]> Git Repo - qemu.git/blame - block/qcow2-refcount.c
qcow2: Name typedef for cluster type
[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
80c71a24 25#include "qemu/osdep.h"
da34e65c 26#include "qapi/error.h"
f7d0fe02 27#include "qemu-common.h"
737e150e 28#include "block/block_int.h"
f7d0fe02 29#include "block/qcow2.h"
a40f1c2a 30#include "qemu/range.h"
58369e22 31#include "qemu/bswap.h"
f7d0fe02 32
bb572aef 33static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size);
92dcb59f 34static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
0e06528e 35 int64_t offset, int64_t length, uint64_t addend,
2aabe7c7 36 bool decrease, enum qcow2_discard_type type);
f7d0fe02 37
59c0cb78
HR
38static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index);
39static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index);
40static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index);
41static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index);
7453c96b 42static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index);
59c0cb78
HR
43static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index);
44static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index);
7453c96b 45
59c0cb78
HR
46static void set_refcount_ro0(void *refcount_array, uint64_t index,
47 uint64_t value);
48static void set_refcount_ro1(void *refcount_array, uint64_t index,
49 uint64_t value);
50static void set_refcount_ro2(void *refcount_array, uint64_t index,
51 uint64_t value);
52static void set_refcount_ro3(void *refcount_array, uint64_t index,
53 uint64_t value);
7453c96b
HR
54static void set_refcount_ro4(void *refcount_array, uint64_t index,
55 uint64_t value);
59c0cb78
HR
56static void set_refcount_ro5(void *refcount_array, uint64_t index,
57 uint64_t value);
58static void set_refcount_ro6(void *refcount_array, uint64_t index,
59 uint64_t value);
60
61
62static Qcow2GetRefcountFunc *const get_refcount_funcs[] = {
63 &get_refcount_ro0,
64 &get_refcount_ro1,
65 &get_refcount_ro2,
66 &get_refcount_ro3,
67 &get_refcount_ro4,
68 &get_refcount_ro5,
69 &get_refcount_ro6
70};
71
72static Qcow2SetRefcountFunc *const set_refcount_funcs[] = {
73 &set_refcount_ro0,
74 &set_refcount_ro1,
75 &set_refcount_ro2,
76 &set_refcount_ro3,
77 &set_refcount_ro4,
78 &set_refcount_ro5,
79 &set_refcount_ro6
80};
7453c96b 81
3b88e52b 82
f7d0fe02
KW
83/*********************************************************/
84/* refcount handling */
85
7061a078
AG
86static void update_max_refcount_table_index(BDRVQcow2State *s)
87{
88 unsigned i = s->refcount_table_size - 1;
89 while (i > 0 && (s->refcount_table[i] & REFT_OFFSET_MASK) == 0) {
90 i--;
91 }
92 /* Set s->max_refcount_table_index to the index of the last used entry */
93 s->max_refcount_table_index = i;
94}
95
ed6ccf0f 96int qcow2_refcount_init(BlockDriverState *bs)
f7d0fe02 97{
ff99129a 98 BDRVQcow2State *s = bs->opaque;
5dab2fad
KW
99 unsigned int refcount_table_size2, i;
100 int ret;
f7d0fe02 101
59c0cb78
HR
102 assert(s->refcount_order >= 0 && s->refcount_order <= 6);
103
104 s->get_refcount = get_refcount_funcs[s->refcount_order];
105 s->set_refcount = set_refcount_funcs[s->refcount_order];
7453c96b 106
5dab2fad 107 assert(s->refcount_table_size <= INT_MAX / sizeof(uint64_t));
f7d0fe02 108 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
de82815d
KW
109 s->refcount_table = g_try_malloc(refcount_table_size2);
110
f7d0fe02 111 if (s->refcount_table_size > 0) {
de82815d 112 if (s->refcount_table == NULL) {
8fcffa98 113 ret = -ENOMEM;
de82815d
KW
114 goto fail;
115 }
66f82cee 116 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
cf2ab8fc 117 ret = bdrv_pread(bs->file, s->refcount_table_offset,
f7d0fe02 118 s->refcount_table, refcount_table_size2);
8fcffa98 119 if (ret < 0) {
f7d0fe02 120 goto fail;
8fcffa98 121 }
f7d0fe02
KW
122 for(i = 0; i < s->refcount_table_size; i++)
123 be64_to_cpus(&s->refcount_table[i]);
7061a078 124 update_max_refcount_table_index(s);
f7d0fe02
KW
125 }
126 return 0;
127 fail:
8fcffa98 128 return ret;
f7d0fe02
KW
129}
130
ed6ccf0f 131void qcow2_refcount_close(BlockDriverState *bs)
f7d0fe02 132{
ff99129a 133 BDRVQcow2State *s = bs->opaque;
7267c094 134 g_free(s->refcount_table);
f7d0fe02
KW
135}
136
137
59c0cb78
HR
138static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index)
139{
140 return (((const uint8_t *)refcount_array)[index / 8] >> (index % 8)) & 0x1;
141}
142
143static void set_refcount_ro0(void *refcount_array, uint64_t index,
144 uint64_t value)
145{
146 assert(!(value >> 1));
147 ((uint8_t *)refcount_array)[index / 8] &= ~(0x1 << (index % 8));
148 ((uint8_t *)refcount_array)[index / 8] |= value << (index % 8);
149}
150
151static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index)
152{
153 return (((const uint8_t *)refcount_array)[index / 4] >> (2 * (index % 4)))
154 & 0x3;
155}
156
157static void set_refcount_ro1(void *refcount_array, uint64_t index,
158 uint64_t value)
159{
160 assert(!(value >> 2));
161 ((uint8_t *)refcount_array)[index / 4] &= ~(0x3 << (2 * (index % 4)));
162 ((uint8_t *)refcount_array)[index / 4] |= value << (2 * (index % 4));
163}
164
165static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index)
166{
167 return (((const uint8_t *)refcount_array)[index / 2] >> (4 * (index % 2)))
168 & 0xf;
169}
170
171static void set_refcount_ro2(void *refcount_array, uint64_t index,
172 uint64_t value)
173{
174 assert(!(value >> 4));
175 ((uint8_t *)refcount_array)[index / 2] &= ~(0xf << (4 * (index % 2)));
176 ((uint8_t *)refcount_array)[index / 2] |= value << (4 * (index % 2));
177}
178
179static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index)
180{
181 return ((const uint8_t *)refcount_array)[index];
182}
183
184static void set_refcount_ro3(void *refcount_array, uint64_t index,
185 uint64_t value)
186{
187 assert(!(value >> 8));
188 ((uint8_t *)refcount_array)[index] = value;
189}
190
7453c96b
HR
191static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index)
192{
193 return be16_to_cpu(((const uint16_t *)refcount_array)[index]);
194}
195
196static void set_refcount_ro4(void *refcount_array, uint64_t index,
197 uint64_t value)
198{
199 assert(!(value >> 16));
200 ((uint16_t *)refcount_array)[index] = cpu_to_be16(value);
201}
202
59c0cb78
HR
203static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index)
204{
205 return be32_to_cpu(((const uint32_t *)refcount_array)[index]);
206}
207
208static void set_refcount_ro5(void *refcount_array, uint64_t index,
209 uint64_t value)
210{
211 assert(!(value >> 32));
212 ((uint32_t *)refcount_array)[index] = cpu_to_be32(value);
213}
214
215static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index)
216{
217 return be64_to_cpu(((const uint64_t *)refcount_array)[index]);
218}
219
220static void set_refcount_ro6(void *refcount_array, uint64_t index,
221 uint64_t value)
222{
223 ((uint64_t *)refcount_array)[index] = cpu_to_be64(value);
224}
225
7453c96b 226
f7d0fe02 227static int load_refcount_block(BlockDriverState *bs,
29c1a730
KW
228 int64_t refcount_block_offset,
229 void **refcount_block)
f7d0fe02 230{
ff99129a 231 BDRVQcow2State *s = bs->opaque;
3b88e52b 232
66f82cee 233 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
9be38598
EH
234 return qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
235 refcount_block);
f7d0fe02
KW
236}
237
018faafd 238/*
7324c10f
HR
239 * Retrieves the refcount of the cluster given by its index and stores it in
240 * *refcount. Returns 0 on success and -errno on failure.
018faafd 241 */
7324c10f 242int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index,
0e06528e 243 uint64_t *refcount)
f7d0fe02 244{
ff99129a 245 BDRVQcow2State *s = bs->opaque;
db8a31d1 246 uint64_t refcount_table_index, block_index;
f7d0fe02 247 int64_t refcount_block_offset;
018faafd 248 int ret;
7453c96b 249 void *refcount_block;
f7d0fe02 250
17bd5f47 251 refcount_table_index = cluster_index >> s->refcount_block_bits;
7324c10f
HR
252 if (refcount_table_index >= s->refcount_table_size) {
253 *refcount = 0;
f7d0fe02 254 return 0;
7324c10f 255 }
26d49c46
HR
256 refcount_block_offset =
257 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
7324c10f
HR
258 if (!refcount_block_offset) {
259 *refcount = 0;
f7d0fe02 260 return 0;
7324c10f 261 }
29c1a730 262
a97c67ee
HR
263 if (offset_into_cluster(s, refcount_block_offset)) {
264 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" PRIx64
265 " unaligned (reftable index: %#" PRIx64 ")",
266 refcount_block_offset, refcount_table_index);
267 return -EIO;
268 }
269
29c1a730 270 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
7453c96b 271 &refcount_block);
29c1a730
KW
272 if (ret < 0) {
273 return ret;
f7d0fe02 274 }
29c1a730 275
17bd5f47 276 block_index = cluster_index & (s->refcount_block_size - 1);
7453c96b 277 *refcount = s->get_refcount(refcount_block, block_index);
29c1a730 278
a3f1afb4 279 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
29c1a730 280
7324c10f 281 return 0;
f7d0fe02
KW
282}
283
05121aed
KW
284/*
285 * Rounds the refcount table size up to avoid growing the table for each single
286 * refcount block that is allocated.
287 */
ff99129a 288static unsigned int next_refcount_table_size(BDRVQcow2State *s,
05121aed
KW
289 unsigned int min_size)
290{
291 unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1;
292 unsigned int refcount_table_clusters =
293 MAX(1, s->refcount_table_size >> (s->cluster_bits - 3));
294
295 while (min_clusters > refcount_table_clusters) {
296 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
297 }
298
299 return refcount_table_clusters << (s->cluster_bits - 3);
300}
301
92dcb59f
KW
302
303/* Checks if two offsets are described by the same refcount block */
ff99129a 304static int in_same_refcount_block(BDRVQcow2State *s, uint64_t offset_a,
92dcb59f
KW
305 uint64_t offset_b)
306{
17bd5f47
HR
307 uint64_t block_a = offset_a >> (s->cluster_bits + s->refcount_block_bits);
308 uint64_t block_b = offset_b >> (s->cluster_bits + s->refcount_block_bits);
92dcb59f
KW
309
310 return (block_a == block_b);
311}
312
313/*
314 * Loads a refcount block. If it doesn't exist yet, it is allocated first
315 * (including growing the refcount table if needed).
316 *
29c1a730 317 * Returns 0 on success or -errno in error case
92dcb59f 318 */
29c1a730 319static int alloc_refcount_block(BlockDriverState *bs,
7453c96b 320 int64_t cluster_index, void **refcount_block)
f7d0fe02 321{
ff99129a 322 BDRVQcow2State *s = bs->opaque;
92dcb59f
KW
323 unsigned int refcount_table_index;
324 int ret;
325
66f82cee 326 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
8252278a 327
92dcb59f 328 /* Find the refcount block for the given cluster */
17bd5f47 329 refcount_table_index = cluster_index >> s->refcount_block_bits;
92dcb59f
KW
330
331 if (refcount_table_index < s->refcount_table_size) {
332
333 uint64_t refcount_block_offset =
76dc9e0c 334 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
92dcb59f
KW
335
336 /* If it's already there, we're done */
337 if (refcount_block_offset) {
a97c67ee
HR
338 if (offset_into_cluster(s, refcount_block_offset)) {
339 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
340 PRIx64 " unaligned (reftable index: "
341 "%#x)", refcount_block_offset,
342 refcount_table_index);
343 return -EIO;
344 }
345
29c1a730 346 return load_refcount_block(bs, refcount_block_offset,
7453c96b 347 refcount_block);
92dcb59f
KW
348 }
349 }
350
351 /*
352 * If we came here, we need to allocate something. Something is at least
353 * a cluster for the new refcount block. It may also include a new refcount
354 * table if the old refcount table is too small.
355 *
356 * Note that allocating clusters here needs some special care:
357 *
358 * - We can't use the normal qcow2_alloc_clusters(), it would try to
359 * increase the refcount and very likely we would end up with an endless
360 * recursion. Instead we must place the refcount blocks in a way that
361 * they can describe them themselves.
362 *
363 * - We need to consider that at this point we are inside update_refcounts
b106ad91
KW
364 * and potentially doing an initial refcount increase. This means that
365 * some clusters have already been allocated by the caller, but their
366 * refcount isn't accurate yet. If we allocate clusters for metadata, we
367 * need to return -EAGAIN to signal the caller that it needs to restart
368 * the search for free clusters.
