]> git.proxmox.com Git - mirror_qemu.git/blob - block/qcow2-refcount.c
Merge tag 'linux-user-for-8.0-pull-request' of https://gitlab.com/laurent_vivier...
[mirror_qemu.git] / block / qcow2-refcount.c
1 /*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include "block/block-io.h"
27 #include "qapi/error.h"
28 #include "qcow2.h"
29 #include "qemu/range.h"
30 #include "qemu/bswap.h"
31 #include "qemu/cutils.h"
32 #include "qemu/memalign.h"
33 #include "trace.h"
34
35 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size,
36 uint64_t max);
37
38 G_GNUC_WARN_UNUSED_RESULT
39 static int update_refcount(BlockDriverState *bs,
40 int64_t offset, int64_t length, uint64_t addend,
41 bool decrease, enum qcow2_discard_type type);
42
43 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index);
44 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index);
45 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index);
46 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index);
47 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index);
48 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index);
49 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index);
50
51 static void set_refcount_ro0(void *refcount_array, uint64_t index,
52 uint64_t value);
53 static void set_refcount_ro1(void *refcount_array, uint64_t index,
54 uint64_t value);
55 static void set_refcount_ro2(void *refcount_array, uint64_t index,
56 uint64_t value);
57 static void set_refcount_ro3(void *refcount_array, uint64_t index,
58 uint64_t value);
59 static void set_refcount_ro4(void *refcount_array, uint64_t index,
60 uint64_t value);
61 static void set_refcount_ro5(void *refcount_array, uint64_t index,
62 uint64_t value);
63 static void set_refcount_ro6(void *refcount_array, uint64_t index,
64 uint64_t value);
65
66
67 static Qcow2GetRefcountFunc *const get_refcount_funcs[] = {
68 &get_refcount_ro0,
69 &get_refcount_ro1,
70 &get_refcount_ro2,
71 &get_refcount_ro3,
72 &get_refcount_ro4,
73 &get_refcount_ro5,
74 &get_refcount_ro6
75 };
76
77 static Qcow2SetRefcountFunc *const set_refcount_funcs[] = {
78 &set_refcount_ro0,
79 &set_refcount_ro1,
80 &set_refcount_ro2,
81 &set_refcount_ro3,
82 &set_refcount_ro4,
83 &set_refcount_ro5,
84 &set_refcount_ro6
85 };
86
87
88 /*********************************************************/
89 /* refcount handling */
90
91 static void update_max_refcount_table_index(BDRVQcow2State *s)
92 {
93 unsigned i = s->refcount_table_size - 1;
94 while (i > 0 && (s->refcount_table[i] & REFT_OFFSET_MASK) == 0) {
95 i--;
96 }
97 /* Set s->max_refcount_table_index to the index of the last used entry */
98 s->max_refcount_table_index = i;
99 }
100
101 int coroutine_fn qcow2_refcount_init(BlockDriverState *bs)
102 {
103 BDRVQcow2State *s = bs->opaque;
104 unsigned int refcount_table_size2, i;
105 int ret;
106
107 assert(s->refcount_order >= 0 && s->refcount_order <= 6);
108
109 s->get_refcount = get_refcount_funcs[s->refcount_order];
110 s->set_refcount = set_refcount_funcs[s->refcount_order];
111
112 assert(s->refcount_table_size <= INT_MAX / REFTABLE_ENTRY_SIZE);
113 refcount_table_size2 = s->refcount_table_size * REFTABLE_ENTRY_SIZE;
114 s->refcount_table = g_try_malloc(refcount_table_size2);
115
116 if (s->refcount_table_size > 0) {
117 if (s->refcount_table == NULL) {
118 ret = -ENOMEM;
119 goto fail;
120 }
121 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
122 ret = bdrv_co_pread(bs->file, s->refcount_table_offset,
123 refcount_table_size2, s->refcount_table, 0);
124 if (ret < 0) {
125 goto fail;
126 }
127 for(i = 0; i < s->refcount_table_size; i++)
128 be64_to_cpus(&s->refcount_table[i]);
129 update_max_refcount_table_index(s);
130 }
131 return 0;
132 fail:
133 return ret;
134 }
135
136 void qcow2_refcount_close(BlockDriverState *bs)
137 {
138 BDRVQcow2State *s = bs->opaque;
139 g_free(s->refcount_table);
140 }
141
142
143 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index)
144 {
145 return (((const uint8_t *)refcount_array)[index / 8] >> (index % 8)) & 0x1;
146 }
147
148 static void set_refcount_ro0(void *refcount_array, uint64_t index,
149 uint64_t value)
150 {
151 assert(!(value >> 1));
152 ((uint8_t *)refcount_array)[index / 8] &= ~(0x1 << (index % 8));
153 ((uint8_t *)refcount_array)[index / 8] |= value << (index % 8);
154 }
155
156 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index)
157 {
158 return (((const uint8_t *)refcount_array)[index / 4] >> (2 * (index % 4)))
159 & 0x3;
160 }
161
162 static void set_refcount_ro1(void *refcount_array, uint64_t index,
163 uint64_t value)
164 {
165 assert(!(value >> 2));
166 ((uint8_t *)refcount_array)[index / 4] &= ~(0x3 << (2 * (index % 4)));
167 ((uint8_t *)refcount_array)[index / 4] |= value << (2 * (index % 4));
168 }
169
170 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index)
171 {
172 return (((const uint8_t *)refcount_array)[index / 2] >> (4 * (index % 2)))
173 & 0xf;
174 }
175
176 static void set_refcount_ro2(void *refcount_array, uint64_t index,
177 uint64_t value)
178 {
179 assert(!(value >> 4));
180 ((uint8_t *)refcount_array)[index / 2] &= ~(0xf << (4 * (index % 2)));
181 ((uint8_t *)refcount_array)[index / 2] |= value << (4 * (index % 2));
182 }
183
184 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index)
185 {
186 return ((const uint8_t *)refcount_array)[index];
187 }
188
189 static void set_refcount_ro3(void *refcount_array, uint64_t index,
190 uint64_t value)
191 {
192 assert(!(value >> 8));
193 ((uint8_t *)refcount_array)[index] = value;
194 }
195
196 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index)
197 {
198 return be16_to_cpu(((const uint16_t *)refcount_array)[index]);
199 }
200
201 static void set_refcount_ro4(void *refcount_array, uint64_t index,
202 uint64_t value)
203 {
204 assert(!(value >> 16));
205 ((uint16_t *)refcount_array)[index] = cpu_to_be16(value);
206 }
207
208 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index)
209 {
210 return be32_to_cpu(((const uint32_t *)refcount_array)[index]);
211 }
212
213 static void set_refcount_ro5(void *refcount_array, uint64_t index,
214 uint64_t value)
215 {
216 assert(!(value >> 32));
217 ((uint32_t *)refcount_array)[index] = cpu_to_be32(value);
218 }
219
220 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index)
221 {
222 return be64_to_cpu(((const uint64_t *)refcount_array)[index]);
223 }
224
225 static void set_refcount_ro6(void *refcount_array, uint64_t index,
226 uint64_t value)
227 {
228 ((uint64_t *)refcount_array)[index] = cpu_to_be64(value);
229 }
230
231
232 static int load_refcount_block(BlockDriverState *bs,
233 int64_t refcount_block_offset,
234 void **refcount_block)
235 {
236 BDRVQcow2State *s = bs->opaque;
237
238 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
239 return qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
240 refcount_block);
241 }
242
243 /*
244 * Retrieves the refcount of the cluster given by its index and stores it in
245 * *refcount. Returns 0 on success and -errno on failure.
246 */
247 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index,
248 uint64_t *refcount)
249 {
250 BDRVQcow2State *s = bs->opaque;
251 uint64_t refcount_table_index, block_index;
252 int64_t refcount_block_offset;
253 int ret;
254 void *refcount_block;
255
256 refcount_table_index = cluster_index >> s->refcount_block_bits;
257 if (refcount_table_index >= s->refcount_table_size) {
258 *refcount = 0;
259 return 0;
260 }
261 refcount_block_offset =
262 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
263 if (!refcount_block_offset) {
264 *refcount = 0;
265 return 0;
266 }
267
268 if (offset_into_cluster(s, refcount_block_offset)) {
269 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" PRIx64
270 " unaligned (reftable index: %#" PRIx64 ")",
271 refcount_block_offset, refcount_table_index);
272 return -EIO;
273 }
274
275 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
276 &refcount_block);
277 if (ret < 0) {
278 return ret;
279 }
280
281 block_index = cluster_index & (s->refcount_block_size - 1);
282 *refcount = s->get_refcount(refcount_block, block_index);
283
284 qcow2_cache_put(s->refcount_block_cache, &refcount_block);
285
286 return 0;
287 }
288
289 /* Checks if two offsets are described by the same refcount block */
290 static int in_same_refcount_block(BDRVQcow2State *s, uint64_t offset_a,
291 uint64_t offset_b)
292 {
293 uint64_t block_a = offset_a >> (s->cluster_bits + s->refcount_block_bits);
294 uint64_t block_b = offset_b >> (s->cluster_bits + s->refcount_block_bits);
295
296 return (block_a == block_b);
297 }
298
299 /*
300 * Loads a refcount block. If it doesn't exist yet, it is allocated first
301 * (including growing the refcount table if needed).
302 *
303 * Returns 0 on success or -errno in error case
304 */
305 static int alloc_refcount_block(BlockDriverState *bs,
306 int64_t cluster_index, void **refcount_block)
307 {
308 BDRVQcow2State *s = bs->opaque;
309 unsigned int refcount_table_index;
310 int64_t ret;
311
312 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
313
314 /* Find the refcount block for the given cluster */
315 refcount_table_index = cluster_index >> s->refcount_block_bits;
316
317 if (refcount_table_index < s->refcount_table_size) {
318
319 uint64_t refcount_block_offset =
320 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
321
322 /* If it's already there, we're done */
323 if (refcount_block_offset) {
324 if (offset_into_cluster(s, refcount_block_offset)) {
325 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
326 PRIx64 " unaligned (reftable index: "
327 "%#x)", refcount_block_offset,
328 refcount_table_index);
329 return -EIO;
330 }
331
332 return load_refcount_block(bs, refcount_block_offset,
333 refcount_block);
334 }
335 }
336
337 /*
338 * If we came here, we need to allocate something. Something is at least
339 * a cluster for the new refcount block. It may also include a new refcount
340 * table if the old refcount table is too small.
341 *
342 * Note that allocating clusters here needs some special care:
343 *
344 * - We can't use the normal qcow2_alloc_clusters(), it would try to
345 * increase the refcount and very likely we would end up with an endless
346 * recursion. Instead we must place the refcount blocks in a way that
347 * they can describe them themselves.
348 *
349 * - We need to consider that at this point we are inside update_refcounts
350 * and potentially doing an initial refcount increase. This means that
351 * some clusters have already been allocated by the caller, but their
352 * refcount isn't accurate yet. If we allocate clusters for metadata, we
353 * need to return -EAGAIN to signal the caller that it needs to restart
354 * the search for free clusters.
355 *
356 * - alloc_clusters_noref and qcow2_free_clusters may load a different
357 * refcount block into the cache
358 */
359
360 *refcount_block = NULL;
361
362 /* We write to the refcount table, so we might depend on L2 tables */
363 ret = qcow2_cache_flush(bs, s->l2_table_cache);
364 if (ret < 0) {
365 return ret;
366 }
367
368 /* Allocate the refcount block itself and mark it as used */
369 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size, INT64_MAX);
370 if (new_block < 0) {
371 return new_block;
372 }
373
374 /* The offset must fit in the offset field of the refcount table entry */
375 assert((new_block & REFT_OFFSET_MASK) == new_block);
376
377 /* If we're allocating the block at offset 0 then something is wrong */
378 if (new_block == 0) {
379 qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid "
380 "allocation of refcount block at offset 0");
381 return -EIO;
382 }
383
384 #ifdef DEBUG_ALLOC2
385 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
386 " at %" PRIx64 "\n",
387 refcount_table_index, cluster_index << s->cluster_bits, new_block);
388 #endif
389
390 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
391 /* Zero the new refcount block before updating it */
392 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
393 refcount_block);
394 if (ret < 0) {
395 goto fail;
396 }
397
398 memset(*refcount_block, 0, s->cluster_size);
399
400 /* The block describes itself, need to update the cache */
401 int block_index = (new_block >> s->cluster_bits) &
402 (s->refcount_block_size - 1);
403 s->set_refcount(*refcount_block, block_index, 1);
404 } else {
405 /* Described somewhere else. This can recurse at most twice before we
406 * arrive at a block that describes itself. */
407 ret = update_refcount(bs, new_block, s->cluster_size, 1, false,
408 QCOW2_DISCARD_NEVER);
409 if (ret < 0) {
410 goto fail;
411 }
412
413 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
414 if (ret < 0) {
415 goto fail;
416 }
417
418 /* Initialize the new refcount block only after updating its refcount,
419 * update_refcount uses the refcount cache itself */
420 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
421 refcount_block);
422 if (ret < 0) {
423 goto fail;
424 }
425
426 memset(*refcount_block, 0, s->cluster_size);
427 }
428
429 /* Now the new refcount block needs to be written to disk */
430 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
431 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block);
432 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
433 if (ret < 0) {
434 goto fail;
435 }
436
437 /* If the refcount table is big enough, just hook the block up there */
438 if (refcount_table_index < s->refcount_table_size) {
439 uint64_t data64 = cpu_to_be64(new_block);
440 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
441 ret = bdrv_pwrite_sync(bs->file, s->refcount_table_offset +
442 refcount_table_index * REFTABLE_ENTRY_SIZE,
443 sizeof(data64), &data64, 0);
444 if (ret < 0) {
445 goto fail;
446 }
447
448 s->refcount_table[refcount_table_index] = new_block;
449 /* If there's a hole in s->refcount_table then it can happen
450 * that refcount_table_index < s->max_refcount_table_index */
451 s->max_refcount_table_index =
452 MAX(s->max_refcount_table_index, refcount_table_index);
453
454 /* The new refcount block may be where the caller intended to put its
455 * data, so let it restart the search. */
456 return -EAGAIN;
457 }
458
459 qcow2_cache_put(s->refcount_block_cache, refcount_block);
460
461 /*
462 * If we come here, we need to grow the refcount table. Again, a new
463 * refcount table needs some space and we can't simply allocate to avoid
464 * endless recursion.
465 *
466 * Therefore let's grab new refcount blocks at the end of the image, which
467 * will describe themselves and the new refcount table. This way we can
468 * reference them only in the new table and do the switch to the new
469 * refcount table at once without producing an inconsistent state in
470 * between.
471 */
472 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
473
474 /* Calculate the number of refcount blocks needed so far; this will be the
475 * basis for calculating the index of the first cluster used for the
476 * self-describing refcount structures which we are about to create.
477 *
478 * Because we reached this point, there cannot be any refcount entries for
479 * cluster_index or higher indices yet. However, because new_block has been
480 * allocated to describe that cluster (and it will assume this role later
481 * on), we cannot use that index; also, new_block may actually have a higher
482 * cluster index than cluster_index, so it needs to be taken into account
483 * here (and 1 needs to be added to its value because that cluster is used).
484 */
485 uint64_t blocks_used = DIV_ROUND_UP(MAX(cluster_index + 1,
486 (new_block >> s->cluster_bits) + 1),
487 s->refcount_block_size);
488
489 /* Create the new refcount table and blocks */
490 uint64_t meta_offset = (blocks_used * s->refcount_block_size) *
491 s->cluster_size;
492
493 ret = qcow2_refcount_area(bs, meta_offset, 0, false,
494 refcount_table_index, new_block);
495 if (ret < 0) {
496 return ret;
497 }
498
499 ret = load_refcount_block(bs, new_block, refcount_block);
500 if (ret < 0) {
501 return ret;
502 }
503
504 /* If we were trying to do the initial refcount update for some cluster
505 * allocation, we might have used the same clusters to store newly
506 * allocated metadata. Make the caller search some new space. */
507 return -EAGAIN;
508
509 fail:
510 if (*refcount_block != NULL) {
511 qcow2_cache_put(s->refcount_block_cache, refcount_block);
512 }
513 return ret;
514 }
515
516 /*
517 * Starting at @start_offset, this function creates new self-covering refcount
518 * structures: A new refcount table and refcount blocks which cover all of
519 * themselves, and a number of @additional_clusters beyond their end.
