]> git.proxmox.com Git - mirror_qemu.git/blob - block/qcow2-refcount.c
Merge remote-tracking branch 'remotes/hreitz/tags/pull-block-2021-09-15' into staging
[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 "qapi/error.h"
27 #include "qcow2.h"
28 #include "qemu/range.h"
29 #include "qemu/bswap.h"
30 #include "qemu/cutils.h"
31 #include "trace.h"
32
33 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size,
34 uint64_t max);
35 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
36 int64_t offset, int64_t length, uint64_t addend,
37 bool decrease, enum qcow2_discard_type type);
38
39 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index);
40 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index);
41 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index);
42 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index);
43 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index);
44 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index);
45 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index);
46
47 static void set_refcount_ro0(void *refcount_array, uint64_t index,
48 uint64_t value);
49 static void set_refcount_ro1(void *refcount_array, uint64_t index,
50 uint64_t value);
51 static void set_refcount_ro2(void *refcount_array, uint64_t index,
52 uint64_t value);
53 static void set_refcount_ro3(void *refcount_array, uint64_t index,
54 uint64_t value);
55 static void set_refcount_ro4(void *refcount_array, uint64_t index,
56 uint64_t value);
57 static void set_refcount_ro5(void *refcount_array, uint64_t index,
58 uint64_t value);
59 static void set_refcount_ro6(void *refcount_array, uint64_t index,
60 uint64_t value);
61
62
63 static Qcow2GetRefcountFunc *const get_refcount_funcs[] = {
64 &get_refcount_ro0,
65 &get_refcount_ro1,
66 &get_refcount_ro2,
67 &get_refcount_ro3,
68 &get_refcount_ro4,
69 &get_refcount_ro5,
70 &get_refcount_ro6
71 };
72
73 static Qcow2SetRefcountFunc *const set_refcount_funcs[] = {
74 &set_refcount_ro0,
75 &set_refcount_ro1,
76 &set_refcount_ro2,
77 &set_refcount_ro3,
78 &set_refcount_ro4,
79 &set_refcount_ro5,
80 &set_refcount_ro6
81 };
82
83
84 /*********************************************************/
85 /* refcount handling */
86
87 static void update_max_refcount_table_index(BDRVQcow2State *s)
88 {
89 unsigned i = s->refcount_table_size - 1;
90 while (i > 0 && (s->refcount_table[i] & REFT_OFFSET_MASK) == 0) {
91 i--;
92 }
93 /* Set s->max_refcount_table_index to the index of the last used entry */
94 s->max_refcount_table_index = i;
95 }
96
97 int qcow2_refcount_init(BlockDriverState *bs)
98 {
99 BDRVQcow2State *s = bs->opaque;
100 unsigned int refcount_table_size2, i;
101 int ret;
102
103 assert(s->refcount_order >= 0 && s->refcount_order <= 6);
104
105 s->get_refcount = get_refcount_funcs[s->refcount_order];
106 s->set_refcount = set_refcount_funcs[s->refcount_order];
107
108 assert(s->refcount_table_size <= INT_MAX / REFTABLE_ENTRY_SIZE);
109 refcount_table_size2 = s->refcount_table_size * REFTABLE_ENTRY_SIZE;
110 s->refcount_table = g_try_malloc(refcount_table_size2);
111
112 if (s->refcount_table_size > 0) {
113 if (s->refcount_table == NULL) {
114 ret = -ENOMEM;
115 goto fail;
116 }
117 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
118 ret = bdrv_pread(bs->file, s->refcount_table_offset,
119 s->refcount_table, refcount_table_size2);
120 if (ret < 0) {
121 goto fail;
122 }
123 for(i = 0; i < s->refcount_table_size; i++)
124 be64_to_cpus(&s->refcount_table[i]);
125 update_max_refcount_table_index(s);
126 }
127 return 0;
128 fail:
129 return ret;
130 }
131
132 void qcow2_refcount_close(BlockDriverState *bs)
133 {
134 BDRVQcow2State *s = bs->opaque;
135 g_free(s->refcount_table);
136 }
137
138
139 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index)
140 {
141 return (((const uint8_t *)refcount_array)[index / 8] >> (index % 8)) & 0x1;
142 }
143
144 static void set_refcount_ro0(void *refcount_array, uint64_t index,
145 uint64_t value)
146 {
147 assert(!(value >> 1));
148 ((uint8_t *)refcount_array)[index / 8] &= ~(0x1 << (index % 8));
149 ((uint8_t *)refcount_array)[index / 8] |= value << (index % 8);
150 }
151
152 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index)
153 {
154 return (((const uint8_t *)refcount_array)[index / 4] >> (2 * (index % 4)))
155 & 0x3;
156 }
157
158 static void set_refcount_ro1(void *refcount_array, uint64_t index,
159 uint64_t value)
160 {
161 assert(!(value >> 2));
162 ((uint8_t *)refcount_array)[index / 4] &= ~(0x3 << (2 * (index % 4)));
163 ((uint8_t *)refcount_array)[index / 4] |= value << (2 * (index % 4));
164 }
165
166 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index)
167 {
168 return (((const uint8_t *)refcount_array)[index / 2] >> (4 * (index % 2)))
169 & 0xf;
170 }
171
172 static void set_refcount_ro2(void *refcount_array, uint64_t index,
173 uint64_t value)
174 {
175 assert(!(value >> 4));
176 ((uint8_t *)refcount_array)[index / 2] &= ~(0xf << (4 * (index % 2)));
177 ((uint8_t *)refcount_array)[index / 2] |= value << (4 * (index % 2));
178 }
179
180 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index)
181 {
182 return ((const uint8_t *)refcount_array)[index];
183 }
184
185 static void set_refcount_ro3(void *refcount_array, uint64_t index,
186 uint64_t value)
187 {
188 assert(!(value >> 8));
189 ((uint8_t *)refcount_array)[index] = value;
190 }
191
192 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index)
193 {
194 return be16_to_cpu(((const uint16_t *)refcount_array)[index]);
195 }
196
197 static void set_refcount_ro4(void *refcount_array, uint64_t index,
198 uint64_t value)
199 {
200 assert(!(value >> 16));
201 ((uint16_t *)refcount_array)[index] = cpu_to_be16(value);
202 }
203
204 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index)
205 {
206 return be32_to_cpu(((const uint32_t *)refcount_array)[index]);
207 }
208
209 static void set_refcount_ro5(void *refcount_array, uint64_t index,
210 uint64_t value)
211 {
212 assert(!(value >> 32));
213 ((uint32_t *)refcount_array)[index] = cpu_to_be32(value);
214 }
215
216 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index)
217 {
218 return be64_to_cpu(((const uint64_t *)refcount_array)[index]);
219 }
220
221 static void set_refcount_ro6(void *refcount_array, uint64_t index,
222 uint64_t value)
223 {
224 ((uint64_t *)refcount_array)[index] = cpu_to_be64(value);
225 }
226
227
228 static int load_refcount_block(BlockDriverState *bs,
229 int64_t refcount_block_offset,
230 void **refcount_block)
231 {
232 BDRVQcow2State *s = bs->opaque;
233
234 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
235 return qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
236 refcount_block);
237 }
238
239 /*
240 * Retrieves the refcount of the cluster given by its index and stores it in
241 * *refcount. Returns 0 on success and -errno on failure.
242 */
243 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index,
244 uint64_t *refcount)
245 {
246 BDRVQcow2State *s = bs->opaque;
247 uint64_t refcount_table_index, block_index;
248 int64_t refcount_block_offset;
249 int ret;
250 void *refcount_block;
251
252 refcount_table_index = cluster_index >> s->refcount_block_bits;
253 if (refcount_table_index >= s->refcount_table_size) {
254 *refcount = 0;
255 return 0;
256 }
257 refcount_block_offset =
258 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
259 if (!refcount_block_offset) {
260 *refcount = 0;
261 return 0;
262 }
263
264 if (offset_into_cluster(s, refcount_block_offset)) {
265 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" PRIx64
266 " unaligned (reftable index: %#" PRIx64 ")",
267 refcount_block_offset, refcount_table_index);
268 return -EIO;
269 }
270
271 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
272 &refcount_block);
273 if (ret < 0) {
274 return ret;
275 }
276
277 block_index = cluster_index & (s->refcount_block_size - 1);
278 *refcount = s->get_refcount(refcount_block, block_index);
279
280 qcow2_cache_put(s->refcount_block_cache, &refcount_block);
281
282 return 0;
283 }
284
285 /* Checks if two offsets are described by the same refcount block */
286 static int in_same_refcount_block(BDRVQcow2State *s, uint64_t offset_a,
287 uint64_t offset_b)
288 {
289 uint64_t block_a = offset_a >> (s->cluster_bits + s->refcount_block_bits);
290 uint64_t block_b = offset_b >> (s->cluster_bits + s->refcount_block_bits);
291
292 return (block_a == block_b);
293 }
294
295 /*
296 * Loads a refcount block. If it doesn't exist yet, it is allocated first
297 * (including growing the refcount table if needed).
298 *
299 * Returns 0 on success or -errno in error case
300 */
301 static int alloc_refcount_block(BlockDriverState *bs,
302 int64_t cluster_index, void **refcount_block)
303 {
304 BDRVQcow2State *s = bs->opaque;
305 unsigned int refcount_table_index;
306 int64_t ret;
307
308 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
309
310 /* Find the refcount block for the given cluster */
311 refcount_table_index = cluster_index >> s->refcount_block_bits;
312
313 if (refcount_table_index < s->refcount_table_size) {
314
315 uint64_t refcount_block_offset =
316 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
317
318 /* If it's already there, we're done */
319 if (refcount_block_offset) {
320 if (offset_into_cluster(s, refcount_block_offset)) {
321 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
322 PRIx64 " unaligned (reftable index: "
323 "%#x)", refcount_block_offset,
324 refcount_table_index);
325 return -EIO;
326 }
327
328 return load_refcount_block(bs, refcount_block_offset,
329 refcount_block);
330 }
331 }
332
333 /*
334 * If we came here, we need to allocate something. Something is at least
335 * a cluster for the new refcount block. It may also include a new refcount
336 * table if the old refcount table is too small.
337 *
338 * Note that allocating clusters here needs some special care:
339 *
340 * - We can't use the normal qcow2_alloc_clusters(), it would try to
341 * increase the refcount and very likely we would end up with an endless
342 * recursion. Instead we must place the refcount blocks in a way that
343 * they can describe them themselves.
344 *
345 * - We need to consider that at this point we are inside update_refcounts
346 * and potentially doing an initial refcount increase. This means that
347 * some clusters have already been allocated by the caller, but their
348 * refcount isn't accurate yet. If we allocate clusters for metadata, we
349 * need to return -EAGAIN to signal the caller that it needs to restart
350 * the search for free clusters.
351 *
352 * - alloc_clusters_noref and qcow2_free_clusters may load a different
353 * refcount block into the cache
354 */
355
356 *refcount_block = NULL;
357
358 /* We write to the refcount table, so we might depend on L2 tables */
359 ret = qcow2_cache_flush(bs, s->l2_table_cache);
360 if (ret < 0) {
361 return ret;
362 }
363
364 /* Allocate the refcount block itself and mark it as used */
365 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size, INT64_MAX);
366 if (new_block < 0) {
367 return new_block;
368 }
369
370 /* The offset must fit in the offset field of the refcount table entry */
371 assert((new_block & REFT_OFFSET_MASK) == new_block);
372
373 /* If we're allocating the block at offset 0 then something is wrong */
374 if (new_block == 0) {
375 qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid "
376 "allocation of refcount block at offset 0");
377 return -EIO;
378 }
379
380 #ifdef DEBUG_ALLOC2
381 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
382 " at %" PRIx64 "\n",
383 refcount_table_index, cluster_index << s->cluster_bits, new_block);
384 #endif
385
386 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
387 /* Zero the new refcount block before updating it */
388 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
389 refcount_block);
390 if (ret < 0) {
391 goto fail;
392 }
393
394 memset(*refcount_block, 0, s->cluster_size);
395
396 /* The block describes itself, need to update the cache */
397 int block_index = (new_block >> s->cluster_bits) &
398 (s->refcount_block_size - 1);
399 s->set_refcount(*refcount_block, block_index, 1);
400 } else {
401 /* Described somewhere else. This can recurse at most twice before we
402 * arrive at a block that describes itself. */
403 ret = update_refcount(bs, new_block, s->cluster_size, 1, false,
404 QCOW2_DISCARD_NEVER);
405 if (ret < 0) {
406 goto fail;
407 }
408
409 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
410 if (ret < 0) {
411 goto fail;
412 }
413
414 /* Initialize the new refcount block only after updating its refcount,
415 * update_refcount uses the refcount cache itself */
416 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
417 refcount_block);
418 if (ret < 0) {
419 goto fail;
420 }
421
422 memset(*refcount_block, 0, s->cluster_size);
423 }
424
425 /* Now the new refcount block needs to be written to disk */
426 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
427 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block);
428 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
429 if (ret < 0) {
430 goto fail;
431 }
432
433 /* If the refcount table is big enough, just hook the block up there */
434 if (refcount_table_index < s->refcount_table_size) {
435 uint64_t data64 = cpu_to_be64(new_block);
436 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
437 ret = bdrv_pwrite_sync(bs->file, s->refcount_table_offset +
438 refcount_table_index * REFTABLE_ENTRY_SIZE,
439 &data64, sizeof(data64));
440 if (ret < 0) {
441 goto fail;
442 }
443
444 s->refcount_table[refcount_table_index] = new_block;
445 /* If there's a hole in s->refcount_table then it can happen
446 * that refcount_table_index < s->max_refcount_table_index */
447 s->max_refcount_table_index =
448 MAX(s->max_refcount_table_index, refcount_table_index);
449
450 /* The new refcount block may be where the caller intended to put its
451 * data, so let it restart the search. */
452 return -EAGAIN;
453 }
454
455 qcow2_cache_put(s->refcount_block_cache, refcount_block);
456
457 /*
458 * If we come here, we need to grow the refcount table. Again, a new
459 * refcount table needs some space and we can't simply allocate to avoid
460 * endless recursion.
461 *
462 * Therefore let's grab new refcount blocks at the end of the image, which
463 * will describe themselves and the new refcount table. This way we can
464 * reference them only in the new table and do the switch to the new
465 * refcount table at once without producing an inconsistent state in
466 * between.
467 */
468 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
469
470 /* Calculate the number of refcount blocks needed so far; this will be the
471 * basis for calculating the index of the first cluster used for the
472 * self-describing refcount structures which we are about to create.
473 *
474 * Because we reached this point, there cannot be any refcount entries for
475 * cluster_index or higher indices yet. However, because new_block has been
476 * allocated to describe that cluster (and it will assume this role later
477 * on), we cannot use that index; also, new_block may actually have a higher
478 * cluster index than cluster_index, so it needs to be taken into account
479 * here (and 1 needs to be added to its value because that cluster is used).
