]> git.proxmox.com Git - mirror_qemu.git/blame - block/qcow2-refcount.c
qcow2: Employ metadata overlap checks
[mirror_qemu.git] / block / qcow2-refcount.c
CommitLineData
f7d0fe02
KW
1/*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25#include "qemu-common.h"
737e150e 26#include "block/block_int.h"
f7d0fe02 27#include "block/qcow2.h"
a40f1c2a
HR
28#include "qemu/range.h"
29#include "qapi/qmp/types.h"
f7d0fe02
KW
30
31static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
92dcb59f 32static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
f7d0fe02 33 int64_t offset, int64_t length,
6cfcb9b8 34 int addend, enum qcow2_discard_type type);
f7d0fe02 35
3b88e52b 36
f7d0fe02
KW
37/*********************************************************/
38/* refcount handling */
39
ed6ccf0f 40int qcow2_refcount_init(BlockDriverState *bs)
f7d0fe02
KW
41{
42 BDRVQcowState *s = bs->opaque;
43 int ret, refcount_table_size2, i;
44
f7d0fe02 45 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
7267c094 46 s->refcount_table = g_malloc(refcount_table_size2);
f7d0fe02 47 if (s->refcount_table_size > 0) {
66f82cee
KW
48 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
49 ret = bdrv_pread(bs->file, s->refcount_table_offset,
f7d0fe02
KW
50 s->refcount_table, refcount_table_size2);
51 if (ret != refcount_table_size2)
52 goto fail;
53 for(i = 0; i < s->refcount_table_size; i++)
54 be64_to_cpus(&s->refcount_table[i]);
55 }
56 return 0;
57 fail:
58 return -ENOMEM;
59}
60
ed6ccf0f 61void qcow2_refcount_close(BlockDriverState *bs)
f7d0fe02
KW
62{
63 BDRVQcowState *s = bs->opaque;
7267c094 64 g_free(s->refcount_table);
f7d0fe02
KW
65}
66
67
68static int load_refcount_block(BlockDriverState *bs,
29c1a730
KW
69 int64_t refcount_block_offset,
70 void **refcount_block)
f7d0fe02
KW
71{
72 BDRVQcowState *s = bs->opaque;
73 int ret;
3b88e52b 74
66f82cee 75 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
29c1a730
KW
76 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
77 refcount_block);
e14e8ba5 78
29c1a730 79 return ret;
f7d0fe02
KW
80}
81
018faafd
KW
82/*
83 * Returns the refcount of the cluster given by its index. Any non-negative
84 * return value is the refcount of the cluster, negative values are -errno
85 * and indicate an error.
86 */
f7d0fe02
KW
87static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
88{
89 BDRVQcowState *s = bs->opaque;
90 int refcount_table_index, block_index;
91 int64_t refcount_block_offset;
018faafd 92 int ret;
29c1a730
KW
93 uint16_t *refcount_block;
94 uint16_t refcount;
f7d0fe02
KW
95
96 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
97 if (refcount_table_index >= s->refcount_table_size)
98 return 0;
99 refcount_block_offset = s->refcount_table[refcount_table_index];
100 if (!refcount_block_offset)
101 return 0;
29c1a730
KW
102
103 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
104 (void**) &refcount_block);
105 if (ret < 0) {
106 return ret;
f7d0fe02 107 }
29c1a730 108
f7d0fe02
KW
109 block_index = cluster_index &
110 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
29c1a730
KW
111 refcount = be16_to_cpu(refcount_block[block_index]);
112
113 ret = qcow2_cache_put(bs, s->refcount_block_cache,
114 (void**) &refcount_block);
115 if (ret < 0) {
116 return ret;
117 }
118
119 return refcount;
f7d0fe02
KW
120}
121
05121aed
KW
122/*
123 * Rounds the refcount table size up to avoid growing the table for each single
124 * refcount block that is allocated.
125 */
126static unsigned int next_refcount_table_size(BDRVQcowState *s,
127 unsigned int min_size)
128{
129 unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1;
130 unsigned int refcount_table_clusters =
131 MAX(1, s->refcount_table_size >> (s->cluster_bits - 3));
132
133 while (min_clusters > refcount_table_clusters) {
134 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
135 }
136
137 return refcount_table_clusters << (s->cluster_bits - 3);
138}
139
92dcb59f
KW
140
141/* Checks if two offsets are described by the same refcount block */
142static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a,
143 uint64_t offset_b)
144{
145 uint64_t block_a = offset_a >> (2 * s->cluster_bits - REFCOUNT_SHIFT);
146 uint64_t block_b = offset_b >> (2 * s->cluster_bits - REFCOUNT_SHIFT);
147
148 return (block_a == block_b);
149}
150
151/*
152 * Loads a refcount block. If it doesn't exist yet, it is allocated first
153 * (including growing the refcount table if needed).
154 *
29c1a730 155 * Returns 0 on success or -errno in error case
92dcb59f 156 */
29c1a730
KW
157static int alloc_refcount_block(BlockDriverState *bs,
158 int64_t cluster_index, uint16_t **refcount_block)
f7d0fe02
KW
159{
160 BDRVQcowState *s = bs->opaque;
92dcb59f
KW
161 unsigned int refcount_table_index;
162 int ret;
163
66f82cee 164 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
8252278a 165
92dcb59f
KW
166 /* Find the refcount block for the given cluster */
167 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
168
169 if (refcount_table_index < s->refcount_table_size) {
170
171 uint64_t refcount_block_offset =
76dc9e0c 172 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
92dcb59f
KW
173
174 /* If it's already there, we're done */
175 if (refcount_block_offset) {
29c1a730
KW
176 return load_refcount_block(bs, refcount_block_offset,
177 (void**) refcount_block);
92dcb59f
KW
178 }
179 }
180
181 /*
182 * If we came here, we need to allocate something. Something is at least
183 * a cluster for the new refcount block. It may also include a new refcount
184 * table if the old refcount table is too small.
185 *
186 * Note that allocating clusters here needs some special care:
187 *
188 * - We can't use the normal qcow2_alloc_clusters(), it would try to
189 * increase the refcount and very likely we would end up with an endless
190 * recursion. Instead we must place the refcount blocks in a way that
191 * they can describe them themselves.
192 *
193 * - We need to consider that at this point we are inside update_refcounts
194 * and doing the initial refcount increase. This means that some clusters
195 * have already been allocated by the caller, but their refcount isn't
196 * accurate yet. free_cluster_index tells us where this allocation ends
197 * as long as we don't overwrite it by freeing clusters.
