root / block / qcow2-refcount.c @ 5d757b56
History | View | Annotate | Download (30 kB)
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-common.h" |
26 |
#include "block_int.h" |
27 |
#include "block/qcow2.h" |
28 |
|
29 |
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
|
30 |
static int update_refcount(BlockDriverState *bs, |
31 |
int64_t offset, int64_t length, |
32 |
int addend);
|
33 |
|
34 |
|
35 |
static int cache_refcount_updates = 0; |
36 |
|
37 |
static int write_refcount_block(BDRVQcowState *s) |
38 |
{ |
39 |
size_t size = s->cluster_size; |
40 |
|
41 |
if (s->refcount_block_cache_offset == 0) { |
42 |
return 0; |
43 |
} |
44 |
|
45 |
if (bdrv_pwrite(s->hd, s->refcount_block_cache_offset,
|
46 |
s->refcount_block_cache, size) != size) |
47 |
{ |
48 |
return -EIO;
|
49 |
} |
50 |
|
51 |
return 0; |
52 |
} |
53 |
|
54 |
/*********************************************************/
|
55 |
/* refcount handling */
|
56 |
|
57 |
int qcow2_refcount_init(BlockDriverState *bs)
|
58 |
{ |
59 |
BDRVQcowState *s = bs->opaque; |
60 |
int ret, refcount_table_size2, i;
|
61 |
|
62 |
s->refcount_block_cache = qemu_malloc(s->cluster_size); |
63 |
refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
|
64 |
s->refcount_table = qemu_malloc(refcount_table_size2); |
65 |
if (s->refcount_table_size > 0) { |
66 |
ret = bdrv_pread(s->hd, s->refcount_table_offset, |
67 |
s->refcount_table, refcount_table_size2); |
68 |
if (ret != refcount_table_size2)
|
69 |
goto fail;
|
70 |
for(i = 0; i < s->refcount_table_size; i++) |
71 |
be64_to_cpus(&s->refcount_table[i]); |
72 |
} |
73 |
return 0; |
74 |
fail:
|
75 |
return -ENOMEM;
|
76 |
} |
77 |
|
78 |
void qcow2_refcount_close(BlockDriverState *bs)
|
79 |
{ |
80 |
BDRVQcowState *s = bs->opaque; |
81 |
qemu_free(s->refcount_block_cache); |
82 |
qemu_free(s->refcount_table); |
83 |
} |
84 |
|
85 |
|
86 |
static int load_refcount_block(BlockDriverState *bs, |
87 |
int64_t refcount_block_offset) |
88 |
{ |
89 |
BDRVQcowState *s = bs->opaque; |
90 |
int ret;
|
91 |
|
92 |
if (cache_refcount_updates) {
|
93 |
write_refcount_block(s); |
94 |
} |
95 |
|
96 |
ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache, |
97 |
s->cluster_size); |
98 |
if (ret != s->cluster_size)
|
99 |
return -EIO;
|
100 |
s->refcount_block_cache_offset = refcount_block_offset; |
101 |
return 0; |
102 |
} |
103 |
|
104 |
static int get_refcount(BlockDriverState *bs, int64_t cluster_index) |
105 |
{ |
106 |
BDRVQcowState *s = bs->opaque; |
107 |
int refcount_table_index, block_index;
|
108 |
int64_t refcount_block_offset; |
109 |
|
110 |
refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); |
111 |
if (refcount_table_index >= s->refcount_table_size)
|
112 |
return 0; |
113 |
refcount_block_offset = s->refcount_table[refcount_table_index]; |
114 |
if (!refcount_block_offset)
|
115 |
return 0; |
116 |
if (refcount_block_offset != s->refcount_block_cache_offset) {
|
117 |
/* better than nothing: return allocated if read error */
|
118 |
if (load_refcount_block(bs, refcount_block_offset) < 0) |
119 |
return 1; |
120 |
} |
121 |
block_index = cluster_index & |
122 |
((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); |
123 |
return be16_to_cpu(s->refcount_block_cache[block_index]);
|
124 |
} |
125 |
|
126 |
static int grow_refcount_table(BlockDriverState *bs, int min_size) |
127 |
{ |
128 |
BDRVQcowState *s = bs->opaque; |
129 |
int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
|
130 |
uint64_t *new_table; |
131 |
int64_t table_offset; |
132 |
uint8_t data[12];
|
133 |
int old_table_size;
|
134 |
int64_t old_table_offset; |
135 |
|
136 |
if (min_size <= s->refcount_table_size)
|
137 |
return 0; |
138 |
/* compute new table size */
|
139 |
refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
|
140 |
for(;;) {
|
141 |
if (refcount_table_clusters == 0) { |
142 |
refcount_table_clusters = 1;
|
143 |
} else {
|
144 |
refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2; |
145 |
} |
146 |
new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
|
147 |
if (min_size <= new_table_size)
|
148 |
break;
|
149 |
} |
150 |
#ifdef DEBUG_ALLOC2
|
151 |
printf("grow_refcount_table from %d to %d\n",
|
152 |
s->refcount_table_size, |
153 |
new_table_size); |
154 |
#endif
|
155 |
new_table_size2 = new_table_size * sizeof(uint64_t);
|
156 |
new_table = qemu_mallocz(new_table_size2); |
157 |
memcpy(new_table, s->refcount_table, |
158 |
s->refcount_table_size * sizeof(uint64_t));
|
159 |
for(i = 0; i < s->refcount_table_size; i++) |
160 |
cpu_to_be64s(&new_table[i]); |
161 |
/* Note: we cannot update the refcount now to avoid recursion */
|
162 |
table_offset = alloc_clusters_noref(bs, new_table_size2); |
163 |
ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2); |
164 |
if (ret != new_table_size2)
|
165 |
goto fail;
|
166 |
for(i = 0; i < s->refcount_table_size; i++) |
167 |
be64_to_cpus(&new_table[i]); |
168 |
|
169 |
cpu_to_be64w((uint64_t*)data, table_offset); |
170 |
cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
|
171 |
ret = bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset), |
172 |
data, sizeof(data));
|
173 |
if (ret != sizeof(data)) { |
174 |
goto fail;
|
175 |
} |
176 |
|
177 |
qemu_free(s->refcount_table); |
178 |
old_table_offset = s->refcount_table_offset; |
179 |
old_table_size = s->refcount_table_size; |
180 |
s->refcount_table = new_table; |
181 |
s->refcount_table_size = new_table_size; |
182 |
s->refcount_table_offset = table_offset; |
183 |
|
184 |
update_refcount(bs, table_offset, new_table_size2, 1);
|
185 |
qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
|
186 |
return 0; |
187 |
fail:
|
188 |
qemu_free(new_table); |
189 |
return ret < 0 ? ret : -EIO; |
190 |
} |
191 |
|
192 |
|
193 |
static int64_t alloc_refcount_block(BlockDriverState *bs, int64_t cluster_index)
|
194 |
{ |
195 |
BDRVQcowState *s = bs->opaque; |
196 |
int64_t offset, refcount_block_offset; |
197 |
unsigned int refcount_table_index; |
198 |
int ret;
|
199 |
uint64_t data64; |
200 |
int cache = cache_refcount_updates;
|
201 |
|
202 |
/* Find L1 index and grow refcount table if needed */
|
203 |
refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); |
204 |
if (refcount_table_index >= s->refcount_table_size) {
|
205 |
ret = grow_refcount_table(bs, refcount_table_index + 1);
|
206 |
if (ret < 0) |
207 |
return ret;
|
208 |
} |
209 |
|
210 |
/* Load or allocate the refcount block */
|
211 |
refcount_block_offset = s->refcount_table[refcount_table_index]; |
212 |
if (!refcount_block_offset) {
|
213 |
if (cache_refcount_updates) {
|
214 |
write_refcount_block(s); |
215 |
cache_refcount_updates = 0;
|
216 |
} |
217 |
/* create a new refcount block */
|
218 |
/* Note: we cannot update the refcount now to avoid recursion */
|
219 |
offset = alloc_clusters_noref(bs, s->cluster_size); |
220 |
memset(s->refcount_block_cache, 0, s->cluster_size);
|
221 |
ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size); |
222 |
if (ret != s->cluster_size)
|
223 |
return -EINVAL;
|
224 |
s->refcount_table[refcount_table_index] = offset; |
225 |
data64 = cpu_to_be64(offset); |
226 |
ret = bdrv_pwrite(s->hd, s->refcount_table_offset + |
227 |
refcount_table_index * sizeof(uint64_t),
|
228 |
&data64, sizeof(data64));
|
229 |
if (ret != sizeof(data64)) |
230 |
return -EINVAL;
|
231 |
|
232 |
refcount_block_offset = offset; |
233 |
s->refcount_block_cache_offset = offset; |
234 |
update_refcount(bs, offset, s->cluster_size, 1);