480 */
481 uint64_t blocks_used = DIV_ROUND_UP(MAX(cluster_index + 1,
482 (new_block >> s->cluster_bits) + 1),
483 s->refcount_block_size);
484
485 /* Create the new refcount table and blocks */
486 uint64_t meta_offset = (blocks_used * s->refcount_block_size) *
487 s->cluster_size;
488
489 ret = qcow2_refcount_area(bs, meta_offset, 0, false,
490 refcount_table_index, new_block);
491 if (ret < 0) {
492 return ret;
493 }
494
495 ret = load_refcount_block(bs, new_block, refcount_block);
496 if (ret < 0) {
497 return ret;
498 }
499
500 /* If we were trying to do the initial refcount update for some cluster
501 * allocation, we might have used the same clusters to store newly
502 * allocated metadata. Make the caller search some new space. */
503 return -EAGAIN;
504
505 fail:
506 if (*refcount_block != NULL) {
507 qcow2_cache_put(s->refcount_block_cache, refcount_block);
508 }
509 return ret;
510 }
511
512 /*
513 * Starting at @start_offset, this function creates new self-covering refcount
514 * structures: A new refcount table and refcount blocks which cover all of
515 * themselves, and a number of @additional_clusters beyond their end.
516 * @start_offset must be at the end of the image file, that is, there must be
517 * only empty space beyond it.
518 * If @exact_size is false, the refcount table will have 50 % more entries than
519 * necessary so it will not need to grow again soon.
520 * If @new_refblock_offset is not zero, it contains the offset of a refcount
521 * block that should be entered into the new refcount table at index
522 * @new_refblock_index.
523 *
524 * Returns: The offset after the new refcount structures (i.e. where the
525 * @additional_clusters may be placed) on success, -errno on error.
526 */
527 int64_t qcow2_refcount_area(BlockDriverState *bs, uint64_t start_offset,
528 uint64_t additional_clusters, bool exact_size,
529 int new_refblock_index,
530 uint64_t new_refblock_offset)
531 {
532 BDRVQcow2State *s = bs->opaque;
533 uint64_t total_refblock_count_u64, additional_refblock_count;
534 int total_refblock_count, table_size, area_reftable_index, table_clusters;
535 int i;
536 uint64_t table_offset, block_offset, end_offset;
537 int ret;
538 uint64_t *new_table;
539
540 assert(!(start_offset % s->cluster_size));
541
542 qcow2_refcount_metadata_size(start_offset / s->cluster_size +
543 additional_clusters,
544 s->cluster_size, s->refcount_order,
545 !exact_size, &total_refblock_count_u64);
546 if (total_refblock_count_u64 > QCOW_MAX_REFTABLE_SIZE) {
547 return -EFBIG;
548 }
549 total_refblock_count = total_refblock_count_u64;
550
551 /* Index in the refcount table of the first refcount block to cover the area
552 * of refcount structures we are about to create; we know that
553 * @total_refblock_count can cover @start_offset, so this will definitely
554 * fit into an int. */
555 area_reftable_index = (start_offset / s->cluster_size) /
556 s->refcount_block_size;
557
558 if (exact_size) {
559 table_size = total_refblock_count;
560 } else {
561 table_size = total_refblock_count +
562 DIV_ROUND_UP(total_refblock_count, 2);
563 }
564 /* The qcow2 file can only store the reftable size in number of clusters */
565 table_size = ROUND_UP(table_size, s->cluster_size / REFTABLE_ENTRY_SIZE);
566 table_clusters = (table_size * REFTABLE_ENTRY_SIZE) / s->cluster_size;
567
568 if (table_size > QCOW_MAX_REFTABLE_SIZE) {
569 return -EFBIG;
570 }
571
572 new_table = g_try_new0(uint64_t, table_size);
573
574 assert(table_size > 0);
575 if (new_table == NULL) {
576 ret = -ENOMEM;
577 goto fail;
578 }
579
580 /* Fill the new refcount table */
581 if (table_size > s->max_refcount_table_index) {
582 /* We're actually growing the reftable */
583 memcpy(new_table, s->refcount_table,
584 (s->max_refcount_table_index + 1) * REFTABLE_ENTRY_SIZE);
585 } else {
586 /* Improbable case: We're shrinking the reftable. However, the caller
587 * has assured us that there is only empty space beyond @start_offset,
588 * so we can simply drop all of the refblocks that won't fit into the
589 * new reftable. */
590 memcpy(new_table, s->refcount_table, table_size * REFTABLE_ENTRY_SIZE);
591 }
592
593 if (new_refblock_offset) {
594 assert(new_refblock_index < total_refblock_count);
595 new_table[new_refblock_index] = new_refblock_offset;
596 }
597
598 /* Count how many new refblocks we have to create */
599 additional_refblock_count = 0;
600 for (i = area_reftable_index; i < total_refblock_count; i++) {
601 if (!new_table[i]) {
602 additional_refblock_count++;
603 }
604 }
605
606 table_offset = start_offset + additional_refblock_count * s->cluster_size;
607 end_offset = table_offset + table_clusters * s->cluster_size;
608
609 /* Fill the refcount blocks, and create new ones, if necessary */
610 block_offset = start_offset;
611 for (i = area_reftable_index; i < total_refblock_count; i++) {
612 void *refblock_data;
613 uint64_t first_offset_covered;
614
615 /* Reuse an existing refblock if possible, create a new one otherwise */
616 if (new_table[i]) {
617 ret = qcow2_cache_get(bs, s->refcount_block_cache, new_table[i],
618 &refblock_data);
619 if (ret < 0) {
620 goto fail;
621 }
622 } else {
623 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache,
624 block_offset, &refblock_data);
625 if (ret < 0) {
626 goto fail;
627 }
628 memset(refblock_data, 0, s->cluster_size);
629 qcow2_cache_entry_mark_dirty(s->refcount_block_cache,
630 refblock_data);
631
632 new_table[i] = block_offset;
633 block_offset += s->cluster_size;
634 }
635
636 /* First host offset covered by this refblock */
637 first_offset_covered = (uint64_t)i * s->refcount_block_size *
638 s->cluster_size;
639 if (first_offset_covered < end_offset) {
640 int j, end_index;
641
642 /* Set the refcount of all of the new refcount structures to 1 */
643
644 if (first_offset_covered < start_offset) {
645 assert(i == area_reftable_index);
646 j = (start_offset - first_offset_covered) / s->cluster_size;
647 assert(j < s->refcount_block_size);
648 } else {
649 j = 0;
650 }
651
652 end_index = MIN((end_offset - first_offset_covered) /
653 s->cluster_size,
654 s->refcount_block_size);
655
656 for (; j < end_index; j++) {
657 /* The caller guaranteed us this space would be empty */
658 assert(s->get_refcount(refblock_data, j) == 0);
659 s->set_refcount(refblock_data, j, 1);
660 }
661
662 qcow2_cache_entry_mark_dirty(s->refcount_block_cache,
663 refblock_data);
664 }
665
666 qcow2_cache_put(s->refcount_block_cache, &refblock_data);
667 }
668
669 assert(block_offset == table_offset);
670
671 /* Write refcount blocks to disk */
672 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
673 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
674 if (ret < 0) {
675 goto fail;
676 }
677
678 /* Write refcount table to disk */
679 for (i = 0; i < total_refblock_count; i++) {
680 cpu_to_be64s(&new_table[i]);
681 }
682
683 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
684 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
685 table_size * REFTABLE_ENTRY_SIZE);
686 if (ret < 0) {
687 goto fail;
688 }
689
690 for (i = 0; i < total_refblock_count; i++) {
691 be64_to_cpus(&new_table[i]);
692 }
693
694 /* Hook up the new refcount table in the qcow2 header */
695 struct QEMU_PACKED {
696 uint64_t d64;
697 uint32_t d32;
698 } data;
699 data.d64 = cpu_to_be64(table_offset);
700 data.d32 = cpu_to_be32(table_clusters);
701 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
702 ret = bdrv_pwrite_sync(bs->file,
703 offsetof(QCowHeader, refcount_table_offset),
704 &data, sizeof(data));
705 if (ret < 0) {
706 goto fail;
707 }
708
709 /* And switch it in memory */
710 uint64_t old_table_offset = s->refcount_table_offset;
711 uint64_t old_table_size = s->refcount_table_size;
712
713 g_free(s->refcount_table);
714 s->refcount_table = new_table;
715 s->refcount_table_size = table_size;
716 s->refcount_table_offset = table_offset;
717 update_max_refcount_table_index(s);
718
719 /* Free old table. */
720 qcow2_free_clusters(bs, old_table_offset,
721 old_table_size * REFTABLE_ENTRY_SIZE,
722 QCOW2_DISCARD_OTHER);
723
724 return end_offset;
725
726 fail:
727 g_free(new_table);
728 return ret;
729 }
730
731 void qcow2_process_discards(BlockDriverState *bs, int ret)
732 {
733 BDRVQcow2State *s = bs->opaque;
734 Qcow2DiscardRegion *d, *next;
735
736 QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) {
737 QTAILQ_REMOVE(&s->discards, d, next);
738
739 /* Discard is optional, ignore the return value */
740 if (ret >= 0) {
741 int r2 = bdrv_pdiscard(bs->file, d->offset, d->bytes);
742 if (r2 < 0) {
743 trace_qcow2_process_discards_failed_region(d->offset, d->bytes,
744 r2);
745 }
746 }
747
748 g_free(d);
749 }
750 }
751
752 static void update_refcount_discard(BlockDriverState *bs,
753 uint64_t offset, uint64_t length)
754 {
755 BDRVQcow2State *s = bs->opaque;
756 Qcow2DiscardRegion *d, *p, *next;
757
758 QTAILQ_FOREACH(d, &s->discards, next) {
759 uint64_t new_start = MIN(offset, d->offset);
760 uint64_t new_end = MAX(offset + length, d->offset + d->bytes);
761
762 if (new_end - new_start <= length + d->bytes) {
763 /* There can't be any overlap, areas ending up here have no
764 * references any more and therefore shouldn't get freed another
765 * time. */
766 assert(d->bytes + length == new_end - new_start);
767 d->offset = new_start;
768 d->bytes = new_end - new_start;
769 goto found;
770 }
771 }
772
773 d = g_malloc(sizeof(*d));
774 *d = (Qcow2DiscardRegion) {
775 .bs = bs,
776 .offset = offset,
777 .bytes = length,
778 };
779 QTAILQ_INSERT_TAIL(&s->discards, d, next);
780
781 found:
782 /* Merge discard requests if they are adjacent now */
783 QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) {
784 if (p == d
785 || p->offset > d->offset + d->bytes
786 || d->offset > p->offset + p->bytes)
787 {
788 continue;
789 }
790
791 /* Still no overlap possible */
792 assert(p->offset == d->offset + d->bytes
793 || d->offset == p->offset + p->bytes);
794
795 QTAILQ_REMOVE(&s->discards, p, next);
796 d->offset = MIN(d->offset, p->offset);
797 d->bytes += p->bytes;
798 g_free(p);
799 }
800 }
801
802 /* XXX: cache several refcount block clusters ? */
803 /* @addend is the absolute value of the addend; if @decrease is set, @addend
804 * will be subtracted from the current refcount, otherwise it will be added */
805 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
806 int64_t offset,
807 int64_t length,
808 uint64_t addend,
809 bool decrease,
810 enum qcow2_discard_type type)
811 {
812 BDRVQcow2State *s = bs->opaque;
813 int64_t start, last, cluster_offset;
814 void *refcount_block = NULL;
815 int64_t old_table_index = -1;
816 int ret;
817
818 #ifdef DEBUG_ALLOC2
819 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64
820 " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "",
821 addend);
822 #endif
823 if (length < 0) {
824 return -EINVAL;
825 } else if (length == 0) {
826 return 0;
827 }
828
829 if (decrease) {
830 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
831 s->l2_table_cache);
832 }
833
834 start = start_of_cluster(s, offset);
835 last = start_of_cluster(s, offset + length - 1);
836 for(cluster_offset = start; cluster_offset <= last;
837 cluster_offset += s->cluster_size)
838 {
839 int block_index;
840 uint64_t refcount;
841 int64_t cluster_index = cluster_offset >> s->cluster_bits;
842 int64_t table_index = cluster_index >> s->refcount_block_bits;
843
844 /* Load the refcount block and allocate it if needed */
845 if (table_index != old_table_index) {
846 if (refcount_block) {
847 qcow2_cache_put(s->refcount_block_cache, &refcount_block);
848 }
849 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
850 /* If the caller needs to restart the search for free clusters,
851 * try the same ones first to see if they're still free. */
852 if (ret == -EAGAIN) {
853 if (s->free_cluster_index > (start >> s->cluster_bits)) {
854 s->free_cluster_index = (start >> s->cluster_bits);
855 }
856 }
857 if (ret < 0) {
858 goto fail;
859 }
860 }
861 old_table_index = table_index;
862
863 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block);
864
865 /* we can update the count and save it */
866 block_index = cluster_index & (s->refcount_block_size - 1);
867
868 refcount = s->get_refcount(refcount_block, block_index);
869 if (decrease ? (refcount - addend > refcount)
870 : (refcount + addend < refcount ||
871 refcount + addend > s->refcount_max))
872 {
873 ret = -EINVAL;
874 goto fail;
875 }
876 if (decrease) {
877 refcount -= addend;
878 } else {
879 refcount += addend;
880 }
881 if (refcount == 0 && cluster_index < s->free_cluster_index) {
882 s->free_cluster_index = cluster_index;
883 }
884 s->set_refcount(refcount_block, block_index, refcount);
885
886 if (refcount == 0) {
887 void *table;
888
889 table = qcow2_cache_is_table_offset(s->refcount_block_cache,
890 offset);
891 if (table != NULL) {
892 qcow2_cache_put(s->refcount_block_cache, &refcount_block);
893 old_table_index = -1;
894 qcow2_cache_discard(s->refcount_block_cache, table);
895 }
896
897 table = qcow2_cache_is_table_offset(s->l2_table_cache, offset);
898 if (table != NULL) {
899 qcow2_cache_discard(s->l2_table_cache, table);
900 }
901
902 if (s->discard_passthrough[type]) {
903 update_refcount_discard(bs, cluster_offset, s->cluster_size);
904 }
905 }
906 }
907
908 ret = 0;
909 fail:
910 if (!s->cache_discards) {
911 qcow2_process_discards(bs, ret);
912 }
913
914 /* Write last changed block to disk */
915 if (refcount_block) {
916 qcow2_cache_put(s->refcount_block_cache, &refcount_block);
917 }
918
919 /*
920 * Try do undo any updates if an error is returned (This may succeed in
921 * some cases like ENOSPC for allocating a new refcount block)
922 */
923 if (ret < 0) {
924 int dummy;
925 dummy = update_refcount(bs, offset, cluster_offset - offset, addend,
926 !decrease, QCOW2_DISCARD_NEVER);
927 (void)dummy;
928 }
929
930 return ret;
931 }
932
933 /*
934 * Increases or decreases the refcount of a given cluster.
935 *
936 * @addend is the absolute value of the addend; if @decrease is set, @addend
937 * will be subtracted from the current refcount, otherwise it will be added.
938 *
939 * On success 0 is returned; on failure -errno is returned.
