_size,
1021 int64_t l1_table_offset, int l1_size,
1022 int check_copied)
1023{
1024 BDRVQcowState *s = bs->opaque;
1025 uint64_t *l1_table, l2_offset, l1_size2;
1026 int i, refcount, ret;
1027 int errors = 0;
1028
1029 l1_size2 = l1_size * sizeof(uint64_t);
1030
1031 /* Mark L1 table as used */
1032 errors += inc_refcounts(bs, refcount_table, refcount_table_size,
1033 l1_table_offset, l1_size2);
1034
1035 /* Read L1 table entries from disk */
702ef63f
KW
1036 if (l1_size2 == 0) {
1037 l1_table = NULL;
1038 } else {
1039 l1_table = qemu_malloc(l1_size2);
66f82cee 1040 if (bdrv_pread(bs->file, l1_table_offset,
702ef63f
KW
1041 l1_table, l1_size2) != l1_size2)
1042 goto fail;
1043 for(i = 0;i < l1_size; i++)
1044 be64_to_cpus(&l1_table[i]);
1045 }
f7d0fe02
KW
1046
1047 /* Do the actual checks */
1048 for(i = 0; i < l1_size; i++) {
1049 l2_offset = l1_table[i];
1050 if (l2_offset) {
1051 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
1052 if (check_copied) {
1053 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
1054 >> s->cluster_bits);
018faafd
KW
1055 if (refcount < 0) {
1056 fprintf(stderr, "Can't get refcount for l2_offset %"
1057 PRIx64 ": %s\n", l2_offset, strerror(-refcount));
1058 }
f7d0fe02
KW
1059 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
1060 fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
1061 " refcount=%d\n", l2_offset, refcount);
1062 errors++;
1063 }
1064 }
1065
1066 /* Mark L2 table as used */
1067 l2_offset &= ~QCOW_OFLAG_COPIED;
1068 errors += inc_refcounts(bs, refcount_table,
1069 refcount_table_size,
1070 l2_offset,
1071 s->cluster_size);
1072
1073 /* L2 tables are cluster aligned */
1074 if (l2_offset & (s->cluster_size - 1)) {
1075 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1076 "cluster aligned; L1 entry corrupted\n", l2_offset);
1077 errors++;
1078 }
1079
1080 /* Process and check L2 entries */
1081 ret = check_refcounts_l2(bs, refcount_table, refcount_table_size,
1082 l2_offset, check_copied);
1083 if (ret < 0) {
1084 goto fail;
1085 }
1086 errors += ret;
1087 }
1088 }
1089 qemu_free(l1_table);
1090 return errors;
1091
1092fail:
1093 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1094 qemu_free(l1_table);
1095 return -EIO;
1096}
1097
1098/*
1099 * Checks an image for refcount consistency.
1100 *
1101 * Returns 0 if no errors are found, the number of errors in case the image is
1102 * detected as corrupted, and -errno when an internal error occured.
1103 */
ed6ccf0f 1104int qcow2_check_refcounts(BlockDriverState *bs)
f7d0fe02
KW
1105{
1106 BDRVQcowState *s = bs->opaque;
1107 int64_t size;
1108 int nb_clusters, refcount1, refcount2, i;
1109 QCowSnapshot *sn;
1110 uint16_t *refcount_table;
1111 int ret, errors = 0;
1112
66f82cee 1113 size = bdrv_getlength(bs->file);
f7d0fe02
KW
1114 nb_clusters = size_to_clusters(s, size);
1115 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
1116
1117 /* header */
1118 errors += inc_refcounts(bs, refcount_table, nb_clusters,
1119 0, s->cluster_size);
1120
1121 /* current L1 table */
1122 ret = check_refcounts_l1(bs, refcount_table, nb_clusters,
1123 s->l1_table_offset, s->l1_size, 1);
1124 if (ret < 0) {
1125 return ret;
1126 }
1127 errors += ret;
1128
1129 /* snapshots */
1130 for(i = 0; i < s->nb_snapshots; i++) {
1131 sn = s->snapshots + i;
1132 check_refcounts_l1(bs, refcount_table, nb_clusters,
1133 sn->l1_table_offset, sn->l1_size, 0);
1134 }
1135 errors += inc_refcounts(bs, refcount_table, nb_clusters,
1136 s->snapshots_offset, s->snapshots_size);
1137
1138 /* refcount data */
1139 errors += inc_refcounts(bs, refcount_table, nb_clusters,
1140 s->refcount_table_offset,
1141 s->refcount_table_size * sizeof(uint64_t));
1142 for(i = 0; i < s->refcount_table_size; i++) {
1143 int64_t offset;
1144 offset = s->refcount_table[i];
746c3cb5
KW
1145
1146 /* Refcount blocks are cluster aligned */
1147 if (offset & (s->cluster_size - 1)) {
1148 fprintf(stderr, "ERROR refcount block %d is not "
1149 "cluster aligned; refcount table entry corrupted\n", i);
1150 errors++;
1151 }
1152
f7d0fe02
KW
1153 if (offset != 0) {
1154 errors += inc_refcounts(bs, refcount_table, nb_clusters,
1155 offset, s->cluster_size);
746c3cb5
KW
1156 if (refcount_table[offset / s->cluster_size] != 1) {
1157 fprintf(stderr, "ERROR refcount block %d refcount=%d\n",
1158 i, refcount_table[offset / s->cluster_size]);
1159 }
f7d0fe02
KW
1160 }
1161 }
1162
1163 /* compare ref counts */
1164 for(i = 0; i < nb_clusters; i++) {
1165 refcount1 = get_refcount(bs, i);
018faafd
KW
1166 if (refcount1 < 0) {
1167 fprintf(stderr, "Can't get refcount for cluster %d: %s\n",
1168 i, strerror(-refcount1));
1169 }
1170
f7d0fe02
KW
1171 refcount2 = refcount_table[i];
1172 if (refcount1 != refcount2) {
1173 fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n",
1174 i, refcount1, refcount2);
1175 errors++;
1176 }
1177 }
1178
1179 qemu_free(refcount_table);
1180
1181 return errors;
1182}
1183