]>
Commit | Line | Data |
---|---|---|
f7d0fe02 KW |
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 | ||
3b88e52b KW |
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 | ||
f7d0fe02 KW |
54 | /*********************************************************/ |
55 | /* refcount handling */ | |
56 | ||
ed6ccf0f | 57 | int qcow2_refcount_init(BlockDriverState *bs) |
f7d0fe02 KW |
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 | ||
ed6ccf0f | 78 | void qcow2_refcount_close(BlockDriverState *bs) |
f7d0fe02 KW |
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; | |
3b88e52b KW |
91 | |
92 | if (cache_refcount_updates) { | |
93 | write_refcount_block(s); | |
94 | } | |
95 | ||
f7d0fe02 KW |
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); | |
f2b7c8b3 KW |
171 | ret = bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset), |
172 | data, sizeof(data)); | |
173 | if (ret != sizeof(data)) { | |
f7d0fe02 | 174 | goto fail; |
f2b7c8b3 KW |
175 | } |
176 | ||
f7d0fe02 KW |
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); | |
ed6ccf0f | 185 | qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t)); |
f7d0fe02 KW |
186 | return 0; |
187 | fail: | |
f7d0fe02 | 188 | qemu_free(new_table); |
f2b7c8b3 | 189 | return ret < 0 ? ret : -EIO; |
f7d0fe02 KW |
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; | |
80ee15a6 KW |
197 | unsigned int refcount_table_index; |
198 | int ret; | |
f7d0fe02 | 199 | uint64_t data64; |
3b88e52b | 200 | int cache = cache_refcount_updates; |
f7d0fe02 KW |
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) { | |
3b88e52b KW |
213 | if (cache_refcount_updates) { |
214 | write_refcount_block(s); | |
215 | cache_refcount_updates = 0; | |
216 | } | |
f7d0fe02 KW |
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); | |
3b88e52b | 235 | cache_refcount_updates = cache; |
f7d0fe02 KW |
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 | ||
9923e05e KW |
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 | ||
3b88e52b KW |
252 | if (cache_refcount_updates) { |
253 | return 0; | |
254 | } | |
255 | ||
9923e05e KW |
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 | ||
f7d0fe02 KW |
271 | /* XXX: cache several refcount block clusters ? */ |
272 | static int update_refcount(BlockDriverState *bs, | |
273 | int64_t offset, int64_t length, | |
274 | int addend) | |
275 | { | |
276 | BDRVQcowState *s = bs->opaque; | |
277 | int64_t start, last, cluster_offset; | |
278 | int64_t refcount_block_offset = 0; | |
279 | int64_t table_index = -1, old_table_index; | |
280 | int first_index = -1, last_index = -1; | |
09508d13 | 281 | int ret; |
f7d0fe02 KW |
282 | |
283 | #ifdef DEBUG_ALLOC2 | |
0bf9e31a | 284 | printf("update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n", |
f7d0fe02 KW |
285 | offset, length, addend); |
286 | #endif | |
7322afe7 | 287 | if (length < 0) { |
f7d0fe02 | 288 | return -EINVAL; |
7322afe7 KW |
289 | } else if (length == 0) { |
290 | return 0; | |
291 | } | |
292 | ||
f7d0fe02 KW |
293 | start = offset & ~(s->cluster_size - 1); |
294 | last = (offset + length - 1) & ~(s->cluster_size - 1); | |
295 | for(cluster_offset = start; cluster_offset <= last; | |
296 | cluster_offset += s->cluster_size) | |
297 | { | |
298 | int block_index, refcount; | |
299 | int64_t cluster_index = cluster_offset >> s->cluster_bits; | |
09508d13 | 300 | int64_t new_block; |
f7d0fe02 KW |
301 | |
302 | /* Only write refcount block to disk when we are done with it */ | |
303 | old_table_index = table_index; | |
304 | table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); | |
