]> git.proxmox.com Git - mirror_qemu.git/blob - block/qcow2-refcount.c
block: release persistent bitmaps on inactivate
[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 "qemu-common.h"
28 #include "block/block_int.h"
29 #include "block/qcow2.h"
30 #include "qemu/range.h"
31 #include "qemu/bswap.h"
32
33 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size);
34 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
35 int64_t offset, int64_t length, uint64_t addend,
36 bool decrease, enum qcow2_discard_type type);
37
38 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index);
39 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index);
40 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index);
41 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index);
42 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index);
43 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index);
44 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index);
45
46 static void set_refcount_ro0(void *refcount_array, uint64_t index,
47 uint64_t value);
48 static void set_refcount_ro1(void *refcount_array, uint64_t index,
49 uint64_t value);
50 static void set_refcount_ro2(void *refcount_array, uint64_t index,
51 uint64_t value);
52 static void set_refcount_ro3(void *refcount_array, uint64_t index,
53 uint64_t value);
54 static void set_refcount_ro4(void *refcount_array, uint64_t index,
55 uint64_t value);
56 static void set_refcount_ro5(void *refcount_array, uint64_t index,
57 uint64_t value);
58 static void set_refcount_ro6(void *refcount_array, uint64_t index,
59 uint64_t value);
60
61
62 static Qcow2GetRefcountFunc *const get_refcount_funcs[] = {
63 &get_refcount_ro0,
64 &get_refcount_ro1,
65 &get_refcount_ro2,
66 &get_refcount_ro3,
67 &get_refcount_ro4,
68 &get_refcount_ro5,
69 &get_refcount_ro6
70 };
71
72 static Qcow2SetRefcountFunc *const set_refcount_funcs[] = {
73 &set_refcount_ro0,
74 &set_refcount_ro1,
75 &set_refcount_ro2,
76 &set_refcount_ro3,
77 &set_refcount_ro4,
78 &set_refcount_ro5,
79 &set_refcount_ro6
80 };
81
82
83 /*********************************************************/
84 /* refcount handling */
85
86 static void update_max_refcount_table_index(BDRVQcow2State *s)
87 {
88 unsigned i = s->refcount_table_size - 1;
89 while (i > 0 && (s->refcount_table[i] & REFT_OFFSET_MASK) == 0) {
90 i--;
91 }
92 /* Set s->max_refcount_table_index to the index of the last used entry */
93 s->max_refcount_table_index = i;
94 }
95
96 int qcow2_refcount_init(BlockDriverState *bs)
97 {
98 BDRVQcow2State *s = bs->opaque;
99 unsigned int refcount_table_size2, i;
100 int ret;
101
102 assert(s->refcount_order >= 0 && s->refcount_order <= 6);
103
104 s->get_refcount = get_refcount_funcs[s->refcount_order];
105 s->set_refcount = set_refcount_funcs[s->refcount_order];
106
107 assert(s->refcount_table_size <= INT_MAX / sizeof(uint64_t));
108 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
109 s->refcount_table = g_try_malloc(refcount_table_size2);
110
111 if (s->refcount_table_size > 0) {
112 if (s->refcount_table == NULL) {
113 ret = -ENOMEM;
114 goto fail;
115 }
116 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
117 ret = bdrv_pread(bs->file, s->refcount_table_offset,
118 s->refcount_table, refcount_table_size2);
119 if (ret < 0) {
120 goto fail;
121 }
122 for(i = 0; i < s->refcount_table_size; i++)
123 be64_to_cpus(&s->refcount_table[i]);
124 update_max_refcount_table_index(s);
125 }
126 return 0;
127 fail:
128 return ret;
129 }
130
131 void qcow2_refcount_close(BlockDriverState *bs)
132 {
133 BDRVQcow2State *s = bs->opaque;
134 g_free(s->refcount_table);
135 }
136
137
138 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index)
139 {
140 return (((const uint8_t *)refcount_array)[index / 8] >> (index % 8)) & 0x1;
141 }
142
143 static void set_refcount_ro0(void *refcount_array, uint64_t index,
144 uint64_t value)
145 {
146 assert(!(value >> 1));
147 ((uint8_t *)refcount_array)[index / 8] &= ~(0x1 << (index % 8));
148 ((uint8_t *)refcount_array)[index / 8] |= value << (index % 8);
149 }
150
151 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index)
152 {
153 return (((const uint8_t *)refcount_array)[index / 4] >> (2 * (index % 4)))
154 & 0x3;
155 }
156
157 static void set_refcount_ro1(void *refcount_array, uint64_t index,
158 uint64_t value)
159 {
160 assert(!(value >> 2));
161 ((uint8_t *)refcount_array)[index / 4] &= ~(0x3 << (2 * (index % 4)));
162 ((uint8_t *)refcount_array)[index / 4] |= value << (2 * (index % 4));
163 }
164
165 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index)
166 {
167 return (((const uint8_t *)refcount_array)[index / 2] >> (4 * (index % 2)))
168 & 0xf;
169 }
170
171 static void set_refcount_ro2(void *refcount_array, uint64_t index,
172 uint64_t value)
173 {
174 assert(!(value >> 4));
175 ((uint8_t *)refcount_array)[index / 2] &= ~(0xf << (4 * (index % 2)));
176 ((uint8_t *)refcount_array)[index / 2] |= value << (4 * (index % 2));
177 }
178
179 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index)
180 {
181 return ((const uint8_t *)refcount_array)[index];
182 }
183
184 static void set_refcount_ro3(void *refcount_array, uint64_t index,
185 uint64_t value)
186 {
187 assert(!(value >> 8));
188 ((uint8_t *)refcount_array)[index] = value;
189 }
190
191 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index)
192 {
193 return be16_to_cpu(((const uint16_t *)refcount_array)[index]);
194 }
195
196 static void set_refcount_ro4(void *refcount_array, uint64_t index,
197 uint64_t value)
198 {
199 assert(!(value >> 16));
200 ((uint16_t *)refcount_array)[index] = cpu_to_be16(value);
201 }
202
203 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index)
204 {
205 return be32_to_cpu(((const uint32_t *)refcount_array)[index]);
206 }
207
208 static void set_refcount_ro5(void *refcount_array, uint64_t index,
209 uint64_t value)
210 {
211 assert(!(value >> 32));
212 ((uint32_t *)refcount_array)[index] = cpu_to_be32(value);
213 }
214
215 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index)
216 {
217 return be64_to_cpu(((const uint64_t *)refcount_array)[index]);
218 }
219
220 static void set_refcount_ro6(void *refcount_array, uint64_t index,
221 uint64_t value)
222 {
223 ((uint64_t *)refcount_array)[index] = cpu_to_be64(value);
224 }
225
226
227 static int load_refcount_block(BlockDriverState *bs,
228 int64_t refcount_block_offset,
229 void **refcount_block)
230 {
231 BDRVQcow2State *s = bs->opaque;
232
233 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
234 return qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
235 refcount_block);
236 }
237
238 /*
239 * Retrieves the refcount of the cluster given by its index and stores it in
240 * *refcount. Returns 0 on success and -errno on failure.
241 */
242 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index,
243 uint64_t *refcount)
244 {
245 BDRVQcow2State *s = bs->opaque;
246 uint64_t refcount_table_index, block_index;
247 int64_t refcount_block_offset;
248 int ret;
249 void *refcount_block;
250
251 refcount_table_index = cluster_index >> s->refcount_block_bits;
252 if (refcount_table_index >= s->refcount_table_size) {
253 *refcount = 0;
254 return 0;
255 }
256 refcount_block_offset =
257 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
258 if (!refcount_block_offset) {
259 *refcount = 0;
260 return 0;
261 }
262
263 if (offset_into_cluster(s, refcount_block_offset)) {
264 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" PRIx64
265 " unaligned (reftable index: %#" PRIx64 ")",
266 refcount_block_offset, refcount_table_index);
267 return -EIO;
268 }
269
270 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
271 &refcount_block);
272 if (ret < 0) {
273 return ret;
274 }
275
276 block_index = cluster_index & (s->refcount_block_size - 1);
277 *refcount = s->get_refcount(refcount_block, block_index);
278
279 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
280
281 return 0;
282 }
283
284 /*
285 * Rounds the refcount table size up to avoid growing the table for each single
286 * refcount block that is allocated.
287 */
288 static unsigned int next_refcount_table_size(BDRVQcow2State *s,
289 unsigned int min_size)
290 {
291 unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1;
292 unsigned int refcount_table_clusters =
293 MAX(1, s->refcount_table_size >> (s->cluster_bits - 3));
294
295 while (min_clusters > refcount_table_clusters) {
296 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
297 }
298
299 return refcount_table_clusters << (s->cluster_bits - 3);
300 }
301
302
303 /* Checks if two offsets are described by the same refcount block */
304 static int in_same_refcount_block(BDRVQcow2State *s, uint64_t offset_a,
305 uint64_t offset_b)
306 {
307 uint64_t block_a = offset_a >> (s->cluster_bits + s->refcount_block_bits);
308 uint64_t block_b = offset_b >> (s->cluster_bits + s->refcount_block_bits);
309
310 return (block_a == block_b);
311 }
312
313 /*
314 * Loads a refcount block. If it doesn't exist yet, it is allocated first
315 * (including growing the refcount table if needed).
316 *
317 * Returns 0 on success or -errno in error case
318 */
319 static int alloc_refcount_block(BlockDriverState *bs,
320 int64_t cluster_index, void **refcount_block)
321 {
322 BDRVQcow2State *s = bs->opaque;
323 unsigned int refcount_table_index;
324 int ret;
325
326 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
327
328 /* Find the refcount block for the given cluster */
329 refcount_table_index = cluster_index >> s->refcount_block_bits;
330
331 if (refcount_table_index < s->refcount_table_size) {
332
333 uint64_t refcount_block_offset =
334 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
335
336 /* If it's already there, we're done */
337 if (refcount_block_offset) {
338 if (offset_into_cluster(s, refcount_block_offset)) {
339 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
340 PRIx64 " unaligned (reftable index: "
341 "%#x)", refcount_block_offset,
342 refcount_table_index);
343 return -EIO;
344 }
345
346 return load_refcount_block(bs, refcount_block_offset,
347 refcount_block);
348 }
349 }
350
351 /*
352 * If we came here, we need to allocate something. Something is at least
353 * a cluster for the new refcount block. It may also include a new refcount
354 * table if the old refcount table is too small.
355 *
356 * Note that allocating clusters here needs some special care:
357 *
358 * - We can't use the normal qcow2_alloc_clusters(), it would try to
359 * increase the refcount and very likely we would end up with an endless
360 * recursion. Instead we must place the refcount blocks in a way that
361 * they can describe them themselves.
362 *
363 * - We need to consider that at this point we are inside update_refcounts
364 * and potentially doing an initial refcount increase. This means that
365 * some clusters have already been allocated by the caller, but their
366 * refcount isn't accurate yet. If we allocate clusters for metadata, we
367 * need to return -EAGAIN to signal the caller that it needs to restart
368 * the search for free clusters.
