]>
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" | |
737e150e | 26 | #include "block/block_int.h" |
f7d0fe02 | 27 | #include "block/qcow2.h" |
a40f1c2a | 28 | #include "qemu/range.h" |
f7d0fe02 | 29 | |
bb572aef | 30 | static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size); |
92dcb59f | 31 | static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, |
0e06528e | 32 | int64_t offset, int64_t length, uint64_t addend, |
2aabe7c7 | 33 | bool decrease, enum qcow2_discard_type type); |
f7d0fe02 | 34 | |
7453c96b HR |
35 | static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index); |
36 | ||
37 | static void set_refcount_ro4(void *refcount_array, uint64_t index, | |
38 | uint64_t value); | |
39 | ||
3b88e52b | 40 | |
f7d0fe02 KW |
41 | /*********************************************************/ |
42 | /* refcount handling */ | |
43 | ||
ed6ccf0f | 44 | int qcow2_refcount_init(BlockDriverState *bs) |
f7d0fe02 KW |
45 | { |
46 | BDRVQcowState *s = bs->opaque; | |
5dab2fad KW |
47 | unsigned int refcount_table_size2, i; |
48 | int ret; | |
f7d0fe02 | 49 | |
7453c96b HR |
50 | s->get_refcount = &get_refcount_ro4; |
51 | s->set_refcount = &set_refcount_ro4; | |
52 | ||
5dab2fad | 53 | assert(s->refcount_table_size <= INT_MAX / sizeof(uint64_t)); |
f7d0fe02 | 54 | refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t); |
de82815d KW |
55 | s->refcount_table = g_try_malloc(refcount_table_size2); |
56 | ||
f7d0fe02 | 57 | if (s->refcount_table_size > 0) { |
de82815d | 58 | if (s->refcount_table == NULL) { |
8fcffa98 | 59 | ret = -ENOMEM; |
de82815d KW |
60 | goto fail; |
61 | } | |
66f82cee KW |
62 | BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD); |
63 | ret = bdrv_pread(bs->file, s->refcount_table_offset, | |
f7d0fe02 | 64 | s->refcount_table, refcount_table_size2); |
8fcffa98 | 65 | if (ret < 0) { |
f7d0fe02 | 66 | goto fail; |
8fcffa98 | 67 | } |
f7d0fe02 KW |
68 | for(i = 0; i < s->refcount_table_size; i++) |
69 | be64_to_cpus(&s->refcount_table[i]); | |
70 | } | |
71 | return 0; | |
72 | fail: | |
8fcffa98 | 73 | return ret; |
f7d0fe02 KW |
74 | } |
75 | ||
ed6ccf0f | 76 | void qcow2_refcount_close(BlockDriverState *bs) |
f7d0fe02 KW |
77 | { |
78 | BDRVQcowState *s = bs->opaque; | |
7267c094 | 79 | g_free(s->refcount_table); |
f7d0fe02 KW |
80 | } |
81 | ||
82 | ||
7453c96b HR |
83 | static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index) |
84 | { | |
85 | return be16_to_cpu(((const uint16_t *)refcount_array)[index]); | |
86 | } | |
87 | ||
88 | static void set_refcount_ro4(void *refcount_array, uint64_t index, | |
89 | uint64_t value) | |
90 | { | |
91 | assert(!(value >> 16)); | |
92 | ((uint16_t *)refcount_array)[index] = cpu_to_be16(value); | |
93 | } | |
94 | ||
95 | ||
f7d0fe02 | 96 | static int load_refcount_block(BlockDriverState *bs, |
29c1a730 KW |
97 | int64_t refcount_block_offset, |
98 | void **refcount_block) | |
f7d0fe02 KW |
99 | { |
100 | BDRVQcowState *s = bs->opaque; | |
101 | int ret; | |
3b88e52b | 102 | |
66f82cee | 103 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD); |
29c1a730 KW |
104 | ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, |
105 | refcount_block); | |
e14e8ba5 | 106 | |
29c1a730 | 107 | return ret; |
f7d0fe02 KW |
108 | } |
109 | ||
018faafd | 110 | /* |
7324c10f HR |
111 | * Retrieves the refcount of the cluster given by its index and stores it in |
112 | * *refcount. Returns 0 on success and -errno on failure. | |
018faafd | 113 | */ |
7324c10f | 114 | int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index, |
0e06528e | 115 | uint64_t *refcount) |
f7d0fe02 KW |
116 | { |
117 | BDRVQcowState *s = bs->opaque; | |
db8a31d1 | 118 | uint64_t refcount_table_index, block_index; |
f7d0fe02 | 119 | int64_t refcount_block_offset; |
018faafd | 120 | int ret; |
7453c96b | 121 | void *refcount_block; |
f7d0fe02 | 122 | |
17bd5f47 | 123 | refcount_table_index = cluster_index >> s->refcount_block_bits; |
7324c10f HR |
124 | if (refcount_table_index >= s->refcount_table_size) { |
125 | *refcount = 0; | |
f7d0fe02 | 126 | return 0; |
7324c10f | 127 | } |
26d49c46 HR |
128 | refcount_block_offset = |
129 | s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK; | |
7324c10f HR |
130 | if (!refcount_block_offset) { |
131 | *refcount = 0; | |
f7d0fe02 | 132 | return 0; |
7324c10f | 133 | } |
29c1a730 | 134 | |
a97c67ee HR |
135 | if (offset_into_cluster(s, refcount_block_offset)) { |
136 | qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" PRIx64 | |
137 | " unaligned (reftable index: %#" PRIx64 ")", | |
138 | refcount_block_offset, refcount_table_index); | |
139 | return -EIO; | |
140 | } | |
141 | ||
29c1a730 | 142 | ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, |
7453c96b | 143 | &refcount_block); |
29c1a730 KW |
144 | if (ret < 0) { |
145 | return ret; | |
f7d0fe02 | 146 | } |
29c1a730 | 147 | |
17bd5f47 | 148 | block_index = cluster_index & (s->refcount_block_size - 1); |