305 | if ((old_table_index >= 0) && (table_index != old_table_index)) { | |
9923e05e KW |
306 | |
307 | if (write_refcount_block_entries(s, refcount_block_offset, | |
308 | first_index, last_index) < 0) | |
f7d0fe02 KW |
309 | { |
310 | return -EIO; | |
311 | } | |
312 | ||
313 | first_index = -1; | |
314 | last_index = -1; | |
315 | } | |
316 | ||
317 | /* Load the refcount block and allocate it if needed */ | |
09508d13 KW |
318 | new_block = alloc_refcount_block(bs, cluster_index); |
319 | if (new_block < 0) { | |
320 | ret = new_block; | |
321 | goto fail; | |
f7d0fe02 | 322 | } |
09508d13 | 323 | refcount_block_offset = new_block; |
f7d0fe02 KW |
324 | |
325 | /* we can update the count and save it */ | |
326 | block_index = cluster_index & | |
327 | ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); | |
328 | if (first_index == -1 || block_index < first_index) { | |
329 | first_index = block_index; | |
330 | } | |
331 | if (block_index > last_index) { | |
332 | last_index = block_index; | |
333 | } | |
334 | ||
335 | refcount = be16_to_cpu(s->refcount_block_cache[block_index]); | |
336 | refcount += addend; | |
09508d13 KW |
337 | if (refcount < 0 || refcount > 0xffff) { |
338 | ret = -EINVAL; | |
339 | goto fail; | |
340 | } | |
f7d0fe02 KW |
341 | if (refcount == 0 && cluster_index < s->free_cluster_index) { |
342 | s->free_cluster_index = cluster_index; | |
343 | } | |
344 | s->refcount_block_cache[block_index] = cpu_to_be16(refcount); | |
345 | } | |
346 | ||
09508d13 KW |
347 | ret = 0; |
348 | fail: | |
349 | ||
f7d0fe02 KW |
350 | /* Write last changed block to disk */ |
351 | if (refcount_block_offset != 0) { | |
9923e05e KW |
352 | if (write_refcount_block_entries(s, refcount_block_offset, |
353 | first_index, last_index) < 0) | |
f7d0fe02 | 354 | { |
09508d13 | 355 | return ret < 0 ? ret : -EIO; |
f7d0fe02 KW |
356 | } |
357 | } | |
358 | ||
09508d13 KW |
359 | /* |
360 | * Try do undo any updates if an error is returned (This may succeed in | |
361 | * some cases like ENOSPC for allocating a new refcount block) | |
362 | */ | |
363 | if (ret < 0) { | |
364 | int dummy; | |
365 | dummy = update_refcount(bs, offset, cluster_offset - offset, -addend); | |
366 | } | |
367 | ||
368 | return ret; | |
f7d0fe02 KW |
369 | } |
370 | ||
371 | /* addend must be 1 or -1 */ | |
372 | static int update_cluster_refcount(BlockDriverState *bs, | |
373 | int64_t cluster_index, | |
374 | int addend) | |
375 | { | |
376 | BDRVQcowState *s = bs->opaque; | |
377 | int ret; | |
378 | ||
379 | ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend); | |
380 | if (ret < 0) { | |
381 | return ret; | |
382 | } | |
383 | ||
384 | return get_refcount(bs, cluster_index); | |
385 | } | |
386 | ||
387 | ||
388 | ||
389 | /*********************************************************/ | |
390 | /* cluster allocation functions */ | |
391 | ||
392 | ||
393 | ||
394 | /* return < 0 if error */ | |
395 | static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size) | |
396 | { | |
397 | BDRVQcowState *s = bs->opaque; | |
398 | int i, nb_clusters; | |
399 | ||
400 | nb_clusters = size_to_clusters(s, size); | |
401 | retry: | |
402 | for(i = 0; i < nb_clusters; i++) { | |
403 | int64_t i = s->free_cluster_index++; | |
404 | if (get_refcount(bs, i) != 0) | |
405 | goto retry; | |
406 | } | |
407 | #ifdef DEBUG_ALLOC2 | |
0bf9e31a | 408 | printf("alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n", |
f7d0fe02 KW |
409 | size, |
410 | (s->free_cluster_index - nb_clusters) << s->cluster_bits); | |
411 | #endif | |