92dcb59f
KW
369 *
370 * - alloc_clusters_noref and qcow2_free_clusters may load a different
371 * refcount block into the cache
372 */
373
29c1a730
KW
374 *refcount_block = NULL;
375
376 /* We write to the refcount table, so we might depend on L2 tables */
9991923b
SH
377 ret = qcow2_cache_flush(bs, s->l2_table_cache);
378 if (ret < 0) {
379 return ret;
380 }
92dcb59f
KW
381
382 /* Allocate the refcount block itself and mark it as used */
2eaa8f63
KW
383 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
384 if (new_block < 0) {
385 return new_block;
386 }
f7d0fe02 387
f7d0fe02 388#ifdef DEBUG_ALLOC2
92dcb59f
KW
389 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
390 " at %" PRIx64 "\n",
391 refcount_table_index, cluster_index << s->cluster_bits, new_block);
f7d0fe02 392#endif
92dcb59f
KW
393
394 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
25408c09 395 /* Zero the new refcount block before updating it */
29c1a730 396 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
7453c96b 397 refcount_block);
29c1a730
KW
398 if (ret < 0) {
399 goto fail_block;
400 }
401
402 memset(*refcount_block, 0, s->cluster_size);
25408c09 403
92dcb59f
KW
404 /* The block describes itself, need to update the cache */
405 int block_index = (new_block >> s->cluster_bits) &
17bd5f47 406 (s->refcount_block_size - 1);
7453c96b 407 s->set_refcount(*refcount_block, block_index, 1);
92dcb59f
KW
408 } else {
409 /* Described somewhere else. This can recurse at most twice before we
410 * arrive at a block that describes itself. */
2aabe7c7 411 ret = update_refcount(bs, new_block, s->cluster_size, 1, false,
6cfcb9b8 412 QCOW2_DISCARD_NEVER);
92dcb59f
KW
413 if (ret < 0) {
414 goto fail_block;
415 }
25408c09 416
9991923b
SH
417 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
418 if (ret < 0) {
419 goto fail_block;
420 }
1c4c2814 421
25408c09
KW
422 /* Initialize the new refcount block only after updating its refcount,
423 * update_refcount uses the refcount cache itself */
29c1a730 424 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
7453c96b 425 refcount_block);
29c1a730
KW
426 if (ret < 0) {
427 goto fail_block;
428 }
429
430 memset(*refcount_block, 0, s->cluster_size);
92dcb59f
KW
431 }
432
433 /* Now the new refcount block needs to be written to disk */
66f82cee 434 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
72e80b89 435 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache, *refcount_block);
29c1a730 436 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
92dcb59f
KW
437 if (ret < 0) {
438 goto fail_block;
439 }
440
441 /* If the refcount table is big enough, just hook the block up there */
442 if (refcount_table_index < s->refcount_table_size) {
443 uint64_t data64 = cpu_to_be64(new_block);
66f82cee 444 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
d9ca2ea2 445 ret = bdrv_pwrite_sync(bs->file,
92dcb59f
KW
446 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
447 &data64, sizeof(data64));
448 if (ret < 0) {
449 goto fail_block;
450 }
451
452 s->refcount_table[refcount_table_index] = new_block;
7061a078
AG
453 /* If there's a hole in s->refcount_table then it can happen
454 * that refcount_table_index < s->max_refcount_table_index */
455 s->max_refcount_table_index =
456 MAX(s->max_refcount_table_index, refcount_table_index);
b106ad91
KW
457
458 /* The new refcount block may be where the caller intended to put its
459 * data, so let it restart the search. */
460 return -EAGAIN;
29c1a730
KW
461 }
462
a3f1afb4 463 qcow2_cache_put(bs, s->refcount_block_cache, refcount_block);
92dcb59f
KW
464
465 /*
466 * If we come here, we need to grow the refcount table. Again, a new
467 * refcount table needs some space and we can't simply allocate to avoid
468 * endless recursion.
469 *
470 * Therefore let's grab new refcount blocks at the end of the image, which
471 * will describe themselves and the new refcount table. This way we can
472 * reference them only in the new table and do the switch to the new
473 * refcount table at once without producing an inconsistent state in
474 * between.
475 */
66f82cee 476 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
8252278a 477
14a58a4e
HR
478 /* Calculate the number of refcount blocks needed so far; this will be the
479 * basis for calculating the index of the first cluster used for the
480 * self-describing refcount structures which we are about to create.
481 *
482 * Because we reached this point, there cannot be any refcount entries for
483 * cluster_index or higher indices yet. However, because new_block has been
484 * allocated to describe that cluster (and it will assume this role later
485 * on), we cannot use that index; also, new_block may actually have a higher
486 * cluster index than cluster_index, so it needs to be taken into account
487 * here (and 1 needs to be added to its value because that cluster is used).
488 */
489 uint64_t blocks_used = DIV_ROUND_UP(MAX(cluster_index + 1,
490 (new_block >> s->cluster_bits) + 1),
491 s->refcount_block_size);
92dcb59f 492
2b5d5953
KW
493 if (blocks_used > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) {
494 return -EFBIG;
495 }
496
92dcb59f
KW
497 /* And now we need at least one block more for the new metadata */
498 uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
499 uint64_t last_table_size;
500 uint64_t blocks_clusters;
501 do {
a3548077
KW
502 uint64_t table_clusters =
503 size_to_clusters(s, table_size * sizeof(uint64_t));
92dcb59f 504 blocks_clusters = 1 +
d737b78c 505 DIV_ROUND_UP(table_clusters, s->refcount_block_size);
92dcb59f
KW
506 uint64_t meta_clusters = table_clusters + blocks_clusters;
507
508 last_table_size = table_size;
509 table_size = next_refcount_table_size(s, blocks_used +
d737b78c 510 DIV_ROUND_UP(meta_clusters, s->refcount_block_size));
92dcb59f
KW
511
512 } while (last_table_size != table_size);
513
514#ifdef DEBUG_ALLOC2
515 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
516 s->refcount_table_size, table_size);
517#endif
518
519 /* Create the new refcount table and blocks */
17bd5f47 520 uint64_t meta_offset = (blocks_used * s->refcount_block_size) *
92dcb59f
KW
521 s->cluster_size;
522 uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
5839e53b 523 uint64_t *new_table = g_try_new0(uint64_t, table_size);
7453c96b 524 void *new_blocks = g_try_malloc0(blocks_clusters * s->cluster_size);
de82815d
KW
525
526 assert(table_size > 0 && blocks_clusters > 0);
527 if (new_table == NULL || new_blocks == NULL) {
528 ret = -ENOMEM;
529 goto fail_table;
530 }
92dcb59f 531
92dcb59f 532 /* Fill the new refcount table */
f7d0fe02 533 memcpy(new_table, s->refcount_table,
92dcb59f
KW
534 s->refcount_table_size * sizeof(uint64_t));
535 new_table[refcount_table_index] = new_block;
536
537 int i;
538 for (i = 0; i < blocks_clusters; i++) {
539 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
540 }
541
542 /* Fill the refcount blocks */
543 uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
544 int block = 0;
545 for (i = 0; i < table_clusters + blocks_clusters; i++) {
7453c96b 546 s->set_refcount(new_blocks, block++, 1);
92dcb59f
KW
547 }
548
549 /* Write refcount blocks to disk */
66f82cee 550 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
d9ca2ea2 551 ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
92dcb59f 552 blocks_clusters * s->cluster_size);
7267c094 553 g_free(new_blocks);
39ba3bf6 554 new_blocks = NULL;
92dcb59f
KW
555 if (ret < 0) {
556 goto fail_table;
557 }
558
559 /* Write refcount table to disk */
560 for(i = 0; i < table_size; i++) {
561 cpu_to_be64s(&new_table[i]);
562 }
563
66f82cee 564 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
d9ca2ea2 565 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
92dcb59f
KW
566 table_size * sizeof(uint64_t));
567 if (ret < 0) {
568 goto fail_table;
569 }
570
571 for(i = 0; i < table_size; i++) {
87267753 572 be64_to_cpus(&new_table[i]);
92dcb59f 573 }
f7d0fe02 574
92dcb59f 575 /* Hook up the new refcount table in the qcow2 header */
95334230
JS
576 struct QEMU_PACKED {
577 uint64_t d64;
578 uint32_t d32;
579 } data;
f1f7a1dd
PM
580 data.d64 = cpu_to_be64(table_offset);
581 data.d32 = cpu_to_be32(table_clusters);
66f82cee 582 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
d9ca2ea2 583 ret = bdrv_pwrite_sync(bs->file,
9a4f4c31 584 offsetof(QCowHeader, refcount_table_offset),
95334230 585 &data, sizeof(data));
92dcb59f
KW
586 if (ret < 0) {
587 goto fail_table;
f2b7c8b3
KW
588 }
589
92dcb59f
KW
590 /* And switch it in memory */
591 uint64_t old_table_offset = s->refcount_table_offset;
592 uint64_t old_table_size = s->refcount_table_size;
593
7267c094 594 g_free(s->refcount_table);
f7d0fe02 595 s->refcount_table = new_table;
92dcb59f 596 s->refcount_table_size = table_size;
f7d0fe02 597 s->refcount_table_offset = table_offset;
7061a078 598 update_max_refcount_table_index(s);
f7d0fe02 599
b106ad91 600 /* Free old table. */
6cfcb9b8
KW
601 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t),
602 QCOW2_DISCARD_OTHER);
f7d0fe02 603
7453c96b 604 ret = load_refcount_block(bs, new_block, refcount_block);
92dcb59f 605 if (ret < 0) {
29c1a730 606 return ret;
f7d0fe02
KW
607 }
608
b106ad91
KW
609 /* If we were trying to do the initial refcount update for some cluster
610 * allocation, we might have used the same clusters to store newly
611 * allocated metadata. Make the caller search some new space. */
612 return -EAGAIN;
f7d0fe02 613
92dcb59f 614fail_table:
de82815d 615 g_free(new_blocks);
7267c094 616 g_free(new_table);
92dcb59f 617fail_block:
29c1a730 618 if (*refcount_block != NULL) {
7453c96b 619 qcow2_cache_put(bs, s->refcount_block_cache, refcount_block);
3b88e52b 620 }
29c1a730 621 return ret;
9923e05e
KW
622}
623
0b919fae
KW
624void qcow2_process_discards(BlockDriverState *bs, int ret)
625{
ff99129a 626 BDRVQcow2State *s = bs->opaque;
0b919fae
KW
627 Qcow2DiscardRegion *d, *next;
628
629 QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) {
630 QTAILQ_REMOVE(&s->discards, d, next);
631
632 /* Discard is optional, ignore the return value */
633 if (ret >= 0) {
0c51a893 634 bdrv_pdiscard(bs->file->bs, d->offset, d->bytes);
0b919fae
KW
635 }
636
637 g_free(d);
638 }
639}
640
641static void update_refcount_discard(BlockDriverState *bs,
642 uint64_t offset, uint64_t length)
643{
ff99129a 644 BDRVQcow2State *s = bs->opaque;
0b919fae
KW
645 Qcow2DiscardRegion *d, *p, *next;
646
647 QTAILQ_FOREACH(d, &s->discards, next) {
648 uint64_t new_start = MIN(offset, d->offset);
649 uint64_t new_end = MAX(offset + length, d->offset + d->bytes);
650
651 if (new_end - new_start <= length + d->bytes) {
652 /* There can't be any overlap, areas ending up here have no
653 * references any more and therefore shouldn't get freed another
654 * time. */
655 assert(d->bytes + length == new_end - new_start);
656 d->offset = new_start;
657 d->bytes = new_end - new_start;
658 goto found;
659 }
660 }
661
662 d = g_malloc(sizeof(*d));
663 *d = (Qcow2DiscardRegion) {
664 .bs = bs,
665 .offset = offset,
666 .bytes = length,
667 };
668 QTAILQ_INSERT_TAIL(&s->discards, d, next);
669
670found:
671 /* Merge discard requests if they are adjacent now */
672 QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) {
673 if (p == d
674 || p->offset > d->offset + d->bytes
675 || d->offset > p->offset + p->bytes)
676 {
677 continue;
678 }
679
680 /* Still no overlap possible */
681 assert(p->offset == d->offset + d->bytes
682 || d->offset == p->offset + p->bytes);
683
684 QTAILQ_REMOVE(&s->discards, p, next);
685 d->offset = MIN(d->offset, p->offset);
686 d->bytes += p->bytes;
d8bb71b6 687 g_free(p);
0b919fae
KW
688 }
689}
690
f7d0fe02 691/* XXX: cache several refcount block clusters ? */
2aabe7c7
HR
692/* @addend is the absolute value of the addend; if @decrease is set, @addend
693 * will be subtracted from the current refcount, otherwise it will be added */
db3a964f 694static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
2aabe7c7
HR
695 int64_t offset,
696 int64_t length,
0e06528e 697 uint64_t addend,
2aabe7c7
HR
698 bool decrease,
699 enum qcow2_discard_type type)
f7d0fe02 700{
ff99129a 701 BDRVQcow2State *s = bs->opaque;
f7d0fe02 702 int64_t start, last, cluster_offset;
7453c96b 703 void *refcount_block = NULL;
29c1a730 704 int64_t old_table_index = -1;
09508d13 705 int ret;
f7d0fe02
KW
706
707#ifdef DEBUG_ALLOC2
2aabe7c7 708 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64
0e06528e 709 " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "",
2aabe7c7 710 addend);
f7d0fe02 711#endif
7322afe7 712 if (length < 0) {
f7d0fe02 713 return -EINVAL;
7322afe7
KW
714 } else if (length == 0) {
715 return 0;
716 }
717
2aabe7c7 718 if (decrease) {
29c1a730
KW
719 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
720 s->l2_table_cache);
721 }
722
ac95acdb
HT
723 start = start_of_cluster(s, offset);
724 last = start_of_cluster(s, offset + length - 1);
f7d0fe02
KW
725 for(cluster_offset = start; cluster_offset <= last;
726 cluster_offset += s->cluster_size)
727 {
2aabe7c7 728 int block_index;
0e06528e 729 uint64_t refcount;
f7d0fe02 730 int64_t cluster_index = cluster_offset >> s->cluster_bits;
17bd5f47 731 int64_t table_index = cluster_index >> s->refcount_block_bits;
f7d0fe02 732
29c1a730
KW
733 /* Load the refcount block and allocate it if needed */
734 if (table_index != old_table_index) {
735 if (refcount_block) {
a3f1afb4 736 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
29c1a730 737 }
29c1a730 738 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
ed0df867 739 if (ret < 0) {
29c1a730 740 goto fail;
f7d0fe02 741 }
f7d0fe02 742 }
29c1a730 743 old_table_index = table_index;
f7d0fe02 744
72e80b89
AG
745 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache,
746 refcount_block);
f7d0fe02
KW
747
748 /* we can update the count and save it */
17bd5f47 749 block_index = cluster_index & (s->refcount_block_size - 1);
f7d0fe02 750
7453c96b 751 refcount = s->get_refcount(refcount_block, block_index);
0e06528e
HR
752 if (decrease ? (refcount - addend > refcount)
753 : (refcount + addend < refcount ||
754 refcount + addend > s->refcount_max))
2aabe7c7 755 {
09508d13
KW
756 ret = -EINVAL;
757 goto fail;
758 }
2aabe7c7
HR
759 if (decrease) {
760 refcount -= addend;
761 } else {
762 refcount += addend;
763 }
f7d0fe02
KW
764 if (refcount == 0 && cluster_index < s->free_cluster_index) {
765 s->free_cluster_index = cluster_index;
766 }
7453c96b 767 s->set_refcount(refcount_block, block_index, refcount);
0b919fae 768
67af674e 769 if (refcount == 0 && s->discard_passthrough[type]) {
0b919fae 770 update_refcount_discard(bs, cluster_offset, s->cluster_size);
67af674e 771 }
f7d0fe02
KW
772 }
773
09508d13
KW
774 ret = 0;
775fail:
0b919fae
KW
776 if (!s->cache_discards) {
777 qcow2_process_discards(bs, ret);
778 }
779
f7d0fe02 780 /* Write last changed block to disk */
29c1a730 781 if (refcount_block) {
a3f1afb4 782 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
f7d0fe02
KW
783 }
784
09508d13
KW
785 /*
786 * Try do undo any updates if an error is returned (This may succeed in
787 * some cases like ENOSPC for allocating a new refcount block)
788 */
789 if (ret < 0) {
790 int dummy;
2aabe7c7
HR
791 dummy = update_refcount(bs, offset, cluster_offset - offset, addend,
792 !decrease, QCOW2_DISCARD_NEVER);
83e3f76c 793 (void)dummy;
09508d13
KW
794 }
795
796 return ret;
f7d0fe02
KW
797}
798
018faafd 799/*
44751917 800 * Increases or decreases the refcount of a given cluster.
