root / block / qcow2-refcount.c @ c5baaa48
History | View | Annotate | Download (28.8 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 |
if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
|
172 |
data, sizeof(data)) != sizeof(data)) |
173 |
goto fail;
|
174 |
qemu_free(s->refcount_table); |
175 |
old_table_offset = s->refcount_table_offset; |
176 |
old_table_size = s->refcount_table_size; |
177 |
s->refcount_table = new_table; |
178 |
s->refcount_table_size = new_table_size; |
179 |
s->refcount_table_offset = table_offset; |
180 |
|
181 |
update_refcount(bs, table_offset, new_table_size2, 1);
|
182 |
qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
|
183 |
return 0; |
184 |
fail:
|
185 |
qemu_free(new_table); |
186 |
return -EIO;
|
187 |
} |
188 |
|
189 |
|
190 |
static int64_t alloc_refcount_block(BlockDriverState *bs, int64_t cluster_index)
|
191 |
{ |
192 |
BDRVQcowState *s = bs->opaque; |
193 |
int64_t offset, refcount_block_offset; |
194 |
unsigned int refcount_table_index; |
195 |
int ret;
|
196 |
uint64_t data64; |
197 |
int cache = cache_refcount_updates;
|
198 |
|
199 |
/* Find L1 index and grow refcount table if needed */
|
200 |
refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); |
201 |
if (refcount_table_index >= s->refcount_table_size) {
|
202 |
ret = grow_refcount_table(bs, refcount_table_index + 1);
|
203 |
if (ret < 0) |
204 |
return ret;
|
205 |
} |
206 |
|
207 |
/* Load or allocate the refcount block */
|
208 |
refcount_block_offset = s->refcount_table[refcount_table_index]; |
209 |
if (!refcount_block_offset) {
|
210 |
if (cache_refcount_updates) {
|
211 |
write_refcount_block(s); |
212 |
cache_refcount_updates = 0;
|
213 |
} |
214 |
/* create a new refcount block */
|
215 |
/* Note: we cannot update the refcount now to avoid recursion */
|
216 |
offset = alloc_clusters_noref(bs, s->cluster_size); |
217 |
memset(s->refcount_block_cache, 0, s->cluster_size);
|
218 |
ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size); |
219 |
if (ret != s->cluster_size)
|
220 |
return -EINVAL;
|
221 |
s->refcount_table[refcount_table_index] = offset; |
222 |
data64 = cpu_to_be64(offset); |
223 |
ret = bdrv_pwrite(s->hd, s->refcount_table_offset + |
224 |
refcount_table_index * sizeof(uint64_t),
|
225 |
&data64, sizeof(data64));
|
226 |
if (ret != sizeof(data64)) |
227 |
return -EINVAL;
|
228 |
|
229 |
refcount_block_offset = offset; |
230 |
s->refcount_block_cache_offset = offset; |
231 |
update_refcount(bs, offset, s->cluster_size, 1);
|
232 |
cache_refcount_updates = cache; |
233 |
} else {
|
234 |
if (refcount_block_offset != s->refcount_block_cache_offset) {
|
235 |
if (load_refcount_block(bs, refcount_block_offset) < 0) |
236 |
return -EIO;
|
237 |
} |
238 |
} |
239 |
|
240 |
return refcount_block_offset;
|
241 |
} |
242 |
|
243 |
#define REFCOUNTS_PER_SECTOR (512 >> REFCOUNT_SHIFT) |
244 |
static int write_refcount_block_entries(BDRVQcowState *s, |
245 |
int64_t refcount_block_offset, int first_index, int last_index) |
246 |
{ |
247 |
size_t size; |
248 |
|
249 |
if (cache_refcount_updates) {
|
250 |
return 0; |
251 |
} |
252 |
|
253 |
first_index &= ~(REFCOUNTS_PER_SECTOR - 1);
|
254 |
last_index = (last_index + REFCOUNTS_PER_SECTOR) |
255 |
& ~(REFCOUNTS_PER_SECTOR - 1);
|
256 |
|
257 |