198 *
199 * - alloc_clusters_noref and qcow2_free_clusters may load a different
200 * refcount block into the cache
201 */
202
29c1a730
KW
203 *refcount_block = NULL;
204
205 /* We write to the refcount table, so we might depend on L2 tables */
9991923b
SH
206 ret = qcow2_cache_flush(bs, s->l2_table_cache);
207 if (ret < 0) {
208 return ret;
209 }
92dcb59f
KW
210
211 /* Allocate the refcount block itself and mark it as used */
2eaa8f63
KW
212 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
213 if (new_block < 0) {
214 return new_block;
215 }
f7d0fe02 216
f7d0fe02 217#ifdef DEBUG_ALLOC2
92dcb59f
KW
218 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
219 " at %" PRIx64 "\n",
220 refcount_table_index, cluster_index << s->cluster_bits, new_block);
f7d0fe02 221#endif
92dcb59f
KW
222
223 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
25408c09 224 /* Zero the new refcount block before updating it */
29c1a730
KW
225 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
226 (void**) refcount_block);
227 if (ret < 0) {
228 goto fail_block;
229 }
230
231 memset(*refcount_block, 0, s->cluster_size);
25408c09 232
92dcb59f
KW
233 /* The block describes itself, need to update the cache */
234 int block_index = (new_block >> s->cluster_bits) &
235 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
29c1a730 236 (*refcount_block)[block_index] = cpu_to_be16(1);
92dcb59f
KW
237 } else {
238 /* Described somewhere else. This can recurse at most twice before we
239 * arrive at a block that describes itself. */
6cfcb9b8
KW
240 ret = update_refcount(bs, new_block, s->cluster_size, 1,
241 QCOW2_DISCARD_NEVER);
92dcb59f
KW
242 if (ret < 0) {
243 goto fail_block;
244 }
25408c09 245
9991923b
SH
246 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
247 if (ret < 0) {
248 goto fail_block;
249 }
1c4c2814 250
25408c09
KW
251 /* Initialize the new refcount block only after updating its refcount,
252 * update_refcount uses the refcount cache itself */
29c1a730
KW
253 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
254 (void**) refcount_block);
255 if (ret < 0) {
256 goto fail_block;
257 }
258
259 memset(*refcount_block, 0, s->cluster_size);
92dcb59f
KW
260 }
261
262 /* Now the new refcount block needs to be written to disk */
66f82cee 263 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
29c1a730
KW
264 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block);
265 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
92dcb59f
KW
266 if (ret < 0) {
267 goto fail_block;
268 }
269
270 /* If the refcount table is big enough, just hook the block up there */
271 if (refcount_table_index < s->refcount_table_size) {
272 uint64_t data64 = cpu_to_be64(new_block);
66f82cee 273 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
8b3b7206 274 ret = bdrv_pwrite_sync(bs->file,
92dcb59f
KW
275 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
276 &data64, sizeof(data64));
277 if (ret < 0) {
278 goto fail_block;
279 }
280
281 s->refcount_table[refcount_table_index] = new_block;
29c1a730
KW
282 return 0;
283 }
284
285 ret = qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
286 if (ret < 0) {
287 goto fail_block;
92dcb59f
KW
288 }
289
290 /*
291 * If we come here, we need to grow the refcount table. Again, a new
292 * refcount table needs some space and we can't simply allocate to avoid
293 * endless recursion.
294 *
295 * Therefore let's grab new refcount blocks at the end of the image, which
296 * will describe themselves and the new refcount table. This way we can
297 * reference them only in the new table and do the switch to the new
298 * refcount table at once without producing an inconsistent state in
299 * between.
300 */
66f82cee 301 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
8252278a 302
92dcb59f
KW
303 /* Calculate the number of refcount blocks needed so far */
304 uint64_t refcount_block_clusters = 1 << (s->cluster_bits - REFCOUNT_SHIFT);
305 uint64_t blocks_used = (s->free_cluster_index +
306 refcount_block_clusters - 1) / refcount_block_clusters;
307
308 /* And now we need at least one block more for the new metadata */
309 uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
310 uint64_t last_table_size;
311 uint64_t blocks_clusters;
312 do {
a3548077
KW
313 uint64_t table_clusters =
314 size_to_clusters(s, table_size * sizeof(uint64_t));
92dcb59f
KW
315 blocks_clusters = 1 +
316 ((table_clusters + refcount_block_clusters - 1)
317 / refcount_block_clusters);
318 uint64_t meta_clusters = table_clusters + blocks_clusters;
319
320 last_table_size = table_size;
321 table_size = next_refcount_table_size(s, blocks_used +
322 ((meta_clusters + refcount_block_clusters - 1)
323 / refcount_block_clusters));
324
325 } while (last_table_size != table_size);
326
327#ifdef DEBUG_ALLOC2
328 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
329 s->refcount_table_size, table_size);
330#endif
331
332 /* Create the new refcount table and blocks */
333 uint64_t meta_offset = (blocks_used * refcount_block_clusters) *
334 s->cluster_size;
335 uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
7267c094
AL
336 uint16_t *new_blocks = g_malloc0(blocks_clusters * s->cluster_size);
337 uint64_t *new_table = g_malloc0(table_size * sizeof(uint64_t));
92dcb59f
KW
338
339 assert(meta_offset >= (s->free_cluster_index * s->cluster_size));
340
341 /* Fill the new refcount table */
f7d0fe02 342 memcpy(new_table, s->refcount_table,
92dcb59f
KW
343 s->refcount_table_size * sizeof(uint64_t));
344 new_table[refcount_table_index] = new_block;
345
346 int i;
347 for (i = 0; i < blocks_clusters; i++) {
348 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
349 }
350
351 /* Fill the refcount blocks */
352 uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
353 int block = 0;
354 for (i = 0; i < table_clusters + blocks_clusters; i++) {
355 new_blocks[block++] = cpu_to_be16(1);
356 }
357
358 /* Write refcount blocks to disk */
66f82cee 359 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
8b3b7206 360 ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
92dcb59f 361 blocks_clusters * s->cluster_size);
7267c094 362 g_free(new_blocks);
92dcb59f
KW
363 if (ret < 0) {
364 goto fail_table;
365 }
366
367 /* Write refcount table to disk */
368 for(i = 0; i < table_size; i++) {
369 cpu_to_be64s(&new_table[i]);
370 }
371
66f82cee 372 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
8b3b7206 373 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
92dcb59f
KW
374 table_size * sizeof(uint64_t));
375 if (ret < 0) {
376 goto fail_table;
377 }
378
379 for(i = 0; i < table_size; i++) {
87267753 380 be64_to_cpus(&new_table[i]);
92dcb59f 381 }
f7d0fe02 382
92dcb59f
KW
383 /* Hook up the new refcount table in the qcow2 header */
384 uint8_t data[12];
f7d0fe02 385 cpu_to_be64w((uint64_t*)data, table_offset);
92dcb59f 386 cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