520 * @start_offset must be at the end of the image file, that is, there must be
521 * only empty space beyond it.
522 * If @exact_size is false, the refcount table will have 50 % more entries than
523 * necessary so it will not need to grow again soon.
524 * If @new_refblock_offset is not zero, it contains the offset of a refcount
525 * block that should be entered into the new refcount table at index
526 * @new_refblock_index.
527 *
528 * Returns: The offset after the new refcount structures (i.e. where the
529 * @additional_clusters may be placed) on success, -errno on error.
530 */
531 int64_t qcow2_refcount_area(BlockDriverState *bs, uint64_t start_offset,
532 uint64_t additional_clusters, bool exact_size,
533 int new_refblock_index,
534 uint64_t new_refblock_offset)
535 {
536 BDRVQcow2State *s = bs->opaque;
537 uint64_t total_refblock_count_u64, additional_refblock_count;
538 int total_refblock_count, table_size, area_reftable_index, table_clusters;
539 int i;
540 uint64_t table_offset, block_offset, end_offset;
541 int ret;
542 uint64_t *new_table;
543
544 assert(!(start_offset % s->cluster_size));
545
546 qcow2_refcount_metadata_size(start_offset / s->cluster_size +
547 additional_clusters,
548 s->cluster_size, s->refcount_order,
549 !exact_size, &total_refblock_count_u64);
550 if (total_refblock_count_u64 > QCOW_MAX_REFTABLE_SIZE) {
551 return -EFBIG;
552 }
553 total_refblock_count = total_refblock_count_u64;
554
555 /* Index in the refcount table of the first refcount block to cover the area
556 * of refcount structures we are about to create; we know that
557 * @total_refblock_count can cover @start_offset, so this will definitely
558 * fit into an int. */
559 area_reftable_index = (start_offset / s->cluster_size) /
560 s->refcount_block_size;
561
562 if (exact_size) {
563 table_size = total_refblock_count;
564 } else {
565 table_size = total_refblock_count +
566 DIV_ROUND_UP(total_refblock_count, 2);
567 }
568 /* The qcow2 file can only store the reftable size in number of clusters */
569 table_size = ROUND_UP(table_size, s->cluster_size / REFTABLE_ENTRY_SIZE);
570 table_clusters = (table_size * REFTABLE_ENTRY_SIZE) / s->cluster_size;
571
572 if (table_size > QCOW_MAX_REFTABLE_SIZE) {
573 return -EFBIG;
574 }
575
576 new_table = g_try_new0(uint64_t, table_size);
577
578 assert(table_size > 0);
579 if (new_table == NULL) {
580 ret = -ENOMEM;
581 goto fail;
582 }
583
584 /* Fill the new refcount table */
585 if (table_size > s->max_refcount_table_index) {
586 /* We're actually growing the reftable */
587 memcpy(new_table, s->refcount_table,
588 (s->max_refcount_table_index + 1) * REFTABLE_ENTRY_SIZE);
589 } else {
590 /* Improbable case: We're shrinking the reftable. However, the caller
591 * has assured us that there is only empty space beyond @start_offset,
592 * so we can simply drop all of the refblocks that won't fit into the
593 * new reftable. */
594 memcpy(new_table, s->refcount_table, table_size * REFTABLE_ENTRY_SIZE);
595 }
596
597 if (new_refblock_offset) {
598 assert(new_refblock_index < total_refblock_count);
599 new_table[new_refblock_index] = new_refblock_offset;
600 }
601
602 /* Count how many new refblocks we have to create */
603 additional_refblock_count = 0;
604 for (i = area_reftable_index; i < total_refblock_count; i++) {
605 if (!new_table[i]) {
606 additional_refblock_count++;
607 }
608 }
609
610 table_offset = start_offset + additional_refblock_count * s->cluster_size;
611 end_offset = table_offset + table_clusters * s->cluster_size;
612
613 /* Fill the refcount blocks, and create new ones, if necessary */
614 block_offset = start_offset;
615 for (i = area_reftable_index; i < total_refblock_count; i++) {
616 void *refblock_data;
617 uint64_t first_offset_covered;
618
619 /* Reuse an existing refblock if possible, create a new one otherwise */
620 if (new_table[i]) {
621 ret = qcow2_cache_get(bs, s->refcount_block_cache, new_table[i],
622 &refblock_data);
623 if (ret < 0) {
624 goto fail;
625 }
626 } else {
627 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache,
628 block_offset, &refblock_data);
629 if (ret < 0) {
630 goto fail;
631 }
632 memset(refblock_data, 0, s->cluster_size);
633 qcow2_cache_entry_mark_dirty(s->refcount_block_cache,
634 refblock_data);
635
636 new_table[i] = block_offset;
637 block_offset += s->cluster_size;
638 }
639
640 /* First host offset covered by this refblock */
641 first_offset_covered = (uint64_t)i * s->refcount_block_size *
642 s->cluster_size;
643 if (first_offset_covered < end_offset) {
644 int j, end_index;
645
646 /* Set the refcount of all of the new refcount structures to 1 */
647
648 if (first_offset_covered < start_offset) {
649 assert(i == area_reftable_index);
650 j = (start_offset - first_offset_covered) / s->cluster_size;
651 assert(j < s->refcount_block_size);
652 } else {
653 j = 0;
654 }
655
656 end_index = MIN((end_offset - first_offset_covered) /
657 s->cluster_size,
658 s->refcount_block_size);
659
660 for (; j < end_index; j++) {
661 /* The caller guaranteed us this space would be empty */
662 assert(s->get_refcount(refblock_data, j) == 0);
663 s->set_refcount(refblock_data, j, 1);
664 }
665
666 qcow2_cache_entry_mark_dirty(s->refcount_block_cache,
667 refblock_data);
668 }
669
670 qcow2_cache_put(s->refcount_block_cache, &refblock_data);
671 }
672
673 assert(block_offset == table_offset);
674
675 /* Write refcount blocks to disk */
676 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
677 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
678 if (ret < 0) {
679 goto fail;
680 }
681
682 /* Write refcount table to disk */
683 for (i = 0; i < total_refblock_count; i++) {
684 cpu_to_be64s(&new_table[i]);
685 }
686
687 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
688 ret = bdrv_pwrite_sync(bs->file, table_offset,
689 table_size * REFTABLE_ENTRY_SIZE, new_table, 0);
690 if (ret < 0) {
691 goto fail;
692 }
693
694 for (i = 0; i < total_refblock_count; i++) {
695 be64_to_cpus(&new_table[i]);
696 }
697
698 /* Hook up the new refcount table in the qcow2 header */
699 struct QEMU_PACKED {
700 uint64_t d64;
701 uint32_t d32;
702 } data;
703 data.d64 = cpu_to_be64(table_offset);
704 data.d32 = cpu_to_be32(table_clusters);
705 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
706 ret = bdrv_pwrite_sync(bs->file,
707 offsetof(QCowHeader, refcount_table_offset),
708 sizeof(data), &data, 0);
709 if (ret < 0) {
710 goto fail;
711 }
712
713 /* And switch it in memory */
714 uint64_t old_table_offset = s->refcount_table_offset;
715 uint64_t old_table_size = s->refcount_table_size;
716
717 g_free(s->refcount_table);
718 s->refcount_table = new_table;
719 s->refcount_table_size = table_size;
720 s->refcount_table_offset = table_offset;
721 update_max_refcount_table_index(s);
722
723 /* Free old table. */
724 qcow2_free_clusters(bs, old_table_offset,
725 old_table_size * REFTABLE_ENTRY_SIZE,
726 QCOW2_DISCARD_OTHER);
727
728 return end_offset;
729
730 fail:
731 g_free(new_table);
732 return ret;
733 }
734
735 void qcow2_process_discards(BlockDriverState *bs, int ret)
736 {
737 BDRVQcow2State *s = bs->opaque;
738 Qcow2DiscardRegion *d, *next;
739
740 QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) {
741 QTAILQ_REMOVE(&s->discards, d, next);
742
743 /* Discard is optional, ignore the return value */
744 if (ret >= 0) {
745 int r2 = bdrv_pdiscard(bs->file, d->offset, d->bytes);
746 if (r2 < 0) {
747 trace_qcow2_process_discards_failed_region(d->offset, d->bytes,
748 r2);
749 }
750 }
751
752 g_free(d);
753 }
754 }
755
756 static void update_refcount_discard(BlockDriverState *bs,
757 uint64_t offset, uint64_t length)
758 {
759 BDRVQcow2State *s = bs->opaque;
760 Qcow2DiscardRegion *d, *p, *next;
761
762 QTAILQ_FOREACH(d, &s->discards, next) {
763 uint64_t new_start = MIN(offset, d->offset);
764 uint64_t new_end = MAX(offset + length, d->offset + d->bytes);
765
766 if (new_end - new_start <= length + d->bytes) {
767 /* There can't be any overlap, areas ending up here have no
768 * references any more and therefore shouldn't get freed another
769 * time. */
770 assert(d->bytes + length == new_end - new_start);
771 d->offset = new_start;
772 d->bytes = new_end - new_start;
773 goto found;
774 }
775 }
776
777 d = g_malloc(sizeof(*d));
778 *d = (Qcow2DiscardRegion) {
779 .bs = bs,
780 .offset = offset,
781 .bytes = length,
782 };
783 QTAILQ_INSERT_TAIL(&s->discards, d, next);
784
785 found:
786 /* Merge discard requests if they are adjacent now */
787 QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) {
788 if (p == d
789 || p->offset > d->offset + d->bytes
790 || d->offset > p->offset + p->bytes)
791 {
792 continue;
793 }
794
795 /* Still no overlap possible */
796 assert(p->offset == d->offset + d->bytes
797 || d->offset == p->offset + p->bytes);
798
799 QTAILQ_REMOVE(&s->discards, p, next);
800 d->offset = MIN(d->offset, p->offset);
801 d->bytes += p->bytes;
802 g_free(p);
803 }
804 }
805
806 /* XXX: cache several refcount block clusters ? */
807 /* @addend is the absolute value of the addend; if @decrease is set, @addend
808 * will be subtracted from the current refcount, otherwise it will be added */
809 static int update_refcount(BlockDriverState *bs,
810 int64_t offset,
811 int64_t length,
812 uint64_t addend,
813 bool decrease,
814 enum qcow2_discard_type type)
815 {
816 BDRVQcow2State *s = bs->opaque;
817 int64_t start, last, cluster_offset;
818 void *refcount_block = NULL;
819 int64_t old_table_index = -1;
820 int ret;
821
822 #ifdef DEBUG_ALLOC2
823 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64
824 " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "",
825 addend);
826 #endif
827 if (length < 0) {
828 return -EINVAL;
829 } else if (length == 0) {
830 return 0;
831 }
832
833 if (decrease) {
834 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
835 s->l2_table_cache);
836 }
837
838 start = start_of_cluster(s, offset);
839 last = start_of_cluster(s, offset + length - 1);
840 for(cluster_offset = start; cluster_offset <= last;
841 cluster_offset += s->cluster_size)
842 {
843 int block_index;
844 uint64_t refcount;
845 int64_t cluster_index = cluster_offset >> s->cluster_bits;
846 int64_t table_index = cluster_index >> s->refcount_block_bits;
847
848 /* Load the refcount block and allocate it if needed */
849 if (table_index != old_table_index) {
850 if (refcount_block) {
851 qcow2_cache_put(s->refcount_block_cache, &refcount_block);
852 }
853 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
854 /* If the caller needs to restart the search for free clusters,
855 * try the same ones first to see if they're still free. */
856 if (ret == -EAGAIN) {
857 if (s->free_cluster_index > (start >> s->cluster_bits)) {
858 s->free_cluster_index = (start >> s->cluster_bits);
859 }
860 }
861 if (ret < 0) {
862 goto fail;
863 }
864 }
865 old_table_index = table_index;
866
867 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
868
869 /* we can update the count and save it */
870 block_index = cluster_index & (s->refcount_block_size - 1);
871
872 refcount = s->get_refcount(refcount_block, block_index);
873 if (decrease ? (refcount - addend > refcount)
874 : (refcount + addend < refcount ||
875 refcount + addend > s->refcount_max))
876 {
877 ret = -EINVAL;
878 goto fail;
879 }
880 if (decrease) {
881 refcount -= addend;
882 } else {
883 refcount += addend;
884 }
885 if (refcount == 0 && cluster_index < s->free_cluster_index) {
886 s->free_cluster_index = cluster_index;
887 }
888 s->set_refcount(refcount_block, block_index, refcount);
889
890 if (refcount == 0) {
891 void *table;
892
893 table = qcow2_cache_is_table_offset(s->refcount_block_cache,
894 offset);
895 if (table != NULL) {
896 qcow2_cache_put(s->refcount_block_cache, &refcount_block);
897 old_table_index = -1;
898 qcow2_cache_discard(s->refcount_block_cache, table);
899 }
900
901 table = qcow2_cache_is_table_offset(s->l2_table_cache, offset);
902 if (table != NULL) {
903 qcow2_cache_discard(s->l2_table_cache, table);
904 }
905
906 if (s->discard_passthrough[type]) {
907 update_refcount_discard(bs, cluster_offset, s->cluster_size);
908 }
909 }
910 }
911
912 ret = 0;
913 fail:
914 if (!s->cache_discards) {
915 qcow2_process_discards(bs, ret);
916 }
917
918 /* Write last changed block to disk */
919 if (refcount_block) {
920 qcow2_cache_put(s->refcount_block_cache, &refcount_block);
921 }
922
923 /*
924 * Try do undo any updates if an error is returned (This may succeed in
925 * some cases like ENOSPC for allocating a new refcount block)
926 */
927 if (ret < 0) {
928 int dummy;
929 dummy = update_refcount(bs, offset, cluster_offset - offset, addend,
930 !decrease, QCOW2_DISCARD_NEVER);
931 (void)dummy;
932 }
933
934 return ret;
935 }
936
937 /*
938 * Increases or decreases the refcount of a given cluster.
939 *
940 * @addend is the absolute value of the addend; if @decrease is set, @addend
941 * will be subtracted from the current refcount, otherwise it will be added.
942 *
943 * On success 0 is returned; on failure -errno is returned.