940 */
941 int qcow2_update_cluster_refcount(BlockDriverState *bs,
942 int64_t cluster_index,
943 uint64_t addend, bool decrease,
944 enum qcow2_discard_type type)
945 {
946 BDRVQcow2State *s = bs->opaque;
947 int ret;
948
949 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend,
950 decrease, type);
951 if (ret < 0) {
952 return ret;
953 }
954
955 return 0;
956 }
957
958
959
960 /*********************************************************/
961 /* cluster allocation functions */
962
963
964
965 /* return < 0 if error */
966 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size,
967 uint64_t max)
968 {
969 BDRVQcow2State *s = bs->opaque;
970 uint64_t i, nb_clusters, refcount;
971 int ret;
972
973 /* We can't allocate clusters if they may still be queued for discard. */
974 if (s->cache_discards) {
975 qcow2_process_discards(bs, 0);
976 }
977
978 nb_clusters = size_to_clusters(s, size);
979 retry:
980 for(i = 0; i < nb_clusters; i++) {
981 uint64_t next_cluster_index = s->free_cluster_index++;
982 ret = qcow2_get_refcount(bs, next_cluster_index, &refcount);
983
984 if (ret < 0) {
985 return ret;
986 } else if (refcount != 0) {
987 goto retry;
988 }
989 }
990
991 /* Make sure that all offsets in the "allocated" range are representable
992 * in the requested max */
993 if (s->free_cluster_index > 0 &&
994 s->free_cluster_index - 1 > (max >> s->cluster_bits))
995 {
996 return -EFBIG;
997 }
998
999 #ifdef DEBUG_ALLOC2
1000 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
1001 size,
1002 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
1003 #endif
1004 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
1005 }
1006
1007 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size)
1008 {
1009 int64_t offset;
1010 int ret;
1011
1012 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
1013 do {
1014 offset = alloc_clusters_noref(bs, size, QCOW_MAX_CLUSTER_OFFSET);
1015 if (offset < 0) {
1016 return offset;
1017 }
1018
1019 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
1020 } while (ret == -EAGAIN);
1021
1022 if (ret < 0) {
1023 return ret;
1024 }
1025
1026 return offset;
1027 }
1028
1029 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
1030 int64_t nb_clusters)
1031 {
1032 BDRVQcow2State *s = bs->opaque;
1033 uint64_t cluster_index, refcount;
1034 uint64_t i;
1035 int ret;
1036
1037 assert(nb_clusters >= 0);
1038 if (nb_clusters == 0) {
1039 return 0;
1040 }
1041
1042 do {
1043 /* Check how many clusters there are free */
1044 cluster_index = offset >> s->cluster_bits;
1045 for(i = 0; i < nb_clusters; i++) {
1046 ret = qcow2_get_refcount(bs, cluster_index++, &refcount);
1047 if (ret < 0) {
1048 return ret;
1049 } else if (refcount != 0) {
1050 break;
1051 }
1052 }
1053
1054 /* And then allocate them */
1055 ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false,
1056 QCOW2_DISCARD_NEVER);
1057 } while (ret == -EAGAIN);
1058
1059 if (ret < 0) {
1060 return ret;
1061 }
1062
1063 return i;
1064 }
1065
1066 /* only used to allocate compressed sectors. We try to allocate
1067 contiguous sectors. size must be <= cluster_size */
1068 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
1069 {
1070 BDRVQcow2State *s = bs->opaque;
1071 int64_t offset;
1072 size_t free_in_cluster;
1073 int ret;
1074
1075 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
1076 assert(size > 0 && size <= s->cluster_size);
1077 assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset));
1078
1079 offset = s->free_byte_offset;
1080
1081 if (offset) {
1082 uint64_t refcount;
1083 ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount);
1084 if (ret < 0) {
1085 return ret;
1086 }
1087
1088 if (refcount == s->refcount_max) {
1089 offset = 0;
1090 }
1091 }
1092
1093 free_in_cluster = s->cluster_size - offset_into_cluster(s, offset);
1094 do {
1095 if (!offset || free_in_cluster < size) {
1096 int64_t new_cluster;
1097
1098 new_cluster = alloc_clusters_noref(bs, s->cluster_size,
1099 MIN(s->cluster_offset_mask,
1100 QCOW_MAX_CLUSTER_OFFSET));
1101 if (new_cluster < 0) {
1102 return new_cluster;
1103 }
1104
1105 if (new_cluster == 0) {
1106 qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid "
1107 "allocation of compressed cluster "
1108 "at offset 0");
1109 return -EIO;
1110 }
1111
1112 if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) {
1113 offset = new_cluster;
1114 free_in_cluster = s->cluster_size;
1115 } else {
1116 free_in_cluster += s->cluster_size;
1117 }
1118 }
1119
1120 assert(offset);
1121 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
1122 if (ret < 0) {
1123 offset = 0;
1124 }
1125 } while (ret == -EAGAIN);
1126 if (ret < 0) {
1127 return ret;
1128 }
1129
1130 /* The cluster refcount was incremented; refcount blocks must be flushed
1131 * before the caller's L2 table updates. */
1132 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
1133
1134 s->free_byte_offset = offset + size;
1135 if (!offset_into_cluster(s, s->free_byte_offset)) {
1136 s->free_byte_offset = 0;
1137 }
1138
1139 return offset;
1140 }
1141
1142 void qcow2_free_clusters(BlockDriverState *bs,
1143 int64_t offset, int64_t size,
1144 enum qcow2_discard_type type)
1145 {
1146 int ret;
1147
1148 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
1149 ret = update_refcount(bs, offset, size, 1, true, type);
1150 if (ret < 0) {
1151 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
1152 /* TODO Remember the clusters to free them later and avoid leaking */
1153 }
1154 }
1155
1156 /*
1157 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
1158 * normal cluster, compressed cluster, etc.)
1159 */
1160 void qcow2_free_any_cluster(BlockDriverState *bs, uint64_t l2_entry,
1161 enum qcow2_discard_type type)
1162 {
1163 BDRVQcow2State *s = bs->opaque;
1164 QCow2ClusterType ctype = qcow2_get_cluster_type(bs, l2_entry);
1165
1166 if (has_data_file(bs)) {
1167 if (s->discard_passthrough[type] &&
1168 (ctype == QCOW2_CLUSTER_NORMAL ||
1169 ctype == QCOW2_CLUSTER_ZERO_ALLOC))
1170 {
1171 bdrv_pdiscard(s->data_file, l2_entry & L2E_OFFSET_MASK,
1172 s->cluster_size);
1173 }
1174 return;
1175 }
1176
1177 switch (ctype) {
1178 case QCOW2_CLUSTER_COMPRESSED:
1179 {
1180 uint64_t coffset;
1181 int csize;
1182
1183 qcow2_parse_compressed_l2_entry(bs, l2_entry, &coffset, &csize);
1184 qcow2_free_clusters(bs, coffset, csize, type);
1185 }
1186 break;
1187 case QCOW2_CLUSTER_NORMAL:
1188 case QCOW2_CLUSTER_ZERO_ALLOC:
1189 if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) {
1190 qcow2_signal_corruption(bs, false, -1, -1,
1191 "Cannot free unaligned cluster %#llx",
1192 l2_entry & L2E_OFFSET_MASK);
1193 } else {
1194 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
1195 s->cluster_size, type);
1196 }
1197 break;
1198 case QCOW2_CLUSTER_ZERO_PLAIN:
1199 case QCOW2_CLUSTER_UNALLOCATED:
1200 break;
1201 default:
1202 abort();
1203 }
1204 }
1205
1206 int coroutine_fn qcow2_write_caches(BlockDriverState *bs)
1207 {
1208 BDRVQcow2State *s = bs->opaque;
1209 int ret;
1210
1211 ret = qcow2_cache_write(bs, s->l2_table_cache);
1212 if (ret < 0) {
1213 return ret;
1214 }
1215
1216 if (qcow2_need_accurate_refcounts(s)) {
1217 ret = qcow2_cache_write(bs, s->refcount_block_cache);
1218 if (ret < 0) {
1219 return ret;
1220 }
1221 }
1222
1223 return 0;
1224 }
1225
1226 int coroutine_fn qcow2_flush_caches(BlockDriverState *bs)
1227 {
1228 int ret = qcow2_write_caches(bs);
1229 if (ret < 0) {
1230 return ret;
1231 }
1232
1233 return bdrv_flush(bs->file->bs);
1234 }
1235
1236 /*********************************************************/
1237 /* snapshots and image creation */
1238
1239
1240
1241 /* update the refcounts of snapshots and the copied flag */
1242 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
1243 int64_t l1_table_offset, int l1_size, int addend)
1244 {
1245 BDRVQcow2State *s = bs->opaque;
1246 uint64_t *l1_table, *l2_slice, l2_offset, entry, l1_size2, refcount;
1247 bool l1_allocated = false;
1248 int64_t old_entry, old_l2_offset;
1249 unsigned slice, slice_size2, n_slices;
1250 int i, j, l1_modified = 0;
1251 int ret;
1252
1253 assert(addend >= -1 && addend <= 1);
1254
1255 l2_slice = NULL;
1256 l1_table = NULL;
1257 l1_size2 = l1_size * L1E_SIZE;
1258 slice_size2 = s->l2_slice_size * l2_entry_size(s);
1259 n_slices = s->cluster_size / slice_size2;
1260
1261 s->cache_discards = true;
1262
1263 /* WARNING: qcow2_snapshot_goto relies on this function not using the
1264 * l1_table_offset when it is the current s->l1_table_offset! Be careful
1265 * when changing this! */
1266 if (l1_table_offset != s->l1_table_offset) {
1267 l1_table = g_try_malloc0(l1_size2);
1268 if (l1_size2 && l1_table == NULL) {
1269 ret = -ENOMEM;
1270 goto fail;
1271 }
1272 l1_allocated = true;
1273
1274 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
1275 if (ret < 0) {
1276 goto fail;
1277 }
1278
1279 for (i = 0; i < l1_size; i++) {
1280 be64_to_cpus(&l1_table[i]);
1281 }
1282 } else {
1283 assert(l1_size == s->l1_size);
1284 l1_table = s->l1_table;
1285 l1_allocated = false;
1286 }
1287
1288 for (i = 0; i < l1_size; i++) {
1289 l2_offset = l1_table[i];
1290 if (l2_offset) {
1291 old_l2_offset = l2_offset;
1292 l2_offset &= L1E_OFFSET_MASK;
1293
1294 if (offset_into_cluster(s, l2_offset)) {
1295 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1296 PRIx64 " unaligned (L1 index: %#x)",
1297 l2_offset, i);
1298 ret = -EIO;
1299 goto fail;
1300 }
1301
1302 for (slice = 0; slice < n_slices; slice++) {
1303 ret = qcow2_cache_get(bs, s->l2_table_cache,
1304 l2_offset + slice * slice_size2,
1305 (void **) &l2_slice);
1306 if (ret < 0) {
1307 goto fail;
1308 }
1309
1310 for (j = 0; j < s->l2_slice_size; j++) {
1311 uint64_t cluster_index;
1312 uint64_t offset;
1313
1314 entry = get_l2_entry(s, l2_slice, j);
1315 old_entry = entry;
1316 entry &= ~QCOW_OFLAG_COPIED;
1317 offset = entry & L2E_OFFSET_MASK;
1318
1319 switch (qcow2_get_cluster_type(bs, entry)) {
1320 case QCOW2_CLUSTER_COMPRESSED:
1321 if (addend != 0) {
1322 uint64_t coffset;
1323 int csize;
1324
1325 qcow2_parse_compressed_l2_entry(bs, entry,
1326 &coffset, &csize);
1327 ret = update_refcount(
1328 bs, coffset, csize,
1329 abs(addend), addend < 0,
1330 QCOW2_DISCARD_SNAPSHOT);
1331 if (ret < 0) {
1332 goto fail;
1333 }
1334 }
1335 /* compressed clusters are never modified */
1336 refcount = 2;
1337 break;
1338
1339 case QCOW2_CLUSTER_NORMAL:
1340 case QCOW2_CLUSTER_ZERO_ALLOC:
1341 if (offset_into_cluster(s, offset)) {
1342 /* Here l2_index means table (not slice) index */
1343 int l2_index = slice * s->l2_slice_size + j;
1344 qcow2_signal_corruption(
1345 bs, true, -1, -1, "Cluster "
1346 "allocation offset %#" PRIx64
1347 " unaligned (L2 offset: %#"
1348 PRIx64 ", L2 index: %#x)",
1349 offset, l2_offset, l2_index);
1350 ret = -EIO;
1351 goto fail;
1352 }
1353
1354 cluster_index = offset >> s->cluster_bits;
1355 assert(cluster_index);
1356 if (addend != 0) {
1357 ret = qcow2_update_cluster_refcount(
1358 bs, cluster_index, abs(addend), addend < 0,
1359 QCOW2_DISCARD_SNAPSHOT);
1360 if (ret < 0) {
1361 goto fail;
1362 }
1363 }
1364
1365 ret = qcow2_get_refcount(bs, cluster_index, &refcount);
1366 if (ret < 0) {
1367 goto fail;
1368 }
1369 break;
1370
1371 case QCOW2_CLUSTER_ZERO_PLAIN:
1372 case QCOW2_CLUSTER_UNALLOCATED:
1373 refcount = 0;
1374 break;
1375
1376 default:
1377 abort();
1378 }
1379
1380 if (refcount == 1) {
1381 entry |= QCOW_OFLAG_COPIED;
1382 }
1383 if (entry != old_entry) {
1384 if (addend > 0) {
1385 qcow2_cache_set_dependency(bs, s->l2_table_cache,
1386 s->refcount_block_cache);
1387 }
1388 set_l2_entry(s, l2_slice, j, entry);
1389 qcow2_cache_entry_mark_dirty(s->l2_table_cache,
1390 l2_slice);
1391 }
1392 }
1393
1394 qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
1395 }
1396
1397 if (addend != 0) {
1398 ret = qcow2_update_cluster_refcount(bs, l2_offset >>
1399 s->cluster_bits,
1400 abs(addend), addend < 0,
1401 QCOW2_DISCARD_SNAPSHOT);
1402 if (ret < 0) {
1403 goto fail;
1404 }
1405 }
1406 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1407 &refcount);
1408 if (ret < 0) {
1409 goto fail;
1410 } else if (refcount == 1) {
1411 l2_offset |= QCOW_OFLAG_COPIED;
1412 }
1413 if (l2_offset != old_l2_offset) {
1414 l1_table[i] = l2_offset;
1415 l1_modified = 1;
1416 }
1417 }
1418 }
1419
1420 ret = bdrv_flush(bs);
1421 fail:
1422 if (l2_slice) {
1423 qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
1424 }
1425
1426 s->cache_discards = false;
1427 qcow2_process_discards(bs, ret);
1428
1429 /* Update L1 only if it isn't deleted anyway (addend = -1) */
1430 if (ret == 0 && addend >= 0 && l1_modified) {
1431 for (i = 0; i < l1_size; i++) {
1432 cpu_to_be64s(&l1_table[i]);
1433 }
1434
1435 ret = bdrv_pwrite_sync(bs->file, l1_table_offset,
1436 l1_table, l1_size2);
1437
1438 for (i = 0; i < l1_size; i++) {
1439 be64_to_cpus(&l1_table[i]);
1440 }
1441 }
1442 if (l1_allocated)
1443 g_free(l1_table);
1444 return ret;
1445 }
1446
1447
1448
1449
1450 /*********************************************************/
1451 /* refcount checking functions */
1452
1453
1454 static uint64_t refcount_array_byte_size(BDRVQcow2State *s, uint64_t entries)
1455 {
1456 /* This assertion holds because there is no way we can address more than
1457 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because
1458 * offsets have to be representable in bytes); due to every cluster
1459 * corresponding to one refcount entry, we are well below that limit */
1460 assert(entries < (UINT64_C(1) << (64 - 9)));
1461
1462 /* Thanks to the assertion this will not overflow, because
1463 * s->refcount_order < 7.