369 *
370 * - alloc_clusters_noref and qcow2_free_clusters may load a different
371 * refcount block into the cache
372 */
373
374 *refcount_block = NULL;
375
376 /* We write to the refcount table, so we might depend on L2 tables */
377 ret = qcow2_cache_flush(bs, s->l2_table_cache);
378 if (ret < 0) {
379 return ret;
380 }
381
382 /* Allocate the refcount block itself and mark it as used */
383 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
384 if (new_block < 0) {
385 return new_block;
386 }
387
388 #ifdef DEBUG_ALLOC2
389 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
390 " at %" PRIx64 "\n",
391 refcount_table_index, cluster_index << s->cluster_bits, new_block);
392 #endif
393
394 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
395 /* Zero the new refcount block before updating it */
396 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
397 refcount_block);
398 if (ret < 0) {
399 goto fail_block;
400 }
401
402 memset(*refcount_block, 0, s->cluster_size);
403
404 /* The block describes itself, need to update the cache */
405 int block_index = (new_block >> s->cluster_bits) &
406 (s->refcount_block_size - 1);
407 s->set_refcount(*refcount_block, block_index, 1);
408 } else {
409 /* Described somewhere else. This can recurse at most twice before we
410 * arrive at a block that describes itself. */
411 ret = update_refcount(bs, new_block, s->cluster_size, 1, false,
412 QCOW2_DISCARD_NEVER);
413 if (ret < 0) {
414 goto fail_block;
415 }
416
417 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
418 if (ret < 0) {
419 goto fail_block;
420 }
421
422 /* Initialize the new refcount block only after updating its refcount,
423 * update_refcount uses the refcount cache itself */
424 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
425 refcount_block);
426 if (ret < 0) {
427 goto fail_block;
428 }
429
430 memset(*refcount_block, 0, s->cluster_size);
431 }
432
433 /* Now the new refcount block needs to be written to disk */
434 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
435 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache, *refcount_block);
436 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
437 if (ret < 0) {
438 goto fail_block;
439 }
440
441 /* If the refcount table is big enough, just hook the block up there */
442 if (refcount_table_index < s->refcount_table_size) {
443 uint64_t data64 = cpu_to_be64(new_block);
444 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
445 ret = bdrv_pwrite_sync(bs->file,
446 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
447 &data64, sizeof(data64));
448 if (ret < 0) {
449 goto fail_block;
450 }
451
452 s->refcount_table[refcount_table_index] = new_block;
453 /* If there's a hole in s->refcount_table then it can happen
454 * that refcount_table_index < s->max_refcount_table_index */
455 s->max_refcount_table_index =
456 MAX(s->max_refcount_table_index, refcount_table_index);
457
458 /* The new refcount block may be where the caller intended to put its
459 * data, so let it restart the search. */
460 return -EAGAIN;
461 }
462
463 qcow2_cache_put(bs, s->refcount_block_cache, refcount_block);
464
465 /*
466 * If we come here, we need to grow the refcount table. Again, a new
467 * refcount table needs some space and we can't simply allocate to avoid
468 * endless recursion.
469 *
470 * Therefore let's grab new refcount blocks at the end of the image, which
471 * will describe themselves and the new refcount table. This way we can
472 * reference them only in the new table and do the switch to the new
473 * refcount table at once without producing an inconsistent state in
474 * between.
475 */
476 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
477
478 /* Calculate the number of refcount blocks needed so far; this will be the
479 * basis for calculating the index of the first cluster used for the
480 * self-describing refcount structures which we are about to create.
481 *
482 * Because we reached this point, there cannot be any refcount entries for
483 * cluster_index or higher indices yet. However, because new_block has been
484 * allocated to describe that cluster (and it will assume this role later
485 * on), we cannot use that index; also, new_block may actually have a higher
486 * cluster index than cluster_index, so it needs to be taken into account
487 * here (and 1 needs to be added to its value because that cluster is used).
488 */
489 uint64_t blocks_used = DIV_ROUND_UP(MAX(cluster_index + 1,
490 (new_block >> s->cluster_bits) + 1),
491 s->refcount_block_size);
492
493 if (blocks_used > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) {
494 return -EFBIG;
495 }
496
497 /* And now we need at least one block more for the new metadata */
498 uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
499 uint64_t last_table_size;
500 uint64_t blocks_clusters;
501 do {
502 uint64_t table_clusters =
503 size_to_clusters(s, table_size * sizeof(uint64_t));
504 blocks_clusters = 1 +
505 DIV_ROUND_UP(table_clusters, s->refcount_block_size);
506 uint64_t meta_clusters = table_clusters + blocks_clusters;
507
508 last_table_size = table_size;
509 table_size = next_refcount_table_size(s, blocks_used +
510 DIV_ROUND_UP(meta_clusters, s->refcount_block_size));
511
512 } while (last_table_size != table_size);
513
514 #ifdef DEBUG_ALLOC2
515 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
516 s->refcount_table_size, table_size);
517 #endif
518
519 /* Create the new refcount table and blocks */
520 uint64_t meta_offset = (blocks_used * s->refcount_block_size) *
521 s->cluster_size;
522 uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
523 uint64_t *new_table = g_try_new0(uint64_t, table_size);
524 void *new_blocks = g_try_malloc0(blocks_clusters * s->cluster_size);
525
526 assert(table_size > 0 && blocks_clusters > 0);
527 if (new_table == NULL || new_blocks == NULL) {
528 ret = -ENOMEM;
529 goto fail_table;
530 }
531
532 /* Fill the new refcount table */
533 memcpy(new_table, s->refcount_table,
534 s->refcount_table_size * sizeof(uint64_t));
535 new_table[refcount_table_index] = new_block;
536
537 int i;
538 for (i = 0; i < blocks_clusters; i++) {
539 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
540 }
541
542 /* Fill the refcount blocks */
543 uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
544 int block = 0;
545 for (i = 0; i < table_clusters + blocks_clusters; i++) {
546 s->set_refcount(new_blocks, block++, 1);
547 }
548
549 /* Write refcount blocks to disk */
550 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
551 ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
552 blocks_clusters * s->cluster_size);
553 g_free(new_blocks);
554 new_blocks = NULL;
555 if (ret < 0) {
556 goto fail_table;
557 }
558
559 /* Write refcount table to disk */
560 for(i = 0; i < table_size; i++) {
561 cpu_to_be64s(&new_table[i]);
562 }
563
564 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
565 ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
566 table_size * sizeof(uint64_t));
567 if (ret < 0) {
568 goto fail_table;
569 }
570
571 for(i = 0; i < table_size; i++) {
572 be64_to_cpus(&new_table[i]);
573 }
574
575 /* Hook up the new refcount table in the qcow2 header */
576 struct QEMU_PACKED {
577 uint64_t d64;
578 uint32_t d32;
579 } data;
580 data.d64 = cpu_to_be64(table_offset);
581 data.d32 = cpu_to_be32(table_clusters);
582 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
583 ret = bdrv_pwrite_sync(bs->file,
584 offsetof(QCowHeader, refcount_table_offset),
585 &data, sizeof(data));
586 if (ret < 0) {
587 goto fail_table;
588 }
589
590 /* And switch it in memory */
591 uint64_t old_table_offset = s->refcount_table_offset;
592 uint64_t old_table_size = s->refcount_table_size;
593
594 g_free(s->refcount_table);
595 s->refcount_table = new_table;
596 s->refcount_table_size = table_size;
597 s->refcount_table_offset = table_offset;
598 update_max_refcount_table_index(s);
599
600 /* Free old table. */
601 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t),
602 QCOW2_DISCARD_OTHER);
603
604 ret = load_refcount_block(bs, new_block, refcount_block);
605 if (ret < 0) {
606 return ret;
607 }
608
609 /* If we were trying to do the initial refcount update for some cluster
610 * allocation, we might have used the same clusters to store newly
611 * allocated metadata. Make the caller search some new space. */
612 return -EAGAIN;
613
614 fail_table:
615 g_free(new_blocks);
616 g_free(new_table);
617 fail_block:
618 if (*refcount_block != NULL) {
619 qcow2_cache_put(bs, s->refcount_block_cache, refcount_block);
620 }
621 return ret;
622 }
623
624 void qcow2_process_discards(BlockDriverState *bs, int ret)
625 {
626 BDRVQcow2State *s = bs->opaque;
627 Qcow2DiscardRegion *d, *next;
628
629 QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) {
630 QTAILQ_REMOVE(&s->discards, d, next);
631
632 /* Discard is optional, ignore the return value */
633 if (ret >= 0) {
634 bdrv_pdiscard(bs->file->bs, d->offset, d->bytes);
635 }
636
637 g_free(d);
638 }
639 }
640
641 static void update_refcount_discard(BlockDriverState *bs,
642 uint64_t offset, uint64_t length)
643 {
644 BDRVQcow2State *s = bs->opaque;
645 Qcow2DiscardRegion *d, *p, *next;
646
647 QTAILQ_FOREACH(d, &s->discards, next) {
648 uint64_t new_start = MIN(offset, d->offset);
649 uint64_t new_end = MAX(offset + length, d->offset + d->bytes);
650
651 if (new_end - new_start <= length + d->bytes) {
652 /* There can't be any overlap, areas ending up here have no
653 * references any more and therefore shouldn't get freed another
654 * time. */
655 assert(d->bytes + length == new_end - new_start);
656 d->offset = new_start;
657 d->bytes = new_end - new_start;
658 goto found;
659 }
660 }
661
662 d = g_malloc(sizeof(*d));
663 *d = (Qcow2DiscardRegion) {
664 .bs = bs,
665 .offset = offset,
666 .bytes = length,
667 };
668 QTAILQ_INSERT_TAIL(&s->discards, d, next);
669
670 found:
671 /* Merge discard requests if they are adjacent now */
672 QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) {
673 if (p == d
674 || p->offset > d->offset + d->bytes
675 || d->offset > p->offset + p->bytes)
676 {
677 continue;
678 }
679
680 /* Still no overlap possible */
681 assert(p->offset == d->offset + d->bytes
682 || d->offset == p->offset + p->bytes);
683
684 QTAILQ_REMOVE(&s->discards, p, next);
685 d->offset = MIN(d->offset, p->offset);
686 d->bytes += p->bytes;
687 g_free(p);
688 }
689 }
690
691 /* XXX: cache several refcount block clusters ? */
692 /* @addend is the absolute value of the addend; if @decrease is set, @addend
693 * will be subtracted from the current refcount, otherwise it will be added */
694 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
695 int64_t offset,
696 int64_t length,
697 uint64_t addend,
698 bool decrease,
699 enum qcow2_discard_type type)
700 {
701 BDRVQcow2State *s = bs->opaque;
702 int64_t start, last, cluster_offset;
703 void *refcount_block = NULL;
704 int64_t old_table_index = -1;
705 int ret;
706
707 #ifdef DEBUG_ALLOC2
708 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64
709 " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "",
710 addend);
711 #endif
712 if (length < 0) {
713 return -EINVAL;
714 } else if (length == 0) {
715 return 0;
716 }
717
718 if (decrease) {
719 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
720 s->l2_table_cache);
721 }
722
723 start = start_of_cluster(s, offset);
724 last = start_of_cluster(s, offset + length - 1);
725 for(cluster_offset = start; cluster_offset <= last;
726 cluster_offset += s->cluster_size)
727 {
728 int block_index;
729 uint64_t refcount;
730 int64_t cluster_index = cluster_offset >> s->cluster_bits;
731 int64_t table_index = cluster_index >> s->refcount_block_bits;
732
733 /* Load the refcount block and allocate it if needed */
734 if (table_index != old_table_index) {
735 if (refcount_block) {
736 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
737 }
738 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
739 if (ret < 0) {
740 goto fail;
741 }
742 }
743 old_table_index = table_index;
744
745 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache,
746 refcount_block);
747
748 /* we can update the count and save it */
749 block_index = cluster_index & (s->refcount_block_size - 1);
750
751 refcount = s->get_refcount(refcount_block, block_index);
752 if (decrease ? (refcount - addend > refcount)
753 : (refcount + addend < refcount ||
754 refcount + addend > s->refcount_max))
755 {
756 ret = -EINVAL;
757 goto fail;
758 }
759 if (decrease) {
760 refcount -= addend;
761 } else {
762 refcount += addend;
763 }
764 if (refcount == 0 && cluster_index < s->free_cluster_index) {
765 s->free_cluster_index = cluster_index;
766 }
767 s->set_refcount(refcount_block, block_index, refcount);
768
769 if (refcount == 0 && s->discard_passthrough[type]) {
770 update_refcount_discard(bs, cluster_offset, s->cluster_size);
771 }
772 }
773
774 ret = 0;
775 fail:
776 if (!s->cache_discards) {
777 qcow2_process_discards(bs, ret);
778 }
779
780 /* Write last changed block to disk */
781 if (refcount_block) {
782 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
783 }
784
785 /*
786 * Try do undo any updates if an error is returned (This may succeed in
787 * some cases like ENOSPC for allocating a new refcount block)
788 */
789 if (ret < 0) {
790 int dummy;
791 dummy = update_refcount(bs, offset, cluster_offset - offset, addend,
792 !decrease, QCOW2_DISCARD_NEVER);
793 (void)dummy;
794 }
795
796 return ret;
797 }
798
799 /*
800 * Increases or decreases the refcount of a given cluster.