66f82cee 387 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
8b3b7206 388 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
92dcb59f
KW
389 data, sizeof(data));
390 if (ret < 0) {
391 goto fail_table;
f2b7c8b3
KW
392 }
393
92dcb59f
KW
394 /* And switch it in memory */
395 uint64_t old_table_offset = s->refcount_table_offset;
396 uint64_t old_table_size = s->refcount_table_size;
397
7267c094 398 g_free(s->refcount_table);
f7d0fe02 399 s->refcount_table = new_table;
92dcb59f 400 s->refcount_table_size = table_size;
f7d0fe02
KW
401 s->refcount_table_offset = table_offset;
402
92dcb59f
KW
403 /* Free old table. Remember, we must not change free_cluster_index */
404 uint64_t old_free_cluster_index = s->free_cluster_index;
6cfcb9b8
KW
405 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t),
406 QCOW2_DISCARD_OTHER);
92dcb59f 407 s->free_cluster_index = old_free_cluster_index;
f7d0fe02 408
29c1a730 409 ret = load_refcount_block(bs, new_block, (void**) refcount_block);
92dcb59f 410 if (ret < 0) {
29c1a730 411 return ret;
f7d0fe02
KW
412 }
413
2795ecf6 414 return 0;
f7d0fe02 415
92dcb59f 416fail_table:
7267c094 417 g_free(new_table);
92dcb59f 418fail_block:
29c1a730
KW
419 if (*refcount_block != NULL) {
420 qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block);
3b88e52b 421 }
29c1a730 422 return ret;
9923e05e
KW
423}
424
0b919fae
KW
425void qcow2_process_discards(BlockDriverState *bs, int ret)
426{
427 BDRVQcowState *s = bs->opaque;
428 Qcow2DiscardRegion *d, *next;
429
430 QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) {
431 QTAILQ_REMOVE(&s->discards, d, next);
432
433 /* Discard is optional, ignore the return value */
434 if (ret >= 0) {
435 bdrv_discard(bs->file,
436 d->offset >> BDRV_SECTOR_BITS,
437 d->bytes >> BDRV_SECTOR_BITS);
438 }
439
440 g_free(d);
441 }
442}
443
444static void update_refcount_discard(BlockDriverState *bs,
445 uint64_t offset, uint64_t length)
446{
447 BDRVQcowState *s = bs->opaque;
448 Qcow2DiscardRegion *d, *p, *next;
449
450 QTAILQ_FOREACH(d, &s->discards, next) {
451 uint64_t new_start = MIN(offset, d->offset);
452 uint64_t new_end = MAX(offset + length, d->offset + d->bytes);
453
454 if (new_end - new_start <= length + d->bytes) {
455 /* There can't be any overlap, areas ending up here have no
456 * references any more and therefore shouldn't get freed another
457 * time. */
458 assert(d->bytes + length == new_end - new_start);
459 d->offset = new_start;
460 d->bytes = new_end - new_start;
461 goto found;
462 }
463 }
464
465 d = g_malloc(sizeof(*d));
466 *d = (Qcow2DiscardRegion) {
467 .bs = bs,
468 .offset = offset,
469 .bytes = length,
470 };
471 QTAILQ_INSERT_TAIL(&s->discards, d, next);
472
473found:
474 /* Merge discard requests if they are adjacent now */
475 QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) {
476 if (p == d
477 || p->offset > d->offset + d->bytes
478 || d->offset > p->offset + p->bytes)
479 {
480 continue;
481 }
482
483 /* Still no overlap possible */
484 assert(p->offset == d->offset + d->bytes
485 || d->offset == p->offset + p->bytes);
486
487 QTAILQ_REMOVE(&s->discards, p, next);
488 d->offset = MIN(d->offset, p->offset);
489 d->bytes += p->bytes;
490 }
491}
492
f7d0fe02 493/* XXX: cache several refcount block clusters ? */
db3a964f 494static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
6cfcb9b8 495 int64_t offset, int64_t length, int addend, enum qcow2_discard_type type)
f7d0fe02
KW
496{
497 BDRVQcowState *s = bs->opaque;
498 int64_t start, last, cluster_offset;
29c1a730
KW
499 uint16_t *refcount_block = NULL;
500 int64_t old_table_index = -1;
09508d13 501 int ret;
f7d0fe02
KW
502
503#ifdef DEBUG_ALLOC2
35ee5e39 504 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
f7d0fe02
KW
505 offset, length, addend);
506#endif
7322afe7 507 if (length < 0) {
f7d0fe02 508 return -EINVAL;
7322afe7
KW
509 } else if (length == 0) {
510 return 0;
511 }
512
29c1a730
KW
513 if (addend < 0) {
514 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
515 s->l2_table_cache);
516 }
517
f7d0fe02
KW
518 start = offset & ~(s->cluster_size - 1);
519 last = (offset + length - 1) & ~(s->cluster_size - 1);
520 for(cluster_offset = start; cluster_offset <= last;
521 cluster_offset += s->cluster_size)
522 {
523 int block_index, refcount;
524 int64_t cluster_index = cluster_offset >> s->cluster_bits;
29c1a730
KW
525 int64_t table_index =
526 cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
f7d0fe02 527
29c1a730
KW
528 /* Load the refcount block and allocate it if needed */
529 if (table_index != old_table_index) {
530 if (refcount_block) {
531 ret = qcow2_cache_put(bs, s->refcount_block_cache,
532 (void**) &refcount_block);
533 if (ret < 0) {
534 goto fail;
535 }
536 }
9923e05e 537
29c1a730 538 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
ed0df867 539 if (ret < 0) {
29c1a730 540 goto fail;
f7d0fe02 541 }
f7d0fe02 542 }
29c1a730 543 old_table_index = table_index;
f7d0fe02 544
29c1a730 545 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
f7d0fe02
KW
546
547 /* we can update the count and save it */
548 block_index = cluster_index &
549 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
f7d0fe02 550
29c1a730 551 refcount = be16_to_cpu(refcount_block[block_index]);
f7d0fe02 552 refcount += addend;
09508d13
KW
553 if (refcount < 0 || refcount > 0xffff) {
554 ret = -EINVAL;
555 goto fail;
556 }
f7d0fe02
KW
557 if (refcount == 0 && cluster_index < s->free_cluster_index) {
558 s->free_cluster_index = cluster_index;
559 }
29c1a730 560 refcount_block[block_index] = cpu_to_be16(refcount);
0b919fae 561
67af674e 562 if (refcount == 0 && s->discard_passthrough[type]) {
0b919fae 563 update_refcount_discard(bs, cluster_offset, s->cluster_size);
67af674e 564 }
f7d0fe02
KW
565 }
566
09508d13
KW
567 ret = 0;
568fail:
0b919fae
KW
569 if (!s->cache_discards) {
570 qcow2_process_discards(bs, ret);
571 }
572
f7d0fe02 573 /* Write last changed block to disk */
29c1a730 574 if (refcount_block) {
ed0df867 575 int wret;
29c1a730
KW
576 wret = qcow2_cache_put(bs, s->refcount_block_cache,
577 (void**) &refcount_block);
ed0df867
KW
578 if (wret < 0) {
579 return ret < 0 ? ret : wret;
f7d0fe02
KW
580 }
581 }
582
09508d13
KW
583 /*
584 * Try do undo any updates if an error is returned (This may succeed in
585 * some cases like ENOSPC for allocating a new refcount block)
586 */
587 if (ret < 0) {
588 int dummy;
6cfcb9b8
KW
589 dummy = update_refcount(bs, offset, cluster_offset - offset, -addend,
590 QCOW2_DISCARD_NEVER);
83e3f76c 591 (void)dummy;
09508d13
KW
592 }
593
594 return ret;
f7d0fe02
KW
595}
596
018faafd
KW
597/*
598 * Increases or decreases the refcount of a given cluster by one.
599 * addend must be 1 or -1.
600 *
601 * If the return value is non-negative, it is the new refcount of the cluster.
602 * If it is negative, it is -errno and indicates an error.