1464 * (note: x << s->refcount_order == x * s->refcount_bits) */
1465 return DIV_ROUND_UP(entries << s->refcount_order, 8);
1466 }
1467
1468 /**
1469 * Reallocates *array so that it can hold new_size entries. *size must contain
1470 * the current number of entries in *array. If the reallocation fails, *array
1471 * and *size will not be modified and -errno will be returned. If the
1472 * reallocation is successful, *array will be set to the new buffer, *size
1473 * will be set to new_size and 0 will be returned. The size of the reallocated
1474 * refcount array buffer will be aligned to a cluster boundary, and the newly
1475 * allocated area will be zeroed.
1476 */
1477 static int realloc_refcount_array(BDRVQcow2State *s, void **array,
1478 int64_t *size, int64_t new_size)
1479 {
1480 int64_t old_byte_size, new_byte_size;
1481 void *new_ptr;
1482
1483 /* Round to clusters so the array can be directly written to disk */
1484 old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size))
1485 * s->cluster_size;
1486 new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size))
1487 * s->cluster_size;
1488
1489 if (new_byte_size == old_byte_size) {
1490 *size = new_size;
1491 return 0;
1492 }
1493
1494 assert(new_byte_size > 0);
1495
1496 if (new_byte_size > SIZE_MAX) {
1497 return -ENOMEM;
1498 }
1499
1500 new_ptr = g_try_realloc(*array, new_byte_size);
1501 if (!new_ptr) {
1502 return -ENOMEM;
1503 }
1504
1505 if (new_byte_size > old_byte_size) {
1506 memset((char *)new_ptr + old_byte_size, 0,
1507 new_byte_size - old_byte_size);
1508 }
1509
1510 *array = new_ptr;
1511 *size = new_size;
1512
1513 return 0;
1514 }
1515
1516 /*
1517 * Increases the refcount for a range of clusters in a given refcount table.
1518 * This is used to construct a temporary refcount table out of L1 and L2 tables
1519 * which can be compared to the refcount table saved in the image.
1520 *
1521 * Modifies the number of errors in res.
1522 */
1523 int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res,
1524 void **refcount_table,
1525 int64_t *refcount_table_size,
1526 int64_t offset, int64_t size)
1527 {
1528 BDRVQcow2State *s = bs->opaque;
1529 uint64_t start, last, cluster_offset, k, refcount;
1530 int64_t file_len;
1531 int ret;
1532
1533 if (size <= 0) {
1534 return 0;
1535 }
1536
1537 file_len = bdrv_getlength(bs->file->bs);
1538 if (file_len < 0) {
1539 return file_len;
1540 }
1541
1542 /*
1543 * Last cluster of qcow2 image may be semi-allocated, so it may be OK to
1544 * reference some space after file end but it should be less than one
1545 * cluster.
1546 */
1547 if (offset + size - file_len >= s->cluster_size) {
1548 fprintf(stderr, "ERROR: counting reference for region exceeding the "
1549 "end of the file by one cluster or more: offset 0x%" PRIx64
1550 " size 0x%" PRIx64 "\n", offset, size);
1551 res->corruptions++;
1552 return 0;
1553 }
1554
1555 start = start_of_cluster(s, offset);
1556 last = start_of_cluster(s, offset + size - 1);
1557 for(cluster_offset = start; cluster_offset <= last;
1558 cluster_offset += s->cluster_size) {
1559 k = cluster_offset >> s->cluster_bits;
1560 if (k >= *refcount_table_size) {
1561 ret = realloc_refcount_array(s, refcount_table,
1562 refcount_table_size, k + 1);
1563 if (ret < 0) {
1564 res->check_errors++;
1565 return ret;
1566 }
1567 }
1568
1569 refcount = s->get_refcount(*refcount_table, k);
1570 if (refcount == s->refcount_max) {
1571 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
1572 "\n", cluster_offset);
1573 fprintf(stderr, "Use qemu-img amend to increase the refcount entry "
1574 "width or qemu-img convert to create a clean copy if the "
1575 "image cannot be opened for writing\n");
1576 res->corruptions++;
1577 continue;
1578 }
1579 s->set_refcount(*refcount_table, k, refcount + 1);
1580 }
1581
1582 return 0;
1583 }
1584
1585 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
1586 enum {
1587 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
1588 };
1589
1590 /*
1591 * Fix L2 entry by making it QCOW2_CLUSTER_ZERO_PLAIN (or making all its present
1592 * subclusters QCOW2_SUBCLUSTER_ZERO_PLAIN).
1593 *
1594 * This function decrements res->corruptions on success, so the caller is
1595 * responsible to increment res->corruptions prior to the call.
1596 *
1597 * On failure in-memory @l2_table may be modified.
1598 */
1599 static int fix_l2_entry_by_zero(BlockDriverState *bs, BdrvCheckResult *res,
1600 uint64_t l2_offset,
1601 uint64_t *l2_table, int l2_index, bool active,
1602 bool *metadata_overlap)
1603 {
1604 BDRVQcow2State *s = bs->opaque;
1605 int ret;
1606 int idx = l2_index * (l2_entry_size(s) / sizeof(uint64_t));
1607 uint64_t l2e_offset = l2_offset + (uint64_t)l2_index * l2_entry_size(s);
1608 int ign = active ? QCOW2_OL_ACTIVE_L2 : QCOW2_OL_INACTIVE_L2;
1609
1610 if (has_subclusters(s)) {
1611 uint64_t l2_bitmap = get_l2_bitmap(s, l2_table, l2_index);
1612
1613 /* Allocated subclusters become zero */
1614 l2_bitmap |= l2_bitmap << 32;
1615 l2_bitmap &= QCOW_L2_BITMAP_ALL_ZEROES;
1616
1617 set_l2_bitmap(s, l2_table, l2_index, l2_bitmap);
1618 set_l2_entry(s, l2_table, l2_index, 0);
1619 } else {
1620 set_l2_entry(s, l2_table, l2_index, QCOW_OFLAG_ZERO);
1621 }
1622
1623 ret = qcow2_pre_write_overlap_check(bs, ign, l2e_offset, l2_entry_size(s),
1624 false);
1625 if (metadata_overlap) {
1626 *metadata_overlap = ret < 0;
1627 }
1628 if (ret < 0) {
1629 fprintf(stderr, "ERROR: Overlap check failed\n");
1630 goto fail;
1631 }
1632
1633 ret = bdrv_pwrite_sync(bs->file, l2e_offset, &l2_table[idx],
1634 l2_entry_size(s));
1635 if (ret < 0) {
1636 fprintf(stderr, "ERROR: Failed to overwrite L2 "
1637 "table entry: %s\n", strerror(-ret));
1638 goto fail;
1639 }
1640
1641 res->corruptions--;
1642 res->corruptions_fixed++;
1643 return 0;
1644
1645 fail:
1646 res->check_errors++;
1647 return ret;
1648 }
1649
1650 /*
1651 * Increases the refcount in the given refcount table for the all clusters
1652 * referenced in the L2 table. While doing so, performs some checks on L2
1653 * entries.
1654 *
1655 * Returns the number of errors found by the checks or -errno if an internal
1656 * error occurred.
1657 */
1658 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
1659 void **refcount_table,
1660 int64_t *refcount_table_size, int64_t l2_offset,
1661 int flags, BdrvCheckMode fix, bool active)
1662 {
1663 BDRVQcow2State *s = bs->opaque;
1664 uint64_t l2_entry, l2_bitmap;
1665 uint64_t next_contiguous_offset = 0;
1666 int i, ret;
1667 size_t l2_size_bytes = s->l2_size * l2_entry_size(s);
1668 g_autofree uint64_t *l2_table = g_malloc(l2_size_bytes);
1669 bool metadata_overlap;
1670
1671 /* Read L2 table from disk */
1672 ret = bdrv_pread(bs->file, l2_offset, l2_table, l2_size_bytes);
1673 if (ret < 0) {
1674 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1675 res->check_errors++;
1676 return ret;
1677 }
1678
1679 /* Do the actual checks */
1680 for (i = 0; i < s->l2_size; i++) {
1681 uint64_t coffset;
1682 int csize;
1683 QCow2ClusterType type;
1684
1685 l2_entry = get_l2_entry(s, l2_table, i);
1686 l2_bitmap = get_l2_bitmap(s, l2_table, i);
1687 type = qcow2_get_cluster_type(bs, l2_entry);
1688
1689 if (type != QCOW2_CLUSTER_COMPRESSED) {
1690 /* Check reserved bits of Standard Cluster Descriptor */
1691 if (l2_entry & L2E_STD_RESERVED_MASK) {
1692 fprintf(stderr, "ERROR found l2 entry with reserved bits set: "
1693 "%" PRIx64 "\n", l2_entry);
1694 res->corruptions++;
1695 }
1696 }
1697
1698 switch (type) {
1699 case QCOW2_CLUSTER_COMPRESSED:
1700 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1701 if (l2_entry & QCOW_OFLAG_COPIED) {
1702 fprintf(stderr, "ERROR: coffset=0x%" PRIx64 ": "
1703 "copied flag must never be set for compressed "
1704 "clusters\n", l2_entry & s->cluster_offset_mask);
1705 l2_entry &= ~QCOW_OFLAG_COPIED;
1706 res->corruptions++;
1707 }
1708
1709 if (has_data_file(bs)) {
1710 fprintf(stderr, "ERROR compressed cluster %d with data file, "
1711 "entry=0x%" PRIx64 "\n", i, l2_entry);
1712 res->corruptions++;
1713 break;
1714 }
1715
1716 if (l2_bitmap) {
1717 fprintf(stderr, "ERROR compressed cluster %d with non-zero "
1718 "subcluster allocation bitmap, entry=0x%" PRIx64 "\n",
1719 i, l2_entry);
1720 res->corruptions++;
1721 break;
1722 }
1723
1724 /* Mark cluster as used */
1725 qcow2_parse_compressed_l2_entry(bs, l2_entry, &coffset, &csize);
1726 ret = qcow2_inc_refcounts_imrt(
1727 bs, res, refcount_table, refcount_table_size, coffset, csize);
1728 if (ret < 0) {
1729 return ret;
1730 }
1731
1732 if (flags & CHECK_FRAG_INFO) {
1733 res->bfi.allocated_clusters++;
1734 res->bfi.compressed_clusters++;
1735
1736 /*
1737 * Compressed clusters are fragmented by nature. Since they
1738 * take up sub-sector space but we only have sector granularity
1739 * I/O we need to re-read the same sectors even for adjacent
1740 * compressed clusters.
1741 */
1742 res->bfi.fragmented_clusters++;
1743 }
1744 break;
1745
1746 case QCOW2_CLUSTER_ZERO_ALLOC:
1747 case QCOW2_CLUSTER_NORMAL:
1748 {
1749 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
1750
1751 if ((l2_bitmap >> 32) & l2_bitmap) {
1752 res->corruptions++;
1753 fprintf(stderr, "ERROR offset=%" PRIx64 ": Allocated "
1754 "cluster has corrupted subcluster allocation bitmap\n",
1755 offset);
1756 }
1757
1758 /* Correct offsets are cluster aligned */
1759 if (offset_into_cluster(s, offset)) {
1760 bool contains_data;
1761 res->corruptions++;
1762
1763 if (has_subclusters(s)) {
1764 contains_data = (l2_bitmap & QCOW_L2_BITMAP_ALL_ALLOC);
1765 } else {
1766 contains_data = !(l2_entry & QCOW_OFLAG_ZERO);
1767 }
1768
1769 if (!contains_data) {
1770 fprintf(stderr, "%s offset=%" PRIx64 ": Preallocated "
1771 "cluster is not properly aligned; L2 entry "
1772 "corrupted.\n",
1773 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR",
1774 offset);
1775 if (fix & BDRV_FIX_ERRORS) {
1776 ret = fix_l2_entry_by_zero(bs, res, l2_offset,
1777 l2_table, i, active,
1778 &metadata_overlap);
1779 if (metadata_overlap) {
1780 /*
1781 * Something is seriously wrong, so abort checking
1782 * this L2 table.
1783 */
1784 return ret;
1785 }
1786
1787 if (ret == 0) {
1788 /*
1789 * Skip marking the cluster as used
1790 * (it is unused now).
1791 */
1792 continue;
1793 }
1794
1795 /*
1796 * Failed to fix.
1797 * Do not abort, continue checking the rest of this
1798 * L2 table's entries.
1799 */
1800 }
1801 } else {
1802 fprintf(stderr, "ERROR offset=%" PRIx64 ": Data cluster is "
1803 "not properly aligned; L2 entry corrupted.\n", offset);
1804 }
1805 }
1806
1807 if (flags & CHECK_FRAG_INFO) {
1808 res->bfi.allocated_clusters++;
1809 if (next_contiguous_offset &&
1810 offset != next_contiguous_offset) {
1811 res->bfi.fragmented_clusters++;
1812 }
1813 next_contiguous_offset = offset + s->cluster_size;
1814 }
1815
1816 /* Mark cluster as used */
1817 if (!has_data_file(bs)) {
1818 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table,
1819 refcount_table_size,
1820 offset, s->cluster_size);
1821 if (ret < 0) {
1822 return ret;
1823 }
1824 }
1825 break;
1826 }
1827
1828 case QCOW2_CLUSTER_ZERO_PLAIN:
1829 /* Impossible when image has subclusters */
1830 assert(!l2_bitmap);
1831 break;
1832
1833 case QCOW2_CLUSTER_UNALLOCATED:
1834 if (l2_bitmap & QCOW_L2_BITMAP_ALL_ALLOC) {
1835 res->corruptions++;
1836 fprintf(stderr, "ERROR: Unallocated "
1837 "cluster has non-zero subcluster allocation map\n");
1838 }
1839 break;
1840
1841 default:
1842 abort();
1843 }
1844 }
1845
1846 return 0;
1847 }
1848
1849 /*
1850 * Increases the refcount for the L1 table, its L2 tables and all referenced
1851 * clusters in the given refcount table. While doing so, performs some checks
1852 * on L1 and L2 entries.
1853 *
1854 * Returns the number of errors found by the checks or -errno if an internal
1855 * error occurred.
