]> git.proxmox.com Git - mirror_qemu.git/blob - block/qcow2-refcount.c
include/qemu/osdep.h: Don't include qapi/error.h
[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
32 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size);
33 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
34 int64_t offset, int64_t length, uint64_t addend,
35 bool decrease, enum qcow2_discard_type type);
36
37 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index);
38 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index);
39 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index);
40 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index);
41 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index);
42 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index);
43 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index);
44
45 static void set_refcount_ro0(void *refcount_array, uint64_t index,
46 uint64_t value);
47 static void set_refcount_ro1(void *refcount_array, uint64_t index,
48 uint64_t value);
49 static void set_refcount_ro2(void *refcount_array, uint64_t index,
50 uint64_t value);
51 static void set_refcount_ro3(void *refcount_array, uint64_t index,
52 uint64_t value);
53 static void set_refcount_ro4(void *refcount_array, uint64_t index,
54 uint64_t value);
55 static void set_refcount_ro5(void *refcount_array, uint64_t index,
56 uint64_t value);
57 static void set_refcount_ro6(void *refcount_array, uint64_t index,
58 uint64_t value);
59
60
61 static Qcow2GetRefcountFunc *const get_refcount_funcs[] = {
62 &get_refcount_ro0,
63 &get_refcount_ro1,
64 &get_refcount_ro2,
65 &get_refcount_ro3,
66 &get_refcount_ro4,
67 &get_refcount_ro5,
68 &get_refcount_ro6
69 };
70
71 static Qcow2SetRefcountFunc *const set_refcount_funcs[] = {
72 &set_refcount_ro0,
73 &set_refcount_ro1,
74 &set_refcount_ro2,
75 &set_refcount_ro3,
76 &set_refcount_ro4,
77 &set_refcount_ro5,
78 &set_refcount_ro6
79 };
80
81
82 /*********************************************************/
83 /* refcount handling */
84
85 int qcow2_refcount_init(BlockDriverState *bs)
86 {
87 BDRVQcow2State *s = bs->opaque;
88 unsigned int refcount_table_size2, i;
89 int ret;
90
91 assert(s->refcount_order >= 0 && s->refcount_order <= 6);
92
93 s->get_refcount = get_refcount_funcs[s->refcount_order];
94 s->set_refcount = set_refcount_funcs[s->refcount_order];
95
96 assert(s->refcount_table_size <= INT_MAX / sizeof(uint64_t));
97 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
98 s->refcount_table = g_try_malloc(refcount_table_size2);
99
100 if (s->refcount_table_size > 0) {
101 if (s->refcount_table == NULL) {
102 ret = -ENOMEM;
103 goto fail;
104 }
105 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
106 ret = bdrv_pread(bs->file->bs, s->refcount_table_offset,
107 s->refcount_table, refcount_table_size2);
108 if (ret < 0) {
109 goto fail;
110 }
111 for(i = 0; i < s->refcount_table_size; i++)
112 be64_to_cpus(&s->refcount_table[i]);
113 }
114 return 0;
115 fail:
116 return ret;
117 }
118
119 void qcow2_refcount_close(BlockDriverState *bs)
120 {
121 BDRVQcow2State *s = bs->opaque;
122 g_free(s->refcount_table);
123 }
124
125
126 static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index)
127 {
128 return (((const uint8_t *)refcount_array)[index / 8] >> (index % 8)) & 0x1;
129 }
130
131 static void set_refcount_ro0(void *refcount_array, uint64_t index,
132 uint64_t value)
133 {
134 assert(!(value >> 1));
135 ((uint8_t *)refcount_array)[index / 8] &= ~(0x1 << (index % 8));
136 ((uint8_t *)refcount_array)[index / 8] |= value << (index % 8);
137 }
138
139 static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index)
140 {
141 return (((const uint8_t *)refcount_array)[index / 4] >> (2 * (index % 4)))
142 & 0x3;
143 }
144
145 static void set_refcount_ro1(void *refcount_array, uint64_t index,
146 uint64_t value)
147 {
148 assert(!(value >> 2));
149 ((uint8_t *)refcount_array)[index / 4] &= ~(0x3 << (2 * (index % 4)));
150 ((uint8_t *)refcount_array)[index / 4] |= value << (2 * (index % 4));
151 }
152
153 static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index)
154 {
155 return (((const uint8_t *)refcount_array)[index / 2] >> (4 * (index % 2)))
156 & 0xf;
157 }
158
159 static void set_refcount_ro2(void *refcount_array, uint64_t index,
160 uint64_t value)
161 {
162 assert(!(value >> 4));
163 ((uint8_t *)refcount_array)[index / 2] &= ~(0xf << (4 * (index % 2)));
164 ((uint8_t *)refcount_array)[index / 2] |= value << (4 * (index % 2));
165 }
166
167 static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index)
168 {
169 return ((const uint8_t *)refcount_array)[index];
170 }
171
172 static void set_refcount_ro3(void *refcount_array, uint64_t index,
173 uint64_t value)
174 {
175 assert(!(value >> 8));
176 ((uint8_t *)refcount_array)[index] = value;
177 }
178
179 static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index)
180 {
181 return be16_to_cpu(((const uint16_t *)refcount_array)[index]);
182 }
183
184 static void set_refcount_ro4(void *refcount_array, uint64_t index,
185 uint64_t value)
186 {
187 assert(!(value >> 16));
188 ((uint16_t *)refcount_array)[index] = cpu_to_be16(value);
189 }
190
191 static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index)
192 {
193 return be32_to_cpu(((const uint32_t *)refcount_array)[index]);
194 }
195
196 static void set_refcount_ro5(void *refcount_array, uint64_t index,
197 uint64_t value)
198 {
199 assert(!(value >> 32));
200 ((uint32_t *)refcount_array)[index] = cpu_to_be32(value);
201 }
202
203 static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index)
204 {
205 return be64_to_cpu(((const uint64_t *)refcount_array)[index]);
206 }
207
208 static void set_refcount_ro6(void *refcount_array, uint64_t index,
209 uint64_t value)
210 {
211 ((uint64_t *)refcount_array)[index] = cpu_to_be64(value);
212 }
213
214
215 static int load_refcount_block(BlockDriverState *bs,
216 int64_t refcount_block_offset,
217 void **refcount_block)
218 {
219 BDRVQcow2State *s = bs->opaque;
220 int ret;
221
222 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
223 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
224 refcount_block);
225
226 return ret;
227 }
228
229 /*
230 * Retrieves the refcount of the cluster given by its index and stores it in
231 * *refcount. Returns 0 on success and -errno on failure.
232 */
233 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index,
234 uint64_t *refcount)
235 {
236 BDRVQcow2State *s = bs->opaque;
237 uint64_t refcount_table_index, block_index;
238 int64_t refcount_block_offset;
239 int ret;
240 void *refcount_block;
241
242 refcount_table_index = cluster_index >> s->refcount_block_bits;
243 if (refcount_table_index >= s->refcount_table_size) {
244 *refcount = 0;
245 return 0;
246 }
247 refcount_block_offset =
248 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
249 if (!refcount_block_offset) {
250 *refcount = 0;
251 return 0;
252 }
253
254 if (offset_into_cluster(s, refcount_block_offset)) {
255 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" PRIx64
256 " unaligned (reftable index: %#" PRIx64 ")",
257 refcount_block_offset, refcount_table_index);
258 return -EIO;
259 }
260
261 ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset,
262 &refcount_block);
263 if (ret < 0) {
264 return ret;
265 }
266
267 block_index = cluster_index & (s->refcount_block_size - 1);
268 *refcount = s->get_refcount(refcount_block, block_index);
269
270 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
271
272 return 0;
273 }
274
275 /*
276 * Rounds the refcount table size up to avoid growing the table for each single
277 * refcount block that is allocated.
278 */
279 static unsigned int next_refcount_table_size(BDRVQcow2State *s,
280 unsigned int min_size)
281 {
282 unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1;
283 unsigned int refcount_table_clusters =
284 MAX(1, s->refcount_table_size >> (s->cluster_bits - 3));
285
286 while (min_clusters > refcount_table_clusters) {
287 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
288 }
289
290 return refcount_table_clusters << (s->cluster_bits - 3);
291 }
292
293
294 /* Checks if two offsets are described by the same refcount block */
295 static int in_same_refcount_block(BDRVQcow2State *s, uint64_t offset_a,
296 uint64_t offset_b)
297 {
298 uint64_t block_a = offset_a >> (s->cluster_bits + s->refcount_block_bits);
299 uint64_t block_b = offset_b >> (s->cluster_bits + s->refcount_block_bits);
300
301 return (block_a == block_b);
302 }
303
304 /*
305 * Loads a refcount block. If it doesn't exist yet, it is allocated first
306 * (including growing the refcount table if needed).
307 *
308 * Returns 0 on success or -errno in error case
309 */
310 static int alloc_refcount_block(BlockDriverState *bs,
311 int64_t cluster_index, void **refcount_block)
312 {
313 BDRVQcow2State *s = bs->opaque;
314 unsigned int refcount_table_index;
315 int ret;
316
317 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
318
319 /* Find the refcount block for the given cluster */
320 refcount_table_index = cluster_index >> s->refcount_block_bits;
321
322 if (refcount_table_index < s->refcount_table_size) {
323
324 uint64_t refcount_block_offset =
325 s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK;
326
327 /* If it's already there, we're done */
328 if (refcount_block_offset) {
329 if (offset_into_cluster(s, refcount_block_offset)) {
330 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
331 PRIx64 " unaligned (reftable index: "
332 "%#x)", refcount_block_offset,
333 refcount_table_index);
334 return -EIO;
335 }
336
337 return load_refcount_block(bs, refcount_block_offset,
338 refcount_block);
339 }
340 }
341
342 /*
343 * If we came here, we need to allocate something. Something is at least
344 * a cluster for the new refcount block. It may also include a new refcount
345 * table if the old refcount table is too small.
346 *
347 * Note that allocating clusters here needs some special care:
348 *
349 * - We can't use the normal qcow2_alloc_clusters(), it would try to
350 * increase the refcount and very likely we would end up with an endless
351 * recursion. Instead we must place the refcount blocks in a way that
352 * they can describe them themselves.
353 *
354 * - We need to consider that at this point we are inside update_refcounts
355 * and potentially doing an initial refcount increase. This means that
356 * some clusters have already been allocated by the caller, but their
357 * refcount isn't accurate yet. If we allocate clusters for metadata, we
358 * need to return -EAGAIN to signal the caller that it needs to restart
359 * the search for free clusters.
360 *
361 * - alloc_clusters_noref and qcow2_free_clusters may load a different
362 * refcount block into the cache
363 */
364
365 *refcount_block = NULL;
366
367 /* We write to the refcount table, so we might depend on L2 tables */
368 ret = qcow2_cache_flush(bs, s->l2_table_cache);
369 if (ret < 0) {
370 return ret;
371 }
372
373 /* Allocate the refcount block itself and mark it as used */
374 int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
375 if (new_block < 0) {
376 return new_block;
377 }
378
379 #ifdef DEBUG_ALLOC2
380 fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
381 " at %" PRIx64 "\n",
382 refcount_table_index, cluster_index << s->cluster_bits, new_block);
383 #endif
384
385 if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
386 /* Zero the new refcount block before updating it */
387 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
388 refcount_block);
389 if (ret < 0) {
390 goto fail_block;
391 }
392
393 memset(*refcount_block, 0, s->cluster_size);
394
395 /* The block describes itself, need to update the cache */
396 int block_index = (new_block >> s->cluster_bits) &
397 (s->refcount_block_size - 1);
398 s->set_refcount(*refcount_block, block_index, 1);
399 } else {
400 /* Described somewhere else. This can recurse at most twice before we
401 * arrive at a block that describes itself. */
402 ret = update_refcount(bs, new_block, s->cluster_size, 1, false,
403 QCOW2_DISCARD_NEVER);
404 if (ret < 0) {
405 goto fail_block;
406 }
407
408 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
409 if (ret < 0) {
410 goto fail_block;
411 }
412
413 /* Initialize the new refcount block only after updating its refcount,
414 * update_refcount uses the refcount cache itself */
415 ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block,
416 refcount_block);
417 if (ret < 0) {
418 goto fail_block;
419 }
420
421 memset(*refcount_block, 0, s->cluster_size);
422 }
423
424 /* Now the new refcount block needs to be written to disk */
425 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
426 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache, *refcount_block);
427 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
428 if (ret < 0) {
429 goto fail_block;
430 }
431
432 /* If the refcount table is big enough, just hook the block up there */
433 if (refcount_table_index < s->refcount_table_size) {
434 uint64_t data64 = cpu_to_be64(new_block);
435 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
436 ret = bdrv_pwrite_sync(bs->file->bs,
437 s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
438 &data64, sizeof(data64));
439 if (ret < 0) {
440 goto fail_block;
441 }
442
443 s->refcount_table[refcount_table_index] = new_block;
444
445 /* The new refcount block may be where the caller intended to put its
446 * data, so let it restart the search. */
447 return -EAGAIN;
448 }
449
450 qcow2_cache_put(bs, s->refcount_block_cache, refcount_block);
451
452 /*
453 * If we come here, we need to grow the refcount table. Again, a new
454 * refcount table needs some space and we can't simply allocate to avoid
455 * endless recursion.