412 | return (s->free_cluster_index - nb_clusters) << s->cluster_bits; | |
413 | } | |
414 | ||
ed6ccf0f | 415 | int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size) |
f7d0fe02 KW |
416 | { |
417 | int64_t offset; | |
418 | ||
419 | offset = alloc_clusters_noref(bs, size); | |
420 | update_refcount(bs, offset, size, 1); | |
421 | return offset; | |
422 | } | |
423 | ||
424 | /* only used to allocate compressed sectors. We try to allocate | |
425 | contiguous sectors. size must be <= cluster_size */ | |
ed6ccf0f | 426 | int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size) |
f7d0fe02 KW |
427 | { |
428 | BDRVQcowState *s = bs->opaque; | |
429 | int64_t offset, cluster_offset; | |
430 | int free_in_cluster; | |
431 | ||
432 | assert(size > 0 && size <= s->cluster_size); | |
433 | if (s->free_byte_offset == 0) { | |
ed6ccf0f | 434 | s->free_byte_offset = qcow2_alloc_clusters(bs, s->cluster_size); |
f7d0fe02 KW |
435 | } |
436 | redo: | |
437 | free_in_cluster = s->cluster_size - | |
438 | (s->free_byte_offset & (s->cluster_size - 1)); | |
439 | if (size <= free_in_cluster) { | |
440 | /* enough space in current cluster */ | |
441 | offset = s->free_byte_offset; | |
442 | s->free_byte_offset += size; | |
443 | free_in_cluster -= size; | |
444 | if (free_in_cluster == 0) | |
445 | s->free_byte_offset = 0; | |
446 | if ((offset & (s->cluster_size - 1)) != 0) | |
447 | update_cluster_refcount(bs, offset >> s->cluster_bits, 1); | |
448 | } else { | |
ed6ccf0f | 449 | offset = qcow2_alloc_clusters(bs, s->cluster_size); |
f7d0fe02 KW |
450 | cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1); |
451 | if ((cluster_offset + s->cluster_size) == offset) { | |
452 | /* we are lucky: contiguous data */ | |
453 | offset = s->free_byte_offset; | |
454 | update_cluster_refcount(bs, offset >> s->cluster_bits, 1); | |
455 | s->free_byte_offset += size; | |
456 | } else { | |
457 | s->free_byte_offset = offset; | |
458 | goto redo; | |
459 | } | |
460 | } | |
461 | return offset; | |
462 | } | |
463 | ||
ed6ccf0f | 464 | void qcow2_free_clusters(BlockDriverState *bs, |
f7d0fe02 KW |
465 | int64_t offset, int64_t size) |
466 | { | |
467 | update_refcount(bs, offset, size, -1); | |
468 | } | |
469 | ||
45aba42f KW |
470 | /* |
471 | * free_any_clusters | |
472 | * | |
473 | * free clusters according to its type: compressed or not | |
474 | * | |
475 | */ | |
476 | ||
ed6ccf0f | 477 | void qcow2_free_any_clusters(BlockDriverState *bs, |
45aba42f KW |
478 | uint64_t cluster_offset, int nb_clusters) |
479 | { | |
480 | BDRVQcowState *s = bs->opaque; | |
481 | ||
482 | /* free the cluster */ | |
483 | ||
484 | if (cluster_offset & QCOW_OFLAG_COMPRESSED) { | |
485 | int nb_csectors; | |
486 | nb_csectors = ((cluster_offset >> s->csize_shift) & | |
487 | s->csize_mask) + 1; | |
ed6ccf0f KW |
488 | qcow2_free_clusters(bs, |
489 | (cluster_offset & s->cluster_offset_mask) & ~511, | |
490 | nb_csectors * 512); | |
45aba42f KW |
491 | return; |
492 | } | |
493 | ||
ed6ccf0f | 494 | qcow2_free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits); |
45aba42f KW |
495 | |
496 | return; | |
497 | } | |
498 | ||
f7d0fe02 KW |
499 | |
500 | ||
501 | /*********************************************************/ | |
502 | /* snapshots and image creation */ | |
503 | ||
504 | ||
505 | ||
ed6ccf0f KW |
506 | void qcow2_create_refcount_update(QCowCreateState *s, int64_t offset, |
507 | int64_t size) | |
f7d0fe02 KW |
508 | { |
509 | int refcount; | |
510 | int64_t start, last, cluster_offset; | |
511 | uint16_t *p; | |
512 | ||
513 | start = offset & ~(s->cluster_size - 1); | |