801 *
802 * @addend is the absolute value of the addend; if @decrease is set, @addend
803 * will be subtracted from the current refcount, otherwise it will be added.
804 *
805 * On success 0 is returned; on failure -errno is returned.
806 */
807 int qcow2_update_cluster_refcount(BlockDriverState *bs,
808 int64_t cluster_index,
809 uint64_t addend, bool decrease,
810 enum qcow2_discard_type type)
811 {
812 BDRVQcow2State *s = bs->opaque;
813 int ret;
814
815 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend,
816 decrease, type);
817 if (ret < 0) {
818 return ret;
819 }
820
821 return 0;
822 }
823
824
825
826 /*********************************************************/
827 /* cluster allocation functions */
828
829
830
831 /* return < 0 if error */
832 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size)
833 {
834 BDRVQcow2State *s = bs->opaque;
835 uint64_t i, nb_clusters, refcount;
836 int ret;
837
838 /* We can't allocate clusters if they may still be queued for discard. */
839 if (s->cache_discards) {
840 qcow2_process_discards(bs, 0);
841 }
842
843 nb_clusters = size_to_clusters(s, size);
844 retry:
845 for(i = 0; i < nb_clusters; i++) {
846 uint64_t next_cluster_index = s->free_cluster_index++;
847 ret = qcow2_get_refcount(bs, next_cluster_index, &refcount);
848
849 if (ret < 0) {
850 return ret;
851 } else if (refcount != 0) {
852 goto retry;
853 }
854 }
855
856 /* Make sure that all offsets in the "allocated" range are representable
857 * in an int64_t */
858 if (s->free_cluster_index > 0 &&
859 s->free_cluster_index - 1 > (INT64_MAX >> s->cluster_bits))
860 {
861 return -EFBIG;
862 }
863
864 #ifdef DEBUG_ALLOC2
865 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
866 size,
867 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
868 #endif
869 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
870 }
871
872 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size)
873 {
874 int64_t offset;
875 int ret;
876
877 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
878 do {
879 offset = alloc_clusters_noref(bs, size);
880 if (offset < 0) {
881 return offset;
882 }
883
884 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
885 } while (ret == -EAGAIN);
886
887 if (ret < 0) {
888 return ret;
889 }
890
891 return offset;
892 }
893
894 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
895 int64_t nb_clusters)
896 {
897 BDRVQcow2State *s = bs->opaque;
898 uint64_t cluster_index, refcount;
899 uint64_t i;
900 int ret;
901
902 assert(nb_clusters >= 0);
903 if (nb_clusters == 0) {
904 return 0;
905 }
906
907 do {
908 /* Check how many clusters there are free */
909 cluster_index = offset >> s->cluster_bits;
910 for(i = 0; i < nb_clusters; i++) {
911 ret = qcow2_get_refcount(bs, cluster_index++, &refcount);
912 if (ret < 0) {
913 return ret;
914 } else if (refcount != 0) {
915 break;
916 }
917 }
918
919 /* And then allocate them */
920 ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false,
921 QCOW2_DISCARD_NEVER);
922 } while (ret == -EAGAIN);
923
924 if (ret < 0) {
925 return ret;
926 }
927
928 return i;
929 }
930
931 /* only used to allocate compressed sectors. We try to allocate
932 contiguous sectors. size must be <= cluster_size */
933 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
934 {
935 BDRVQcow2State *s = bs->opaque;
936 int64_t offset;
937 size_t free_in_cluster;
938 int ret;
939
940 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
941 assert(size > 0 && size <= s->cluster_size);
942 assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset));
943
944 offset = s->free_byte_offset;
945
946 if (offset) {
947 uint64_t refcount;
948 ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount);
949 if (ret < 0) {
950 return ret;
951 }
952
953 if (refcount == s->refcount_max) {
954 offset = 0;
955 }
956 }
957
958 free_in_cluster = s->cluster_size - offset_into_cluster(s, offset);
959 do {
960 if (!offset || free_in_cluster < size) {
961 int64_t new_cluster = alloc_clusters_noref(bs, s->cluster_size);
962 if (new_cluster < 0) {
963 return new_cluster;
964 }
965
966 if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) {
967 offset = new_cluster;
968 free_in_cluster = s->cluster_size;
969 } else {
970 free_in_cluster += s->cluster_size;
971 }
972 }
973
974 assert(offset);
975 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
976 if (ret < 0) {
977 offset = 0;
978 }
979 } while (ret == -EAGAIN);
980 if (ret < 0) {
981 return ret;
982 }
983
984 /* The cluster refcount was incremented; refcount blocks must be flushed
985 * before the caller's L2 table updates. */
986 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
987
988 s->free_byte_offset = offset + size;
989 if (!offset_into_cluster(s, s->free_byte_offset)) {
990 s->free_byte_offset = 0;
991 }
992
993 return offset;
994 }
995
996 void qcow2_free_clusters(BlockDriverState *bs,
997 int64_t offset, int64_t size,
998 enum qcow2_discard_type type)
999 {
1000 int ret;
1001
1002 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
1003 ret = update_refcount(bs, offset, size, 1, true, type);
1004 if (ret < 0) {
1005 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
1006 /* TODO Remember the clusters to free them later and avoid leaking */
1007 }
1008 }
1009
1010 /*
1011 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
1012 * normal cluster, compressed cluster, etc.)
1013 */
1014 void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
1015 int nb_clusters, enum qcow2_discard_type type)
1016 {
1017 BDRVQcow2State *s = bs->opaque;
1018
1019 switch (qcow2_get_cluster_type(l2_entry)) {
1020 case QCOW2_CLUSTER_COMPRESSED:
1021 {
1022 int nb_csectors;
1023 nb_csectors = ((l2_entry >> s->csize_shift) &
1024 s->csize_mask) + 1;
1025 qcow2_free_clusters(bs,
1026 (l2_entry & s->cluster_offset_mask) & ~511,
1027 nb_csectors * 512, type);
1028 }
1029 break;
1030 case QCOW2_CLUSTER_NORMAL:
1031 case QCOW2_CLUSTER_ZERO_ALLOC:
1032 if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) {
1033 qcow2_signal_corruption(bs, false, -1, -1,
1034 "Cannot free unaligned cluster %#llx",
1035 l2_entry & L2E_OFFSET_MASK);
1036 } else {
1037 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
1038 nb_clusters << s->cluster_bits, type);
1039 }
1040 break;
1041 case QCOW2_CLUSTER_ZERO_PLAIN:
1042 case QCOW2_CLUSTER_UNALLOCATED:
1043 break;
1044 default:
1045 abort();
1046 }
1047 }
1048
1049
1050
1051 /*********************************************************/
1052 /* snapshots and image creation */
1053
1054
1055
1056 /* update the refcounts of snapshots and the copied flag */
1057 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
1058 int64_t l1_table_offset, int l1_size, int addend)
1059 {
1060 BDRVQcow2State *s = bs->opaque;
1061 uint64_t *l1_table, *l2_table, l2_offset, entry, l1_size2, refcount;
1062 bool l1_allocated = false;
1063 int64_t old_entry, old_l2_offset;
1064 int i, j, l1_modified = 0, nb_csectors;
1065 int ret;
1066
1067 assert(addend >= -1 && addend <= 1);
1068
1069 l2_table = NULL;
1070 l1_table = NULL;
1071 l1_size2 = l1_size * sizeof(uint64_t);
1072
1073 s->cache_discards = true;
1074
1075 /* WARNING: qcow2_snapshot_goto relies on this function not using the
1076 * l1_table_offset when it is the current s->l1_table_offset! Be careful
1077 * when changing this! */
1078 if (l1_table_offset != s->l1_table_offset) {
1079 l1_table = g_try_malloc0(align_offset(l1_size2, 512));
1080 if (l1_size2 && l1_table == NULL) {
1081 ret = -ENOMEM;
1082 goto fail;
1083 }
1084 l1_allocated = true;
1085
1086 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
1087 if (ret < 0) {
1088 goto fail;
1089 }
1090
1091 for (i = 0; i < l1_size; i++) {
1092 be64_to_cpus(&l1_table[i]);
1093 }
1094 } else {
1095 assert(l1_size == s->l1_size);
1096 l1_table = s->l1_table;
1097 l1_allocated = false;
1098 }
1099
1100 for (i = 0; i < l1_size; i++) {
1101 l2_offset = l1_table[i];
1102 if (l2_offset) {
1103 old_l2_offset = l2_offset;
1104 l2_offset &= L1E_OFFSET_MASK;
1105
1106 if (offset_into_cluster(s, l2_offset)) {
1107 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1108 PRIx64 " unaligned (L1 index: %#x)",
1109 l2_offset, i);
1110 ret = -EIO;
1111 goto fail;
1112 }
1113
1114 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1115 (void**) &l2_table);
1116 if (ret < 0) {
1117 goto fail;
1118 }
1119
1120 for (j = 0; j < s->l2_size; j++) {
1121 uint64_t cluster_index;
1122 uint64_t offset;
1123
1124 entry = be64_to_cpu(l2_table[j]);
1125 old_entry = entry;
1126 entry &= ~QCOW_OFLAG_COPIED;
1127 offset = entry & L2E_OFFSET_MASK;
1128
1129 switch (qcow2_get_cluster_type(entry)) {
1130 case QCOW2_CLUSTER_COMPRESSED:
1131 nb_csectors = ((entry >> s->csize_shift) &
1132 s->csize_mask) + 1;
1133 if (addend != 0) {
1134 ret = update_refcount(bs,
1135 (entry & s->cluster_offset_mask) & ~511,
1136 nb_csectors * 512, abs(addend), addend < 0,
1137 QCOW2_DISCARD_SNAPSHOT);
1138 if (ret < 0) {
1139 goto fail;
1140 }
1141 }
1142 /* compressed clusters are never modified */
1143 refcount = 2;
1144 break;
1145
1146 case QCOW2_CLUSTER_NORMAL:
1147 case QCOW2_CLUSTER_ZERO_ALLOC:
1148 if (offset_into_cluster(s, offset)) {
1149 qcow2_signal_corruption(bs, true, -1, -1, "Cluster "
1150 "allocation offset %#" PRIx64
1151 " unaligned (L2 offset: %#"
1152 PRIx64 ", L2 index: %#x)",
1153 offset, l2_offset, j);
1154 ret = -EIO;
1155 goto fail;
1156 }
1157
1158 cluster_index = offset >> s->cluster_bits;
1159 assert(cluster_index);
1160 if (addend != 0) {
1161 ret = qcow2_update_cluster_refcount(bs,
1162 cluster_index, abs(addend), addend < 0,
1163 QCOW2_DISCARD_SNAPSHOT);
1164 if (ret < 0) {
1165 goto fail;
1166 }
1167 }
1168
1169 ret = qcow2_get_refcount(bs, cluster_index, &refcount);
1170 if (ret < 0) {
1171 goto fail;
1172 }
1173 break;
1174
1175 case QCOW2_CLUSTER_ZERO_PLAIN:
1176 case QCOW2_CLUSTER_UNALLOCATED:
1177 refcount = 0;
1178 break;
1179
1180 default:
1181 abort();
1182 }
1183
1184 if (refcount == 1) {
1185 entry |= QCOW_OFLAG_COPIED;
1186 }
1187 if (entry != old_entry) {
1188 if (addend > 0) {
1189 qcow2_cache_set_dependency(bs, s->l2_table_cache,
1190 s->refcount_block_cache);
1191 }
1192 l2_table[j] = cpu_to_be64(entry);
1193 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache,
1194 l2_table);
1195 }
1196 }
1197
1198 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1199
1200 if (addend != 0) {
1201 ret = qcow2_update_cluster_refcount(bs, l2_offset >>
1202 s->cluster_bits,
1203 abs(addend), addend < 0,
1204 QCOW2_DISCARD_SNAPSHOT);
1205 if (ret < 0) {
1206 goto fail;
1207 }
1208 }
1209 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1210 &refcount);
1211 if (ret < 0) {
1212 goto fail;
1213 } else if (refcount == 1) {
1214 l2_offset |= QCOW_OFLAG_COPIED;
1215 }
1216 if (l2_offset != old_l2_offset) {
1217 l1_table[i] = l2_offset;
1218 l1_modified = 1;
1219 }
1220 }
1221 }
1222
1223 ret = bdrv_flush(bs);
1224 fail:
1225 if (l2_table) {
1226 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1227 }
1228
1229 s->cache_discards = false;
1230 qcow2_process_discards(bs, ret);
1231
1232 /* Update L1 only if it isn't deleted anyway (addend = -1) */
1233 if (ret == 0 && addend >= 0 && l1_modified) {
1234 for (i = 0; i < l1_size; i++) {
1235 cpu_to_be64s(&l1_table[i]);
1236 }
1237
1238 ret = bdrv_pwrite_sync(bs->file, l1_table_offset,
1239 l1_table, l1_size2);
1240
1241 for (i = 0; i < l1_size; i++) {
1242 be64_to_cpus(&l1_table[i]);
1243 }
1244 }
1245 if (l1_allocated)
1246 g_free(l1_table);
1247 return ret;
1248 }
1249
1250
1251
1252
1253 /*********************************************************/
1254 /* refcount checking functions */
1255
1256
1257 static uint64_t refcount_array_byte_size(BDRVQcow2State *s, uint64_t entries)
1258 {
1259 /* This assertion holds because there is no way we can address more than
1260 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because
1261 * offsets have to be representable in bytes); due to every cluster
1262 * corresponding to one refcount entry, we are well below that limit */
1263 assert(entries < (UINT64_C(1) << (64 - 9)));
1264
1265 /* Thanks to the assertion this will not overflow, because
1266 * s->refcount_order < 7.