944 */
945 int qcow2_update_cluster_refcount(BlockDriverState *bs,
946 int64_t cluster_index,
947 uint64_t addend, bool decrease,
948 enum qcow2_discard_type type)
949 {
950 BDRVQcow2State *s = bs->opaque;
951 int ret;
952
953 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend,
954 decrease, type);
955 if (ret < 0) {
956 return ret;
957 }
958
959 return 0;
960 }
961
962
963
964 /*********************************************************/
965 /* cluster allocation functions */
966
967
968
969 /* return < 0 if error */
970 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size,
971 uint64_t max)
972 {
973 BDRVQcow2State *s = bs->opaque;
974 uint64_t i, nb_clusters, refcount;
975 int ret;
976
977 /* We can't allocate clusters if they may still be queued for discard. */
978 if (s->cache_discards) {
979 qcow2_process_discards(bs, 0);
980 }
981
982 nb_clusters = size_to_clusters(s, size);
983 retry:
984 for(i = 0; i < nb_clusters; i++) {
985 uint64_t next_cluster_index = s->free_cluster_index++;
986 ret = qcow2_get_refcount(bs, next_cluster_index, &refcount);
987
988 if (ret < 0) {
989 return ret;
990 } else if (refcount != 0) {
991 goto retry;
992 }
993 }
994
995 /* Make sure that all offsets in the "allocated" range are representable
996 * in the requested max */
997 if (s->free_cluster_index > 0 &&
998 s->free_cluster_index - 1 > (max >> s->cluster_bits))
999 {
1000 return -EFBIG;
1001 }
1002
1003 #ifdef DEBUG_ALLOC2
1004 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
1005 size,
1006 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
1007 #endif
1008 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
1009 }
1010
1011 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size)
1012 {
1013 int64_t offset;
1014 int ret;
1015
1016 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
1017 do {
1018 offset = alloc_clusters_noref(bs, size, QCOW_MAX_CLUSTER_OFFSET);
1019 if (offset < 0) {
1020 return offset;
1021 }
1022
1023 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
1024 } while (ret == -EAGAIN);
1025
1026 if (ret < 0) {
1027 return ret;
1028 }
1029
1030 return offset;
1031 }
1032
1033 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
1034 int64_t nb_clusters)
1035 {
1036 BDRVQcow2State *s = bs->opaque;
1037 uint64_t cluster_index, refcount;
1038 uint64_t i;
1039 int ret;
1040
1041 assert(nb_clusters >= 0);
1042 if (nb_clusters == 0) {
1043 return 0;
1044 }
1045
1046 do {
1047 /* Check how many clusters there are free */
1048 cluster_index = offset >> s->cluster_bits;
1049 for(i = 0; i < nb_clusters; i++) {
1050 ret = qcow2_get_refcount(bs, cluster_index++, &refcount);
1051 if (ret < 0) {
1052 return ret;
1053 } else if (refcount != 0) {
1054 break;
1055 }
1056 }
1057
1058 /* And then allocate them */
1059 ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false,
1060 QCOW2_DISCARD_NEVER);
1061 } while (ret == -EAGAIN);
1062
1063 if (ret < 0) {
1064 return ret;
1065 }
1066
1067 return i;
1068 }
1069
1070 /* only used to allocate compressed sectors. We try to allocate
1071 contiguous sectors. size must be <= cluster_size */
1072 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
1073 {
1074 BDRVQcow2State *s = bs->opaque;
1075 int64_t offset;
1076 size_t free_in_cluster;
1077 int ret;
1078
1079 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
1080 assert(size > 0 && size <= s->cluster_size);
1081 assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset));
1082
1083 offset = s->free_byte_offset;
1084
1085 if (offset) {
1086 uint64_t refcount;
1087 ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount);
1088 if (ret < 0) {
1089 return ret;
1090 }
1091
1092 if (refcount == s->refcount_max) {
1093 offset = 0;
1094 }
1095 }
1096
1097 free_in_cluster = s->cluster_size - offset_into_cluster(s, offset);
1098 do {
1099 if (!offset || free_in_cluster < size) {
1100 int64_t new_cluster;
1101
1102 new_cluster = alloc_clusters_noref(bs, s->cluster_size,
1103 MIN(s->cluster_offset_mask,
1104 QCOW_MAX_CLUSTER_OFFSET));
1105 if (new_cluster < 0) {
1106 return new_cluster;
1107 }
1108
1109 if (new_cluster == 0) {
1110 qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid "
1111 "allocation of compressed cluster "
1112 "at offset 0");
1113 return -EIO;
1114 }
1115
1116 if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) {
1117 offset = new_cluster;
1118 free_in_cluster = s->cluster_size;
1119 } else {
1120 free_in_cluster += s->cluster_size;
1121 }
1122 }
1123
1124 assert(offset);
1125 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
1126 if (ret < 0) {
1127 offset = 0;
1128 }
1129 } while (ret == -EAGAIN);
1130 if (ret < 0) {
1131 return ret;
1132 }
1133
1134 /* The cluster refcount was incremented; refcount blocks must be flushed
1135 * before the caller's L2 table updates. */
1136 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
1137
1138 s->free_byte_offset = offset + size;
1139 if (!offset_into_cluster(s, s->free_byte_offset)) {
1140 s->free_byte_offset = 0;
1141 }
1142
1143 return offset;
1144 }
1145
1146 void qcow2_free_clusters(BlockDriverState *bs,
1147 int64_t offset, int64_t size,
1148 enum qcow2_discard_type type)
1149 {
1150 int ret;
1151
1152 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
1153 ret = update_refcount(bs, offset, size, 1, true, type);
1154 if (ret < 0) {
1155 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
1156 /* TODO Remember the clusters to free them later and avoid leaking */
1157 }
1158 }
1159
1160 /*
1161 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
1162 * normal cluster, compressed cluster, etc.)
1163 */
1164 void qcow2_free_any_cluster(BlockDriverState *bs, uint64_t l2_entry,
1165 enum qcow2_discard_type type)
1166 {
1167 BDRVQcow2State *s = bs->opaque;
1168 QCow2ClusterType ctype = qcow2_get_cluster_type(bs, l2_entry);
1169
1170 if (has_data_file(bs)) {
1171 if (s->discard_passthrough[type] &&
1172 (ctype == QCOW2_CLUSTER_NORMAL ||
1173 ctype == QCOW2_CLUSTER_ZERO_ALLOC))
1174 {
1175 bdrv_pdiscard(s->data_file, l2_entry & L2E_OFFSET_MASK,
1176 s->cluster_size);
1177 }
1178 return;
1179 }
1180
1181 switch (ctype) {
1182 case QCOW2_CLUSTER_COMPRESSED:
1183 {
1184 uint64_t coffset;
1185 int csize;
1186
1187 qcow2_parse_compressed_l2_entry(bs, l2_entry, &coffset, &csize);
1188 qcow2_free_clusters(bs, coffset, csize, type);
1189 }
1190 break;
1191 case QCOW2_CLUSTER_NORMAL:
1192 case QCOW2_CLUSTER_ZERO_ALLOC:
1193 if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) {
1194 qcow2_signal_corruption(bs, false, -1, -1,
1195 "Cannot free unaligned cluster %#llx",
1196 l2_entry & L2E_OFFSET_MASK);
1197 } else {
1198 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
1199 s->cluster_size, type);
1200 }
1201 break;
1202 case QCOW2_CLUSTER_ZERO_PLAIN:
1203 case QCOW2_CLUSTER_UNALLOCATED:
1204 break;
1205 default:
1206 abort();
1207 }
1208 }
1209
1210 int qcow2_write_caches(BlockDriverState *bs)
1211 {
1212 BDRVQcow2State *s = bs->opaque;
1213 int ret;
1214
1215 ret = qcow2_cache_write(bs, s->l2_table_cache);
1216 if (ret < 0) {
1217 return ret;
1218 }
1219
1220 if (qcow2_need_accurate_refcounts(s)) {
1221 ret = qcow2_cache_write(bs, s->refcount_block_cache);
1222 if (ret < 0) {
1223 return ret;
1224 }
1225 }
1226
1227 return 0;
1228 }
1229
1230 int qcow2_flush_caches(BlockDriverState *bs)
1231 {
1232 int ret = qcow2_write_caches(bs);
1233 if (ret < 0) {
1234 return ret;
1235 }
1236
1237 return bdrv_flush(bs->file->bs);
1238 }
1239
1240 /*********************************************************/
1241 /* snapshots and image creation */
1242
1243
1244
1245 /* update the refcounts of snapshots and the copied flag */
1246 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
1247 int64_t l1_table_offset, int l1_size, int addend)
1248 {
1249 BDRVQcow2State *s = bs->opaque;
1250 uint64_t *l1_table, *l2_slice, l2_offset, entry, l1_size2, refcount;
1251 bool l1_allocated = false;
1252 int64_t old_entry, old_l2_offset;
1253 unsigned slice, slice_size2, n_slices;
1254 int i, j, l1_modified = 0;
1255 int ret;
1256
1257 assert(addend >= -1 && addend <= 1);
1258
1259 l2_slice = NULL;
1260 l1_table = NULL;
1261 l1_size2 = l1_size * L1E_SIZE;
1262 slice_size2 = s->l2_slice_size * l2_entry_size(s);
1263 n_slices = s->cluster_size / slice_size2;
1264
1265 s->cache_discards = true;
1266
1267 /* WARNING: qcow2_snapshot_goto relies on this function not using the
1268 * l1_table_offset when it is the current s->l1_table_offset! Be careful
1269 * when changing this! */
1270 if (l1_table_offset != s->l1_table_offset) {
1271 l1_table = g_try_malloc0(l1_size2);
1272 if (l1_size2 && l1_table == NULL) {
1273 ret = -ENOMEM;
1274 goto fail;
1275 }
1276 l1_allocated = true;
1277
1278 ret = bdrv_pread(bs->file, l1_table_offset, l1_size2, l1_table, 0);
1279 if (ret < 0) {
1280 goto fail;
1281 }
1282
1283 for (i = 0; i < l1_size; i++) {
1284 be64_to_cpus(&l1_table[i]);
1285 }
1286 } else {
1287 assert(l1_size == s->l1_size);
1288 l1_table = s->l1_table;
1289 l1_allocated = false;
1290 }
1291
1292 for (i = 0; i < l1_size; i++) {
1293 l2_offset = l1_table[i];
1294 if (l2_offset) {
1295 old_l2_offset = l2_offset;
1296 l2_offset &= L1E_OFFSET_MASK;
1297
1298 if (offset_into_cluster(s, l2_offset)) {
1299 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1300 PRIx64 " unaligned (L1 index: %#x)",
1301 l2_offset, i);
1302 ret = -EIO;
1303 goto fail;
1304 }
1305
1306 for (slice = 0; slice < n_slices; slice++) {
1307 ret = qcow2_cache_get(bs, s->l2_table_cache,
1308 l2_offset + slice * slice_size2,
1309 (void **) &l2_slice);
1310 if (ret < 0) {
1311 goto fail;
1312 }
1313
1314 for (j = 0; j < s->l2_slice_size; j++) {
1315 uint64_t cluster_index;
1316 uint64_t offset;
1317
1318 entry = get_l2_entry(s, l2_slice, j);
1319 old_entry = entry;
1320 entry &= ~QCOW_OFLAG_COPIED;
1321 offset = entry & L2E_OFFSET_MASK;
1322
1323 switch (qcow2_get_cluster_type(bs, entry)) {
1324 case QCOW2_CLUSTER_COMPRESSED:
1325 if (addend != 0) {
1326 uint64_t coffset;
1327 int csize;
1328
1329 qcow2_parse_compressed_l2_entry(bs, entry,
1330 &coffset, &csize);
1331 ret = update_refcount(
1332 bs, coffset, csize,
1333 abs(addend), addend < 0,
1334 QCOW2_DISCARD_SNAPSHOT);
1335 if (ret < 0) {
1336 goto fail;
1337 }
1338 }
1339 /* compressed clusters are never modified */
1340 refcount = 2;
1341 break;
1342
1343 case QCOW2_CLUSTER_NORMAL:
1344 case QCOW2_CLUSTER_ZERO_ALLOC:
1345 if (offset_into_cluster(s, offset)) {
1346 /* Here l2_index means table (not slice) index */
1347 int l2_index = slice * s->l2_slice_size + j;
1348 qcow2_signal_corruption(
1349 bs, true, -1, -1, "Cluster "
1350 "allocation offset %#" PRIx64
1351 " unaligned (L2 offset: %#"
1352 PRIx64 ", L2 index: %#x)",
1353 offset, l2_offset, l2_index);
1354 ret = -EIO;
1355 goto fail;
1356 }
1357
1358 cluster_index = offset >> s->cluster_bits;
1359 assert(cluster_index);
1360 if (addend != 0) {
1361 ret = qcow2_update_cluster_refcount(
1362 bs, cluster_index, abs(addend), addend < 0,
1363 QCOW2_DISCARD_SNAPSHOT);
1364 if (ret < 0) {
1365 goto fail;
1366 }
1367 }
1368
1369 ret = qcow2_get_refcount(bs, cluster_index, &refcount);
1370 if (ret < 0) {
1371 goto fail;
1372 }
1373 break;
1374
1375 case QCOW2_CLUSTER_ZERO_PLAIN:
1376 case QCOW2_CLUSTER_UNALLOCATED:
1377 refcount = 0;
1378 break;
1379
1380 default:
1381 abort();
1382 }
1383
1384 if (refcount == 1) {
1385 entry |= QCOW_OFLAG_COPIED;
1386 }
1387 if (entry != old_entry) {
1388 if (addend > 0) {
1389 qcow2_cache_set_dependency(bs, s->l2_table_cache,
1390 s->refcount_block_cache);
1391 }
1392 set_l2_entry(s, l2_slice, j, entry);
1393 qcow2_cache_entry_mark_dirty(s->l2_table_cache,
1394 l2_slice);
1395 }
1396 }
1397
1398 qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
1399 }
1400
1401 if (addend != 0) {
1402 ret = qcow2_update_cluster_refcount(bs, l2_offset >>
1403 s->cluster_bits,
1404 abs(addend), addend < 0,
1405 QCOW2_DISCARD_SNAPSHOT);
1406 if (ret < 0) {
1407 goto fail;
1408 }
1409 }
1410 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1411 &refcount);
1412 if (ret < 0) {
1413 goto fail;
1414 } else if (refcount == 1) {
1415 l2_offset |= QCOW_OFLAG_COPIED;
1416 }
1417 if (l2_offset != old_l2_offset) {
1418 l1_table[i] = l2_offset;
1419 l1_modified = 1;
1420 }
1421 }
1422 }
1423
1424 ret = bdrv_flush(bs);
1425 fail:
1426 if (l2_slice) {
1427 qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
1428 }
1429
1430 s->cache_discards = false;
1431 qcow2_process_discards(bs, ret);
1432
1433 /* Update L1 only if it isn't deleted anyway (addend = -1) */
1434 if (ret == 0 && addend >= 0 && l1_modified) {
1435 for (i = 0; i < l1_size; i++) {
1436 cpu_to_be64s(&l1_table[i]);
1437 }
1438
1439 ret = bdrv_pwrite_sync(bs->file, l1_table_offset, l1_size2, l1_table,
1440 0);
1441
1442 for (i = 0; i < l1_size; i++) {
1443 be64_to_cpus(&l1_table[i]);
1444 }
1445 }
1446 if (l1_allocated)
1447 g_free(l1_table);
1448 return ret;
1449 }
1450
1451
1452
1453
1454 /*********************************************************/
1455 /* refcount checking functions */
1456
1457
1458 static uint64_t refcount_array_byte_size(BDRVQcow2State *s, uint64_t entries)
1459 {
1460 /* This assertion holds because there is no way we can address more than
1461 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because
1462 * offsets have to be representable in bytes); due to every cluster
1463 * corresponding to one refcount entry, we are well below that limit */
1464 assert(entries < (UINT64_C(1) << (64 - 9)));
1465
1466 /* Thanks to the assertion this will not overflow, because
1467 * s->refcount_order < 7.
1468 * (note: x << s->refcount_order == x * s->refcount_bits) */
1469 return DIV_ROUND_UP(entries << s->refcount_order, 8);
1470 }
1471
1472 /**
1473 * Reallocates *array so that it can hold new_size entries. *size must contain
1474 * the current number of entries in *array. If the reallocation fails, *array
1475 * and *size will not be modified and -errno will be returned. If the
1476 * reallocation is successful, *array will be set to the new buffer, *size
1477 * will be set to new_size and 0 will be returned. The size of the reallocated
1478 * refcount array buffer will be aligned to a cluster boundary, and the newly
1479 * allocated area will be zeroed.
1480 */
1481 static int realloc_refcount_array(BDRVQcow2State *s, void **array,
1482 int64_t *size, int64_t new_size)
1483 {
1484 int64_t old_byte_size, new_byte_size;
1485 void *new_ptr;
1486
1487 /* Round to clusters so the array can be directly written to disk */
1488 old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size))
1489 * s->cluster_size;
1490 new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size))
1491 * s->cluster_size;
1492
1493 if (new_byte_size == old_byte_size) {
1494 *size = new_size;
1495 return 0;
1496 }
1497
1498 assert(new_byte_size > 0);
1499
1500 if (new_byte_size > SIZE_MAX) {
1501 return -ENOMEM;
1502 }
1503
1504 new_ptr = g_try_realloc(*array, new_byte_size);
1505 if (!new_ptr) {
1506 return -ENOMEM;
1507 }
1508
1509 if (new_byte_size > old_byte_size) {
1510 memset((char *)new_ptr + old_byte_size, 0,
1511 new_byte_size - old_byte_size);
1512 }
1513
1514 *array = new_ptr;
1515 *size = new_size;
1516
1517 return 0;
1518 }
1519
1520 /*
1521 * Increases the refcount for a range of clusters in a given refcount table.