size = (last_index - first_index) << REFCOUNT_SHIFT; |
258 |
if (bdrv_pwrite(s->hd,
|
259 |
refcount_block_offset + (first_index << REFCOUNT_SHIFT), |
260 |
&s->refcount_block_cache[first_index], size) != size) |
261 |
{ |
262 |
return -EIO;
|
263 |
} |
264 |
|
265 |
return 0; |
266 |
} |
267 |
|
268 |
/* XXX: cache several refcount block clusters ? */
|
269 |
static int update_refcount(BlockDriverState *bs, |
270 |
int64_t offset, int64_t length, |
271 |
int addend)
|
272 |
{ |
273 |
BDRVQcowState *s = bs->opaque; |
274 |
int64_t start, last, cluster_offset; |
275 |
int64_t refcount_block_offset = 0;
|
276 |
int64_t table_index = -1, old_table_index;
|
277 |
int first_index = -1, last_index = -1; |
278 |
|
279 |
#ifdef DEBUG_ALLOC2
|
280 |
printf("update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n", |
281 |
offset, length, addend); |
282 |
#endif
|
283 |
if (length <= 0) |
284 |
return -EINVAL;
|
285 |
start = offset & ~(s->cluster_size - 1);
|
286 |
last = (offset + length - 1) & ~(s->cluster_size - 1); |
287 |
for(cluster_offset = start; cluster_offset <= last;
|
288 |
cluster_offset += s->cluster_size) |
289 |
{ |
290 |
int block_index, refcount;
|
291 |
int64_t cluster_index = cluster_offset >> s->cluster_bits; |
292 |
|
293 |
/* Only write refcount block to disk when we are done with it */
|
294 |
old_table_index = table_index; |
295 |
table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); |
296 |
if ((old_table_index >= 0) && (table_index != old_table_index)) { |
297 |
|
298 |
if (write_refcount_block_entries(s, refcount_block_offset,
|
299 |
first_index, last_index) < 0)
|
300 |
{ |
301 |
return -EIO;
|
302 |
} |
303 |
|
304 |
first_index = -1;
|
305 |
last_index = -1;
|
306 |
} |
307 |
|
308 |
/* Load the refcount block and allocate it if needed */
|
309 |
refcount_block_offset = alloc_refcount_block(bs, cluster_index); |
310 |
if (refcount_block_offset < 0) { |
311 |
return refcount_block_offset;
|
312 |
} |
313 |
|
314 |
/* we can update the count and save it */
|
315 |
block_index = cluster_index & |
316 |
((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); |
317 |
if (first_index == -1 || block_index < first_index) { |
318 |
first_index = block_index; |
319 |
} |
320 |
if (block_index > last_index) {
|
321 |
last_index = block_index; |
322 |
} |
323 |
|
324 |
refcount = be16_to_cpu(s->refcount_block_cache[block_index]); |
325 |
refcount += addend; |
326 |
if (refcount < 0 || refcount > 0xffff) |
327 |
return -EINVAL;
|
328 |
if (refcount == 0 && cluster_index < s->free_cluster_index) { |
329 |
s->free_cluster_index = cluster_index; |
330 |
} |
331 |
s->refcount_block_cache[block_index] = cpu_to_be16(refcount); |
332 |
} |
333 |
|
334 |
/* Write last changed block to disk */
|
335 |
if (refcount_block_offset != 0) { |
336 |
if (write_refcount_block_entries(s, refcount_block_offset,
|
337 |
first_index, last_index) < 0)
|
338 |
{ |
339 |
return -EIO;
|
340 |
} |
341 |
} |
342 |
|
343 |
return 0; |
344 |
} |
345 |
|
346 |
/* addend must be 1 or -1 */
|
347 |
static int update_cluster_refcount(BlockDriverState *bs, |
348 |
int64_t cluster_index, |
349 |
int addend)
|
350 |
{ |
351 |
BDRVQcowState *s = bs->opaque; |
352 |
int ret;
|
353 |
|
354 |
ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
|
355 |
if (ret < 0) { |