|
235 |
cache_refcount_updates = cache; |
236 |
} else {
|
237 |
if (refcount_block_offset != s->refcount_block_cache_offset) {
|
238 |
if (load_refcount_block(bs, refcount_block_offset) < 0) |
239 |
return -EIO;
|
240 |
} |
241 |
} |
242 |
|
243 |
return refcount_block_offset;
|
244 |
} |
245 |
|
246 |
#define REFCOUNTS_PER_SECTOR (512 >> REFCOUNT_SHIFT) |
247 |
static int write_refcount_block_entries(BDRVQcowState *s, |
248 |
int64_t refcount_block_offset, int first_index, int last_index) |
249 |
{ |
250 |
size_t size; |
251 |
|
252 |
if (cache_refcount_updates) {
|
253 |
return 0; |
254 |
} |
255 |
|
256 |
first_index &= ~(REFCOUNTS_PER_SECTOR - 1);
|
257 |
last_index = (last_index + REFCOUNTS_PER_SECTOR) |
258 |
& ~(REFCOUNTS_PER_SECTOR - 1);
|
259 |
|
260 |
size = (last_index - first_index) << REFCOUNT_SHIFT; |
261 |
if (bdrv_pwrite(s->hd,
|
262 |
refcount_block_offset + (first_index << REFCOUNT_SHIFT), |
263 |
&s->refcount_block_cache[first_index], size) != size) |
264 |
{ |
265 |
return -EIO;
|
266 |
} |
267 |
|
268 |
return 0; |
269 |
} |
270 |
|
271 |
/* XXX: cache several refcount block clusters ? */
|
272 |
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, |
273 |
int64_t offset, int64_t length, int addend)
|
274 |
{ |
275 |
BDRVQcowState *s = bs->opaque; |
276 |
int64_t start, last, cluster_offset; |
277 |
int64_t refcount_block_offset = 0;
|
278 |
int64_t table_index = -1, old_table_index;
|
279 |
int first_index = -1, last_index = -1; |
280 |
int ret;
|
281 |
|
282 |
#ifdef DEBUG_ALLOC2
|
283 |
printf("update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n", |
284 |
offset, length, addend); |
285 |
#endif
|
286 |
if (length < 0) { |
287 |
return -EINVAL;
|
288 |
} else if (length == 0) { |
289 |
return 0; |
290 |
} |
291 |
|
292 |
start = offset & ~(s->cluster_size - 1);
|
293 |
last = (offset + length - 1) & ~(s->cluster_size - 1); |
294 |
for(cluster_offset = start; cluster_offset <= last;
|
295 |
cluster_offset += s->cluster_size) |
296 |
{ |
297 |
int block_index, refcount;
|
298 |
int64_t cluster_index = cluster_offset >> s->cluster_bits; |
299 |
int64_t new_block; |
300 |
|
301 |
/* Only write refcount block to disk when we are done with it */
|
302 |
old_table_index = table_index; |
303 |
table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); |
304 |
if ((old_table_index >= 0) && (table_index != old_table_index)) { |
305 |
|
306 |
if (write_refcount_block_entries(s, refcount_block_offset,
|
307 |
first_index, last_index) < 0)
|
308 |
{ |
309 |
return -EIO;
|
310 |
} |
311 |
|
312 |
first_index = -1;
|
313 |
last_index = -1;
|
314 |
} |
315 |
|
316 |
/* Load the refcount block and allocate it if needed */
|
317 |
new_block = alloc_refcount_block(bs, cluster_index); |
318 |
if (new_block < 0) { |
319 |
ret = new_block; |
320 |
goto fail;
|
321 |
} |
322 |
refcount_block_offset = new_block; |
323 |
|
324 |
/* we can update the count and save it */
|
325 |
block_index = cluster_index & |
326 |
((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); |
327 |
if (first_index == -1 || block_index < first_index) { |
328 |
first_index = block_index; |
329 |
} |
330 |
if (block_index > last_index) {
|
331 |
last_index = block_index; |
332 |
} |
333 |
|
334 |
refcount = be16_to_cpu(s->refcount_block_cache[block_index]); |
335 |
refcount += addend; |
336 |
if (refcount < 0 || refcount > 0xffff) { |
337 |
ret = -EINVAL; |
338 |
goto fail;
|
339 |
} |
340 |
if (refcount == 0 && cluster_index < s->free_cluster_index) { |
341 |
s->free_cluster_index = cluster_index; |
342 |
} |
343 |
s->refcount_block_cache[block_index] = cpu_to_be16(refcount); |
344 |
} |
345 |
|
346 |
ret = 0;
|
347 |
fail:
|
348 |
|
349 |