514 | last = (offset + size - 1) & ~(s->cluster_size - 1); | |
515 | for(cluster_offset = start; cluster_offset <= last; | |
516 | cluster_offset += s->cluster_size) { | |
517 | p = &s->refcount_block[cluster_offset >> s->cluster_bits]; | |
518 | refcount = be16_to_cpu(*p); | |
519 | refcount++; | |
520 | *p = cpu_to_be16(refcount); | |
521 | } | |
522 | } | |
523 | ||
524 | /* update the refcounts of snapshots and the copied flag */ | |
ed6ccf0f KW |
525 | int qcow2_update_snapshot_refcount(BlockDriverState *bs, |
526 | int64_t l1_table_offset, int l1_size, int addend) | |
f7d0fe02 KW |
527 | { |
528 | BDRVQcowState *s = bs->opaque; | |
529 | uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated; | |
530 | int64_t old_offset, old_l2_offset; | |
531 | int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount; | |
532 | ||
ed6ccf0f | 533 | qcow2_l2_cache_reset(bs); |
3b88e52b | 534 | cache_refcount_updates = 1; |
f7d0fe02 KW |
535 | |
536 | l2_table = NULL; | |
537 | l1_table = NULL; | |
538 | l1_size2 = l1_size * sizeof(uint64_t); | |
f7d0fe02 | 539 | if (l1_table_offset != s->l1_table_offset) { |
702ef63f KW |
540 | if (l1_size2 != 0) { |
541 | l1_table = qemu_mallocz(align_offset(l1_size2, 512)); | |
542 | } else { | |
543 | l1_table = NULL; | |
544 | } | |
f7d0fe02 KW |
545 | l1_allocated = 1; |
546 | if (bdrv_pread(s->hd, l1_table_offset, | |
547 | l1_table, l1_size2) != l1_size2) | |
548 | goto fail; | |
549 | for(i = 0;i < l1_size; i++) | |
550 | be64_to_cpus(&l1_table[i]); | |
551 | } else { | |
552 | assert(l1_size == s->l1_size); | |
553 | l1_table = s->l1_table; | |
554 | l1_allocated = 0; | |
555 | } | |
556 | ||
557 | l2_size = s->l2_size * sizeof(uint64_t); | |
558 | l2_table = qemu_malloc(l2_size); | |
559 | l1_modified = 0; | |
560 | for(i = 0; i < l1_size; i++) { | |
561 | l2_offset = l1_table[i]; | |
562 | if (l2_offset) { | |
563 | old_l2_offset = l2_offset; | |
564 | l2_offset &= ~QCOW_OFLAG_COPIED; | |
565 | l2_modified = 0; | |
566 | if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size) | |
567 | goto fail; | |
568 | for(j = 0; j < s->l2_size; j++) { | |
569 | offset = be64_to_cpu(l2_table[j]); | |
570 | if (offset != 0) { | |
571 | old_offset = offset; | |
572 | offset &= ~QCOW_OFLAG_COPIED; | |
573 | if (offset & QCOW_OFLAG_COMPRESSED) { | |
574 | nb_csectors = ((offset >> s->csize_shift) & | |
575 | s->csize_mask) + 1; | |
576 | if (addend != 0) | |
577 | update_refcount(bs, (offset & s->cluster_offset_mask) & ~511, | |
578 | nb_csectors * 512, addend); | |
579 | /* compressed clusters are never modified */ | |
580 | refcount = 2; | |
581 | } else { | |
582 | if (addend != 0) { | |
583 | refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend); | |
584 | } else { | |
585 | refcount = get_refcount(bs, offset >> s->cluster_bits); | |
586 | } | |
587 | } | |
588 | ||
589 | if (refcount == 1) { | |
590 | offset |= QCOW_OFLAG_COPIED; | |
591 | } | |
592 | if (offset != old_offset) { | |
593 | l2_table[j] = cpu_to_be64(offset); | |
594 | l2_modified = 1; | |
595 | } | |
596 | } | |
597 | } | |
598 | if (l2_modified) { | |
599 | if (bdrv_pwrite(s->hd, | |
600 | l2_offset, l2_table, l2_size) != l2_size) | |
601 | goto fail; | |
602 | } | |
603 | ||
604 | if (addend != 0) { | |
605 | refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend); | |
606 | } else { | |
607 | refcount = get_refcount(bs, l2_offset >> s->cluster_bits); | |
608 | } | |
609 | if (refcount == 1) { | |
610 | l2_offset |= QCOW_OFLAG_COPIED; | |
611 | } | |
612 | if (l2_offset != old_l2_offset) { | |