1522 * This is used to construct a temporary refcount table out of L1 and L2 tables
1523 * which can be compared to the refcount table saved in the image.
1524 *
1525 * Modifies the number of errors in res.
1526 */
1527 int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res,
1528 void **refcount_table,
1529 int64_t *refcount_table_size,
1530 int64_t offset, int64_t size)
1531 {
1532 BDRVQcow2State *s = bs->opaque;
1533 uint64_t start, last, cluster_offset, k, refcount;
1534 int64_t file_len;
1535 int ret;
1536
1537 if (size <= 0) {
1538 return 0;
1539 }
1540
1541 file_len = bdrv_getlength(bs->file->bs);
1542 if (file_len < 0) {
1543 return file_len;
1544 }
1545
1546 /*
1547 * Last cluster of qcow2 image may be semi-allocated, so it may be OK to
1548 * reference some space after file end but it should be less than one
1549 * cluster.
1550 */
1551 if (offset + size - file_len >= s->cluster_size) {
1552 fprintf(stderr, "ERROR: counting reference for region exceeding the "
1553 "end of the file by one cluster or more: offset 0x%" PRIx64
1554 " size 0x%" PRIx64 "\n", offset, size);
1555 res->corruptions++;
1556 return 0;
1557 }
1558
1559 start = start_of_cluster(s, offset);
1560 last = start_of_cluster(s, offset + size - 1);
1561 for(cluster_offset = start; cluster_offset <= last;
1562 cluster_offset += s->cluster_size) {
1563 k = cluster_offset >> s->cluster_bits;
1564 if (k >= *refcount_table_size) {
1565 ret = realloc_refcount_array(s, refcount_table,
1566 refcount_table_size, k + 1);
1567 if (ret < 0) {
1568 res->check_errors++;
1569 return ret;
1570 }
1571 }
1572
1573 refcount = s->get_refcount(*refcount_table, k);
1574 if (refcount == s->refcount_max) {
1575 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
1576 "\n", cluster_offset);
1577 fprintf(stderr, "Use qemu-img amend to increase the refcount entry "
1578 "width or qemu-img convert to create a clean copy if the "
1579 "image cannot be opened for writing\n");
1580 res->corruptions++;
1581 continue;
1582 }
1583 s->set_refcount(*refcount_table, k, refcount + 1);
1584 }
1585
1586 return 0;
1587 }
1588
1589 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
1590 enum {
1591 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
1592 };
1593
1594 /*
1595 * Fix L2 entry by making it QCOW2_CLUSTER_ZERO_PLAIN (or making all its present
1596 * subclusters QCOW2_SUBCLUSTER_ZERO_PLAIN).
1597 *
1598 * This function decrements res->corruptions on success, so the caller is
1599 * responsible to increment res->corruptions prior to the call.
1600 *
1601 * On failure in-memory @l2_table may be modified.
1602 */
1603 static int fix_l2_entry_by_zero(BlockDriverState *bs, BdrvCheckResult *res,
1604 uint64_t l2_offset,
1605 uint64_t *l2_table, int l2_index, bool active,
1606 bool *metadata_overlap)
1607 {
1608 BDRVQcow2State *s = bs->opaque;
1609 int ret;
1610 int idx = l2_index * (l2_entry_size(s) / sizeof(uint64_t));
1611 uint64_t l2e_offset = l2_offset + (uint64_t)l2_index * l2_entry_size(s);
1612 int ign = active ? QCOW2_OL_ACTIVE_L2 : QCOW2_OL_INACTIVE_L2;
1613
1614 if (has_subclusters(s)) {
1615 uint64_t l2_bitmap = get_l2_bitmap(s, l2_table, l2_index);
1616
1617 /* Allocated subclusters become zero */
1618 l2_bitmap |= l2_bitmap << 32;
1619 l2_bitmap &= QCOW_L2_BITMAP_ALL_ZEROES;
1620
1621 set_l2_bitmap(s, l2_table, l2_index, l2_bitmap);
1622 set_l2_entry(s, l2_table, l2_index, 0);
1623 } else {
1624 set_l2_entry(s, l2_table, l2_index, QCOW_OFLAG_ZERO);
1625 }
1626
1627 ret = qcow2_pre_write_overlap_check(bs, ign, l2e_offset, l2_entry_size(s),
1628 false);
1629 if (metadata_overlap) {
1630 *metadata_overlap = ret < 0;
1631 }
1632 if (ret < 0) {
1633 fprintf(stderr, "ERROR: Overlap check failed\n");
1634 goto fail;
1635 }
1636
1637 ret = bdrv_pwrite_sync(bs->file, l2e_offset, l2_entry_size(s),
1638 &l2_table[idx], 0);
1639 if (ret < 0) {
1640 fprintf(stderr, "ERROR: Failed to overwrite L2 "
1641 "table entry: %s\n", strerror(-ret));
1642 goto fail;
1643 }
1644
1645 res->corruptions--;
1646 res->corruptions_fixed++;
1647 return 0;
1648
1649 fail:
1650 res->check_errors++;
1651 return ret;
1652 }
1653
1654 /*
1655 * Increases the refcount in the given refcount table for the all clusters
1656 * referenced in the L2 table. While doing so, performs some checks on L2
1657 * entries.
1658 *
1659 * Returns the number of errors found by the checks or -errno if an internal
1660 * error occurred.
1661 */
1662 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
1663 void **refcount_table,
1664 int64_t *refcount_table_size, int64_t l2_offset,
1665 int flags, BdrvCheckMode fix, bool active)
1666 {
1667 BDRVQcow2State *s = bs->opaque;
1668 uint64_t l2_entry, l2_bitmap;
1669 uint64_t next_contiguous_offset = 0;
1670 int i, ret;
1671 size_t l2_size_bytes = s->l2_size * l2_entry_size(s);
1672 g_autofree uint64_t *l2_table = g_malloc(l2_size_bytes);
1673 bool metadata_overlap;
1674
1675 /* Read L2 table from disk */
1676 ret = bdrv_pread(bs->file, l2_offset, l2_size_bytes, l2_table, 0);
1677 if (ret < 0) {
1678 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1679 res->check_errors++;
1680 return ret;
1681 }
1682
1683 /* Do the actual checks */
1684 for (i = 0; i < s->l2_size; i++) {
1685 uint64_t coffset;
1686 int csize;
1687 QCow2ClusterType type;
1688
1689 l2_entry = get_l2_entry(s, l2_table, i);
1690 l2_bitmap = get_l2_bitmap(s, l2_table, i);
1691 type = qcow2_get_cluster_type(bs, l2_entry);
1692
1693 if (type != QCOW2_CLUSTER_COMPRESSED) {
1694 /* Check reserved bits of Standard Cluster Descriptor */
1695 if (l2_entry & L2E_STD_RESERVED_MASK) {
1696 fprintf(stderr, "ERROR found l2 entry with reserved bits set: "
1697 "%" PRIx64 "\n", l2_entry);
1698 res->corruptions++;
1699 }
1700 }
1701
1702 switch (type) {
1703 case QCOW2_CLUSTER_COMPRESSED:
1704 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1705 if (l2_entry & QCOW_OFLAG_COPIED) {
1706 fprintf(stderr, "ERROR: coffset=0x%" PRIx64 ": "
1707 "copied flag must never be set for compressed "
1708 "clusters\n", l2_entry & s->cluster_offset_mask);
1709 l2_entry &= ~QCOW_OFLAG_COPIED;
1710 res->corruptions++;
1711 }
1712
1713 if (has_data_file(bs)) {
1714 fprintf(stderr, "ERROR compressed cluster %d with data file, "
1715 "entry=0x%" PRIx64 "\n", i, l2_entry);
1716 res->corruptions++;
1717 break;
1718 }
1719
1720 if (l2_bitmap) {
1721 fprintf(stderr, "ERROR compressed cluster %d with non-zero "
1722 "subcluster allocation bitmap, entry=0x%" PRIx64 "\n",
1723 i, l2_entry);
1724 res->corruptions++;
1725 break;
1726 }
1727
1728 /* Mark cluster as used */
1729 qcow2_parse_compressed_l2_entry(bs, l2_entry, &coffset, &csize);
1730 ret = qcow2_inc_refcounts_imrt(
1731 bs, res, refcount_table, refcount_table_size, coffset, csize);
1732 if (ret < 0) {
1733 return ret;
1734 }
1735
1736 if (flags & CHECK_FRAG_INFO) {
1737 res->bfi.allocated_clusters++;
1738 res->bfi.compressed_clusters++;
1739
1740 /*
1741 * Compressed clusters are fragmented by nature. Since they
1742 * take up sub-sector space but we only have sector granularity
1743 * I/O we need to re-read the same sectors even for adjacent
1744 * compressed clusters.
1745 */
1746 res->bfi.fragmented_clusters++;
1747 }
1748 break;
1749
1750 case QCOW2_CLUSTER_ZERO_ALLOC:
1751 case QCOW2_CLUSTER_NORMAL:
1752 {
1753 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
1754
1755 if ((l2_bitmap >> 32) & l2_bitmap) {
1756 res->corruptions++;
1757 fprintf(stderr, "ERROR offset=%" PRIx64 ": Allocated "
1758 "cluster has corrupted subcluster allocation bitmap\n",
1759 offset);
1760 }
1761
1762 /* Correct offsets are cluster aligned */
1763 if (offset_into_cluster(s, offset)) {
1764 bool contains_data;
1765 res->corruptions++;
1766
1767 if (has_subclusters(s)) {
1768 contains_data = (l2_bitmap & QCOW_L2_BITMAP_ALL_ALLOC);
1769 } else {
1770 contains_data = !(l2_entry & QCOW_OFLAG_ZERO);
1771 }
1772
1773 if (!contains_data) {
1774 fprintf(stderr, "%s offset=%" PRIx64 ": Preallocated "
1775 "cluster is not properly aligned; L2 entry "
1776 "corrupted.\n",
1777 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR",
1778 offset);
1779 if (fix & BDRV_FIX_ERRORS) {
1780 ret = fix_l2_entry_by_zero(bs, res, l2_offset,
1781 l2_table, i, active,
1782 &metadata_overlap);
1783 if (metadata_overlap) {
1784 /*
1785 * Something is seriously wrong, so abort checking
1786 * this L2 table.
1787 */
1788 return ret;
1789 }
1790
1791 if (ret == 0) {
1792 /*
1793 * Skip marking the cluster as used
1794 * (it is unused now).
1795 */
1796 continue;
1797 }
1798
1799 /*
1800 * Failed to fix.
1801 * Do not abort, continue checking the rest of this
1802 * L2 table's entries.
1803 */
1804 }
1805 } else {
1806 fprintf(stderr, "ERROR offset=%" PRIx64 ": Data cluster is "
1807 "not properly aligned; L2 entry corrupted.\n", offset);
1808 }
1809 }
1810
1811 if (flags & CHECK_FRAG_INFO) {
1812 res->bfi.allocated_clusters++;
1813 if (next_contiguous_offset &&
1814 offset != next_contiguous_offset) {
1815 res->bfi.fragmented_clusters++;
1816 }
1817 next_contiguous_offset = offset + s->cluster_size;
1818 }
1819
1820 /* Mark cluster as used */
1821 if (!has_data_file(bs)) {
1822 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table,
1823 refcount_table_size,
1824 offset, s->cluster_size);
1825 if (ret < 0) {
1826 return ret;
1827 }
1828 }
1829 break;
1830 }
1831
1832 case QCOW2_CLUSTER_ZERO_PLAIN:
1833 /* Impossible when image has subclusters */
1834 assert(!l2_bitmap);
1835 break;
1836
1837 case QCOW2_CLUSTER_UNALLOCATED:
1838 if (l2_bitmap & QCOW_L2_BITMAP_ALL_ALLOC) {
1839 res->corruptions++;
1840 fprintf(stderr, "ERROR: Unallocated "
1841 "cluster has non-zero subcluster allocation map\n");
1842 }
1843 break;
1844
1845 default:
1846 abort();
1847 }
1848 }
1849
1850 return 0;
1851 }
1852
1853 /*
1854 * Increases the refcount for the L1 table, its L2 tables and all referenced
1855 * clusters in the given refcount table. While doing so, performs some checks
1856 * on L1 and L2 entries.
1857 *
1858 * Returns the number of errors found by the checks or -errno if an internal
1859 * error occurred.
1860 */
1861 static int check_refcounts_l1(BlockDriverState *bs,
1862 BdrvCheckResult *res,
1863 void **refcount_table,
1864 int64_t *refcount_table_size,
1865 int64_t l1_table_offset, int l1_size,
1866 int flags, BdrvCheckMode fix, bool active)
1867 {
1868 BDRVQcow2State *s = bs->opaque;
1869 size_t l1_size_bytes = l1_size * L1E_SIZE;
1870 g_autofree uint64_t *l1_table = NULL;
1871 uint64_t l2_offset;
1872 int i, ret;
1873
1874 if (!l1_size) {
1875 return 0;
1876 }
1877
1878 /* Mark L1 table as used */
1879 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, refcount_table_size,
1880 l1_table_offset, l1_size_bytes);
1881 if (ret < 0) {
1882 return ret;
1883 }
1884
1885 l1_table = g_try_malloc(l1_size_bytes);
1886 if (l1_table == NULL) {
1887 res->check_errors++;
1888 return -ENOMEM;
1889 }
1890
1891 /* Read L1 table entries from disk */
1892 ret = bdrv_pread(bs->file, l1_table_offset, l1_size_bytes, l1_table, 0);
1893 if (ret < 0) {
1894 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1895 res->check_errors++;
1896 return ret;
1897 }
1898
1899 for (i = 0; i < l1_size; i++) {
1900 be64_to_cpus(&l1_table[i]);
1901 }
1902
1903 /* Do the actual checks */
1904 for (i = 0; i < l1_size; i++) {
1905 if (!l1_table[i]) {
1906 continue;
1907 }
1908
1909 if (l1_table[i] & L1E_RESERVED_MASK) {
1910 fprintf(stderr, "ERROR found L1 entry with reserved bits set: "
1911 "%" PRIx64 "\n", l1_table[i]);
1912 res->corruptions++;
1913 }
1914
1915 l2_offset = l1_table[i] & L1E_OFFSET_MASK;
1916
1917 /* Mark L2 table as used */
1918 ret = qcow2_inc_refcounts_imrt(bs, res,
1919 refcount_table, refcount_table_size,
1920 l2_offset, s->cluster_size);
1921 if (ret < 0) {
1922 return ret;
1923 }
1924
1925 /* L2 tables are cluster aligned */
1926 if (offset_into_cluster(s, l2_offset)) {
1927 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1928 "cluster aligned; L1 entry corrupted\n", l2_offset);
1929 res->corruptions++;
1930 }
1931
1932 /* Process and check L2 entries */
1933 ret = check_refcounts_l2(bs, res, refcount_table,
1934 refcount_table_size, l2_offset, flags,
1935 fix, active);
1936 if (ret < 0) {
1937 return ret;
1938 }
1939 }
1940
1941 return 0;
1942 }
1943
1944 /*
1945 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1946 *
1947 * This function does not print an error message nor does it increment
1948 * check_errors if qcow2_get_refcount fails (this is because such an error will
1949 * have been already detected and sufficiently signaled by the calling function
1950 * (qcow2_check_refcounts) by the time this function is called).