/* Write last changed block to disk */
|
350 |
if (refcount_block_offset != 0) { |
351 |
if (write_refcount_block_entries(s, refcount_block_offset,
|
352 |
first_index, last_index) < 0)
|
353 |
{ |
354 |
return ret < 0 ? ret : -EIO; |
355 |
} |
356 |
} |
357 |
|
358 |
/*
|
359 |
* Try do undo any updates if an error is returned (This may succeed in
|
360 |
* some cases like ENOSPC for allocating a new refcount block)
|
361 |
*/
|
362 |
if (ret < 0) { |
363 |
int dummy;
|
364 |
dummy = update_refcount(bs, offset, cluster_offset - offset, -addend); |
365 |
} |
366 |
|
367 |
return ret;
|
368 |
} |
369 |
|
370 |
/* addend must be 1 or -1 */
|
371 |
static int update_cluster_refcount(BlockDriverState *bs, |
372 |
int64_t cluster_index, |
373 |
int addend)
|
374 |
{ |
375 |
BDRVQcowState *s = bs->opaque; |
376 |
int ret;
|
377 |
|
378 |
ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
|
379 |
if (ret < 0) { |
380 |
return ret;
|
381 |
} |
382 |
|
383 |
return get_refcount(bs, cluster_index);
|
384 |
} |
385 |
|
386 |
|
387 |
|
388 |
/*********************************************************/
|
389 |
/* cluster allocation functions */
|
390 |
|
391 |
|
392 |
|
393 |
/* return < 0 if error */
|
394 |
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
|
395 |
{ |
396 |
BDRVQcowState *s = bs->opaque; |
397 |
int i, nb_clusters;
|
398 |
|
399 |
nb_clusters = size_to_clusters(s, size); |
400 |
retry:
|
401 |
for(i = 0; i < nb_clusters; i++) { |
402 |
int64_t i = s->free_cluster_index++; |
403 |
if (get_refcount(bs, i) != 0) |
404 |
goto retry;
|
405 |
} |
406 |
#ifdef DEBUG_ALLOC2
|
407 |
printf("alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n", |
408 |
size, |
409 |
(s->free_cluster_index - nb_clusters) << s->cluster_bits); |
410 |
#endif
|
411 |
return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
|
412 |
} |
413 |
|
414 |
int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size) |
415 |
{ |
416 |
int64_t offset; |
417 |
int ret;
|
418 |
|
419 |
offset = alloc_clusters_noref(bs, size); |
420 |
ret = update_refcount(bs, offset, size, 1);
|
421 |
if (ret < 0) { |
422 |
return ret;
|
423 |
} |
424 |
return offset;
|
425 |
} |
426 |
|
427 |
/* only used to allocate compressed sectors. We try to allocate
|
428 |
contiguous sectors. size must be <= cluster_size */
|
429 |
int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
|
430 |
{ |
431 |
BDRVQcowState *s = bs->opaque; |
432 |
int64_t offset, cluster_offset; |
433 |
int free_in_cluster;
|
434 |
|
435 |
assert(size > 0 && size <= s->cluster_size);
|
436 |
if (s->free_byte_offset == 0) { |
437 |
s->free_byte_offset = qcow2_alloc_clusters(bs, s->cluster_size); |
438 |
if (s->free_byte_offset < 0) { |
439 |
return s->free_byte_offset;
|
440 |
} |
441 |
} |
442 |
redo:
|
443 |
free_in_cluster = s->cluster_size - |
444 |
(s->free_byte_offset & (s->cluster_size - 1));
|
445 |
if (size <= free_in_cluster) {
|
446 |
/* enough space in current cluster */
|
447 |
offset = s->free_byte_offset; |
448 |
s->free_byte_offset += size; |
449 |
free_in_cluster -= size; |
450 |
if (free_in_cluster == 0) |
451 |
s->free_byte_offset = 0;
|
452 |
if ((offset & (s->cluster_size - 1)) != 0) |
453 |
update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
|
454 |
} else {
|
455 |
offset = qcow2_alloc_clusters(bs, s->cluster_size); |
456 |
if (offset < 0) { |
457 |
return offset;
|
458 |
} |
459 |
cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
|
460 |
if ((cluster_offset + s->cluster_size) == offset) {
|
461 |
/* we are lucky: contiguous data */
|
462 |
offset = s->free_byte_offset; |
463 |
update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
|
464 |
s->free_byte_offset += size; |
465 |
} else {
|
466 |