7453c96b | 149 | *refcount = s->get_refcount(refcount_block, block_index); |
29c1a730 | 150 | |
7453c96b | 151 | ret = qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block); |
29c1a730 KW |
152 | if (ret < 0) { |
153 | return ret; | |
154 | } | |
155 | ||
7324c10f | 156 | return 0; |
f7d0fe02 KW |
157 | } |
158 | ||
05121aed KW |
159 | /* |
160 | * Rounds the refcount table size up to avoid growing the table for each single | |
161 | * refcount block that is allocated. | |
162 | */ | |
163 | static unsigned int next_refcount_table_size(BDRVQcowState *s, | |
164 | unsigned int min_size) | |
165 | { | |
166 | unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1; | |
167 | unsigned int refcount_table_clusters = | |
168 | MAX(1, s->refcount_table_size >> (s->cluster_bits - 3)); | |
169 | ||
170 | while (min_clusters > refcount_table_clusters) { | |
171 | refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2; | |
172 | } | |
173 | ||
174 | return refcount_table_clusters << (s->cluster_bits - 3); | |
175 | } | |
176 | ||
92dcb59f KW |
177 | |
178 | /* Checks if two offsets are described by the same refcount block */ | |
179 | static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a, | |
180 | uint64_t offset_b) | |
181 | { | |
17bd5f47 HR |
182 | uint64_t block_a = offset_a >> (s->cluster_bits + s->refcount_block_bits); |
183 | uint64_t block_b = offset_b >> (s->cluster_bits + s->refcount_block_bits); | |
92dcb59f KW |
184 | |
185 | return (block_a == block_b); | |
186 | } | |
187 | ||
188 | /* | |
189 | * Loads a refcount block. If it doesn't exist yet, it is allocated first | |
190 | * (including growing the refcount table if needed). | |
191 | * | |
29c1a730 | 192 | * Returns 0 on success or -errno in error case |
92dcb59f | 193 | */ |
29c1a730 | 194 | static int alloc_refcount_block(BlockDriverState *bs, |
7453c96b | 195 | int64_t cluster_index, void **refcount_block) |
f7d0fe02 KW |
196 | { |
197 | BDRVQcowState *s = bs->opaque; | |
92dcb59f KW |
198 | unsigned int refcount_table_index; |
199 | int ret; | |
200 | ||
66f82cee | 201 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); |
8252278a | 202 | |
92dcb59f | 203 | /* Find the refcount block for the given cluster */ |
17bd5f47 | 204 | refcount_table_index = cluster_index >> s->refcount_block_bits; |
92dcb59f KW |
205 | |
206 | if (refcount_table_index < s->refcount_table_size) { | |
207 | ||
208 | uint64_t refcount_block_offset = | |
76dc9e0c | 209 | s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK; |
92dcb59f KW |
210 | |
211 | /* If it's already there, we're done */ | |
212 | if (refcount_block_offset) { | |
a97c67ee HR |
213 | if (offset_into_cluster(s, refcount_block_offset)) { |
214 | qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" | |
215 | PRIx64 " unaligned (reftable index: " | |
216 | "%#x)", refcount_block_offset, | |
217 | refcount_table_index); | |
218 | return -EIO; | |
219 | } | |
220 | ||
29c1a730 | 221 | return load_refcount_block(bs, refcount_block_offset, |
7453c96b | 222 | refcount_block); |
92dcb59f KW |
223 | } |
224 | } | |
225 | ||
226 | /* | |
227 | * If we came here, we need to allocate something. Something is at least | |
228 | * a cluster for the new refcount block. It may also include a new refcount | |
229 | * table if the old refcount table is too small. | |
230 | * | |
231 | * Note that allocating clusters here needs some special care: | |
232 | * | |
233 | * - We can't use the normal qcow2_alloc_clusters(), it would try to | |
234 | * increase the refcount and very likely we would end up with an endless | |
235 | * recursion. Instead we must place the refcount blocks in a way that | |
236 | * they can describe them themselves. | |
237 | * | |
238 | * - We need to consider that at this point we are inside update_refcounts | |
b106ad91 KW |
239 | * and potentially doing an initial refcount increase. This means that |
240 | * some clusters have already been allocated by the caller, but their | |
241 | * refcount isn't accurate yet. If we allocate clusters for metadata, we | |
242 | * need to return -EAGAIN to signal the caller that it needs to restart | |
243 | * the search for free clusters. | |
92dcb59f KW |
244 | * |
245 | * - alloc_clusters_noref and qcow2_free_clusters may load a different | |
246 | * refcount block into the cache | |
247 | */ | |
248 | ||
29c1a730 KW |
249 | *refcount_block = NULL; |
250 | ||
251 | /* We write to the refcount table, so we might depend on L2 tables */ | |
9991923b SH |
252 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
253 | if (ret < 0) { | |
254 | return ret; | |
255 | } | |
92dcb59f KW |
256 | |
257 | /* Allocate the refcount block itself and mark it as used */ | |
2eaa8f63 KW |
258 | int64_t new_block = alloc_clusters_noref(bs, s->cluster_size); |
259 | if (new_block < 0) { | |
260 | return new_block; | |
261 | } | |
f7d0fe02 | 262 | |
f7d0fe02 | 263 | #ifdef DEBUG_ALLOC2 |
92dcb59f KW |