1267 * (note: x << s->refcount_order == x * s->refcount_bits) */
1268 return DIV_ROUND_UP(entries << s->refcount_order, 8);
1269 }
1270
1271 /**
1272 * Reallocates *array so that it can hold new_size entries. *size must contain
1273 * the current number of entries in *array. If the reallocation fails, *array
1274 * and *size will not be modified and -errno will be returned. If the
1275 * reallocation is successful, *array will be set to the new buffer, *size
1276 * will be set to new_size and 0 will be returned. The size of the reallocated
1277 * refcount array buffer will be aligned to a cluster boundary, and the newly
1278 * allocated area will be zeroed.
1279 */
1280 static int realloc_refcount_array(BDRVQcow2State *s, void **array,
1281 int64_t *size, int64_t new_size)
1282 {
1283 int64_t old_byte_size, new_byte_size;
1284 void *new_ptr;
1285
1286 /* Round to clusters so the array can be directly written to disk */
1287 old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size))
1288 * s->cluster_size;
1289 new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size))
1290 * s->cluster_size;
1291
1292 if (new_byte_size == old_byte_size) {
1293 *size = new_size;
1294 return 0;
1295 }
1296
1297 assert(new_byte_size > 0);
1298
1299 if (new_byte_size > SIZE_MAX) {
1300 return -ENOMEM;
1301 }
1302
1303 new_ptr = g_try_realloc(*array, new_byte_size);
1304 if (!new_ptr) {
1305 return -ENOMEM;
1306 }
1307
1308 if (new_byte_size > old_byte_size) {
1309 memset((char *)new_ptr + old_byte_size, 0,
1310 new_byte_size - old_byte_size);
1311 }
1312
1313 *array = new_ptr;
1314 *size = new_size;
1315
1316 return 0;
1317 }
1318
1319 /*
1320 * Increases the refcount for a range of clusters in a given refcount table.
1321 * This is used to construct a temporary refcount table out of L1 and L2 tables
1322 * which can be compared to the refcount table saved in the image.
1323 *
1324 * Modifies the number of errors in res.
1325 */
1326 int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res,
1327 void **refcount_table,
1328 int64_t *refcount_table_size,
1329 int64_t offset, int64_t size)
1330 {
1331 BDRVQcow2State *s = bs->opaque;
1332 uint64_t start, last, cluster_offset, k, refcount;
1333 int ret;
1334
1335 if (size <= 0) {
1336 return 0;
1337 }
1338
1339 start = start_of_cluster(s, offset);
1340 last = start_of_cluster(s, offset + size - 1);
1341 for(cluster_offset = start; cluster_offset <= last;
1342 cluster_offset += s->cluster_size) {
1343 k = cluster_offset >> s->cluster_bits;
1344 if (k >= *refcount_table_size) {
1345 ret = realloc_refcount_array(s, refcount_table,
1346 refcount_table_size, k + 1);
1347 if (ret < 0) {
1348 res->check_errors++;
1349 return ret;
1350 }
1351 }
1352
1353 refcount = s->get_refcount(*refcount_table, k);
1354 if (refcount == s->refcount_max) {
1355 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
1356 "\n", cluster_offset);
1357 fprintf(stderr, "Use qemu-img amend to increase the refcount entry "
1358 "width or qemu-img convert to create a clean copy if the "
1359 "image cannot be opened for writing\n");
1360 res->corruptions++;
1361 continue;
1362 }
1363 s->set_refcount(*refcount_table, k, refcount + 1);
1364 }
1365
1366 return 0;
1367 }
1368
1369 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
1370 enum {
1371 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
1372 };
1373
1374 /*
1375 * Increases the refcount in the given refcount table for the all clusters
1376 * referenced in the L2 table. While doing so, performs some checks on L2
1377 * entries.
1378 *
1379 * Returns the number of errors found by the checks or -errno if an internal
1380 * error occurred.
1381 */
1382 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
1383 void **refcount_table,
1384 int64_t *refcount_table_size, int64_t l2_offset,
1385 int flags)
1386 {
1387 BDRVQcow2State *s = bs->opaque;
1388 uint64_t *l2_table, l2_entry;
1389 uint64_t next_contiguous_offset = 0;
1390 int i, l2_size, nb_csectors, ret;
1391
1392 /* Read L2 table from disk */
1393 l2_size = s->l2_size * sizeof(uint64_t);
1394 l2_table = g_malloc(l2_size);
1395
1396 ret = bdrv_pread(bs->file, l2_offset, l2_table, l2_size);
1397 if (ret < 0) {
1398 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1399 res->check_errors++;
1400 goto fail;
1401 }
1402
1403 /* Do the actual checks */
1404 for(i = 0; i < s->l2_size; i++) {
1405 l2_entry = be64_to_cpu(l2_table[i]);
1406
1407 switch (qcow2_get_cluster_type(l2_entry)) {
1408 case QCOW2_CLUSTER_COMPRESSED:
1409 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1410 if (l2_entry & QCOW_OFLAG_COPIED) {
1411 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
1412 "copied flag must never be set for compressed "
1413 "clusters\n", l2_entry >> s->cluster_bits);
1414 l2_entry &= ~QCOW_OFLAG_COPIED;
1415 res->corruptions++;
1416 }
1417
1418 /* Mark cluster as used */
1419 nb_csectors = ((l2_entry >> s->csize_shift) &
1420 s->csize_mask) + 1;
1421 l2_entry &= s->cluster_offset_mask;
1422 ret = qcow2_inc_refcounts_imrt(bs, res,
1423 refcount_table, refcount_table_size,
1424 l2_entry & ~511, nb_csectors * 512);
1425 if (ret < 0) {
1426 goto fail;
1427 }
1428
1429 if (flags & CHECK_FRAG_INFO) {
1430 res->bfi.allocated_clusters++;
1431 res->bfi.compressed_clusters++;
1432
1433 /* Compressed clusters are fragmented by nature. Since they
1434 * take up sub-sector space but we only have sector granularity
1435 * I/O we need to re-read the same sectors even for adjacent
1436 * compressed clusters.
1437 */
1438 res->bfi.fragmented_clusters++;
1439 }
1440 break;
1441
1442 case QCOW2_CLUSTER_ZERO_ALLOC:
1443 case QCOW2_CLUSTER_NORMAL:
1444 {
1445 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
1446
1447 if (flags & CHECK_FRAG_INFO) {
1448 res->bfi.allocated_clusters++;
1449 if (next_contiguous_offset &&
1450 offset != next_contiguous_offset) {
1451 res->bfi.fragmented_clusters++;
1452 }
1453 next_contiguous_offset = offset + s->cluster_size;
1454 }
1455
1456 /* Mark cluster as used */
1457 ret = qcow2_inc_refcounts_imrt(bs, res,
1458 refcount_table, refcount_table_size,
1459 offset, s->cluster_size);
1460 if (ret < 0) {
1461 goto fail;
1462 }
1463
1464 /* Correct offsets are cluster aligned */
1465 if (offset_into_cluster(s, offset)) {
1466 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
1467 "properly aligned; L2 entry corrupted.\n", offset);
1468 res->corruptions++;
1469 }
1470 break;
1471 }
1472
1473 case QCOW2_CLUSTER_ZERO_PLAIN:
1474 case QCOW2_CLUSTER_UNALLOCATED:
1475 break;
1476
1477 default:
1478 abort();
1479 }
1480 }
1481
1482 g_free(l2_table);
1483 return 0;
1484
1485 fail:
1486 g_free(l2_table);
1487 return ret;
1488 }
1489
1490 /*
1491 * Increases the refcount for the L1 table, its L2 tables and all referenced
1492 * clusters in the given refcount table. While doing so, performs some checks
1493 * on L1 and L2 entries.
1494 *
1495 * Returns the number of errors found by the checks or -errno if an internal
1496 * error occurred.
1497 */
1498 static int check_refcounts_l1(BlockDriverState *bs,
1499 BdrvCheckResult *res,
1500 void **refcount_table,
1501 int64_t *refcount_table_size,
1502 int64_t l1_table_offset, int l1_size,
1503 int flags)
1504 {
1505 BDRVQcow2State *s = bs->opaque;
1506 uint64_t *l1_table = NULL, l2_offset, l1_size2;
1507 int i, ret;
1508
1509 l1_size2 = l1_size * sizeof(uint64_t);
1510
1511 /* Mark L1 table as used */
1512 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, refcount_table_size,
1513 l1_table_offset, l1_size2);
1514 if (ret < 0) {
1515 goto fail;
1516 }
1517
1518 /* Read L1 table entries from disk */
1519 if (l1_size2 > 0) {
1520 l1_table = g_try_malloc(l1_size2);
1521 if (l1_table == NULL) {
1522 ret = -ENOMEM;
1523 res->check_errors++;
1524 goto fail;
1525 }
1526 ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2);
1527 if (ret < 0) {
1528 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1529 res->check_errors++;
1530 goto fail;
1531 }
1532 for(i = 0;i < l1_size; i++)
1533 be64_to_cpus(&l1_table[i]);
1534 }
1535
1536 /* Do the actual checks */
1537 for(i = 0; i < l1_size; i++) {
1538 l2_offset = l1_table[i];
1539 if (l2_offset) {
1540 /* Mark L2 table as used */
1541 l2_offset &= L1E_OFFSET_MASK;
1542 ret = qcow2_inc_refcounts_imrt(bs, res,
1543 refcount_table, refcount_table_size,
1544 l2_offset, s->cluster_size);
1545 if (ret < 0) {
1546 goto fail;
1547 }
1548
1549 /* L2 tables are cluster aligned */
1550 if (offset_into_cluster(s, l2_offset)) {
1551 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1552 "cluster aligned; L1 entry corrupted\n", l2_offset);
1553 res->corruptions++;
1554 }
1555
1556 /* Process and check L2 entries */
1557 ret = check_refcounts_l2(bs, res, refcount_table,
1558 refcount_table_size, l2_offset, flags);
1559 if (ret < 0) {
1560 goto fail;
1561 }
1562 }
1563 }
1564 g_free(l1_table);
1565 return 0;
1566
1567 fail:
1568 g_free(l1_table);
1569 return ret;
1570 }
1571
1572 /*
1573 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1574 *
1575 * This function does not print an error message nor does it increment
1576 * check_errors if qcow2_get_refcount fails (this is because such an error will
1577 * have been already detected and sufficiently signaled by the calling function
1578 * (qcow2_check_refcounts) by the time this function is called).