613 | l1_table[i] = l2_offset; | |
614 | l1_modified = 1; | |
615 | } | |
616 | } | |
617 | } | |
618 | if (l1_modified) { | |
619 | for(i = 0; i < l1_size; i++) | |
620 | cpu_to_be64s(&l1_table[i]); | |
621 | if (bdrv_pwrite(s->hd, l1_table_offset, l1_table, | |
622 | l1_size2) != l1_size2) | |
623 | goto fail; | |
624 | for(i = 0; i < l1_size; i++) | |
625 | be64_to_cpus(&l1_table[i]); | |
626 | } | |
627 | if (l1_allocated) | |
628 | qemu_free(l1_table); | |
629 | qemu_free(l2_table); | |
3b88e52b KW |
630 | cache_refcount_updates = 0; |
631 | write_refcount_block(s); | |
f7d0fe02 KW |
632 | return 0; |
633 | fail: | |
634 | if (l1_allocated) | |
635 | qemu_free(l1_table); | |
636 | qemu_free(l2_table); | |
3b88e52b KW |
637 | cache_refcount_updates = 0; |
638 | write_refcount_block(s); | |
f7d0fe02 KW |
639 | return -EIO; |
640 | } | |
641 | ||
642 | ||
643 | ||
644 | ||
645 | /*********************************************************/ | |
646 | /* refcount checking functions */ | |
647 | ||
648 | ||
649 | ||
650 | /* | |
651 | * Increases the refcount for a range of clusters in a given refcount table. | |
652 | * This is used to construct a temporary refcount table out of L1 and L2 tables | |
653 | * which can be compared the the refcount table saved in the image. | |
654 | * | |
655 | * Returns the number of errors in the image that were found | |
656 | */ | |
657 | static int inc_refcounts(BlockDriverState *bs, | |
658 | uint16_t *refcount_table, | |
659 | int refcount_table_size, | |
660 | int64_t offset, int64_t size) | |
661 | { | |
662 | BDRVQcowState *s = bs->opaque; | |
663 | int64_t start, last, cluster_offset; | |
664 | int k; | |
665 | int errors = 0; | |
666 | ||
667 | if (size <= 0) | |
668 | return 0; | |
669 | ||
670 | start = offset & ~(s->cluster_size - 1); | |
671 | last = (offset + size - 1) & ~(s->cluster_size - 1); | |
672 | for(cluster_offset = start; cluster_offset <= last; | |
673 | cluster_offset += s->cluster_size) { | |
674 | k = cluster_offset >> s->cluster_bits; | |
675 | if (k < 0 || k >= refcount_table_size) { | |
676 | fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n", | |
677 | cluster_offset); | |
678 | errors++; | |
679 | } else { | |
680 | if (++refcount_table[k] == 0) { | |
681 | fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64 | |
682 | "\n", cluster_offset); | |
683 | errors++; | |
684 | } | |
685 | } | |
686 | } | |
687 | ||
688 | return errors; | |
689 | } | |
690 | ||
691 | /* | |
692 | * Increases the refcount in the given refcount table for the all clusters | |
693 | * referenced in the L2 table. While doing so, performs some checks on L2 | |
694 | * entries. | |
695 | * | |
696 | * Returns the number of errors found by the checks or -errno if an internal | |
697 | * error occurred. | |
698 | */ | |
699 | static int check_refcounts_l2(BlockDriverState *bs, | |
700 | uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset, | |
701 | int check_copied) | |
702 | { | |
703 | BDRVQcowState *s = bs->opaque; | |
704 | uint64_t *l2_table, offset; | |
705 | int i, l2_size, nb_csectors, refcount; | |
706 | int errors = 0; | |
707 | ||
708 | /* Read L2 table from disk */ | |
709 | l2_size = s->l2_size * sizeof(uint64_t); | |
710 | l2_table = qemu_malloc(l2_size); | |
711 | ||
712 | if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size) | |
713 | goto fail; | |
714 | ||
715 | /* Do the actual checks */ | |
716 | for(i = 0; i < s->l2_size; i++) { | |
717 | offset = be64_to_cpu(l2_table[i]); | |
718 | if (offset != 0) { | |
719 | if (offset & QCOW_OFLAG_COMPRESSED) { | |