264 | fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64 |
265 | " at %" PRIx64 "\n", | |
266 | refcount_table_index, cluster_index << s->cluster_bits, new_block); | |
f7d0fe02 | 267 | #endif |
92dcb59f KW |
268 | |
269 | if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) { | |
25408c09 | 270 | /* Zero the new refcount block before updating it */ |
29c1a730 | 271 | ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, |
7453c96b | 272 | refcount_block); |
29c1a730 KW |
273 | if (ret < 0) { |
274 | goto fail_block; | |
275 | } | |
276 | ||
277 | memset(*refcount_block, 0, s->cluster_size); | |
25408c09 | 278 | |
92dcb59f KW |
279 | /* The block describes itself, need to update the cache */ |
280 | int block_index = (new_block >> s->cluster_bits) & | |
17bd5f47 | 281 | (s->refcount_block_size - 1); |
7453c96b | 282 | s->set_refcount(*refcount_block, block_index, 1); |
92dcb59f KW |
283 | } else { |
284 | /* Described somewhere else. This can recurse at most twice before we | |
285 | * arrive at a block that describes itself. */ | |
2aabe7c7 | 286 | ret = update_refcount(bs, new_block, s->cluster_size, 1, false, |
6cfcb9b8 | 287 | QCOW2_DISCARD_NEVER); |
92dcb59f KW |
288 | if (ret < 0) { |
289 | goto fail_block; | |
290 | } | |
25408c09 | 291 | |
9991923b SH |
292 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
293 | if (ret < 0) { | |
294 | goto fail_block; | |
295 | } | |
1c4c2814 | 296 | |
25408c09 KW |
297 | /* Initialize the new refcount block only after updating its refcount, |
298 | * update_refcount uses the refcount cache itself */ | |
29c1a730 | 299 | ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, |
7453c96b | 300 | refcount_block); |
29c1a730 KW |
301 | if (ret < 0) { |
302 | goto fail_block; | |
303 | } | |
304 | ||
305 | memset(*refcount_block, 0, s->cluster_size); | |
92dcb59f KW |
306 | } |
307 | ||
308 | /* Now the new refcount block needs to be written to disk */ | |
66f82cee | 309 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE); |
29c1a730 KW |
310 | qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block); |
311 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
92dcb59f KW |
312 | if (ret < 0) { |
313 | goto fail_block; | |
314 | } | |
315 | ||
316 | /* If the refcount table is big enough, just hook the block up there */ | |
317 | if (refcount_table_index < s->refcount_table_size) { | |
318 | uint64_t data64 = cpu_to_be64(new_block); | |
66f82cee | 319 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP); |
8b3b7206 | 320 | ret = bdrv_pwrite_sync(bs->file, |
92dcb59f KW |
321 | s->refcount_table_offset + refcount_table_index * sizeof(uint64_t), |
322 | &data64, sizeof(data64)); | |
323 | if (ret < 0) { | |
324 | goto fail_block; | |
325 | } | |
326 | ||
327 | s->refcount_table[refcount_table_index] = new_block; | |
b106ad91 KW |
328 | |
329 | /* The new refcount block may be where the caller intended to put its | |
330 | * data, so let it restart the search. */ | |
331 | return -EAGAIN; | |
29c1a730 KW |
332 | } |
333 | ||
7453c96b | 334 | ret = qcow2_cache_put(bs, s->refcount_block_cache, refcount_block); |
29c1a730 KW |
335 | if (ret < 0) { |
336 | goto fail_block; | |
92dcb59f KW |
337 | } |
338 | ||
339 | /* | |
340 | * If we come here, we need to grow the refcount table. Again, a new | |
341 | * refcount table needs some space and we can't simply allocate to avoid | |
342 | * endless recursion. | |
343 | * | |
344 | * Therefore let's grab new refcount blocks at the end of the image, which | |
345 | * will describe themselves and the new refcount table. This way we can | |
346 | * reference them only in the new table and do the switch to the new | |
347 | * refcount table at once without producing an inconsistent state in | |
348 | * between. | |
349 | */ | |
66f82cee | 350 | BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW); |
8252278a | 351 | |
92dcb59f | 352 | /* Calculate the number of refcount blocks needed so far */ |
17bd5f47 | 353 | uint64_t blocks_used = DIV_ROUND_UP(cluster_index, s->refcount_block_size); |
92dcb59f | 354 | |
2b5d5953 KW |
355 | if (blocks_used > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) { |
356 | return -EFBIG; | |
357 | } | |
358 | ||
92dcb59f KW |
359 | /* And now we need at least one block more for the new metadata */ |
360 | uint64_t table_size = next_refcount_table_size(s, blocks_used + 1); | |
361 | uint64_t last_table_size; | |
362 | uint64_t blocks_clusters; | |
363 | do { | |
a3548077 KW |
364 | uint64_t table_clusters = |
365 | size_to_clusters(s, table_size * sizeof(uint64_t)); | |
92dcb59f | 366 | blocks_clusters = 1 + |
17bd5f47 HR |
367 | ((table_clusters + s->refcount_block_size - 1) |
368 | / s->refcount_block_size); | |
92dcb59f KW |
369 | uint64_t meta_clusters = table_clusters + blocks_clusters; |
370 | ||
371 | last_table_size = table_size; | |
372 | table_size = next_refcount_table_size(s, blocks_used + | |
17bd5f47 HR |
373 | ((meta_clusters + s->refcount_block_size - 1) |