603 */
f7d0fe02
KW
604static int update_cluster_refcount(BlockDriverState *bs,
605 int64_t cluster_index,
6cfcb9b8
KW
606 int addend,
607 enum qcow2_discard_type type)
f7d0fe02
KW
608{
609 BDRVQcowState *s = bs->opaque;
610 int ret;
611
6cfcb9b8
KW
612 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend,
613 type);
f7d0fe02
KW
614 if (ret < 0) {
615 return ret;
616 }
617
618 return get_refcount(bs, cluster_index);
619}
620
621
622
623/*********************************************************/
624/* cluster allocation functions */
625
626
627
628/* return < 0 if error */
629static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
630{
631 BDRVQcowState *s = bs->opaque;
2eaa8f63 632 int i, nb_clusters, refcount;
f7d0fe02
KW
633
634 nb_clusters = size_to_clusters(s, size);
635retry:
636 for(i = 0; i < nb_clusters; i++) {
508e0893 637 int64_t next_cluster_index = s->free_cluster_index++;
2eaa8f63
KW
638 refcount = get_refcount(bs, next_cluster_index);
639
640 if (refcount < 0) {
641 return refcount;
642 } else if (refcount != 0) {
f7d0fe02 643 goto retry;
2eaa8f63 644 }
f7d0fe02
KW
645 }
646#ifdef DEBUG_ALLOC2
35ee5e39 647 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
f7d0fe02
KW
648 size,
649 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
650#endif
651 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
652}
653
ed6ccf0f 654int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
f7d0fe02
KW
655{
656 int64_t offset;
db3a964f 657 int ret;
f7d0fe02 658
66f82cee 659 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
f7d0fe02 660 offset = alloc_clusters_noref(bs, size);
2eaa8f63
KW
661 if (offset < 0) {
662 return offset;
663 }
664
6cfcb9b8 665 ret = update_refcount(bs, offset, size, 1, QCOW2_DISCARD_NEVER);
db3a964f
KW
666 if (ret < 0) {
667 return ret;
668 }
1c4c2814 669
f7d0fe02
KW
670 return offset;
671}
672
256900b1
KW
673int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
674 int nb_clusters)
675{
676 BDRVQcowState *s = bs->opaque;
677 uint64_t cluster_index;
f24423bd 678 uint64_t old_free_cluster_index;
256900b1
KW
679 int i, refcount, ret;
680
681 /* Check how many clusters there are free */
682 cluster_index = offset >> s->cluster_bits;
683 for(i = 0; i < nb_clusters; i++) {
684 refcount = get_refcount(bs, cluster_index++);
685
686 if (refcount < 0) {
687 return refcount;
688 } else if (refcount != 0) {
689 break;
690 }
691 }
692
693 /* And then allocate them */
f24423bd
KW
694 old_free_cluster_index = s->free_cluster_index;
695 s->free_cluster_index = cluster_index + i;
696
6cfcb9b8
KW
697 ret = update_refcount(bs, offset, i << s->cluster_bits, 1,
698 QCOW2_DISCARD_NEVER);
256900b1
KW
699 if (ret < 0) {
700 return ret;
701 }
702
f24423bd
KW
703 s->free_cluster_index = old_free_cluster_index;
704
256900b1
KW
705 return i;
706}
707
f7d0fe02
KW
708/* only used to allocate compressed sectors. We try to allocate
709 contiguous sectors. size must be <= cluster_size */
ed6ccf0f 710int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
f7d0fe02
KW
711{
712 BDRVQcowState *s = bs->opaque;
713 int64_t offset, cluster_offset;
714 int free_in_cluster;
715
66f82cee 716 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
f7d0fe02
KW
717 assert(size > 0 && size <= s->cluster_size);
718 if (s->free_byte_offset == 0) {
206e6d85
SH
719 offset = qcow2_alloc_clusters(bs, s->cluster_size);
720 if (offset < 0) {
721 return offset;
5d757b56 722 }
206e6d85 723 s->free_byte_offset = offset;
f7d0fe02
KW
724 }
725 redo:
726 free_in_cluster = s->cluster_size -
727 (s->free_byte_offset & (s->cluster_size - 1));
728 if (size <= free_in_cluster) {
729 /* enough space in current cluster */
730 offset = s->free_byte_offset;
731 s->free_byte_offset += size;
732 free_in_cluster -= size;
733 if (free_in_cluster == 0)
734 s->free_byte_offset = 0;
735 if ((offset & (s->cluster_size - 1)) != 0)
6cfcb9b8
KW
736 update_cluster_refcount(bs, offset >> s->cluster_bits, 1,
737 QCOW2_DISCARD_NEVER);
f7d0fe02 738 } else {
ed6ccf0f 739 offset = qcow2_alloc_clusters(bs, s->cluster_size);
5d757b56
KW
740 if (offset < 0) {
741 return offset;
742 }
f7d0fe02
KW
743 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
744 if ((cluster_offset + s->cluster_size) == offset) {
745 /* we are lucky: contiguous data */
746 offset = s->free_byte_offset;
6cfcb9b8
KW
747 update_cluster_refcount(bs, offset >> s->cluster_bits, 1,
748 QCOW2_DISCARD_NEVER);
f7d0fe02
KW
749 s->free_byte_offset += size;
750 } else {
751 s->free_byte_offset = offset;
752 goto redo;
753 }
754 }
29216ed1 755
c1f5bafd
SH
756 /* The cluster refcount was incremented, either by qcow2_alloc_clusters()
757 * or explicitly by update_cluster_refcount(). Refcount blocks must be
758 * flushed before the caller's L2 table updates.
759 */
760 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
f7d0fe02
KW
761 return offset;
762}
763
ed6ccf0f 764void qcow2_free_clusters(BlockDriverState *bs,
6cfcb9b8
KW
765 int64_t offset, int64_t size,
766 enum qcow2_discard_type type)
f7d0fe02 767{
db3a964f
KW
768 int ret;
769
66f82cee 770 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
6cfcb9b8 771 ret = update_refcount(bs, offset, size, -1, type);
db3a964f
KW
772 if (ret < 0) {
773 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
003fad6e 774 /* TODO Remember the clusters to free them later and avoid leaking */
db3a964f 775 }
f7d0fe02
KW
776}
777
45aba42f 778/*
c7a4c37a
KW
779 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
780 * normal cluster, compressed cluster, etc.)