720 | /* Compressed clusters don't have QCOW_OFLAG_COPIED */ | |
721 | if (offset & QCOW_OFLAG_COPIED) { | |
722 | fprintf(stderr, "ERROR: cluster %" PRId64 ": " | |
723 | "copied flag must never be set for compressed " | |
724 | "clusters\n", offset >> s->cluster_bits); | |
725 | offset &= ~QCOW_OFLAG_COPIED; | |
726 | errors++; | |
727 | } | |
728 | ||
729 | /* Mark cluster as used */ | |
730 | nb_csectors = ((offset >> s->csize_shift) & | |
731 | s->csize_mask) + 1; | |
732 | offset &= s->cluster_offset_mask; | |
733 | errors += inc_refcounts(bs, refcount_table, | |
734 | refcount_table_size, | |
735 | offset & ~511, nb_csectors * 512); | |
736 | } else { | |
737 | /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */ | |
738 | if (check_copied) { | |
739 | uint64_t entry = offset; | |
740 | offset &= ~QCOW_OFLAG_COPIED; | |
741 | refcount = get_refcount(bs, offset >> s->cluster_bits); | |
742 | if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) { | |
743 | fprintf(stderr, "ERROR OFLAG_COPIED: offset=%" | |
744 | PRIx64 " refcount=%d\n", entry, refcount); | |
745 | errors++; | |
746 | } | |
747 | } | |
748 | ||
749 | /* Mark cluster as used */ | |
750 | offset &= ~QCOW_OFLAG_COPIED; | |
751 | errors += inc_refcounts(bs, refcount_table, | |
752 | refcount_table_size, | |
753 | offset, s->cluster_size); | |
754 | ||
755 | /* Correct offsets are cluster aligned */ | |
756 | if (offset & (s->cluster_size - 1)) { | |
757 | fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not " | |
758 | "properly aligned; L2 entry corrupted.\n", offset); | |
759 | errors++; | |
760 | } | |
761 | } | |
762 | } | |
763 | } | |
764 | ||
765 | qemu_free(l2_table); | |
766 | return errors; | |
767 | ||
768 | fail: | |
769 | fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n"); | |
770 | qemu_free(l2_table); | |
771 | return -EIO; | |
772 | } | |
773 | ||
774 | /* | |
775 | * Increases the refcount for the L1 table, its L2 tables and all referenced | |
776 | * clusters in the given refcount table. While doing so, performs some checks | |
777 | * on L1 and L2 entries. | |
778 | * | |
779 | * Returns the number of errors found by the checks or -errno if an internal | |
780 | * error occurred. | |
781 | */ | |
782 | static int check_refcounts_l1(BlockDriverState *bs, | |
783 | uint16_t *refcount_table, | |
784 | int refcount_table_size, | |
785 | int64_t l1_table_offset, int l1_size, | |
786 | int check_copied) | |
787 | { | |
788 | BDRVQcowState *s = bs->opaque; | |
789 | uint64_t *l1_table, l2_offset, l1_size2; | |
790 | int i, refcount, ret; | |
791 | int errors = 0; | |
792 | ||
793 | l1_size2 = l1_size * sizeof(uint64_t); | |
794 | ||
795 | /* Mark L1 table as used */ | |
796 | errors += inc_refcounts(bs, refcount_table, refcount_table_size, | |
797 | l1_table_offset, l1_size2); | |
798 | ||
799 | /* Read L1 table entries from disk */ | |
702ef63f KW |
800 | if (l1_size2 == 0) { |
801 | l1_table = NULL; | |
802 | } else { | |
803 | l1_table = qemu_malloc(l1_size2); | |
804 | if (bdrv_pread(s->hd, l1_table_offset, | |
805 | l1_table, l1_size2) != l1_size2) | |
806 | goto fail; | |
807 | for(i = 0;i < l1_size; i++) | |
808 | be64_to_cpus(&l1_table[i]); | |
809 | } | |
f7d0fe02 KW |
810 | |
811 | /* Do the actual checks */ | |
812 | for(i = 0; i < l1_size; i++) { | |
813 | l2_offset = l1_table[i]; | |
814 | if (l2_offset) { | |
815 | /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */ | |
816 | if (check_copied) { | |
817 | refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) | |
818 | >> s->cluster_bits); | |