018faafd 801 *
2aabe7c7
HR
802 * @addend is the absolute value of the addend; if @decrease is set, @addend
803 * will be subtracted from the current refcount, otherwise it will be added.
804 *
c6e9d8ae 805 * On success 0 is returned; on failure -errno is returned.
018faafd 806 */
32b6444d
HR
807int qcow2_update_cluster_refcount(BlockDriverState *bs,
808 int64_t cluster_index,
0e06528e 809 uint64_t addend, bool decrease,
32b6444d 810 enum qcow2_discard_type type)
f7d0fe02 811{
ff99129a 812 BDRVQcow2State *s = bs->opaque;
f7d0fe02
KW
813 int ret;
814
6cfcb9b8 815 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend,
2aabe7c7 816 decrease, type);
f7d0fe02
KW
817 if (ret < 0) {
818 return ret;
819 }
820
c6e9d8ae 821 return 0;
f7d0fe02
KW
822}
823
824
825
826/*********************************************************/
827/* cluster allocation functions */
828
829
830
831/* return < 0 if error */
bb572aef 832static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size)
f7d0fe02 833{
ff99129a 834 BDRVQcow2State *s = bs->opaque;
0e06528e 835 uint64_t i, nb_clusters, refcount;
7324c10f 836 int ret;
f7d0fe02 837
ecbda7a2
KW
838 /* We can't allocate clusters if they may still be queued for discard. */
839 if (s->cache_discards) {
840 qcow2_process_discards(bs, 0);
841 }
842
f7d0fe02
KW
843 nb_clusters = size_to_clusters(s, size);
844retry:
845 for(i = 0; i < nb_clusters; i++) {
bb572aef 846 uint64_t next_cluster_index = s->free_cluster_index++;
7324c10f 847 ret = qcow2_get_refcount(bs, next_cluster_index, &refcount);
2eaa8f63 848
7324c10f
HR
849 if (ret < 0) {
850 return ret;
2eaa8f63 851 } else if (refcount != 0) {
f7d0fe02 852 goto retry;
2eaa8f63 853 }
f7d0fe02 854 }
91f827dc
HR
855
856 /* Make sure that all offsets in the "allocated" range are representable
857 * in an int64_t */
65f33bc0
HR
858 if (s->free_cluster_index > 0 &&
859 s->free_cluster_index - 1 > (INT64_MAX >> s->cluster_bits))
860 {
91f827dc
HR
861 return -EFBIG;
862 }
863
f7d0fe02 864#ifdef DEBUG_ALLOC2
35ee5e39 865 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
f7d0fe02
KW
866 size,
867 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
868#endif
869 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
870}
871
bb572aef 872int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size)
f7d0fe02
KW
873{
874 int64_t offset;
db3a964f 875 int ret;
f7d0fe02 876
66f82cee 877 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
b106ad91
KW
878 do {
879 offset = alloc_clusters_noref(bs, size);
880 if (offset < 0) {
881 return offset;
882 }
883
2aabe7c7 884 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
b106ad91 885 } while (ret == -EAGAIN);
2eaa8f63 886
db3a964f
KW
887 if (ret < 0) {
888 return ret;
889 }
1c4c2814 890
f7d0fe02
KW
891 return offset;
892}
893
b6d36def
HR
894int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
895 int64_t nb_clusters)
256900b1 896{
ff99129a 897 BDRVQcow2State *s = bs->opaque;
0e06528e 898 uint64_t cluster_index, refcount;
33304ec9 899 uint64_t i;
7324c10f 900 int ret;
33304ec9
HT
901
902 assert(nb_clusters >= 0);
903 if (nb_clusters == 0) {
904 return 0;
905 }
256900b1 906
b106ad91
KW
907 do {
908 /* Check how many clusters there are free */
909 cluster_index = offset >> s->cluster_bits;
910 for(i = 0; i < nb_clusters; i++) {
7324c10f
HR
911 ret = qcow2_get_refcount(bs, cluster_index++, &refcount);
912 if (ret < 0) {
913 return ret;
b106ad91
KW
914 } else if (refcount != 0) {
915 break;
916 }
256900b1 917 }
256900b1 918
b106ad91 919 /* And then allocate them */
2aabe7c7 920 ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false,
b106ad91
KW
921 QCOW2_DISCARD_NEVER);
922 } while (ret == -EAGAIN);
f24423bd 923
256900b1
KW
924 if (ret < 0) {
925 return ret;
926 }
927
928 return i;
929}
930
f7d0fe02
KW
931/* only used to allocate compressed sectors. We try to allocate
932 contiguous sectors. size must be <= cluster_size */
ed6ccf0f 933int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
f7d0fe02 934{
ff99129a 935 BDRVQcow2State *s = bs->opaque;
8c44dfbc
HR
936 int64_t offset;
937 size_t free_in_cluster;
938 int ret;
f7d0fe02 939
66f82cee 940 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
f7d0fe02 941 assert(size > 0 && size <= s->cluster_size);
8c44dfbc
HR
942 assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset));
943
944 offset = s->free_byte_offset;
945
946 if (offset) {
0e06528e 947 uint64_t refcount;
7324c10f
HR
948 ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount);
949 if (ret < 0) {
950 return ret;
5d757b56 951 }
8c44dfbc 952
346a53df 953 if (refcount == s->refcount_max) {
8c44dfbc 954 offset = 0;
5d757b56 955 }
8c44dfbc
HR
956 }
957
958 free_in_cluster = s->cluster_size - offset_into_cluster(s, offset);
3e5feb62
JM
959 do {
960 if (!offset || free_in_cluster < size) {
961 int64_t new_cluster = alloc_clusters_noref(bs, s->cluster_size);
962 if (new_cluster < 0) {
963 return new_cluster;
964 }
8c44dfbc 965
3e5feb62
JM
966 if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) {
967 offset = new_cluster;
2ac01520
HR
968 free_in_cluster = s->cluster_size;
969 } else {
970 free_in_cluster += s->cluster_size;
3e5feb62 971 }
f7d0fe02 972 }
29216ed1 973
3e5feb62
JM
974 assert(offset);
975 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
2ac01520
HR
976 if (ret < 0) {
977 offset = 0;
978 }
3e5feb62 979 } while (ret == -EAGAIN);
8c44dfbc
HR
980 if (ret < 0) {
981 return ret;
982 }
983
984 /* The cluster refcount was incremented; refcount blocks must be flushed
985 * before the caller's L2 table updates. */
c1f5bafd 986 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
8c44dfbc
HR
987
988 s->free_byte_offset = offset + size;
989 if (!offset_into_cluster(s, s->free_byte_offset)) {
990 s->free_byte_offset = 0;
991 }
992
f7d0fe02
KW
993 return offset;
994}
995
ed6ccf0f 996void qcow2_free_clusters(BlockDriverState *bs,
6cfcb9b8
KW
997 int64_t offset, int64_t size,
998 enum qcow2_discard_type type)
f7d0fe02 999{
db3a964f
KW
1000 int ret;
1001
66f82cee 1002 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
2aabe7c7 1003 ret = update_refcount(bs, offset, size, 1, true, type);
db3a964f
KW
1004 if (ret < 0) {
1005 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
003fad6e 1006 /* TODO Remember the clusters to free them later and avoid leaking */
db3a964f 1007 }
f7d0fe02
KW
1008}
1009
45aba42f 1010/*
c7a4c37a
KW
1011 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
1012 * normal cluster, compressed cluster, etc.)
45aba42f 1013 */
6cfcb9b8
KW
1014void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
1015 int nb_clusters, enum qcow2_discard_type type)
45aba42f 1016{
ff99129a 1017 BDRVQcow2State *s = bs->opaque;
45aba42f 1018
c7a4c37a
KW
1019 switch (qcow2_get_cluster_type(l2_entry)) {
1020 case QCOW2_CLUSTER_COMPRESSED:
1021 {
1022 int nb_csectors;
1023 nb_csectors = ((l2_entry >> s->csize_shift) &
1024 s->csize_mask) + 1;
1025 qcow2_free_clusters(bs,
1026 (l2_entry & s->cluster_offset_mask) & ~511,
6cfcb9b8 1027 nb_csectors * 512, type);
c7a4c37a
KW
1028 }
1029 break;
1030 case QCOW2_CLUSTER_NORMAL:
8f730dd2
HR
1031 case QCOW2_CLUSTER_ZERO:
1032 if (l2_entry & L2E_OFFSET_MASK) {
a97c67ee
HR
1033 if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) {
1034 qcow2_signal_corruption(bs, false, -1, -1,
1035 "Cannot free unaligned cluster %#llx",
1036 l2_entry & L2E_OFFSET_MASK);
1037 } else {
1038 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
1039 nb_clusters << s->cluster_bits, type);
1040 }
8f730dd2 1041 }
c7a4c37a
KW
1042 break;
1043 case QCOW2_CLUSTER_UNALLOCATED:
1044 break;
1045 default:
1046 abort();
45aba42f 1047 }
45aba42f
KW
1048}
1049
f7d0fe02
KW
1050
1051
1052/*********************************************************/
1053/* snapshots and image creation */
1054
1055
1056
f7d0fe02 1057/* update the refcounts of snapshots and the copied flag */
ed6ccf0f
KW
1058int qcow2_update_snapshot_refcount(BlockDriverState *bs,
1059 int64_t l1_table_offset, int l1_size, int addend)
f7d0fe02 1060{
ff99129a 1061 BDRVQcow2State *s = bs->opaque;
b32cbae1 1062 uint64_t *l1_table, *l2_table, l2_offset, entry, l1_size2, refcount;
de82815d 1063 bool l1_allocated = false;
b32cbae1 1064 int64_t old_entry, old_l2_offset;
7324c10f 1065 int i, j, l1_modified = 0, nb_csectors;
29c1a730 1066 int ret;
f7d0fe02 1067
2aabe7c7
HR
1068 assert(addend >= -1 && addend <= 1);
1069
f7d0fe02
KW
1070 l2_table = NULL;
1071 l1_table = NULL;
1072 l1_size2 = l1_size * sizeof(uint64_t);
43a0cac4 1073
0b919fae
KW
1074 s->cache_discards = true;
1075
43a0cac4
KW
1076 /* WARNING: qcow2_snapshot_goto relies on this function not using the
1077 * l1_table_offset when it is the current s->l1_table_offset! Be careful
1078 * when changing this! */
f7d0fe02 1079 if (l1_table_offset != s->l1_table_offset) {
de82815d
KW
1080 l1_table = g_try_malloc0(align_offset(l1_size2, 512));
1081 if (l1_size2 && l1_table == NULL) {
1082 ret = -ENOMEM;
1083 goto fail;
1084 }
1085 l1_allocated = true;
c2bc78b6 1086
cf2ab8fc 1087 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
c2bc78b6 1088 if (ret < 0) {
f7d0fe02 1089 goto fail;
93913dfd
KW
1090 }
1091
b32cbae1 1092 for (i = 0; i < l1_size; i++) {
f7d0fe02 1093 be64_to_cpus(&l1_table[i]);
b32cbae1 1094 }
f7d0fe02
KW
1095 } else {
1096 assert(l1_size == s->l1_size);
1097 l1_table = s->l1_table;
de82815d 1098 l1_allocated = false;
f7d0fe02
KW
1099 }
1100
b32cbae1 1101 for (i = 0; i < l1_size; i++) {
f7d0fe02
KW
1102 l2_offset = l1_table[i];
1103 if (l2_offset) {
1104 old_l2_offset = l2_offset;
8e37f681 1105 l2_offset &= L1E_OFFSET_MASK;
29c1a730 1106
a97c67ee
HR
1107 if (offset_into_cluster(s, l2_offset)) {
1108 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1109 PRIx64 " unaligned (L1 index: %#x)",
1110 l2_offset, i);
1111 ret = -EIO;
1112 goto fail;
1113 }
1114
29c1a730
KW
1115 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1116 (void**) &l2_table);
1117 if (ret < 0) {
f7d0fe02 1118 goto fail;
29c1a730
KW
1119 }
1120
b32cbae1 1121 for (j = 0; j < s->l2_size; j++) {
8b81a7b6 1122 uint64_t cluster_index;
b32cbae1 1123 uint64_t offset;
8b81a7b6 1124
b32cbae1
EB
1125 entry = be64_to_cpu(l2_table[j]);
1126 old_entry = entry;
1127 entry &= ~QCOW_OFLAG_COPIED;
1128 offset = entry & L2E_OFFSET_MASK;
8b81a7b6 1129
b32cbae1 1130 switch (qcow2_get_cluster_type(entry)) {
bbd995d8
EB