356 |
return ret;
|
357 |
} |
358 |
|
359 |
return get_refcount(bs, cluster_index);
|
360 |
} |
361 |
|
362 |
|
363 |
|
364 |
/*********************************************************/
|
365 |
/* cluster allocation functions */
|
366 |
|
367 |
|
368 |
|
369 |
/* return < 0 if error */
|
370 |
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
|
371 |
{ |
372 |
BDRVQcowState *s = bs->opaque; |
373 |
int i, nb_clusters;
|
374 |
|
375 |
nb_clusters = size_to_clusters(s, size); |
376 |
retry:
|
377 |
for(i = 0; i < nb_clusters; i++) { |
378 |
int64_t i = s->free_cluster_index++; |
379 |
if (get_refcount(bs, i) != 0) |
380 |
goto retry;
|
381 |
} |
382 |
#ifdef DEBUG_ALLOC2
|
383 |
printf("alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n", |
384 |
size, |
385 |
(s->free_cluster_index - nb_clusters) << s->cluster_bits); |
386 |
#endif
|
387 |
return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
|
388 |
} |
389 |
|
390 |
int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size) |
391 |
{ |
392 |
int64_t offset; |
393 |
|
394 |
offset = alloc_clusters_noref(bs, size); |
395 |
update_refcount(bs, offset, size, 1);
|
396 |
return offset;
|
397 |
} |
398 |
|
399 |
/* only used to allocate compressed sectors. We try to allocate
|
400 |
contiguous sectors. size must be <= cluster_size */
|
401 |
int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
|
402 |
{ |
403 |
BDRVQcowState *s = bs->opaque; |
404 |
int64_t offset, cluster_offset; |
405 |
int free_in_cluster;
|
406 |
|
407 |
assert(size > 0 && size <= s->cluster_size);
|
408 |
if (s->free_byte_offset == 0) { |
409 |
s->free_byte_offset = qcow2_alloc_clusters(bs, s->cluster_size); |
410 |
} |
411 |
redo:
|
412 |
free_in_cluster = s->cluster_size - |
413 |
(s->free_byte_offset & (s->cluster_size - 1));
|
414 |
if (size <= free_in_cluster) {
|
415 |
/* enough space in current cluster */
|
416 |
offset = s->free_byte_offset; |
417 |
s->free_byte_offset += size; |
418 |
free_in_cluster -= size; |
419 |
if (free_in_cluster == 0) |
420 |
s->free_byte_offset = 0;
|
421 |
if ((offset & (s->cluster_size - 1)) != 0) |
422 |
update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
|
423 |
} else {
|
424 |
offset = qcow2_alloc_clusters(bs, s->cluster_size); |
425 |
cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
|
426 |
if ((cluster_offset + s->cluster_size) == offset) {
|
427 |
/* we are lucky: contiguous data */
|
428 |
offset = s->free_byte_offset; |
429 |
update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
|
430 |
s->free_byte_offset += size; |
431 |
} else {
|
432 |
s->free_byte_offset = offset; |
433 |
goto redo;
|
434 |
} |
435 |
} |
436 |
return offset;
|
437 |
} |
438 |
|
439 |
void qcow2_free_clusters(BlockDriverState *bs,
|
440 |
int64_t offset, int64_t size) |
441 |
{ |
442 |
update_refcount(bs, offset, size, -1);
|
443 |
} |
444 |
|
445 |
/*
|
446 |
* free_any_clusters
|
447 |
*
|
448 |
* free clusters according to its type: compressed or not
|
449 |
*
|
450 |
*/
|
451 |
|
452 |
void qcow2_free_any_clusters(BlockDriverState *bs,
|
453 |
uint64_t cluster_offset, int nb_clusters)
|
454 |
{ |
455 |
BDRVQcowState *s = bs->opaque; |
456 |
|
457 |
/* free the cluster */
|
458 |
|
459 |
if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
|
460 |
