root / block / qcow2-refcount.c @ f7d0fe02
History | View | Annotate | Download (27 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 |
/* refcount handling */
|
36 |
|
37 |
int refcount_init(BlockDriverState *bs)
|
38 |
{ |
39 |
BDRVQcowState *s = bs->opaque; |
40 |
int ret, refcount_table_size2, i;
|
41 |
|
42 |
s->refcount_block_cache = qemu_malloc(s->cluster_size); |
43 |
refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
|
44 |
s->refcount_table = qemu_malloc(refcount_table_size2); |
45 |
if (s->refcount_table_size > 0) { |
46 |
ret = bdrv_pread(s->hd, s->refcount_table_offset, |
47 |
s->refcount_table, refcount_table_size2); |
48 |
if (ret != refcount_table_size2)
|
49 |
goto fail;
|
50 |
for(i = 0; i < s->refcount_table_size; i++) |
51 |
be64_to_cpus(&s->refcount_table[i]); |
52 |
} |
53 |
return 0; |
54 |
fail:
|
55 |
return -ENOMEM;
|
56 |
} |
57 |
|
58 |
void refcount_close(BlockDriverState *bs)
|
59 |
{ |
60 |
BDRVQcowState *s = bs->opaque; |
61 |
qemu_free(s->refcount_block_cache); |
62 |
qemu_free(s->refcount_table); |
63 |
} |
64 |
|
65 |
|
66 |
static int load_refcount_block(BlockDriverState *bs, |
67 |
int64_t refcount_block_offset) |
68 |
{ |
69 |
BDRVQcowState *s = bs->opaque; |
70 |
int ret;
|
71 |
ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache, |
72 |
s->cluster_size); |
73 |
if (ret != s->cluster_size)
|
74 |
return -EIO;
|
75 |
s->refcount_block_cache_offset = refcount_block_offset; |
76 |
return 0; |
77 |
} |
78 |
|
79 |
static int get_refcount(BlockDriverState *bs, int64_t cluster_index) |
80 |
{ |
81 |
BDRVQcowState *s = bs->opaque; |
82 |
int refcount_table_index, block_index;
|
83 |
int64_t refcount_block_offset; |
84 |
|
85 |
refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); |
86 |
if (refcount_table_index >= s->refcount_table_size)
|
87 |
return 0; |
88 |
refcount_block_offset = s->refcount_table[refcount_table_index]; |
89 |
if (!refcount_block_offset)
|
90 |
return 0; |
91 |
if (refcount_block_offset != s->refcount_block_cache_offset) {
|
92 |
/* better than nothing: return allocated if read error */
|
93 |
if (load_refcount_block(bs, refcount_block_offset) < 0) |
94 |
return 1; |
95 |
} |
96 |
block_index = cluster_index & |
97 |
((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); |
98 |
return be16_to_cpu(s->refcount_block_cache[block_index]);
|
99 |
} |
100 |
|
101 |
static int grow_refcount_table(BlockDriverState *bs, int min_size) |
102 |
{ |
103 |
BDRVQcowState *s = bs->opaque; |
104 |
int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
|
105 |
uint64_t *new_table; |
106 |
int64_t table_offset; |
107 |
uint8_t data[12];
|
108 |
int old_table_size;
|
109 |
int64_t old_table_offset; |
110 |
|
111 |
if (min_size <= s->refcount_table_size)
|
112 |
return 0; |
113 |
/* compute new table size */
|
114 |
refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
|
115 |
for(;;) {
|
116 |
if (refcount_table_clusters == 0) { |
117 |
refcount_table_clusters = 1;
|
118 |
} else {
|
119 |
refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2; |
120 |
} |
121 |
new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
|
122 |
if (min_size <= new_table_size)
|
123 |
break;
|
124 |
} |
125 |
#ifdef DEBUG_ALLOC2
|
126 |
printf("grow_refcount_table from %d to %d\n",
|
127 |
s->refcount_table_size, |
128 |
new_table_size); |
129 |
#endif
|
130 |
new_table_size2 = new_table_size * sizeof(uint64_t);
|
131 |