456 *
457 * Therefore let's grab new refcount blocks at the end of the image, which
458 * will describe themselves and the new refcount table. This way we can
459 * reference them only in the new table and do the switch to the new
460 * refcount table at once without producing an inconsistent state in
461 * between.
462 */
463 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
464
465 /* Calculate the number of refcount blocks needed so far; this will be the
466 * basis for calculating the index of the first cluster used for the
467 * self-describing refcount structures which we are about to create.
468 *
469 * Because we reached this point, there cannot be any refcount entries for
470 * cluster_index or higher indices yet. However, because new_block has been
471 * allocated to describe that cluster (and it will assume this role later
472 * on), we cannot use that index; also, new_block may actually have a higher
473 * cluster index than cluster_index, so it needs to be taken into account
474 * here (and 1 needs to be added to its value because that cluster is used).
475 */
476 uint64_t blocks_used = DIV_ROUND_UP(MAX(cluster_index + 1,
477 (new_block >> s->cluster_bits) + 1),
478 s->refcount_block_size);
479
480 if (blocks_used > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) {
481 return -EFBIG;
482 }
483
484 /* And now we need at least one block more for the new metadata */
485 uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
486 uint64_t last_table_size;
487 uint64_t blocks_clusters;
488 do {
489 uint64_t table_clusters =
490 size_to_clusters(s, table_size * sizeof(uint64_t));
491 blocks_clusters = 1 +
492 ((table_clusters + s->refcount_block_size - 1)
493 / s->refcount_block_size);
494 uint64_t meta_clusters = table_clusters + blocks_clusters;
495
496 last_table_size = table_size;
497 table_size = next_refcount_table_size(s, blocks_used +
498 ((meta_clusters + s->refcount_block_size - 1)
499 / s->refcount_block_size));
500
501 } while (last_table_size != table_size);
502
503 #ifdef DEBUG_ALLOC2
504 fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
505 s->refcount_table_size, table_size);
506 #endif
507
508 /* Create the new refcount table and blocks */
509 uint64_t meta_offset = (blocks_used * s->refcount_block_size) *
510 s->cluster_size;
511 uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
512 uint64_t *new_table = g_try_new0(uint64_t, table_size);
513 void *new_blocks = g_try_malloc0(blocks_clusters * s->cluster_size);
514
515 assert(table_size > 0 && blocks_clusters > 0);
516 if (new_table == NULL || new_blocks == NULL) {
517 ret = -ENOMEM;
518 goto fail_table;
519 }
520
521 /* Fill the new refcount table */
522 memcpy(new_table, s->refcount_table,
523 s->refcount_table_size * sizeof(uint64_t));
524 new_table[refcount_table_index] = new_block;
525
526 int i;
527 for (i = 0; i < blocks_clusters; i++) {
528 new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
529 }
530
531 /* Fill the refcount blocks */
532 uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
533 int block = 0;
534 for (i = 0; i < table_clusters + blocks_clusters; i++) {
535 s->set_refcount(new_blocks, block++, 1);
536 }
537
538 /* Write refcount blocks to disk */
539 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
540 ret = bdrv_pwrite_sync(bs->file->bs, meta_offset, new_blocks,
541 blocks_clusters * s->cluster_size);
542 g_free(new_blocks);
543 new_blocks = NULL;
544 if (ret < 0) {
545 goto fail_table;
546 }
547
548 /* Write refcount table to disk */
549 for(i = 0; i < table_size; i++) {
550 cpu_to_be64s(&new_table[i]);
551 }
552
553 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
554 ret = bdrv_pwrite_sync(bs->file->bs, table_offset, new_table,
555 table_size * sizeof(uint64_t));
556 if (ret < 0) {
557 goto fail_table;
558 }
559
560 for(i = 0; i < table_size; i++) {
561 be64_to_cpus(&new_table[i]);
562 }
563
564 /* Hook up the new refcount table in the qcow2 header */
565 struct QEMU_PACKED {
566 uint64_t d64;
567 uint32_t d32;
568 } data;
569 cpu_to_be64w(&data.d64, table_offset);
570 cpu_to_be32w(&data.d32, table_clusters);
571 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
572 ret = bdrv_pwrite_sync(bs->file->bs,
573 offsetof(QCowHeader, refcount_table_offset),
574 &data, sizeof(data));
575 if (ret < 0) {
576 goto fail_table;
577 }
578
579 /* And switch it in memory */
580 uint64_t old_table_offset = s->refcount_table_offset;
581 uint64_t old_table_size = s->refcount_table_size;
582
583 g_free(s->refcount_table);
584 s->refcount_table = new_table;
585 s->refcount_table_size = table_size;
586 s->refcount_table_offset = table_offset;
587
588 /* Free old table. */
589 qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t),
590 QCOW2_DISCARD_OTHER);
591
592 ret = load_refcount_block(bs, new_block, refcount_block);
593 if (ret < 0) {
594 return ret;
595 }
596
597 /* If we were trying to do the initial refcount update for some cluster
598 * allocation, we might have used the same clusters to store newly
599 * allocated metadata. Make the caller search some new space. */
600 return -EAGAIN;
601
602 fail_table:
603 g_free(new_blocks);
604 g_free(new_table);
605 fail_block:
606 if (*refcount_block != NULL) {
607 qcow2_cache_put(bs, s->refcount_block_cache, refcount_block);
608 }
609 return ret;
610 }
611
612 void qcow2_process_discards(BlockDriverState *bs, int ret)
613 {
614 BDRVQcow2State *s = bs->opaque;
615 Qcow2DiscardRegion *d, *next;
616
617 QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) {
618 QTAILQ_REMOVE(&s->discards, d, next);
619
620 /* Discard is optional, ignore the return value */
621 if (ret >= 0) {
622 bdrv_discard(bs->file->bs,
623 d->offset >> BDRV_SECTOR_BITS,
624 d->bytes >> BDRV_SECTOR_BITS);
625 }
626
627 g_free(d);
628 }
629 }
630
631 static void update_refcount_discard(BlockDriverState *bs,
632 uint64_t offset, uint64_t length)
633 {
634 BDRVQcow2State *s = bs->opaque;
635 Qcow2DiscardRegion *d, *p, *next;
636
637 QTAILQ_FOREACH(d, &s->discards, next) {
638 uint64_t new_start = MIN(offset, d->offset);
639 uint64_t new_end = MAX(offset + length, d->offset + d->bytes);
640
641 if (new_end - new_start <= length + d->bytes) {
642 /* There can't be any overlap, areas ending up here have no
643 * references any more and therefore shouldn't get freed another
644 * time. */
645 assert(d->bytes + length == new_end - new_start);
646 d->offset = new_start;
647 d->bytes = new_end - new_start;
648 goto found;
649 }
650 }
651
652 d = g_malloc(sizeof(*d));
653 *d = (Qcow2DiscardRegion) {
654 .bs = bs,
655 .offset = offset,
656 .bytes = length,
657 };
658 QTAILQ_INSERT_TAIL(&s->discards, d, next);
659
660 found:
661 /* Merge discard requests if they are adjacent now */
662 QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) {
663 if (p == d
664 || p->offset > d->offset + d->bytes
665 || d->offset > p->offset + p->bytes)
666 {
667 continue;
668 }
669
670 /* Still no overlap possible */
671 assert(p->offset == d->offset + d->bytes
672 || d->offset == p->offset + p->bytes);
673
674 QTAILQ_REMOVE(&s->discards, p, next);
675 d->offset = MIN(d->offset, p->offset);
676 d->bytes += p->bytes;
677 g_free(p);
678 }
679 }
680
681 /* XXX: cache several refcount block clusters ? */
682 /* @addend is the absolute value of the addend; if @decrease is set, @addend
683 * will be subtracted from the current refcount, otherwise it will be added */
684 static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
685 int64_t offset,
686 int64_t length,
687 uint64_t addend,
688 bool decrease,
689 enum qcow2_discard_type type)
690 {
691 BDRVQcow2State *s = bs->opaque;
692 int64_t start, last, cluster_offset;
693 void *refcount_block = NULL;
694 int64_t old_table_index = -1;
695 int ret;
696
697 #ifdef DEBUG_ALLOC2
698 fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64
699 " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "",
700 addend);
701 #endif
702 if (length < 0) {
703 return -EINVAL;
704 } else if (length == 0) {
705 return 0;
706 }
707
708 if (decrease) {
709 qcow2_cache_set_dependency(bs, s->refcount_block_cache,
710 s->l2_table_cache);
711 }
712
713 start = start_of_cluster(s, offset);
714 last = start_of_cluster(s, offset + length - 1);
715 for(cluster_offset = start; cluster_offset <= last;
716 cluster_offset += s->cluster_size)
717 {
718 int block_index;
719 uint64_t refcount;
720 int64_t cluster_index = cluster_offset >> s->cluster_bits;
721 int64_t table_index = cluster_index >> s->refcount_block_bits;
722
723 /* Load the refcount block and allocate it if needed */
724 if (table_index != old_table_index) {
725 if (refcount_block) {
726 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
727 }
728 ret = alloc_refcount_block(bs, cluster_index, &refcount_block);
729 if (ret < 0) {
730 goto fail;
731 }
732 }
733 old_table_index = table_index;
734
735 qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache,
736 refcount_block);
737
738 /* we can update the count and save it */
739 block_index = cluster_index & (s->refcount_block_size - 1);
740
741 refcount = s->get_refcount(refcount_block, block_index);
742 if (decrease ? (refcount - addend > refcount)
743 : (refcount + addend < refcount ||
744 refcount + addend > s->refcount_max))
745 {
746 ret = -EINVAL;
747 goto fail;
748 }
749 if (decrease) {
750 refcount -= addend;
751 } else {
752 refcount += addend;
753 }
754 if (refcount == 0 && cluster_index < s->free_cluster_index) {
755 s->free_cluster_index = cluster_index;
756 }
757 s->set_refcount(refcount_block, block_index, refcount);
758
759 if (refcount == 0 && s->discard_passthrough[type]) {
760 update_refcount_discard(bs, cluster_offset, s->cluster_size);
761 }
762 }
763
764 ret = 0;
765 fail:
766 if (!s->cache_discards) {
767 qcow2_process_discards(bs, ret);
768 }
769
770 /* Write last changed block to disk */
771 if (refcount_block) {
772 qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block);
773 }
774
775 /*
776 * Try do undo any updates if an error is returned (This may succeed in
777 * some cases like ENOSPC for allocating a new refcount block)
778 */
779 if (ret < 0) {
780 int dummy;
781 dummy = update_refcount(bs, offset, cluster_offset - offset, addend,
782 !decrease, QCOW2_DISCARD_NEVER);
783 (void)dummy;
784 }
785
786 return ret;
787 }
788
789 /*
790 * Increases or decreases the refcount of a given cluster.