s->free_byte_offset = offset; |
467 |
goto redo;
|
468 |
} |
469 |
} |
470 |
return offset;
|
471 |
} |
472 |
|
473 |
void qcow2_free_clusters(BlockDriverState *bs,
|
474 |
int64_t offset, int64_t size) |
475 |
{ |
476 |
int ret;
|
477 |
|
478 |
ret = update_refcount(bs, offset, size, -1);
|
479 |
if (ret < 0) { |
480 |
fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
|
481 |
abort(); |
482 |
} |
483 |
} |
484 |
|
485 |
/*
|
486 |
* free_any_clusters
|
487 |
*
|
488 |
* free clusters according to its type: compressed or not
|
489 |
*
|
490 |
*/
|
491 |
|
492 |
void qcow2_free_any_clusters(BlockDriverState *bs,
|
493 |
uint64_t cluster_offset, int nb_clusters)
|
494 |
{ |
495 |
BDRVQcowState *s = bs->opaque; |
496 |
|
497 |
/* free the cluster */
|
498 |
|
499 |
if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
|
500 |
int nb_csectors;
|
501 |
nb_csectors = ((cluster_offset >> s->csize_shift) & |
502 |
s->csize_mask) + 1;
|
503 |
qcow2_free_clusters(bs, |
504 |
(cluster_offset & s->cluster_offset_mask) & ~511,
|
505 |
nb_csectors * 512);
|
506 |
return;
|
507 |
} |
508 |
|
509 |
qcow2_free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits); |
510 |
|
511 |
return;
|
512 |
} |
513 |
|
514 |
|
515 |
|
516 |
/*********************************************************/
|
517 |
/* snapshots and image creation */
|
518 |
|
519 |
|
520 |
|
521 |
void qcow2_create_refcount_update(QCowCreateState *s, int64_t offset,
|
522 |
int64_t size) |
523 |
{ |
524 |
int refcount;
|
525 |
int64_t start, last, cluster_offset; |
526 |
uint16_t *p; |
527 |
|
528 |
start = offset & ~(s->cluster_size - 1);
|
529 |
last = (offset + size - 1) & ~(s->cluster_size - 1); |
530 |
for(cluster_offset = start; cluster_offset <= last;
|
531 |
cluster_offset += s->cluster_size) { |
532 |
p = &s->refcount_block[cluster_offset >> s->cluster_bits]; |
533 |
refcount = be16_to_cpu(*p); |
534 |
refcount++; |
535 |
*p = cpu_to_be16(refcount); |
536 |
} |
537 |
} |
538 |
|
539 |
/* update the refcounts of snapshots and the copied flag */
|
540 |
int qcow2_update_snapshot_refcount(BlockDriverState *bs,
|
541 |
int64_t l1_table_offset, int l1_size, int addend) |
542 |
{ |
543 |
BDRVQcowState *s = bs->opaque; |
544 |
uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated; |
545 |
int64_t old_offset, old_l2_offset; |
546 |
int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
|
547 |
|
548 |
qcow2_l2_cache_reset(bs); |
549 |
cache_refcount_updates = 1;
|
550 |
|
551 |
l2_table = NULL;
|
552 |
l1_table = NULL;
|
553 |
l1_size2 = l1_size * sizeof(uint64_t);
|
554 |
if (l1_table_offset != s->l1_table_offset) {
|
555 |
if (l1_size2 != 0) { |
556 |
l1_table = qemu_mallocz(align_offset(l1_size2, 512));
|
557 |
} else {
|
558 |
l1_table = NULL;
|
559 |
} |
560 |
l1_allocated = 1;
|
561 |
if (bdrv_pread(s->hd, l1_table_offset,
|
562 |
l1_table, l1_size2) != l1_size2) |
563 |
goto fail;
|
564 |
for(i = 0;i < l1_size; i++) |
565 |
be64_to_cpus(&l1_table[i]); |
566 |
} else {
|
567 |
assert(l1_size == s->l1_size); |
568 |
l1_table = s->l1_table; |
569 |
l1_allocated = 0;
|
570 |
} |
571 |
|
572 |
l2_size = s->l2_size * sizeof(uint64_t);
|
573 |
l2_table = qemu_malloc(l2_size); |
574 |
l1_modified = 0;
|
575 |
for(i = 0; i < l1_size; i++) { |
576 |
l2_offset = l1_table[i]; |
577 |
if (l2_offset) {
|
578 |
old_l2_offset = l2_offset; |
579 |
l2_offset &= ~QCOW_OFLAG_COPIED; |
580 |
l2_modified = 0;
|
581 |
if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
|
582 |
goto fail;
|
583 |
for(j = 0; j < s->l2_size; j++) { |
584 |
offset = be64_to_cpu(l2_table[j]); |
585 |
if (offset != 0) { |
586 |
old_offset = offset; |
587 |
offset &= ~QCOW_OFLAG_COPIED; |
588 |
if (offset & QCOW_OFLAG_COMPRESSED) {
|
589 |