new_table = qemu_mallocz(new_table_size2); |
132 |
memcpy(new_table, s->refcount_table, |
133 |
s->refcount_table_size * sizeof(uint64_t));
|
134 |
for(i = 0; i < s->refcount_table_size; i++) |
135 |
cpu_to_be64s(&new_table[i]); |
136 |
/* Note: we cannot update the refcount now to avoid recursion */
|
137 |
table_offset = alloc_clusters_noref(bs, new_table_size2); |
138 |
ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2); |
139 |
if (ret != new_table_size2)
|
140 |
goto fail;
|
141 |
for(i = 0; i < s->refcount_table_size; i++) |
142 |
be64_to_cpus(&new_table[i]); |
143 |
|
144 |
cpu_to_be64w((uint64_t*)data, table_offset); |
145 |
cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
|
146 |
if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
|
147 |
data, sizeof(data)) != sizeof(data)) |
148 |
goto fail;
|
149 |
qemu_free(s->refcount_table); |
150 |
old_table_offset = s->refcount_table_offset; |
151 |
old_table_size = s->refcount_table_size; |
152 |
s->refcount_table = new_table; |
153 |
s->refcount_table_size = new_table_size; |
154 |
s->refcount_table_offset = table_offset; |
155 |
|
156 |
update_refcount(bs, table_offset, new_table_size2, 1);
|
157 |
free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
|
158 |
return 0; |
159 |
fail:
|
160 |
free_clusters(bs, table_offset, new_table_size2); |
161 |
qemu_free(new_table); |
162 |
return -EIO;
|
163 |
} |
164 |
|
165 |
|
166 |
static int64_t alloc_refcount_block(BlockDriverState *bs, int64_t cluster_index)
|
167 |
{ |
168 |
BDRVQcowState *s = bs->opaque; |
169 |
int64_t offset, refcount_block_offset; |
170 |
int ret, refcount_table_index;
|
171 |
uint64_t data64; |
172 |
|
173 |
/* Find L1 index and grow refcount table if needed */
|
174 |
refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); |
175 |
if (refcount_table_index >= s->refcount_table_size) {
|
176 |
ret = grow_refcount_table(bs, refcount_table_index + 1);
|
177 |
if (ret < 0) |
178 |
return ret;
|
179 |
} |
180 |
|
181 |
/* Load or allocate the refcount block */
|
182 |
refcount_block_offset = s->refcount_table[refcount_table_index]; |
183 |
if (!refcount_block_offset) {
|
184 |
/* create a new refcount block */
|
185 |
/* Note: we cannot update the refcount now to avoid recursion */
|
186 |
offset = alloc_clusters_noref(bs, s->cluster_size); |
187 |
memset(s->refcount_block_cache, 0, s->cluster_size);
|
188 |
ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size); |
189 |
if (ret != s->cluster_size)
|
190 |
return -EINVAL;
|
191 |
s->refcount_table[refcount_table_index] = offset; |
192 |
data64 = cpu_to_be64(offset); |
193 |
ret = bdrv_pwrite(s->hd, s->refcount_table_offset + |
194 |
refcount_table_index * sizeof(uint64_t),
|
195 |
&data64, sizeof(data64));
|
196 |
if (ret != sizeof(data64)) |
197 |
return -EINVAL;
|
198 |
|
199 |
refcount_block_offset = offset; |
200 |
s->refcount_block_cache_offset = offset; |
201 |
update_refcount(bs, offset, s->cluster_size, 1);
|
202 |
} else {
|
203 |
if (refcount_block_offset != s->refcount_block_cache_offset) {
|
204 |
if (load_refcount_block(bs, refcount_block_offset) < 0) |
205 |
return -EIO;
|
206 |
} |
207 |
} |
208 |
|
209 |
return refcount_block_offset;
|
210 |
} |
211 |
|
212 |
/* XXX: cache several refcount block clusters ? */
|
213 |
static int update_refcount(BlockDriverState *bs, |
214 |
int64_t offset, int64_t length, |
215 |
int addend)
|
216 |
{ |
217 |
BDRVQcowState *s = bs->opaque; |