374 | / s->refcount_block_size)); | |
92dcb59f KW |
375 | |
376 | } while (last_table_size != table_size); | |
377 | ||
378 | #ifdef DEBUG_ALLOC2 | |
379 | fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n", | |
380 | s->refcount_table_size, table_size); | |
381 | #endif | |
382 | ||
383 | /* Create the new refcount table and blocks */ | |
17bd5f47 | 384 | uint64_t meta_offset = (blocks_used * s->refcount_block_size) * |
92dcb59f KW |
385 | s->cluster_size; |
386 | uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size; | |
5839e53b | 387 | uint64_t *new_table = g_try_new0(uint64_t, table_size); |
7453c96b | 388 | void *new_blocks = g_try_malloc0(blocks_clusters * s->cluster_size); |
de82815d KW |
389 | |
390 | assert(table_size > 0 && blocks_clusters > 0); | |
391 | if (new_table == NULL || new_blocks == NULL) { | |
392 | ret = -ENOMEM; | |
393 | goto fail_table; | |
394 | } | |
92dcb59f | 395 | |
92dcb59f | 396 | /* Fill the new refcount table */ |
f7d0fe02 | 397 | memcpy(new_table, s->refcount_table, |
92dcb59f KW |
398 | s->refcount_table_size * sizeof(uint64_t)); |
399 | new_table[refcount_table_index] = new_block; | |
400 | ||
401 | int i; | |
402 | for (i = 0; i < blocks_clusters; i++) { | |
403 | new_table[blocks_used + i] = meta_offset + (i * s->cluster_size); | |
404 | } | |
405 | ||
406 | /* Fill the refcount blocks */ | |
407 | uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t)); | |
408 | int block = 0; | |
409 | for (i = 0; i < table_clusters + blocks_clusters; i++) { | |
7453c96b | 410 | s->set_refcount(new_blocks, block++, 1); |
92dcb59f KW |
411 | } |
412 | ||
413 | /* Write refcount blocks to disk */ | |
66f82cee | 414 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS); |
8b3b7206 | 415 | ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks, |
92dcb59f | 416 | blocks_clusters * s->cluster_size); |
7267c094 | 417 | g_free(new_blocks); |
39ba3bf6 | 418 | new_blocks = NULL; |
92dcb59f KW |
419 | if (ret < 0) { |
420 | goto fail_table; | |
421 | } | |
422 | ||
423 | /* Write refcount table to disk */ | |
424 | for(i = 0; i < table_size; i++) { | |
425 | cpu_to_be64s(&new_table[i]); | |
426 | } | |
427 | ||
66f82cee | 428 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE); |
8b3b7206 | 429 | ret = bdrv_pwrite_sync(bs->file, table_offset, new_table, |
92dcb59f KW |
430 | table_size * sizeof(uint64_t)); |
431 | if (ret < 0) { | |
432 | goto fail_table; | |
433 | } | |
434 | ||
435 | for(i = 0; i < table_size; i++) { | |
87267753 | 436 | be64_to_cpus(&new_table[i]); |
92dcb59f | 437 | } |
f7d0fe02 | 438 | |
92dcb59f KW |
439 | /* Hook up the new refcount table in the qcow2 header */ |
440 | uint8_t data[12]; | |
f7d0fe02 | 441 | cpu_to_be64w((uint64_t*)data, table_offset); |
92dcb59f | 442 | cpu_to_be32w((uint32_t*)(data + 8), table_clusters); |
66f82cee | 443 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE); |
8b3b7206 | 444 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset), |
92dcb59f KW |
445 | data, sizeof(data)); |
446 | if (ret < 0) { | |
447 | goto fail_table; | |
f2b7c8b3 KW |
448 | } |
449 | ||
92dcb59f KW |
450 | /* And switch it in memory */ |
451 | uint64_t old_table_offset = s->refcount_table_offset; | |
452 | uint64_t old_table_size = s->refcount_table_size; | |
453 | ||
7267c094 | 454 | g_free(s->refcount_table); |
f7d0fe02 | 455 | s->refcount_table = new_table; |
92dcb59f | 456 | s->refcount_table_size = table_size; |
f7d0fe02 KW |
457 | s->refcount_table_offset = table_offset; |
458 | ||
b106ad91 | 459 | /* Free old table. */ |
6cfcb9b8 KW |
460 | qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t), |
461 | QCOW2_DISCARD_OTHER); | |
f7d0fe02 | 462 | |
7453c96b | 463 | ret = load_refcount_block(bs, new_block, refcount_block); |
92dcb59f | 464 | if (ret < 0) { |
29c1a730 | 465 | return ret; |
f7d0fe02 KW |
466 | } |
467 | ||
b106ad91 KW |
468 | /* If we were trying to do the initial refcount update for some cluster |
469 | * allocation, we might have used the same clusters to store newly | |
470 | * allocated metadata. Make the caller search some new space. */ | |
471 | return -EAGAIN; | |
f7d0fe02 | 472 | |
92dcb59f | 473 | fail_table: |
de82815d | 474 | g_free(new_blocks); |
7267c094 | 475 | g_free(new_table); |
92dcb59f | 476 | fail_block: |
29c1a730 | 477 | if (*refcount_block != NULL) { |
7453c96b | 478 | qcow2_cache_put(bs, s->refcount_block_cache, refcount_block); |
3b88e52b | 479 | } |
29c1a730 | 480 | return ret; |
9923e05e KW |
481 | } |
482 | ||
0b919fae KW |
483 | void qcow2_process_discards(BlockDriverState *bs, int ret) |
484 | { | |
485 | BDRVQcowState *s = bs->opaque; | |
486 | Qcow2DiscardRegion *d, *next; | |
487 | ||
488 | QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) { | |
489 | QTAILQ_REMOVE(&s->discards, d, next); | |
490 | ||
491 | /* Discard is optional, ignore the return value */ | |
492 | if (ret >= 0) { | |
493 | bdrv_discard(bs->file, | |
494 | d->offset >> BDRV_SECTOR_BITS, | |