nb_csectors = ((offset >> s->csize_shift) & |
590 |
s->csize_mask) + 1;
|
591 |
if (addend != 0) { |
592 |
int ret;
|
593 |
ret = update_refcount(bs, |
594 |
(offset & s->cluster_offset_mask) & ~511,
|
595 |
nb_csectors * 512, addend);
|
596 |
if (ret < 0) { |
597 |
goto fail;
|
598 |
} |
599 |
} |
600 |
/* compressed clusters are never modified */
|
601 |
refcount = 2;
|
602 |
} else {
|
603 |
if (addend != 0) { |
604 |
refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend); |
605 |
} else {
|
606 |
refcount = get_refcount(bs, offset >> s->cluster_bits); |
607 |
} |
608 |
} |
609 |
|
610 |
if (refcount == 1) { |
611 |
offset |= QCOW_OFLAG_COPIED; |
612 |
} |
613 |
if (offset != old_offset) {
|
614 |
l2_table[j] = cpu_to_be64(offset); |
615 |
l2_modified = 1;
|
616 |
} |
617 |
} |
618 |
} |
619 |
if (l2_modified) {
|
620 |
if (bdrv_pwrite(s->hd,
|
621 |
l2_offset, l2_table, l2_size) != l2_size) |
622 |
goto fail;
|
623 |
} |
624 |
|
625 |
if (addend != 0) { |
626 |
refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend); |
627 |
} else {
|
628 |
refcount = get_refcount(bs, l2_offset >> s->cluster_bits); |
629 |
} |
630 |
if (refcount == 1) { |
631 |
l2_offset |= QCOW_OFLAG_COPIED; |
632 |
} |
633 |
if (l2_offset != old_l2_offset) {
|
634 |
l1_table[i] = l2_offset; |
635 |
l1_modified = 1;
|
636 |
} |
637 |
} |
638 |
} |
639 |
if (l1_modified) {
|
640 |
for(i = 0; i < l1_size; i++) |
641 |
cpu_to_be64s(&l1_table[i]); |
642 |
if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
|
643 |
l1_size2) != l1_size2) |
644 |
goto fail;
|
645 |
for(i = 0; i < l1_size; i++) |
646 |
be64_to_cpus(&l1_table[i]); |
647 |
} |
648 |
if (l1_allocated)
|
649 |
qemu_free(l1_table); |
650 |
qemu_free(l2_table); |
651 |
cache_refcount_updates = 0;
|
652 |
write_refcount_block(s); |
653 |
return 0; |
654 |
fail:
|
655 |
if (l1_allocated)
|
656 |
qemu_free(l1_table); |
657 |
qemu_free(l2_table); |
658 |
cache_refcount_updates = 0;
|
659 |
write_refcount_block(s); |
660 |
return -EIO;
|
661 |
} |
662 |
|
663 |
|
664 |
|
665 |
|
666 |
/*********************************************************/
|
667 |
/* refcount checking functions */
|
668 |
|
669 |
|
670 |
|
671 |
/*
|
672 |
* Increases the refcount for a range of clusters in a given refcount table.
|
673 |
* This is used to construct a temporary refcount table out of L1 and L2 tables
|
674 |
* which can be compared the the refcount table saved in the image.
|
675 |
*
|
676 |
* Returns the number of errors in the image that were found
|
677 |
*/
|
678 |
static int inc_refcounts(BlockDriverState *bs, |
679 |
uint16_t *refcount_table, |
680 |
int refcount_table_size,
|
681 |
int64_t offset, int64_t size) |
682 |
{ |
683 |
BDRVQcowState *s = bs->opaque; |
684 |
int64_t start, last, cluster_offset; |
685 |
int k;
|
686 |
int errors = 0; |
687 |
|
688 |
if (size <= 0) |
689 |
return 0; |
690 |
|
691 |
start = offset & ~(s->cluster_size - 1);
|
692 |
last = (offset + size - 1) & ~(s->cluster_size - 1); |
693 |
for(cluster_offset = start; cluster_offset <= last;
|
694 |
cluster_offset += s->cluster_size) { |
695 |
k = cluster_offset >> s->cluster_bits; |
696 |
if (k < 0 || k >= refcount_table_size) { |
697 |
fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n", |
698 |
cluster_offset); |
699 |
errors++; |
700 |
} else {
|
701 |
if (++refcount_table[k] == 0) { |
702 |
fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
|
703 |
"\n", cluster_offset);
|
704 |
errors++; |
705 |
} |
706 |
} |
707 |
} |
708 |
|
709 |
return errors;
|
710 |
} |
711 |
|
712 |
/*
|
713 |
* Increases the refcount in the given refcount table for the all clusters
|
714 |
* referenced in the L2 table. While doing so, performs some checks on L2
|
715 |
* entries.