218 |
int64_t start, last, cluster_offset; |
219 |
int64_t refcount_block_offset = 0;
|
220 |
int64_t table_index = -1, old_table_index;
|
221 |
int first_index = -1, last_index = -1; |
222 |
|
223 |
#ifdef DEBUG_ALLOC2
|
224 |
printf("update_refcount: offset=%lld size=%lld addend=%d\n",
|
225 |
offset, length, addend); |
226 |
#endif
|
227 |
if (length <= 0) |
228 |
return -EINVAL;
|
229 |
start = offset & ~(s->cluster_size - 1);
|
230 |
last = (offset + length - 1) & ~(s->cluster_size - 1); |
231 |
for(cluster_offset = start; cluster_offset <= last;
|
232 |
cluster_offset += s->cluster_size) |
233 |
{ |
234 |
int block_index, refcount;
|
235 |
int64_t cluster_index = cluster_offset >> s->cluster_bits; |
236 |
|
237 |
/* Only write refcount block to disk when we are done with it */
|
238 |
old_table_index = table_index; |
239 |
table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); |
240 |
if ((old_table_index >= 0) && (table_index != old_table_index)) { |
241 |
size_t size = (last_index - first_index + 1) << REFCOUNT_SHIFT;
|
242 |
if (bdrv_pwrite(s->hd,
|
243 |
refcount_block_offset + (first_index << REFCOUNT_SHIFT), |
244 |
&s->refcount_block_cache[first_index], size) != size) |
245 |
{ |
246 |
return -EIO;
|
247 |
} |
248 |
|
249 |
first_index = -1;
|
250 |
last_index = -1;
|
251 |
} |
252 |
|
253 |
/* Load the refcount block and allocate it if needed */
|
254 |
refcount_block_offset = alloc_refcount_block(bs, cluster_index); |
255 |
if (refcount_block_offset < 0) { |
256 |
return refcount_block_offset;
|
257 |
} |
258 |
|
259 |
/* we can update the count and save it */
|
260 |
block_index = cluster_index & |
261 |
((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); |
262 |
if (first_index == -1 || block_index < first_index) { |
263 |
first_index = block_index; |
264 |
} |
265 |
if (block_index > last_index) {
|
266 |
last_index = block_index; |
267 |
} |
268 |
|
269 |
refcount = be16_to_cpu(s->refcount_block_cache[block_index]); |
270 |
refcount += addend; |
271 |
if (refcount < 0 || refcount > 0xffff) |
272 |
return -EINVAL;
|
273 |
if (refcount == 0 && cluster_index < s->free_cluster_index) { |
274 |
s->free_cluster_index = cluster_index; |
275 |
} |
276 |
s->refcount_block_cache[block_index] = cpu_to_be16(refcount); |
277 |
} |
278 |
|
279 |
/* Write last changed block to disk */
|
280 |
if (refcount_block_offset != 0) { |
281 |
size_t size = (last_index - first_index + 1) << REFCOUNT_SHIFT;
|
282 |
if (bdrv_pwrite(s->hd,
|
283 |
refcount_block_offset + (first_index << REFCOUNT_SHIFT), |
284 |
&s->refcount_block_cache[first_index], size) != size) |
285 |
{ |
286 |
return -EIO;
|
287 |
} |
288 |
} |
289 |
|
290 |
return 0; |
291 |
} |
292 |
|
293 |
/* addend must be 1 or -1 */
|
294 |
static int update_cluster_refcount(BlockDriverState *bs, |
295 |
int64_t cluster_index, |
296 |
int addend)
|
297 |
{ |
298 |
BDRVQcowState *s = bs->opaque; |
299 |
int ret;
|
300 |
|
301 |
ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
|
302 |
if (ret < 0) { |
303 |
return ret;
|
304 |
} |
305 |
|
306 |
return get_refcount(bs, cluster_index);
|
307 |
} |
308 |
|
309 |
|
310 |
|
311 |
/*********************************************************/
|
312 |
/* cluster allocation functions */
|
313 |
|
314 |
|
315 |
|
316 |
/* return < 0 if error */
|
317 |
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
|
318 |
{ |
319 |
BDRVQcowState *s = bs->opaque; |