1856 */
1857 static int check_refcounts_l1(BlockDriverState *bs,
1858 BdrvCheckResult *res,
1859 void **refcount_table,
1860 int64_t *refcount_table_size,
1861 int64_t l1_table_offset, int l1_size,
1862 int flags, BdrvCheckMode fix, bool active)
1863 {
1864 BDRVQcow2State *s = bs->opaque;
1865 size_t l1_size_bytes = l1_size * L1E_SIZE;
1866 g_autofree uint64_t *l1_table = NULL;
1867 uint64_t l2_offset;
1868 int i, ret;
1869
1870 if (!l1_size) {
1871 return 0;
1872 }
1873
1874 /* Mark L1 table as used */
1875 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, refcount_table_size,
1876 l1_table_offset, l1_size_bytes);
1877 if (ret < 0) {
1878 return ret;
1879 }
1880
1881 l1_table = g_try_malloc(l1_size_bytes);
1882 if (l1_table == NULL) {
1883 res->check_errors++;
1884 return -ENOMEM;
1885 }
1886
1887 /* Read L1 table entries from disk */
1888 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size_bytes);
1889 if (ret < 0) {
1890 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1891 res->check_errors++;
1892 return ret;
1893 }
1894
1895 for (i = 0; i < l1_size; i++) {
1896 be64_to_cpus(&l1_table[i]);
1897 }
1898
1899 /* Do the actual checks */
1900 for (i = 0; i < l1_size; i++) {
1901 if (!l1_table[i]) {
1902 continue;
1903 }
1904
1905 if (l1_table[i] & L1E_RESERVED_MASK) {
1906 fprintf(stderr, "ERROR found L1 entry with reserved bits set: "
1907 "%" PRIx64 "\n", l1_table[i]);
1908 res->corruptions++;
1909 }
1910
1911 l2_offset = l1_table[i] & L1E_OFFSET_MASK;
1912
1913 /* Mark L2 table as used */
1914 ret = qcow2_inc_refcounts_imrt(bs, res,
1915 refcount_table, refcount_table_size,
1916 l2_offset, s->cluster_size);
1917 if (ret < 0) {
1918 return ret;
1919 }
1920
1921 /* L2 tables are cluster aligned */
1922 if (offset_into_cluster(s, l2_offset)) {
1923 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1924 "cluster aligned; L1 entry corrupted\n", l2_offset);
1925 res->corruptions++;
1926 }
1927
1928 /* Process and check L2 entries */
1929 ret = check_refcounts_l2(bs, res, refcount_table,
1930 refcount_table_size, l2_offset, flags,
1931 fix, active);
1932 if (ret < 0) {
1933 return ret;
1934 }
1935 }
1936
1937 return 0;
1938 }
1939
1940 /*
1941 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1942 *
1943 * This function does not print an error message nor does it increment
1944 * check_errors if qcow2_get_refcount fails (this is because such an error will
1945 * have been already detected and sufficiently signaled by the calling function
1946 * (qcow2_check_refcounts) by the time this function is called).
1947 */
1948 static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
1949 BdrvCheckMode fix)
1950 {
1951 BDRVQcow2State *s = bs->opaque;
1952 uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size);
1953 int ret;
1954 uint64_t refcount;
1955 int i, j;
1956 bool repair;
1957
1958 if (fix & BDRV_FIX_ERRORS) {
1959 /* Always repair */
1960 repair = true;
1961 } else if (fix & BDRV_FIX_LEAKS) {
1962 /* Repair only if that seems safe: This function is always
1963 * called after the refcounts have been fixed, so the refcount
1964 * is accurate if that repair was successful */
1965 repair = !res->check_errors && !res->corruptions && !res->leaks;
1966 } else {
1967 repair = false;
1968 }
1969
1970 for (i = 0; i < s->l1_size; i++) {
1971 uint64_t l1_entry = s->l1_table[i];
1972 uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK;
1973 int l2_dirty = 0;
1974
1975 if (!l2_offset) {
1976 continue;
1977 }
1978
1979 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1980 &refcount);
1981 if (ret < 0) {
1982 /* don't print message nor increment check_errors */
1983 continue;
1984 }
1985 if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) {
1986 res->corruptions++;
1987 fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1988 "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
1989 repair ? "Repairing" : "ERROR", i, l1_entry, refcount);
1990 if (repair) {
1991 s->l1_table[i] = refcount == 1
1992 ? l1_entry | QCOW_OFLAG_COPIED
1993 : l1_entry & ~QCOW_OFLAG_COPIED;
1994 ret = qcow2_write_l1_entry(bs, i);
1995 if (ret < 0) {
1996 res->check_errors++;
1997 goto fail;
1998 }
1999 res->corruptions--;
2000 res->corruptions_fixed++;
2001 }
2002 }
2003
2004 ret = bdrv_pread(bs->file, l2_offset, l2_table,
2005 s->l2_size * l2_entry_size(s));
2006 if (ret < 0) {
2007 fprintf(stderr, "ERROR: Could not read L2 table: %s\n",
2008 strerror(-ret));
2009 res->check_errors++;
2010 goto fail;
2011 }
2012
2013 for (j = 0; j < s->l2_size; j++) {
2014 uint64_t l2_entry = get_l2_entry(s, l2_table, j);
2015 uint64_t data_offset = l2_entry & L2E_OFFSET_MASK;
2016 QCow2ClusterType cluster_type = qcow2_get_cluster_type(bs, l2_entry);
2017
2018 if (cluster_type == QCOW2_CLUSTER_NORMAL ||
2019 cluster_type == QCOW2_CLUSTER_ZERO_ALLOC) {
2020 if (has_data_file(bs)) {
2021 refcount = 1;
2022 } else {
2023 ret = qcow2_get_refcount(bs,
2024 data_offset >> s->cluster_bits,
2025 &refcount);
2026 if (ret < 0) {
2027 /* don't print message nor increment check_errors */
2028 continue;
2029 }
2030 }
2031 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
2032 res->corruptions++;
2033 fprintf(stderr, "%s OFLAG_COPIED data cluster: "
2034 "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
2035 repair ? "Repairing" : "ERROR", l2_entry, refcount);
2036 if (repair) {
2037 set_l2_entry(s, l2_table, j,
2038 refcount == 1 ?
2039 l2_entry | QCOW_OFLAG_COPIED :
2040 l2_entry & ~QCOW_OFLAG_COPIED);
2041 l2_dirty++;
2042 }
2043 }
2044 }
2045 }
2046
2047 if (l2_dirty > 0) {
2048 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
2049 l2_offset, s->cluster_size,
2050 false);
2051 if (ret < 0) {
2052 fprintf(stderr, "ERROR: Could not write L2 table; metadata "
2053 "overlap check failed: %s\n", strerror(-ret));
2054 res->check_errors++;
2055 goto fail;
2056 }
2057
2058 ret = bdrv_pwrite(bs->file, l2_offset, l2_table,
2059 s->cluster_size);
2060 if (ret < 0) {
2061 fprintf(stderr, "ERROR: Could not write L2 table: %s\n",
2062 strerror(-ret));
2063 res->check_errors++;
2064 goto fail;
2065 }
2066 res->corruptions -= l2_dirty;
2067 res->corruptions_fixed += l2_dirty;
2068 }
2069 }
2070
2071 ret = 0;
2072
2073 fail:
2074 qemu_vfree(l2_table);
2075 return ret;
2076 }
2077
2078 /*
2079 * Checks consistency of refblocks and accounts for each refblock in
2080 * *refcount_table.
2081 */
2082 static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res,
2083 BdrvCheckMode fix, bool *rebuild,
2084 void **refcount_table, int64_t *nb_clusters)
2085 {
2086 BDRVQcow2State *s = bs->opaque;
2087 int64_t i, size;
2088 int ret;
2089
2090 for(i = 0; i < s->refcount_table_size; i++) {
2091 uint64_t offset, cluster;
2092 offset = s->refcount_table[i] & REFT_OFFSET_MASK;
2093 cluster = offset >> s->cluster_bits;
2094
2095 if (s->refcount_table[i] & REFT_RESERVED_MASK) {
2096 fprintf(stderr, "ERROR refcount table entry %" PRId64 " has "
2097 "reserved bits set\n", i);
2098 res->corruptions++;
2099 *rebuild = true;
2100 continue;
2101 }
2102
2103 /* Refcount blocks are cluster aligned */
2104 if (offset_into_cluster(s, offset)) {
2105 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
2106 "cluster aligned; refcount table entry corrupted\n", i);
2107 res->corruptions++;
2108 *rebuild = true;
2109 continue;
2110 }
2111
2112 if (cluster >= *nb_clusters) {
2113 res->corruptions++;
2114 fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n",
2115 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
2116
2117 if (fix & BDRV_FIX_ERRORS) {
2118 int64_t new_nb_clusters;
2119 Error *local_err = NULL;
2120
2121 if (offset > INT64_MAX - s->cluster_size) {
2122 ret = -EINVAL;
2123 goto resize_fail;
2124 }
2125
2126 ret = bdrv_truncate(bs->file, offset + s->cluster_size, false,
2127 PREALLOC_MODE_OFF, 0, &local_err);
2128 if (ret < 0) {
2129 error_report_err(local_err);
2130 goto resize_fail;
2131 }
2132 size = bdrv_getlength(bs->file->bs);
2133 if (size < 0) {
2134 ret = size;
2135 goto resize_fail;
2136 }
2137
2138 new_nb_clusters = size_to_clusters(s, size);
2139 assert(new_nb_clusters >= *nb_clusters);
2140
2141 ret = realloc_refcount_array(s, refcount_table,
2142 nb_clusters, new_nb_clusters);
2143 if (ret < 0) {
2144 res->check_errors++;
2145 return ret;
2146 }
2147
2148 if (cluster >= *nb_clusters) {
2149 ret = -EINVAL;
2150 goto resize_fail;
2151 }
2152
2153 res->corruptions--;
2154 res->corruptions_fixed++;
2155 ret = qcow2_inc_refcounts_imrt(bs, res,
2156 refcount_table, nb_clusters,
2157 offset, s->cluster_size);
2158 if (ret < 0) {
2159 return ret;
2160 }
2161 /* No need to check whether the refcount is now greater than 1:
2162 * This area was just allocated and zeroed, so it can only be
2163 * exactly 1 after qcow2_inc_refcounts_imrt() */
2164 continue;
2165
2166 resize_fail:
2167 *rebuild = true;
2168 fprintf(stderr, "ERROR could not resize image: %s\n",
2169 strerror(-ret));
2170 }
2171 continue;
2172 }
2173
2174 if (offset != 0) {
2175 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2176 offset, s->cluster_size);
2177 if (ret < 0) {
2178 return ret;
2179 }
2180 if (s->get_refcount(*refcount_table, cluster) != 1) {
2181 fprintf(stderr, "ERROR refcount block %" PRId64
2182 " refcount=%" PRIu64 "\n", i,
2183 s->get_refcount(*refcount_table, cluster));
2184 res->corruptions++;
2185 *rebuild = true;
2186 }
2187 }
2188 }
2189
2190 return 0;
2191 }
2192
2193 /*
2194 * Calculates an in-memory refcount table.
2195 */
2196 static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2197 BdrvCheckMode fix, bool *rebuild,
2198 void **refcount_table, int64_t *nb_clusters)
2199 {
2200 BDRVQcow2State *s = bs->opaque;
2201 int64_t i;
2202 QCowSnapshot *sn;
2203 int ret;
2204
2205 if (!*refcount_table) {
2206 int64_t old_size = 0;
2207 ret = realloc_refcount_array(s, refcount_table,
2208 &old_size, *nb_clusters);
2209 if (ret < 0) {
2210 res->check_errors++;
2211 return ret;
2212 }
2213 }
2214
2215 /* header */
2216 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2217 0, s->cluster_size);
2218 if (ret < 0) {
2219 return ret;
2220 }
2221
2222 /* current L1 table */
2223 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
2224 s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO,
2225 fix, true);
2226 if (ret < 0) {
2227 return ret;
2228 }
2229
2230 /* snapshots */
2231 if (has_data_file(bs) && s->nb_snapshots) {
2232 fprintf(stderr, "ERROR %d snapshots in image with data file\n",
2233 s->nb_snapshots);
2234 res->corruptions++;
2235 }
2236
2237 for (i = 0; i < s->nb_snapshots; i++) {
2238 sn = s->snapshots + i;
2239 if (offset_into_cluster(s, sn->l1_table_offset)) {
2240 fprintf(stderr, "ERROR snapshot %s (%s) l1_offset=%#" PRIx64 ": "
2241 "L1 table is not cluster aligned; snapshot table entry "
2242 "corrupted\n", sn->id_str, sn->name, sn->l1_table_offset);
2243 res->corruptions++;
2244 continue;
2245 }
2246 if (sn->l1_size > QCOW_MAX_L1_SIZE / L1E_SIZE) {
2247 fprintf(stderr, "ERROR snapshot %s (%s) l1_size=%#" PRIx32 ": "
2248 "L1 table is too large; snapshot table entry corrupted\n",
2249 sn->id_str, sn->name, sn->l1_size);
2250 res->corruptions++;
2251 continue;
2252 }
2253 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
2254 sn->l1_table_offset, sn->l1_size, 0, fix,
2255 false);
2256 if (ret < 0) {
2257 return ret;
2258 }
2259 }
2260 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2261 s->snapshots_offset, s->snapshots_size);
2262 if (ret < 0) {
2263 return ret;
2264 }
2265
2266 /* refcount data */
2267 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2268 s->refcount_table_offset,
2269 s->refcount_table_size *
2270 REFTABLE_ENTRY_SIZE);
2271 if (ret < 0) {
2272 return ret;
2273 }
2274
2275 /* encryption */
2276 if (s->crypto_header.length) {
2277 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
2278 s->crypto_header.offset,
2279 s->crypto_header.length);
2280 if (ret < 0) {
2281 return ret;
2282 }
2283 }
2284
2285 /* bitmaps */
2286 ret = qcow2_check_bitmaps_refcounts(bs, res, refcount_table, nb_clusters);
2287 if (ret < 0) {
2288 return ret;
2289 }
2290
2291 return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters);
2292 }
2293
2294 /*
2295 * Compares the actual reference count for each cluster in the image against the
2296 * refcount as reported by the refcount structures on-disk.
2297 */
2298 static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2299 BdrvCheckMode fix, bool *rebuild,
2300 int64_t *highest_cluster,
2301 void *refcount_table, int64_t nb_clusters)
2302 {
2303 BDRVQcow2State *s = bs->opaque;
2304 int64_t i;
2305 uint64_t refcount1, refcount2;
2306 int ret;
2307
2308 for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) {
2309 ret = qcow2_get_refcount(bs, i, &refcount1);
2310 if (ret < 0) {
2311 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
2312 i, strerror(-ret));
2313 res->check_errors++;
2314 continue;
2315 }
2316
2317 refcount2 = s->get_refcount(refcount_table, i);
2318
2319 if (refcount1 > 0 || refcount2 > 0) {
2320 *highest_cluster = i;
2321 }
2322
2323 if (refcount1 != refcount2) {
2324 /* Check if we're allowed to fix the mismatch */
2325 int *num_fixed = NULL;
2326 if (refcount1 == 0) {
2327 *rebuild = true;
2328 } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
2329 num_fixed = &res->leaks_fixed;
2330 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
2331 num_fixed = &res->corruptions_fixed;
2332 }
2333
2334 fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64
2335 " reference=%" PRIu64 "\n",
2336 num_fixed != NULL ? "Repairing" :
2337 refcount1 < refcount2 ? "ERROR" :
2338 "Leaked",
2339 i, refcount1, refcount2);
2340
2341 if (num_fixed) {
2342 ret = update_refcount(bs, i << s->cluster_bits, 1,
2343 refcount_diff(refcount1, refcount2),
2344 refcount1 > refcount2,
2345 QCOW2_DISCARD_ALWAYS);
2346 if (ret >= 0) {
2347 (*num_fixed)++;
2348 continue;
2349 }
2350 }
2351
2352 /* And if we couldn't, print an error */
2353 if (refcount1 < refcount2) {
2354 res->corruptions++;
2355 } else {
2356 res->leaks++;
2357 }
2358 }
2359 }
2360 }
2361
2362 /*
2363 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to
2364 * the on-disk refcount structures.