495 | d->bytes >> BDRV_SECTOR_BITS); | |
496 | } | |
497 | ||
498 | g_free(d); | |
499 | } | |
500 | } | |
501 | ||
502 | static void update_refcount_discard(BlockDriverState *bs, | |
503 | uint64_t offset, uint64_t length) | |
504 | { | |
505 | BDRVQcowState *s = bs->opaque; | |
506 | Qcow2DiscardRegion *d, *p, *next; | |
507 | ||
508 | QTAILQ_FOREACH(d, &s->discards, next) { | |
509 | uint64_t new_start = MIN(offset, d->offset); | |
510 | uint64_t new_end = MAX(offset + length, d->offset + d->bytes); | |
511 | ||
512 | if (new_end - new_start <= length + d->bytes) { | |
513 | /* There can't be any overlap, areas ending up here have no | |
514 | * references any more and therefore shouldn't get freed another | |
515 | * time. */ | |
516 | assert(d->bytes + length == new_end - new_start); | |
517 | d->offset = new_start; | |
518 | d->bytes = new_end - new_start; | |
519 | goto found; | |
520 | } | |
521 | } | |
522 | ||
523 | d = g_malloc(sizeof(*d)); | |
524 | *d = (Qcow2DiscardRegion) { | |
525 | .bs = bs, | |
526 | .offset = offset, | |
527 | .bytes = length, | |
528 | }; | |
529 | QTAILQ_INSERT_TAIL(&s->discards, d, next); | |
530 | ||
531 | found: | |
532 | /* Merge discard requests if they are adjacent now */ | |
533 | QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) { | |
534 | if (p == d | |
535 | || p->offset > d->offset + d->bytes | |
536 | || d->offset > p->offset + p->bytes) | |
537 | { | |
538 | continue; | |
539 | } | |
540 | ||
541 | /* Still no overlap possible */ | |
542 | assert(p->offset == d->offset + d->bytes | |
543 | || d->offset == p->offset + p->bytes); | |
544 | ||
545 | QTAILQ_REMOVE(&s->discards, p, next); | |
546 | d->offset = MIN(d->offset, p->offset); | |
547 | d->bytes += p->bytes; | |
d8bb71b6 | 548 | g_free(p); |
0b919fae KW |
549 | } |
550 | } | |
551 | ||
f7d0fe02 | 552 | /* XXX: cache several refcount block clusters ? */ |
2aabe7c7 HR |
553 | /* @addend is the absolute value of the addend; if @decrease is set, @addend |
554 | * will be subtracted from the current refcount, otherwise it will be added */ | |
db3a964f | 555 | static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, |
2aabe7c7 HR |
556 | int64_t offset, |
557 | int64_t length, | |
0e06528e | 558 | uint64_t addend, |
2aabe7c7 HR |
559 | bool decrease, |
560 | enum qcow2_discard_type type) | |
f7d0fe02 KW |
561 | { |
562 | BDRVQcowState *s = bs->opaque; | |
563 | int64_t start, last, cluster_offset; | |
7453c96b | 564 | void *refcount_block = NULL; |
29c1a730 | 565 | int64_t old_table_index = -1; |
09508d13 | 566 | int ret; |
f7d0fe02 KW |
567 | |
568 | #ifdef DEBUG_ALLOC2 | |
2aabe7c7 | 569 | fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 |
0e06528e | 570 | " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "", |
2aabe7c7 | 571 | addend); |
f7d0fe02 | 572 | #endif |
7322afe7 | 573 | if (length < 0) { |
f7d0fe02 | 574 | return -EINVAL; |
7322afe7 KW |
575 | } else if (length == 0) { |
576 | return 0; | |
577 | } | |
578 | ||
2aabe7c7 | 579 | if (decrease) { |
29c1a730 KW |
580 | qcow2_cache_set_dependency(bs, s->refcount_block_cache, |
581 | s->l2_table_cache); | |
582 | } | |
583 | ||
ac95acdb HT |
584 | start = start_of_cluster(s, offset); |
585 | last = start_of_cluster(s, offset + length - 1); | |
f7d0fe02 KW |
586 | for(cluster_offset = start; cluster_offset <= last; |
587 | cluster_offset += s->cluster_size) | |
588 | { | |
2aabe7c7 | 589 | int block_index; |
0e06528e | 590 | uint64_t refcount; |
f7d0fe02 | 591 | int64_t cluster_index = cluster_offset >> s->cluster_bits; |
17bd5f47 | 592 | int64_t table_index = cluster_index >> s->refcount_block_bits; |
f7d0fe02 | 593 | |
29c1a730 KW |
594 | /* Load the refcount block and allocate it if needed */ |
595 | if (table_index != old_table_index) { | |
596 | if (refcount_block) { | |
597 | ret = qcow2_cache_put(bs, s->refcount_block_cache, | |
7453c96b | 598 | &refcount_block); |
29c1a730 KW |
599 | if (ret < 0) { |
600 | goto fail; | |
601 | } | |
602 | } | |
9923e05e | 603 | |
29c1a730 | 604 | ret = alloc_refcount_block(bs, cluster_index, &refcount_block); |
ed0df867 | 605 | if (ret < 0) { |
29c1a730 | 606 | goto fail; |
f7d0fe02 | 607 | } |
f7d0fe02 | 608 | } |
29c1a730 | 609 | old_table_index = table_index; |
f7d0fe02 | 610 | |
29c1a730 | 611 | qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block); |
f7d0fe02 KW |
612 | |
613 | /* we can update the count and save it */ | |
17bd5f47 | 614 | block_index = cluster_index & (s->refcount_block_size - 1); |
f7d0fe02 | 615 | |
7453c96b | 616 | refcount = s->get_refcount(refcount_block, block_index); |
0e06528e HR |
617 | if (decrease ? (refcount - addend > refcount) |
618 | : (refcount + addend < refcount || | |
619 | refcount + addend > s->refcount_max)) | |
2aabe7c7 | 620 | { |
09508d13 KW |
621 | ret = -EINVAL; |
622 | goto fail; | |
623 | } | |
2aabe7c7 HR |
624 | if (decrease) { |
625 | refcount -= addend; | |
626 | } else { | |