int nb_csectors;
|
461 |
nb_csectors = ((cluster_offset >> s->csize_shift) & |
462 |
s->csize_mask) + 1;
|
463 |
qcow2_free_clusters(bs, |
464 |
(cluster_offset & s->cluster_offset_mask) & ~511,
|
465 |
nb_csectors * 512);
|
466 |
return;
|
467 |
} |
468 |
|
469 |
qcow2_free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits); |
470 |
|
471 |
return;
|
472 |
} |
473 |
|
474 |
|
475 |
|
476 |
/*********************************************************/
|
477 |
/* snapshots and image creation */
|
478 |
|
479 |
|
480 |
|
481 |
void qcow2_create_refcount_update(QCowCreateState *s, int64_t offset,
|
482 |
int64_t size) |
483 |
{ |
484 |
int refcount;
|
485 |
int64_t start, last, cluster_offset; |
486 |
uint16_t *p; |
487 |
|
488 |
start = offset & ~(s->cluster_size - 1);
|
489 |
last = (offset + size - 1) & ~(s->cluster_size - 1); |
490 |
for(cluster_offset = start; cluster_offset <= last;
|
491 |
cluster_offset += s->cluster_size) { |
492 |
p = &s->refcount_block[cluster_offset >> s->cluster_bits]; |
493 |
refcount = be16_to_cpu(*p); |
494 |
refcount++; |
495 |
*p = cpu_to_be16(refcount); |
496 |
} |
497 |
} |
498 |
|
499 |
/* update the refcounts of snapshots and the copied flag */
|
500 |
int qcow2_update_snapshot_refcount(BlockDriverState *bs,
|
501 |
int64_t l1_table_offset, int l1_size, int addend) |
502 |
{ |
503 |
BDRVQcowState *s = bs->opaque; |
504 |
uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated; |
505 |
int64_t old_offset, old_l2_offset; |
506 |
int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
|
507 |
|
508 |
qcow2_l2_cache_reset(bs); |
509 |
cache_refcount_updates = 1;
|
510 |
|
511 |
l2_table = NULL;
|
512 |
l1_table = NULL;
|
513 |
l1_size2 = l1_size * sizeof(uint64_t);
|
514 |
l1_allocated = 0;
|
515 |
if (l1_table_offset != s->l1_table_offset) {
|
516 |
l1_table = qemu_mallocz(align_offset(l1_size2, 512));
|
517 |
l1_allocated = 1;
|
518 |
if (bdrv_pread(s->hd, l1_table_offset,
|
519 |
l1_table, l1_size2) != l1_size2) |
520 |
goto fail;
|
521 |
for(i = 0;i < l1_size; i++) |
522 |
be64_to_cpus(&l1_table[i]); |
523 |
} else {
|
524 |
assert(l1_size == s->l1_size); |
525 |
l1_table = s->l1_table; |
526 |
l1_allocated = 0;
|
527 |
} |
528 |
|
529 |
l2_size = s->l2_size * sizeof(uint64_t);
|
530 |
l2_table = qemu_malloc(l2_size); |
531 |
l1_modified = 0;
|
532 |
for(i = 0; i < l1_size; i++) { |
533 |
l2_offset = l1_table[i]; |
534 |
if (l2_offset) {
|
535 |
old_l2_offset = l2_offset; |
536 |
l2_offset &= ~QCOW_OFLAG_COPIED; |
537 |
l2_modified = 0;
|
538 |
if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
|
539 |
goto fail;
|
540 |
for(j = 0; j < s->l2_size; j++) { |
541 |
offset = be64_to_cpu(l2_table[j]); |
542 |
if (offset != 0) { |
543 |
old_offset = offset; |
544 |
offset &= ~QCOW_OFLAG_COPIED; |
545 |
if (offset & QCOW_OFLAG_COMPRESSED) {
|
546 |
nb_csectors = ((offset >> s->csize_shift) & |
547 |
s->csize_mask) + 1;
|
548 |
if (addend != 0) |
549 |
update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
|
550 |
nb_csectors * 512, addend);
|
551 |
/* compressed clusters are never modified */
|
552 |
refcount = 2;
|
553 |
} else {
|
554 |
if (addend != 0) { |
555 |
refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend); |
556 |
} else {
|
557 |
refcount = get_refcount(bs, offset >> s->cluster_bits); |