2365 *
2366 * On input, *first_free_cluster tells where to start looking, and need not
2367 * actually be a free cluster; the returned offset will not be before that
2368 * cluster. On output, *first_free_cluster points to the first gap found, even
2369 * if that gap was too small to be used as the returned offset.
2370 *
2371 * Note that *first_free_cluster is a cluster index whereas the return value is
2372 * an offset.
2373 */
2374 static int64_t alloc_clusters_imrt(BlockDriverState *bs,
2375 int cluster_count,
2376 void **refcount_table,
2377 int64_t *imrt_nb_clusters,
2378 int64_t *first_free_cluster)
2379 {
2380 BDRVQcow2State *s = bs->opaque;
2381 int64_t cluster = *first_free_cluster, i;
2382 bool first_gap = true;
2383 int contiguous_free_clusters;
2384 int ret;
2385
2386 /* Starting at *first_free_cluster, find a range of at least cluster_count
2387 * continuously free clusters */
2388 for (contiguous_free_clusters = 0;
2389 cluster < *imrt_nb_clusters &&
2390 contiguous_free_clusters < cluster_count;
2391 cluster++)
2392 {
2393 if (!s->get_refcount(*refcount_table, cluster)) {
2394 contiguous_free_clusters++;
2395 if (first_gap) {
2396 /* If this is the first free cluster found, update
2397 * *first_free_cluster accordingly */
2398 *first_free_cluster = cluster;
2399 first_gap = false;
2400 }
2401 } else if (contiguous_free_clusters) {
2402 contiguous_free_clusters = 0;
2403 }
2404 }
2405
2406 /* If contiguous_free_clusters is greater than zero, it contains the number
2407 * of continuously free clusters until the current cluster; the first free
2408 * cluster in the current "gap" is therefore
2409 * cluster - contiguous_free_clusters */
2410
2411 /* If no such range could be found, grow the in-memory refcount table
2412 * accordingly to append free clusters at the end of the image */
2413 if (contiguous_free_clusters < cluster_count) {
2414 /* contiguous_free_clusters clusters are already empty at the image end;
2415 * we need cluster_count clusters; therefore, we have to allocate
2416 * cluster_count - contiguous_free_clusters new clusters at the end of
2417 * the image (which is the current value of cluster; note that cluster
2418 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond
2419 * the image end) */
2420 ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters,
2421 cluster + cluster_count
2422 - contiguous_free_clusters);
2423 if (ret < 0) {
2424 return ret;
2425 }
2426 }
2427
2428 /* Go back to the first free cluster */
2429 cluster -= contiguous_free_clusters;
2430 for (i = 0; i < cluster_count; i++) {
2431 s->set_refcount(*refcount_table, cluster + i, 1);
2432 }
2433
2434 return cluster << s->cluster_bits;
2435 }
2436
2437 /*
2438 * Creates a new refcount structure based solely on the in-memory information
2439 * given through *refcount_table. All necessary allocations will be reflected
2440 * in that array.
2441 *
2442 * On success, the old refcount structure is leaked (it will be covered by the
2443 * new refcount structure).
2444 */
2445 static int rebuild_refcount_structure(BlockDriverState *bs,
2446 BdrvCheckResult *res,
2447 void **refcount_table,
2448 int64_t *nb_clusters)
2449 {
2450 BDRVQcow2State *s = bs->opaque;
2451 int64_t first_free_cluster = 0, reftable_offset = -1, cluster = 0;
2452 int64_t refblock_offset, refblock_start, refblock_index;
2453 uint32_t reftable_size = 0;
2454 uint64_t *on_disk_reftable = NULL;
2455 void *on_disk_refblock;
2456 int ret = 0;
2457 struct {
2458 uint64_t reftable_offset;
2459 uint32_t reftable_clusters;
2460 } QEMU_PACKED reftable_offset_and_clusters;
2461
2462 qcow2_cache_empty(bs, s->refcount_block_cache);
2463
2464 write_refblocks:
2465 for (; cluster < *nb_clusters; cluster++) {
2466 if (!s->get_refcount(*refcount_table, cluster)) {
2467 continue;
2468 }
2469
2470 refblock_index = cluster >> s->refcount_block_bits;
2471 refblock_start = refblock_index << s->refcount_block_bits;
2472
2473 /* Don't allocate a cluster in a refblock already written to disk */
2474 if (first_free_cluster < refblock_start) {
2475 first_free_cluster = refblock_start;
2476 }
2477 refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table,
2478 nb_clusters, &first_free_cluster);
2479 if (refblock_offset < 0) {
2480 fprintf(stderr, "ERROR allocating refblock: %s\n",
2481 strerror(-refblock_offset));
2482 res->check_errors++;
2483 ret = refblock_offset;
2484 goto fail;
2485 }
2486
2487 if (reftable_size <= refblock_index) {
2488 uint32_t old_reftable_size = reftable_size;
2489 uint64_t *new_on_disk_reftable;
2490
2491 reftable_size = ROUND_UP((refblock_index + 1) * REFTABLE_ENTRY_SIZE,
2492 s->cluster_size) / REFTABLE_ENTRY_SIZE;
2493 new_on_disk_reftable = g_try_realloc(on_disk_reftable,
2494 reftable_size *
2495 REFTABLE_ENTRY_SIZE);
2496 if (!new_on_disk_reftable) {
2497 res->check_errors++;
2498 ret = -ENOMEM;
2499 goto fail;
2500 }
2501 on_disk_reftable = new_on_disk_reftable;
2502
2503 memset(on_disk_reftable + old_reftable_size, 0,
2504 (reftable_size - old_reftable_size) * REFTABLE_ENTRY_SIZE);
2505
2506 /* The offset we have for the reftable is now no longer valid;
2507 * this will leak that range, but we can easily fix that by running
2508 * a leak-fixing check after this rebuild operation */
2509 reftable_offset = -1;
2510 } else {
2511 assert(on_disk_reftable);
2512 }
2513 on_disk_reftable[refblock_index] = refblock_offset;
2514
2515 /* If this is apparently the last refblock (for now), try to squeeze the
2516 * reftable in */
2517 if (refblock_index == (*nb_clusters - 1) >> s->refcount_block_bits &&
2518 reftable_offset < 0)
2519 {
2520 uint64_t reftable_clusters = size_to_clusters(s, reftable_size *
2521 REFTABLE_ENTRY_SIZE);
2522 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2523 refcount_table, nb_clusters,
2524 &first_free_cluster);
2525 if (reftable_offset < 0) {
2526 fprintf(stderr, "ERROR allocating reftable: %s\n",
2527 strerror(-reftable_offset));
2528 res->check_errors++;
2529 ret = reftable_offset;
2530 goto fail;
2531 }
2532 }
2533
2534 ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset,
2535 s->cluster_size, false);
2536 if (ret < 0) {
2537 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2538 goto fail;
2539 }
2540
2541 /* The size of *refcount_table is always cluster-aligned, therefore the
2542 * write operation will not overflow */
2543 on_disk_refblock = (void *)((char *) *refcount_table +
2544 refblock_index * s->cluster_size);
2545
2546 ret = bdrv_pwrite(bs->file, refblock_offset, on_disk_refblock,
2547 s->cluster_size);
2548 if (ret < 0) {
2549 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2550 goto fail;
2551 }
2552
2553 /* Go to the end of this refblock */
2554 cluster = refblock_start + s->refcount_block_size - 1;
2555 }
2556
2557 if (reftable_offset < 0) {
2558 uint64_t post_refblock_start, reftable_clusters;
2559
2560 post_refblock_start = ROUND_UP(*nb_clusters, s->refcount_block_size);
2561 reftable_clusters =
2562 size_to_clusters(s, reftable_size * REFTABLE_ENTRY_SIZE);
2563 /* Not pretty but simple */
2564 if (first_free_cluster < post_refblock_start) {
2565 first_free_cluster = post_refblock_start;
2566 }
2567 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2568 refcount_table, nb_clusters,
2569 &first_free_cluster);
2570 if (reftable_offset < 0) {
2571 fprintf(stderr, "ERROR allocating reftable: %s\n",
2572 strerror(-reftable_offset));
2573 res->check_errors++;
2574 ret = reftable_offset;
2575 goto fail;
2576 }
2577
2578 goto write_refblocks;
2579 }
2580
2581 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2582 cpu_to_be64s(&on_disk_reftable[refblock_index]);
2583 }
2584
2585 ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset,
2586 reftable_size * REFTABLE_ENTRY_SIZE,
2587 false);
2588 if (ret < 0) {
2589 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2590 goto fail;
2591 }
2592
2593 assert(reftable_size < INT_MAX / REFTABLE_ENTRY_SIZE);
2594 ret = bdrv_pwrite(bs->file, reftable_offset, on_disk_reftable,
2595 reftable_size * REFTABLE_ENTRY_SIZE);
2596 if (ret < 0) {
2597 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2598 goto fail;
2599 }
2600
2601 /* Enter new reftable into the image header */
2602 reftable_offset_and_clusters.reftable_offset = cpu_to_be64(reftable_offset);
2603 reftable_offset_and_clusters.reftable_clusters =
2604 cpu_to_be32(size_to_clusters(s, reftable_size * REFTABLE_ENTRY_SIZE));
2605 ret = bdrv_pwrite_sync(bs->file,
2606 offsetof(QCowHeader, refcount_table_offset),
2607 &reftable_offset_and_clusters,
2608 sizeof(reftable_offset_and_clusters));
2609 if (ret < 0) {
2610 fprintf(stderr, "ERROR setting reftable: %s\n", strerror(-ret));
2611 goto fail;
2612 }
2613
2614 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2615 be64_to_cpus(&on_disk_reftable[refblock_index]);
2616 }
2617 s->refcount_table = on_disk_reftable;
2618 s->refcount_table_offset = reftable_offset;
2619 s->refcount_table_size = reftable_size;
2620 update_max_refcount_table_index(s);
2621
2622 return 0;
2623
2624 fail:
2625 g_free(on_disk_reftable);
2626 return ret;
2627 }
2628
2629 /*
2630 * Checks an image for refcount consistency.
2631 *
2632 * Returns 0 if no errors are found, the number of errors in case the image is
2633 * detected as corrupted, and -errno when an internal error occurred.
2634 */
2635 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2636 BdrvCheckMode fix)
2637 {
2638 BDRVQcow2State *s = bs->opaque;
2639 BdrvCheckResult pre_compare_res;
2640 int64_t size, highest_cluster, nb_clusters;
2641 void *refcount_table = NULL;
2642 bool rebuild = false;
2643 int ret;
2644
2645 size = bdrv_getlength(bs->file->bs);
2646 if (size < 0) {
2647 res->check_errors++;
2648 return size;
2649 }
2650
2651 nb_clusters = size_to_clusters(s, size);
2652 if (nb_clusters > INT_MAX) {
2653 res->check_errors++;
2654 return -EFBIG;
2655 }
2656
2657 res->bfi.total_clusters =
2658 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE);
2659
2660 ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table,
2661 &nb_clusters);
2662 if (ret < 0) {
2663 goto fail;
2664 }
2665
2666 /* In case we don't need to rebuild the refcount structure (but want to fix
2667 * something), this function is immediately called again, in which case the
2668 * result should be ignored */
2669 pre_compare_res = *res;
2670 compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table,
2671 nb_clusters);
2672
2673 if (rebuild && (fix & BDRV_FIX_ERRORS)) {
2674 BdrvCheckResult old_res = *res;
2675 int fresh_leaks = 0;
2676
2677 fprintf(stderr, "Rebuilding refcount structure\n");
2678 ret = rebuild_refcount_structure(bs, res, &refcount_table,
2679 &nb_clusters);
2680 if (ret < 0) {
2681 goto fail;
2682 }
2683
2684 res->corruptions = 0;
2685 res->leaks = 0;
2686
2687 /* Because the old reftable has been exchanged for a new one the
2688 * references have to be recalculated */
2689 rebuild = false;
2690 memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters));
2691 ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table,
2692 &nb_clusters);
2693 if (ret < 0) {
2694 goto fail;
2695 }
2696
2697 if (fix & BDRV_FIX_LEAKS) {
2698 /* The old refcount structures are now leaked, fix it; the result
2699 * can be ignored, aside from leaks which were introduced by
2700 * rebuild_refcount_structure() that could not be fixed */
2701 BdrvCheckResult saved_res = *res;
2702 *res = (BdrvCheckResult){ 0 };
2703
2704 compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild,
2705 &highest_cluster, refcount_table, nb_clusters);
2706 if (rebuild) {
2707 fprintf(stderr, "ERROR rebuilt refcount structure is still "
2708 "broken\n");
2709 }
2710
2711 /* Any leaks accounted for here were introduced by
2712 * rebuild_refcount_structure() because that function has created a
2713 * new refcount structure from scratch */
2714 fresh_leaks = res->leaks;
2715 *res = saved_res;
2716 }
2717
2718 if (res->corruptions < old_res.corruptions) {
2719 res->corruptions_fixed += old_res.corruptions - res->corruptions;
2720 }
2721 if (res->leaks < old_res.leaks) {
2722 res->leaks_fixed += old_res.leaks - res->leaks;
2723 }
2724 res->leaks += fresh_leaks;
2725 } else if (fix) {
2726 if (rebuild) {
2727 fprintf(stderr, "ERROR need to rebuild refcount structures\n");
2728 res->check_errors++;
2729 ret = -EIO;
2730 goto fail;
2731 }
2732
2733 if (res->leaks || res->corruptions) {
2734 *res = pre_compare_res;
2735 compare_refcounts(bs, res, fix, &rebuild, &highest_cluster,
2736 refcount_table, nb_clusters);
2737 }
2738 }
2739
2740 /* check OFLAG_COPIED */
2741 ret = check_oflag_copied(bs, res, fix);
2742 if (ret < 0) {
2743 goto fail;
2744 }
2745
2746 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
2747 ret = 0;
2748
2749 fail:
2750 g_free(refcount_table);
2751
2752 return ret;
2753 }
2754
2755 #define overlaps_with(ofs, sz) \
2756 ranges_overlap(offset, size, ofs, sz)
2757
2758 /*
2759 * Checks if the given offset into the image file is actually free to use by
2760 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
2761 * i.e. a sanity check without relying on the refcount tables.
2762 *
2763 * The ign parameter specifies what checks not to perform (being a bitmask of
2764 * QCow2MetadataOverlap values), i.e., what sections to ignore.