791 *
792 * @addend is the absolute value of the addend; if @decrease is set, @addend
793 * will be subtracted from the current refcount, otherwise it will be added.
794 *
795 * On success 0 is returned; on failure -errno is returned.
796 */
797 int qcow2_update_cluster_refcount(BlockDriverState *bs,
798 int64_t cluster_index,
799 uint64_t addend, bool decrease,
800 enum qcow2_discard_type type)
801 {
802 BDRVQcow2State *s = bs->opaque;
803 int ret;
804
805 ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend,
806 decrease, type);
807 if (ret < 0) {
808 return ret;
809 }
810
811 return 0;
812 }
813
814
815
816 /*********************************************************/
817 /* cluster allocation functions */
818
819
820
821 /* return < 0 if error */
822 static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size)
823 {
824 BDRVQcow2State *s = bs->opaque;
825 uint64_t i, nb_clusters, refcount;
826 int ret;
827
828 /* We can't allocate clusters if they may still be queued for discard. */
829 if (s->cache_discards) {
830 qcow2_process_discards(bs, 0);
831 }
832
833 nb_clusters = size_to_clusters(s, size);
834 retry:
835 for(i = 0; i < nb_clusters; i++) {
836 uint64_t next_cluster_index = s->free_cluster_index++;
837 ret = qcow2_get_refcount(bs, next_cluster_index, &refcount);
838
839 if (ret < 0) {
840 return ret;
841 } else if (refcount != 0) {
842 goto retry;
843 }
844 }
845
846 /* Make sure that all offsets in the "allocated" range are representable
847 * in an int64_t */
848 if (s->free_cluster_index > 0 &&
849 s->free_cluster_index - 1 > (INT64_MAX >> s->cluster_bits))
850 {
851 return -EFBIG;
852 }
853
854 #ifdef DEBUG_ALLOC2
855 fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
856 size,
857 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
858 #endif
859 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
860 }
861
862 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size)
863 {
864 int64_t offset;
865 int ret;
866
867 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
868 do {
869 offset = alloc_clusters_noref(bs, size);
870 if (offset < 0) {
871 return offset;
872 }
873
874 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
875 } while (ret == -EAGAIN);
876
877 if (ret < 0) {
878 return ret;
879 }
880
881 return offset;
882 }
883
884 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
885 int64_t nb_clusters)
886 {
887 BDRVQcow2State *s = bs->opaque;
888 uint64_t cluster_index, refcount;
889 uint64_t i;
890 int ret;
891
892 assert(nb_clusters >= 0);
893 if (nb_clusters == 0) {
894 return 0;
895 }
896
897 do {
898 /* Check how many clusters there are free */
899 cluster_index = offset >> s->cluster_bits;
900 for(i = 0; i < nb_clusters; i++) {
901 ret = qcow2_get_refcount(bs, cluster_index++, &refcount);
902 if (ret < 0) {
903 return ret;
904 } else if (refcount != 0) {
905 break;
906 }
907 }
908
909 /* And then allocate them */
910 ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false,
911 QCOW2_DISCARD_NEVER);
912 } while (ret == -EAGAIN);
913
914 if (ret < 0) {
915 return ret;
916 }
917
918 return i;
919 }
920
921 /* only used to allocate compressed sectors. We try to allocate
922 contiguous sectors. size must be <= cluster_size */
923 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
924 {
925 BDRVQcow2State *s = bs->opaque;
926 int64_t offset;
927 size_t free_in_cluster;
928 int ret;
929
930 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
931 assert(size > 0 && size <= s->cluster_size);
932 assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset));
933
934 offset = s->free_byte_offset;
935
936 if (offset) {
937 uint64_t refcount;
938 ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount);
939 if (ret < 0) {
940 return ret;
941 }
942
943 if (refcount == s->refcount_max) {
944 offset = 0;
945 }
946 }
947
948 free_in_cluster = s->cluster_size - offset_into_cluster(s, offset);
949 do {
950 if (!offset || free_in_cluster < size) {
951 int64_t new_cluster = alloc_clusters_noref(bs, s->cluster_size);
952 if (new_cluster < 0) {
953 return new_cluster;
954 }
955
956 if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) {
957 offset = new_cluster;
958 free_in_cluster = s->cluster_size;
959 } else {
960 free_in_cluster += s->cluster_size;
961 }
962 }
963
964 assert(offset);
965 ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
966 if (ret < 0) {
967 offset = 0;
968 }
969 } while (ret == -EAGAIN);
970 if (ret < 0) {
971 return ret;
972 }
973
974 /* The cluster refcount was incremented; refcount blocks must be flushed
975 * before the caller's L2 table updates. */
976 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
977
978 s->free_byte_offset = offset + size;
979 if (!offset_into_cluster(s, s->free_byte_offset)) {
980 s->free_byte_offset = 0;
981 }
982
983 return offset;
984 }
985
986 void qcow2_free_clusters(BlockDriverState *bs,
987 int64_t offset, int64_t size,
988 enum qcow2_discard_type type)
989 {
990 int ret;
991
992 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
993 ret = update_refcount(bs, offset, size, 1, true, type);
994 if (ret < 0) {
995 fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
996 /* TODO Remember the clusters to free them later and avoid leaking */
997 }
998 }
999
1000 /*
1001 * Free a cluster using its L2 entry (handles clusters of all types, e.g.
1002 * normal cluster, compressed cluster, etc.)
1003 */
1004 void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
1005 int nb_clusters, enum qcow2_discard_type type)
1006 {
1007 BDRVQcow2State *s = bs->opaque;
1008
1009 switch (qcow2_get_cluster_type(l2_entry)) {
1010 case QCOW2_CLUSTER_COMPRESSED:
1011 {
1012 int nb_csectors;
1013 nb_csectors = ((l2_entry >> s->csize_shift) &
1014 s->csize_mask) + 1;
1015 qcow2_free_clusters(bs,
1016 (l2_entry & s->cluster_offset_mask) & ~511,
1017 nb_csectors * 512, type);
1018 }
1019 break;
1020 case QCOW2_CLUSTER_NORMAL:
1021 case QCOW2_CLUSTER_ZERO:
1022 if (l2_entry & L2E_OFFSET_MASK) {
1023 if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) {
1024 qcow2_signal_corruption(bs, false, -1, -1,
1025 "Cannot free unaligned cluster %#llx",
1026 l2_entry & L2E_OFFSET_MASK);
1027 } else {
1028 qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
1029 nb_clusters << s->cluster_bits, type);
1030 }
1031 }
1032 break;
1033 case QCOW2_CLUSTER_UNALLOCATED:
1034 break;
1035 default:
1036 abort();
1037 }
1038 }
1039
1040
1041
1042 /*********************************************************/
1043 /* snapshots and image creation */
1044
1045
1046
1047 /* update the refcounts of snapshots and the copied flag */
1048 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
1049 int64_t l1_table_offset, int l1_size, int addend)
1050 {
1051 BDRVQcow2State *s = bs->opaque;
1052 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, refcount;
1053 bool l1_allocated = false;
1054 int64_t old_offset, old_l2_offset;
1055 int i, j, l1_modified = 0, nb_csectors;
1056 int ret;
1057
1058 assert(addend >= -1 && addend <= 1);
1059
1060 l2_table = NULL;
1061 l1_table = NULL;
1062 l1_size2 = l1_size * sizeof(uint64_t);
1063
1064 s->cache_discards = true;
1065
1066 /* WARNING: qcow2_snapshot_goto relies on this function not using the
1067 * l1_table_offset when it is the current s->l1_table_offset! Be careful
1068 * when changing this! */
1069 if (l1_table_offset != s->l1_table_offset) {
1070 l1_table = g_try_malloc0(align_offset(l1_size2, 512));
1071 if (l1_size2 && l1_table == NULL) {
1072 ret = -ENOMEM;
1073 goto fail;
1074 }
1075 l1_allocated = true;
1076
1077 ret = bdrv_pread(bs->file->bs, l1_table_offset, l1_table, l1_size2);
1078 if (ret < 0) {
1079 goto fail;
1080 }
1081
1082 for(i = 0;i < l1_size; i++)
1083 be64_to_cpus(&l1_table[i]);
1084 } else {
1085 assert(l1_size == s->l1_size);
1086 l1_table = s->l1_table;
1087 l1_allocated = false;
1088 }
1089
1090 for(i = 0; i < l1_size; i++) {
1091 l2_offset = l1_table[i];
1092 if (l2_offset) {
1093 old_l2_offset = l2_offset;
1094 l2_offset &= L1E_OFFSET_MASK;
1095
1096 if (offset_into_cluster(s, l2_offset)) {
1097 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1098 PRIx64 " unaligned (L1 index: %#x)",
1099 l2_offset, i);
1100 ret = -EIO;
1101 goto fail;
1102 }
1103
1104 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1105 (void**) &l2_table);
1106 if (ret < 0) {
1107 goto fail;
1108 }
1109
1110 for(j = 0; j < s->l2_size; j++) {
1111 uint64_t cluster_index;
1112
1113 offset = be64_to_cpu(l2_table[j]);
1114 old_offset = offset;
1115 offset &= ~QCOW_OFLAG_COPIED;
1116
1117 switch (qcow2_get_cluster_type(offset)) {
1118 case QCOW2_CLUSTER_COMPRESSED:
1119 nb_csectors = ((offset >> s->csize_shift) &
1120 s->csize_mask) + 1;
1121 if (addend != 0) {
1122 ret = update_refcount(bs,
1123 (offset & s->cluster_offset_mask) & ~511,
1124 nb_csectors * 512, abs(addend), addend < 0,
1125 QCOW2_DISCARD_SNAPSHOT);
1126 if (ret < 0) {
1127 goto fail;
1128 }
1129 }
1130 /* compressed clusters are never modified */
1131 refcount = 2;
1132 break;
1133
1134 case QCOW2_CLUSTER_NORMAL:
1135 case QCOW2_CLUSTER_ZERO:
1136 if (offset_into_cluster(s, offset & L2E_OFFSET_MASK)) {
1137 qcow2_signal_corruption(bs, true, -1, -1, "Data "
1138 "cluster offset %#llx "
1139 "unaligned (L2 offset: %#"
1140 PRIx64 ", L2 index: %#x)",
1141 offset & L2E_OFFSET_MASK,
1142 l2_offset, j);
1143 ret = -EIO;
1144 goto fail;
1145 }
1146
1147 cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
1148 if (!cluster_index) {
1149 /* unallocated */
1150 refcount = 0;
1151 break;
1152 }
1153 if (addend != 0) {
1154 ret = qcow2_update_cluster_refcount(bs,
1155 cluster_index, abs(addend), addend < 0,
1156 QCOW2_DISCARD_SNAPSHOT);
1157 if (ret < 0) {
1158 goto fail;
1159 }
1160 }
1161
1162 ret = qcow2_get_refcount(bs, cluster_index, &refcount);
1163 if (ret < 0) {
1164 goto fail;
1165 }
1166 break;
1167
1168 case QCOW2_CLUSTER_UNALLOCATED:
1169 refcount = 0;
1170 break;
1171
1172 default:
1173 abort();
1174 }
1175
1176 if (refcount == 1) {
1177 offset |= QCOW_OFLAG_COPIED;
1178 }
1179 if (offset != old_offset) {
1180 if (addend > 0) {
1181 qcow2_cache_set_dependency(bs, s->l2_table_cache,
1182 s->refcount_block_cache);
1183 }
1184 l2_table[j] = cpu_to_be64(offset);
1185 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache,
1186 l2_table);
1187 }
1188 }
1189
1190 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1191
1192 if (addend != 0) {
1193 ret = qcow2_update_cluster_refcount(bs, l2_offset >>
1194 s->cluster_bits,
1195 abs(addend), addend < 0,
1196 QCOW2_DISCARD_SNAPSHOT);
1197 if (ret < 0) {
1198 goto fail;
1199 }
1200 }
1201 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1202 &refcount);
1203 if (ret < 0) {
1204 goto fail;
1205 } else if (refcount == 1) {
1206 l2_offset |= QCOW_OFLAG_COPIED;
1207 }
1208 if (l2_offset != old_l2_offset) {
1209 l1_table[i] = l2_offset;
1210 l1_modified = 1;
1211 }
1212 }
1213 }
1214
1215 ret = bdrv_flush(bs);
1216 fail:
1217 if (l2_table) {
1218 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1219 }
1220
1221 s->cache_discards = false;
1222 qcow2_process_discards(bs, ret);
1223
1224 /* Update L1 only if it isn't deleted anyway (addend = -1) */
1225 if (ret == 0 && addend >= 0 && l1_modified) {
1226 for (i = 0; i < l1_size; i++) {
1227 cpu_to_be64s(&l1_table[i]);
1228 }
1229
1230 ret = bdrv_pwrite_sync(bs->file->bs, l1_table_offset,
1231 l1_table, l1_size2);
1232
1233 for (i = 0; i < l1_size; i++) {
1234 be64_to_cpus(&l1_table[i]);
1235 }
1236 }
1237 if (l1_allocated)
1238 g_free(l1_table);
1239 return ret;
1240 }
1241
1242
1243
1244
1245 /*********************************************************/
1246 /* refcount checking functions */
1247
1248
1249 static uint64_t refcount_array_byte_size(BDRVQcow2State *s, uint64_t entries)
1250 {
1251 /* This assertion holds because there is no way we can address more than
1252 * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because
1253 * offsets have to be representable in bytes); due to every cluster
1254 * corresponding to one refcount entry, we are well below that limit */
1255 assert(entries < (UINT64_C(1) << (64 - 9)));
1256
1257 /* Thanks to the assertion this will not overflow, because
1258 * s->refcount_order < 7.