558 |
} |
559 |
} |
560 |
|
561 |
if (refcount == 1) { |
562 |
offset |= QCOW_OFLAG_COPIED; |
563 |
} |
564 |
if (offset != old_offset) {
|
565 |
l2_table[j] = cpu_to_be64(offset); |
566 |
l2_modified = 1;
|
567 |
} |
568 |
} |
569 |
} |
570 |
if (l2_modified) {
|
571 |
if (bdrv_pwrite(s->hd,
|
572 |
l2_offset, l2_table, l2_size) != l2_size) |
573 |
goto fail;
|
574 |
} |
575 |
|
576 |
if (addend != 0) { |
577 |
refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend); |
578 |
} else {
|
579 |
refcount = get_refcount(bs, l2_offset >> s->cluster_bits); |
580 |
} |
581 |
if (refcount == 1) { |
582 |
l2_offset |= QCOW_OFLAG_COPIED; |
583 |
} |
584 |
if (l2_offset != old_l2_offset) {
|
585 |
l1_table[i] = l2_offset; |
586 |
l1_modified = 1;
|
587 |
} |
588 |
} |
589 |
} |
590 |
if (l1_modified) {
|
591 |
for(i = 0; i < l1_size; i++) |
592 |
cpu_to_be64s(&l1_table[i]); |
593 |
if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
|
594 |
l1_size2) != l1_size2) |
595 |
goto fail;
|
596 |
for(i = 0; i < l1_size; i++) |
597 |
be64_to_cpus(&l1_table[i]); |
598 |
} |
599 |
if (l1_allocated)
|
600 |
qemu_free(l1_table); |
601 |
qemu_free(l2_table); |
602 |
cache_refcount_updates = 0;
|
603 |
write_refcount_block(s); |
604 |
return 0; |
605 |
fail:
|
606 |
if (l1_allocated)
|
607 |
qemu_free(l1_table); |
608 |
qemu_free(l2_table); |
609 |
cache_refcount_updates = 0;
|
610 |
write_refcount_block(s); |
611 |
return -EIO;
|
612 |
} |
613 |
|
614 |
|
615 |
|
616 |
|
617 |
/*********************************************************/
|
618 |
/* refcount checking functions */
|
619 |
|
620 |
|
621 |
|
622 |
/*
|
623 |
* Increases the refcount for a range of clusters in a given refcount table.
|
624 |
* This is used to construct a temporary refcount table out of L1 and L2 tables
|
625 |
* which can be compared the the refcount table saved in the image.
|
626 |
*
|
627 |
* Returns the number of errors in the image that were found
|
628 |
*/
|
629 |
static int inc_refcounts(BlockDriverState *bs, |
630 |
uint16_t *refcount_table, |
631 |
int refcount_table_size,
|
632 |
int64_t offset, int64_t size) |
633 |
{ |
634 |
BDRVQcowState *s = bs->opaque; |
635 |
int64_t start, last, cluster_offset; |
636 |
int k;
|
637 |
int errors = 0; |
638 |
|
639 |
if (size <= 0) |
640 |
return 0; |
641 |
|
642 |
start = offset & ~(s->cluster_size - 1);
|
643 |
last = (offset + size - 1) & ~(s->cluster_size - 1); |
644 |
for(cluster_offset = start; cluster_offset <= last;
|
645 |
cluster_offset += s->cluster_size) { |
646 |
k = cluster_offset >> s->cluster_bits; |
647 |
if (k < 0 || k >= refcount_table_size) { |
648 |
fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n", |
649 |
cluster_offset); |
650 |
errors++; |
651 |
} else {
|
652 |
if (++refcount_table[k] == 0) { |
653 |
fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
|
654 |
"\n", cluster_offset);
|
655 |
errors++; |
656 |
} |
657 |
} |
658 |
} |
659 |
|
660 |
return errors;
|
661 |
} |
662 |
|
663 |
/*
|
664 |
* Increases the refcount in the given refcount table for the all clusters
|
665 |
* referenced in the L2 table. While doing so, performs some checks on L2
|
666 |
* entries.
|
667 |
*
|
668 |
* Returns the number of errors found by the checks or -errno if an internal
|
669 |
* error occurred.