45aba42f 781 */
6cfcb9b8
KW
782void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
783 int nb_clusters, enum qcow2_discard_type type)
45aba42f
KW
784{
785 BDRVQcowState *s = bs->opaque;
786
c7a4c37a
KW
787 switch (qcow2_get_cluster_type(l2_entry)) {
788 case QCOW2_CLUSTER_COMPRESSED:
789 {
790 int nb_csectors;
791 nb_csectors = ((l2_entry >> s->csize_shift) &
792 s->csize_mask) + 1;
793 qcow2_free_clusters(bs,
794 (l2_entry & s->cluster_offset_mask) & ~511,
6cfcb9b8 795 nb_csectors * 512, type);
c7a4c37a
KW
796 }
797 break;
798 case QCOW2_CLUSTER_NORMAL:
799 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
6cfcb9b8 800 nb_clusters << s->cluster_bits, type);
c7a4c37a
KW
801 break;
802 case QCOW2_CLUSTER_UNALLOCATED:
6377af48 803 case QCOW2_CLUSTER_ZERO:
c7a4c37a
KW
804 break;
805 default:
806 abort();
45aba42f 807 }
45aba42f
KW
808}
809
f7d0fe02
KW
810
811
812/*********************************************************/
813/* snapshots and image creation */
814
815
816
f7d0fe02 817/* update the refcounts of snapshots and the copied flag */
ed6ccf0f
KW
818int qcow2_update_snapshot_refcount(BlockDriverState *bs,
819 int64_t l1_table_offset, int l1_size, int addend)
f7d0fe02
KW
820{
821 BDRVQcowState *s = bs->opaque;
822 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
823 int64_t old_offset, old_l2_offset;
93913dfd 824 int i, j, l1_modified = 0, nb_csectors, refcount;
29c1a730 825 int ret;
f7d0fe02
KW
826
827 l2_table = NULL;
828 l1_table = NULL;
829 l1_size2 = l1_size * sizeof(uint64_t);
43a0cac4 830
0b919fae
KW
831 s->cache_discards = true;
832
43a0cac4
KW
833 /* WARNING: qcow2_snapshot_goto relies on this function not using the
834 * l1_table_offset when it is the current s->l1_table_offset! Be careful
835 * when changing this! */
f7d0fe02 836 if (l1_table_offset != s->l1_table_offset) {
6528499f 837 l1_table = g_malloc0(align_offset(l1_size2, 512));
f7d0fe02 838 l1_allocated = 1;
c2bc78b6
KW
839
840 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
841 if (ret < 0) {
f7d0fe02 842 goto fail;
93913dfd
KW
843 }
844
f7d0fe02
KW
845 for(i = 0;i < l1_size; i++)
846 be64_to_cpus(&l1_table[i]);
847 } else {
848 assert(l1_size == s->l1_size);
849 l1_table = s->l1_table;
850 l1_allocated = 0;
851 }
852
f7d0fe02
KW
853 for(i = 0; i < l1_size; i++) {
854 l2_offset = l1_table[i];
855 if (l2_offset) {
856 old_l2_offset = l2_offset;
8e37f681 857 l2_offset &= L1E_OFFSET_MASK;
29c1a730
KW
858
859 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
860 (void**) &l2_table);
861 if (ret < 0) {
f7d0fe02 862 goto fail;
29c1a730
KW
863 }
864
f7d0fe02 865 for(j = 0; j < s->l2_size; j++) {
8b81a7b6
HR
866 uint64_t cluster_index;
867
f7d0fe02 868 offset = be64_to_cpu(l2_table[j]);
8b81a7b6
HR
869 old_offset = offset;
870 offset &= ~QCOW_OFLAG_COPIED;
871
872 switch (qcow2_get_cluster_type(offset)) {
873 case QCOW2_CLUSTER_COMPRESSED:
f7d0fe02
KW
874 nb_csectors = ((offset >> s->csize_shift) &
875 s->csize_mask) + 1;
db3a964f
KW
876 if (addend != 0) {
877 int ret;
878 ret = update_refcount(bs,
879 (offset & s->cluster_offset_mask) & ~511,
6cfcb9b8
KW
880 nb_csectors * 512, addend,
881 QCOW2_DISCARD_SNAPSHOT);
db3a964f
KW
882 if (ret < 0) {
883 goto fail;
884 }
885 }
f7d0fe02
KW
886 /* compressed clusters are never modified */
887 refcount = 2;
8b81a7b6
HR
888 break;
889
890 case QCOW2_CLUSTER_NORMAL:
891 case QCOW2_CLUSTER_ZERO:
892 cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
893 if (!cluster_index) {
894 /* unallocated */
895 refcount = 0;
896 break;
897 }
f7d0fe02 898 if (addend != 0) {
6cfcb9b8
KW
899 refcount = update_cluster_refcount(bs, cluster_index, addend,
900 QCOW2_DISCARD_SNAPSHOT);
f7d0fe02 901 } else {
8e37f681 902 refcount = get_refcount(bs, cluster_index);
f7d0fe02 903 }
018faafd
KW
904
905 if (refcount < 0) {
c2bc78b6 906 ret = refcount;
018faafd
KW
907 goto fail;
908 }
8b81a7b6 909 break;
f7d0fe02 910
8b81a7b6
HR
911 case QCOW2_CLUSTER_UNALLOCATED:
912 refcount = 0;
913 break;
914
915 default:
916 abort();
917 }
918
919 if (refcount == 1) {
920 offset |= QCOW_OFLAG_COPIED;
921 }
922 if (offset != old_offset) {
923 if (addend > 0) {
924 qcow2_cache_set_dependency(bs, s->l2_table_cache,
925 s->refcount_block_cache);
f7d0fe02 926 }
8b81a7b6
HR
927 l2_table[j] = cpu_to_be64(offset);
928 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
f7d0fe02
KW
929 }
930 }
29c1a730
KW
931
932 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
933 if (ret < 0) {
934 goto fail;
f7d0fe02
KW
935 }
936
29c1a730 937
f7d0fe02 938 if (addend != 0) {
6cfcb9b8
KW
939 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend,
940 QCOW2_DISCARD_SNAPSHOT);
f7d0fe02
KW
941 } else {
942 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
943 }
018faafd 944 if (refcount < 0) {
c2bc78b6 945 ret = refcount;
018faafd
KW
946 goto fail;
947 } else if (refcount == 1) {
f7d0fe02
KW
948 l2_offset |= QCOW_OFLAG_COPIED;
949 }
950 if (l2_offset != old_l2_offset) {
951 l1_table[i] = l2_offset;
952 l1_modified = 1;
953 }
954 }
955 }
93913dfd 956
2154f24e 957 ret = bdrv_flush(bs);
93913dfd
KW
958fail:
959 if (l2_table) {
960 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
961 }
962
0b919fae
KW
963 s->cache_discards = false;
964 qcow2_process_discards(bs, ret);
965
43a0cac4 966 /* Update L1 only if it isn't deleted anyway (addend = -1) */
c2b6ff51
KW
967 if (ret == 0 && addend >= 0 && l1_modified) {
968 for (i = 0; i < l1_size; i++) {
f7d0fe02 969 cpu_to_be64s(&l1_table[i]);
c2b6ff51
KW
970 }
971
972 ret = bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table, l1_size2);
973
974 for (i = 0; i < l1_size; i++) {
f7d0fe02 975 be64_to_cpus(&l1_table[i]);
c2b6ff51 976 }
f7d0fe02
KW
977 }
978 if (l1_allocated)
7267c094 979 g_free(l1_table);
93913dfd 980 return ret;
f7d0fe02
KW
981}
982
983
984
985
986/*********************************************************/
987/* refcount checking functions */
988
989
990
991/*
992 * Increases the refcount for a range of clusters in a given refcount table.
993 * This is used to construct a temporary refcount table out of L1 and L2 tables
994 * which can be compared the the refcount table saved in the image.