1951 */
1952 static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
1953 BdrvCheckMode fix)
1954 {
1955 BDRVQcow2State *s = bs->opaque;
1956 uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size);
1957 int ret;
1958 uint64_t refcount;
1959 int i, j;
1960 bool repair;
1961
1962 if (fix & BDRV_FIX_ERRORS) {
1963 /* Always repair */
1964 repair = true;
1965 } else if (fix & BDRV_FIX_LEAKS) {
1966 /* Repair only if that seems safe: This function is always
1967 * called after the refcounts have been fixed, so the refcount
1968 * is accurate if that repair was successful */
1969 repair = !res->check_errors && !res->corruptions && !res->leaks;
1970 } else {
1971 repair = false;
1972 }
1973
1974 for (i = 0; i < s->l1_size; i++) {
1975 uint64_t l1_entry = s->l1_table[i];
1976 uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK;
1977 int l2_dirty = 0;
1978
1979 if (!l2_offset) {
1980 continue;
1981 }
1982
1983 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1984 &refcount);
1985 if (ret < 0) {
1986 /* don't print message nor increment check_errors */
1987 continue;
1988 }
1989 if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) {
1990 res->corruptions++;
1991 fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1992 "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
1993 repair ? "Repairing" : "ERROR", i, l1_entry, refcount);
1994 if (repair) {
1995 s->l1_table[i] = refcount == 1
1996 ? l1_entry | QCOW_OFLAG_COPIED
1997 : l1_entry & ~QCOW_OFLAG_COPIED;
1998 ret = qcow2_write_l1_entry(bs, i);
1999 if (ret < 0) {
2000 res->check_errors++;
2001 goto fail;
2002 }
2003 res->corruptions--;
2004 res->corruptions_fixed++;
2005 }
2006 }
2007
2008 ret = bdrv_pread(bs->file, l2_offset, s->l2_size * l2_entry_size(s),
2009 l2_table, 0);
2010 if (ret < 0) {
2011 fprintf(stderr, "ERROR: Could not read L2 table: %s\n",
2012 strerror(-ret));
2013 res->check_errors++;
2014 goto fail;
2015 }
2016
2017 for (j = 0; j < s->l2_size; j++) {
2018 uint64_t l2_entry = get_l2_entry(s, l2_table, j);
2019 uint64_t data_offset = l2_entry & L2E_OFFSET_MASK;
2020 QCow2ClusterType cluster_type = qcow2_get_cluster_type(bs, l2_entry);
2021
2022 if (cluster_type == QCOW2_CLUSTER_NORMAL ||
2023 cluster_type == QCOW2_CLUSTER_ZERO_ALLOC) {
2024 if (has_data_file(bs)) {
2025 refcount = 1;
2026 } else {
2027 ret = qcow2_get_refcount(bs,
2028 data_offset >> s->cluster_bits,
2029 &refcount);
2030 if (ret < 0) {
2031 /* don't print message nor increment check_errors */
2032 continue;
2033 }
2034 }
2035 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
2036 res->corruptions++;
2037 fprintf(stderr, "%s OFLAG_COPIED data cluster: "
2038 "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
2039 repair ? "Repairing" : "ERROR", l2_entry, refcount);
2040 if (repair) {
2041 set_l2_entry(s, l2_table, j,
2042 refcount == 1 ?
2043 l2_entry | QCOW_OFLAG_COPIED :
2044 l2_entry & ~QCOW_OFLAG_COPIED);
2045 l2_dirty++;
2046 }
2047 }
2048 }
2049 }
2050
2051 if (l2_dirty > 0) {
2052 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
2053 l2_offset, s->cluster_size,
2054 false);
2055 if (ret < 0) {
2056 fprintf(stderr, "ERROR: Could not write L2 table; metadata "
2057 "overlap check failed: %s\n", strerror(-ret));
2058 res->check_errors++;
2059 goto fail;
2060 }
2061
2062 ret = bdrv_pwrite(bs->file, l2_offset, s->cluster_size, l2_table,
2063 0);
2064 if (ret < 0) {
2065 fprintf(stderr, "ERROR: Could not write L2 table: %s\n",
2066 strerror(-ret));
2067 res->check_errors++;
2068 goto fail;
2069 }
2070 res->corruptions -= l2_dirty;
2071 res->corruptions_fixed += l2_dirty;
2072 }
2073 }
2074
2075 ret = 0;
2076
2077 fail:
2078 qemu_vfree(l2_table);
2079 return ret;
2080 }
2081
2082 /*
2083 * Checks consistency of refblocks and accounts for each refblock in
2084 * *refcount_table.
2085 */
2086 static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res,
2087 BdrvCheckMode fix, bool *rebuild,
2088 void **refcount_table, int64_t *nb_clusters)
2089 {
2090 BDRVQcow2State *s = bs->opaque;
2091 int64_t i, size;
2092 int ret;
2093
2094 for(i = 0; i < s->refcount_table_size; i++) {
2095 uint64_t offset, cluster;
2096 offset = s->refcount_table[i] & REFT_OFFSET_MASK;
2097 cluster = offset >> s->cluster_bits;
2098
2099 if (s->refcount_table[i] & REFT_RESERVED_MASK) {
2100 fprintf(stderr, "ERROR refcount table entry %" PRId64 " has "
2101 "reserved bits set\n", i);
2102 res->corruptions++;
2103 *rebuild = true;
2104 continue;
2105 }
2106
2107 /* Refcount blocks are cluster aligned */
2108 if (offset_into_cluster(s, offset)) {
2109 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
2110 "cluster aligned; refcount table entry corrupted\n", i);
2111 res->corruptions++;
2112 *rebuild = true;
2113 continue;
2114 }
2115
2116 if (cluster >= *nb_clusters) {
2117 res->corruptions++;
2118 fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n",
2119 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
2120
2121 if (fix & BDRV_FIX_ERRORS) {
2122 int64_t new_nb_clusters;
2123 Error *local_err = NULL;
2124
2125 if (offset > INT64_MAX - s->cluster_size) {
2126 ret = -EINVAL;
2127 goto resize_fail;
2128 }
2129
2130 ret = bdrv_truncate(bs->file, offset + s->cluster_size, false,
2131 PREALLOC_MODE_OFF, 0, &local_err);
2132 if (ret < 0) {
2133 error_report_err(local_err);
2134 goto resize_fail;
2135 }
2136 size = bdrv_getlength(bs->file->bs);
2137 if (size < 0) {
2138 ret = size;
2139 goto resize_fail;
2140 }
2141
2142 new_nb_clusters = size_to_clusters(s, size);
2143 assert(new_nb_clusters >= *nb_clusters);
2144
2145 ret = realloc_refcount_array(s, refcount_table,
2146 nb_clusters, new_nb_clusters);
2147 if (ret < 0) {
2148 res->check_errors++;
2149 return ret;
2150 }
2151
2152 if (cluster >= *nb_clusters) {
2153 ret = -EINVAL;
2154 goto resize_fail;
2155 }
2156
2157 res->corruptions--;
2158 res->corruptions_fixed++;
2159 ret = qcow2_inc_refcounts_imrt(bs, res,
2160 refcount_table, nb_clusters,
2161 offset, s->cluster_size);
2162 if (ret < 0) {
2163 return ret;
2164 }
2165 /* No need to check whether the refcount is now greater than 1:
2166 * This area was just allocated and zeroed, so it can only be
2167 * exactly 1 after qcow2_inc_refcounts_imrt() */
2168 continue;
2169
2170 resize_fail:
2171 *rebuild = true;
2172 fprintf(stderr, "ERROR could not resize image: %s\n",
2173 strerror(-ret));
2174 }
2175 continue;
2176 }
2177
2178 if (offset != 0) {
2179 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2180 offset, s->cluster_size);
2181 if (ret < 0) {
2182 return ret;
2183 }
2184 if (s->get_refcount(*refcount_table, cluster) != 1) {
2185 fprintf(stderr, "ERROR refcount block %" PRId64
2186 " refcount=%" PRIu64 "\n", i,
2187 s->get_refcount(*refcount_table, cluster));
2188 res->corruptions++;
2189 *rebuild = true;
2190 }
2191 }
2192 }
2193
2194 return 0;
2195 }
2196
2197 /*
2198 * Calculates an in-memory refcount table.
2199 */
2200 static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2201 BdrvCheckMode fix, bool *rebuild,
2202 void **refcount_table, int64_t *nb_clusters)
2203 {
2204 BDRVQcow2State *s = bs->opaque;
2205 int64_t i;
2206 QCowSnapshot *sn;
2207 int ret;
2208
2209 if (!*refcount_table) {
2210 int64_t old_size = 0;
2211 ret = realloc_refcount_array(s, refcount_table,
2212 &old_size, *nb_clusters);
2213 if (ret < 0) {
2214 res->check_errors++;
2215 return ret;
2216 }
2217 }
2218
2219 /* header */
2220 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2221 0, s->cluster_size);
2222 if (ret < 0) {
2223 return ret;
2224 }
2225
2226 /* current L1 table */
2227 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
2228 s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO,
2229 fix, true);
2230 if (ret < 0) {
2231 return ret;
2232 }
2233
2234 /* snapshots */
2235 if (has_data_file(bs) && s->nb_snapshots) {
2236 fprintf(stderr, "ERROR %d snapshots in image with data file\n",
2237 s->nb_snapshots);
2238 res->corruptions++;
2239 }
2240
2241 for (i = 0; i < s->nb_snapshots; i++) {
2242 sn = s->snapshots + i;
2243 if (offset_into_cluster(s, sn->l1_table_offset)) {
2244 fprintf(stderr, "ERROR snapshot %s (%s) l1_offset=%#" PRIx64 ": "
2245 "L1 table is not cluster aligned; snapshot table entry "
2246 "corrupted\n", sn->id_str, sn->name, sn->l1_table_offset);
2247 res->corruptions++;
2248 continue;
2249 }
2250 if (sn->l1_size > QCOW_MAX_L1_SIZE / L1E_SIZE) {
2251 fprintf(stderr, "ERROR snapshot %s (%s) l1_size=%#" PRIx32 ": "
2252 "L1 table is too large; snapshot table entry corrupted\n",
2253 sn->id_str, sn->name, sn->l1_size);
2254 res->corruptions++;
2255 continue;
2256 }
2257 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
2258 sn->l1_table_offset, sn->l1_size, 0, fix,
2259 false);
2260 if (ret < 0) {
2261 return ret;
2262 }
2263 }
2264 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2265 s->snapshots_offset, s->snapshots_size);
2266 if (ret < 0) {
2267 return ret;
2268 }
2269
2270 /* refcount data */
2271 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2272 s->refcount_table_offset,
2273 s->refcount_table_size *
2274 REFTABLE_ENTRY_SIZE);
2275 if (ret < 0) {
2276 return ret;
2277 }
2278
2279 /* encryption */
2280 if (s->crypto_header.length) {
2281 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2282 s->crypto_header.offset,
2283 s->crypto_header.length);
2284 if (ret < 0) {
2285 return ret;
2286 }
2287 }
2288
2289 /* bitmaps */
2290 ret = qcow2_check_bitmaps_refcounts(bs, res, refcount_table, nb_clusters);
2291 if (ret < 0) {
2292 return ret;
2293 }
2294
2295 return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters);
2296 }
2297
2298 /*
2299 * Compares the actual reference count for each cluster in the image against the
2300 * refcount as reported by the refcount structures on-disk.
2301 */
2302 static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2303 BdrvCheckMode fix, bool *rebuild,
2304 int64_t *highest_cluster,
2305 void *refcount_table, int64_t nb_clusters)
2306 {
2307 BDRVQcow2State *s = bs->opaque;
2308 int64_t i;
2309 uint64_t refcount1, refcount2;
2310 int ret;
2311
2312 for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) {
2313 ret = qcow2_get_refcount(bs, i, &refcount1);
2314 if (ret < 0) {
2315 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
2316 i, strerror(-ret));
2317 res->check_errors++;
2318 continue;
2319 }
2320
2321 refcount2 = s->get_refcount(refcount_table, i);
2322
2323 if (refcount1 > 0 || refcount2 > 0) {
2324 *highest_cluster = i;
2325 }
2326
2327 if (refcount1 != refcount2) {
2328 /* Check if we're allowed to fix the mismatch */
2329 int *num_fixed = NULL;
2330 if (refcount1 == 0) {
2331 *rebuild = true;
2332 } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
2333 num_fixed = &res->leaks_fixed;
2334 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
2335 num_fixed = &res->corruptions_fixed;
2336 }
2337
2338 fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64
2339 " reference=%" PRIu64 "\n",
2340 num_fixed != NULL ? "Repairing" :
2341 refcount1 < refcount2 ? "ERROR" :
2342 "Leaked",
2343 i, refcount1, refcount2);
2344
2345 if (num_fixed) {
2346 ret = update_refcount(bs, i << s->cluster_bits, 1,
2347 refcount_diff(refcount1, refcount2),
2348 refcount1 > refcount2,
2349 QCOW2_DISCARD_ALWAYS);
2350 if (ret >= 0) {
2351 (*num_fixed)++;
2352 continue;
2353 }
2354 }
2355
2356 /* And if we couldn't, print an error */
2357 if (refcount1 < refcount2) {
2358 res->corruptions++;
2359 } else {
2360 res->leaks++;
2361 }
2362 }
2363 }
2364 }
2365
2366 /*
2367 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to
2368 * the on-disk refcount structures.
2369 *
2370 * On input, *first_free_cluster tells where to start looking, and need not
2371 * actually be a free cluster; the returned offset will not be before that
2372 * cluster. On output, *first_free_cluster points to the first gap found, even
2373 * if that gap was too small to be used as the returned offset.
2374 *
2375 * Note that *first_free_cluster is a cluster index whereas the return value is
2376 * an offset.
2377 */
2378 static int64_t alloc_clusters_imrt(BlockDriverState *bs,
2379 int cluster_count,
2380 void **refcount_table,
2381 int64_t *imrt_nb_clusters,
2382 int64_t *first_free_cluster)
2383 {
2384 BDRVQcow2State *s = bs->opaque;
2385 int64_t cluster = *first_free_cluster, i;
2386 bool first_gap = true;
2387 int contiguous_free_clusters;
2388 int ret;
2389
2390 /* Starting at *first_free_cluster, find a range of at least cluster_count
2391 * continuously free clusters */
2392 for (contiguous_free_clusters = 0;
2393 cluster < *imrt_nb_clusters &&
2394 contiguous_free_clusters < cluster_count;
2395 cluster++)
2396 {
2397 if (!s->get_refcount(*refcount_table, cluster)) {
2398 contiguous_free_clusters++;
2399 if (first_gap) {
2400 /* If this is the first free cluster found, update
2401 * *first_free_cluster accordingly */
2402 *first_free_cluster = cluster;
2403 first_gap = false;
2404 }
2405 } else if (contiguous_free_clusters) {
2406 contiguous_free_clusters = 0;
2407 }
2408 }
2409
2410 /* If contiguous_free_clusters is greater than zero, it contains the number
2411 * of continuously free clusters until the current cluster; the first free
2412 * cluster in the current "gap" is therefore
2413 * cluster - contiguous_free_clusters */
2414
2415 /* If no such range could be found, grow the in-memory refcount table
2416 * accordingly to append free clusters at the end of the image */
2417 if (contiguous_free_clusters < cluster_count) {
2418 /* contiguous_free_clusters clusters are already empty at the image end;
2419 * we need cluster_count clusters; therefore, we have to allocate
2420 * cluster_count - contiguous_free_clusters new clusters at the end of
2421 * the image (which is the current value of cluster; note that cluster
2422 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond
2423 * the image end) */
2424 ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters,
2425 cluster + cluster_count
2426 - contiguous_free_clusters);
2427 if (ret < 0) {
2428 return ret;
2429 }
2430 }
2431
2432 /* Go back to the first free cluster */
2433 cluster -= contiguous_free_clusters;
2434 for (i = 0; i < cluster_count; i++) {
2435 s->set_refcount(*refcount_table, cluster + i, 1);
2436 }
2437
2438 return cluster << s->cluster_bits;
2439 }
2440
2441 /*
2442 * Helper function for rebuild_refcount_structure().