627 | refcount += addend; | |
628 | } | |
f7d0fe02 KW |
629 | if (refcount == 0 && cluster_index < s->free_cluster_index) { |
630 | s->free_cluster_index = cluster_index; | |
631 | } | |
7453c96b | 632 | s->set_refcount(refcount_block, block_index, refcount); |
0b919fae | 633 | |
67af674e | 634 | if (refcount == 0 && s->discard_passthrough[type]) { |
0b919fae | 635 | update_refcount_discard(bs, cluster_offset, s->cluster_size); |
67af674e | 636 | } |
f7d0fe02 KW |
637 | } |
638 | ||
09508d13 KW |
639 | ret = 0; |
640 | fail: | |
0b919fae KW |
641 | if (!s->cache_discards) { |
642 | qcow2_process_discards(bs, ret); | |
643 | } | |
644 | ||
f7d0fe02 | 645 | /* Write last changed block to disk */ |
29c1a730 | 646 | if (refcount_block) { |
ed0df867 | 647 | int wret; |
7453c96b | 648 | wret = qcow2_cache_put(bs, s->refcount_block_cache, &refcount_block); |
ed0df867 KW |
649 | if (wret < 0) { |
650 | return ret < 0 ? ret : wret; | |
f7d0fe02 KW |
651 | } |
652 | } | |
653 | ||
09508d13 KW |
654 | /* |
655 | * Try do undo any updates if an error is returned (This may succeed in | |
656 | * some cases like ENOSPC for allocating a new refcount block) | |
657 | */ | |
658 | if (ret < 0) { | |
659 | int dummy; | |
2aabe7c7 HR |
660 | dummy = update_refcount(bs, offset, cluster_offset - offset, addend, |
661 | !decrease, QCOW2_DISCARD_NEVER); | |
83e3f76c | 662 | (void)dummy; |
09508d13 KW |
663 | } |
664 | ||
665 | return ret; | |
f7d0fe02 KW |
666 | } |
667 | ||
018faafd | 668 | /* |
44751917 | 669 | * Increases or decreases the refcount of a given cluster. |
018faafd | 670 | * |
2aabe7c7 HR |
671 | * @addend is the absolute value of the addend; if @decrease is set, @addend |
672 | * will be subtracted from the current refcount, otherwise it will be added. | |
673 | * | |
c6e9d8ae | 674 | * On success 0 is returned; on failure -errno is returned. |
018faafd | 675 | */ |
32b6444d HR |
676 | int qcow2_update_cluster_refcount(BlockDriverState *bs, |
677 | int64_t cluster_index, | |
0e06528e | 678 | uint64_t addend, bool decrease, |
32b6444d | 679 | enum qcow2_discard_type type) |
f7d0fe02 KW |
680 | { |
681 | BDRVQcowState *s = bs->opaque; | |
682 | int ret; | |
683 | ||
6cfcb9b8 | 684 | ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend, |
2aabe7c7 | 685 | decrease, type); |
f7d0fe02 KW |
686 | if (ret < 0) { |
687 | return ret; | |
688 | } | |
689 | ||
c6e9d8ae | 690 | return 0; |
f7d0fe02 KW |
691 | } |
692 | ||
693 | ||
694 | ||
695 | /*********************************************************/ | |
696 | /* cluster allocation functions */ | |
697 | ||
698 | ||
699 | ||
700 | /* return < 0 if error */ | |
bb572aef | 701 | static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size) |
f7d0fe02 KW |
702 | { |
703 | BDRVQcowState *s = bs->opaque; | |
0e06528e | 704 | uint64_t i, nb_clusters, refcount; |
7324c10f | 705 | int ret; |
f7d0fe02 KW |
706 | |
707 | nb_clusters = size_to_clusters(s, size); | |
708 | retry: | |
709 | for(i = 0; i < nb_clusters; i++) { | |
bb572aef | 710 | uint64_t next_cluster_index = s->free_cluster_index++; |
7324c10f | 711 | ret = qcow2_get_refcount(bs, next_cluster_index, &refcount); |
2eaa8f63 | 712 | |
7324c10f HR |
713 | if (ret < 0) { |
714 | return ret; | |
2eaa8f63 | 715 | } else if (refcount != 0) { |
f7d0fe02 | 716 | goto retry; |
2eaa8f63 | 717 | } |
f7d0fe02 | 718 | } |
91f827dc HR |
719 | |
720 | /* Make sure that all offsets in the "allocated" range are representable | |
721 | * in an int64_t */ | |
65f33bc0 HR |
722 | if (s->free_cluster_index > 0 && |
723 | s->free_cluster_index - 1 > (INT64_MAX >> s->cluster_bits)) | |
724 | { | |
91f827dc HR |
725 | return -EFBIG; |
726 | } | |
727 | ||
f7d0fe02 | 728 | #ifdef DEBUG_ALLOC2 |
35ee5e39 | 729 | fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n", |
f7d0fe02 KW |
730 | size, |
731 | (s->free_cluster_index - nb_clusters) << s->cluster_bits); | |
732 | #endif | |
733 | return (s->free_cluster_index - nb_clusters) << s->cluster_bits; | |
734 | } | |
735 | ||
bb572aef | 736 | int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size) |
f7d0fe02 KW |
737 | { |
738 | int64_t offset; | |
db3a964f | 739 | int ret; |
f7d0fe02 | 740 | |
66f82cee | 741 | BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC); |
b106ad91 KW |
742 | do { |
743 | offset = alloc_clusters_noref(bs, size); | |
744 | if (offset < 0) { | |
745 | return offset; | |
746 | } | |
747 | ||
2aabe7c7 | 748 | ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER); |
b106ad91 | 749 | } while (ret == -EAGAIN); |
2eaa8f63 | 750 | |
db3a964f KW |
751 | if (ret < 0) { |
752 | return ret; | |
753 | } | |
1c4c2814 | 754 | |
f7d0fe02 KW |
755 | return offset; |
756 | } | |
757 | ||
256900b1 KW |
758 | int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset, |
759 | int nb_clusters) | |
760 | { | |
761 | BDRVQcowState *s = bs->opaque; | |
0e06528e | 762 | uint64_t cluster_index, refcount; |
33304ec9 | 763 | uint64_t i; |
7324c10f | 764 | int ret; |