819 | if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) { | |
820 | fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64 | |
821 | " refcount=%d\n", l2_offset, refcount); | |
822 | errors++; | |
823 | } | |
824 | } | |
825 | ||
826 | /* Mark L2 table as used */ | |
827 | l2_offset &= ~QCOW_OFLAG_COPIED; | |
828 | errors += inc_refcounts(bs, refcount_table, | |
829 | refcount_table_size, | |
830 | l2_offset, | |
831 | s->cluster_size); | |
832 | ||
833 | /* L2 tables are cluster aligned */ | |
834 | if (l2_offset & (s->cluster_size - 1)) { | |
835 | fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not " | |
836 | "cluster aligned; L1 entry corrupted\n", l2_offset); | |
837 | errors++; | |
838 | } | |
839 | ||
840 | /* Process and check L2 entries */ | |
841 | ret = check_refcounts_l2(bs, refcount_table, refcount_table_size, | |
842 | l2_offset, check_copied); | |
843 | if (ret < 0) { | |
844 | goto fail; | |
845 | } | |
846 | errors += ret; | |
847 | } | |
848 | } | |
849 | qemu_free(l1_table); | |
850 | return errors; | |
851 | ||
852 | fail: | |
853 | fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n"); | |
854 | qemu_free(l1_table); | |
855 | return -EIO; | |
856 | } | |
857 | ||
858 | /* | |
859 | * Checks an image for refcount consistency. | |
860 | * | |
861 | * Returns 0 if no errors are found, the number of errors in case the image is | |
862 | * detected as corrupted, and -errno when an internal error occured. | |
863 | */ | |
ed6ccf0f | 864 | int qcow2_check_refcounts(BlockDriverState *bs) |
f7d0fe02 KW |
865 | { |
866 | BDRVQcowState *s = bs->opaque; | |
867 | int64_t size; | |
868 | int nb_clusters, refcount1, refcount2, i; | |
869 | QCowSnapshot *sn; | |
870 | uint16_t *refcount_table; | |
871 | int ret, errors = 0; | |
872 | ||
873 | size = bdrv_getlength(s->hd); | |
874 | nb_clusters = size_to_clusters(s, size); | |
875 | refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t)); | |
876 | ||
877 | /* header */ | |
878 | errors += inc_refcounts(bs, refcount_table, nb_clusters, | |
879 | 0, s->cluster_size); | |
880 | ||
881 | /* current L1 table */ | |
882 | ret = check_refcounts_l1(bs, refcount_table, nb_clusters, | |
883 | s->l1_table_offset, s->l1_size, 1); | |
884 | if (ret < 0) { | |
885 | return ret; | |
886 | } | |
887 | errors += ret; | |
888 | ||
889 | /* snapshots */ | |
890 | for(i = 0; i < s->nb_snapshots; i++) { | |
891 | sn = s->snapshots + i; | |
892 | check_refcounts_l1(bs, refcount_table, nb_clusters, | |
893 | sn->l1_table_offset, sn->l1_size, 0); | |
894 | } | |
895 | errors += inc_refcounts(bs, refcount_table, nb_clusters, | |
896 | s->snapshots_offset, s->snapshots_size); | |
897 | ||
898 | /* refcount data */ | |
899 | errors += inc_refcounts(bs, refcount_table, nb_clusters, | |
900 | s->refcount_table_offset, | |
901 | s->refcount_table_size * sizeof(uint64_t)); | |
902 | for(i = 0; i < s->refcount_table_size; i++) { | |
903 | int64_t offset; | |
904 | offset = s->refcount_table[i]; | |
905 | if (offset != 0) { | |
906 | errors += inc_refcounts(bs, refcount_table, nb_clusters, | |
907 | offset, s->cluster_size); | |
908 | } | |
909 | } | |
910 | ||
911 | /* compare ref counts */ | |
912 | for(i = 0; i < nb_clusters; i++) { | |
913 | refcount1 = get_refcount(bs, i); | |
914 | refcount2 = refcount_table[i]; | |
915 | if (refcount1 != refcount2) { | |
916 | fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n", | |
917 | i, refcount1, refcount2); | |
918 | errors++; | |
919 | } | |
920 | } | |
921 | ||
922 | qemu_free(refcount_table); | |
923 | ||
924 | return errors; | |
925 | } | |
926 |