320 |
int i, nb_clusters;
|
321 |
|
322 |
nb_clusters = size_to_clusters(s, size); |
323 |
retry:
|
324 |
for(i = 0; i < nb_clusters; i++) { |
325 |
int64_t i = s->free_cluster_index++; |
326 |
if (get_refcount(bs, i) != 0) |
327 |
goto retry;
|
328 |
} |
329 |
#ifdef DEBUG_ALLOC2
|
330 |
printf("alloc_clusters: size=%lld -> %lld\n",
|
331 |
size, |
332 |
(s->free_cluster_index - nb_clusters) << s->cluster_bits); |
333 |
#endif
|
334 |
return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
|
335 |
} |
336 |
|
337 |
int64_t alloc_clusters(BlockDriverState *bs, int64_t size) |
338 |
{ |
339 |
int64_t offset; |
340 |
|
341 |
offset = alloc_clusters_noref(bs, size); |
342 |
update_refcount(bs, offset, size, 1);
|
343 |
return offset;
|
344 |
} |
345 |
|
346 |
/* only used to allocate compressed sectors. We try to allocate
|
347 |
contiguous sectors. size must be <= cluster_size */
|
348 |
int64_t alloc_bytes(BlockDriverState *bs, int size)
|
349 |
{ |
350 |
BDRVQcowState *s = bs->opaque; |
351 |
int64_t offset, cluster_offset; |
352 |
int free_in_cluster;
|
353 |
|
354 |
assert(size > 0 && size <= s->cluster_size);
|
355 |
if (s->free_byte_offset == 0) { |
356 |
s->free_byte_offset = alloc_clusters(bs, s->cluster_size); |
357 |
} |
358 |
redo:
|
359 |
free_in_cluster = s->cluster_size - |
360 |
(s->free_byte_offset & (s->cluster_size - 1));
|
361 |
if (size <= free_in_cluster) {
|
362 |
/* enough space in current cluster */
|
363 |
offset = s->free_byte_offset; |
364 |
s->free_byte_offset += size; |
365 |
free_in_cluster -= size; |
366 |
if (free_in_cluster == 0) |
367 |
s->free_byte_offset = 0;
|
368 |
if ((offset & (s->cluster_size - 1)) != 0) |
369 |
update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
|
370 |
} else {
|
371 |
offset = alloc_clusters(bs, s->cluster_size); |
372 |
cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
|
373 |
if ((cluster_offset + s->cluster_size) == offset) {
|
374 |
/* we are lucky: contiguous data */
|
375 |
offset = s->free_byte_offset; |
376 |
update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
|
377 |
s->free_byte_offset += size; |
378 |
} else {
|
379 |
s->free_byte_offset = offset; |
380 |
goto redo;
|
381 |
} |
382 |
} |
383 |
return offset;
|
384 |
} |
385 |
|
386 |
void free_clusters(BlockDriverState *bs,
|
387 |
int64_t offset, int64_t size) |
388 |
{ |
389 |
update_refcount(bs, offset, size, -1);
|
390 |
} |
391 |
|
392 |
|
393 |
|
394 |
/*********************************************************/
|
395 |
/* snapshots and image creation */
|
396 |
|
397 |
|
398 |
|
399 |
void create_refcount_update(QCowCreateState *s, int64_t offset, int64_t size)
|
400 |
{ |
401 |
int refcount;
|
402 |
int64_t start, last, cluster_offset; |
403 |
uint16_t *p; |
404 |
|
405 |
start = offset & ~(s->cluster_size - 1);
|
406 |
last = (offset + size - 1) & ~(s->cluster_size - 1); |
407 |
for(cluster_offset = start; cluster_offset <= last;
|
408 |
cluster_offset += s->cluster_size) { |
409 |
p = &s->refcount_block[cluster_offset >> s->cluster_bits]; |
410 |
refcount = be16_to_cpu(*p); |
411 |
refcount++; |
412 |
*p = cpu_to_be16(refcount); |
413 |
} |
414 |
} |
415 |
|
416 |
/* update the refcounts of snapshots and the copied flag */
|
417 |
int update_snapshot_refcount(BlockDriverState *bs,
|
418 |
int64_t l1_table_offset, |
419 |
int l1_size,
|
420 |
int addend)
|
421 |
{ |
422 |
BDRVQcowState *s = bs->opaque; |
423 |
uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated; |
424 |
int64_t old_offset, old_l2_offset; |
425 |
int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
|
426 |
|
427 |
l2_cache_reset(bs); |
428 |
|
429 |
l2_table = NULL;
|
430 |
l1_table = NULL;
|
431 |
l1_size2 = l1_size * sizeof(uint64_t);
|
432 |
l1_allocated = 0;
|
433 |
if (l1_table_offset != s->l1_table_offset) {
|
434 |
l1_table = qemu_malloc(l1_size2); |
435 |
l1_allocated = 1;
|
436 |
if (bdrv_pread(s->hd, l1_table_offset,
|
437 |
l1_table, l1_size2) != l1_size2) |
438 |
goto fail;
|
439 |
for(i = 0;i < l1_size; i++) |
440 |
be64_to_cpus(&l1_table[i]); |
441 |
} else {
|
442 |
assert(l1_size == s->l1_size); |
443 |
l1_table = s->l1_table; |
444 |
l1_allocated = 0;
|
445 |
} |
446 |
|
447 |
l2_size = s->l2_size * sizeof(uint64_t);
|
448 |
l2_table = qemu_malloc(l2_size); |
449 |
l1_modified = 0;
|
450 |
for(i = 0; i < l1_size; i++) { |
451 |
l2_offset = l1_table[i]; |
452 |
if (l2_offset) {
|
453 |
old_l2_offset = l2_offset; |
454 |
l2_offset &= ~QCOW_OFLAG_COPIED; |
455 |
l2_modified = 0;
|
456 |
if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
|
457 |
goto fail;
|
458 |
for(j = 0; j < s->l2_size; j++) { |
459 |
offset = be64_to_cpu(l2_table[j]); |
460 |
if (offset != 0) { |
461 |
old_offset = offset; |
462 |
offset &= ~QCOW_OFLAG_COPIED; |
463 |
if (offset & QCOW_OFLAG_COMPRESSED) {
|
464 |
nb_csectors = ((offset >> s->csize_shift) & |
465 |
s->csize_mask) + 1;
|
466 |
if (addend != 0) |
467 |
update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
|
468 |
nb_csectors * 512, addend);
|
469 |
/* compressed clusters are never modified */
|
470 |
refcount = 2;
|
471 |
} else {
|
472 |
if (addend != 0) { |
473 |
refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend); |
474 |
} else {
|
475 |
refcount = get_refcount(bs, offset >> s->cluster_bits); |
476 |
} |
477 |
} |
478 |
|
479 |
if (refcount == 1) { |
480 |
offset |= QCOW_OFLAG_COPIED; |
481 |
} |
482 |
if (offset != old_offset) {
|
483 |
l2_table[j] = cpu_to_be64(offset); |
484 |
l2_modified = 1;
|
485 |
} |
486 |
} |
487 |
} |
488 |
if (l2_modified) {
|
489 |
if (bdrv_pwrite(s->hd,
|
490 |
l2_offset, l2_table, l2_size) != l2_size) |
491 |
goto fail;
|
492 |
} |
493 |
|
494 |
if (addend != 0) { |
495 |
refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend); |
496 |
} else {
|
497 |
refcount = get_refcount(bs, l2_offset >> s->cluster_bits); |
498 |
} |
499 |
if (refcount == 1) { |
500 |
l2_offset |= QCOW_OFLAG_COPIED; |
501 |
} |
502 |
if (l2_offset != old_l2_offset) {
|
503 |
l1_table[i] = l2_offset; |
504 |
l1_modified = 1;
|
505 |
} |
506 |
} |
507 |
} |
508 |
if (l1_modified) {
|
509 |
for(i = 0; i < l1_size; i++) |
510 |
cpu_to_be64s(&l1_table[i]); |
511 |
if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
|
512 |
l1_size2) != l1_size2) |
513 |
goto fail;
|
514 |
for(i = 0; i < l1_size; i++) |
515 |
be64_to_cpus(&l1_table[i]); |
516 |
} |
517 |
if (l1_allocated)
|
518 |
qemu_free(l1_table); |
519 |
qemu_free(l2_table); |
520 |
return 0; |
521 |
fail:
|
522 |
if (l1_allocated)
|
523 |
qemu_free(l1_table); |
524 |
qemu_free(l2_table); |
525 |
return -EIO;
|
526 |
} |
527 |
|
528 |
|
529 |
|
530 |
|
531 |
/*********************************************************/
|
532 |
/* refcount checking functions */
|
533 |
|
534 |
|
535 |
|
536 |
/*
|
537 |
* Increases the refcount for a range of clusters in a given refcount table.