33304ec9 HT |
765 | |
766 | assert(nb_clusters >= 0); | |
767 | if (nb_clusters == 0) { | |
768 | return 0; | |
769 | } | |
256900b1 | 770 | |
b106ad91 KW |
771 | do { |
772 | /* Check how many clusters there are free */ | |
773 | cluster_index = offset >> s->cluster_bits; | |
774 | for(i = 0; i < nb_clusters; i++) { | |
7324c10f HR |
775 | ret = qcow2_get_refcount(bs, cluster_index++, &refcount); |
776 | if (ret < 0) { | |
777 | return ret; | |
b106ad91 KW |
778 | } else if (refcount != 0) { |
779 | break; | |
780 | } | |
256900b1 | 781 | } |
256900b1 | 782 | |
b106ad91 | 783 | /* And then allocate them */ |
2aabe7c7 | 784 | ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false, |
b106ad91 KW |
785 | QCOW2_DISCARD_NEVER); |
786 | } while (ret == -EAGAIN); | |
f24423bd | 787 | |
256900b1 KW |
788 | if (ret < 0) { |
789 | return ret; | |
790 | } | |
791 | ||
792 | return i; | |
793 | } | |
794 | ||
f7d0fe02 KW |
795 | /* only used to allocate compressed sectors. We try to allocate |
796 | contiguous sectors. size must be <= cluster_size */ | |
ed6ccf0f | 797 | int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size) |
f7d0fe02 KW |
798 | { |
799 | BDRVQcowState *s = bs->opaque; | |
8c44dfbc HR |
800 | int64_t offset; |
801 | size_t free_in_cluster; | |
802 | int ret; | |
f7d0fe02 | 803 | |
66f82cee | 804 | BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES); |
f7d0fe02 | 805 | assert(size > 0 && size <= s->cluster_size); |
8c44dfbc HR |
806 | assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset)); |
807 | ||
808 | offset = s->free_byte_offset; | |
809 | ||
810 | if (offset) { | |
0e06528e | 811 | uint64_t refcount; |
7324c10f HR |
812 | ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount); |
813 | if (ret < 0) { | |
814 | return ret; | |
5d757b56 | 815 | } |
8c44dfbc | 816 | |
346a53df | 817 | if (refcount == s->refcount_max) { |
8c44dfbc | 818 | offset = 0; |
5d757b56 | 819 | } |
8c44dfbc HR |
820 | } |
821 | ||
822 | free_in_cluster = s->cluster_size - offset_into_cluster(s, offset); | |
823 | if (!offset || free_in_cluster < size) { | |
824 | int64_t new_cluster = alloc_clusters_noref(bs, s->cluster_size); | |
825 | if (new_cluster < 0) { | |
826 | return new_cluster; | |
827 | } | |
828 | ||
829 | if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) { | |
830 | offset = new_cluster; | |
f7d0fe02 KW |
831 | } |
832 | } | |
29216ed1 | 833 | |
8c44dfbc | 834 | assert(offset); |
2aabe7c7 | 835 | ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER); |
8c44dfbc HR |
836 | if (ret < 0) { |
837 | return ret; | |
838 | } | |
839 | ||
840 | /* The cluster refcount was incremented; refcount blocks must be flushed | |
841 | * before the caller's L2 table updates. */ | |
c1f5bafd | 842 | qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache); |
8c44dfbc HR |
843 | |
844 | s->free_byte_offset = offset + size; | |
845 | if (!offset_into_cluster(s, s->free_byte_offset)) { | |
846 | s->free_byte_offset = 0; | |
847 | } | |
848 | ||
f7d0fe02 KW |
849 | return offset; |
850 | } | |
851 | ||
ed6ccf0f | 852 | void qcow2_free_clusters(BlockDriverState *bs, |
6cfcb9b8 KW |
853 | int64_t offset, int64_t size, |
854 | enum qcow2_discard_type type) | |
f7d0fe02 | 855 | { |
db3a964f KW |
856 | int ret; |
857 | ||
66f82cee | 858 | BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE); |
2aabe7c7 | 859 | ret = update_refcount(bs, offset, size, 1, true, type); |
db3a964f KW |
860 | if (ret < 0) { |
861 | fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret)); | |
003fad6e | 862 | /* TODO Remember the clusters to free them later and avoid leaking */ |
db3a964f | 863 | } |
f7d0fe02 KW |
864 | } |
865 | ||
45aba42f | 866 | /* |
c7a4c37a KW |
867 | * Free a cluster using its L2 entry (handles clusters of all types, e.g. |
868 | * normal cluster, compressed cluster, etc.) | |
45aba42f | 869 | */ |
6cfcb9b8 KW |
870 | void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry, |
871 | int nb_clusters, enum qcow2_discard_type type) | |
45aba42f KW |
872 | { |
873 | BDRVQcowState *s = bs->opaque; | |
874 | ||
c7a4c37a KW |
875 | switch (qcow2_get_cluster_type(l2_entry)) { |
876 | case QCOW2_CLUSTER_COMPRESSED: | |
877 | { | |
878 | int nb_csectors; | |
879 | nb_csectors = ((l2_entry >> s->csize_shift) & | |
880 | s->csize_mask) + 1; | |
881 | qcow2_free_clusters(bs, | |
882 | (l2_entry & s->cluster_offset_mask) & ~511, | |
6cfcb9b8 | 883 | nb_csectors * 512, type); |
c7a4c37a KW |
884 | } |
885 | break; | |
886 | case QCOW2_CLUSTER_NORMAL: | |
8f730dd2 HR |
887 | case QCOW2_CLUSTER_ZERO: |
888 | if (l2_entry & L2E_OFFSET_MASK) { | |
a97c67ee HR |
889 | if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) { |
890 | qcow2_signal_corruption(bs, false, -1, -1, | |
891 | "Cannot free unaligned cluster %#llx", | |
892 | l2_entry & L2E_OFFSET_MASK); | |
893 | } else { | |
894 | qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK, | |