2765 *
2766 * Returns:
2767 * - 0 if writing to this offset will not affect the mentioned metadata
2768 * - a positive QCow2MetadataOverlap value indicating one overlapping section
2769 * - a negative value (-errno) indicating an error while performing a check,
2770 * e.g. when bdrv_pread failed on QCOW2_OL_INACTIVE_L2
2771 */
2772 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset,
2773 int64_t size)
2774 {
2775 BDRVQcow2State *s = bs->opaque;
2776 int chk = s->overlap_check & ~ign;
2777 int i, j;
2778
2779 if (!size) {
2780 return 0;
2781 }
2782
2783 if (chk & QCOW2_OL_MAIN_HEADER) {
2784 if (offset < s->cluster_size) {
2785 return QCOW2_OL_MAIN_HEADER;
2786 }
2787 }
2788
2789 /* align range to test to cluster boundaries */
2790 size = ROUND_UP(offset_into_cluster(s, offset) + size, s->cluster_size);
2791 offset = start_of_cluster(s, offset);
2792
2793 if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) {
2794 if (overlaps_with(s->l1_table_offset, s->l1_size * L1E_SIZE)) {
2795 return QCOW2_OL_ACTIVE_L1;
2796 }
2797 }
2798
2799 if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) {
2800 if (overlaps_with(s->refcount_table_offset,
2801 s->refcount_table_size * REFTABLE_ENTRY_SIZE)) {
2802 return QCOW2_OL_REFCOUNT_TABLE;
2803 }
2804 }
2805
2806 if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) {
2807 if (overlaps_with(s->snapshots_offset, s->snapshots_size)) {
2808 return QCOW2_OL_SNAPSHOT_TABLE;
2809 }
2810 }
2811
2812 if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) {
2813 for (i = 0; i < s->nb_snapshots; i++) {
2814 if (s->snapshots[i].l1_size &&
2815 overlaps_with(s->snapshots[i].l1_table_offset,
2816 s->snapshots[i].l1_size * L1E_SIZE)) {
2817 return QCOW2_OL_INACTIVE_L1;
2818 }
2819 }
2820 }
2821
2822 if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) {
2823 for (i = 0; i < s->l1_size; i++) {
2824 if ((s->l1_table[i] & L1E_OFFSET_MASK) &&
2825 overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK,
2826 s->cluster_size)) {
2827 return QCOW2_OL_ACTIVE_L2;
2828 }
2829 }
2830 }
2831
2832 if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) {
2833 unsigned last_entry = s->max_refcount_table_index;
2834 assert(last_entry < s->refcount_table_size);
2835 assert(last_entry + 1 == s->refcount_table_size ||
2836 (s->refcount_table[last_entry + 1] & REFT_OFFSET_MASK) == 0);
2837 for (i = 0; i <= last_entry; i++) {
2838 if ((s->refcount_table[i] & REFT_OFFSET_MASK) &&
2839 overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK,
2840 s->cluster_size)) {
2841 return QCOW2_OL_REFCOUNT_BLOCK;
2842 }
2843 }
2844 }
2845
2846 if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) {
2847 for (i = 0; i < s->nb_snapshots; i++) {
2848 uint64_t l1_ofs = s->snapshots[i].l1_table_offset;
2849 uint32_t l1_sz = s->snapshots[i].l1_size;
2850 uint64_t l1_sz2 = l1_sz * L1E_SIZE;
2851 uint64_t *l1;
2852 int ret;
2853
2854 ret = qcow2_validate_table(bs, l1_ofs, l1_sz, L1E_SIZE,
2855 QCOW_MAX_L1_SIZE, "", NULL);
2856 if (ret < 0) {
2857 return ret;
2858 }
2859
2860 l1 = g_try_malloc(l1_sz2);
2861
2862 if (l1_sz2 && l1 == NULL) {
2863 return -ENOMEM;
2864 }
2865
2866 ret = bdrv_pread(bs->file, l1_ofs, l1, l1_sz2);
2867 if (ret < 0) {
2868 g_free(l1);
2869 return ret;
2870 }
2871
2872 for (j = 0; j < l1_sz; j++) {
2873 uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK;
2874 if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) {
2875 g_free(l1);
2876 return QCOW2_OL_INACTIVE_L2;
2877 }
2878 }
2879
2880 g_free(l1);
2881 }
2882 }
2883
2884 if ((chk & QCOW2_OL_BITMAP_DIRECTORY) &&
2885 (s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS))
2886 {
2887 if (overlaps_with(s->bitmap_directory_offset,
2888 s->bitmap_directory_size))
2889 {
2890 return QCOW2_OL_BITMAP_DIRECTORY;
2891 }
2892 }
2893
2894 return 0;
2895 }
2896
2897 static const char *metadata_ol_names[] = {
2898 [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header",
2899 [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table",
2900 [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table",
2901 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table",
2902 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block",
2903 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table",
2904 [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table",
2905 [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table",
2906 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = "bitmap directory",
2907 };
2908 QEMU_BUILD_BUG_ON(QCOW2_OL_MAX_BITNR != ARRAY_SIZE(metadata_ol_names));
2909
2910 /*
2911 * First performs a check for metadata overlaps (through
2912 * qcow2_check_metadata_overlap); if that fails with a negative value (error
2913 * while performing a check), that value is returned. If an impending overlap
2914 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
2915 * and -EIO returned.
2916 *
2917 * Returns 0 if there were neither overlaps nor errors while checking for
2918 * overlaps; or a negative value (-errno) on error.
2919 */
2920 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset,
2921 int64_t size, bool data_file)
2922 {
2923 int ret;
2924
2925 if (data_file && has_data_file(bs)) {
2926 return 0;
2927 }
2928
2929 ret = qcow2_check_metadata_overlap(bs, ign, offset, size);
2930 if (ret < 0) {
2931 return ret;
2932 } else if (ret > 0) {
2933 int metadata_ol_bitnr = ctz32(ret);
2934 assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR);
2935
2936 qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid "
2937 "write on metadata (overlaps with %s)",
2938 metadata_ol_names[metadata_ol_bitnr]);
2939 return -EIO;
2940 }
2941
2942 return 0;
2943 }
2944
2945 /* A pointer to a function of this type is given to walk_over_reftable(). That
2946 * function will create refblocks and pass them to a RefblockFinishOp once they
2947 * are completed (@refblock). @refblock_empty is set if the refblock is
2948 * completely empty.
2949 *
2950 * Along with the refblock, a corresponding reftable entry is passed, in the
2951 * reftable @reftable (which may be reallocated) at @reftable_index.
2952 *
2953 * @allocated should be set to true if a new cluster has been allocated.
2954 */
2955 typedef int (RefblockFinishOp)(BlockDriverState *bs, uint64_t **reftable,
2956 uint64_t reftable_index, uint64_t *reftable_size,
2957 void *refblock, bool refblock_empty,
2958 bool *allocated, Error **errp);
2959
2960 /**
2961 * This "operation" for walk_over_reftable() allocates the refblock on disk (if
2962 * it is not empty) and inserts its offset into the new reftable. The size of
2963 * this new reftable is increased as required.
2964 */
2965 static int alloc_refblock(BlockDriverState *bs, uint64_t **reftable,
2966 uint64_t reftable_index, uint64_t *reftable_size,
2967 void *refblock, bool refblock_empty, bool *allocated,
2968 Error **errp)
2969 {
2970 BDRVQcow2State *s = bs->opaque;
2971 int64_t offset;
2972
2973 if (!refblock_empty && reftable_index >= *reftable_size) {
2974 uint64_t *new_reftable;
2975 uint64_t new_reftable_size;
2976
2977 new_reftable_size = ROUND_UP(reftable_index + 1,
2978 s->cluster_size / REFTABLE_ENTRY_SIZE);
2979 if (new_reftable_size > QCOW_MAX_REFTABLE_SIZE / REFTABLE_ENTRY_SIZE) {
2980 error_setg(errp,
2981 "This operation would make the refcount table grow "
2982 "beyond the maximum size supported by QEMU, aborting");
2983 return -ENOTSUP;
2984 }
2985
2986 new_reftable = g_try_realloc(*reftable, new_reftable_size *
2987 REFTABLE_ENTRY_SIZE);
2988 if (!new_reftable) {
2989 error_setg(errp, "Failed to increase reftable buffer size");
2990 return -ENOMEM;
2991 }
2992
2993 memset(new_reftable + *reftable_size, 0,
2994 (new_reftable_size - *reftable_size) * REFTABLE_ENTRY_SIZE);
2995
2996 *reftable = new_reftable;
2997 *reftable_size = new_reftable_size;
2998 }
2999
3000 if (!refblock_empty && !(*reftable)[reftable_index]) {
3001 offset = qcow2_alloc_clusters(bs, s->cluster_size);
3002 if (offset < 0) {
3003 error_setg_errno(errp, -offset, "Failed to allocate refblock");
3004 return offset;
3005 }
3006 (*reftable)[reftable_index] = offset;
3007 *allocated = true;
3008 }
3009
3010 return 0;
3011 }
3012
3013 /**
3014 * This "operation" for walk_over_reftable() writes the refblock to disk at the
3015 * offset specified by the new reftable's entry. It does not modify the new
3016 * reftable or change any refcounts.
3017 */
3018 static int flush_refblock(BlockDriverState *bs, uint64_t **reftable,
3019 uint64_t reftable_index, uint64_t *reftable_size,
3020 void *refblock, bool refblock_empty, bool *allocated,
3021 Error **errp)
3022 {
3023 BDRVQcow2State *s = bs->opaque;
3024 int64_t offset;
3025 int ret;
3026
3027 if (reftable_index < *reftable_size && (*reftable)[reftable_index]) {
3028 offset = (*reftable)[reftable_index];
3029
3030 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size,
3031 false);
3032 if (ret < 0) {
3033 error_setg_errno(errp, -ret, "Overlap check failed");
3034 return ret;
3035 }
3036
3037 ret = bdrv_pwrite(bs->file, offset, refblock, s->cluster_size);
3038 if (ret < 0) {
3039 error_setg_errno(errp, -ret, "Failed to write refblock");
3040 return ret;
3041 }
3042 } else {
3043 assert(refblock_empty);
3044 }
3045
3046 return 0;
3047 }
3048
3049 /**
3050 * This function walks over the existing reftable and every referenced refblock;
3051 * if @new_set_refcount is non-NULL, it is called for every refcount entry to
3052 * create an equal new entry in the passed @new_refblock. Once that
3053 * @new_refblock is completely filled, @operation will be called.
3054 *
3055 * @status_cb and @cb_opaque are used for the amend operation's status callback.
3056 * @index is the index of the walk_over_reftable() calls and @total is the total
3057 * number of walk_over_reftable() calls per amend operation. Both are used for
3058 * calculating the parameters for the status callback.
3059 *
3060 * @allocated is set to true if a new cluster has been allocated.