|
670 |
*/
|
671 |
static int check_refcounts_l2(BlockDriverState *bs, |
672 |
uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
|
673 |
int check_copied)
|
674 |
{ |
675 |
BDRVQcowState *s = bs->opaque; |
676 |
uint64_t *l2_table, offset; |
677 |
int i, l2_size, nb_csectors, refcount;
|
678 |
int errors = 0; |
679 |
|
680 |
/* Read L2 table from disk */
|
681 |
l2_size = s->l2_size * sizeof(uint64_t);
|
682 |
l2_table = qemu_malloc(l2_size); |
683 |
|
684 |
if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
|
685 |
goto fail;
|
686 |
|
687 |
/* Do the actual checks */
|
688 |
for(i = 0; i < s->l2_size; i++) { |
689 |
offset = be64_to_cpu(l2_table[i]); |
690 |
if (offset != 0) { |
691 |
if (offset & QCOW_OFLAG_COMPRESSED) {
|
692 |
/* Compressed clusters don't have QCOW_OFLAG_COPIED */
|
693 |
if (offset & QCOW_OFLAG_COPIED) {
|
694 |
fprintf(stderr, "ERROR: cluster %" PRId64 ": " |
695 |
"copied flag must never be set for compressed "
|
696 |
"clusters\n", offset >> s->cluster_bits);
|
697 |
offset &= ~QCOW_OFLAG_COPIED; |
698 |
errors++; |
699 |
} |
700 |
|
701 |
/* Mark cluster as used */
|
702 |
nb_csectors = ((offset >> s->csize_shift) & |
703 |
s->csize_mask) + 1;
|
704 |
offset &= s->cluster_offset_mask; |
705 |
errors += inc_refcounts(bs, refcount_table, |
706 |
refcount_table_size, |
707 |
offset & ~511, nb_csectors * 512); |
708 |
} else {
|
709 |
/* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
|
710 |
if (check_copied) {
|
711 |
uint64_t entry = offset; |
712 |
offset &= ~QCOW_OFLAG_COPIED; |
713 |
refcount = get_refcount(bs, offset >> s->cluster_bits); |
714 |
if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) { |
715 |
fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
|
716 |
PRIx64 " refcount=%d\n", entry, refcount);
|
717 |
errors++; |
718 |
} |
719 |
} |
720 |
|
721 |
/* Mark cluster as used */
|
722 |
offset &= ~QCOW_OFLAG_COPIED; |
723 |
errors += inc_refcounts(bs, refcount_table, |
724 |
refcount_table_size, |
725 |
offset, s->cluster_size); |
726 |
|
727 |
/* Correct offsets are cluster aligned */
|
728 |
if (offset & (s->cluster_size - 1)) { |
729 |
fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not " |
730 |
"properly aligned; L2 entry corrupted.\n", offset);
|
731 |
errors++; |
732 |
} |
733 |
} |
734 |
} |
735 |
} |
736 |
|
737 |
qemu_free(l2_table); |
738 |
return errors;
|
739 |
|
740 |
fail:
|
741 |
fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
|
742 |
qemu_free(l2_table); |
743 |
return -EIO;
|
744 |
} |
745 |
|
746 |
/*
|
747 |
* Increases the refcount for the L1 table, its L2 tables and all referenced
|
748 |
* clusters in the given refcount table. While doing so, performs some checks
|
749 |
* on L1 and L2 entries.
|
750 |
*
|
751 |
* Returns the number of errors found by the checks or -errno if an internal
|
752 |
* error occurred.
|
753 |
*/
|
754 |
static int check_refcounts_l1(BlockDriverState *bs, |
755 |
uint16_t *refcount_table, |
756 |
int refcount_table_size,
|
757 |
int64_t l1_table_offset, int l1_size,
|
758 |
int check_copied)
|
759 |
{ |
760 |
BDRVQcowState *s = bs->opaque; |
761 |
uint64_t *l1_table, l2_offset, l1_size2; |
762 |
int i, refcount, ret;
|
763 |
int errors = 0; |
764 |
|
765 |
l1_size2 = l1_size * sizeof(uint64_t);
|
766 |
|
767 |
/* Mark L1 table as used */
|
768 |
errors += inc_refcounts(bs, refcount_table, refcount_table_size, |
769 |
l1_table_offset, l1_size2); |
770 |
|
771 |
/* Read L1 table entries from disk */
|
772 |
l1_table = qemu_malloc(l1_size2); |
773 |
if (bdrv_pread(s->hd, l1_table_offset,
|
774 |
l1_table, l1_size2) != l1_size2) |
775 |
goto fail;
|
776 |
for(i = 0;i < l1_size; i++) |
777 |
be64_to_cpus(&l1_table[i]); |