1579 */
1580 static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
1581 BdrvCheckMode fix)
1582 {
1583 BDRVQcow2State *s = bs->opaque;
1584 uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size);
1585 int ret;
1586 uint64_t refcount;
1587 int i, j;
1588
1589 for (i = 0; i < s->l1_size; i++) {
1590 uint64_t l1_entry = s->l1_table[i];
1591 uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK;
1592 bool l2_dirty = false;
1593
1594 if (!l2_offset) {
1595 continue;
1596 }
1597
1598 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1599 &refcount);
1600 if (ret < 0) {
1601 /* don't print message nor increment check_errors */
1602 continue;
1603 }
1604 if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) {
1605 fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1606 "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
1607 fix & BDRV_FIX_ERRORS ? "Repairing" :
1608 "ERROR",
1609 i, l1_entry, refcount);
1610 if (fix & BDRV_FIX_ERRORS) {
1611 s->l1_table[i] = refcount == 1
1612 ? l1_entry | QCOW_OFLAG_COPIED
1613 : l1_entry & ~QCOW_OFLAG_COPIED;
1614 ret = qcow2_write_l1_entry(bs, i);
1615 if (ret < 0) {
1616 res->check_errors++;
1617 goto fail;
1618 }
1619 res->corruptions_fixed++;
1620 } else {
1621 res->corruptions++;
1622 }
1623 }
1624
1625 ret = bdrv_pread(bs->file, l2_offset, l2_table,
1626 s->l2_size * sizeof(uint64_t));
1627 if (ret < 0) {
1628 fprintf(stderr, "ERROR: Could not read L2 table: %s\n",
1629 strerror(-ret));
1630 res->check_errors++;
1631 goto fail;
1632 }
1633
1634 for (j = 0; j < s->l2_size; j++) {
1635 uint64_t l2_entry = be64_to_cpu(l2_table[j]);
1636 uint64_t data_offset = l2_entry & L2E_OFFSET_MASK;
1637 QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry);
1638
1639 if (cluster_type == QCOW2_CLUSTER_NORMAL ||
1640 cluster_type == QCOW2_CLUSTER_ZERO_ALLOC) {
1641 ret = qcow2_get_refcount(bs,
1642 data_offset >> s->cluster_bits,
1643 &refcount);
1644 if (ret < 0) {
1645 /* don't print message nor increment check_errors */
1646 continue;
1647 }
1648 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
1649 fprintf(stderr, "%s OFLAG_COPIED data cluster: "
1650 "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
1651 fix & BDRV_FIX_ERRORS ? "Repairing" :
1652 "ERROR",
1653 l2_entry, refcount);
1654 if (fix & BDRV_FIX_ERRORS) {
1655 l2_table[j] = cpu_to_be64(refcount == 1
1656 ? l2_entry | QCOW_OFLAG_COPIED
1657 : l2_entry & ~QCOW_OFLAG_COPIED);
1658 l2_dirty = true;
1659 res->corruptions_fixed++;
1660 } else {
1661 res->corruptions++;
1662 }
1663 }
1664 }
1665 }
1666
1667 if (l2_dirty) {
1668 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
1669 l2_offset, s->cluster_size);
1670 if (ret < 0) {
1671 fprintf(stderr, "ERROR: Could not write L2 table; metadata "
1672 "overlap check failed: %s\n", strerror(-ret));
1673 res->check_errors++;
1674 goto fail;
1675 }
1676
1677 ret = bdrv_pwrite(bs->file, l2_offset, l2_table,
1678 s->cluster_size);
1679 if (ret < 0) {
1680 fprintf(stderr, "ERROR: Could not write L2 table: %s\n",
1681 strerror(-ret));
1682 res->check_errors++;
1683 goto fail;
1684 }
1685 }
1686 }
1687
1688 ret = 0;
1689
1690 fail:
1691 qemu_vfree(l2_table);
1692 return ret;
1693 }
1694
1695 /*
1696 * Checks consistency of refblocks and accounts for each refblock in
1697 * *refcount_table.
1698 */
1699 static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res,
1700 BdrvCheckMode fix, bool *rebuild,
1701 void **refcount_table, int64_t *nb_clusters)
1702 {
1703 BDRVQcow2State *s = bs->opaque;
1704 int64_t i, size;
1705 int ret;
1706
1707 for(i = 0; i < s->refcount_table_size; i++) {
1708 uint64_t offset, cluster;
1709 offset = s->refcount_table[i];
1710 cluster = offset >> s->cluster_bits;
1711
1712 /* Refcount blocks are cluster aligned */
1713 if (offset_into_cluster(s, offset)) {
1714 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1715 "cluster aligned; refcount table entry corrupted\n", i);
1716 res->corruptions++;
1717 *rebuild = true;
1718 continue;
1719 }
1720
1721 if (cluster >= *nb_clusters) {
1722 fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n",
1723 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
1724
1725 if (fix & BDRV_FIX_ERRORS) {
1726 int64_t new_nb_clusters;
1727 Error *local_err = NULL;
1728
1729 if (offset > INT64_MAX - s->cluster_size) {
1730 ret = -EINVAL;
1731 goto resize_fail;
1732 }
1733
1734 ret = bdrv_truncate(bs->file, offset + s->cluster_size,
1735 &local_err);
1736 if (ret < 0) {
1737 error_report_err(local_err);
1738 goto resize_fail;
1739 }
1740 size = bdrv_getlength(bs->file->bs);
1741 if (size < 0) {
1742 ret = size;
1743 goto resize_fail;
1744 }
1745
1746 new_nb_clusters = size_to_clusters(s, size);
1747 assert(new_nb_clusters >= *nb_clusters);
1748
1749 ret = realloc_refcount_array(s, refcount_table,
1750 nb_clusters, new_nb_clusters);
1751 if (ret < 0) {
1752 res->check_errors++;
1753 return ret;
1754 }
1755
1756 if (cluster >= *nb_clusters) {
1757 ret = -EINVAL;
1758 goto resize_fail;
1759 }
1760
1761 res->corruptions_fixed++;
1762 ret = qcow2_inc_refcounts_imrt(bs, res,
1763 refcount_table, nb_clusters,
1764 offset, s->cluster_size);
1765 if (ret < 0) {
1766 return ret;
1767 }
1768 /* No need to check whether the refcount is now greater than 1:
1769 * This area was just allocated and zeroed, so it can only be
1770 * exactly 1 after qcow2_inc_refcounts_imrt() */
1771 continue;
1772
1773 resize_fail:
1774 res->corruptions++;
1775 *rebuild = true;
1776 fprintf(stderr, "ERROR could not resize image: %s\n",
1777 strerror(-ret));
1778 } else {
1779 res->corruptions++;
1780 }
1781 continue;
1782 }
1783
1784 if (offset != 0) {
1785 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
1786 offset, s->cluster_size);
1787 if (ret < 0) {
1788 return ret;
1789 }
1790 if (s->get_refcount(*refcount_table, cluster) != 1) {
1791 fprintf(stderr, "ERROR refcount block %" PRId64
1792 " refcount=%" PRIu64 "\n", i,
1793 s->get_refcount(*refcount_table, cluster));
1794 res->corruptions++;
1795 *rebuild = true;
1796 }
1797 }
1798 }
1799
1800 return 0;
1801 }
1802
1803 /*
1804 * Calculates an in-memory refcount table.
1805 */
1806 static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1807 BdrvCheckMode fix, bool *rebuild,
1808 void **refcount_table, int64_t *nb_clusters)
1809 {
1810 BDRVQcow2State *s = bs->opaque;
1811 int64_t i;
1812 QCowSnapshot *sn;
1813 int ret;
1814
1815 if (!*refcount_table) {
1816 int64_t old_size = 0;
1817 ret = realloc_refcount_array(s, refcount_table,
1818 &old_size, *nb_clusters);
1819 if (ret < 0) {
1820 res->check_errors++;
1821 return ret;
1822 }
1823 }
1824
1825 /* header */
1826 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
1827 0, s->cluster_size);
1828 if (ret < 0) {
1829 return ret;
1830 }
1831
1832 /* current L1 table */
1833 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1834 s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO);
1835 if (ret < 0) {
1836 return ret;
1837 }
1838
1839 /* snapshots */
1840 for (i = 0; i < s->nb_snapshots; i++) {
1841 sn = s->snapshots + i;
1842 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1843 sn->l1_table_offset, sn->l1_size, 0);
1844 if (ret < 0) {
1845 return ret;
1846 }
1847 }
1848 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
1849 s->snapshots_offset, s->snapshots_size);
1850 if (ret < 0) {
1851 return ret;
1852 }
1853
1854 /* refcount data */
1855 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
1856 s->refcount_table_offset,
1857 s->refcount_table_size * sizeof(uint64_t));
1858 if (ret < 0) {
1859 return ret;
1860 }
1861
1862 /* encryption */
1863 if (s->crypto_header.length) {
1864 ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters,
1865 s->crypto_header.offset,
1866 s->crypto_header.length);
1867 if (ret < 0) {
1868 return ret;
1869 }
1870 }
1871
1872 /* bitmaps */
1873 ret = qcow2_check_bitmaps_refcounts(bs, res, refcount_table, nb_clusters);
1874 if (ret < 0) {
1875 return ret;
1876 }
1877
1878 return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters);
1879 }
1880
1881 /*
1882 * Compares the actual reference count for each cluster in the image against the
1883 * refcount as reported by the refcount structures on-disk.
1884 */
1885 static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1886 BdrvCheckMode fix, bool *rebuild,
1887 int64_t *highest_cluster,
1888 void *refcount_table, int64_t nb_clusters)
1889 {
1890 BDRVQcow2State *s = bs->opaque;
1891 int64_t i;
1892 uint64_t refcount1, refcount2;
1893 int ret;
1894
1895 for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) {
1896 ret = qcow2_get_refcount(bs, i, &refcount1);
1897 if (ret < 0) {
1898 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
1899 i, strerror(-ret));
1900 res->check_errors++;
1901 continue;
1902 }
1903
1904 refcount2 = s->get_refcount(refcount_table, i);
1905
1906 if (refcount1 > 0 || refcount2 > 0) {
1907 *highest_cluster = i;
1908 }
1909
1910 if (refcount1 != refcount2) {
1911 /* Check if we're allowed to fix the mismatch */
1912 int *num_fixed = NULL;
1913 if (refcount1 == 0) {
1914 *rebuild = true;
1915 } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
1916 num_fixed = &res->leaks_fixed;
1917 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
1918 num_fixed = &res->corruptions_fixed;
1919 }
1920
1921 fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64
1922 " reference=%" PRIu64 "\n",
1923 num_fixed != NULL ? "Repairing" :
1924 refcount1 < refcount2 ? "ERROR" :
1925 "Leaked",
1926 i, refcount1, refcount2);
1927
1928 if (num_fixed) {
1929 ret = update_refcount(bs, i << s->cluster_bits, 1,
1930 refcount_diff(refcount1, refcount2),
1931 refcount1 > refcount2,
1932 QCOW2_DISCARD_ALWAYS);
1933 if (ret >= 0) {
1934 (*num_fixed)++;
1935 continue;
1936 }
1937 }
1938
1939 /* And if we couldn't, print an error */
1940 if (refcount1 < refcount2) {
1941 res->corruptions++;
1942 } else {
1943 res->leaks++;
1944 }
1945 }
1946 }
1947 }
1948
1949 /*
1950 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to
1951 * the on-disk refcount structures.