995 *
9ac228e0 996 * Modifies the number of errors in res.
f7d0fe02 997 */
9ac228e0
KW
998static void inc_refcounts(BlockDriverState *bs,
999 BdrvCheckResult *res,
f7d0fe02
KW
1000 uint16_t *refcount_table,
1001 int refcount_table_size,
1002 int64_t offset, int64_t size)
1003{
1004 BDRVQcowState *s = bs->opaque;
1005 int64_t start, last, cluster_offset;
1006 int k;
f7d0fe02
KW
1007
1008 if (size <= 0)
9ac228e0 1009 return;
f7d0fe02
KW
1010
1011 start = offset & ~(s->cluster_size - 1);
1012 last = (offset + size - 1) & ~(s->cluster_size - 1);
1013 for(cluster_offset = start; cluster_offset <= last;
1014 cluster_offset += s->cluster_size) {
1015 k = cluster_offset >> s->cluster_bits;
9ac228e0 1016 if (k < 0) {
f7d0fe02
KW
1017 fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
1018 cluster_offset);
9ac228e0
KW
1019 res->corruptions++;
1020 } else if (k >= refcount_table_size) {
1021 fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after "
1022 "the end of the image file, can't properly check refcounts.\n",
1023 cluster_offset);
1024 res->check_errors++;
f7d0fe02
KW
1025 } else {
1026 if (++refcount_table[k] == 0) {
1027 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
1028 "\n", cluster_offset);
9ac228e0 1029 res->corruptions++;
f7d0fe02
KW
1030 }
1031 }
1032 }
f7d0fe02
KW
1033}
1034
801f7044
SH
1035/* Flags for check_refcounts_l1() and check_refcounts_l2() */
1036enum {
1037 CHECK_OFLAG_COPIED = 0x1, /* check QCOW_OFLAG_COPIED matches refcount */
fba31bae 1038 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
801f7044
SH
1039};
1040
f7d0fe02
KW
1041/*
1042 * Increases the refcount in the given refcount table for the all clusters
1043 * referenced in the L2 table. While doing so, performs some checks on L2
1044 * entries.
1045 *
1046 * Returns the number of errors found by the checks or -errno if an internal
1047 * error occurred.
1048 */
9ac228e0 1049static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
f7d0fe02 1050 uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
801f7044 1051 int flags)
f7d0fe02
KW
1052{
1053 BDRVQcowState *s = bs->opaque;
afdf0abe 1054 uint64_t *l2_table, l2_entry;
fba31bae 1055 uint64_t next_contiguous_offset = 0;
f7d0fe02 1056 int i, l2_size, nb_csectors, refcount;
f7d0fe02
KW
1057
1058 /* Read L2 table from disk */
1059 l2_size = s->l2_size * sizeof(uint64_t);
7267c094 1060 l2_table = g_malloc(l2_size);
f7d0fe02 1061
66f82cee 1062 if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
f7d0fe02
KW
1063 goto fail;
1064
1065 /* Do the actual checks */
1066 for(i = 0; i < s->l2_size; i++) {
afdf0abe
KW
1067 l2_entry = be64_to_cpu(l2_table[i]);
1068
1069 switch (qcow2_get_cluster_type(l2_entry)) {
1070 case QCOW2_CLUSTER_COMPRESSED:
1071 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1072 if (l2_entry & QCOW_OFLAG_COPIED) {
1073 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
1074 "copied flag must never be set for compressed "
1075 "clusters\n", l2_entry >> s->cluster_bits);
1076 l2_entry &= ~QCOW_OFLAG_COPIED;
1077 res->corruptions++;
1078 }
f7d0fe02 1079
afdf0abe
KW
1080 /* Mark cluster as used */
1081 nb_csectors = ((l2_entry >> s->csize_shift) &
1082 s->csize_mask) + 1;
1083 l2_entry &= s->cluster_offset_mask;
1084 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1085 l2_entry & ~511, nb_csectors * 512);
fba31bae
SH
1086
1087 if (flags & CHECK_FRAG_INFO) {
1088 res->bfi.allocated_clusters++;
4db35162 1089 res->bfi.compressed_clusters++;
fba31bae
SH
1090
1091 /* Compressed clusters are fragmented by nature. Since they
1092 * take up sub-sector space but we only have sector granularity
1093 * I/O we need to re-read the same sectors even for adjacent
1094 * compressed clusters.
1095 */
1096 res->bfi.fragmented_clusters++;
1097 }
afdf0abe 1098 break;
f7d0fe02 1099
6377af48
KW
1100 case QCOW2_CLUSTER_ZERO:
1101 if ((l2_entry & L2E_OFFSET_MASK) == 0) {
1102 break;
1103 }
1104 /* fall through */
1105
afdf0abe
KW
1106 case QCOW2_CLUSTER_NORMAL:
1107 {
1108 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
1109 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
f7d0fe02 1110
801f7044 1111 if (flags & CHECK_OFLAG_COPIED) {
afdf0abe
KW
1112 refcount = get_refcount(bs, offset >> s->cluster_bits);
1113 if (refcount < 0) {
1114 fprintf(stderr, "Can't get refcount for offset %"
1115 PRIx64 ": %s\n", l2_entry, strerror(-refcount));
1116 goto fail;
1117 }
1118 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
1119 fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
1120 PRIx64 " refcount=%d\n", l2_entry, refcount);
9ac228e0 1121 res->corruptions++;
f7d0fe02
KW
1122 }
1123 }
afdf0abe 1124
fba31bae
SH
1125 if (flags & CHECK_FRAG_INFO) {
1126 res->bfi.allocated_clusters++;
1127 if (next_contiguous_offset &&
1128 offset != next_contiguous_offset) {
1129 res->bfi.fragmented_clusters++;
1130 }
1131 next_contiguous_offset = offset + s->cluster_size;
1132 }
1133
afdf0abe
KW
1134 /* Mark cluster as used */
1135 inc_refcounts(bs, res, refcount_table,refcount_table_size,
1136 offset, s->cluster_size);
1137
1138 /* Correct offsets are cluster aligned */
1139 if (offset & (s->cluster_size - 1)) {
1140 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
1141 "properly aligned; L2 entry corrupted.\n", offset);
1142 res->corruptions++;
1143 }
1144 break;
1145 }
1146
1147 case QCOW2_CLUSTER_UNALLOCATED:
1148 break;
1149
1150 default:
1151 abort();
f7d0fe02
KW
1152 }
1153 }
1154
7267c094 1155 g_free(l2_table);
9ac228e0 1156 return 0;
f7d0fe02
KW
1157
1158fail:
9ac228e0 1159 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
7267c094 1160 g_free(l2_table);
f7d0fe02
KW
1161 return -EIO;
1162}
1163
1164/*
1165 * Increases the refcount for the L1 table, its L2 tables and all referenced
1166 * clusters in the given refcount table. While doing so, performs some checks
1167 * on L1 and L2 entries.
1168 *
1169 * Returns the number of errors found by the checks or -errno if an internal
1170 * error occurred.