|
538 |
* This is used to construct a temporary refcount table out of L1 and L2 tables
|
539 |
* which can be compared the the refcount table saved in the image.
|
540 |
*
|
541 |
* Returns the number of errors in the image that were found
|
542 |
*/
|
543 |
static int inc_refcounts(BlockDriverState *bs, |
544 |
uint16_t *refcount_table, |
545 |
int refcount_table_size,
|
546 |
int64_t offset, int64_t size) |
547 |
{ |
548 |
BDRVQcowState *s = bs->opaque; |
549 |
int64_t start, last, cluster_offset; |
550 |
int k;
|
551 |
int errors = 0; |
552 |
|
553 |
if (size <= 0) |
554 |
return 0; |
555 |
|
556 |
start = offset & ~(s->cluster_size - 1);
|
557 |
last = (offset + size - 1) & ~(s->cluster_size - 1); |
558 |
for(cluster_offset = start; cluster_offset <= last;
|
559 |
cluster_offset += s->cluster_size) { |
560 |
k = cluster_offset >> s->cluster_bits; |
561 |
if (k < 0 || k >= refcount_table_size) { |
562 |
fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n", |
563 |
cluster_offset); |
564 |
errors++; |
565 |
} else {
|
566 |
if (++refcount_table[k] == 0) { |
567 |
fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
|
568 |
"\n", cluster_offset);
|
569 |
errors++; |
570 |
} |
571 |
} |
572 |
} |
573 |
|
574 |
return errors;
|
575 |
} |
576 |
|
577 |
/*
|
578 |
* Increases the refcount in the given refcount table for the all clusters
|
579 |
* referenced in the L2 table. While doing so, performs some checks on L2
|
580 |
* entries.
|
581 |
*
|
582 |
* Returns the number of errors found by the checks or -errno if an internal
|
583 |
* error occurred.