1952 *
1953 * On input, *first_free_cluster tells where to start looking, and need not
1954 * actually be a free cluster; the returned offset will not be before that
1955 * cluster. On output, *first_free_cluster points to the first gap found, even
1956 * if that gap was too small to be used as the returned offset.
1957 *
1958 * Note that *first_free_cluster is a cluster index whereas the return value is
1959 * an offset.
1960 */
1961 static int64_t alloc_clusters_imrt(BlockDriverState *bs,
1962 int cluster_count,
1963 void **refcount_table,
1964 int64_t *imrt_nb_clusters,
1965 int64_t *first_free_cluster)
1966 {
1967 BDRVQcow2State *s = bs->opaque;
1968 int64_t cluster = *first_free_cluster, i;
1969 bool first_gap = true;
1970 int contiguous_free_clusters;
1971 int ret;
1972
1973 /* Starting at *first_free_cluster, find a range of at least cluster_count
1974 * continuously free clusters */
1975 for (contiguous_free_clusters = 0;
1976 cluster < *imrt_nb_clusters &&
1977 contiguous_free_clusters < cluster_count;
1978 cluster++)
1979 {
1980 if (!s->get_refcount(*refcount_table, cluster)) {
1981 contiguous_free_clusters++;
1982 if (first_gap) {
1983 /* If this is the first free cluster found, update
1984 * *first_free_cluster accordingly */
1985 *first_free_cluster = cluster;
1986 first_gap = false;
1987 }
1988 } else if (contiguous_free_clusters) {
1989 contiguous_free_clusters = 0;
1990 }
1991 }
1992
1993 /* If contiguous_free_clusters is greater than zero, it contains the number
1994 * of continuously free clusters until the current cluster; the first free
1995 * cluster in the current "gap" is therefore
1996 * cluster - contiguous_free_clusters */
1997
1998 /* If no such range could be found, grow the in-memory refcount table
1999 * accordingly to append free clusters at the end of the image */
2000 if (contiguous_free_clusters < cluster_count) {
2001 /* contiguous_free_clusters clusters are already empty at the image end;
2002 * we need cluster_count clusters; therefore, we have to allocate
2003 * cluster_count - contiguous_free_clusters new clusters at the end of
2004 * the image (which is the current value of cluster; note that cluster
2005 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond
2006 * the image end) */
2007 ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters,
2008 cluster + cluster_count
2009 - contiguous_free_clusters);
2010 if (ret < 0) {
2011 return ret;
2012 }
2013 }
2014
2015 /* Go back to the first free cluster */
2016 cluster -= contiguous_free_clusters;
2017 for (i = 0; i < cluster_count; i++) {
2018 s->set_refcount(*refcount_table, cluster + i, 1);
2019 }
2020
2021 return cluster << s->cluster_bits;
2022 }
2023
2024 /*
2025 * Creates a new refcount structure based solely on the in-memory information
2026 * given through *refcount_table. All necessary allocations will be reflected
2027 * in that array.
2028 *
2029 * On success, the old refcount structure is leaked (it will be covered by the
2030 * new refcount structure).
2031 */
2032 static int rebuild_refcount_structure(BlockDriverState *bs,
2033 BdrvCheckResult *res,
2034 void **refcount_table,
2035 int64_t *nb_clusters)
2036 {
2037 BDRVQcow2State *s = bs->opaque;
2038 int64_t first_free_cluster = 0, reftable_offset = -1, cluster = 0;
2039 int64_t refblock_offset, refblock_start, refblock_index;
2040 uint32_t reftable_size = 0;
2041 uint64_t *on_disk_reftable = NULL;
2042 void *on_disk_refblock;
2043 int ret = 0;
2044 struct {
2045 uint64_t reftable_offset;
2046 uint32_t reftable_clusters;
2047 } QEMU_PACKED reftable_offset_and_clusters;
2048
2049 qcow2_cache_empty(bs, s->refcount_block_cache);
2050
2051 write_refblocks:
2052 for (; cluster < *nb_clusters; cluster++) {
2053 if (!s->get_refcount(*refcount_table, cluster)) {
2054 continue;
2055 }
2056
2057 refblock_index = cluster >> s->refcount_block_bits;
2058 refblock_start = refblock_index << s->refcount_block_bits;
2059
2060 /* Don't allocate a cluster in a refblock already written to disk */
2061 if (first_free_cluster < refblock_start) {
2062 first_free_cluster = refblock_start;
2063 }
2064 refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table,
2065 nb_clusters, &first_free_cluster);
2066 if (refblock_offset < 0) {
2067 fprintf(stderr, "ERROR allocating refblock: %s\n",
2068 strerror(-refblock_offset));
2069 res->check_errors++;
2070 ret = refblock_offset;
2071 goto fail;
2072 }
2073
2074 if (reftable_size <= refblock_index) {
2075 uint32_t old_reftable_size = reftable_size;
2076 uint64_t *new_on_disk_reftable;
2077
2078 reftable_size = ROUND_UP((refblock_index + 1) * sizeof(uint64_t),
2079 s->cluster_size) / sizeof(uint64_t);
2080 new_on_disk_reftable = g_try_realloc(on_disk_reftable,
2081 reftable_size *
2082 sizeof(uint64_t));
2083 if (!new_on_disk_reftable) {
2084 res->check_errors++;
2085 ret = -ENOMEM;
2086 goto fail;
2087 }
2088 on_disk_reftable = new_on_disk_reftable;
2089
2090 memset(on_disk_reftable + old_reftable_size, 0,
2091 (reftable_size - old_reftable_size) * sizeof(uint64_t));
2092
2093 /* The offset we have for the reftable is now no longer valid;
2094 * this will leak that range, but we can easily fix that by running
2095 * a leak-fixing check after this rebuild operation */
2096 reftable_offset = -1;
2097 }
2098 on_disk_reftable[refblock_index] = refblock_offset;
2099
2100 /* If this is apparently the last refblock (for now), try to squeeze the
2101 * reftable in */
2102 if (refblock_index == (*nb_clusters - 1) >> s->refcount_block_bits &&
2103 reftable_offset < 0)
2104 {
2105 uint64_t reftable_clusters = size_to_clusters(s, reftable_size *
2106 sizeof(uint64_t));
2107 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2108 refcount_table, nb_clusters,
2109 &first_free_cluster);
2110 if (reftable_offset < 0) {
2111 fprintf(stderr, "ERROR allocating reftable: %s\n",
2112 strerror(-reftable_offset));
2113 res->check_errors++;
2114 ret = reftable_offset;
2115 goto fail;
2116 }
2117 }
2118
2119 ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset,
2120 s->cluster_size);
2121 if (ret < 0) {
2122 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2123 goto fail;
2124 }
2125
2126 /* The size of *refcount_table is always cluster-aligned, therefore the
2127 * write operation will not overflow */
2128 on_disk_refblock = (void *)((char *) *refcount_table +
2129 refblock_index * s->cluster_size);
2130
2131 ret = bdrv_write(bs->file, refblock_offset / BDRV_SECTOR_SIZE,
2132 on_disk_refblock, s->cluster_sectors);
2133 if (ret < 0) {
2134 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2135 goto fail;
2136 }
2137
2138 /* Go to the end of this refblock */
2139 cluster = refblock_start + s->refcount_block_size - 1;
2140 }
2141
2142 if (reftable_offset < 0) {
2143 uint64_t post_refblock_start, reftable_clusters;
2144
2145 post_refblock_start = ROUND_UP(*nb_clusters, s->refcount_block_size);
2146 reftable_clusters = size_to_clusters(s,
2147 reftable_size * sizeof(uint64_t));
2148 /* Not pretty but simple */
2149 if (first_free_cluster < post_refblock_start) {
2150 first_free_cluster = post_refblock_start;
2151 }
2152 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2153 refcount_table, nb_clusters,
2154 &first_free_cluster);
2155 if (reftable_offset < 0) {
2156 fprintf(stderr, "ERROR allocating reftable: %s\n",
2157 strerror(-reftable_offset));
2158 res->check_errors++;
2159 ret = reftable_offset;
2160 goto fail;
2161 }
2162
2163 goto write_refblocks;
2164 }
2165
2166 assert(on_disk_reftable);
2167
2168 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2169 cpu_to_be64s(&on_disk_reftable[refblock_index]);
2170 }
2171
2172 ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset,
2173 reftable_size * sizeof(uint64_t));
2174 if (ret < 0) {
2175 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2176 goto fail;
2177 }
2178
2179 assert(reftable_size < INT_MAX / sizeof(uint64_t));
2180 ret = bdrv_pwrite(bs->file, reftable_offset, on_disk_reftable,
2181 reftable_size * sizeof(uint64_t));
2182 if (ret < 0) {
2183 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2184 goto fail;
2185 }
2186
2187 /* Enter new reftable into the image header */
2188 reftable_offset_and_clusters.reftable_offset = cpu_to_be64(reftable_offset);
2189 reftable_offset_and_clusters.reftable_clusters =
2190 cpu_to_be32(size_to_clusters(s, reftable_size * sizeof(uint64_t)));
2191 ret = bdrv_pwrite_sync(bs->file,
2192 offsetof(QCowHeader, refcount_table_offset),
2193 &reftable_offset_and_clusters,
2194 sizeof(reftable_offset_and_clusters));
2195 if (ret < 0) {
2196 fprintf(stderr, "ERROR setting reftable: %s\n", strerror(-ret));
2197 goto fail;
2198 }
2199
2200 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2201 be64_to_cpus(&on_disk_reftable[refblock_index]);
2202 }
2203 s->refcount_table = on_disk_reftable;
2204 s->refcount_table_offset = reftable_offset;
2205 s->refcount_table_size = reftable_size;
2206 update_max_refcount_table_index(s);
2207
2208 return 0;
2209
2210 fail:
2211 g_free(on_disk_reftable);
2212 return ret;
2213 }
2214
2215 /*
2216 * Checks an image for refcount consistency.
2217 *
2218 * Returns 0 if no errors are found, the number of errors in case the image is
2219 * detected as corrupted, and -errno when an internal error occurred.