1259 * (note: x << s->refcount_order == x * s->refcount_bits) */
1260 return DIV_ROUND_UP(entries << s->refcount_order, 8);
1261 }
1262
1263 /**
1264 * Reallocates *array so that it can hold new_size entries. *size must contain
1265 * the current number of entries in *array. If the reallocation fails, *array
1266 * and *size will not be modified and -errno will be returned. If the
1267 * reallocation is successful, *array will be set to the new buffer, *size
1268 * will be set to new_size and 0 will be returned. The size of the reallocated
1269 * refcount array buffer will be aligned to a cluster boundary, and the newly
1270 * allocated area will be zeroed.
1271 */
1272 static int realloc_refcount_array(BDRVQcow2State *s, void **array,
1273 int64_t *size, int64_t new_size)
1274 {
1275 int64_t old_byte_size, new_byte_size;
1276 void *new_ptr;
1277
1278 /* Round to clusters so the array can be directly written to disk */
1279 old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size))
1280 * s->cluster_size;
1281 new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size))
1282 * s->cluster_size;
1283
1284 if (new_byte_size == old_byte_size) {
1285 *size = new_size;
1286 return 0;
1287 }
1288
1289 assert(new_byte_size > 0);
1290
1291 if (new_byte_size > SIZE_MAX) {
1292 return -ENOMEM;
1293 }
1294
1295 new_ptr = g_try_realloc(*array, new_byte_size);
1296 if (!new_ptr) {
1297 return -ENOMEM;
1298 }
1299
1300 if (new_byte_size > old_byte_size) {
1301 memset((char *)new_ptr + old_byte_size, 0,
1302 new_byte_size - old_byte_size);
1303 }
1304
1305 *array = new_ptr;
1306 *size = new_size;
1307
1308 return 0;
1309 }
1310
1311 /*
1312 * Increases the refcount for a range of clusters in a given refcount table.
1313 * This is used to construct a temporary refcount table out of L1 and L2 tables
1314 * which can be compared to the refcount table saved in the image.
1315 *
1316 * Modifies the number of errors in res.
1317 */
1318 static int inc_refcounts(BlockDriverState *bs,
1319 BdrvCheckResult *res,
1320 void **refcount_table,
1321 int64_t *refcount_table_size,
1322 int64_t offset, int64_t size)
1323 {
1324 BDRVQcow2State *s = bs->opaque;
1325 uint64_t start, last, cluster_offset, k, refcount;
1326 int ret;
1327
1328 if (size <= 0) {
1329 return 0;
1330 }
1331
1332 start = start_of_cluster(s, offset);
1333 last = start_of_cluster(s, offset + size - 1);
1334 for(cluster_offset = start; cluster_offset <= last;
1335 cluster_offset += s->cluster_size) {
1336 k = cluster_offset >> s->cluster_bits;
1337 if (k >= *refcount_table_size) {
1338 ret = realloc_refcount_array(s, refcount_table,
1339 refcount_table_size, k + 1);
1340 if (ret < 0) {
1341 res->check_errors++;
1342 return ret;
1343 }
1344 }
1345
1346 refcount = s->get_refcount(*refcount_table, k);
1347 if (refcount == s->refcount_max) {
1348 fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
1349 "\n", cluster_offset);
1350 fprintf(stderr, "Use qemu-img amend to increase the refcount entry "
1351 "width or qemu-img convert to create a clean copy if the "
1352 "image cannot be opened for writing\n");
1353 res->corruptions++;
1354 continue;
1355 }
1356 s->set_refcount(*refcount_table, k, refcount + 1);
1357 }
1358
1359 return 0;
1360 }
1361
1362 /* Flags for check_refcounts_l1() and check_refcounts_l2() */
1363 enum {
1364 CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */
1365 };
1366
1367 /*
1368 * Increases the refcount in the given refcount table for the all clusters
1369 * referenced in the L2 table. While doing so, performs some checks on L2
1370 * entries.
1371 *
1372 * Returns the number of errors found by the checks or -errno if an internal
1373 * error occurred.
1374 */
1375 static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
1376 void **refcount_table,
1377 int64_t *refcount_table_size, int64_t l2_offset,
1378 int flags)
1379 {
1380 BDRVQcow2State *s = bs->opaque;
1381 uint64_t *l2_table, l2_entry;
1382 uint64_t next_contiguous_offset = 0;
1383 int i, l2_size, nb_csectors, ret;
1384
1385 /* Read L2 table from disk */
1386 l2_size = s->l2_size * sizeof(uint64_t);
1387 l2_table = g_malloc(l2_size);
1388
1389 ret = bdrv_pread(bs->file->bs, l2_offset, l2_table, l2_size);
1390 if (ret < 0) {
1391 fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n");
1392 res->check_errors++;
1393 goto fail;
1394 }
1395
1396 /* Do the actual checks */
1397 for(i = 0; i < s->l2_size; i++) {
1398 l2_entry = be64_to_cpu(l2_table[i]);
1399
1400 switch (qcow2_get_cluster_type(l2_entry)) {
1401 case QCOW2_CLUSTER_COMPRESSED:
1402 /* Compressed clusters don't have QCOW_OFLAG_COPIED */
1403 if (l2_entry & QCOW_OFLAG_COPIED) {
1404 fprintf(stderr, "ERROR: cluster %" PRId64 ": "
1405 "copied flag must never be set for compressed "
1406 "clusters\n", l2_entry >> s->cluster_bits);
1407 l2_entry &= ~QCOW_OFLAG_COPIED;
1408 res->corruptions++;
1409 }
1410
1411 /* Mark cluster as used */
1412 nb_csectors = ((l2_entry >> s->csize_shift) &
1413 s->csize_mask) + 1;
1414 l2_entry &= s->cluster_offset_mask;
1415 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1416 l2_entry & ~511, nb_csectors * 512);
1417 if (ret < 0) {
1418 goto fail;
1419 }
1420
1421 if (flags & CHECK_FRAG_INFO) {
1422 res->bfi.allocated_clusters++;
1423 res->bfi.compressed_clusters++;
1424
1425 /* Compressed clusters are fragmented by nature. Since they
1426 * take up sub-sector space but we only have sector granularity
1427 * I/O we need to re-read the same sectors even for adjacent
1428 * compressed clusters.
1429 */
1430 res->bfi.fragmented_clusters++;
1431 }
1432 break;
1433
1434 case QCOW2_CLUSTER_ZERO:
1435 if ((l2_entry & L2E_OFFSET_MASK) == 0) {
1436 break;
1437 }
1438 /* fall through */
1439
1440 case QCOW2_CLUSTER_NORMAL:
1441 {
1442 uint64_t offset = l2_entry & L2E_OFFSET_MASK;
1443
1444 if (flags & CHECK_FRAG_INFO) {
1445 res->bfi.allocated_clusters++;
1446 if (next_contiguous_offset &&
1447 offset != next_contiguous_offset) {
1448 res->bfi.fragmented_clusters++;
1449 }
1450 next_contiguous_offset = offset + s->cluster_size;
1451 }
1452
1453 /* Mark cluster as used */
1454 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1455 offset, s->cluster_size);
1456 if (ret < 0) {
1457 goto fail;
1458 }
1459
1460 /* Correct offsets are cluster aligned */
1461 if (offset_into_cluster(s, offset)) {
1462 fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
1463 "properly aligned; L2 entry corrupted.\n", offset);
1464 res->corruptions++;
1465 }
1466 break;
1467 }
1468
1469 case QCOW2_CLUSTER_UNALLOCATED:
1470 break;
1471
1472 default:
1473 abort();
1474 }
1475 }
1476
1477 g_free(l2_table);
1478 return 0;
1479
1480 fail:
1481 g_free(l2_table);
1482 return ret;
1483 }
1484
1485 /*
1486 * Increases the refcount for the L1 table, its L2 tables and all referenced
1487 * clusters in the given refcount table. While doing so, performs some checks
1488 * on L1 and L2 entries.
1489 *
1490 * Returns the number of errors found by the checks or -errno if an internal
1491 * error occurred.
1492 */
1493 static int check_refcounts_l1(BlockDriverState *bs,
1494 BdrvCheckResult *res,
1495 void **refcount_table,
1496 int64_t *refcount_table_size,
1497 int64_t l1_table_offset, int l1_size,
1498 int flags)
1499 {
1500 BDRVQcow2State *s = bs->opaque;
1501 uint64_t *l1_table = NULL, l2_offset, l1_size2;
1502 int i, ret;
1503
1504 l1_size2 = l1_size * sizeof(uint64_t);
1505
1506 /* Mark L1 table as used */
1507 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1508 l1_table_offset, l1_size2);
1509 if (ret < 0) {
1510 goto fail;
1511 }
1512
1513 /* Read L1 table entries from disk */
1514 if (l1_size2 > 0) {
1515 l1_table = g_try_malloc(l1_size2);
1516 if (l1_table == NULL) {
1517 ret = -ENOMEM;
1518 res->check_errors++;
1519 goto fail;
1520 }
1521 ret = bdrv_pread(bs->file->bs, l1_table_offset, l1_table, l1_size2);
1522 if (ret < 0) {
1523 fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
1524 res->check_errors++;
1525 goto fail;
1526 }
1527 for(i = 0;i < l1_size; i++)
1528 be64_to_cpus(&l1_table[i]);
1529 }
1530
1531 /* Do the actual checks */
1532 for(i = 0; i < l1_size; i++) {
1533 l2_offset = l1_table[i];
1534 if (l2_offset) {
1535 /* Mark L2 table as used */
1536 l2_offset &= L1E_OFFSET_MASK;
1537 ret = inc_refcounts(bs, res, refcount_table, refcount_table_size,
1538 l2_offset, s->cluster_size);
1539 if (ret < 0) {
1540 goto fail;
1541 }
1542
1543 /* L2 tables are cluster aligned */
1544 if (offset_into_cluster(s, l2_offset)) {
1545 fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
1546 "cluster aligned; L1 entry corrupted\n", l2_offset);
1547 res->corruptions++;
1548 }
1549
1550 /* Process and check L2 entries */
1551 ret = check_refcounts_l2(bs, res, refcount_table,
1552 refcount_table_size, l2_offset, flags);
1553 if (ret < 0) {
1554 goto fail;
1555 }
1556 }
1557 }
1558 g_free(l1_table);
1559 return 0;
1560
1561 fail:
1562 g_free(l1_table);
1563 return ret;
1564 }
1565
1566 /*
1567 * Checks the OFLAG_COPIED flag for all L1 and L2 entries.
