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