1131 case QCOW2_CLUSTER_COMPRESSED:
1132 nb_csectors = ((entry >> s->csize_shift) &
1133 s->csize_mask) + 1;
1134 if (addend != 0) {
1135 ret = update_refcount(bs,
b32cbae1 1136 (entry & s->cluster_offset_mask) & ~511,
2aabe7c7 1137 nb_csectors * 512, abs(addend), addend < 0,
6cfcb9b8 1138 QCOW2_DISCARD_SNAPSHOT);
bbd995d8 1139 if (ret < 0) {
a97c67ee
HR
1140 goto fail;
1141 }
bbd995d8
EB
1142 }
1143 /* compressed clusters are never modified */
1144 refcount = 2;
1145 break;
1146
1147 case QCOW2_CLUSTER_NORMAL:
1148 case QCOW2_CLUSTER_ZERO:
1149 if (offset_into_cluster(s, offset)) {
1150 qcow2_signal_corruption(bs, true, -1, -1, "Data "
1151 "cluster offset %#" PRIx64
1152 " unaligned (L2 offset: %#"
1153 PRIx64 ", L2 index: %#x)",
1154 offset, l2_offset, j);
1155 ret = -EIO;
1156 goto fail;
1157 }
a97c67ee 1158
bbd995d8
EB
1159 cluster_index = offset >> s->cluster_bits;
1160 if (!cluster_index) {
1161 /* unallocated */
1162 refcount = 0;
1163 break;
1164 }
1165 if (addend != 0) {
1166 ret = qcow2_update_cluster_refcount(bs,
2aabe7c7 1167 cluster_index, abs(addend), addend < 0,
32b6444d 1168 QCOW2_DISCARD_SNAPSHOT);
7324c10f 1169 if (ret < 0) {
018faafd
KW
1170 goto fail;
1171 }
bbd995d8 1172 }
f7d0fe02 1173
bbd995d8
EB
1174 ret = qcow2_get_refcount(bs, cluster_index, &refcount);
1175 if (ret < 0) {
1176 goto fail;
1177 }
1178 break;
1179
1180 case QCOW2_CLUSTER_UNALLOCATED:
1181 refcount = 0;
1182 break;
8b81a7b6 1183
bbd995d8
EB
1184 default:
1185 abort();
8b81a7b6
HR
1186 }
1187
1188 if (refcount == 1) {
b32cbae1 1189 entry |= QCOW_OFLAG_COPIED;
8b81a7b6 1190 }
b32cbae1 1191 if (entry != old_entry) {
8b81a7b6
HR
1192 if (addend > 0) {
1193 qcow2_cache_set_dependency(bs, s->l2_table_cache,
1194 s->refcount_block_cache);
f7d0fe02 1195 }
b32cbae1 1196 l2_table[j] = cpu_to_be64(entry);
72e80b89
AG
1197 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache,
1198 l2_table);
f7d0fe02
KW
1199 }
1200 }
29c1a730 1201
a3f1afb4 1202 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
29c1a730 1203
f7d0fe02 1204 if (addend != 0) {
c6e9d8ae
HR
1205 ret = qcow2_update_cluster_refcount(bs, l2_offset >>
1206 s->cluster_bits,
2aabe7c7 1207 abs(addend), addend < 0,
c6e9d8ae
HR
1208 QCOW2_DISCARD_SNAPSHOT);
1209 if (ret < 0) {
1210 goto fail;
1211 }
f7d0fe02 1212 }
7324c10f
HR
1213 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1214 &refcount);
1215 if (ret < 0) {
018faafd
KW
1216 goto fail;
1217 } else if (refcount == 1) {
f7d0fe02
KW
1218 l2_offset |= QCOW_OFLAG_COPIED;
1219 }
1220 if (l2_offset != old_l2_offset) {
1221 l1_table[i] = l2_offset;
1222 l1_modified = 1;
1223 }
1224 }
1225 }
93913dfd 1226
2154f24e 1227 ret = bdrv_flush(bs);
93913dfd
KW
1228fail:
1229 if (l2_table) {
1230 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1231 }
1232
0b919fae
KW
1233 s->cache_discards = false;
1234 qcow2_process_discards(bs, ret);
1235
43a0cac4 1236 /* Update L1 only if it isn't deleted anyway (addend = -1) */
c2b6ff51
KW
1237 if (ret == 0 && addend >= 0 && l1_modified) {
1238 for (i = 0; i < l1_size; i++) {
f7d0fe02 1239 cpu_to_be64s(&l1_table[i]);
c2b6ff51
KW
1240 }
1241
d9ca2ea2 1242 ret = bdrv_pwrite_sync(bs->file, l1_table_offset,
9a4f4c31 1243 l1_table, l1_size2);
c2b6ff51
KW
1244
1245 for (i = 0; i < l1_size; i++) {
f7d0fe02 1246 be64_to_cpus(&l1_table[i]);
c2b6ff51 1247 }
f7d0fe02
KW
1248 }
1249 if (l1_allocated)
7267c094 1250 g_free(l1_table);
93913dfd 1251 return ret;
f7d0fe02
KW
1252}
1253
1254
1255
1256
1257/*********************************************************/
1258/* refcount checking functions */
1259
1260
c2551b47 1261static uint64_t refcount_array_byte_size(BDRVQcow2State *s, uint64_t entries)
5fee192e
HR
1262{
1263 /* This assertion holds because there is no way we can address more than
1264 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because
1265 * offsets have to be representable in bytes); due to every cluster
1266 * corresponding to one refcount entry, we are well below that limit */
1267 assert(entries < (UINT64_C(1) << (64 - 9)));
1268
1269 /* Thanks to the assertion this will not overflow, because
1270 * s->refcount_order < 7.
1271 * (note: x << s->refcount_order == x * s->refcount_bits) */
1272 return DIV_ROUND_UP(entries << s->refcount_order, 8);
1273}
1274
1275/**
1276 * Reallocates *array so that it can hold new_size entries. *size must contain
1277 * the current number of entries in *array. If the reallocation fails, *array
1278 * and *size will not be modified and -errno will be returned. If the
1279 * reallocation is successful, *array will be set to the new buffer, *size
1280 * will be set to new_size and 0 will be returned. The size of the reallocated
1281 * refcount array buffer will be aligned to a cluster boundary, and the newly
1282 * allocated area will be zeroed.
1283 */
ff99129a 1284static int realloc_refcount_array(BDRVQcow2State *s, void **array,
5fee192e
HR
1285 int64_t *size, int64_t new_size)
1286{
b6d36def 1287 int64_t old_byte_size, new_byte_size;
7453c96b 1288 void *new_ptr;
5fee192e
HR
1289
1290 /* Round to clusters so the array can be directly written to disk */
1291 old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size))
1292 * s->cluster_size;
1293 new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size))
1294 * s->cluster_size;
1295
1296 if (new_byte_size == old_byte_size) {
1297 *size = new_size;
1298 return 0;
1299 }
1300
1301 assert(new_byte_size > 0);
1302
b6d36def
HR
1303 if (new_byte_size > SIZE_MAX) {
1304 return -ENOMEM;
1305 }
1306
5fee192e
HR
1307 new_ptr = g_try_realloc(*array, new_byte_size);
1308 if (!new_ptr) {
1309 return -ENOMEM;
1310 }
1311
1312 if (new_byte_size > old_byte_size) {
b6d36def 1313 memset((char *)new_ptr + old_byte_size, 0,
5fee192e
HR
1314 new_byte_size - old_byte_size);
1315 }
1316
1317 *array = new_ptr;
1318 *size = new_size;
1319
1320 return 0;
1321}
f7d0fe02
KW
1322
1323/*
1324 * Increases the refcount for a range of clusters in a given refcount table.
1325 * This is used to construct a temporary refcount table out of L1 and L2 tables
b6af0975 1326 * which can be compared to the refcount table saved in the image.
f7d0fe02 1327 *
9ac228e0 1328 * Modifies the number of errors in res.
f7d0fe02 1329 */
fef4d3d5
HR
1330static int inc_refcounts(BlockDriverState *bs,
1331 BdrvCheckResult *res,
7453c96b 1332 void **refcount_table,
641bb63c 1333 int64_t *refcount_table_size,
fef4d3d5 1334 int64_t offset, int64_t size)
f7d0fe02 1335{
ff99129a 1336 BDRVQcow2State *s = bs->opaque;
7453c96b 1337 uint64_t start, last, cluster_offset, k, refcount;
5fee192e 1338 int ret;
f7d0fe02 1339
fef4d3d5
HR
1340 if (size <= 0) {
1341 return 0;
1342 }
f7d0fe02 1343
ac95acdb
HT
1344 start = start_of_cluster(s, offset);
1345 last = start_of_cluster(s, offset + size - 1);
f7d0fe02
KW
1346 for(cluster_offset = start; cluster_offset <= last;
1347 cluster_offset += s->cluster_size) {
1348 k = cluster_offset >> s->cluster_bits;
641bb63c 1349 if (k >= *refcount_table_size) {
5fee192e
HR
1350 ret = realloc_refcount_array(s, refcount_table,
1351 refcount_table_size, k + 1);
1352 if (ret < 0) {
641bb63c 1353 res->check_errors++;
5fee192e 1354 return ret;
f7d0fe02 1355 }
641bb63c
HR
1356 }
1357
7453c96b
HR
1358 refcount = s->get_refcount(*refcount_table, k);
1359 if (refcount == s->refcount_max) {
641bb63c
HR
1360 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
1361 "\n", cluster_offset);
03bb78ed
HR
1362 fprintf(stderr, "Use qemu-img amend to increase the refcount entry "
1363 "width or qemu-img convert to create a clean copy if the "
1364 "image cannot be opened for writing\n");
641bb63c 1365 res->corruptions++;
7453c96b 1366 continue;
f7d0fe02 1367 }
7453c96b 1368 s->set_refcount(*refcount_table, k, refcount + 1);
f7d0fe02 1369 }
fef4d3d5
HR
1370
1371 return 0;
f7d0fe02
KW
1372}
1373
801f7044
SH
1374/* Flags for check_refcounts_l1() and check_refcounts_l2() */
1375enum {
fba31bae 1376 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
801f7044
SH
1377};
1378
f7d0fe02
KW
1379/*
1380 * Increases the refcount in the given refcount table for the all clusters
1381 * referenced in the L2 table. While doing so, performs some checks on L2
1382 * entries.
1383 *
1384 * Returns the number of errors found by the checks or -errno if an internal
1385 * error occurred.
1386 */
9ac228e0 1387static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
7453c96b
HR
1388 void **refcount_table,
1389 int64_t *refcount_table_size, int64_t l2_offset,
1390 int flags)
f7d0fe02 1391{
ff99129a 1392 BDRVQcow2State *s = bs->opaque;
afdf0abe 1393 uint64_t *l2_table, l2_entry;
fba31bae 1394 uint64_t next_contiguous_offset = 0;
ad27390c 1395 int i, l2_size, nb_csectors, ret;
f7d0fe02
KW
1396
1397 /* Read L2 table from disk */
1398 l2_size = s->l2_size * sizeof(uint64_t);
7267c094 1399 l2_table = g_malloc(l2_size);
f7d0fe02 1400
cf2ab8fc 1401 ret = bdrv_pread(bs->file, l2_offset, l2_table, l2_size);
ad27390c
HR
1402 if (ret < 0) {
1403 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1404 res->check_errors++;
f7d0fe02 1405 goto fail;
ad27390c 1406 }
f7d0fe02
KW
1407
1408 /* Do the actual checks */
1409 for(i = 0; i < s->l2_size; i++) {
afdf0abe
KW
1410 l2_entry = be64_to_cpu(l2_table[i]);
1411
1412 switch (qcow2_get_cluster_type(l2_entry)) {
1413 case QCOW2_CLUSTER_COMPRESSED:
1414 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1415 if (l2_entry & QCOW_OFLAG_COPIED) {
1416 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
1417 "copied flag must never be set for compressed "
1418 "clusters\n", l2_entry >> s->cluster_bits);
1419 l2_entry &= ~QCOW_OFLAG_COPIED;
1420 res->corruptions++;
1421 }
f7d0fe02 1422
afdf0abe
KW
1423 /* Mark cluster as used */
1424 nb_csectors = ((l2_entry >> s->csize_shift) &
1425 s->csize_mask) + 1;
1426 l2_entry &= s->cluster_offset_mask;
fef4d3d5
HR
1427 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1428 l2_entry & ~511, nb_csectors * 512);
1429 if (ret < 0) {
1430 goto fail;
1431 }
fba31bae
SH
1432
1433 if (flags & CHECK_FRAG_INFO) {
1434 res->bfi.allocated_clusters++;
4db35162 1435 res->bfi.compressed_clusters++;
fba31bae
SH
1436
1437 /* Compressed clusters are fragmented by nature. Since they
1438 * take up sub-sector space but we only have sector granularity
1439 * I/O we need to re-read the same sectors even for adjacent
1440 * compressed clusters.