1568 *
1569 * This function does not print an error message nor does it increment
1570 * check_errors if qcow2_get_refcount fails (this is because such an error will
1571 * have been already detected and sufficiently signaled by the calling function
1572 * (qcow2_check_refcounts) by the time this function is called).
1573 */
1574 static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
1575 BdrvCheckMode fix)
1576 {
1577 BDRVQcow2State *s = bs->opaque;
1578 uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size);
1579 int ret;
1580 uint64_t refcount;
1581 int i, j;
1582
1583 for (i = 0; i < s->l1_size; i++) {
1584 uint64_t l1_entry = s->l1_table[i];
1585 uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK;
1586 bool l2_dirty = false;
1587
1588 if (!l2_offset) {
1589 continue;
1590 }
1591
1592 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1593 &refcount);
1594 if (ret < 0) {
1595 /* don't print message nor increment check_errors */
1596 continue;
1597 }
1598 if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) {
1599 fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d "
1600 "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
1601 fix & BDRV_FIX_ERRORS ? "Repairing" :
1602 "ERROR",
1603 i, l1_entry, refcount);
1604 if (fix & BDRV_FIX_ERRORS) {
1605 s->l1_table[i] = refcount == 1
1606 ? l1_entry | QCOW_OFLAG_COPIED
1607 : l1_entry & ~QCOW_OFLAG_COPIED;
1608 ret = qcow2_write_l1_entry(bs, i);
1609 if (ret < 0) {
1610 res->check_errors++;
1611 goto fail;
1612 }
1613 res->corruptions_fixed++;
1614 } else {
1615 res->corruptions++;
1616 }
1617 }
1618
1619 ret = bdrv_pread(bs->file->bs, l2_offset, l2_table,
1620 s->l2_size * sizeof(uint64_t));
1621 if (ret < 0) {
1622 fprintf(stderr, "ERROR: Could not read L2 table: %s\n",
1623 strerror(-ret));
1624 res->check_errors++;
1625 goto fail;
1626 }
1627
1628 for (j = 0; j < s->l2_size; j++) {
1629 uint64_t l2_entry = be64_to_cpu(l2_table[j]);
1630 uint64_t data_offset = l2_entry & L2E_OFFSET_MASK;
1631 int cluster_type = qcow2_get_cluster_type(l2_entry);
1632
1633 if ((cluster_type == QCOW2_CLUSTER_NORMAL) ||
1634 ((cluster_type == QCOW2_CLUSTER_ZERO) && (data_offset != 0))) {
1635 ret = qcow2_get_refcount(bs,
1636 data_offset >> s->cluster_bits,
1637 &refcount);
1638 if (ret < 0) {
1639 /* don't print message nor increment check_errors */
1640 continue;
1641 }
1642 if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
1643 fprintf(stderr, "%s OFLAG_COPIED data cluster: "
1644 "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
1645 fix & BDRV_FIX_ERRORS ? "Repairing" :
1646 "ERROR",
1647 l2_entry, refcount);
1648 if (fix & BDRV_FIX_ERRORS) {
1649 l2_table[j] = cpu_to_be64(refcount == 1
1650 ? l2_entry | QCOW_OFLAG_COPIED
1651 : l2_entry & ~QCOW_OFLAG_COPIED);
1652 l2_dirty = true;
1653 res->corruptions_fixed++;
1654 } else {
1655 res->corruptions++;
1656 }
1657 }
1658 }
1659 }
1660
1661 if (l2_dirty) {
1662 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
1663 l2_offset, s->cluster_size);
1664 if (ret < 0) {
1665 fprintf(stderr, "ERROR: Could not write L2 table; metadata "
1666 "overlap check failed: %s\n", strerror(-ret));
1667 res->check_errors++;
1668 goto fail;
1669 }
1670
1671 ret = bdrv_pwrite(bs->file->bs, l2_offset, l2_table,
1672 s->cluster_size);
1673 if (ret < 0) {
1674 fprintf(stderr, "ERROR: Could not write L2 table: %s\n",
1675 strerror(-ret));
1676 res->check_errors++;
1677 goto fail;
1678 }
1679 }
1680 }
1681
1682 ret = 0;
1683
1684 fail:
1685 qemu_vfree(l2_table);
1686 return ret;
1687 }
1688
1689 /*
1690 * Checks consistency of refblocks and accounts for each refblock in
1691 * *refcount_table.
1692 */
1693 static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res,
1694 BdrvCheckMode fix, bool *rebuild,
1695 void **refcount_table, int64_t *nb_clusters)
1696 {
1697 BDRVQcow2State *s = bs->opaque;
1698 int64_t i, size;
1699 int ret;
1700
1701 for(i = 0; i < s->refcount_table_size; i++) {
1702 uint64_t offset, cluster;
1703 offset = s->refcount_table[i];
1704 cluster = offset >> s->cluster_bits;
1705
1706 /* Refcount blocks are cluster aligned */
1707 if (offset_into_cluster(s, offset)) {
1708 fprintf(stderr, "ERROR refcount block %" PRId64 " is not "
1709 "cluster aligned; refcount table entry corrupted\n", i);
1710 res->corruptions++;
1711 *rebuild = true;
1712 continue;
1713 }
1714
1715 if (cluster >= *nb_clusters) {
1716 fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n",
1717 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
1718
1719 if (fix & BDRV_FIX_ERRORS) {
1720 int64_t new_nb_clusters;
1721
1722 if (offset > INT64_MAX - s->cluster_size) {
1723 ret = -EINVAL;
1724 goto resize_fail;
1725 }
1726
1727 ret = bdrv_truncate(bs->file->bs, offset + s->cluster_size);
1728 if (ret < 0) {
1729 goto resize_fail;
1730 }
1731 size = bdrv_getlength(bs->file->bs);
1732 if (size < 0) {
1733 ret = size;
1734 goto resize_fail;
1735 }
1736
1737 new_nb_clusters = size_to_clusters(s, size);
1738 assert(new_nb_clusters >= *nb_clusters);
1739
1740 ret = realloc_refcount_array(s, refcount_table,
1741 nb_clusters, new_nb_clusters);
1742 if (ret < 0) {
1743 res->check_errors++;
1744 return ret;
1745 }
1746
1747 if (cluster >= *nb_clusters) {
1748 ret = -EINVAL;
1749 goto resize_fail;
1750 }
1751
1752 res->corruptions_fixed++;
1753 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1754 offset, s->cluster_size);
1755 if (ret < 0) {
1756 return ret;
1757 }
1758 /* No need to check whether the refcount is now greater than 1:
1759 * This area was just allocated and zeroed, so it can only be
1760 * exactly 1 after inc_refcounts() */
1761 continue;
1762
1763 resize_fail:
1764 res->corruptions++;
1765 *rebuild = true;
1766 fprintf(stderr, "ERROR could not resize image: %s\n",
1767 strerror(-ret));
1768 } else {
1769 res->corruptions++;
1770 }
1771 continue;
1772 }
1773
1774 if (offset != 0) {
1775 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1776 offset, s->cluster_size);
1777 if (ret < 0) {
1778 return ret;
1779 }
1780 if (s->get_refcount(*refcount_table, cluster) != 1) {
1781 fprintf(stderr, "ERROR refcount block %" PRId64
1782 " refcount=%" PRIu64 "\n", i,
1783 s->get_refcount(*refcount_table, cluster));
1784 res->corruptions++;
1785 *rebuild = true;
1786 }
1787 }
1788 }
1789
1790 return 0;
1791 }
1792
1793 /*
1794 * Calculates an in-memory refcount table.
1795 */
1796 static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1797 BdrvCheckMode fix, bool *rebuild,
1798 void **refcount_table, int64_t *nb_clusters)
1799 {
1800 BDRVQcow2State *s = bs->opaque;
1801 int64_t i;
1802 QCowSnapshot *sn;
1803 int ret;
1804
1805 if (!*refcount_table) {
1806 int64_t old_size = 0;
1807 ret = realloc_refcount_array(s, refcount_table,
1808 &old_size, *nb_clusters);
1809 if (ret < 0) {
1810 res->check_errors++;
1811 return ret;
1812 }
1813 }
1814
1815 /* header */
1816 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1817 0, s->cluster_size);
1818 if (ret < 0) {
1819 return ret;
1820 }
1821
1822 /* current L1 table */
1823 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1824 s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO);
1825 if (ret < 0) {
1826 return ret;
1827 }
1828
1829 /* snapshots */
1830 for (i = 0; i < s->nb_snapshots; i++) {
1831 sn = s->snapshots + i;
1832 ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters,
1833 sn->l1_table_offset, sn->l1_size, 0);
1834 if (ret < 0) {
1835 return ret;
1836 }
1837 }
1838 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1839 s->snapshots_offset, s->snapshots_size);
1840 if (ret < 0) {
1841 return ret;
1842 }
1843
1844 /* refcount data */
1845 ret = inc_refcounts(bs, res, refcount_table, nb_clusters,
1846 s->refcount_table_offset,
1847 s->refcount_table_size * sizeof(uint64_t));
1848 if (ret < 0) {
1849 return ret;
1850 }
1851
1852 return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters);
1853 }
1854
1855 /*
1856 * Compares the actual reference count for each cluster in the image against the
1857 * refcount as reported by the refcount structures on-disk.
1858 */
1859 static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
1860 BdrvCheckMode fix, bool *rebuild,
1861 int64_t *highest_cluster,
1862 void *refcount_table, int64_t nb_clusters)
1863 {
1864 BDRVQcow2State *s = bs->opaque;
1865 int64_t i;
1866 uint64_t refcount1, refcount2;
1867 int ret;
1868
1869 for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) {
1870 ret = qcow2_get_refcount(bs, i, &refcount1);
1871 if (ret < 0) {
1872 fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n",
1873 i, strerror(-ret));
1874 res->check_errors++;
1875 continue;
1876 }
1877
1878 refcount2 = s->get_refcount(refcount_table, i);
1879
1880 if (refcount1 > 0 || refcount2 > 0) {
1881 *highest_cluster = i;
1882 }
1883
1884 if (refcount1 != refcount2) {
1885 /* Check if we're allowed to fix the mismatch */
1886 int *num_fixed = NULL;
1887 if (refcount1 == 0) {
1888 *rebuild = true;
1889 } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) {
1890 num_fixed = &res->leaks_fixed;
1891 } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) {
1892 num_fixed = &res->corruptions_fixed;
1893 }
1894
1895 fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64
1896 " reference=%" PRIu64 "\n",
1897 num_fixed != NULL ? "Repairing" :
1898 refcount1 < refcount2 ? "ERROR" :
1899 "Leaked",
1900 i, refcount1, refcount2);
1901
1902 if (num_fixed) {
1903 ret = update_refcount(bs, i << s->cluster_bits, 1,
1904 refcount_diff(refcount1, refcount2),
1905 refcount1 > refcount2,
1906 QCOW2_DISCARD_ALWAYS);
1907 if (ret >= 0) {
1908 (*num_fixed)++;
1909 continue;
1910 }
1911 }
1912
1913 /* And if we couldn't, print an error */
1914 if (refcount1 < refcount2) {
1915 res->corruptions++;
1916 } else {
1917 res->leaks++;
1918 }
1919 }
1920 }
1921 }
1922
1923 /*
1924 * Allocates clusters using an in-memory refcount table (IMRT) in contrast to
1925 * the on-disk refcount structures.