2220 */
2221 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2222 BdrvCheckMode fix)
2223 {
2224 BDRVQcow2State *s = bs->opaque;
2225 BdrvCheckResult pre_compare_res;
2226 int64_t size, highest_cluster, nb_clusters;
2227 void *refcount_table = NULL;
2228 bool rebuild = false;
2229 int ret;
2230
2231 size = bdrv_getlength(bs->file->bs);
2232 if (size < 0) {
2233 res->check_errors++;
2234 return size;
2235 }
2236
2237 nb_clusters = size_to_clusters(s, size);
2238 if (nb_clusters > INT_MAX) {
2239 res->check_errors++;
2240 return -EFBIG;
2241 }
2242
2243 res->bfi.total_clusters =
2244 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE);
2245
2246 ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table,
2247 &nb_clusters);
2248 if (ret < 0) {
2249 goto fail;
2250 }
2251
2252 /* In case we don't need to rebuild the refcount structure (but want to fix
2253 * something), this function is immediately called again, in which case the
2254 * result should be ignored */
2255 pre_compare_res = *res;
2256 compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table,
2257 nb_clusters);
2258
2259 if (rebuild && (fix & BDRV_FIX_ERRORS)) {
2260 BdrvCheckResult old_res = *res;
2261 int fresh_leaks = 0;
2262
2263 fprintf(stderr, "Rebuilding refcount structure\n");
2264 ret = rebuild_refcount_structure(bs, res, &refcount_table,
2265 &nb_clusters);
2266 if (ret < 0) {
2267 goto fail;
2268 }
2269
2270 res->corruptions = 0;
2271 res->leaks = 0;
2272
2273 /* Because the old reftable has been exchanged for a new one the
2274 * references have to be recalculated */
2275 rebuild = false;
2276 memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters));
2277 ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table,
2278 &nb_clusters);
2279 if (ret < 0) {
2280 goto fail;
2281 }
2282
2283 if (fix & BDRV_FIX_LEAKS) {
2284 /* The old refcount structures are now leaked, fix it; the result
2285 * can be ignored, aside from leaks which were introduced by
2286 * rebuild_refcount_structure() that could not be fixed */
2287 BdrvCheckResult saved_res = *res;
2288 *res = (BdrvCheckResult){ 0 };
2289
2290 compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild,
2291 &highest_cluster, refcount_table, nb_clusters);
2292 if (rebuild) {
2293 fprintf(stderr, "ERROR rebuilt refcount structure is still "
2294 "broken\n");
2295 }
2296
2297 /* Any leaks accounted for here were introduced by
2298 * rebuild_refcount_structure() because that function has created a
2299 * new refcount structure from scratch */
2300 fresh_leaks = res->leaks;
2301 *res = saved_res;
2302 }
2303
2304 if (res->corruptions < old_res.corruptions) {
2305 res->corruptions_fixed += old_res.corruptions - res->corruptions;
2306 }
2307 if (res->leaks < old_res.leaks) {
2308 res->leaks_fixed += old_res.leaks - res->leaks;
2309 }
2310 res->leaks += fresh_leaks;
2311 } else if (fix) {
2312 if (rebuild) {
2313 fprintf(stderr, "ERROR need to rebuild refcount structures\n");
2314 res->check_errors++;
2315 ret = -EIO;
2316 goto fail;
2317 }
2318
2319 if (res->leaks || res->corruptions) {
2320 *res = pre_compare_res;
2321 compare_refcounts(bs, res, fix, &rebuild, &highest_cluster,
2322 refcount_table, nb_clusters);
2323 }
2324 }
2325
2326 /* check OFLAG_COPIED */
2327 ret = check_oflag_copied(bs, res, fix);
2328 if (ret < 0) {
2329 goto fail;
2330 }
2331
2332 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
2333 ret = 0;
2334
2335 fail:
2336 g_free(refcount_table);
2337
2338 return ret;
2339 }
2340
2341 #define overlaps_with(ofs, sz) \
2342 ranges_overlap(offset, size, ofs, sz)
2343
2344 /*
2345 * Checks if the given offset into the image file is actually free to use by
2346 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
2347 * i.e. a sanity check without relying on the refcount tables.
2348 *
2349 * The ign parameter specifies what checks not to perform (being a bitmask of
2350 * QCow2MetadataOverlap values), i.e., what sections to ignore.
2351 *
2352 * Returns:
2353 * - 0 if writing to this offset will not affect the mentioned metadata
2354 * - a positive QCow2MetadataOverlap value indicating one overlapping section
2355 * - a negative value (-errno) indicating an error while performing a check,
2356 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
2357 */
2358 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset,
2359 int64_t size)
2360 {
2361 BDRVQcow2State *s = bs->opaque;
2362 int chk = s->overlap_check & ~ign;
2363 int i, j;
2364
2365 if (!size) {
2366 return 0;
2367 }
2368
2369 if (chk & QCOW2_OL_MAIN_HEADER) {
2370 if (offset < s->cluster_size) {
2371 return QCOW2_OL_MAIN_HEADER;
2372 }
2373 }
2374
2375 /* align range to test to cluster boundaries */
2376 size = align_offset(offset_into_cluster(s, offset) + size, s->cluster_size);
2377 offset = start_of_cluster(s, offset);
2378
2379 if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) {
2380 if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) {
2381 return QCOW2_OL_ACTIVE_L1;
2382 }
2383 }
2384
2385 if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) {
2386 if (overlaps_with(s->refcount_table_offset,
2387 s->refcount_table_size * sizeof(uint64_t))) {
2388 return QCOW2_OL_REFCOUNT_TABLE;
2389 }
2390 }
2391
2392 if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) {
2393 if (overlaps_with(s->snapshots_offset, s->snapshots_size)) {
2394 return QCOW2_OL_SNAPSHOT_TABLE;
2395 }
2396 }
2397
2398 if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) {
2399 for (i = 0; i < s->nb_snapshots; i++) {
2400 if (s->snapshots[i].l1_size &&
2401 overlaps_with(s->snapshots[i].l1_table_offset,
2402 s->snapshots[i].l1_size * sizeof(uint64_t))) {
2403 return QCOW2_OL_INACTIVE_L1;
2404 }
2405 }
2406 }
2407
2408 if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) {
2409 for (i = 0; i < s->l1_size; i++) {
2410 if ((s->l1_table[i] & L1E_OFFSET_MASK) &&
2411 overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK,
2412 s->cluster_size)) {
2413 return QCOW2_OL_ACTIVE_L2;
2414 }
2415 }
2416 }
2417
2418 if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) {
2419 unsigned last_entry = s->max_refcount_table_index;
2420 assert(last_entry < s->refcount_table_size);
2421 assert(last_entry + 1 == s->refcount_table_size ||
2422 (s->refcount_table[last_entry + 1] & REFT_OFFSET_MASK) == 0);
2423 for (i = 0; i <= last_entry; i++) {
2424 if ((s->refcount_table[i] & REFT_OFFSET_MASK) &&
2425 overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK,
2426 s->cluster_size)) {
2427 return QCOW2_OL_REFCOUNT_BLOCK;
2428 }
2429 }
2430 }
2431
2432 if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) {
2433 for (i = 0; i < s->nb_snapshots; i++) {
2434 uint64_t l1_ofs = s->snapshots[i].l1_table_offset;
2435 uint32_t l1_sz = s->snapshots[i].l1_size;
2436 uint64_t l1_sz2 = l1_sz * sizeof(uint64_t);
2437 uint64_t *l1 = g_try_malloc(l1_sz2);
2438 int ret;
2439
2440 if (l1_sz2 && l1 == NULL) {
2441 return -ENOMEM;
2442 }
2443
2444 ret = bdrv_pread(bs->file, l1_ofs, l1, l1_sz2);
2445 if (ret < 0) {
2446 g_free(l1);
2447 return ret;
2448 }
2449
2450 for (j = 0; j < l1_sz; j++) {
2451 uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK;
2452 if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) {
2453 g_free(l1);
2454 return QCOW2_OL_INACTIVE_L2;
2455 }
2456 }
2457
2458 g_free(l1);
2459 }
2460 }
2461
2462 return 0;
2463 }
2464
2465 static const char *metadata_ol_names[] = {
2466 [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header",
2467 [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table",
2468 [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table",
2469 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table",
2470 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block",
2471 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table",
2472 [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table",
2473 [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table",
2474 };
2475
2476 /*
2477 * First performs a check for metadata overlaps (through
2478 * qcow2_check_metadata_overlap); if that fails with a negative value (error
2479 * while performing a check), that value is returned. If an impending overlap
2480 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
2481 * and -EIO returned.
2482 *
2483 * Returns 0 if there were neither overlaps nor errors while checking for
2484 * overlaps; or a negative value (-errno) on error.
2485 */
2486 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset,
2487 int64_t size)
2488 {
2489 int ret = qcow2_check_metadata_overlap(bs, ign, offset, size);
2490
2491 if (ret < 0) {
2492 return ret;
2493 } else if (ret > 0) {
2494 int metadata_ol_bitnr = ctz32(ret);
2495 assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR);
2496
2497 qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid "
2498 "write on metadata (overlaps with %s)",
2499 metadata_ol_names[metadata_ol_bitnr]);
2500 return -EIO;
2501 }
2502
2503 return 0;
2504 }
2505
2506 /* A pointer to a function of this type is given to walk_over_reftable(). That
2507 * function will create refblocks and pass them to a RefblockFinishOp once they
2508 * are completed (@refblock). @refblock_empty is set if the refblock is
2509 * completely empty.
2510 *
2511 * Along with the refblock, a corresponding reftable entry is passed, in the
2512 * reftable @reftable (which may be reallocated) at @reftable_index.
2513 *
2514 * @allocated should be set to true if a new cluster has been allocated.
2515 */
2516 typedef int (RefblockFinishOp)(BlockDriverState *bs, uint64_t **reftable,
2517 uint64_t reftable_index, uint64_t *reftable_size,
2518 void *refblock, bool refblock_empty,
2519 bool *allocated, Error **errp);
2520
2521 /**
2522 * This "operation" for walk_over_reftable() allocates the refblock on disk (if
2523 * it is not empty) and inserts its offset into the new reftable. The size of
2524 * this new reftable is increased as required.
2525 */
2526 static int alloc_refblock(BlockDriverState *bs, uint64_t **reftable,
2527 uint64_t reftable_index, uint64_t *reftable_size,
2528 void *refblock, bool refblock_empty, bool *allocated,
2529 Error **errp)
2530 {
2531 BDRVQcow2State *s = bs->opaque;
2532 int64_t offset;
2533
2534 if (!refblock_empty && reftable_index >= *reftable_size) {
2535 uint64_t *new_reftable;
2536 uint64_t new_reftable_size;
2537
2538 new_reftable_size = ROUND_UP(reftable_index + 1,
2539 s->cluster_size / sizeof(uint64_t));
2540 if (new_reftable_size > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) {
2541 error_setg(errp,
2542 "This operation would make the refcount table grow "
2543 "beyond the maximum size supported by QEMU, aborting");
2544 return -ENOTSUP;
2545 }
2546
2547 new_reftable = g_try_realloc(*reftable, new_reftable_size *
2548 sizeof(uint64_t));
2549 if (!new_reftable) {
2550 error_setg(errp, "Failed to increase reftable buffer size");
2551 return -ENOMEM;
2552 }
2553
2554 memset(new_reftable + *reftable_size, 0,
2555 (new_reftable_size - *reftable_size) * sizeof(uint64_t));
2556
2557 *reftable = new_reftable;
2558 *reftable_size = new_reftable_size;
2559 }
2560
2561 if (!refblock_empty && !(*reftable)[reftable_index]) {
2562 offset = qcow2_alloc_clusters(bs, s->cluster_size);
2563 if (offset < 0) {
2564 error_setg_errno(errp, -offset, "Failed to allocate refblock");
2565 return offset;
2566 }
2567 (*reftable)[reftable_index] = offset;
2568 *allocated = true;
2569 }
2570
2571 return 0;
2572 }
2573
2574 /**
2575 * This "operation" for walk_over_reftable() writes the refblock to disk at the
2576 * offset specified by the new reftable's entry. It does not modify the new
2577 * reftable or change any refcounts.
2578 */
2579 static int flush_refblock(BlockDriverState *bs, uint64_t **reftable,
2580 uint64_t reftable_index, uint64_t *reftable_size,
2581 void *refblock, bool refblock_empty, bool *allocated,
2582 Error **errp)
2583 {
2584 BDRVQcow2State *s = bs->opaque;
2585 int64_t offset;
2586 int ret;
2587
2588 if (reftable_index < *reftable_size && (*reftable)[reftable_index]) {
2589 offset = (*reftable)[reftable_index];
2590
2591 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
2592 if (ret < 0) {
2593 error_setg_errno(errp, -ret, "Overlap check failed");
2594 return ret;
2595 }
2596
2597 ret = bdrv_pwrite(bs->file, offset, refblock, s->cluster_size);
2598 if (ret < 0) {
2599 error_setg_errno(errp, -ret, "Failed to write refblock");
2600 return ret;
2601 }
2602 } else {
2603 assert(refblock_empty);
2604 }
2605
2606 return 0;
2607 }
2608
2609 /**
2610 * This function walks over the existing reftable and every referenced refblock;
2611 * if @new_set_refcount is non-NULL, it is called for every refcount entry to
2612 * create an equal new entry in the passed @new_refblock. Once that
2613 * @new_refblock is completely filled, @operation will be called.
2614 *
2615 * @status_cb and @cb_opaque are used for the amend operation's status callback.
2616 * @index is the index of the walk_over_reftable() calls and @total is the total
2617 * number of walk_over_reftable() calls per amend operation. Both are used for
2618 * calculating the parameters for the status callback.
2619 *
2620 * @allocated is set to true if a new cluster has been allocated.