2443 *
2444 * Scan the range of clusters [first_cluster, end_cluster) for allocated
2445 * clusters and write all corresponding refblocks to disk. The refblock
2446 * and allocation data is taken from the in-memory refcount table
2447 * *refcount_table[] (of size *nb_clusters), which is basically one big
2448 * (unlimited size) refblock for the whole image.
2449 *
2450 * For these refblocks, clusters are allocated using said in-memory
2451 * refcount table. Care is taken that these allocations are reflected
2452 * in the refblocks written to disk.
2453 *
2454 * The refblocks' offsets are written into a reftable, which is
2455 * *on_disk_reftable_ptr[] (of size *on_disk_reftable_entries_ptr). If
2456 * that reftable is of insufficient size, it will be resized to fit.
2457 * This reftable is not written to disk.
2458 *
2459 * (If *on_disk_reftable_ptr is not NULL, the entries within are assumed
2460 * to point to existing valid refblocks that do not need to be allocated
2461 * again.)
2462 *
2463 * Return whether the on-disk reftable array was resized (true/false),
2464 * or -errno on error.
2465 */
2466 static int rebuild_refcounts_write_refblocks(
2467 BlockDriverState *bs, void **refcount_table, int64_t *nb_clusters,
2468 int64_t first_cluster, int64_t end_cluster,
2469 uint64_t **on_disk_reftable_ptr, uint32_t *on_disk_reftable_entries_ptr,
2470 Error **errp
2471 )
2472 {
2473 BDRVQcow2State *s = bs->opaque;
2474 int64_t cluster;
2475 int64_t refblock_offset, refblock_start, refblock_index;
2476 int64_t first_free_cluster = 0;
2477 uint64_t *on_disk_reftable = *on_disk_reftable_ptr;
2478 uint32_t on_disk_reftable_entries = *on_disk_reftable_entries_ptr;
2479 void *on_disk_refblock;
2480 bool reftable_grown = false;
2481 int ret;
2482
2483 for (cluster = first_cluster; cluster < end_cluster; cluster++) {
2484 /* Check all clusters to find refblocks that contain non-zero entries */
2485 if (!s->get_refcount(*refcount_table, cluster)) {
2486 continue;
2487 }
2488
2489 /*
2490 * This cluster is allocated, so we need to create a refblock
2491 * for it. The data we will write to disk is just the
2492 * respective slice from *refcount_table, so it will contain
2493 * accurate refcounts for all clusters belonging to this
2494 * refblock. After we have written it, we will therefore skip
2495 * all remaining clusters in this refblock.
2496 */
2497
2498 refblock_index = cluster >> s->refcount_block_bits;
2499 refblock_start = refblock_index << s->refcount_block_bits;
2500
2501 if (on_disk_reftable_entries > refblock_index &&
2502 on_disk_reftable[refblock_index])
2503 {
2504 /*
2505 * We can get here after a `goto write_refblocks`: We have a
2506 * reftable from a previous run, and the refblock is already
2507 * allocated. No need to allocate it again.
2508 */
2509 refblock_offset = on_disk_reftable[refblock_index];
2510 } else {
2511 int64_t refblock_cluster_index;
2512
2513 /* Don't allocate a cluster in a refblock already written to disk */
2514 if (first_free_cluster < refblock_start) {
2515 first_free_cluster = refblock_start;
2516 }
2517 refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table,
2518 nb_clusters,
2519 &first_free_cluster);
2520 if (refblock_offset < 0) {
2521 error_setg_errno(errp, -refblock_offset,
2522 "ERROR allocating refblock");
2523 return refblock_offset;
2524 }
2525
2526 refblock_cluster_index = refblock_offset / s->cluster_size;
2527 if (refblock_cluster_index >= end_cluster) {
2528 /*
2529 * We must write the refblock that holds this refblock's
2530 * refcount
2531 */
2532 end_cluster = refblock_cluster_index + 1;
2533 }
2534
2535 if (on_disk_reftable_entries <= refblock_index) {
2536 on_disk_reftable_entries =
2537 ROUND_UP((refblock_index + 1) * REFTABLE_ENTRY_SIZE,
2538 s->cluster_size) / REFTABLE_ENTRY_SIZE;
2539 on_disk_reftable =
2540 g_try_realloc(on_disk_reftable,
2541 on_disk_reftable_entries *
2542 REFTABLE_ENTRY_SIZE);
2543 if (!on_disk_reftable) {
2544 error_setg(errp, "ERROR allocating reftable memory");
2545 return -ENOMEM;
2546 }
2547
2548 memset(on_disk_reftable + *on_disk_reftable_entries_ptr, 0,
2549 (on_disk_reftable_entries -
2550 *on_disk_reftable_entries_ptr) *
2551 REFTABLE_ENTRY_SIZE);
2552
2553 *on_disk_reftable_ptr = on_disk_reftable;
2554 *on_disk_reftable_entries_ptr = on_disk_reftable_entries;
2555
2556 reftable_grown = true;
2557 } else {
2558 assert(on_disk_reftable);
2559 }
2560 on_disk_reftable[refblock_index] = refblock_offset;
2561 }
2562
2563 /* Refblock is allocated, write it to disk */
2564
2565 ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset,
2566 s->cluster_size, false);
2567 if (ret < 0) {
2568 error_setg_errno(errp, -ret, "ERROR writing refblock");
2569 return ret;
2570 }
2571
2572 /*
2573 * The refblock is simply a slice of *refcount_table.
2574 * Note that the size of *refcount_table is always aligned to
2575 * whole clusters, so the write operation will not result in
2576 * out-of-bounds accesses.
2577 */
2578 on_disk_refblock = (void *)((char *) *refcount_table +
2579 refblock_index * s->cluster_size);
2580
2581 ret = bdrv_pwrite(bs->file, refblock_offset, s->cluster_size,
2582 on_disk_refblock, 0);
2583 if (ret < 0) {
2584 error_setg_errno(errp, -ret, "ERROR writing refblock");
2585 return ret;
2586 }
2587
2588 /* This refblock is done, skip to its end */
2589 cluster = refblock_start + s->refcount_block_size - 1;
2590 }
2591
2592 return reftable_grown;
2593 }
2594
2595 /*
2596 * Creates a new refcount structure based solely on the in-memory information
2597 * given through *refcount_table (this in-memory information is basically just
2598 * the concatenation of all refblocks). All necessary allocations will be
2599 * reflected in that array.
2600 *
2601 * On success, the old refcount structure is leaked (it will be covered by the
2602 * new refcount structure).
2603 */
2604 static int rebuild_refcount_structure(BlockDriverState *bs,
2605 BdrvCheckResult *res,
2606 void **refcount_table,
2607 int64_t *nb_clusters,
2608 Error **errp)
2609 {
2610 BDRVQcow2State *s = bs->opaque;
2611 int64_t reftable_offset = -1;
2612 int64_t reftable_length = 0;
2613 int64_t reftable_clusters;
2614 int64_t refblock_index;
2615 uint32_t on_disk_reftable_entries = 0;
2616 uint64_t *on_disk_reftable = NULL;
2617 int ret = 0;
2618 int reftable_size_changed = 0;
2619 struct {
2620 uint64_t reftable_offset;
2621 uint32_t reftable_clusters;
2622 } QEMU_PACKED reftable_offset_and_clusters;
2623
2624 qcow2_cache_empty(bs, s->refcount_block_cache);
2625
2626 /*
2627 * For each refblock containing entries, we try to allocate a
2628 * cluster (in the in-memory refcount table) and write its offset
2629 * into on_disk_reftable[]. We then write the whole refblock to
2630 * disk (as a slice of the in-memory refcount table).
2631 * This is done by rebuild_refcounts_write_refblocks().
2632 *
2633 * Once we have scanned all clusters, we try to find space for the
2634 * reftable. This will dirty the in-memory refcount table (i.e.
2635 * make it differ from the refblocks we have already written), so we
2636 * need to run rebuild_refcounts_write_refblocks() again for the
2637 * range of clusters where the reftable has been allocated.
2638 *
2639 * This second run might make the reftable grow again, in which case
2640 * we will need to allocate another space for it, which is why we
2641 * repeat all this until the reftable stops growing.
2642 *
2643 * (This loop will terminate, because with every cluster the
2644 * reftable grows, it can accomodate a multitude of more refcounts,
2645 * so that at some point this must be able to cover the reftable
2646 * and all refblocks describing it.)
2647 *
2648 * We then convert the reftable to big-endian and write it to disk.
2649 *
2650 * Note that we never free any reftable allocations. Doing so would
2651 * needlessly complicate the algorithm: The eventual second check
2652 * run we do will clean up all leaks we have caused.
2653 */
2654
2655 reftable_size_changed =
2656 rebuild_refcounts_write_refblocks(bs, refcount_table, nb_clusters,
2657 0, *nb_clusters,
2658 &on_disk_reftable,
2659 &on_disk_reftable_entries, errp);
2660 if (reftable_size_changed < 0) {
2661 res->check_errors++;
2662 ret = reftable_size_changed;
2663 goto fail;
2664 }
2665
2666 /*
2667 * There was no reftable before, so rebuild_refcounts_write_refblocks()
2668 * must have increased its size (from 0 to something).
2669 */
2670 assert(reftable_size_changed);
2671
2672 do {
2673 int64_t reftable_start_cluster, reftable_end_cluster;
2674 int64_t first_free_cluster = 0;
2675
2676 reftable_length = on_disk_reftable_entries * REFTABLE_ENTRY_SIZE;
2677 reftable_clusters = size_to_clusters(s, reftable_length);
2678
2679 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2680 refcount_table, nb_clusters,
2681 &first_free_cluster);
2682 if (reftable_offset < 0) {
2683 error_setg_errno(errp, -reftable_offset,
2684 "ERROR allocating reftable");
2685 res->check_errors++;
2686 ret = reftable_offset;
2687 goto fail;
2688 }
2689
2690 /*
2691 * We need to update the affected refblocks, so re-run the
2692 * write_refblocks loop for the reftable's range of clusters.
2693 */
2694 assert(offset_into_cluster(s, reftable_offset) == 0);
2695 reftable_start_cluster = reftable_offset / s->cluster_size;
2696 reftable_end_cluster = reftable_start_cluster + reftable_clusters;
2697 reftable_size_changed =
2698 rebuild_refcounts_write_refblocks(bs, refcount_table, nb_clusters,
2699 reftable_start_cluster,
2700 reftable_end_cluster,
2701 &on_disk_reftable,
2702 &on_disk_reftable_entries, errp);
2703 if (reftable_size_changed < 0) {
2704 res->check_errors++;
2705 ret = reftable_size_changed;
2706 goto fail;
2707 }
2708
2709 /*
2710 * If the reftable size has changed, we will need to find a new
2711 * allocation, repeating the loop.
2712 */
2713 } while (reftable_size_changed);
2714
2715 /* The above loop must have run at least once */
2716 assert(reftable_offset >= 0);
2717
2718 /*
2719 * All allocations are done, all refblocks are written, convert the
2720 * reftable to big-endian and write it to disk.
2721 */
2722
2723 for (refblock_index = 0; refblock_index < on_disk_reftable_entries;
2724 refblock_index++)
2725 {
2726 cpu_to_be64s(&on_disk_reftable[refblock_index]);
2727 }
2728
2729 ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset, reftable_length,
2730 false);
2731 if (ret < 0) {
2732 error_setg_errno(errp, -ret, "ERROR writing reftable");
2733 goto fail;
2734 }
2735
2736 assert(reftable_length < INT_MAX);
2737 ret = bdrv_pwrite(bs->file, reftable_offset, reftable_length,
2738 on_disk_reftable, 0);
2739 if (ret < 0) {
2740 error_setg_errno(errp, -ret, "ERROR writing reftable");
2741 goto fail;
2742 }
2743
2744 /* Enter new reftable into the image header */
2745 reftable_offset_and_clusters.reftable_offset = cpu_to_be64(reftable_offset);
2746 reftable_offset_and_clusters.reftable_clusters =
2747 cpu_to_be32(reftable_clusters);
2748 ret = bdrv_pwrite_sync(bs->file,
2749 offsetof(QCowHeader, refcount_table_offset),
2750 sizeof(reftable_offset_and_clusters),
2751 &reftable_offset_and_clusters, 0);
2752 if (ret < 0) {
2753 error_setg_errno(errp, -ret, "ERROR setting reftable");
2754 goto fail;
2755 }
2756
2757 for (refblock_index = 0; refblock_index < on_disk_reftable_entries;
2758 refblock_index++)
2759 {
2760 be64_to_cpus(&on_disk_reftable[refblock_index]);
2761 }
2762 s->refcount_table = on_disk_reftable;
2763 s->refcount_table_offset = reftable_offset;
2764 s->refcount_table_size = on_disk_reftable_entries;
2765 update_max_refcount_table_index(s);
2766
2767 return 0;
2768
2769 fail:
2770 g_free(on_disk_reftable);
2771 return ret;
2772 }
2773
2774 /*
2775 * Checks an image for refcount consistency.
2776 *
2777 * Returns 0 if no errors are found, the number of errors in case the image is
2778 * detected as corrupted, and -errno when an internal error occurred.
2779 */
2780 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2781 BdrvCheckMode fix)
2782 {
2783 BDRVQcow2State *s = bs->opaque;
2784 BdrvCheckResult pre_compare_res;
2785 int64_t size, highest_cluster, nb_clusters;
2786 void *refcount_table = NULL;
2787 bool rebuild = false;
2788 int ret;
2789
2790 size = bdrv_getlength(bs->file->bs);
2791 if (size < 0) {
2792 res->check_errors++;
2793 return size;
2794 }
2795
2796 nb_clusters = size_to_clusters(s, size);
2797 if (nb_clusters > INT_MAX) {
2798 res->check_errors++;
2799 return -EFBIG;
2800 }
2801
2802 res->bfi.total_clusters =
2803 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE);
2804
2805 ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table,
2806 &nb_clusters);
2807 if (ret < 0) {
2808 goto fail;
2809 }
2810
2811 /* In case we don't need to rebuild the refcount structure (but want to fix
2812 * something), this function is immediately called again, in which case the
2813 * result should be ignored */
2814 pre_compare_res = *res;
2815 compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table,
2816 nb_clusters);
2817
2818 if (rebuild && (fix & BDRV_FIX_ERRORS)) {
2819 BdrvCheckResult old_res = *res;
2820 int fresh_leaks = 0;
2821 Error *local_err = NULL;
2822
2823 fprintf(stderr, "Rebuilding refcount structure\n");
2824 ret = rebuild_refcount_structure(bs, res, &refcount_table,
2825 &nb_clusters, &local_err);
2826 if (ret < 0) {
2827 error_report_err(local_err);
2828 goto fail;
2829 }
2830
2831 res->corruptions = 0;
2832 res->leaks = 0;
2833
2834 /* Because the old reftable has been exchanged for a new one the
2835 * references have to be recalculated */
2836 rebuild = false;
2837 memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters));
2838 ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table,
2839 &nb_clusters);
2840 if (ret < 0) {
2841 goto fail;
2842 }
2843
2844 if (fix & BDRV_FIX_LEAKS) {
2845 /* The old refcount structures are now leaked, fix it; the result
2846 * can be ignored, aside from leaks which were introduced by
2847 * rebuild_refcount_structure() that could not be fixed */
2848 BdrvCheckResult saved_res = *res;
2849 *res = (BdrvCheckResult){ 0 };
2850
2851 compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild,
2852 &highest_cluster, refcount_table, nb_clusters);
2853 if (rebuild) {
2854 fprintf(stderr, "ERROR rebuilt refcount structure is still "
2855 "broken\n");
2856 }
2857
2858 /* Any leaks accounted for here were introduced by
2859 * rebuild_refcount_structure() because that function has created a
2860 * new refcount structure from scratch */
2861 fresh_leaks = res->leaks;
2862 *res = saved_res;
2863 }
2864
2865 if (res->corruptions < old_res.corruptions) {
2866 res->corruptions_fixed += old_res.corruptions - res->corruptions;
2867 }
2868 if (res->leaks < old_res.leaks) {
2869 res->leaks_fixed += old_res.leaks - res->leaks;
2870 }
2871 res->leaks += fresh_leaks;
2872 } else if (fix) {
2873 if (rebuild) {
2874 fprintf(stderr, "ERROR need to rebuild refcount structures\n");
2875 res->check_errors++;
2876 ret = -EIO;
2877 goto fail;
2878 }
2879
2880 if (res->leaks || res->corruptions) {
2881 *res = pre_compare_res;
2882 compare_refcounts(bs, res, fix, &rebuild, &highest_cluster,
2883 refcount_table, nb_clusters);
2884 }
2885 }
2886
2887 /* check OFLAG_COPIED */
2888 ret = check_oflag_copied(bs, res, fix);
2889 if (ret < 0) {
2890 goto fail;
2891 }
2892
2893 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
2894 ret = 0;
2895
2896 fail:
2897 g_free(refcount_table);
2898
2899 return ret;
2900 }
2901
2902 #define overlaps_with(ofs, sz) \
2903 ranges_overlap(offset, size, ofs, sz)
2904
2905 /*
2906 * Checks if the given offset into the image file is actually free to use by
2907 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
2908 * i.e. a sanity check without relying on the refcount tables.