1926 *
1927 * On input, *first_free_cluster tells where to start looking, and need not
1928 * actually be a free cluster; the returned offset will not be before that
1929 * cluster. On output, *first_free_cluster points to the first gap found, even
1930 * if that gap was too small to be used as the returned offset.
1931 *
1932 * Note that *first_free_cluster is a cluster index whereas the return value is
1933 * an offset.
1934 */
1935 static int64_t alloc_clusters_imrt(BlockDriverState *bs,
1936 int cluster_count,
1937 void **refcount_table,
1938 int64_t *imrt_nb_clusters,
1939 int64_t *first_free_cluster)
1940 {
1941 BDRVQcow2State *s = bs->opaque;
1942 int64_t cluster = *first_free_cluster, i;
1943 bool first_gap = true;
1944 int contiguous_free_clusters;
1945 int ret;
1946
1947 /* Starting at *first_free_cluster, find a range of at least cluster_count
1948 * continuously free clusters */
1949 for (contiguous_free_clusters = 0;
1950 cluster < *imrt_nb_clusters &&
1951 contiguous_free_clusters < cluster_count;
1952 cluster++)
1953 {
1954 if (!s->get_refcount(*refcount_table, cluster)) {
1955 contiguous_free_clusters++;
1956 if (first_gap) {
1957 /* If this is the first free cluster found, update
1958 * *first_free_cluster accordingly */
1959 *first_free_cluster = cluster;
1960 first_gap = false;
1961 }
1962 } else if (contiguous_free_clusters) {
1963 contiguous_free_clusters = 0;
1964 }
1965 }
1966
1967 /* If contiguous_free_clusters is greater than zero, it contains the number
1968 * of continuously free clusters until the current cluster; the first free
1969 * cluster in the current "gap" is therefore
1970 * cluster - contiguous_free_clusters */
1971
1972 /* If no such range could be found, grow the in-memory refcount table
1973 * accordingly to append free clusters at the end of the image */
1974 if (contiguous_free_clusters < cluster_count) {
1975 /* contiguous_free_clusters clusters are already empty at the image end;
1976 * we need cluster_count clusters; therefore, we have to allocate
1977 * cluster_count - contiguous_free_clusters new clusters at the end of
1978 * the image (which is the current value of cluster; note that cluster
1979 * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond
1980 * the image end) */
1981 ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters,
1982 cluster + cluster_count
1983 - contiguous_free_clusters);
1984 if (ret < 0) {
1985 return ret;
1986 }
1987 }
1988
1989 /* Go back to the first free cluster */
1990 cluster -= contiguous_free_clusters;
1991 for (i = 0; i < cluster_count; i++) {
1992 s->set_refcount(*refcount_table, cluster + i, 1);
1993 }
1994
1995 return cluster << s->cluster_bits;
1996 }
1997
1998 /*
1999 * Creates a new refcount structure based solely on the in-memory information
2000 * given through *refcount_table. All necessary allocations will be reflected
2001 * in that array.
2002 *
2003 * On success, the old refcount structure is leaked (it will be covered by the
2004 * new refcount structure).
2005 */
2006 static int rebuild_refcount_structure(BlockDriverState *bs,
2007 BdrvCheckResult *res,
2008 void **refcount_table,
2009 int64_t *nb_clusters)
2010 {
2011 BDRVQcow2State *s = bs->opaque;
2012 int64_t first_free_cluster = 0, reftable_offset = -1, cluster = 0;
2013 int64_t refblock_offset, refblock_start, refblock_index;
2014 uint32_t reftable_size = 0;
2015 uint64_t *on_disk_reftable = NULL;
2016 void *on_disk_refblock;
2017 int ret = 0;
2018 struct {
2019 uint64_t reftable_offset;
2020 uint32_t reftable_clusters;
2021 } QEMU_PACKED reftable_offset_and_clusters;
2022
2023 qcow2_cache_empty(bs, s->refcount_block_cache);
2024
2025 write_refblocks:
2026 for (; cluster < *nb_clusters; cluster++) {
2027 if (!s->get_refcount(*refcount_table, cluster)) {
2028 continue;
2029 }
2030
2031 refblock_index = cluster >> s->refcount_block_bits;
2032 refblock_start = refblock_index << s->refcount_block_bits;
2033
2034 /* Don't allocate a cluster in a refblock already written to disk */
2035 if (first_free_cluster < refblock_start) {
2036 first_free_cluster = refblock_start;
2037 }
2038 refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table,
2039 nb_clusters, &first_free_cluster);
2040 if (refblock_offset < 0) {
2041 fprintf(stderr, "ERROR allocating refblock: %s\n",
2042 strerror(-refblock_offset));
2043 res->check_errors++;
2044 ret = refblock_offset;
2045 goto fail;
2046 }
2047
2048 if (reftable_size <= refblock_index) {
2049 uint32_t old_reftable_size = reftable_size;
2050 uint64_t *new_on_disk_reftable;
2051
2052 reftable_size = ROUND_UP((refblock_index + 1) * sizeof(uint64_t),
2053 s->cluster_size) / sizeof(uint64_t);
2054 new_on_disk_reftable = g_try_realloc(on_disk_reftable,
2055 reftable_size *
2056 sizeof(uint64_t));
2057 if (!new_on_disk_reftable) {
2058 res->check_errors++;
2059 ret = -ENOMEM;
2060 goto fail;
2061 }
2062 on_disk_reftable = new_on_disk_reftable;
2063
2064 memset(on_disk_reftable + old_reftable_size, 0,
2065 (reftable_size - old_reftable_size) * sizeof(uint64_t));
2066
2067 /* The offset we have for the reftable is now no longer valid;
2068 * this will leak that range, but we can easily fix that by running
2069 * a leak-fixing check after this rebuild operation */
2070 reftable_offset = -1;
2071 }
2072 on_disk_reftable[refblock_index] = refblock_offset;
2073
2074 /* If this is apparently the last refblock (for now), try to squeeze the
2075 * reftable in */
2076 if (refblock_index == (*nb_clusters - 1) >> s->refcount_block_bits &&
2077 reftable_offset < 0)
2078 {
2079 uint64_t reftable_clusters = size_to_clusters(s, reftable_size *
2080 sizeof(uint64_t));
2081 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2082 refcount_table, nb_clusters,
2083 &first_free_cluster);
2084 if (reftable_offset < 0) {
2085 fprintf(stderr, "ERROR allocating reftable: %s\n",
2086 strerror(-reftable_offset));
2087 res->check_errors++;
2088 ret = reftable_offset;
2089 goto fail;
2090 }
2091 }
2092
2093 ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset,
2094 s->cluster_size);
2095 if (ret < 0) {
2096 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2097 goto fail;
2098 }
2099
2100 /* The size of *refcount_table is always cluster-aligned, therefore the
2101 * write operation will not overflow */
2102 on_disk_refblock = (void *)((char *) *refcount_table +
2103 refblock_index * s->cluster_size);
2104
2105 ret = bdrv_write(bs->file->bs, refblock_offset / BDRV_SECTOR_SIZE,
2106 on_disk_refblock, s->cluster_sectors);
2107 if (ret < 0) {
2108 fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret));
2109 goto fail;
2110 }
2111
2112 /* Go to the end of this refblock */
2113 cluster = refblock_start + s->refcount_block_size - 1;
2114 }
2115
2116 if (reftable_offset < 0) {
2117 uint64_t post_refblock_start, reftable_clusters;
2118
2119 post_refblock_start = ROUND_UP(*nb_clusters, s->refcount_block_size);
2120 reftable_clusters = size_to_clusters(s,
2121 reftable_size * sizeof(uint64_t));
2122 /* Not pretty but simple */
2123 if (first_free_cluster < post_refblock_start) {
2124 first_free_cluster = post_refblock_start;
2125 }
2126 reftable_offset = alloc_clusters_imrt(bs, reftable_clusters,
2127 refcount_table, nb_clusters,
2128 &first_free_cluster);
2129 if (reftable_offset < 0) {
2130 fprintf(stderr, "ERROR allocating reftable: %s\n",
2131 strerror(-reftable_offset));
2132 res->check_errors++;
2133 ret = reftable_offset;
2134 goto fail;
2135 }
2136
2137 goto write_refblocks;
2138 }
2139
2140 assert(on_disk_reftable);
2141
2142 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2143 cpu_to_be64s(&on_disk_reftable[refblock_index]);
2144 }
2145
2146 ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset,
2147 reftable_size * sizeof(uint64_t));
2148 if (ret < 0) {
2149 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2150 goto fail;
2151 }
2152
2153 assert(reftable_size < INT_MAX / sizeof(uint64_t));
2154 ret = bdrv_pwrite(bs->file->bs, reftable_offset, on_disk_reftable,
2155 reftable_size * sizeof(uint64_t));
2156 if (ret < 0) {
2157 fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret));
2158 goto fail;
2159 }
2160
2161 /* Enter new reftable into the image header */
2162 cpu_to_be64w(&reftable_offset_and_clusters.reftable_offset,
2163 reftable_offset);
2164 cpu_to_be32w(&reftable_offset_and_clusters.reftable_clusters,
2165 size_to_clusters(s, reftable_size * sizeof(uint64_t)));
2166 ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader,
2167 refcount_table_offset),
2168 &reftable_offset_and_clusters,
2169 sizeof(reftable_offset_and_clusters));
2170 if (ret < 0) {
2171 fprintf(stderr, "ERROR setting reftable: %s\n", strerror(-ret));
2172 goto fail;
2173 }
2174
2175 for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) {
2176 be64_to_cpus(&on_disk_reftable[refblock_index]);
2177 }
2178 s->refcount_table = on_disk_reftable;
2179 s->refcount_table_offset = reftable_offset;
2180 s->refcount_table_size = reftable_size;
2181
2182 return 0;
2183
2184 fail:
2185 g_free(on_disk_reftable);
2186 return ret;
2187 }
2188
2189 /*
2190 * Checks an image for refcount consistency.
2191 *
2192 * Returns 0 if no errors are found, the number of errors in case the image is
2193 * detected as corrupted, and -errno when an internal error occurred.