|
584 |
*/
|
585 |
static int check_refcounts_l2(BlockDriverState *bs, |
586 |
uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
|
587 |
int check_copied)
|
588 |
{ |
589 |
BDRVQcowState *s = bs->opaque; |
590 |
uint64_t *l2_table, offset; |
591 |
int i, l2_size, nb_csectors, refcount;
|
592 |
int errors = 0; |
593 |
|
594 |
/* Read L2 table from disk */
|
595 |
l2_size = s->l2_size * sizeof(uint64_t);
|
596 |
l2_table = qemu_malloc(l2_size); |
597 |
|
598 |
if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
|
599 |
goto fail;
|
600 |
|
601 |
/* Do the actual checks */
|
602 |
for(i = 0; i < s->l2_size; i++) { |
603 |
offset = be64_to_cpu(l2_table[i]); |
604 |
if (offset != 0) { |
605 |
if (offset & QCOW_OFLAG_COMPRESSED) {
|
606 |
/* Compressed clusters don't have QCOW_OFLAG_COPIED */
|
607 |
if (offset & QCOW_OFLAG_COPIED) {
|
608 |
fprintf(stderr, "ERROR: cluster %" PRId64 ": " |
609 |
"copied flag must never be set for compressed "
|
610 |
"clusters\n", offset >> s->cluster_bits);
|
611 |
offset &= ~QCOW_OFLAG_COPIED; |
612 |
errors++; |
613 |
} |
614 |
|
615 |
/* Mark cluster as used */
|
616 |
nb_csectors = ((offset >> s->csize_shift) & |
617 |
s->csize_mask) + 1;
|
618 |
offset &= s->cluster_offset_mask; |
619 |
errors += inc_refcounts(bs, refcount_table, |
620 |
refcount_table_size, |
621 |
offset & ~511, nb_csectors * 512); |
622 |
} else {
|
623 |
/* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
|
624 |
if (check_copied) {
|
625 |
uint64_t entry = offset; |
626 |
offset &= ~QCOW_OFLAG_COPIED; |
627 |
refcount = get_refcount(bs, offset >> s->cluster_bits); |
628 |
if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) { |
629 |
fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
|
630 |
PRIx64 " refcount=%d\n", entry, refcount);
|
631 |
errors++; |
632 |
} |
633 |
} |
634 |
|
635 |
/* Mark cluster as used */
|
636 |
offset &= ~QCOW_OFLAG_COPIED; |
637 |
errors += inc_refcounts(bs, refcount_table, |
638 |
refcount_table_size, |
639 |
offset, s->cluster_size); |
640 |
|
641 |
/* Correct offsets are cluster aligned */
|
642 |
if (offset & (s->cluster_size - 1)) { |
643 |
fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not " |
644 |
"properly aligned; L2 entry corrupted.\n", offset);
|
645 |
errors++; |
646 |
} |
647 |
} |
648 |
} |
649 |
} |
650 |
|
651 |
qemu_free(l2_table); |
652 |
return errors;
|
653 |
|
654 |
fail:
|
655 |
fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
|
656 |
qemu_free(l2_table); |
657 |
return -EIO;
|
658 |
} |
659 |
|
660 |
/*
|
661 |
* Increases the refcount for the L1 table, its L2 tables and all referenced
|
662 |
* clusters in the given refcount table. While doing so, performs some checks
|
663 |
* on L1 and L2 entries.
|
664 |
*
|
665 |
* Returns the number of errors found by the checks or -errno if an internal
|
666 |
* error occurred.
|
667 |
*/
|
668 |
static int check_refcounts_l1(BlockDriverState *bs, |
669 |
uint16_t *refcount_table, |
670 |
int refcount_table_size,
|
671 |
int64_t l1_table_offset, int l1_size,
|
672 |
int check_copied)
|
673 |
{ |
674 |
BDRVQcowState *s = bs->opaque; |
675 |
uint64_t *l1_table, l2_offset, l1_size2; |
676 |
int i, refcount, ret;
|
677 |
int errors = 0; |
678 |
|
679 |
l1_size2 = l1_size * sizeof(uint64_t);
|
680 |
|
681 |
/* Mark L1 table as used */
|
682 |
errors += inc_refcounts(bs, refcount_table, refcount_table_size, |
683 |
l1_table_offset, l1_size2); |
684 |
|
685 |
/* Read L1 table entries from disk */
|
686 |
l1_table = qemu_malloc(l1_size2); |
687 |
if (bdrv_pread(s->hd, l1_table_offset,
|
688 |
l1_table, l1_size2) != l1_size2) |
689 |
goto fail;
|
690 |
for(i = 0;i < l1_size; i++) |
691 |
be64_to_cpus(&l1_table[i]); |