2621 */
2622 static int walk_over_reftable(BlockDriverState *bs, uint64_t **new_reftable,
2623 uint64_t *new_reftable_index,
2624 uint64_t *new_reftable_size,
2625 void *new_refblock, int new_refblock_size,
2626 int new_refcount_bits,
2627 RefblockFinishOp *operation, bool *allocated,
2628 Qcow2SetRefcountFunc *new_set_refcount,
2629 BlockDriverAmendStatusCB *status_cb,
2630 void *cb_opaque, int index, int total,
2631 Error **errp)
2632 {
2633 BDRVQcow2State *s = bs->opaque;
2634 uint64_t reftable_index;
2635 bool new_refblock_empty = true;
2636 int refblock_index;
2637 int new_refblock_index = 0;
2638 int ret;
2639
2640 for (reftable_index = 0; reftable_index < s->refcount_table_size;
2641 reftable_index++)
2642 {
2643 uint64_t refblock_offset = s->refcount_table[reftable_index]
2644 & REFT_OFFSET_MASK;
2645
2646 status_cb(bs, (uint64_t)index * s->refcount_table_size + reftable_index,
2647 (uint64_t)total * s->refcount_table_size, cb_opaque);
2648
2649 if (refblock_offset) {
2650 void *refblock;
2651
2652 if (offset_into_cluster(s, refblock_offset)) {
2653 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
2654 PRIx64 " unaligned (reftable index: %#"
2655 PRIx64 ")", refblock_offset,
2656 reftable_index);
2657 error_setg(errp,
2658 "Image is corrupt (unaligned refblock offset)");
2659 return -EIO;
2660 }
2661
2662 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offset,
2663 &refblock);
2664 if (ret < 0) {
2665 error_setg_errno(errp, -ret, "Failed to retrieve refblock");
2666 return ret;
2667 }
2668
2669 for (refblock_index = 0; refblock_index < s->refcount_block_size;
2670 refblock_index++)
2671 {
2672 uint64_t refcount;
2673
2674 if (new_refblock_index >= new_refblock_size) {
2675 /* new_refblock is now complete */
2676 ret = operation(bs, new_reftable, *new_reftable_index,
2677 new_reftable_size, new_refblock,
2678 new_refblock_empty, allocated, errp);
2679 if (ret < 0) {
2680 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2681 return ret;
2682 }
2683
2684 (*new_reftable_index)++;
2685 new_refblock_index = 0;
2686 new_refblock_empty = true;
2687 }
2688
2689 refcount = s->get_refcount(refblock, refblock_index);
2690 if (new_refcount_bits < 64 && refcount >> new_refcount_bits) {
2691 uint64_t offset;
2692
2693 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2694
2695 offset = ((reftable_index << s->refcount_block_bits)
2696 + refblock_index) << s->cluster_bits;
2697
2698 error_setg(errp, "Cannot decrease refcount entry width to "
2699 "%i bits: Cluster at offset %#" PRIx64 " has a "
2700 "refcount of %" PRIu64, new_refcount_bits,
2701 offset, refcount);
2702 return -EINVAL;
2703 }
2704
2705 if (new_set_refcount) {
2706 new_set_refcount(new_refblock, new_refblock_index++,
2707 refcount);
2708 } else {
2709 new_refblock_index++;
2710 }
2711 new_refblock_empty = new_refblock_empty && refcount == 0;
2712 }
2713
2714 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2715 } else {
2716 /* No refblock means every refcount is 0 */
2717 for (refblock_index = 0; refblock_index < s->refcount_block_size;
2718 refblock_index++)
2719 {
2720 if (new_refblock_index >= new_refblock_size) {
2721 /* new_refblock is now complete */
2722 ret = operation(bs, new_reftable, *new_reftable_index,
2723 new_reftable_size, new_refblock,
2724 new_refblock_empty, allocated, errp);
2725 if (ret < 0) {
2726 return ret;
2727 }
2728
2729 (*new_reftable_index)++;
2730 new_refblock_index = 0;
2731 new_refblock_empty = true;
2732 }
2733
2734 if (new_set_refcount) {
2735 new_set_refcount(new_refblock, new_refblock_index++, 0);
2736 } else {
2737 new_refblock_index++;
2738 }
2739 }
2740 }
2741 }
2742
2743 if (new_refblock_index > 0) {
2744 /* Complete the potentially existing partially filled final refblock */
2745 if (new_set_refcount) {
2746 for (; new_refblock_index < new_refblock_size;
2747 new_refblock_index++)
2748 {
2749 new_set_refcount(new_refblock, new_refblock_index, 0);
2750 }
2751 }
2752
2753 ret = operation(bs, new_reftable, *new_reftable_index,
2754 new_reftable_size, new_refblock, new_refblock_empty,
2755 allocated, errp);
2756 if (ret < 0) {
2757 return ret;
2758 }
2759
2760 (*new_reftable_index)++;
2761 }
2762
2763 status_cb(bs, (uint64_t)(index + 1) * s->refcount_table_size,
2764 (uint64_t)total * s->refcount_table_size, cb_opaque);
2765
2766 return 0;
2767 }
2768
2769 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order,
2770 BlockDriverAmendStatusCB *status_cb,
2771 void *cb_opaque, Error **errp)
2772 {
2773 BDRVQcow2State *s = bs->opaque;
2774 Qcow2GetRefcountFunc *new_get_refcount;
2775 Qcow2SetRefcountFunc *new_set_refcount;
2776 void *new_refblock = qemu_blockalign(bs->file->bs, s->cluster_size);
2777 uint64_t *new_reftable = NULL, new_reftable_size = 0;
2778 uint64_t *old_reftable, old_reftable_size, old_reftable_offset;
2779 uint64_t new_reftable_index = 0;
2780 uint64_t i;
2781 int64_t new_reftable_offset = 0, allocated_reftable_size = 0;
2782 int new_refblock_size, new_refcount_bits = 1 << refcount_order;
2783 int old_refcount_order;
2784 int walk_index = 0;
2785 int ret;
2786 bool new_allocation;
2787
2788 assert(s->qcow_version >= 3);
2789 assert(refcount_order >= 0 && refcount_order <= 6);
2790
2791 /* see qcow2_open() */
2792 new_refblock_size = 1 << (s->cluster_bits - (refcount_order - 3));
2793
2794 new_get_refcount = get_refcount_funcs[refcount_order];
2795 new_set_refcount = set_refcount_funcs[refcount_order];
2796
2797
2798 do {
2799 int total_walks;
2800
2801 new_allocation = false;
2802
2803 /* At least we have to do this walk and the one which writes the
2804 * refblocks; also, at least we have to do this loop here at least
2805 * twice (normally), first to do the allocations, and second to
2806 * determine that everything is correctly allocated, this then makes
2807 * three walks in total */
2808 total_walks = MAX(walk_index + 2, 3);
2809
2810 /* First, allocate the structures so they are present in the refcount
2811 * structures */
2812 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
2813 &new_reftable_size, NULL, new_refblock_size,
2814 new_refcount_bits, &alloc_refblock,
2815 &new_allocation, NULL, status_cb, cb_opaque,
2816 walk_index++, total_walks, errp);
2817 if (ret < 0) {
2818 goto done;
2819 }
2820
2821 new_reftable_index = 0;
2822
2823 if (new_allocation) {
2824 if (new_reftable_offset) {
2825 qcow2_free_clusters(bs, new_reftable_offset,
2826 allocated_reftable_size * sizeof(uint64_t),
2827 QCOW2_DISCARD_NEVER);
2828 }
2829
2830 new_reftable_offset = qcow2_alloc_clusters(bs, new_reftable_size *
2831 sizeof(uint64_t));
2832 if (new_reftable_offset < 0) {
2833 error_setg_errno(errp, -new_reftable_offset,
2834 "Failed to allocate the new reftable");
2835 ret = new_reftable_offset;
2836 goto done;
2837 }
2838 allocated_reftable_size = new_reftable_size;
2839 }
2840 } while (new_allocation);
2841
2842 /* Second, write the new refblocks */
2843 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
2844 &new_reftable_size, new_refblock,
2845 new_refblock_size, new_refcount_bits,
2846 &flush_refblock, &new_allocation, new_set_refcount,
2847 status_cb, cb_opaque, walk_index, walk_index + 1,
2848 errp);
2849 if (ret < 0) {
2850 goto done;
2851 }
2852 assert(!new_allocation);
2853
2854
2855 /* Write the new reftable */
2856 ret = qcow2_pre_write_overlap_check(bs, 0, new_reftable_offset,
2857 new_reftable_size * sizeof(uint64_t));
2858 if (ret < 0) {
2859 error_setg_errno(errp, -ret, "Overlap check failed");
2860 goto done;
2861 }
2862
2863 for (i = 0; i < new_reftable_size; i++) {
2864 cpu_to_be64s(&new_reftable[i]);
2865 }
2866
2867 ret = bdrv_pwrite(bs->file, new_reftable_offset, new_reftable,
2868 new_reftable_size * sizeof(uint64_t));
2869
2870 for (i = 0; i < new_reftable_size; i++) {
2871 be64_to_cpus(&new_reftable[i]);
2872 }
2873
2874 if (ret < 0) {
2875 error_setg_errno(errp, -ret, "Failed to write the new reftable");
2876 goto done;
2877 }
2878
2879
2880 /* Empty the refcount cache */
2881 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2882 if (ret < 0) {
2883 error_setg_errno(errp, -ret, "Failed to flush the refblock cache");
2884 goto done;
2885 }
2886
2887 /* Update the image header to point to the new reftable; this only updates
2888 * the fields which are relevant to qcow2_update_header(); other fields
2889 * such as s->refcount_table or s->refcount_bits stay stale for now
2890 * (because we have to restore everything if qcow2_update_header() fails) */
2891 old_refcount_order = s->refcount_order;
2892 old_reftable_size = s->refcount_table_size;
2893 old_reftable_offset = s->refcount_table_offset;
2894
2895 s->refcount_order = refcount_order;
2896 s->refcount_table_size = new_reftable_size;
2897 s->refcount_table_offset = new_reftable_offset;
2898
2899 ret = qcow2_update_header(bs);
2900 if (ret < 0) {
2901 s->refcount_order = old_refcount_order;
2902 s->refcount_table_size = old_reftable_size;
2903 s->refcount_table_offset = old_reftable_offset;
2904 error_setg_errno(errp, -ret, "Failed to update the qcow2 header");
2905 goto done;
2906 }
2907
2908 /* Now update the rest of the in-memory information */
2909 old_reftable = s->refcount_table;
2910 s->refcount_table = new_reftable;
2911 update_max_refcount_table_index(s);
2912
2913 s->refcount_bits = 1 << refcount_order;
2914 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
2915 s->refcount_max += s->refcount_max - 1;
2916
2917 s->refcount_block_bits = s->cluster_bits - (refcount_order - 3);
2918 s->refcount_block_size = 1 << s->refcount_block_bits;
2919
2920 s->get_refcount = new_get_refcount;
2921 s->set_refcount = new_set_refcount;
2922
2923 /* For cleaning up all old refblocks and the old reftable below the "done"
2924 * label */
2925 new_reftable = old_reftable;
2926 new_reftable_size = old_reftable_size;
2927 new_reftable_offset = old_reftable_offset;
2928
2929 done:
2930 if (new_reftable) {
2931 /* On success, new_reftable actually points to the old reftable (and
2932 * new_reftable_size is the old reftable's size); but that is just
2933 * fine */
2934 for (i = 0; i < new_reftable_size; i++) {
2935 uint64_t offset = new_reftable[i] & REFT_OFFSET_MASK;
2936 if (offset) {
2937 qcow2_free_clusters(bs, offset, s->cluster_size,
2938 QCOW2_DISCARD_OTHER);
2939 }
2940 }
2941 g_free(new_reftable);
2942
2943 if (new_reftable_offset > 0) {
2944 qcow2_free_clusters(bs, new_reftable_offset,
2945 new_reftable_size * sizeof(uint64_t),
2946 QCOW2_DISCARD_OTHER);
2947 }
2948 }
2949
2950 qemu_vfree(new_refblock);
2951 return ret;
2952 }