|
716 |
*
|
717 |
* Returns the number of errors found by the checks or -errno if an internal
|
718 |
* error occurred.
|
719 |
*/
|
720 |
static int check_refcounts_l2(BlockDriverState *bs, |
721 |
uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
|
722 |
int check_copied)
|
723 |
{ |
724 |
BDRVQcowState *s = bs->opaque; |
725 |
uint64_t *l2_table, offset; |
726 |
int i, l2_size, nb_csectors, refcount;
|
727 |
int errors = 0; |
728 |
|
729 |
/* Read L2 table from disk */
|
730 |
l2_size = s->l2_size * sizeof(uint64_t);
|
731 |
l2_table = qemu_malloc(l2_size); |
732 |
|
733 |
if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
|
734 |
goto fail;
|
735 |
|
736 |
/* Do the actual checks */
|
737 |
for(i = 0; i < s->l2_size; i++) { |
738 |
offset = be64_to_cpu(l2_table[i]); |
739 |
if (offset != 0) { |
740 |
if (offset & QCOW_OFLAG_COMPRESSED) {
|
741 |
/* Compressed clusters don't have QCOW_OFLAG_COPIED */
|
742 |
if (offset & QCOW_OFLAG_COPIED) {
|
743 |
fprintf(stderr, "ERROR: cluster %" PRId64 ": " |
744 |
"copied flag must never be set for compressed "
|
745 |
"clusters\n", offset >> s->cluster_bits);
|
746 |
offset &= ~QCOW_OFLAG_COPIED; |
747 |
errors++; |
748 |
} |
749 |
|
750 |
/* Mark cluster as used */
|
751 |
nb_csectors = ((offset >> s->csize_shift) & |
752 |
s->csize_mask) + 1;
|
753 |
offset &= s->cluster_offset_mask; |
754 |
errors += inc_refcounts(bs, refcount_table, |
755 |
refcount_table_size, |
756 |
offset & ~511, nb_csectors * 512); |
757 |
} else {
|
758 |
/* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
|
759 |
if (check_copied) {
|
760 |
uint64_t entry = offset; |
761 |
offset &= ~QCOW_OFLAG_COPIED; |
762 |
refcount = get_refcount(bs, offset >> s->cluster_bits); |
763 |
if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) { |
764 |
fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
|
765 |
PRIx64 " refcount=%d\n", entry, refcount);
|
766 |
errors++; |
767 |
} |
768 |
} |
769 |
|
770 |
/* Mark cluster as used */
|
771 |
offset &= ~QCOW_OFLAG_COPIED; |
772 |
errors += inc_refcounts(bs, refcount_table, |
773 |
refcount_table_size, |
774 |
offset, s->cluster_size); |
775 |
|
776 |
/* Correct offsets are cluster aligned */
|
777 |
if (offset & (s->cluster_size - 1)) { |
778 |
fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not " |
779 |
"properly aligned; L2 entry corrupted.\n", offset);
|
780 |
errors++; |
781 |
} |
782 |
} |
783 |
} |
784 |
} |
785 |
|
786 |
qemu_free(l2_table); |
787 |
return errors;
|
788 |
|
789 |
fail:
|
790 |
fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
|
791 |
qemu_free(l2_table); |
792 |
return -EIO;
|
793 |
} |
794 |
|
795 |
/*
|
796 |
* Increases the refcount for the L1 table, its L2 tables and all referenced
|
797 |
* clusters in the given refcount table. While doing so, performs some checks
|
798 |
* on L1 and L2 entries.
|
799 |
*
|
800 |
* Returns the number of errors found by the checks or -errno if an internal
|
801 |
* error occurred.