1171 */
1172static int check_refcounts_l1(BlockDriverState *bs,
9ac228e0 1173 BdrvCheckResult *res,
f7d0fe02
KW
1174 uint16_t *refcount_table,
1175 int refcount_table_size,
1176 int64_t l1_table_offset, int l1_size,
801f7044 1177 int flags)
f7d0fe02
KW
1178{
1179 BDRVQcowState *s = bs->opaque;
1180 uint64_t *l1_table, l2_offset, l1_size2;
1181 int i, refcount, ret;
f7d0fe02
KW
1182
1183 l1_size2 = l1_size * sizeof(uint64_t);
1184
1185 /* Mark L1 table as used */
9ac228e0
KW
1186 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1187 l1_table_offset, l1_size2);
f7d0fe02
KW
1188
1189 /* Read L1 table entries from disk */
702ef63f
KW
1190 if (l1_size2 == 0) {
1191 l1_table = NULL;
1192 } else {
7267c094 1193 l1_table = g_malloc(l1_size2);
66f82cee 1194 if (bdrv_pread(bs->file, l1_table_offset,
702ef63f
KW
1195 l1_table, l1_size2) != l1_size2)
1196 goto fail;
1197 for(i = 0;i < l1_size; i++)
1198 be64_to_cpus(&l1_table[i]);
1199 }
f7d0fe02
KW
1200
1201 /* Do the actual checks */
1202 for(i = 0; i < l1_size; i++) {
1203 l2_offset = l1_table[i];
1204 if (l2_offset) {
1205 /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
801f7044 1206 if (flags & CHECK_OFLAG_COPIED) {
f7d0fe02
KW
1207 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
1208 >> s->cluster_bits);
018faafd
KW
1209 if (refcount < 0) {
1210 fprintf(stderr, "Can't get refcount for l2_offset %"
1211 PRIx64 ": %s\n", l2_offset, strerror(-refcount));
9ac228e0 1212 goto fail;
018faafd 1213 }
f7d0fe02
KW
1214 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
1215 fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
1216 " refcount=%d\n", l2_offset, refcount);
9ac228e0 1217 res->corruptions++;
f7d0fe02
KW
1218 }
1219 }
1220
1221 /* Mark L2 table as used */
afdf0abe 1222 l2_offset &= L1E_OFFSET_MASK;
9ac228e0
KW
1223 inc_refcounts(bs, res, refcount_table, refcount_table_size,
1224 l2_offset, s->cluster_size);
f7d0fe02
KW
1225
1226 /* L2 tables are cluster aligned */
1227 if (l2_offset & (s->cluster_size - 1)) {
1228 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1229 "cluster aligned; L1 entry corrupted\n", l2_offset);
9ac228e0 1230 res->corruptions++;
f7d0fe02
KW
1231 }
1232
1233 /* Process and check L2 entries */
9ac228e0 1234 ret = check_refcounts_l2(bs, res, refcount_table,
801f7044 1235 refcount_table_size, l2_offset, flags);
f7d0fe02
KW
1236 if (ret < 0) {
1237 goto fail;
1238 }
f7d0fe02
KW
1239 }
1240 }
7267c094 1241 g_free(l1_table);
9ac228e0 1242 return 0;
f7d0fe02
KW
1243
1244fail:
1245 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
9ac228e0 1246 res->check_errors++;
7267c094 1247 g_free(l1_table);
f7d0fe02
KW
1248 return -EIO;
1249}
1250
1251/*
1252 * Checks an image for refcount consistency.
1253 *
1254 * Returns 0 if no errors are found, the number of errors in case the image is
a1c7273b 1255 * detected as corrupted, and -errno when an internal error occurred.
f7d0fe02 1256 */
166acf54
KW
1257int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1258 BdrvCheckMode fix)
f7d0fe02
KW
1259{
1260 BDRVQcowState *s = bs->opaque;
c6bb9ad1 1261 int64_t size, i, highest_cluster;
166acf54 1262 int nb_clusters, refcount1, refcount2;
f7d0fe02
KW
1263 QCowSnapshot *sn;
1264 uint16_t *refcount_table;
9ac228e0 1265 int ret;
f7d0fe02 1266
66f82cee 1267 size = bdrv_getlength(bs->file);
f7d0fe02 1268 nb_clusters = size_to_clusters(s, size);
7267c094 1269 refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t));
f7d0fe02 1270
c349ca4b
KW
1271 res->bfi.total_clusters =
1272 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE);
1273
f7d0fe02 1274 /* header */
9ac228e0
KW
1275 inc_refcounts(bs, res, refcount_table, nb_clusters,
1276 0, s->cluster_size);
f7d0fe02
KW
1277
1278 /* current L1 table */
9ac228e0 1279 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
801f7044 1280 s->l1_table_offset, s->l1_size,
fba31bae 1281 CHECK_OFLAG_COPIED | CHECK_FRAG_INFO);
f7d0fe02 1282 if (ret < 0) {
80fa3341 1283 goto fail;
f7d0fe02 1284 }
f7d0fe02
KW
1285
1286 /* snapshots */
1287 for(i = 0; i < s->nb_snapshots; i++) {
1288 sn = s->snapshots + i;
9ac228e0
KW
1289 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1290 sn->l1_table_offset, sn->l1_size, 0);
1291 if (ret < 0) {
80fa3341 1292 goto fail;
9ac228e0 1293 }
f7d0fe02 1294 }
9ac228e0
KW
1295 inc_refcounts(bs, res, refcount_table, nb_clusters,
1296 s->snapshots_offset, s->snapshots_size);
f7d0fe02
KW
1297
1298 /* refcount data */
9ac228e0
KW
1299 inc_refcounts(bs, res, refcount_table, nb_clusters,
1300 s->refcount_table_offset,
1301 s->refcount_table_size * sizeof(uint64_t));
1302
f7d0fe02 1303 for(i = 0; i < s->refcount_table_size; i++) {
6882c8fa 1304 uint64_t offset, cluster;
f7d0fe02 1305 offset = s->refcount_table[i];
6882c8fa 1306 cluster = offset >> s->cluster_bits;
746c3cb5
KW
1307
1308 /* Refcount blocks are cluster aligned */
1309 if (offset & (s->cluster_size - 1)) {
166acf54 1310 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
746c3cb5 1311 "cluster aligned; refcount table entry corrupted\n", i);
9ac228e0 1312 res->corruptions++;
6882c8fa
KW
1313 continue;
1314 }
1315
1316 if (cluster >= nb_clusters) {
166acf54
KW
1317 fprintf(stderr, "ERROR refcount block %" PRId64
1318 " is outside image\n", i);
9ac228e0 1319 res->corruptions++;
6882c8fa 1320 continue;
746c3cb5
KW
1321 }
1322
f7d0fe02 1323 if (offset != 0) {
9ac228e0
KW
1324 inc_refcounts(bs, res, refcount_table, nb_clusters,
1325 offset, s->cluster_size);
6882c8fa 1326 if (refcount_table[cluster] != 1) {
166acf54
KW
1327 fprintf(stderr, "ERROR refcount block %" PRId64
1328 " refcount=%d\n",
6882c8fa 1329 i, refcount_table[cluster]);
9ac228e0 1330 res->corruptions++;
746c3cb5 1331 }
f7d0fe02
KW
1332 }
1333 }
1334
1335 /* compare ref counts */
c6bb9ad1 1336 for (i = 0, highest_cluster = 0; i < nb_clusters; i++) {
f7d0fe02 1337 refcount1 = get_refcount(bs, i);
018faafd 1338 if (refcount1 < 0) {
166acf54 1339 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
018faafd 1340 i, strerror(-refcount1));
9ac228e0 1341 res->check_errors++;
f74550fd 1342 continue;
018faafd
KW
1343 }
1344
f7d0fe02 1345 refcount2 = refcount_table[i];
c6bb9ad1
FS
1346
1347 if (refcount1 > 0 || refcount2 > 0) {
1348 highest_cluster = i;
1349 }
1350
f7d0fe02 1351 if (refcount1 != refcount2) {
166acf54
KW
1352
1353 /* Check if we're allowed to fix the mismatch */
1354 int *num_fixed = NULL;
1355 if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
1356 num_fixed = &res->leaks_fixed;
1357 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
1358 num_fixed = &res->corruptions_fixed;
1359 }
1360
1361 fprintf(stderr, "%s cluster %" PRId64 " refcount=%d reference=%d\n",
1362 num_fixed != NULL ? "Repairing" :
1363 refcount1 < refcount2 ? "ERROR" :
1364 "Leaked",
f7d0fe02 1365 i, refcount1, refcount2);
166acf54
KW
1366
1367 if (num_fixed) {
1368 ret = update_refcount(bs, i << s->cluster_bits, 1,
6cfcb9b8
KW
1369 refcount2 - refcount1,
1370 QCOW2_DISCARD_ALWAYS);
166acf54
KW
1371 if (ret >= 0) {
1372 (*num_fixed)++;
1373 continue;
1374 }
1375 }
1376
1377 /* And if we couldn't, print an error */
9ac228e0
KW
1378 if (refcount1 < refcount2) {
1379 res->corruptions++;
1380 } else {
1381 res->leaks++;
1382 }
f7d0fe02
KW
1383 }
1384 }
1385
c6bb9ad1 1386 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
80fa3341
KW
1387 ret = 0;
1388
1389fail:
7267c094 1390 g_free(refcount_table);
f7d0fe02 1391
80fa3341 1392 return ret;
f7d0fe02
KW
1393}
1394
a40f1c2a
HR
1395#define overlaps_with(ofs, sz) \
1396 ranges_overlap(offset, size, ofs, sz)
1397
1398/*
1399 * Checks if the given offset into the image file is actually free to use by
1400 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
1401 * i.e. a sanity check without relying on the refcount tables.