2909 *
2910 * The ign parameter specifies what checks not to perform (being a bitmask of
2911 * QCow2MetadataOverlap values), i.e., what sections to ignore.
2912 *
2913 * Returns:
2914 * - 0 if writing to this offset will not affect the mentioned metadata
2915 * - a positive QCow2MetadataOverlap value indicating one overlapping section
2916 * - a negative value (-errno) indicating an error while performing a check,
2917 * e.g. when bdrv_pread failed on QCOW2_OL_INACTIVE_L2
2918 */
2919 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset,
2920 int64_t size)
2921 {
2922 BDRVQcow2State *s = bs->opaque;
2923 int chk = s->overlap_check & ~ign;
2924 int i, j;
2925
2926 if (!size) {
2927 return 0;
2928 }
2929
2930 if (chk & QCOW2_OL_MAIN_HEADER) {
2931 if (offset < s->cluster_size) {
2932 return QCOW2_OL_MAIN_HEADER;
2933 }
2934 }
2935
2936 /* align range to test to cluster boundaries */
2937 size = ROUND_UP(offset_into_cluster(s, offset) + size, s->cluster_size);
2938 offset = start_of_cluster(s, offset);
2939
2940 if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) {
2941 if (overlaps_with(s->l1_table_offset, s->l1_size * L1E_SIZE)) {
2942 return QCOW2_OL_ACTIVE_L1;
2943 }
2944 }
2945
2946 if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) {
2947 if (overlaps_with(s->refcount_table_offset,
2948 s->refcount_table_size * REFTABLE_ENTRY_SIZE)) {
2949 return QCOW2_OL_REFCOUNT_TABLE;
2950 }
2951 }
2952
2953 if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) {
2954 if (overlaps_with(s->snapshots_offset, s->snapshots_size)) {
2955 return QCOW2_OL_SNAPSHOT_TABLE;
2956 }
2957 }
2958
2959 if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) {
2960 for (i = 0; i < s->nb_snapshots; i++) {
2961 if (s->snapshots[i].l1_size &&
2962 overlaps_with(s->snapshots[i].l1_table_offset,
2963 s->snapshots[i].l1_size * L1E_SIZE)) {
2964 return QCOW2_OL_INACTIVE_L1;
2965 }
2966 }
2967 }
2968
2969 if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) {
2970 for (i = 0; i < s->l1_size; i++) {
2971 if ((s->l1_table[i] & L1E_OFFSET_MASK) &&
2972 overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK,
2973 s->cluster_size)) {
2974 return QCOW2_OL_ACTIVE_L2;
2975 }
2976 }
2977 }
2978
2979 if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) {
2980 unsigned last_entry = s->max_refcount_table_index;
2981 assert(last_entry < s->refcount_table_size);
2982 assert(last_entry + 1 == s->refcount_table_size ||
2983 (s->refcount_table[last_entry + 1] & REFT_OFFSET_MASK) == 0);
2984 for (i = 0; i <= last_entry; i++) {
2985 if ((s->refcount_table[i] & REFT_OFFSET_MASK) &&
2986 overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK,
2987 s->cluster_size)) {
2988 return QCOW2_OL_REFCOUNT_BLOCK;
2989 }
2990 }
2991 }
2992
2993 if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) {
2994 for (i = 0; i < s->nb_snapshots; i++) {
2995 uint64_t l1_ofs = s->snapshots[i].l1_table_offset;
2996 uint32_t l1_sz = s->snapshots[i].l1_size;
2997 uint64_t l1_sz2 = l1_sz * L1E_SIZE;
2998 uint64_t *l1;
2999 int ret;
3000
3001 ret = qcow2_validate_table(bs, l1_ofs, l1_sz, L1E_SIZE,
3002 QCOW_MAX_L1_SIZE, "", NULL);
3003 if (ret < 0) {
3004 return ret;
3005 }
3006
3007 l1 = g_try_malloc(l1_sz2);
3008
3009 if (l1_sz2 && l1 == NULL) {
3010 return -ENOMEM;
3011 }
3012
3013 ret = bdrv_pread(bs->file, l1_ofs, l1_sz2, l1, 0);
3014 if (ret < 0) {
3015 g_free(l1);
3016 return ret;
3017 }
3018
3019 for (j = 0; j < l1_sz; j++) {
3020 uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK;
3021 if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) {
3022 g_free(l1);
3023 return QCOW2_OL_INACTIVE_L2;
3024 }
3025 }
3026
3027 g_free(l1);
3028 }
3029 }
3030
3031 if ((chk & QCOW2_OL_BITMAP_DIRECTORY) &&
3032 (s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS))
3033 {
3034 if (overlaps_with(s->bitmap_directory_offset,
3035 s->bitmap_directory_size))
3036 {
3037 return QCOW2_OL_BITMAP_DIRECTORY;
3038 }
3039 }
3040
3041 return 0;
3042 }
3043
3044 static const char *metadata_ol_names[] = {
3045 [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header",
3046 [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table",
3047 [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table",
3048 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table",
3049 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block",
3050 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table",
3051 [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table",
3052 [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table",
3053 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = "bitmap directory",
3054 };
3055 QEMU_BUILD_BUG_ON(QCOW2_OL_MAX_BITNR != ARRAY_SIZE(metadata_ol_names));
3056
3057 /*
3058 * First performs a check for metadata overlaps (through
3059 * qcow2_check_metadata_overlap); if that fails with a negative value (error
3060 * while performing a check), that value is returned. If an impending overlap
3061 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
3062 * and -EIO returned.
3063 *
3064 * Returns 0 if there were neither overlaps nor errors while checking for
3065 * overlaps; or a negative value (-errno) on error.
3066 */
3067 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset,
3068 int64_t size, bool data_file)
3069 {
3070 int ret;
3071
3072 if (data_file && has_data_file(bs)) {
3073 return 0;
3074 }
3075
3076 ret = qcow2_check_metadata_overlap(bs, ign, offset, size);
3077 if (ret < 0) {
3078 return ret;
3079 } else if (ret > 0) {
3080 int metadata_ol_bitnr = ctz32(ret);
3081 assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR);
3082
3083 qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid "
3084 "write on metadata (overlaps with %s)",
3085 metadata_ol_names[metadata_ol_bitnr]);
3086 return -EIO;
3087 }
3088
3089 return 0;
3090 }
3091
3092 /* A pointer to a function of this type is given to walk_over_reftable(). That
3093 * function will create refblocks and pass them to a RefblockFinishOp once they
3094 * are completed (@refblock). @refblock_empty is set if the refblock is
3095 * completely empty.
3096 *
3097 * Along with the refblock, a corresponding reftable entry is passed, in the
3098 * reftable @reftable (which may be reallocated) at @reftable_index.
3099 *
3100 * @allocated should be set to true if a new cluster has been allocated.
3101 */
3102 typedef int (RefblockFinishOp)(BlockDriverState *bs, uint64_t **reftable,
3103 uint64_t reftable_index, uint64_t *reftable_size,
3104 void *refblock, bool refblock_empty,
3105 bool *allocated, Error **errp);
3106
3107 /**
3108 * This "operation" for walk_over_reftable() allocates the refblock on disk (if
3109 * it is not empty) and inserts its offset into the new reftable. The size of
3110 * this new reftable is increased as required.
3111 */
3112 static int alloc_refblock(BlockDriverState *bs, uint64_t **reftable,
3113 uint64_t reftable_index, uint64_t *reftable_size,
3114 void *refblock, bool refblock_empty, bool *allocated,
3115 Error **errp)
3116 {
3117 BDRVQcow2State *s = bs->opaque;
3118 int64_t offset;
3119
3120 if (!refblock_empty && reftable_index >= *reftable_size) {
3121 uint64_t *new_reftable;
3122 uint64_t new_reftable_size;
3123
3124 new_reftable_size = ROUND_UP(reftable_index + 1,
3125 s->cluster_size / REFTABLE_ENTRY_SIZE);
3126 if (new_reftable_size > QCOW_MAX_REFTABLE_SIZE / REFTABLE_ENTRY_SIZE) {
3127 error_setg(errp,
3128 "This operation would make the refcount table grow "
3129 "beyond the maximum size supported by QEMU, aborting");
3130 return -ENOTSUP;
3131 }
3132
3133 new_reftable = g_try_realloc(*reftable, new_reftable_size *
3134 REFTABLE_ENTRY_SIZE);
3135 if (!new_reftable) {
3136 error_setg(errp, "Failed to increase reftable buffer size");
3137 return -ENOMEM;
3138 }
3139
3140 memset(new_reftable + *reftable_size, 0,
3141 (new_reftable_size - *reftable_size) * REFTABLE_ENTRY_SIZE);
3142
3143 *reftable = new_reftable;
3144 *reftable_size = new_reftable_size;
3145 }
3146
3147 if (!refblock_empty && !(*reftable)[reftable_index]) {
3148 offset = qcow2_alloc_clusters(bs, s->cluster_size);
3149 if (offset < 0) {
3150 error_setg_errno(errp, -offset, "Failed to allocate refblock");
3151 return offset;
3152 }
3153 (*reftable)[reftable_index] = offset;
3154 *allocated = true;
3155 }
3156
3157 return 0;
3158 }
3159
3160 /**
3161 * This "operation" for walk_over_reftable() writes the refblock to disk at the
3162 * offset specified by the new reftable's entry. It does not modify the new
3163 * reftable or change any refcounts.
3164 */
3165 static int flush_refblock(BlockDriverState *bs, uint64_t **reftable,
3166 uint64_t reftable_index, uint64_t *reftable_size,
3167 void *refblock, bool refblock_empty, bool *allocated,
3168 Error **errp)
3169 {
3170 BDRVQcow2State *s = bs->opaque;
3171 int64_t offset;
3172 int ret;
3173
3174 if (reftable_index < *reftable_size && (*reftable)[reftable_index]) {
3175 offset = (*reftable)[reftable_index];
3176
3177 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size,
3178 false);
3179 if (ret < 0) {
3180 error_setg_errno(errp, -ret, "Overlap check failed");
3181 return ret;
3182 }
3183
3184 ret = bdrv_pwrite(bs->file, offset, s->cluster_size, refblock, 0);
3185 if (ret < 0) {
3186 error_setg_errno(errp, -ret, "Failed to write refblock");
3187 return ret;
3188 }
3189 } else {
3190 assert(refblock_empty);
3191 }
3192
3193 return 0;
3194 }
3195
3196 /**
3197 * This function walks over the existing reftable and every referenced refblock;
3198 * if @new_set_refcount is non-NULL, it is called for every refcount entry to
3199 * create an equal new entry in the passed @new_refblock. Once that
3200 * @new_refblock is completely filled, @operation will be called.
3201 *
3202 * @status_cb and @cb_opaque are used for the amend operation's status callback.
3203 * @index is the index of the walk_over_reftable() calls and @total is the total
3204 * number of walk_over_reftable() calls per amend operation. Both are used for
3205 * calculating the parameters for the status callback.
3206 *
3207 * @allocated is set to true if a new cluster has been allocated.