692 |
|
693 |
/* Do the actual checks */
|
694 |
for(i = 0; i < l1_size; i++) { |
695 |
l2_offset = l1_table[i]; |
696 |
if (l2_offset) {
|
697 |
/* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
|
698 |
if (check_copied) {
|
699 |
refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) |
700 |
>> s->cluster_bits); |
701 |
if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) { |
702 |
fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
|
703 |
" refcount=%d\n", l2_offset, refcount);
|
704 |
errors++; |
705 |
} |
706 |
} |
707 |
|
708 |
/* Mark L2 table as used */
|
709 |
l2_offset &= ~QCOW_OFLAG_COPIED; |
710 |
errors += inc_refcounts(bs, refcount_table, |
711 |
refcount_table_size, |
712 |
l2_offset, |
713 |
s->cluster_size); |
714 |
|
715 |
/* L2 tables are cluster aligned */
|
716 |
if (l2_offset & (s->cluster_size - 1)) { |
717 |
fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not " |
718 |
"cluster aligned; L1 entry corrupted\n", l2_offset);
|
719 |
errors++; |
720 |
} |
721 |
|
722 |
/* Process and check L2 entries */
|
723 |
ret = check_refcounts_l2(bs, refcount_table, refcount_table_size, |
724 |
l2_offset, check_copied); |
725 |
if (ret < 0) { |
726 |
goto fail;
|
727 |
} |
728 |
errors += ret; |
729 |
} |
730 |
} |
731 |
qemu_free(l1_table); |
732 |
return errors;
|
733 |
|
734 |
fail:
|
735 |
fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
|
736 |
qemu_free(l1_table); |
737 |
return -EIO;
|
738 |
} |
739 |
|
740 |
/*
|
741 |
* Checks an image for refcount consistency.
|
742 |
*
|
743 |
* Returns 0 if no errors are found, the number of errors in case the image is
|
744 |
* detected as corrupted, and -errno when an internal error occured.
|
745 |
*/
|
746 |
int check_refcounts(BlockDriverState *bs)
|
747 |
{ |
748 |
BDRVQcowState *s = bs->opaque; |
749 |
int64_t size; |
750 |
int nb_clusters, refcount1, refcount2, i;
|
751 |
QCowSnapshot *sn; |
752 |
uint16_t *refcount_table; |
753 |
int ret, errors = 0; |
754 |
|
755 |
size = bdrv_getlength(s->hd); |
756 |
nb_clusters = size_to_clusters(s, size); |
757 |
refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
|
758 |
|
759 |
/* header */
|
760 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
761 |
0, s->cluster_size);
|
762 |
|
763 |
/* current L1 table */
|
764 |
ret = check_refcounts_l1(bs, refcount_table, nb_clusters, |
765 |
s->l1_table_offset, s->l1_size, 1);
|
766 |
if (ret < 0) { |
767 |
return ret;
|
768 |
} |
769 |
errors += ret; |
770 |
|
771 |
/* snapshots */
|
772 |
for(i = 0; i < s->nb_snapshots; i++) { |
773 |
sn = s->snapshots + i; |
774 |
check_refcounts_l1(bs, refcount_table, nb_clusters, |
775 |
sn->l1_table_offset, sn->l1_size, 0);
|
776 |
} |
777 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
778 |
s->snapshots_offset, s->snapshots_size); |
779 |
|
780 |
/* refcount data */
|
781 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
782 |
s->refcount_table_offset, |
783 |
s->refcount_table_size * sizeof(uint64_t));
|
784 |
for(i = 0; i < s->refcount_table_size; i++) { |
785 |
int64_t offset; |
786 |
offset = s->refcount_table[i]; |
787 |
if (offset != 0) { |
788 |
errors += inc_refcounts(bs, refcount_table, nb_clusters, |
789 |
offset, s->cluster_size); |
790 |
} |
791 |
} |
792 |
|
793 |
/* compare ref counts */
|
794 |
for(i = 0; i < nb_clusters; i++) { |
795 |
refcount1 = get_refcount(bs, i); |
796 |
refcount2 = refcount_table[i]; |
797 |
if (refcount1 != refcount2) {
|
798 |
fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n",
|
799 |
i, refcount1, refcount2); |
800 |
errors++; |
801 |
} |
802 |
} |
803 |
|
804 |
qemu_free(refcount_table); |
805 |
|
806 |
return errors;
|
807 |
} |
808 |
|