|
802 |
*/
|
803 |
static int check_refcounts_l1(BlockDriverState *bs, |
804 |
uint16_t *refcount_table, |
805 |
int refcount_table_size,
|
806 |
int64_t l1_table_offset, int l1_size,
|
807 |
int check_copied)
|
808 |
{ |
809 |
BDRVQcowState *s = bs->opaque; |
810 |
uint64_t *l1_table, l2_offset, l1_size2; |
811 |
int i, refcount, ret;
|
812 |
int errors = 0; |
813 |
|
814 |
l1_size2 = l1_size * sizeof(uint64_t);
|
815 |
|
816 |
/* Mark L1 table as used */
|
817 |
errors += inc_refcounts(bs, refcount_table, refcount_table_size, |
818 |
l1_table_offset, l1_size2); |
819 |
|
820 |
/* Read L1 table entries from disk */
|
821 |
if (l1_size2 == 0) { |
822 |
l1_table = NULL;
|
823 |
} else {
|
824 |
l1_table = qemu_malloc(l1_size2); |
825 |
if (bdrv_pread(s->hd, l1_table_offset,
|
826 |
l1_table, l1_size2) != l1_size2) |
827 |
goto fail;
|
828 |
for(i = 0;i < l1_size; i++) |
829 |
be64_to_cpus(&l1_table[i]); |
830 |
} |
831 |
|
832 |
/* Do the actual checks */
|
833 |
for(i = 0; i < l1_size; i++) { |
834 |
l2_offset = l1_table[i]; |
835 |
if (l2_offset) {
|
836 |
/* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
|
837 |
if (check_copied) {
|
838 |
refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) |
839 |
>> s->cluster_bits); |
840 |
if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) { |
841 |
fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
|
842 |
" refcount=%d\n", l2_offset, refcount);
|
843 |
errors++; |
844 |
} |
845 |
} |
846 |
|
847 |
/* Mark L2 table as used */
|
848 |
l2_offset &= ~QCOW_OFLAG_COPIED; |
849 |
errors += inc_refcounts(bs, refcount_table, |
850 |
refcount_table_size, |
851 |
l2_offset, |
852 |
s->cluster_size); |
853 |
|
854 |
/* L2 tables are cluster aligned */
|
855 |
if (l2_offset & (s->cluster_size - 1)) { |
856 |
fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not " |
857 |
"cluster aligned; L1 entry corrupted\n", l2_offset);
|
858 |
errors++; |
859 |
} |
860 |
|
861 |
/* Process and check L2 entries */
|
862 |
ret = check_refcounts_l2(bs, refcount_table, refcount_table_size, |
863 |
l2_offset, check_copied); |
864 |
if (ret < 0) { |
865 |
goto fail;
|
866 |
} |
867 |
errors += ret; |
868 |
} |
869 |
} |
870 |
qemu_free(l1_table); |
871 |
return errors;
|
872 |
|
873 |
fail:
|
874 |
fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
|
875 |
qemu_free(l1_table); |
876 |
return -EIO;
|
877 |
} |
878 |
|
879 |
/*
|
880 |
* Checks an image for refcount consistency.
|
881 |
*
|
882 |
* Returns 0 if no errors are found, the number of errors in case the image is
|
883 |
* detected as corrupted, and -errno when an internal error occured.
|
884 |
*/
|
885 |
int qcow2_check_refcounts(BlockDriverState *bs)
|
886 |
{ |
887 |
BDRVQcowState *s = bs->opaque; |
888 |
int64_t size; |
889 |
int nb_clusters, refcount1, refcount2, i;
|
890 |
QCowSnapshot *sn; |
891 |
uint16_t *refcount_table; |
892 |
int ret, errors = 0; |
893 |
|
894 |
size = bdrv_getlength(s->hd); |
895 |
nb_clusters = size_to_clusters(s, size); |
896 |
refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
|
897 |
|
898 |
/* header */
|
899 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
900 |
0, s->cluster_size);
|
901 |
|
902 |
/* current L1 table */
|
903 |
ret = check_refcounts_l1(bs, refcount_table, nb_clusters, |
904 |
s->l1_table_offset, s->l1_size, 1);
|
905 |
if (ret < 0) { |
906 |
return ret;
|
907 |
} |
908 |
errors += ret; |
909 |
|
910 |
/* snapshots */
|
911 |
for(i = 0; i < s->nb_snapshots; i++) { |
912 |
sn = s->snapshots + i; |
913 |
check_refcounts_l1(bs, refcount_table, nb_clusters, |
914 |
sn->l1_table_offset, sn->l1_size, 0);
|
915 |
} |
916 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
917 |
s->snapshots_offset, s->snapshots_size); |
918 |
|
919 |
/* refcount data */
|
920 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
921 |
s->refcount_table_offset, |
922 |
s->refcount_table_size * sizeof(uint64_t));
|
923 |
for(i = 0; i < s->refcount_table_size; i++) { |
924 |
int64_t offset; |
925 |
offset = s->refcount_table[i]; |
926 |
if (offset != 0) { |
927 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
928 |
offset, s->cluster_size); |
929 |
} |
930 |
} |
931 |
|
932 |
/* compare ref counts */
|
933 |
for(i = 0; i < nb_clusters; i++) { |
934 |
refcount1 = get_refcount(bs, i); |
935 |
refcount2 = refcount_table[i]; |
936 |
if (refcount1 != refcount2) {
|
937 |
fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n",
|
938 |
i, refcount1, refcount2); |
939 |
errors++; |
940 |
} |
941 |
} |
942 |
|
943 |
qemu_free(refcount_table); |
944 |
|
945 |
return errors;
|
946 |
} |
947 |
|