778 |
|
779 |
/* Do the actual checks */
|
780 |
for(i = 0; i < l1_size; i++) { |
781 |
l2_offset = l1_table[i]; |
782 |
if (l2_offset) {
|
783 |
/* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
|
784 |
if (check_copied) {
|
785 |
refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) |
786 |
>> s->cluster_bits); |
787 |
if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) { |
788 |
fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
|
789 |
" refcount=%d\n", l2_offset, refcount);
|
790 |
errors++; |
791 |
} |
792 |
} |
793 |
|
794 |
/* Mark L2 table as used */
|
795 |
l2_offset &= ~QCOW_OFLAG_COPIED; |
796 |
errors += inc_refcounts(bs, refcount_table, |
797 |
refcount_table_size, |
798 |
l2_offset, |
799 |
s->cluster_size); |
800 |
|
801 |
/* L2 tables are cluster aligned */
|
802 |
if (l2_offset & (s->cluster_size - 1)) { |
803 |
fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not " |
804 |
"cluster aligned; L1 entry corrupted\n", l2_offset);
|
805 |
errors++; |
806 |
} |
807 |
|
808 |
/* Process and check L2 entries */
|
809 |
ret = check_refcounts_l2(bs, refcount_table, refcount_table_size, |
810 |
l2_offset, check_copied); |
811 |
if (ret < 0) { |
812 |
goto fail;
|
813 |
} |
814 |
errors += ret; |
815 |
} |
816 |
} |
817 |
qemu_free(l1_table); |
818 |
return errors;
|
819 |
|
820 |
fail:
|
821 |
fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
|
822 |
qemu_free(l1_table); |
823 |
return -EIO;
|
824 |
} |
825 |
|
826 |
/*
|
827 |
* Checks an image for refcount consistency.
|
828 |
*
|
829 |
* Returns 0 if no errors are found, the number of errors in case the image is
|
830 |
* detected as corrupted, and -errno when an internal error occured.
|
831 |
*/
|
832 |
int qcow2_check_refcounts(BlockDriverState *bs)
|
833 |
{ |
834 |
BDRVQcowState *s = bs->opaque; |
835 |
int64_t size; |
836 |
int nb_clusters, refcount1, refcount2, i;
|
837 |
QCowSnapshot *sn; |
838 |
uint16_t *refcount_table; |
839 |
int ret, errors = 0; |
840 |
|
841 |
size = bdrv_getlength(s->hd); |
842 |
nb_clusters = size_to_clusters(s, size); |
843 |
refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
|
844 |
|
845 |
/* header */
|
846 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
847 |
0, s->cluster_size);
|
848 |
|
849 |
/* current L1 table */
|
850 |
ret = check_refcounts_l1(bs, refcount_table, nb_clusters, |
851 |
s->l1_table_offset, s->l1_size, 1);
|
852 |
if (ret < 0) { |
853 |
return ret;
|
854 |
} |
855 |
errors += ret; |
856 |
|
857 |
/* snapshots */
|
858 |
for(i = 0; i < s->nb_snapshots; i++) { |
859 |
sn = s->snapshots + i; |
860 |
check_refcounts_l1(bs, refcount_table, nb_clusters, |
861 |
sn->l1_table_offset, sn->l1_size, 0);
|
862 |
} |
863 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
864 |
s->snapshots_offset, s->snapshots_size); |
865 |
|
866 |
/* refcount data */
|
867 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
868 |
s->refcount_table_offset, |
869 |
s->refcount_table_size * sizeof(uint64_t));
|
870 |
for(i = 0; i < s->refcount_table_size; i++) { |
871 |
int64_t offset; |
872 |
offset = s->refcount_table[i]; |
873 |
if (offset != 0) { |
874 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
875 |
offset, s->cluster_size); |
876 |
} |
877 |
} |
878 |
|
879 |
/* compare ref counts */
|
880 |
for(i = 0; i < nb_clusters; i++) { |
881 |
refcount1 = get_refcount(bs, i); |
882 |
refcount2 = refcount_table[i]; |
883 |
if (refcount1 != refcount2) {
|
884 |
fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n",
|
885 |
i, refcount1, refcount2); |
886 |
errors++; |
887 |
} |
888 |
} |
889 |
|
890 |
qemu_free(refcount_table); |
891 |
|
892 |
return errors;
|
893 |
} |
894 |
|