3061 */
3062 static int walk_over_reftable(BlockDriverState *bs, uint64_t **new_reftable,
3063 uint64_t *new_reftable_index,
3064 uint64_t *new_reftable_size,
3065 void *new_refblock, int new_refblock_size,
3066 int new_refcount_bits,
3067 RefblockFinishOp *operation, bool *allocated,
3068 Qcow2SetRefcountFunc *new_set_refcount,
3069 BlockDriverAmendStatusCB *status_cb,
3070 void *cb_opaque, int index, int total,
3071 Error **errp)
3072 {
3073 BDRVQcow2State *s = bs->opaque;
3074 uint64_t reftable_index;
3075 bool new_refblock_empty = true;
3076 int refblock_index;
3077 int new_refblock_index = 0;
3078 int ret;
3079
3080 for (reftable_index = 0; reftable_index < s->refcount_table_size;
3081 reftable_index++)
3082 {
3083 uint64_t refblock_offset = s->refcount_table[reftable_index]
3084 & REFT_OFFSET_MASK;
3085
3086 status_cb(bs, (uint64_t)index * s->refcount_table_size + reftable_index,
3087 (uint64_t)total * s->refcount_table_size, cb_opaque);
3088
3089 if (refblock_offset) {
3090 void *refblock;
3091
3092 if (offset_into_cluster(s, refblock_offset)) {
3093 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
3094 PRIx64 " unaligned (reftable index: %#"
3095 PRIx64 ")", refblock_offset,
3096 reftable_index);
3097 error_setg(errp,
3098 "Image is corrupt (unaligned refblock offset)");
3099 return -EIO;
3100 }
3101
3102 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offset,
3103 &refblock);
3104 if (ret < 0) {
3105 error_setg_errno(errp, -ret, "Failed to retrieve refblock");
3106 return ret;
3107 }
3108
3109 for (refblock_index = 0; refblock_index < s->refcount_block_size;
3110 refblock_index++)
3111 {
3112 uint64_t refcount;
3113
3114 if (new_refblock_index >= new_refblock_size) {
3115 /* new_refblock is now complete */
3116 ret = operation(bs, new_reftable, *new_reftable_index,
3117 new_reftable_size, new_refblock,
3118 new_refblock_empty, allocated, errp);
3119 if (ret < 0) {
3120 qcow2_cache_put(s->refcount_block_cache, &refblock);
3121 return ret;
3122 }
3123
3124 (*new_reftable_index)++;
3125 new_refblock_index = 0;
3126 new_refblock_empty = true;
3127 }
3128
3129 refcount = s->get_refcount(refblock, refblock_index);
3130 if (new_refcount_bits < 64 && refcount >> new_refcount_bits) {
3131 uint64_t offset;
3132
3133 qcow2_cache_put(s->refcount_block_cache, &refblock);
3134
3135 offset = ((reftable_index << s->refcount_block_bits)
3136 + refblock_index) << s->cluster_bits;
3137
3138 error_setg(errp, "Cannot decrease refcount entry width to "
3139 "%i bits: Cluster at offset %#" PRIx64 " has a "
3140 "refcount of %" PRIu64, new_refcount_bits,
3141 offset, refcount);
3142 return -EINVAL;
3143 }
3144
3145 if (new_set_refcount) {
3146 new_set_refcount(new_refblock, new_refblock_index++,
3147 refcount);
3148 } else {
3149 new_refblock_index++;
3150 }
3151 new_refblock_empty = new_refblock_empty && refcount == 0;
3152 }
3153
3154 qcow2_cache_put(s->refcount_block_cache, &refblock);
3155 } else {
3156 /* No refblock means every refcount is 0 */
3157 for (refblock_index = 0; refblock_index < s->refcount_block_size;
3158 refblock_index++)
3159 {
3160 if (new_refblock_index >= new_refblock_size) {
3161 /* new_refblock is now complete */
3162 ret = operation(bs, new_reftable, *new_reftable_index,
3163 new_reftable_size, new_refblock,
3164 new_refblock_empty, allocated, errp);
3165 if (ret < 0) {
3166 return ret;
3167 }
3168
3169 (*new_reftable_index)++;
3170 new_refblock_index = 0;
3171 new_refblock_empty = true;
3172 }
3173
3174 if (new_set_refcount) {
3175 new_set_refcount(new_refblock, new_refblock_index++, 0);
3176 } else {
3177 new_refblock_index++;
3178 }
3179 }
3180 }
3181 }
3182
3183 if (new_refblock_index > 0) {
3184 /* Complete the potentially existing partially filled final refblock */
3185 if (new_set_refcount) {
3186 for (; new_refblock_index < new_refblock_size;
3187 new_refblock_index++)
3188 {
3189 new_set_refcount(new_refblock, new_refblock_index, 0);
3190 }
3191 }
3192
3193 ret = operation(bs, new_reftable, *new_reftable_index,
3194 new_reftable_size, new_refblock, new_refblock_empty,
3195 allocated, errp);
3196 if (ret < 0) {
3197 return ret;
3198 }
3199
3200 (*new_reftable_index)++;
3201 }
3202
3203 status_cb(bs, (uint64_t)(index + 1) * s->refcount_table_size,
3204 (uint64_t)total * s->refcount_table_size, cb_opaque);
3205
3206 return 0;
3207 }
3208
3209 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order,
3210 BlockDriverAmendStatusCB *status_cb,
3211 void *cb_opaque, Error **errp)
3212 {
3213 BDRVQcow2State *s = bs->opaque;
3214 Qcow2GetRefcountFunc *new_get_refcount;
3215 Qcow2SetRefcountFunc *new_set_refcount;
3216 void *new_refblock = qemu_blockalign(bs->file->bs, s->cluster_size);
3217 uint64_t *new_reftable = NULL, new_reftable_size = 0;
3218 uint64_t *old_reftable, old_reftable_size, old_reftable_offset;
3219 uint64_t new_reftable_index = 0;
3220 uint64_t i;
3221 int64_t new_reftable_offset = 0, allocated_reftable_size = 0;
3222 int new_refblock_size, new_refcount_bits = 1 << refcount_order;
3223 int old_refcount_order;
3224 int walk_index = 0;
3225 int ret;
3226 bool new_allocation;
3227
3228 assert(s->qcow_version >= 3);
3229 assert(refcount_order >= 0 && refcount_order <= 6);
3230
3231 /* see qcow2_open() */
3232 new_refblock_size = 1 << (s->cluster_bits - (refcount_order - 3));
3233
3234 new_get_refcount = get_refcount_funcs[refcount_order];
3235 new_set_refcount = set_refcount_funcs[refcount_order];
3236
3237
3238 do {
3239 int total_walks;
3240
3241 new_allocation = false;
3242
3243 /* At least we have to do this walk and the one which writes the
3244 * refblocks; also, at least we have to do this loop here at least
3245 * twice (normally), first to do the allocations, and second to
3246 * determine that everything is correctly allocated, this then makes
3247 * three walks in total */
3248 total_walks = MAX(walk_index + 2, 3);
3249
3250 /* First, allocate the structures so they are present in the refcount
3251 * structures */
3252 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
3253 &new_reftable_size, NULL, new_refblock_size,
3254 new_refcount_bits, &alloc_refblock,
3255 &new_allocation, NULL, status_cb, cb_opaque,
3256 walk_index++, total_walks, errp);
3257 if (ret < 0) {
3258 goto done;
3259 }
3260
3261 new_reftable_index = 0;
3262
3263 if (new_allocation) {
3264 if (new_reftable_offset) {
3265 qcow2_free_clusters(
3266 bs, new_reftable_offset,
3267 allocated_reftable_size * REFTABLE_ENTRY_SIZE,
3268 QCOW2_DISCARD_NEVER);
3269 }
3270
3271 new_reftable_offset = qcow2_alloc_clusters(bs, new_reftable_size *
3272 REFTABLE_ENTRY_SIZE);
3273 if (new_reftable_offset < 0) {
3274 error_setg_errno(errp, -new_reftable_offset,
3275 "Failed to allocate the new reftable");
3276 ret = new_reftable_offset;
3277 goto done;
3278 }
3279 allocated_reftable_size = new_reftable_size;
3280 }
3281 } while (new_allocation);
3282
3283 /* Second, write the new refblocks */
3284 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
3285 &new_reftable_size, new_refblock,
3286 new_refblock_size, new_refcount_bits,
3287 &flush_refblock, &new_allocation, new_set_refcount,
3288 status_cb, cb_opaque, walk_index, walk_index + 1,
3289 errp);
3290 if (ret < 0) {
3291 goto done;
3292 }
3293 assert(!new_allocation);
3294
3295
3296 /* Write the new reftable */
3297 ret = qcow2_pre_write_overlap_check(bs, 0, new_reftable_offset,
3298 new_reftable_size * REFTABLE_ENTRY_SIZE,
3299 false);
3300 if (ret < 0) {
3301 error_setg_errno(errp, -ret, "Overlap check failed");
3302 goto done;
3303 }
3304
3305 for (i = 0; i < new_reftable_size; i++) {
3306 cpu_to_be64s(&new_reftable[i]);
3307 }
3308
3309 ret = bdrv_pwrite(bs->file, new_reftable_offset, new_reftable,
3310 new_reftable_size * REFTABLE_ENTRY_SIZE);
3311
3312 for (i = 0; i < new_reftable_size; i++) {
3313 be64_to_cpus(&new_reftable[i]);
3314 }
3315
3316 if (ret < 0) {
3317 error_setg_errno(errp, -ret, "Failed to write the new reftable");
3318 goto done;
3319 }
3320
3321
3322 /* Empty the refcount cache */
3323 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
3324 if (ret < 0) {
3325 error_setg_errno(errp, -ret, "Failed to flush the refblock cache");
3326 goto done;
3327 }
3328
3329 /* Update the image header to point to the new reftable; this only updates
3330 * the fields which are relevant to qcow2_update_header(); other fields
3331 * such as s->refcount_table or s->refcount_bits stay stale for now
3332 * (because we have to restore everything if qcow2_update_header() fails) */
3333 old_refcount_order = s->refcount_order;
3334 old_reftable_size = s->refcount_table_size;
3335 old_reftable_offset = s->refcount_table_offset;
3336
3337 s->refcount_order = refcount_order;
3338 s->refcount_table_size = new_reftable_size;
3339 s->refcount_table_offset = new_reftable_offset;
3340
3341 ret = qcow2_update_header(bs);
3342 if (ret < 0) {
3343 s->refcount_order = old_refcount_order;
3344 s->refcount_table_size = old_reftable_size;
3345 s->refcount_table_offset = old_reftable_offset;
3346 error_setg_errno(errp, -ret, "Failed to update the qcow2 header");
3347 goto done;
3348 }
3349
3350 /* Now update the rest of the in-memory information */
3351 old_reftable = s->refcount_table;
3352 s->refcount_table = new_reftable;
3353 update_max_refcount_table_index(s);
3354
3355 s->refcount_bits = 1 << refcount_order;
3356 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
3357 s->refcount_max += s->refcount_max - 1;
3358
3359 s->refcount_block_bits = s->cluster_bits - (refcount_order - 3);
3360 s->refcount_block_size = 1 << s->refcount_block_bits;
3361
3362 s->get_refcount = new_get_refcount;
3363 s->set_refcount = new_set_refcount;
3364
3365 /* For cleaning up all old refblocks and the old reftable below the "done"
3366 * label */
3367 new_reftable = old_reftable;
3368 new_reftable_size = old_reftable_size;
3369 new_reftable_offset = old_reftable_offset;
3370
3371 done:
3372 if (new_reftable) {
3373 /* On success, new_reftable actually points to the old reftable (and
3374 * new_reftable_size is the old reftable's size); but that is just
3375 * fine */
3376 for (i = 0; i < new_reftable_size; i++) {
3377 uint64_t offset = new_reftable[i] & REFT_OFFSET_MASK;
3378 if (offset) {
3379 qcow2_free_clusters(bs, offset, s->cluster_size,
3380 QCOW2_DISCARD_OTHER);
3381 }
3382 }
3383 g_free(new_reftable);
3384
3385 if (new_reftable_offset > 0) {
3386 qcow2_free_clusters(bs, new_reftable_offset,
3387 new_reftable_size * REFTABLE_ENTRY_SIZE,
3388 QCOW2_DISCARD_OTHER);
3389 }
3390 }
3391
3392 qemu_vfree(new_refblock);
3393 return ret;
3394 }
3395
3396 static int64_t get_refblock_offset(BlockDriverState *bs, uint64_t offset)
3397 {
3398 BDRVQcow2State *s = bs->opaque;
3399 uint32_t index = offset_to_reftable_index(s, offset);
3400 int64_t covering_refblock_offset = 0;
3401
3402 if (index < s->refcount_table_size) {
3403 covering_refblock_offset = s->refcount_table[index] & REFT_OFFSET_MASK;
3404 }
3405 if (!covering_refblock_offset) {
3406 qcow2_signal_corruption(bs, true, -1, -1, "Refblock at %#" PRIx64 " is "
3407 "not covered by the refcount structures",
3408 offset);
3409 return -EIO;
3410 }
3411
3412 return covering_refblock_offset;
3413 }
3414
3415 static int qcow2_discard_refcount_block(BlockDriverState *bs,
3416 uint64_t discard_block_offs)
3417 {
3418 BDRVQcow2State *s = bs->opaque;
3419 int64_t refblock_offs;
3420 uint64_t cluster_index = discard_block_offs >> s->cluster_bits;
3421 uint32_t block_index = cluster_index & (s->refcount_block_size - 1);
3422 void *refblock;
3423 int ret;
3424
3425 refblock_offs = get_refblock_offset(bs, discard_block_offs);
3426 if (refblock_offs < 0) {
3427 return refblock_offs;
3428 }
3429
3430 assert(discard_block_offs != 0);
3431
3432 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs,
3433 &refblock);
3434 if (ret < 0) {
3435 return ret;
3436 }
3437
3438 if (s->get_refcount(refblock, block_index) != 1) {
3439 qcow2_signal_corruption(bs, true, -1, -1, "Invalid refcount:"
3440 " refblock offset %#" PRIx64
3441 ", reftable index %u"
3442 ", block offset %#" PRIx64
3443 ", refcount %#" PRIx64,
3444 refblock_offs,
3445 offset_to_reftable_index(s, discard_block_offs),
3446 discard_block_offs,
3447 s->get_refcount(refblock, block_index));
3448 qcow2_cache_put(s->refcount_block_cache, &refblock);
3449 return -EINVAL;
3450 }
3451 s->set_refcount(refblock, block_index, 0);
3452
3453 qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refblock);
3454
3455 qcow2_cache_put(s->refcount_block_cache, &refblock);
3456
3457 if (cluster_index < s->free_cluster_index) {
3458 s->free_cluster_index = cluster_index;
3459 }
3460
3461 refblock = qcow2_cache_is_table_offset(s->refcount_block_cache,
3462 discard_block_offs);
3463 if (refblock) {
3464 /* discard refblock from the cache if refblock is cached */
3465 qcow2_cache_discard(s->refcount_block_cache, refblock);
3466 }
3467 update_refcount_discard(bs, discard_block_offs, s->cluster_size);
3468
3469 return 0;
3470 }
3471
3472 int qcow2_shrink_reftable(BlockDriverState *bs)
3473 {
3474 BDRVQcow2State *s = bs->opaque;
3475 uint64_t *reftable_tmp =
3476 g_malloc(s->refcount_table_size * REFTABLE_ENTRY_SIZE);
3477 int i, ret;
3478
3479 for (i = 0; i < s->refcount_table_size; i++) {
3480 int64_t refblock_offs = s->refcount_table[i] & REFT_OFFSET_MASK;
3481 void *refblock;
3482 bool unused_block;
3483
3484 if (refblock_offs == 0) {
3485 reftable_tmp[i] = 0;
3486 continue;
3487 }
3488 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs,
3489 &refblock);
3490 if (ret < 0) {
3491 goto out;
3492 }
3493
3494 /* the refblock has own reference */
3495 if (i == offset_to_reftable_index(s, refblock_offs)) {
3496 uint64_t block_index = (refblock_offs >> s->cluster_bits) &
3497 (s->refcount_block_size - 1);
3498 uint64_t refcount = s->get_refcount(refblock, block_index);
3499
3500 s->set_refcount(refblock, block_index, 0);
3501
3502 unused_block = buffer_is_zero(refblock, s->cluster_size);
3503
3504 s->set_refcount(refblock, block_index, refcount);
3505 } else {
3506 unused_block = buffer_is_zero(refblock, s->cluster_size);
3507 }
3508 qcow2_cache_put(s->refcount_block_cache, &refblock);
3509
3510 reftable_tmp[i] = unused_block ? 0 : cpu_to_be64(s->refcount_table[i]);
3511 }
3512
3513 ret = bdrv_pwrite_sync(bs->file, s->refcount_table_offset, reftable_tmp,
3514 s->refcount_table_size * REFTABLE_ENTRY_SIZE);
3515 /*
3516 * If the write in the reftable failed the image may contain a partially
3517 * overwritten reftable. In this case it would be better to clear the
3518 * reftable in memory to avoid possible image corruption.
3519 */
3520 for (i = 0; i < s->refcount_table_size; i++) {
3521 if (s->refcount_table[i] && !reftable_tmp[i]) {
3522 if (ret == 0) {
3523 ret = qcow2_discard_refcount_block(bs, s->refcount_table[i] &
3524 REFT_OFFSET_MASK);
3525 }
3526 s->refcount_table[i] = 0;
3527 }
3528 }
3529
3530 if (!s->cache_discards) {
3531 qcow2_process_discards(bs, ret);
3532 }
3533
3534 out:
3535 g_free(reftable_tmp);
3536 return ret;
3537 }
3538
3539 int64_t qcow2_get_last_cluster(BlockDriverState *bs, int64_t size)
3540 {
3541 BDRVQcow2State *s = bs->opaque;
3542 int64_t i;
3543
3544 for (i = size_to_clusters(s, size) - 1; i >= 0; i--) {
3545 uint64_t refcount;
3546 int ret = qcow2_get_refcount(bs, i, &refcount);
3547 if (ret < 0) {
3548 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
3549 i, strerror(-ret));
3550 return ret;
3551 }
3552 if (refcount > 0) {
3553 return i;
3554 }
3555 }
3556 qcow2_signal_corruption(bs, true, -1, -1,
3557 "There are no references in the refcount table.");
3558 return -EIO;
3559 }
3560
3561 int qcow2_detect_metadata_preallocation(BlockDriverState *bs)
3562 {
3563 BDRVQcow2State *s = bs->opaque;
3564 int64_t i, end_cluster, cluster_count = 0, threshold;
3565 int64_t file_length, real_allocation, real_clusters;
3566
3567 qemu_co_mutex_assert_locked(&s->lock);
3568
3569 file_length = bdrv_getlength(bs->file->bs);
3570 if (file_length < 0) {
3571 return file_length;
3572 }
3573
3574 real_allocation = bdrv_get_allocated_file_size(bs->file->bs);
3575 if (real_allocation < 0) {
3576 return real_allocation;
3577 }
3578
3579 real_clusters = real_allocation / s->cluster_size;
3580 threshold = MAX(real_clusters * 10 / 9, real_clusters + 2);
3581
3582 end_cluster = size_to_clusters(s, file_length);
3583 for (i = 0; i < end_cluster && cluster_count < threshold; i++) {
3584 uint64_t refcount;
3585 int ret = qcow2_get_refcount(bs, i, &refcount);
3586 if (ret < 0) {
3587 return ret;
3588 }
3589 cluster_count += !!refcount;
3590 }
3591
3592 return cluster_count >= threshold;
3593 }