2194 */
2195 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
2196 BdrvCheckMode fix)
2197 {
2198 BDRVQcow2State *s = bs->opaque;
2199 BdrvCheckResult pre_compare_res;
2200 int64_t size, highest_cluster, nb_clusters;
2201 void *refcount_table = NULL;
2202 bool rebuild = false;
2203 int ret;
2204
2205 size = bdrv_getlength(bs->file->bs);
2206 if (size < 0) {
2207 res->check_errors++;
2208 return size;
2209 }
2210
2211 nb_clusters = size_to_clusters(s, size);
2212 if (nb_clusters > INT_MAX) {
2213 res->check_errors++;
2214 return -EFBIG;
2215 }
2216
2217 res->bfi.total_clusters =
2218 size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE);
2219
2220 ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table,
2221 &nb_clusters);
2222 if (ret < 0) {
2223 goto fail;
2224 }
2225
2226 /* In case we don't need to rebuild the refcount structure (but want to fix
2227 * something), this function is immediately called again, in which case the
2228 * result should be ignored */
2229 pre_compare_res = *res;
2230 compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table,
2231 nb_clusters);
2232
2233 if (rebuild && (fix & BDRV_FIX_ERRORS)) {
2234 BdrvCheckResult old_res = *res;
2235 int fresh_leaks = 0;
2236
2237 fprintf(stderr, "Rebuilding refcount structure\n");
2238 ret = rebuild_refcount_structure(bs, res, &refcount_table,
2239 &nb_clusters);
2240 if (ret < 0) {
2241 goto fail;
2242 }
2243
2244 res->corruptions = 0;
2245 res->leaks = 0;
2246
2247 /* Because the old reftable has been exchanged for a new one the
2248 * references have to be recalculated */
2249 rebuild = false;
2250 memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters));
2251 ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table,
2252 &nb_clusters);
2253 if (ret < 0) {
2254 goto fail;
2255 }
2256
2257 if (fix & BDRV_FIX_LEAKS) {
2258 /* The old refcount structures are now leaked, fix it; the result
2259 * can be ignored, aside from leaks which were introduced by
2260 * rebuild_refcount_structure() that could not be fixed */
2261 BdrvCheckResult saved_res = *res;
2262 *res = (BdrvCheckResult){ 0 };
2263
2264 compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild,
2265 &highest_cluster, refcount_table, nb_clusters);
2266 if (rebuild) {
2267 fprintf(stderr, "ERROR rebuilt refcount structure is still "
2268 "broken\n");
2269 }
2270
2271 /* Any leaks accounted for here were introduced by
2272 * rebuild_refcount_structure() because that function has created a
2273 * new refcount structure from scratch */
2274 fresh_leaks = res->leaks;
2275 *res = saved_res;
2276 }
2277
2278 if (res->corruptions < old_res.corruptions) {
2279 res->corruptions_fixed += old_res.corruptions - res->corruptions;
2280 }
2281 if (res->leaks < old_res.leaks) {
2282 res->leaks_fixed += old_res.leaks - res->leaks;
2283 }
2284 res->leaks += fresh_leaks;
2285 } else if (fix) {
2286 if (rebuild) {
2287 fprintf(stderr, "ERROR need to rebuild refcount structures\n");
2288 res->check_errors++;
2289 ret = -EIO;
2290 goto fail;
2291 }
2292
2293 if (res->leaks || res->corruptions) {
2294 *res = pre_compare_res;
2295 compare_refcounts(bs, res, fix, &rebuild, &highest_cluster,
2296 refcount_table, nb_clusters);
2297 }
2298 }
2299
2300 /* check OFLAG_COPIED */
2301 ret = check_oflag_copied(bs, res, fix);
2302 if (ret < 0) {
2303 goto fail;
2304 }
2305
2306 res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
2307 ret = 0;
2308
2309 fail:
2310 g_free(refcount_table);
2311
2312 return ret;
2313 }
2314
2315 #define overlaps_with(ofs, sz) \
2316 ranges_overlap(offset, size, ofs, sz)
2317
2318 /*
2319 * Checks if the given offset into the image file is actually free to use by
2320 * looking for overlaps with important metadata sections (L1/L2 tables etc.),
2321 * i.e. a sanity check without relying on the refcount tables.
2322 *
2323 * The ign parameter specifies what checks not to perform (being a bitmask of
2324 * QCow2MetadataOverlap values), i.e., what sections to ignore.
2325 *
2326 * Returns:
2327 * - 0 if writing to this offset will not affect the mentioned metadata
2328 * - a positive QCow2MetadataOverlap value indicating one overlapping section
2329 * - a negative value (-errno) indicating an error while performing a check,
2330 * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
2331 */
2332 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset,
2333 int64_t size)
2334 {
2335 BDRVQcow2State *s = bs->opaque;
2336 int chk = s->overlap_check & ~ign;
2337 int i, j;
2338
2339 if (!size) {
2340 return 0;
2341 }
2342
2343 if (chk & QCOW2_OL_MAIN_HEADER) {
2344 if (offset < s->cluster_size) {
2345 return QCOW2_OL_MAIN_HEADER;
2346 }
2347 }
2348
2349 /* align range to test to cluster boundaries */
2350 size = align_offset(offset_into_cluster(s, offset) + size, s->cluster_size);
2351 offset = start_of_cluster(s, offset);
2352
2353 if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) {
2354 if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) {
2355 return QCOW2_OL_ACTIVE_L1;
2356 }
2357 }
2358
2359 if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) {
2360 if (overlaps_with(s->refcount_table_offset,
2361 s->refcount_table_size * sizeof(uint64_t))) {
2362 return QCOW2_OL_REFCOUNT_TABLE;
2363 }
2364 }
2365
2366 if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) {
2367 if (overlaps_with(s->snapshots_offset, s->snapshots_size)) {
2368 return QCOW2_OL_SNAPSHOT_TABLE;
2369 }
2370 }
2371
2372 if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) {
2373 for (i = 0; i < s->nb_snapshots; i++) {
2374 if (s->snapshots[i].l1_size &&
2375 overlaps_with(s->snapshots[i].l1_table_offset,
2376 s->snapshots[i].l1_size * sizeof(uint64_t))) {
2377 return QCOW2_OL_INACTIVE_L1;
2378 }
2379 }
2380 }
2381
2382 if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) {
2383 for (i = 0; i < s->l1_size; i++) {
2384 if ((s->l1_table[i] & L1E_OFFSET_MASK) &&
2385 overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK,
2386 s->cluster_size)) {
2387 return QCOW2_OL_ACTIVE_L2;
2388 }
2389 }
2390 }
2391
2392 if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) {
2393 for (i = 0; i < s->refcount_table_size; i++) {
2394 if ((s->refcount_table[i] & REFT_OFFSET_MASK) &&
2395 overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK,
2396 s->cluster_size)) {
2397 return QCOW2_OL_REFCOUNT_BLOCK;
2398 }
2399 }
2400 }
2401
2402 if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) {
2403 for (i = 0; i < s->nb_snapshots; i++) {
2404 uint64_t l1_ofs = s->snapshots[i].l1_table_offset;
2405 uint32_t l1_sz = s->snapshots[i].l1_size;
2406 uint64_t l1_sz2 = l1_sz * sizeof(uint64_t);
2407 uint64_t *l1 = g_try_malloc(l1_sz2);
2408 int ret;
2409
2410 if (l1_sz2 && l1 == NULL) {
2411 return -ENOMEM;
2412 }
2413
2414 ret = bdrv_pread(bs->file->bs, l1_ofs, l1, l1_sz2);
2415 if (ret < 0) {
2416 g_free(l1);
2417 return ret;
2418 }
2419
2420 for (j = 0; j < l1_sz; j++) {
2421 uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK;
2422 if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) {
2423 g_free(l1);
2424 return QCOW2_OL_INACTIVE_L2;
2425 }
2426 }
2427
2428 g_free(l1);
2429 }
2430 }
2431
2432 return 0;
2433 }
2434
2435 static const char *metadata_ol_names[] = {
2436 [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header",
2437 [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table",
2438 [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table",
2439 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table",
2440 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block",
2441 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table",
2442 [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table",
2443 [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table",
2444 };
2445
2446 /*
2447 * First performs a check for metadata overlaps (through
2448 * qcow2_check_metadata_overlap); if that fails with a negative value (error
2449 * while performing a check), that value is returned. If an impending overlap
2450 * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
2451 * and -EIO returned.
2452 *
2453 * Returns 0 if there were neither overlaps nor errors while checking for
2454 * overlaps; or a negative value (-errno) on error.
2455 */
2456 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset,
2457 int64_t size)
2458 {
2459 int ret = qcow2_check_metadata_overlap(bs, ign, offset, size);
2460
2461 if (ret < 0) {
2462 return ret;
2463 } else if (ret > 0) {
2464 int metadata_ol_bitnr = ctz32(ret);
2465 assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR);
2466
2467 qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid "
2468 "write on metadata (overlaps with %s)",
2469 metadata_ol_names[metadata_ol_bitnr]);
2470 return -EIO;
2471 }
2472
2473 return 0;
2474 }
2475
2476 /* A pointer to a function of this type is given to walk_over_reftable(). That
2477 * function will create refblocks and pass them to a RefblockFinishOp once they
2478 * are completed (@refblock). @refblock_empty is set if the refblock is
2479 * completely empty.
2480 *
2481 * Along with the refblock, a corresponding reftable entry is passed, in the
2482 * reftable @reftable (which may be reallocated) at @reftable_index.
2483 *
2484 * @allocated should be set to true if a new cluster has been allocated.
2485 */
2486 typedef int (RefblockFinishOp)(BlockDriverState *bs, uint64_t **reftable,
2487 uint64_t reftable_index, uint64_t *reftable_size,
2488 void *refblock, bool refblock_empty,
2489 bool *allocated, Error **errp);
2490
2491 /**
2492 * This "operation" for walk_over_reftable() allocates the refblock on disk (if
2493 * it is not empty) and inserts its offset into the new reftable. The size of
2494 * this new reftable is increased as required.
2495 */
2496 static int alloc_refblock(BlockDriverState *bs, uint64_t **reftable,
2497 uint64_t reftable_index, uint64_t *reftable_size,
2498 void *refblock, bool refblock_empty, bool *allocated,
2499 Error **errp)
2500 {
2501 BDRVQcow2State *s = bs->opaque;
2502 int64_t offset;
2503
2504 if (!refblock_empty && reftable_index >= *reftable_size) {
2505 uint64_t *new_reftable;
2506 uint64_t new_reftable_size;
2507
2508 new_reftable_size = ROUND_UP(reftable_index + 1,
2509 s->cluster_size / sizeof(uint64_t));
2510 if (new_reftable_size > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) {
2511 error_setg(errp,
2512 "This operation would make the refcount table grow "
2513 "beyond the maximum size supported by QEMU, aborting");
2514 return -ENOTSUP;
2515 }
2516
2517 new_reftable = g_try_realloc(*reftable, new_reftable_size *
2518 sizeof(uint64_t));
2519 if (!new_reftable) {
2520 error_setg(errp, "Failed to increase reftable buffer size");
2521 return -ENOMEM;
2522 }
2523
2524 memset(new_reftable + *reftable_size, 0,
2525 (new_reftable_size - *reftable_size) * sizeof(uint64_t));
2526
2527 *reftable = new_reftable;
2528 *reftable_size = new_reftable_size;
2529 }
2530
2531 if (!refblock_empty && !(*reftable)[reftable_index]) {
2532 offset = qcow2_alloc_clusters(bs, s->cluster_size);
2533 if (offset < 0) {
2534 error_setg_errno(errp, -offset, "Failed to allocate refblock");
2535 return offset;
2536 }
2537 (*reftable)[reftable_index] = offset;
2538 *allocated = true;
2539 }
2540
2541 return 0;
2542 }
2543
2544 /**
2545 * This "operation" for walk_over_reftable() writes the refblock to disk at the
2546 * offset specified by the new reftable's entry. It does not modify the new
2547 * reftable or change any refcounts.
2548 */
2549 static int flush_refblock(BlockDriverState *bs, uint64_t **reftable,
2550 uint64_t reftable_index, uint64_t *reftable_size,
2551 void *refblock, bool refblock_empty, bool *allocated,
2552 Error **errp)
2553 {
2554 BDRVQcow2State *s = bs->opaque;
2555 int64_t offset;
2556 int ret;
2557
2558 if (reftable_index < *reftable_size && (*reftable)[reftable_index]) {
2559 offset = (*reftable)[reftable_index];
2560
2561 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
2562 if (ret < 0) {
2563 error_setg_errno(errp, -ret, "Overlap check failed");
2564 return ret;
2565 }
2566
2567 ret = bdrv_pwrite(bs->file->bs, offset, refblock, s->cluster_size);
2568 if (ret < 0) {
2569 error_setg_errno(errp, -ret, "Failed to write refblock");
2570 return ret;
2571 }
2572 } else {
2573 assert(refblock_empty);
2574 }
2575
2576 return 0;
2577 }
2578
2579 /**
2580 * This function walks over the existing reftable and every referenced refblock;
2581 * if @new_set_refcount is non-NULL, it is called for every refcount entry to
2582 * create an equal new entry in the passed @new_refblock. Once that
2583 * @new_refblock is completely filled, @operation will be called.
2584 *
2585 * @status_cb and @cb_opaque are used for the amend operation's status callback.
2586 * @index is the index of the walk_over_reftable() calls and @total is the total
2587 * number of walk_over_reftable() calls per amend operation. Both are used for
2588 * calculating the parameters for the status callback.
2589 *
2590 * @allocated is set to true if a new cluster has been allocated.