1441 */
1442 res->bfi.fragmented_clusters++;
1443 }
afdf0abe 1444 break;
f7d0fe02 1445
6377af48
KW
1446 case QCOW2_CLUSTER_ZERO:
1447 if ((l2_entry & L2E_OFFSET_MASK) == 0) {
1448 break;
1449 }
1450 /* fall through */
1451
afdf0abe
KW
1452 case QCOW2_CLUSTER_NORMAL:
1453 {
afdf0abe 1454 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
f7d0fe02 1455
fba31bae
SH
1456 if (flags & CHECK_FRAG_INFO) {
1457 res->bfi.allocated_clusters++;
1458 if (next_contiguous_offset &&
1459 offset != next_contiguous_offset) {
1460 res->bfi.fragmented_clusters++;
1461 }
1462 next_contiguous_offset = offset + s->cluster_size;
1463 }
1464
afdf0abe 1465 /* Mark cluster as used */
fef4d3d5
HR
1466 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1467 offset, s->cluster_size);
1468 if (ret < 0) {
1469 goto fail;
1470 }
afdf0abe
KW
1471
1472 /* Correct offsets are cluster aligned */
ac95acdb 1473 if (offset_into_cluster(s, offset)) {
afdf0abe
KW
1474 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
1475 "properly aligned; L2 entry corrupted.\n", offset);
1476 res->corruptions++;
1477 }
1478 break;
1479 }
1480
1481 case QCOW2_CLUSTER_UNALLOCATED:
1482 break;
1483
1484 default:
1485 abort();
f7d0fe02
KW
1486 }
1487 }
1488
7267c094 1489 g_free(l2_table);
9ac228e0 1490 return 0;
f7d0fe02
KW
1491
1492fail:
7267c094 1493 g_free(l2_table);
ad27390c 1494 return ret;
f7d0fe02
KW
1495}
1496
1497/*
1498 * Increases the refcount for the L1 table, its L2 tables and all referenced
1499 * clusters in the given refcount table. While doing so, performs some checks
1500 * on L1 and L2 entries.
1501 *
1502 * Returns the number of errors found by the checks or -errno if an internal
1503 * error occurred.
1504 */
1505static int check_refcounts_l1(BlockDriverState *bs,
9ac228e0 1506 BdrvCheckResult *res,
7453c96b 1507 void **refcount_table,
641bb63c 1508 int64_t *refcount_table_size,
f7d0fe02 1509 int64_t l1_table_offset, int l1_size,
801f7044 1510 int flags)
f7d0fe02 1511{
ff99129a 1512 BDRVQcow2State *s = bs->opaque;
fef4d3d5 1513 uint64_t *l1_table = NULL, l2_offset, l1_size2;
4f6ed88c 1514 int i, ret;
f7d0fe02
KW
1515
1516 l1_size2 = l1_size * sizeof(uint64_t);
1517
1518 /* Mark L1 table as used */
fef4d3d5
HR
1519 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1520 l1_table_offset, l1_size2);
1521 if (ret < 0) {
1522 goto fail;
1523 }
f7d0fe02
KW
1524
1525 /* Read L1 table entries from disk */
fef4d3d5 1526 if (l1_size2 > 0) {
de82815d
KW
1527 l1_table = g_try_malloc(l1_size2);
1528 if (l1_table == NULL) {
1529 ret = -ENOMEM;
ad27390c 1530 res->check_errors++;
de82815d
KW
1531 goto fail;
1532 }
cf2ab8fc 1533 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
ad27390c
HR
1534 if (ret < 0) {
1535 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1536 res->check_errors++;
702ef63f 1537 goto fail;
ad27390c 1538 }
702ef63f
KW
1539 for(i = 0;i < l1_size; i++)
1540 be64_to_cpus(&l1_table[i]);
1541 }
f7d0fe02
KW
1542
1543 /* Do the actual checks */
1544 for(i = 0; i < l1_size; i++) {
1545 l2_offset = l1_table[i];
1546 if (l2_offset) {
f7d0fe02 1547 /* Mark L2 table as used */
afdf0abe 1548 l2_offset &= L1E_OFFSET_MASK;
fef4d3d5
HR
1549 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1550 l2_offset, s->cluster_size);
1551 if (ret < 0) {
1552 goto fail;
1553 }
f7d0fe02
KW
1554
1555 /* L2 tables are cluster aligned */
ac95acdb 1556 if (offset_into_cluster(s, l2_offset)) {
f7d0fe02
KW
1557 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1558 "cluster aligned; L1 entry corrupted\n", l2_offset);
9ac228e0 1559 res->corruptions++;
f7d0fe02
KW
1560 }
1561
1562 /* Process and check L2 entries */
9ac228e0 1563 ret = check_refcounts_l2(bs, res, refcount_table,
801f7044 1564 refcount_table_size, l2_offset, flags);
f7d0fe02
KW
1565 if (ret < 0) {
1566 goto fail;
1567 }
f7d0fe02
KW
1568 }
1569 }
7267c094 1570 g_free(l1_table);
9ac228e0 1571 return 0;
f7d0fe02
KW
1572
1573fail:
7267c094 1574 g_free(l1_table);
ad27390c 1575 return ret;
f7d0fe02
KW
1576}
1577
4f6ed88c
HR
1578/*
1579 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1580 *
1581 * This function does not print an error message nor does it increment
44751917
HR
1582 * check_errors if qcow2_get_refcount fails (this is because such an error will
1583 * have been already detected and sufficiently signaled by the calling function
4f6ed88c
HR
1584 * (qcow2_check_refcounts) by the time this function is called).
1585 */
e23e400e
HR
1586static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
1587 BdrvCheckMode fix)
4f6ed88c 1588{
ff99129a 1589 BDRVQcow2State *s = bs->opaque;
4f6ed88c
HR
1590 uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size);
1591 int ret;
0e06528e 1592 uint64_t refcount;
4f6ed88c
HR
1593 int i, j;
1594
1595 for (i = 0; i < s->l1_size; i++) {
1596 uint64_t l1_entry = s->l1_table[i];
1597 uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK;
e23e400e 1598 bool l2_dirty = false;
4f6ed88c
HR
1599
1600 if (!l2_offset) {
1601 continue;
1602 }
1603
7324c10f
HR
1604 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1605 &refcount);
1606 if (ret < 0) {
4f6ed88c
HR
1607 /* don't print message nor increment check_errors */
1608 continue;
1609 }
1610 if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) {
e23e400e 1611 fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
0e06528e 1612 "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
e23e400e
HR
1613 fix & BDRV_FIX_ERRORS ? "Repairing" :
1614 "ERROR",
4f6ed88c 1615 i, l1_entry, refcount);
e23e400e
HR
1616 if (fix & BDRV_FIX_ERRORS) {
1617 s->l1_table[i] = refcount == 1
1618 ? l1_entry | QCOW_OFLAG_COPIED
1619 : l1_entry & ~QCOW_OFLAG_COPIED;
1620 ret = qcow2_write_l1_entry(bs, i);
1621 if (ret < 0) {
1622 res->check_errors++;
1623 goto fail;
1624 }
1625 res->corruptions_fixed++;
1626 } else {
1627 res->corruptions++;
1628 }
4f6ed88c
HR
1629 }
1630
cf2ab8fc 1631 ret = bdrv_pread(bs->file, l2_offset, l2_table,
4f6ed88c
HR
1632 s->l2_size * sizeof(uint64_t));
1633 if (ret < 0) {
1634 fprintf(stderr, "ERROR: Could not read L2 table: %s\n",
1635 strerror(-ret));
1636 res->check_errors++;
1637 goto fail;
1638 }
1639
1640 for (j = 0; j < s->l2_size; j++) {
1641 uint64_t l2_entry = be64_to_cpu(l2_table[j]);
1642 uint64_t data_offset = l2_entry & L2E_OFFSET_MASK;
3ef95218 1643 QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry);
4f6ed88c
HR
1644
1645 if ((cluster_type == QCOW2_CLUSTER_NORMAL) ||
1646 ((cluster_type == QCOW2_CLUSTER_ZERO) && (data_offset != 0))) {
7324c10f
HR
1647 ret = qcow2_get_refcount(bs,
1648 data_offset >> s->cluster_bits,
1649 &refcount);
1650 if (ret < 0) {
4f6ed88c
HR
1651 /* don't print message nor increment check_errors */
1652 continue;
1653 }
1654 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
e23e400e 1655 fprintf(stderr, "%s OFLAG_COPIED data cluster: "
0e06528e 1656 "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
e23e400e
HR
1657 fix & BDRV_FIX_ERRORS ? "Repairing" :
1658 "ERROR",
4f6ed88c 1659 l2_entry, refcount);
e23e400e
HR
1660 if (fix & BDRV_FIX_ERRORS) {
1661 l2_table[j] = cpu_to_be64(refcount == 1
1662 ? l2_entry | QCOW_OFLAG_COPIED
1663 : l2_entry & ~QCOW_OFLAG_COPIED);
1664 l2_dirty = true;
1665 res->corruptions_fixed++;
1666 } else {
1667 res->corruptions++;
1668 }
4f6ed88c
HR
1669 }
1670 }
1671 }
e23e400e
HR
1672
1673 if (l2_dirty) {
231bb267
HR
1674 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
1675 l2_offset, s->cluster_size);
e23e400e
HR
1676 if (ret < 0) {
1677 fprintf(stderr, "ERROR: Could not write L2 table; metadata "
1678 "overlap check failed: %s\n", strerror(-ret));
1679 res->check_errors++;
1680 goto fail;
1681 }
1682
d9ca2ea2 1683 ret = bdrv_pwrite(bs->file, l2_offset, l2_table,
9a4f4c31 1684 s->cluster_size);
e23e400e
HR
1685 if (ret < 0) {
1686 fprintf(stderr, "ERROR: Could not write L2 table: %s\n",
1687 strerror(-ret));
1688 res->check_errors++;
1689 goto fail;
1690 }
1691 }
4f6ed88c
HR
1692 }
1693
1694 ret = 0;
1695
1696fail:
1697 qemu_vfree(l2_table);
1698 return ret;
1699}
1700
6ca56bf5
HR
1701/*
1702 * Checks consistency of refblocks and accounts for each refblock in
1703 * *refcount_table.
1704 */
1705static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res,
f307b255 1706 BdrvCheckMode fix, bool *rebuild,
7453c96b 1707 void **refcount_table, int64_t *nb_clusters)
6ca56bf5 1708{
ff99129a 1709 BDRVQcow2State *s = bs->opaque;
001c158d 1710 int64_t i, size;
fef4d3d5 1711 int ret;
6ca56bf5 1712
f7d0fe02 1713 for(i = 0; i < s->refcount_table_size; i++) {
6882c8fa 1714 uint64_t offset, cluster;
f7d0fe02 1715 offset = s->refcount_table[i];
6882c8fa 1716 cluster = offset >> s->cluster_bits;
746c3cb5
KW
1717
1718 /* Refcount blocks are cluster aligned */
ac95acdb 1719 if (offset_into_cluster(s, offset)) {
166acf54 1720 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
746c3cb5 1721 "cluster aligned; refcount table entry corrupted\n", i);
9ac228e0 1722 res->corruptions++;
f307b255 1723 *rebuild = true;
6882c8fa
KW
1724 continue;
1725 }
1726
6ca56bf5 1727 if (cluster >= *nb_clusters) {
001c158d
HR
1728 fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n",
1729 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
1730
1731 if (fix & BDRV_FIX_ERRORS) {
5fee192e 1732 int64_t new_nb_clusters;
ed3d2ec9 1733 Error *local_err = NULL;
001c158d
HR
1734
1735 if (offset > INT64_MAX - s->cluster_size) {
1736 ret = -EINVAL;
1737 goto resize_fail;
1738 }
1739
ed3d2ec9
HR
1740 ret = bdrv_truncate(bs->file, offset + s->cluster_size,
1741 &local_err);
001c158d 1742 if (ret < 0) {
ed3d2ec9 1743 error_report_err(local_err);
001c158d
HR
1744 goto resize_fail;
1745 }
9a4f4c31 1746 size = bdrv_getlength(bs->file->bs);
001c158d
HR
1747 if (size < 0) {
1748 ret = size;
1749 goto resize_fail;
1750 }
1751
5fee192e
HR
1752 new_nb_clusters = size_to_clusters(s, size);
1753 assert(new_nb_clusters >= *nb_clusters);
001c158d 1754
5fee192e
HR
1755 ret = realloc_refcount_array(s, refcount_table,
1756 nb_clusters, new_nb_clusters);
1757 if (ret < 0) {
001c158d 1758 res->check_errors++;
5fee192e 1759 return ret;
001c158d 1760 }
001c158d
HR
1761
1762 if (cluster >= *nb_clusters) {
1763 ret = -EINVAL;
1764 goto resize_fail;
1765 }
1766
1767 res->corruptions_fixed++;
1768 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1769 offset, s->cluster_size);
1770 if (ret < 0) {
1771 return ret;
1772 }
1773 /* No need to check whether the refcount is now greater than 1:
1774 * This area was just allocated and zeroed, so it can only be
1775 * exactly 1 after inc_refcounts() */
1776 continue;
1777
1778resize_fail:
1779 res->corruptions++;
f307b255 1780 *rebuild = true;
001c158d
HR
1781 fprintf(stderr, "ERROR could not resize image: %s\n",
1782 strerror(-ret));
1783 } else {
1784 res->corruptions++;
1785 }
6882c8fa 1786 continue;
746c3cb5
KW
1787 }
1788
f7d0fe02 1789 if (offset != 0) {
641bb63c 1790 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
fef4d3d5
HR
1791 offset, s->cluster_size);
1792 if (ret < 0) {
1793 return ret;
1794 }
7453c96b 1795 if (s->get_refcount(*refcount_table, cluster) != 1) {
f307b255 1796 fprintf(stderr, "ERROR refcount block %" PRId64
7453c96b
HR
1797 " refcount=%" PRIu64 "\n", i,
1798 s->get_refcount(*refcount_table, cluster));
f307b255
HR
1799 res->corruptions++;
1800 *rebuild = true;
746c3cb5 1801 }
f7d0fe02
KW
1802 }
1803 }
1804
6ca56bf5
HR
1805 return 0;
1806}
1807
057a3fe5
HR
1808/*
1809 * Calculates an in-memory refcount table.