3208 */
3209 static int walk_over_reftable(BlockDriverState *bs, uint64_t **new_reftable,
3210 uint64_t *new_reftable_index,
3211 uint64_t *new_reftable_size,
3212 void *new_refblock, int new_refblock_size,
3213 int new_refcount_bits,
3214 RefblockFinishOp *operation, bool *allocated,
3215 Qcow2SetRefcountFunc *new_set_refcount,
3216 BlockDriverAmendStatusCB *status_cb,
3217 void *cb_opaque, int index, int total,
3218 Error **errp)
3219 {
3220 BDRVQcow2State *s = bs->opaque;
3221 uint64_t reftable_index;
3222 bool new_refblock_empty = true;
3223 int refblock_index;
3224 int new_refblock_index = 0;
3225 int ret;
3226
3227 for (reftable_index = 0; reftable_index < s->refcount_table_size;
3228 reftable_index++)
3229 {
3230 uint64_t refblock_offset = s->refcount_table[reftable_index]
3231 & REFT_OFFSET_MASK;
3232
3233 status_cb(bs, (uint64_t)index * s->refcount_table_size + reftable_index,
3234 (uint64_t)total * s->refcount_table_size, cb_opaque);
3235
3236 if (refblock_offset) {
3237 void *refblock;
3238
3239 if (offset_into_cluster(s, refblock_offset)) {
3240 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
3241 PRIx64 " unaligned (reftable index: %#"
3242 PRIx64 ")", refblock_offset,
3243 reftable_index);
3244 error_setg(errp,
3245 "Image is corrupt (unaligned refblock offset)");
3246 return -EIO;
3247 }
3248
3249 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offset,
3250 &refblock);
3251 if (ret < 0) {
3252 error_setg_errno(errp, -ret, "Failed to retrieve refblock");
3253 return ret;
3254 }
3255
3256 for (refblock_index = 0; refblock_index < s->refcount_block_size;
3257 refblock_index++)
3258 {
3259 uint64_t refcount;
3260
3261 if (new_refblock_index >= new_refblock_size) {
3262 /* new_refblock is now complete */
3263 ret = operation(bs, new_reftable, *new_reftable_index,
3264 new_reftable_size, new_refblock,
3265 new_refblock_empty, allocated, errp);
3266 if (ret < 0) {
3267 qcow2_cache_put(s->refcount_block_cache, &refblock);
3268 return ret;
3269 }
3270
3271 (*new_reftable_index)++;
3272 new_refblock_index = 0;
3273 new_refblock_empty = true;
3274 }
3275
3276 refcount = s->get_refcount(refblock, refblock_index);
3277 if (new_refcount_bits < 64 && refcount >> new_refcount_bits) {
3278 uint64_t offset;
3279
3280 qcow2_cache_put(s->refcount_block_cache, &refblock);
3281
3282 offset = ((reftable_index << s->refcount_block_bits)
3283 + refblock_index) << s->cluster_bits;
3284
3285 error_setg(errp, "Cannot decrease refcount entry width to "
3286 "%i bits: Cluster at offset %#" PRIx64 " has a "
3287 "refcount of %" PRIu64, new_refcount_bits,
3288 offset, refcount);
3289 return -EINVAL;
3290 }
3291
3292 if (new_set_refcount) {
3293 new_set_refcount(new_refblock, new_refblock_index++,
3294 refcount);
3295 } else {
3296 new_refblock_index++;
3297 }
3298 new_refblock_empty = new_refblock_empty && refcount == 0;
3299 }
3300
3301 qcow2_cache_put(s->refcount_block_cache, &refblock);
3302 } else {
3303 /* No refblock means every refcount is 0 */
3304 for (refblock_index = 0; refblock_index < s->refcount_block_size;
3305 refblock_index++)
3306 {
3307 if (new_refblock_index >= new_refblock_size) {
3308 /* new_refblock is now complete */
3309 ret = operation(bs, new_reftable, *new_reftable_index,
3310 new_reftable_size, new_refblock,
3311 new_refblock_empty, allocated, errp);
3312 if (ret < 0) {
3313 return ret;
3314 }
3315
3316 (*new_reftable_index)++;
3317 new_refblock_index = 0;
3318 new_refblock_empty = true;
3319 }
3320
3321 if (new_set_refcount) {
3322 new_set_refcount(new_refblock, new_refblock_index++, 0);
3323 } else {
3324 new_refblock_index++;
3325 }
3326 }
3327 }
3328 }
3329
3330 if (new_refblock_index > 0) {
3331 /* Complete the potentially existing partially filled final refblock */
3332 if (new_set_refcount) {
3333 for (; new_refblock_index < new_refblock_size;
3334 new_refblock_index++)
3335 {
3336 new_set_refcount(new_refblock, new_refblock_index, 0);
3337 }
3338 }
3339
3340 ret = operation(bs, new_reftable, *new_reftable_index,
3341 new_reftable_size, new_refblock, new_refblock_empty,
3342 allocated, errp);
3343 if (ret < 0) {
3344 return ret;
3345 }
3346
3347 (*new_reftable_index)++;
3348 }
3349
3350 status_cb(bs, (uint64_t)(index + 1) * s->refcount_table_size,
3351 (uint64_t)total * s->refcount_table_size, cb_opaque);
3352
3353 return 0;
3354 }
3355
3356 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order,
3357 BlockDriverAmendStatusCB *status_cb,
3358 void *cb_opaque, Error **errp)
3359 {
3360 BDRVQcow2State *s = bs->opaque;
3361 Qcow2GetRefcountFunc *new_get_refcount;
3362 Qcow2SetRefcountFunc *new_set_refcount;
3363 void *new_refblock = qemu_blockalign(bs->file->bs, s->cluster_size);
3364 uint64_t *new_reftable = NULL, new_reftable_size = 0;
3365 uint64_t *old_reftable, old_reftable_size, old_reftable_offset;
3366 uint64_t new_reftable_index = 0;
3367 uint64_t i;
3368 int64_t new_reftable_offset = 0, allocated_reftable_size = 0;
3369 int new_refblock_size, new_refcount_bits = 1 << refcount_order;
3370 int old_refcount_order;
3371 int walk_index = 0;
3372 int ret;
3373 bool new_allocation;
3374
3375 assert(s->qcow_version >= 3);
3376 assert(refcount_order >= 0 && refcount_order <= 6);
3377
3378 /* see qcow2_open() */
3379 new_refblock_size = 1 << (s->cluster_bits - (refcount_order - 3));
3380
3381 new_get_refcount = get_refcount_funcs[refcount_order];
3382 new_set_refcount = set_refcount_funcs[refcount_order];
3383
3384
3385 do {
3386 int total_walks;
3387
3388 new_allocation = false;
3389
3390 /* At least we have to do this walk and the one which writes the
3391 * refblocks; also, at least we have to do this loop here at least
3392 * twice (normally), first to do the allocations, and second to
3393 * determine that everything is correctly allocated, this then makes
3394 * three walks in total */
3395 total_walks = MAX(walk_index + 2, 3);
3396
3397 /* First, allocate the structures so they are present in the refcount
3398 * structures */
3399 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
3400 &new_reftable_size, NULL, new_refblock_size,
3401 new_refcount_bits, &alloc_refblock,
3402 &new_allocation, NULL, status_cb, cb_opaque,
3403 walk_index++, total_walks, errp);
3404 if (ret < 0) {
3405 goto done;
3406 }
3407
3408 new_reftable_index = 0;
3409
3410 if (new_allocation) {
3411 if (new_reftable_offset) {
3412 qcow2_free_clusters(
3413 bs, new_reftable_offset,
3414 allocated_reftable_size * REFTABLE_ENTRY_SIZE,
3415 QCOW2_DISCARD_NEVER);
3416 }
3417
3418 new_reftable_offset = qcow2_alloc_clusters(bs, new_reftable_size *
3419 REFTABLE_ENTRY_SIZE);
3420 if (new_reftable_offset < 0) {
3421 error_setg_errno(errp, -new_reftable_offset,
3422 "Failed to allocate the new reftable");
3423 ret = new_reftable_offset;
3424 goto done;
3425 }
3426 allocated_reftable_size = new_reftable_size;
3427 }
3428 } while (new_allocation);
3429
3430 /* Second, write the new refblocks */
3431 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
3432 &new_reftable_size, new_refblock,
3433 new_refblock_size, new_refcount_bits,
3434 &flush_refblock, &new_allocation, new_set_refcount,
3435 status_cb, cb_opaque, walk_index, walk_index + 1,
3436 errp);
3437 if (ret < 0) {
3438 goto done;
3439 }
3440 assert(!new_allocation);
3441
3442
3443 /* Write the new reftable */
3444 ret = qcow2_pre_write_overlap_check(bs, 0, new_reftable_offset,
3445 new_reftable_size * REFTABLE_ENTRY_SIZE,
3446 false);
3447 if (ret < 0) {
3448 error_setg_errno(errp, -ret, "Overlap check failed");
3449 goto done;
3450 }
3451
3452 for (i = 0; i < new_reftable_size; i++) {
3453 cpu_to_be64s(&new_reftable[i]);
3454 }
3455
3456 ret = bdrv_pwrite(bs->file, new_reftable_offset,
3457 new_reftable_size * REFTABLE_ENTRY_SIZE, new_reftable,
3458 0);
3459
3460 for (i = 0; i < new_reftable_size; i++) {
3461 be64_to_cpus(&new_reftable[i]);
3462 }
3463
3464 if (ret < 0) {
3465 error_setg_errno(errp, -ret, "Failed to write the new reftable");
3466 goto done;
3467 }
3468
3469
3470 /* Empty the refcount cache */
3471 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
3472 if (ret < 0) {
3473 error_setg_errno(errp, -ret, "Failed to flush the refblock cache");
3474 goto done;
3475 }
3476
3477 /* Update the image header to point to the new reftable; this only updates
3478 * the fields which are relevant to qcow2_update_header(); other fields
3479 * such as s->refcount_table or s->refcount_bits stay stale for now
3480 * (because we have to restore everything if qcow2_update_header() fails) */
3481 old_refcount_order = s->refcount_order;
3482 old_reftable_size = s->refcount_table_size;
3483 old_reftable_offset = s->refcount_table_offset;
3484
3485 s->refcount_order = refcount_order;
3486 s->refcount_table_size = new_reftable_size;
3487 s->refcount_table_offset = new_reftable_offset;
3488
3489 ret = qcow2_update_header(bs);
3490 if (ret < 0) {
3491 s->refcount_order = old_refcount_order;
3492 s->refcount_table_size = old_reftable_size;
3493 s->refcount_table_offset = old_reftable_offset;
3494 error_setg_errno(errp, -ret, "Failed to update the qcow2 header");
3495 goto done;
3496 }
3497
3498 /* Now update the rest of the in-memory information */
3499 old_reftable = s->refcount_table;
3500 s->refcount_table = new_reftable;
3501 update_max_refcount_table_index(s);
3502
3503 s->refcount_bits = 1 << refcount_order;
3504 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
3505 s->refcount_max += s->refcount_max - 1;
3506
3507 s->refcount_block_bits = s->cluster_bits - (refcount_order - 3);
3508 s->refcount_block_size = 1 << s->refcount_block_bits;
3509
3510 s->get_refcount = new_get_refcount;
3511 s->set_refcount = new_set_refcount;
3512
3513 /* For cleaning up all old refblocks and the old reftable below the "done"
3514 * label */
3515 new_reftable = old_reftable;
3516 new_reftable_size = old_reftable_size;
3517 new_reftable_offset = old_reftable_offset;
3518
3519 done:
3520 if (new_reftable) {
3521 /* On success, new_reftable actually points to the old reftable (and
3522 * new_reftable_size is the old reftable's size); but that is just
3523 * fine */
3524 for (i = 0; i < new_reftable_size; i++) {
3525 uint64_t offset = new_reftable[i] & REFT_OFFSET_MASK;
3526 if (offset) {
3527 qcow2_free_clusters(bs, offset, s->cluster_size,
3528 QCOW2_DISCARD_OTHER);
3529 }
3530 }
3531 g_free(new_reftable);
3532
3533 if (new_reftable_offset > 0) {
3534 qcow2_free_clusters(bs, new_reftable_offset,
3535 new_reftable_size * REFTABLE_ENTRY_SIZE,
3536 QCOW2_DISCARD_OTHER);
3537 }
3538 }
3539
3540 qemu_vfree(new_refblock);
3541 return ret;
3542 }
3543
3544 static int64_t get_refblock_offset(BlockDriverState *bs, uint64_t offset)
3545 {
3546 BDRVQcow2State *s = bs->opaque;
3547 uint32_t index = offset_to_reftable_index(s, offset);
3548 int64_t covering_refblock_offset = 0;
3549
3550 if (index < s->refcount_table_size) {
3551 covering_refblock_offset = s->refcount_table[index] & REFT_OFFSET_MASK;
3552 }
3553 if (!covering_refblock_offset) {
3554 qcow2_signal_corruption(bs, true, -1, -1, "Refblock at %#" PRIx64 " is "
3555 "not covered by the refcount structures",
3556 offset);
3557 return -EIO;
3558 }
3559
3560 return covering_refblock_offset;
3561 }
3562
3563 static int coroutine_fn
3564 qcow2_discard_refcount_block(BlockDriverState *bs, uint64_t discard_block_offs)
3565 {
3566 BDRVQcow2State *s = bs->opaque;
3567 int64_t refblock_offs;
3568 uint64_t cluster_index = discard_block_offs >> s->cluster_bits;
3569 uint32_t block_index = cluster_index & (s->refcount_block_size - 1);
3570 void *refblock;
3571 int ret;
3572
3573 refblock_offs = get_refblock_offset(bs, discard_block_offs);
3574 if (refblock_offs < 0) {
3575 return refblock_offs;
3576 }
3577
3578 assert(discard_block_offs != 0);
3579
3580 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs,
3581 &refblock);
3582 if (ret < 0) {
3583 return ret;
3584 }
3585
3586 if (s->get_refcount(refblock, block_index) != 1) {
3587 qcow2_signal_corruption(bs, true, -1, -1, "Invalid refcount:"
3588 " refblock offset %#" PRIx64
3589 ", reftable index %u"
3590 ", block offset %#" PRIx64
3591 ", refcount %#" PRIx64,
3592 refblock_offs,
3593 offset_to_reftable_index(s, discard_block_offs),
3594 discard_block_offs,
3595 s->get_refcount(refblock, block_index));
3596 qcow2_cache_put(s->refcount_block_cache, &refblock);
3597 return -EINVAL;
3598 }
3599 s->set_refcount(refblock, block_index, 0);
3600
3601 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refblock);
3602
3603 qcow2_cache_put(s->refcount_block_cache, &refblock);
3604
3605 if (cluster_index < s->free_cluster_index) {
3606 s->free_cluster_index = cluster_index;
3607 }
3608
3609 refblock = qcow2_cache_is_table_offset(s->refcount_block_cache,
3610 discard_block_offs);
3611 if (refblock) {
3612 /* discard refblock from the cache if refblock is cached */
3613 qcow2_cache_discard(s->refcount_block_cache, refblock);
3614 }
3615 update_refcount_discard(bs, discard_block_offs, s->cluster_size);
3616
3617 return 0;
3618 }
3619
3620 int coroutine_fn qcow2_shrink_reftable(BlockDriverState *bs)
3621 {
3622 BDRVQcow2State *s = bs->opaque;
3623 uint64_t *reftable_tmp =
3624 g_malloc(s->refcount_table_size * REFTABLE_ENTRY_SIZE);
3625 int i, ret;
3626
3627 for (i = 0; i < s->refcount_table_size; i++) {
3628 int64_t refblock_offs = s->refcount_table[i] & REFT_OFFSET_MASK;
3629 void *refblock;
3630 bool unused_block;
3631
3632 if (refblock_offs == 0) {
3633 reftable_tmp[i] = 0;
3634 continue;
3635 }
3636 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs,
3637 &refblock);
3638 if (ret < 0) {
3639 goto out;
3640 }
3641
3642 /* the refblock has own reference */
3643 if (i == offset_to_reftable_index(s, refblock_offs)) {
3644 uint64_t block_index = (refblock_offs >> s->cluster_bits) &
3645 (s->refcount_block_size - 1);
3646 uint64_t refcount = s->get_refcount(refblock, block_index);
3647
3648 s->set_refcount(refblock, block_index, 0);
3649
3650 unused_block = buffer_is_zero(refblock, s->cluster_size);
3651
3652 s->set_refcount(refblock, block_index, refcount);
3653 } else {
3654 unused_block = buffer_is_zero(refblock, s->cluster_size);
3655 }
3656 qcow2_cache_put(s->refcount_block_cache, &refblock);
3657
3658 reftable_tmp[i] = unused_block ? 0 : cpu_to_be64(s->refcount_table[i]);
3659 }
3660
3661 ret = bdrv_co_pwrite_sync(bs->file, s->refcount_table_offset,
3662 s->refcount_table_size * REFTABLE_ENTRY_SIZE,
3663 reftable_tmp, 0);
3664 /*
3665 * If the write in the reftable failed the image may contain a partially
3666 * overwritten reftable. In this case it would be better to clear the
3667 * reftable in memory to avoid possible image corruption.
3668 */
3669 for (i = 0; i < s->refcount_table_size; i++) {
3670 if (s->refcount_table[i] && !reftable_tmp[i]) {
3671 if (ret == 0) {
3672 ret = qcow2_discard_refcount_block(bs, s->refcount_table[i] &
3673 REFT_OFFSET_MASK);
3674 }
3675 s->refcount_table[i] = 0;
3676 }
3677 }
3678
3679 if (!s->cache_discards) {
3680 qcow2_process_discards(bs, ret);
3681 }
3682
3683 out:
3684 g_free(reftable_tmp);
3685 return ret;
3686 }
3687
3688 int64_t qcow2_get_last_cluster(BlockDriverState *bs, int64_t size)
3689 {
3690 BDRVQcow2State *s = bs->opaque;
3691 int64_t i;
3692
3693 for (i = size_to_clusters(s, size) - 1; i >= 0; i--) {
3694 uint64_t refcount;
3695 int ret = qcow2_get_refcount(bs, i, &refcount);
3696 if (ret < 0) {
3697 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
3698 i, strerror(-ret));
3699 return ret;
3700 }
3701 if (refcount > 0) {
3702 return i;
3703 }
3704 }
3705 qcow2_signal_corruption(bs, true, -1, -1,
3706 "There are no references in the refcount table.");
3707 return -EIO;
3708 }
3709
3710 int coroutine_fn qcow2_detect_metadata_preallocation(BlockDriverState *bs)
3711 {
3712 BDRVQcow2State *s = bs->opaque;
3713 int64_t i, end_cluster, cluster_count = 0, threshold;
3714 int64_t file_length, real_allocation, real_clusters;
3715
3716 qemu_co_mutex_assert_locked(&s->lock);
3717
3718 file_length = bdrv_getlength(bs->file->bs);
3719 if (file_length < 0) {
3720 return file_length;
3721 }
3722
3723 real_allocation = bdrv_co_get_allocated_file_size(bs->file->bs);
3724 if (real_allocation < 0) {
3725 return real_allocation;
3726 }
3727
3728 real_clusters = real_allocation / s->cluster_size;
3729 threshold = MAX(real_clusters * 10 / 9, real_clusters + 2);
3730
3731 end_cluster = size_to_clusters(s, file_length);
3732 for (i = 0; i < end_cluster && cluster_count < threshold; i++) {
3733 uint64_t refcount;
3734 int ret = qcow2_get_refcount(bs, i, &refcount);
3735 if (ret < 0) {
3736 return ret;
3737 }
3738 cluster_count += !!refcount;
3739 }
3740
3741 return cluster_count >= threshold;
3742 }