2591 */
2592 static int walk_over_reftable(BlockDriverState *bs, uint64_t **new_reftable,
2593 uint64_t *new_reftable_index,
2594 uint64_t *new_reftable_size,
2595 void *new_refblock, int new_refblock_size,
2596 int new_refcount_bits,
2597 RefblockFinishOp *operation, bool *allocated,
2598 Qcow2SetRefcountFunc *new_set_refcount,
2599 BlockDriverAmendStatusCB *status_cb,
2600 void *cb_opaque, int index, int total,
2601 Error **errp)
2602 {
2603 BDRVQcow2State *s = bs->opaque;
2604 uint64_t reftable_index;
2605 bool new_refblock_empty = true;
2606 int refblock_index;
2607 int new_refblock_index = 0;
2608 int ret;
2609
2610 for (reftable_index = 0; reftable_index < s->refcount_table_size;
2611 reftable_index++)
2612 {
2613 uint64_t refblock_offset = s->refcount_table[reftable_index]
2614 & REFT_OFFSET_MASK;
2615
2616 status_cb(bs, (uint64_t)index * s->refcount_table_size + reftable_index,
2617 (uint64_t)total * s->refcount_table_size, cb_opaque);
2618
2619 if (refblock_offset) {
2620 void *refblock;
2621
2622 if (offset_into_cluster(s, refblock_offset)) {
2623 qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#"
2624 PRIx64 " unaligned (reftable index: %#"
2625 PRIx64 ")", refblock_offset,
2626 reftable_index);
2627 error_setg(errp,
2628 "Image is corrupt (unaligned refblock offset)");
2629 return -EIO;
2630 }
2631
2632 ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offset,
2633 &refblock);
2634 if (ret < 0) {
2635 error_setg_errno(errp, -ret, "Failed to retrieve refblock");
2636 return ret;
2637 }
2638
2639 for (refblock_index = 0; refblock_index < s->refcount_block_size;
2640 refblock_index++)
2641 {
2642 uint64_t refcount;
2643
2644 if (new_refblock_index >= new_refblock_size) {
2645 /* new_refblock is now complete */
2646 ret = operation(bs, new_reftable, *new_reftable_index,
2647 new_reftable_size, new_refblock,
2648 new_refblock_empty, allocated, errp);
2649 if (ret < 0) {
2650 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2651 return ret;
2652 }
2653
2654 (*new_reftable_index)++;
2655 new_refblock_index = 0;
2656 new_refblock_empty = true;
2657 }
2658
2659 refcount = s->get_refcount(refblock, refblock_index);
2660 if (new_refcount_bits < 64 && refcount >> new_refcount_bits) {
2661 uint64_t offset;
2662
2663 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2664
2665 offset = ((reftable_index << s->refcount_block_bits)
2666 + refblock_index) << s->cluster_bits;
2667
2668 error_setg(errp, "Cannot decrease refcount entry width to "
2669 "%i bits: Cluster at offset %#" PRIx64 " has a "
2670 "refcount of %" PRIu64, new_refcount_bits,
2671 offset, refcount);
2672 return -EINVAL;
2673 }
2674
2675 if (new_set_refcount) {
2676 new_set_refcount(new_refblock, new_refblock_index++,
2677 refcount);
2678 } else {
2679 new_refblock_index++;
2680 }
2681 new_refblock_empty = new_refblock_empty && refcount == 0;
2682 }
2683
2684 qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
2685 } else {
2686 /* No refblock means every refcount is 0 */
2687 for (refblock_index = 0; refblock_index < s->refcount_block_size;
2688 refblock_index++)
2689 {
2690 if (new_refblock_index >= new_refblock_size) {
2691 /* new_refblock is now complete */
2692 ret = operation(bs, new_reftable, *new_reftable_index,
2693 new_reftable_size, new_refblock,
2694 new_refblock_empty, allocated, errp);
2695 if (ret < 0) {
2696 return ret;
2697 }
2698
2699 (*new_reftable_index)++;
2700 new_refblock_index = 0;
2701 new_refblock_empty = true;
2702 }
2703
2704 if (new_set_refcount) {
2705 new_set_refcount(new_refblock, new_refblock_index++, 0);
2706 } else {
2707 new_refblock_index++;
2708 }
2709 }
2710 }
2711 }
2712
2713 if (new_refblock_index > 0) {
2714 /* Complete the potentially existing partially filled final refblock */
2715 if (new_set_refcount) {
2716 for (; new_refblock_index < new_refblock_size;
2717 new_refblock_index++)
2718 {
2719 new_set_refcount(new_refblock, new_refblock_index, 0);
2720 }
2721 }
2722
2723 ret = operation(bs, new_reftable, *new_reftable_index,
2724 new_reftable_size, new_refblock, new_refblock_empty,
2725 allocated, errp);
2726 if (ret < 0) {
2727 return ret;
2728 }
2729
2730 (*new_reftable_index)++;
2731 }
2732
2733 status_cb(bs, (uint64_t)(index + 1) * s->refcount_table_size,
2734 (uint64_t)total * s->refcount_table_size, cb_opaque);
2735
2736 return 0;
2737 }
2738
2739 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order,
2740 BlockDriverAmendStatusCB *status_cb,
2741 void *cb_opaque, Error **errp)
2742 {
2743 BDRVQcow2State *s = bs->opaque;
2744 Qcow2GetRefcountFunc *new_get_refcount;
2745 Qcow2SetRefcountFunc *new_set_refcount;
2746 void *new_refblock = qemu_blockalign(bs->file->bs, s->cluster_size);
2747 uint64_t *new_reftable = NULL, new_reftable_size = 0;
2748 uint64_t *old_reftable, old_reftable_size, old_reftable_offset;
2749 uint64_t new_reftable_index = 0;
2750 uint64_t i;
2751 int64_t new_reftable_offset = 0, allocated_reftable_size = 0;
2752 int new_refblock_size, new_refcount_bits = 1 << refcount_order;
2753 int old_refcount_order;
2754 int walk_index = 0;
2755 int ret;
2756 bool new_allocation;
2757
2758 assert(s->qcow_version >= 3);
2759 assert(refcount_order >= 0 && refcount_order <= 6);
2760
2761 /* see qcow2_open() */
2762 new_refblock_size = 1 << (s->cluster_bits - (refcount_order - 3));
2763
2764 new_get_refcount = get_refcount_funcs[refcount_order];
2765 new_set_refcount = set_refcount_funcs[refcount_order];
2766
2767
2768 do {
2769 int total_walks;
2770
2771 new_allocation = false;
2772
2773 /* At least we have to do this walk and the one which writes the
2774 * refblocks; also, at least we have to do this loop here at least
2775 * twice (normally), first to do the allocations, and second to
2776 * determine that everything is correctly allocated, this then makes
2777 * three walks in total */
2778 total_walks = MAX(walk_index + 2, 3);
2779
2780 /* First, allocate the structures so they are present in the refcount
2781 * structures */
2782 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
2783 &new_reftable_size, NULL, new_refblock_size,
2784 new_refcount_bits, &alloc_refblock,
2785 &new_allocation, NULL, status_cb, cb_opaque,
2786 walk_index++, total_walks, errp);
2787 if (ret < 0) {
2788 goto done;
2789 }
2790
2791 new_reftable_index = 0;
2792
2793 if (new_allocation) {
2794 if (new_reftable_offset) {
2795 qcow2_free_clusters(bs, new_reftable_offset,
2796 allocated_reftable_size * sizeof(uint64_t),
2797 QCOW2_DISCARD_NEVER);
2798 }
2799
2800 new_reftable_offset = qcow2_alloc_clusters(bs, new_reftable_size *
2801 sizeof(uint64_t));
2802 if (new_reftable_offset < 0) {
2803 error_setg_errno(errp, -new_reftable_offset,
2804 "Failed to allocate the new reftable");
2805 ret = new_reftable_offset;
2806 goto done;
2807 }
2808 allocated_reftable_size = new_reftable_size;
2809 }
2810 } while (new_allocation);
2811
2812 /* Second, write the new refblocks */
2813 ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index,
2814 &new_reftable_size, new_refblock,
2815 new_refblock_size, new_refcount_bits,
2816 &flush_refblock, &new_allocation, new_set_refcount,
2817 status_cb, cb_opaque, walk_index, walk_index + 1,
2818 errp);
2819 if (ret < 0) {
2820 goto done;
2821 }
2822 assert(!new_allocation);
2823
2824
2825 /* Write the new reftable */
2826 ret = qcow2_pre_write_overlap_check(bs, 0, new_reftable_offset,
2827 new_reftable_size * sizeof(uint64_t));
2828 if (ret < 0) {
2829 error_setg_errno(errp, -ret, "Overlap check failed");
2830 goto done;
2831 }
2832
2833 for (i = 0; i < new_reftable_size; i++) {
2834 cpu_to_be64s(&new_reftable[i]);
2835 }
2836
2837 ret = bdrv_pwrite(bs->file->bs, new_reftable_offset, new_reftable,
2838 new_reftable_size * sizeof(uint64_t));
2839
2840 for (i = 0; i < new_reftable_size; i++) {
2841 be64_to_cpus(&new_reftable[i]);
2842 }
2843
2844 if (ret < 0) {
2845 error_setg_errno(errp, -ret, "Failed to write the new reftable");
2846 goto done;
2847 }
2848
2849
2850 /* Empty the refcount cache */
2851 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2852 if (ret < 0) {
2853 error_setg_errno(errp, -ret, "Failed to flush the refblock cache");
2854 goto done;
2855 }
2856
2857 /* Update the image header to point to the new reftable; this only updates
2858 * the fields which are relevant to qcow2_update_header(); other fields
2859 * such as s->refcount_table or s->refcount_bits stay stale for now
2860 * (because we have to restore everything if qcow2_update_header() fails) */
2861 old_refcount_order = s->refcount_order;
2862 old_reftable_size = s->refcount_table_size;
2863 old_reftable_offset = s->refcount_table_offset;
2864
2865 s->refcount_order = refcount_order;
2866 s->refcount_table_size = new_reftable_size;
2867 s->refcount_table_offset = new_reftable_offset;
2868
2869 ret = qcow2_update_header(bs);
2870 if (ret < 0) {
2871 s->refcount_order = old_refcount_order;
2872 s->refcount_table_size = old_reftable_size;
2873 s->refcount_table_offset = old_reftable_offset;
2874 error_setg_errno(errp, -ret, "Failed to update the qcow2 header");
2875 goto done;
2876 }
2877
2878 /* Now update the rest of the in-memory information */
2879 old_reftable = s->refcount_table;
2880 s->refcount_table = new_reftable;
2881
2882 s->refcount_bits = 1 << refcount_order;
2883 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
2884 s->refcount_max += s->refcount_max - 1;
2885
2886 s->refcount_block_bits = s->cluster_bits - (refcount_order - 3);
2887 s->refcount_block_size = 1 << s->refcount_block_bits;
2888
2889 s->get_refcount = new_get_refcount;
2890 s->set_refcount = new_set_refcount;
2891
2892 /* For cleaning up all old refblocks and the old reftable below the "done"
2893 * label */
2894 new_reftable = old_reftable;
2895 new_reftable_size = old_reftable_size;
2896 new_reftable_offset = old_reftable_offset;
2897
2898 done:
2899 if (new_reftable) {
2900 /* On success, new_reftable actually points to the old reftable (and
2901 * new_reftable_size is the old reftable's size); but that is just
2902 * fine */
2903 for (i = 0; i < new_reftable_size; i++) {
2904 uint64_t offset = new_reftable[i] & REFT_OFFSET_MASK;
2905 if (offset) {
2906 qcow2_free_clusters(bs, offset, s->cluster_size,
2907 QCOW2_DISCARD_OTHER);
2908 }
2909 }
2910 g_free(new_reftable);
2911
2912 if (new_reftable_offset > 0) {
2913 qcow2_free_clusters(bs, new_reftable_offset,
2914 new_reftable_size * sizeof(uint64_t),
2915 QCOW2_DISCARD_OTHER);
2916 }
2917 }
2918
2919 qemu_vfree(new_refblock);
2920 return ret;
2921 }