1810 */
1811static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
f307b255 1812 BdrvCheckMode fix, bool *rebuild,
7453c96b 1813 void **refcount_table, int64_t *nb_clusters)
057a3fe5 1814{
ff99129a 1815 BDRVQcow2State *s = bs->opaque;
057a3fe5
HR
1816 int64_t i;
1817 QCowSnapshot *sn;
1818 int ret;
1819
9696df21 1820 if (!*refcount_table) {
5fee192e
HR
1821 int64_t old_size = 0;
1822 ret = realloc_refcount_array(s, refcount_table,
1823 &old_size, *nb_clusters);
1824 if (ret < 0) {
9696df21 1825 res->check_errors++;
5fee192e 1826 return ret;
9696df21 1827 }
057a3fe5
HR
1828 }
1829
1830 /* header */
641bb63c 1831 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
fef4d3d5
HR
1832 0, s->cluster_size);
1833 if (ret < 0) {
1834 return ret;
1835 }
057a3fe5
HR
1836
1837 /* current L1 table */
641bb63c 1838 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
057a3fe5
HR
1839 s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO);
1840 if (ret < 0) {
1841 return ret;
1842 }
1843
1844 /* snapshots */
1845 for (i = 0; i < s->nb_snapshots; i++) {
1846 sn = s->snapshots + i;
641bb63c 1847 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
fef4d3d5 1848 sn->l1_table_offset, sn->l1_size, 0);
057a3fe5
HR
1849 if (ret < 0) {
1850 return ret;
1851 }
1852 }
641bb63c 1853 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
fef4d3d5
HR
1854 s->snapshots_offset, s->snapshots_size);
1855 if (ret < 0) {
1856 return ret;
1857 }
057a3fe5
HR
1858
1859 /* refcount data */
641bb63c 1860 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
fef4d3d5
HR
1861 s->refcount_table_offset,
1862 s->refcount_table_size * sizeof(uint64_t));
1863 if (ret < 0) {
1864 return ret;
1865 }
057a3fe5 1866
f307b255 1867 return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters);
057a3fe5
HR
1868}
1869
6ca56bf5
HR
1870/*
1871 * Compares the actual reference count for each cluster in the image against the
1872 * refcount as reported by the refcount structures on-disk.
1873 */
1874static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
f307b255
HR
1875 BdrvCheckMode fix, bool *rebuild,
1876 int64_t *highest_cluster,
7453c96b 1877 void *refcount_table, int64_t nb_clusters)
6ca56bf5 1878{
ff99129a 1879 BDRVQcow2State *s = bs->opaque;
6ca56bf5 1880 int64_t i;
0e06528e 1881 uint64_t refcount1, refcount2;
7324c10f 1882 int ret;
6ca56bf5
HR
1883
1884 for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) {
7324c10f
HR
1885 ret = qcow2_get_refcount(bs, i, &refcount1);
1886 if (ret < 0) {
166acf54 1887 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
7324c10f 1888 i, strerror(-ret));
9ac228e0 1889 res->check_errors++;
f74550fd 1890 continue;
018faafd
KW
1891 }
1892
7453c96b 1893 refcount2 = s->get_refcount(refcount_table, i);
c6bb9ad1
FS
1894
1895 if (refcount1 > 0 || refcount2 > 0) {
6ca56bf5 1896 *highest_cluster = i;
c6bb9ad1
FS
1897 }
1898
f7d0fe02 1899 if (refcount1 != refcount2) {
166acf54
KW
1900 /* Check if we're allowed to fix the mismatch */
1901 int *num_fixed = NULL;
f307b255
HR
1902 if (refcount1 == 0) {
1903 *rebuild = true;
1904 } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
166acf54
KW
1905 num_fixed = &res->leaks_fixed;
1906 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
1907 num_fixed = &res->corruptions_fixed;
1908 }
1909
0e06528e
HR
1910 fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64
1911 " reference=%" PRIu64 "\n",
166acf54
KW
1912 num_fixed != NULL ? "Repairing" :
1913 refcount1 < refcount2 ? "ERROR" :
1914 "Leaked",
f7d0fe02 1915 i, refcount1, refcount2);
166acf54
KW
1916
1917 if (num_fixed) {
1918 ret = update_refcount(bs, i << s->cluster_bits, 1,
2aabe7c7
HR
1919 refcount_diff(refcount1, refcount2),
1920 refcount1 > refcount2,
6cfcb9b8 1921 QCOW2_DISCARD_ALWAYS);
166acf54
KW
1922 if (ret >= 0) {
1923 (*num_fixed)++;
1924 continue;
1925 }
1926 }
1927
1928 /* And if we couldn't, print an error */
9ac228e0
KW
1929 if (refcount1 < refcount2) {
1930 res->corruptions++;
1931 } else {
1932 res->leaks++;
1933 }
f7d0fe02
KW
1934 }
1935 }
6ca56bf5
HR
1936}
1937
c7c0681b
HR
1938/*
1939 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to
1940 * the on-disk refcount structures.
1941 *
1942 * On input, *first_free_cluster tells where to start looking, and need not
1943 * actually be a free cluster; the returned offset will not be before that
1944 * cluster. On output, *first_free_cluster points to the first gap found, even
1945 * if that gap was too small to be used as the returned offset.
1946 *
1947 * Note that *first_free_cluster is a cluster index whereas the return value is
1948 * an offset.
1949 */
1950static int64_t alloc_clusters_imrt(BlockDriverState *bs,
1951 int cluster_count,
7453c96b 1952 void **refcount_table,
c7c0681b
HR
1953 int64_t *imrt_nb_clusters,
1954 int64_t *first_free_cluster)
1955{
ff99129a 1956 BDRVQcow2State *s = bs->opaque;
c7c0681b
HR
1957 int64_t cluster = *first_free_cluster, i;
1958 bool first_gap = true;
1959 int contiguous_free_clusters;
5fee192e 1960 int ret;
c7c0681b
HR
1961
1962 /* Starting at *first_free_cluster, find a range of at least cluster_count
1963 * continuously free clusters */
1964 for (contiguous_free_clusters = 0;
1965 cluster < *imrt_nb_clusters &&
1966 contiguous_free_clusters < cluster_count;
1967 cluster++)
1968 {
7453c96b 1969 if (!s->get_refcount(*refcount_table, cluster)) {
c7c0681b
HR
1970 contiguous_free_clusters++;
1971 if (first_gap) {
1972 /* If this is the first free cluster found, update
1973 * *first_free_cluster accordingly */
1974 *first_free_cluster = cluster;
1975 first_gap = false;
1976 }
1977 } else if (contiguous_free_clusters) {
1978 contiguous_free_clusters = 0;
1979 }
1980 }
1981
1982 /* If contiguous_free_clusters is greater than zero, it contains the number
1983 * of continuously free clusters until the current cluster; the first free
1984 * cluster in the current "gap" is therefore
1985 * cluster - contiguous_free_clusters */
1986
1987 /* If no such range could be found, grow the in-memory refcount table
1988 * accordingly to append free clusters at the end of the image */
1989 if (contiguous_free_clusters < cluster_count) {
c7c0681b
HR
1990 /* contiguous_free_clusters clusters are already empty at the image end;
1991 * we need cluster_count clusters; therefore, we have to allocate
1992 * cluster_count - contiguous_free_clusters new clusters at the end of
1993 * the image (which is the current value of cluster; note that cluster
1994 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond
1995 * the image end) */
5fee192e
HR
1996 ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters,
1997 cluster + cluster_count
1998 - contiguous_free_clusters);
1999 if (ret < 0) {
2000 return ret;
c7c0681b 2001 }
c7c0681b
HR
2002 }
2003
2004 /* Go back to the first free cluster */
2005 cluster -= contiguous_free_clusters;
2006 for (i = 0; i < cluster_count; i++) {
7453c96b 2007 s->set_refcount(*refcount_table, cluster + i, 1);
c7c0681b
HR
2008 }
2009
2010 return cluster << s->cluster_bits;
2011}
2012
2013/*
2014 * Creates a new refcount structure based solely on the in-memory information
2015 * given through *refcount_table. All necessary allocations will be reflected
2016 * in that array.
2017 *
2018 * On success, the old refcount structure is leaked (it will be covered by the
2019 * new refcount structure).
2020 */
2021static int rebuild_refcount_structure(BlockDriverState *bs,
2022 BdrvCheckResult *res,
7453c96b 2023 void **refcount_table,
c7c0681b
HR
2024 int64_t *nb_clusters)
2025{
ff99129a 2026 BDRVQcow2State *s = bs->opaque;
c7c0681b
HR
2027 int64_t first_free_cluster = 0, reftable_offset = -1, cluster = 0;
2028 int64_t refblock_offset, refblock_start, refblock_index;
2029 uint32_t reftable_size = 0;
2030 uint64_t *on_disk_reftable = NULL;
7453c96b
HR
2031 void *on_disk_refblock;
2032 int ret = 0;
c7c0681b
HR
2033 struct {
2034 uint64_t reftable_offset;
2035 uint32_t reftable_clusters;
2036 } QEMU_PACKED reftable_offset_and_clusters;
2037
2038 qcow2_cache_empty(bs, s->refcount_block_cache);
2039
2040write_refblocks:
2041 for (; cluster < *nb_clusters; cluster++) {
7453c96b 2042 if (!s->get_refcount(*refcount_table, cluster)) {
c7c0681b
HR
2043 continue;
2044 }
2045
2046 refblock_index = cluster >> s->refcount_block_bits;
2047 refblock_start = refblock_index << s->refcount_block_bits;
2048
2049 /* Don't allocate a cluster in a refblock already written to disk */
2050 if (first_free_cluster < refblock_start) {
2051 first_free_cluster = refblock_start;
2052 }
2053 refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table,
2054 nb_clusters, &first_free_cluster);
2055 if (refblock_offset < 0) {
2056 fprintf(stderr, "ERROR allocating refblock: %s\n",
2057 strerror(-refblock_offset));
2058 res->check_errors++;
2059 ret = refblock_offset;
2060 goto fail;
2061 }
2062
2063 if (reftable_size <= refblock_index) {
2064 uint32_t old_reftable_size = reftable_size;
2065 uint64_t *new_on_disk_reftable;
2066
2067 reftable_size = ROUND_UP((refblock_index + 1) * sizeof(uint64_t),
2068 s->cluster_size) / sizeof(uint64_t);
2069 new_on_disk_reftable = g_try_realloc(on_disk_reftable,
2070 reftable_size *
2071 sizeof(uint64_t));
2072 if (!new_on_disk_reftable) {
2073 res->check_errors++;
2074 ret = -ENOMEM;
2075 goto fail;
2076 }
2077 on_disk_reftable = new_on_disk_reftable;
2078
2079 memset(on_disk_reftable + old_reftable_size, 0,
2080 (reftable_size - old_reftable_size) * sizeof(uint64_t));
2081
2082 /* The offset we have for the reftable is now no longer valid;
2083 * this will leak that range, but we can easily fix that by running
2084 * a leak-fixing check after this rebuild operation */
2085 reftable_offset = -1;
2086 }
2087 on_disk_reftable[refblock_index] = refblock_offset;
2088
2089 /* If this is apparently the last refblock (for now), try to squeeze the
2090 * reftable in */
2091 if (refblock_index == (*nb_clusters - 1) >> s->refcount_block_bits &&
2092 reftable_offset < 0)
2093 {
2094 uint64_t reftable_clusters = size_to_clusters(s, reftable_size *
2095 sizeof(uint64_t));
2096 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2097 refcount_table, nb_clusters,
2098 &first_free_cluster);
2099 if (reftable_offset < 0) {
2100 fprintf(stderr, "ERROR allocating reftable: %s\n",
2101 strerror(-reftable_offset));
2102 res->check_errors++;
2103 ret = reftable_offset;
2104 goto fail;
2105 }
2106 }
2107
2108 ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset,
2109 s->cluster_size);
2110 if (ret < 0) {
2111 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2112 goto fail;
2113 }
2114
7453c96b
HR
2115 /* The size of *refcount_table is always cluster-aligned, therefore the
2116 * write operation will not overflow */
2117 on_disk_refblock = (void *)((char *) *refcount_table +
2118 refblock_index * s->cluster_size);
c7c0681b 2119
18d51c4b 2120 ret = bdrv_write(bs->file, refblock_offset / BDRV_SECTOR_SIZE,
7453c96b 2121 on_disk_refblock, s->cluster_sectors);
c7c0681b
HR
2122 if (ret < 0) {
2123 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2124 goto fail;
2125 }
2126
2127 /* Go to the end of this refblock */
2128 cluster = refblock_start + s->refcount_block_size - 1;
2129 }
2130
2131 if (reftable_offset < 0) {
2132 uint64_t post_refblock_start, reftable_clusters;
2133
2134 post_refblock_start = ROUND_UP(*nb_clusters, s->refcount_block_size);
2135 reftable_clusters = size_to_clusters(s,
2136 reftable_size * sizeof(uint64_t));
2137 /* Not pretty but simple */
2138 if (first_free_cluster < post_refblock_start) {
2139 first_free_cluster = post_refblock_start;
2140 }
2141 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2142 refcount_table, nb_clusters,
2143 &first_free_cluster);
2144 if (reftable_offset < 0) {
2145 fprintf(stderr, "ERROR allocating reftable: %s\n",
2146 strerror(-reftable_offset));
2147 res->check_errors++;
2148 ret = reftable_offset;
2149 goto fail;
2150 }
2151
2152 goto write_refblocks;
2153 }
2154
2155 assert(on_disk_reftable);
2156
2157 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2158 cpu_to_be64s(&on_disk_reftable[refblock_index]);
2159 }
2160
2161 ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset,
2162 reftable_size * sizeof(uint64_t));
2163 if (ret < 0) {
2164 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2165 goto fail;
2166 }
2167
2168 assert(reftable_size < INT_MAX / sizeof(uint64_t));
d9ca2ea2 2169 ret = bdrv_pwrite(bs->file, reftable_offset, on_disk_reftable,
c7c0681b
HR
2170 reftable_size * sizeof(uint64_t));
2171 if (ret < 0) {
2172 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2173 goto fail;
2174 }
2175
2176 /* Enter new reftable into the image header */
f1f7a1dd
PM
2177 reftable_offset_and_clusters.reftable_offset = cpu_to_be64(reftable_offset);
2178 reftable_offset_and_clusters.reftable_clusters =
2179 cpu_to_be32(size_to_clusters(s, reftable_size * sizeof(uint64_t)));
d9ca2ea2
KW
2180 ret = bdrv_pwrite_sync(bs->file,
2181 offsetof(QCowHeader, refcount_table_offset),
c7c0681b
HR
2182 &reftable_offset_and_clusters,
2183 sizeof(reftable_offset_and_clusters));
2184 if (ret < 0) {
2185 fprintf(stderr, "ERROR setting reftable: %s\n", strerror(-ret));
2186 goto fail;
2187 }
2188
2189 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2190 be64_to_cpus(&on_disk_reftable[refblock_index]);
2191 }
2192 s->refcount_table = on_disk_reftable;
2193 s->refcount_table_offset = reftable_offset;
2194 s->refcount_table_size = reftable_size;
7061a078 2195 update_max_refcount_table_index(s);
c7c0681b
HR
2196
2197 return 0;
2198
2199fail:
2200 g_free(on_disk_reftable);
2201 return ret;
2202}
2203
6ca56bf5
HR
2204/*
2205 * Checks an image for refcount consistency.