1402 *
1403 * The chk parameter specifies exactly what checks to perform (being a bitmask
1404 * of QCow2MetadataOverlap values).
1405 *
1406 * Returns:
1407 * - 0 if writing to this offset will not affect the mentioned metadata
1408 * - a positive QCow2MetadataOverlap value indicating one overlapping section
1409 * - a negative value (-errno) indicating an error while performing a check,
1410 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
1411 */
1412int qcow2_check_metadata_overlap(BlockDriverState *bs, int chk, int64_t offset,
1413 int64_t size)
1414{
1415 BDRVQcowState *s = bs->opaque;
1416 int i, j;
1417
1418 if (!size) {
1419 return 0;
1420 }
1421
1422 if (chk & QCOW2_OL_MAIN_HEADER) {
1423 if (offset < s->cluster_size) {
1424 return QCOW2_OL_MAIN_HEADER;
1425 }
1426 }
1427
1428 /* align range to test to cluster boundaries */
1429 size = align_offset(offset_into_cluster(s, offset) + size, s->cluster_size);
1430 offset = start_of_cluster(s, offset);
1431
1432 if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) {
1433 if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) {
1434 return QCOW2_OL_ACTIVE_L1;
1435 }
1436 }
1437
1438 if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) {
1439 if (overlaps_with(s->refcount_table_offset,
1440 s->refcount_table_size * sizeof(uint64_t))) {
1441 return QCOW2_OL_REFCOUNT_TABLE;
1442 }
1443 }
1444
1445 if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) {
1446 if (overlaps_with(s->snapshots_offset, s->snapshots_size)) {
1447 return QCOW2_OL_SNAPSHOT_TABLE;
1448 }
1449 }
1450
1451 if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) {
1452 for (i = 0; i < s->nb_snapshots; i++) {
1453 if (s->snapshots[i].l1_size &&
1454 overlaps_with(s->snapshots[i].l1_table_offset,
1455 s->snapshots[i].l1_size * sizeof(uint64_t))) {
1456 return QCOW2_OL_INACTIVE_L1;
1457 }
1458 }
1459 }
1460
1461 if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) {
1462 for (i = 0; i < s->l1_size; i++) {
1463 if ((s->l1_table[i] & L1E_OFFSET_MASK) &&
1464 overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK,
1465 s->cluster_size)) {
1466 return QCOW2_OL_ACTIVE_L2;
1467 }
1468 }
1469 }
1470
1471 if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) {
1472 for (i = 0; i < s->refcount_table_size; i++) {
1473 if ((s->refcount_table[i] & REFT_OFFSET_MASK) &&
1474 overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK,
1475 s->cluster_size)) {
1476 return QCOW2_OL_REFCOUNT_BLOCK;
1477 }
1478 }
1479 }
1480
1481 if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) {
1482 for (i = 0; i < s->nb_snapshots; i++) {
1483 uint64_t l1_ofs = s->snapshots[i].l1_table_offset;
1484 uint32_t l1_sz = s->snapshots[i].l1_size;
1485 uint64_t *l1 = g_malloc(l1_sz * sizeof(uint64_t));
1486 int ret;
1487
1488 ret = bdrv_read(bs->file, l1_ofs / BDRV_SECTOR_SIZE, (uint8_t *)l1,
1489 l1_sz * sizeof(uint64_t) / BDRV_SECTOR_SIZE);
1490
1491 if (ret < 0) {
1492 g_free(l1);
1493 return ret;
1494 }
1495
1496 for (j = 0; j < l1_sz; j++) {
1497 if ((l1[j] & L1E_OFFSET_MASK) &&
1498 overlaps_with(l1[j] & L1E_OFFSET_MASK, s->cluster_size)) {
1499 g_free(l1);
1500 return QCOW2_OL_INACTIVE_L2;
1501 }
1502 }
1503
1504 g_free(l1);
1505 }
1506 }
1507
1508 return 0;
1509}
1510
1511static const char *metadata_ol_names[] = {
1512 [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header",
1513 [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table",
1514 [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table",
1515 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table",
1516 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block",
1517 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table",
1518 [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table",
1519 [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table",
1520};
1521
1522/*
1523 * First performs a check for metadata overlaps (through
1524 * qcow2_check_metadata_overlap); if that fails with a negative value (error
1525 * while performing a check), that value is returned. If an impending overlap
1526 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
1527 * and -EIO returned.
1528 *
1529 * Returns 0 if there were neither overlaps nor errors while checking for
1530 * overlaps; or a negative value (-errno) on error.
1531 */
1532int qcow2_pre_write_overlap_check(BlockDriverState *bs, int chk, int64_t offset,
1533 int64_t size)
1534{
1535 int ret = qcow2_check_metadata_overlap(bs, chk, offset, size);
1536
1537 if (ret < 0) {
1538 return ret;
1539 } else if (ret > 0) {
1540 int metadata_ol_bitnr = ffs(ret) - 1;
1541 char *message;
1542 QObject *data;
1543
1544 assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR);
1545
1546 fprintf(stderr, "qcow2: Preventing invalid write on metadata (overlaps "
1547 "with %s); image marked as corrupt.\n",
1548 metadata_ol_names[metadata_ol_bitnr]);
1549 message = g_strdup_printf("Prevented %s overwrite",
1550 metadata_ol_names[metadata_ol_bitnr]);
1551 data = qobject_from_jsonf("{ 'device': %s, 'msg': %s, 'offset': %"
1552 PRId64 ", 'size': %" PRId64 " }", bs->device_name, message,
1553 offset, size);
1554 monitor_protocol_event(QEVENT_BLOCK_IMAGE_CORRUPTED, data);
1555 g_free(message);
1556 qobject_decref(data);
1557
1558 qcow2_mark_corrupt(bs);
1559 bs->drv = NULL; /* make BDS unusable */
1560 return -EIO;
1561 }
1562
1563 return 0;
1564}