2206 *
2207 * Returns 0 if no errors are found, the number of errors in case the image is
2208 * detected as corrupted, and -errno when an internal error occurred.
2209 */
2210int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2211 BdrvCheckMode fix)
2212{
ff99129a 2213 BDRVQcow2State *s = bs->opaque;
c7c0681b 2214 BdrvCheckResult pre_compare_res;
6ca56bf5 2215 int64_t size, highest_cluster, nb_clusters;
7453c96b 2216 void *refcount_table = NULL;
f307b255 2217 bool rebuild = false;
6ca56bf5
HR
2218 int ret;
2219
9a4f4c31 2220 size = bdrv_getlength(bs->file->bs);
6ca56bf5
HR
2221 if (size < 0) {
2222 res->check_errors++;
2223 return size;
2224 }
2225
2226 nb_clusters = size_to_clusters(s, size);
2227 if (nb_clusters > INT_MAX) {
2228 res->check_errors++;
2229 return -EFBIG;
2230 }
2231
2232 res->bfi.total_clusters =
2233 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE);
2234
f307b255
HR
2235 ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table,
2236 &nb_clusters);
6ca56bf5
HR
2237 if (ret < 0) {
2238 goto fail;
2239 }
2240
c7c0681b
HR
2241 /* In case we don't need to rebuild the refcount structure (but want to fix
2242 * something), this function is immediately called again, in which case the
2243 * result should be ignored */
2244 pre_compare_res = *res;
2245 compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table,
6ca56bf5 2246 nb_clusters);
f7d0fe02 2247
c7c0681b 2248 if (rebuild && (fix & BDRV_FIX_ERRORS)) {
791230d8
HR
2249 BdrvCheckResult old_res = *res;
2250 int fresh_leaks = 0;
2251
c7c0681b
HR
2252 fprintf(stderr, "Rebuilding refcount structure\n");
2253 ret = rebuild_refcount_structure(bs, res, &refcount_table,
2254 &nb_clusters);
2255 if (ret < 0) {
2256 goto fail;
2257 }
791230d8
HR
2258
2259 res->corruptions = 0;
2260 res->leaks = 0;
2261
2262 /* Because the old reftable has been exchanged for a new one the
2263 * references have to be recalculated */
2264 rebuild = false;
7453c96b 2265 memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters));
791230d8
HR
2266 ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table,
2267 &nb_clusters);
2268 if (ret < 0) {
2269 goto fail;
2270 }
2271
2272 if (fix & BDRV_FIX_LEAKS) {
2273 /* The old refcount structures are now leaked, fix it; the result
2274 * can be ignored, aside from leaks which were introduced by
2275 * rebuild_refcount_structure() that could not be fixed */
2276 BdrvCheckResult saved_res = *res;
2277 *res = (BdrvCheckResult){ 0 };
2278
2279 compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild,
2280 &highest_cluster, refcount_table, nb_clusters);
2281 if (rebuild) {
2282 fprintf(stderr, "ERROR rebuilt refcount structure is still "
2283 "broken\n");
2284 }
2285
2286 /* Any leaks accounted for here were introduced by
2287 * rebuild_refcount_structure() because that function has created a
2288 * new refcount structure from scratch */
2289 fresh_leaks = res->leaks;
2290 *res = saved_res;
2291 }
2292
2293 if (res->corruptions < old_res.corruptions) {
2294 res->corruptions_fixed += old_res.corruptions - res->corruptions;
2295 }
2296 if (res->leaks < old_res.leaks) {
2297 res->leaks_fixed += old_res.leaks - res->leaks;
2298 }
2299 res->leaks += fresh_leaks;
c7c0681b
HR
2300 } else if (fix) {
2301 if (rebuild) {
2302 fprintf(stderr, "ERROR need to rebuild refcount structures\n");
2303 res->check_errors++;
2304 ret = -EIO;
2305 goto fail;
2306 }
2307
2308 if (res->leaks || res->corruptions) {
2309 *res = pre_compare_res;
2310 compare_refcounts(bs, res, fix, &rebuild, &highest_cluster,
2311 refcount_table, nb_clusters);
2312 }
f307b255
HR
2313 }
2314
4f6ed88c 2315 /* check OFLAG_COPIED */
e23e400e 2316 ret = check_oflag_copied(bs, res, fix);
4f6ed88c
HR
2317 if (ret < 0) {
2318 goto fail;
2319 }
2320
c6bb9ad1 2321 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
80fa3341
KW
2322 ret = 0;
2323
2324fail:
7267c094 2325 g_free(refcount_table);
f7d0fe02 2326
80fa3341 2327 return ret;
f7d0fe02
KW
2328}
2329
a40f1c2a
HR
2330#define overlaps_with(ofs, sz) \
2331 ranges_overlap(offset, size, ofs, sz)
2332
2333/*
2334 * Checks if the given offset into the image file is actually free to use by
2335 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
2336 * i.e. a sanity check without relying on the refcount tables.
2337 *
231bb267
HR
2338 * The ign parameter specifies what checks not to perform (being a bitmask of
2339 * QCow2MetadataOverlap values), i.e., what sections to ignore.
a40f1c2a
HR
2340 *
2341 * Returns:
2342 * - 0 if writing to this offset will not affect the mentioned metadata
2343 * - a positive QCow2MetadataOverlap value indicating one overlapping section
2344 * - a negative value (-errno) indicating an error while performing a check,
2345 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
2346 */
231bb267 2347int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset,
a40f1c2a
HR
2348 int64_t size)
2349{
ff99129a 2350 BDRVQcow2State *s = bs->opaque;
3e355390 2351 int chk = s->overlap_check & ~ign;
a40f1c2a
HR
2352 int i, j;
2353
2354 if (!size) {
2355 return 0;
2356 }
2357
2358 if (chk & QCOW2_OL_MAIN_HEADER) {
2359 if (offset < s->cluster_size) {
2360 return QCOW2_OL_MAIN_HEADER;
2361 }
2362 }
2363
2364 /* align range to test to cluster boundaries */
2365 size = align_offset(offset_into_cluster(s, offset) + size, s->cluster_size);
2366 offset = start_of_cluster(s, offset);
2367
2368 if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) {
2369 if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) {
2370 return QCOW2_OL_ACTIVE_L1;
2371 }
2372 }
2373
2374 if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) {
2375 if (overlaps_with(s->refcount_table_offset,
2376 s->refcount_table_size * sizeof(uint64_t))) {
2377 return QCOW2_OL_REFCOUNT_TABLE;
2378 }
2379 }
2380
2381 if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) {
2382 if (overlaps_with(s->snapshots_offset, s->snapshots_size)) {
2383 return QCOW2_OL_SNAPSHOT_TABLE;
2384 }
2385 }
2386
2387 if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) {
2388 for (i = 0; i < s->nb_snapshots; i++) {
2389 if (s->snapshots[i].l1_size &&
2390 overlaps_with(s->snapshots[i].l1_table_offset,
2391 s->snapshots[i].l1_size * sizeof(uint64_t))) {
2392 return QCOW2_OL_INACTIVE_L1;
2393 }
2394 }
2395 }
2396
2397 if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) {
2398 for (i = 0; i < s->l1_size; i++) {
2399 if ((s->l1_table[i] & L1E_OFFSET_MASK) &&
2400 overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK,
2401 s->cluster_size)) {
2402 return QCOW2_OL_ACTIVE_L2;
2403 }
2404 }
2405 }
2406
2407 if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) {
7061a078
AG
2408 unsigned last_entry = s->max_refcount_table_index;
2409 assert(last_entry < s->refcount_table_size);
2410 assert(last_entry + 1 == s->refcount_table_size ||
2411 (s->refcount_table[last_entry + 1] & REFT_OFFSET_MASK) == 0);
2412 for (i = 0; i <= last_entry; i++) {
a40f1c2a
HR
2413 if ((s->refcount_table[i] & REFT_OFFSET_MASK) &&
2414 overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK,
2415 s->cluster_size)) {
2416 return QCOW2_OL_REFCOUNT_BLOCK;
2417 }
2418 }
2419 }
2420
2421 if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) {
2422 for (i = 0; i < s->nb_snapshots; i++) {
2423 uint64_t l1_ofs = s->snapshots[i].l1_table_offset;
2424 uint32_t l1_sz = s->snapshots[i].l1_size;
998b959c 2425 uint64_t l1_sz2 = l1_sz * sizeof(uint64_t);
de82815d 2426 uint64_t *l1 = g_try_malloc(l1_sz2);
a40f1c2a
HR
2427 int ret;
2428
de82815d
KW
2429 if (l1_sz2 && l1 == NULL) {
2430 return -ENOMEM;
2431 }
2432
cf2ab8fc 2433 ret = bdrv_pread(bs->file, l1_ofs, l1, l1_sz2);
a40f1c2a
HR
2434 if (ret < 0) {
2435 g_free(l1);
2436 return ret;
2437 }
2438
2439 for (j = 0; j < l1_sz; j++) {
1e242b55
HR
2440 uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK;
2441 if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) {
a40f1c2a
HR
2442 g_free(l1);
2443 return QCOW2_OL_INACTIVE_L2;
2444 }
2445 }
2446
2447 g_free(l1);
2448 }
2449 }
2450
2451 return 0;
2452}
2453
2454static const char *metadata_ol_names[] = {
2455 [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header",
2456 [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table",
2457 [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table",
2458 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table",
2459 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block",
2460 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table",
2461 [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table",
2462 [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table",
2463};
2464
2465/*
2466 * First performs a check for metadata overlaps (through
2467 * qcow2_check_metadata_overlap); if that fails with a negative value (error
2468 * while performing a check), that value is returned. If an impending overlap
2469 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
2470 * and -EIO returned.
2471 *
2472 * Returns 0 if there were neither overlaps nor errors while checking for
2473 * overlaps; or a negative value (-errno) on error.
2474 */
231bb267 2475int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset,
a40f1c2a
HR
2476 int64_t size)
2477{
231bb267 2478 int ret = qcow2_check_metadata_overlap(bs, ign, offset, size);
a40f1c2a
HR
2479
2480 if (ret < 0) {
2481 return ret;
2482 } else if (ret > 0) {
786a4ea8 2483 int metadata_ol_bitnr = ctz32(ret);
a40f1c2a
HR
2484 assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR);
2485
adb43552
HR
2486 qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid "
2487 "write on metadata (overlaps with %s)",
2488 metadata_ol_names[metadata_ol_bitnr]);
a40f1c2a
HR
2489 return -EIO;
2490 }
2491
2492 return 0;
2493}
791c9a00
HR
2494
2495/* A pointer to a function of this type is given to walk_over_reftable(). That
2496 * function will create refblocks and pass them to a RefblockFinishOp once they
2497 * are completed (@refblock). @refblock_empty is set if the refblock is
2498 * completely empty.
2499 *
2500 * Along with the refblock, a corresponding reftable entry is passed, in the
2501 * reftable @reftable (which may be reallocated) at @reftable_index.
2502 *
2503 * @allocated should be set to true if a new cluster has been allocated.
2504 */
2505typedef int (RefblockFinishOp)(BlockDriverState *bs, uint64_t **reftable,
2506 uint64_t reftable_index, uint64_t *reftable_size,
2507 void *refblock, bool refblock_empty,
2508 bool *allocated, Error **errp);
2509
2510/**
2511 * This "operation" for walk_over_reftable() allocates the refblock on disk (if
2512 * it is not empty) and inserts its offset into the new reftable. The size of
2513 * this new reftable is increased as required.
2514 */
2515static int alloc_refblock(BlockDriverState *bs, uint64_t **reftable,
2516 uint64_t reftable_index, uint64_t *reftable_size,
2517 void *refblock, bool refblock_empty, bool *allocated,
2518 Error **errp)
2519{
2520 BDRVQcow2State *s = bs->opaque;
2521 int64_t offset;
2522
2523 if (!refblock_empty && reftable_index >= *reftable_size) {
2524 uint64_t *new_reftable;
2525 uint64_t new_reftable_size;
2526
2527 new_reftable_size = ROUND_UP(reftable_index + 1,
2528 s->cluster_size / sizeof(uint64_t));
2529 if (new_reftable_size > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) {
2530 error_setg(errp,
2531 "This operation would make the refcount table grow "
2532 "beyond the maximum size supported by QEMU, aborting");
2533 return -ENOTSUP;
2534 }
2535
2536 new_reftable = g_try_realloc(*reftable, new_reftable_size *
2537 sizeof(uint64_t));
2538 if (!new_reftable) {
2539 error_setg(errp, "Failed to increase reftable buffer size");
2540 return -ENOMEM;
2541 }
2542
2543 memset(new_reftable + *reftable_size, 0,
2544 (new_reftable_size - *reftable_size) * sizeof(uint64_t));
2545
2546 *reftable = new_reftable;
2547 *reftable_size = new_reftable_size;
2548 }
2549
2550 if (!refblock_empty && !(*reftable)[reftable_index]) {
2551 offset = qcow2_alloc_clusters(bs, s->cluster_size);
2552 if (offset < 0) {
2553 error_setg_errno(errp, -offset, "Failed to allocate refblock");
2554 return offset;
2555 }
2556 (*reftable)[reftable_index] = offset;
2557 *allocated = true;
2558 }
2559
2560 return 0;
2561}
2562
2563/**
2564 * This "operation" for walk_over_reftable() writes the refblock to disk at the
2565 * offset specified by the new reftable's entry. It does not modify the new
2566 * reftable or change any refcounts.
2567 */
2568static int flush_refblock(BlockDriverState *bs, uint64_t **reftable,
2569 uint64_t reftable_index, uint64_t *reftable_size,
2570 void *refblock, bool refblock_empty, bool *allocated,
2571 Error **errp)
2572{
2573 BDRVQcow2State *s = bs->opaque;
2574 int64_t offset;
2575 int ret;
2576
2577 if (reftable_index < *reftable_size && (*reftable)[reftable_index]) {
2578 offset = (*reftable)[reftable_index];
2579
2580 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
2581 if (ret < 0) {
2582 error_setg_errno(errp, -ret, "Overlap check failed");
2583 return ret;
2584 }
2585
d9ca2ea2 2586 ret = bdrv_pwrite(bs->file, offset, refblock, s->cluster_size);
791c9a00
HR
2587 if (ret < 0) {
2588 error_setg_errno(errp, -ret, "Failed to write refblock");
2589 return ret;
2590 }
2591 } else {
2592 assert(refblock_empty);
2593 }
2594
2595 return 0;
2596}
2597
2598/**
2599 * This function walks over the existing reftable and every referenced refblock;
2600 * if @new_set_refcount is non-NULL, it is called for every refcount entry to
2601 * create an equal new entry in the passed @new_refblock. Once that
2602 * @new_refblock is completely filled, @operation will be called.
2603 *
2604 * @status_cb and @cb_opaque are used for the amend operation's status callback.
2605 * @index is the index of the walk_over_reftable() calls and @total is the total
2606 * number of walk_over_reftable() calls per amend operation. Both are used for
2607 * calculating the parameters for the status callback.
2608 *
2609 * @allocated is set to true if a new cluster has been allocated.
2610 */
2611static int walk_over_reftable(BlockDriverState *bs, uint64_t **new_reftable,
2612 uint64_t *new_reftable_index,
2613 uint64_t *new_reftable_size,
2614 void *new_refblock, int new_refblock_size,
2615 int new_refcount_bits,
2616 RefblockFinishOp *operation, bool *allocated,
2617 Qcow2SetRefcountFunc *new_set_refcount,
2618 BlockDriverAmendStatusCB *status_cb,
2619 void *cb_opaque, int index, int total,
2620 Error **errp)
2621{
2622 BDRVQcow2State *s = bs->opaque;
2623 uint64_t reftable_index;
2624 bool new_refblock_empty = true;
2625 int refblock_index;
2626 int new_refblock_index = 0;
2627 int ret;
2628
2629 for (reftable_index = 0; reftable_index < s->refcount_table_size;
2630 reftable_index++)
2631 {
2632 uint64_t refblock_offset = s->refcount_table[reftable_index]
2633 & REFT_OFFSET_MASK;
2634
2635 status_cb(bs, (uint64_t)index * s->refcount_table_size + reftable_index,
2636 (uint64_t)total * s->refcount_table_size, cb_opaque);
2637
2638 if (refblock_offset) {
2639 void *refblock;
2640
2641 if (offset_into_cluster(s, refblock_offset)) {
2642 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
2643 PRIx64 " unaligned (reftable index: %#"
2644 PRIx64 ")", refblock_offset,
2645 reftable_index);
2646 error_setg(errp,
2647 "Image is corrupt (unaligned refblock offset)");
2648 return -EIO;
2649 }
2650
2651 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offset,
2652 &refblock);
2653 if (ret < 0) {
2654 error_setg_errno(errp, -ret, "Failed to retrieve refblock");
2655 return ret;
2656 }
2657
2658 for (refblock_index = 0; refblock_index < s->refcount_block_size;
2659 refblock_index++)
2660 {
2661 uint64_t refcount;
2662
2663 if (new_refblock_index >= new_refblock_size) {
2664 /* new_refblock is now complete */
2665 ret = operation(bs, new_reftable, *new_reftable_index,
2666 new_reftable_size, new_refblock,
2667 new_refblock_empty, allocated, errp);
2668 if (ret < 0) {
2669 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2670 return ret;
2671 }
2672
2673 (*new_reftable_index)++;
2674 new_refblock_index = 0;
2675 new_refblock_empty = true;
2676 }
2677
2678 refcount = s->get_refcount(refblock, refblock_index);
2679 if (new_refcount_bits < 64 && refcount >> new_refcount_bits) {
2680 uint64_t offset;
2681
2682 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2683
2684 offset = ((reftable_index << s->refcount_block_bits)
2685 + refblock_index) << s->cluster_bits;
2686
2687 error_setg(errp, "Cannot decrease refcount entry width to "
2688 "%i bits: Cluster at offset %#" PRIx64 " has a "
2689 "refcount of %" PRIu64, new_refcount_bits,
2690 offset, refcount);
2691 return -EINVAL;
2692 }
2693
2694 if (new_set_refcount) {
2695 new_set_refcount(new_refblock, new_refblock_index++,
2696 refcount);
2697 } else {
2698 new_refblock_index++;
2699 }
2700 new_refblock_empty = new_refblock_empty && refcount == 0;
2701 }
2702
2703 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2704 } else {
2705 /* No refblock means every refcount is 0 */
2706 for (refblock_index = 0; refblock_index < s->refcount_block_size;
2707 refblock_index++)
2708 {
2709 if (new_refblock_index >= new_refblock_size) {
2710 /* new_refblock is now complete */
2711 ret = operation(bs, new_reftable, *new_reftable_index,
2712 new_reftable_size, new_refblock,
2713 new_refblock_empty, allocated, errp);
2714 if (ret < 0) {
2715 return ret;
2716 }
2717
2718 (*new_reftable_index)++;
2719 new_refblock_index = 0;
2720 new_refblock_empty = true;
2721 }
2722
2723 if (new_set_refcount) {
2724 new_set_refcount(new_refblock, new_refblock_index++, 0);
2725 } else {
2726 new_refblock_index++;
2727 }
2728 }
2729 }
2730 }
2731
2732 if (new_refblock_index > 0) {
2733 /* Complete the potentially existing partially filled final refblock */
2734 if (new_set_refcount) {
2735 for (; new_refblock_index < new_refblock_size;
2736 new_refblock_index++)
2737 {
2738 new_set_refcount(new_refblock, new_refblock_index, 0);
2739 }
2740 }
2741
2742 ret = operation(bs, new_reftable, *new_reftable_index,
2743 new_reftable_size, new_refblock, new_refblock_empty,
2744 allocated, errp);
2745 if (ret < 0) {
2746 return ret;
2747 }
2748
2749 (*new_reftable_index)++;
2750 }
2751
2752 status_cb(bs, (uint64_t)(index + 1) * s->refcount_table_size,
2753 (uint64_t)total * s->refcount_table_size, cb_opaque);
2754
2755 return 0;
2756}
2757
2758int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order,
2759 BlockDriverAmendStatusCB *status_cb,
2760 void *cb_opaque, Error **errp)
2761{
2762 BDRVQcow2State *s = bs->opaque;
2763 Qcow2GetRefcountFunc *new_get_refcount;
2764 Qcow2SetRefcountFunc *new_set_refcount;
2765 void *new_refblock = qemu_blockalign(bs->file->bs, s->cluster_size);
2766 uint64_t *new_reftable = NULL, new_reftable_size = 0;
2767 uint64_t *old_reftable, old_reftable_size, old_reftable_offset;
2768 uint64_t new_reftable_index = 0;
2769 uint64_t i;
2770 int64_t new_reftable_offset = 0, allocated_reftable_size = 0;
2771 int new_refblock_size, new_refcount_bits = 1 << refcount_order;
2772 int old_refcount_order;
2773 int walk_index = 0;
2774 int ret;
2775 bool new_allocation;
2776
2777 assert(s->qcow_version >= 3);
2778 assert(refcount_order >= 0 && refcount_order <= 6);
2779
2780 /* see qcow2_open() */
2781 new_refblock_size = 1 << (s->cluster_bits - (refcount_order - 3));
2782
2783 new_get_refcount = get_refcount_funcs[refcount_order];
2784 new_set_refcount = set_refcount_funcs[refcount_order];
2785
2786
2787 do {
2788 int total_walks;
2789
2790 new_allocation = false;
2791
2792 /* At least we have to do this walk and the one which writes the
2793 * refblocks; also, at least we have to do this loop here at least
2794 * twice (normally), first to do the allocations, and second to
2795 * determine that everything is correctly allocated, this then makes
2796 * three walks in total */
2797 total_walks = MAX(walk_index + 2, 3);
2798
2799 /* First, allocate the structures so they are present in the refcount
2800 * structures */
2801 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
2802 &new_reftable_size, NULL, new_refblock_size,
2803 new_refcount_bits, &alloc_refblock,
2804 &new_allocation, NULL, status_cb, cb_opaque,
2805 walk_index++, total_walks, errp);
2806 if (ret < 0) {
2807 goto done;
2808 }
2809
2810 new_reftable_index = 0;
2811
2812 if (new_allocation) {
2813 if (new_reftable_offset) {
2814 qcow2_free_clusters(bs, new_reftable_offset,
2815 allocated_reftable_size * sizeof(uint64_t),
2816 QCOW2_DISCARD_NEVER);
2817 }
2818
2819 new_reftable_offset = qcow2_alloc_clusters(bs, new_reftable_size *
2820 sizeof(uint64_t));
2821 if (new_reftable_offset < 0) {
2822 error_setg_errno(errp, -new_reftable_offset,
2823 "Failed to allocate the new reftable");
2824 ret = new_reftable_offset;
2825 goto done;
2826 }
2827 allocated_reftable_size = new_reftable_size;
2828 }
2829 } while (new_allocation);
2830
2831 /* Second, write the new refblocks */
2832 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
2833 &new_reftable_size, new_refblock,
2834 new_refblock_size, new_refcount_bits,
2835 &flush_refblock, &new_allocation, new_set_refcount,
2836 status_cb, cb_opaque, walk_index, walk_index + 1,
2837 errp);
2838 if (ret < 0) {
2839 goto done;
2840 }
2841 assert(!new_allocation);
2842
2843
2844 /* Write the new reftable */
2845 ret = qcow2_pre_write_overlap_check(bs, 0, new_reftable_offset,
2846 new_reftable_size * sizeof(uint64_t));
2847 if (ret < 0) {
2848 error_setg_errno(errp, -ret, "Overlap check failed");
2849 goto done;
2850 }
2851
2852 for (i = 0; i < new_reftable_size; i++) {
2853 cpu_to_be64s(&new_reftable[i]);
2854 }
2855
d9ca2ea2 2856 ret = bdrv_pwrite(bs->file, new_reftable_offset, new_reftable,
791c9a00
HR
2857 new_reftable_size * sizeof(uint64_t));
2858
2859 for (i = 0; i < new_reftable_size; i++) {
2860 be64_to_cpus(&new_reftable[i]);
2861 }
2862
2863 if (ret < 0) {
2864 error_setg_errno(errp, -ret, "Failed to write the new reftable");
2865 goto done;
2866 }
2867
2868
2869 /* Empty the refcount cache */
2870 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2871 if (ret < 0) {
2872 error_setg_errno(errp, -ret, "Failed to flush the refblock cache");
2873 goto done;
2874 }
2875
2876 /* Update the image header to point to the new reftable; this only updates
2877 * the fields which are relevant to qcow2_update_header(); other fields
2878 * such as s->refcount_table or s->refcount_bits stay stale for now
2879 * (because we have to restore everything if qcow2_update_header() fails) */
2880 old_refcount_order = s->refcount_order;
2881 old_reftable_size = s->refcount_table_size;
2882 old_reftable_offset = s->refcount_table_offset;
2883
2884 s->refcount_order = refcount_order;
2885 s->refcount_table_size = new_reftable_size;
2886 s->refcount_table_offset = new_reftable_offset;
2887
2888 ret = qcow2_update_header(bs);
2889 if (ret < 0) {
2890 s->refcount_order = old_refcount_order;
2891 s->refcount_table_size = old_reftable_size;
2892 s->refcount_table_offset = old_reftable_offset;
2893 error_setg_errno(errp, -ret, "Failed to update the qcow2 header");
2894 goto done;
2895 }
2896
2897 /* Now update the rest of the in-memory information */
2898 old_reftable = s->refcount_table;
2899 s->refcount_table = new_reftable;
7061a078 2900 update_max_refcount_table_index(s);
791c9a00
HR
2901
2902 s->refcount_bits = 1 << refcount_order;
2903 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
2904 s->refcount_max += s->refcount_max - 1;
2905
2906 s->refcount_block_bits = s->cluster_bits - (refcount_order - 3);
2907 s->refcount_block_size = 1 << s->refcount_block_bits;
2908
2909 s->get_refcount = new_get_refcount;
2910 s->set_refcount = new_set_refcount;
2911
2912 /* For cleaning up all old refblocks and the old reftable below the "done"
2913 * label */
2914 new_reftable = old_reftable;
2915 new_reftable_size = old_reftable_size;
2916 new_reftable_offset = old_reftable_offset;
2917
2918done:
2919 if (new_reftable) {
2920 /* On success, new_reftable actually points to the old reftable (and
2921 * new_reftable_size is the old reftable's size); but that is just
2922 * fine */
2923 for (i = 0; i < new_reftable_size; i++) {
2924 uint64_t offset = new_reftable[i] & REFT_OFFSET_MASK;
2925 if (offset) {
2926 qcow2_free_clusters(bs, offset, s->cluster_size,
2927 QCOW2_DISCARD_OTHER);
2928 }
2929 }
2930 g_free(new_reftable);
2931
2932 if (new_reftable_offset > 0) {
2933 qcow2_free_clusters(bs, new_reftable_offset,
2934 new_reftable_size * sizeof(uint64_t),
2935 QCOW2_DISCARD_OTHER);
2936 }
2937 }
2938
2939 qemu_vfree(new_refblock);
2940 return ret;
2941}
This page took 0.930089 seconds and 4 git commands to generate.