895 | nb_clusters << s->cluster_bits, type); | |
896 | } | |
8f730dd2 | 897 | } |
c7a4c37a KW |
898 | break; |
899 | case QCOW2_CLUSTER_UNALLOCATED: | |
900 | break; | |
901 | default: | |
902 | abort(); | |
45aba42f | 903 | } |
45aba42f KW |
904 | } |
905 | ||
f7d0fe02 KW |
906 | |
907 | ||
908 | /*********************************************************/ | |
909 | /* snapshots and image creation */ | |
910 | ||
911 | ||
912 | ||
f7d0fe02 | 913 | /* update the refcounts of snapshots and the copied flag */ |
ed6ccf0f KW |
914 | int qcow2_update_snapshot_refcount(BlockDriverState *bs, |
915 | int64_t l1_table_offset, int l1_size, int addend) | |
f7d0fe02 KW |
916 | { |
917 | BDRVQcowState *s = bs->opaque; | |
0e06528e | 918 | uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, refcount; |
de82815d | 919 | bool l1_allocated = false; |
f7d0fe02 | 920 | int64_t old_offset, old_l2_offset; |
7324c10f | 921 | int i, j, l1_modified = 0, nb_csectors; |
29c1a730 | 922 | int ret; |
f7d0fe02 | 923 | |
2aabe7c7 HR |
924 | assert(addend >= -1 && addend <= 1); |
925 | ||
f7d0fe02 KW |
926 | l2_table = NULL; |
927 | l1_table = NULL; | |
928 | l1_size2 = l1_size * sizeof(uint64_t); | |
43a0cac4 | 929 | |
0b919fae KW |
930 | s->cache_discards = true; |
931 | ||
43a0cac4 KW |
932 | /* WARNING: qcow2_snapshot_goto relies on this function not using the |
933 | * l1_table_offset when it is the current s->l1_table_offset! Be careful | |
934 | * when changing this! */ | |
f7d0fe02 | 935 | if (l1_table_offset != s->l1_table_offset) { |
de82815d KW |
936 | l1_table = g_try_malloc0(align_offset(l1_size2, 512)); |
937 | if (l1_size2 && l1_table == NULL) { | |
938 | ret = -ENOMEM; | |
939 | goto fail; | |
940 | } | |
941 | l1_allocated = true; | |
c2bc78b6 KW |
942 | |
943 | ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2); | |
944 | if (ret < 0) { | |
f7d0fe02 | 945 | goto fail; |
93913dfd KW |
946 | } |
947 | ||
f7d0fe02 KW |
948 | for(i = 0;i < l1_size; i++) |
949 | be64_to_cpus(&l1_table[i]); | |
950 | } else { | |
951 | assert(l1_size == s->l1_size); | |
952 | l1_table = s->l1_table; | |
de82815d | 953 | l1_allocated = false; |
f7d0fe02 KW |
954 | } |
955 | ||
f7d0fe02 KW |
956 | for(i = 0; i < l1_size; i++) { |
957 | l2_offset = l1_table[i]; | |
958 | if (l2_offset) { | |
959 | old_l2_offset = l2_offset; | |
8e37f681 | 960 | l2_offset &= L1E_OFFSET_MASK; |
29c1a730 | 961 | |
a97c67ee HR |
962 | if (offset_into_cluster(s, l2_offset)) { |
963 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
964 | PRIx64 " unaligned (L1 index: %#x)", | |
965 | l2_offset, i); | |
966 | ret = -EIO; | |
967 | goto fail; | |
968 | } | |
969 | ||
29c1a730 KW |
970 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, |
971 | (void**) &l2_table); | |
972 | if (ret < 0) { | |
f7d0fe02 | 973 | goto fail; |
29c1a730 KW |
974 | } |
975 | ||
f7d0fe02 | 976 | for(j = 0; j < s->l2_size; j++) { |
8b81a7b6 HR |
977 | uint64_t cluster_index; |
978 | ||
f7d0fe02 | 979 | offset = be64_to_cpu(l2_table[j]); |
8b81a7b6 HR |
980 | old_offset = offset; |
981 | offset &= ~QCOW_OFLAG_COPIED; | |
982 | ||
983 | switch (qcow2_get_cluster_type(offset)) { | |
984 | case QCOW2_CLUSTER_COMPRESSED: | |
f7d0fe02 KW |
985 | nb_csectors = ((offset >> s->csize_shift) & |
986 | s->csize_mask) + 1; | |
db3a964f | 987 | if (addend != 0) { |
db3a964f KW |
988 | ret = update_refcount(bs, |
989 | (offset & s->cluster_offset_mask) & ~511, | |
2aabe7c7 | 990 | nb_csectors * 512, abs(addend), addend < 0, |
6cfcb9b8 | 991 | QCOW2_DISCARD_SNAPSHOT); |
db3a964f KW |
992 | if (ret < 0) { |
993 | goto fail; | |
994 | } | |
995 | } | |
f7d0fe02 KW |
996 | /* compressed clusters are never modified */ |
997 | refcount = 2; | |
8b81a7b6 HR |
998 | break; |
999 | ||
1000 | case QCOW2_CLUSTER_NORMAL: | |
1001 | case QCOW2_CLUSTER_ZERO: | |
a97c67ee HR |
1002 | if (offset_into_cluster(s, offset & L2E_OFFSET_MASK)) { |
1003 | qcow2_signal_corruption(bs, true, -1, -1, "Data " | |
1004 | "cluster offset %#llx " | |
1005 | "unaligned (L2 offset: %#" | |
1006 | PRIx64 ", L2 index: %#x)", | |
1007 | offset & L2E_OFFSET_MASK, | |
1008 | l2_offset, j); | |
1009 | ret = -EIO; | |
1010 | goto fail; | |
1011 | } | |
1012 | ||
8b81a7b6 HR |
1013 | cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits; |
1014 | if (!cluster_index) { | |
1015 | /* unallocated */ | |
1016 | refcount = 0; | |
1017 | break; | |
1018 | } | |
f7d0fe02 | 1019 | if (addend != 0) { |
c6e9d8ae | 1020 | ret = qcow2_update_cluster_refcount(bs, |
2aabe7c7 | 1021 | cluster_index, abs(addend), addend < 0, |
32b6444d | 1022 | QCOW2_DISCARD_SNAPSHOT); |
c6e9d8ae HR |
1023 | if (ret < 0) { |
1024 | goto fail; | |
1025 | } | |
f7d0fe02 | 1026 | } |
018faafd | 1027 | |
7324c10f HR |
1028 | ret = qcow2_get_refcount(bs, cluster_index, &refcount); |
1029 | if (ret < 0) { | |
018faafd KW |
1030 | goto fail; |
1031 | } | |
8b81a7b6 | 1032 | break; |
f7d0fe02 | 1033 | |
8b81a7b6 HR |
1034 | case QCOW2_CLUSTER_UNALLOCATED: |
1035 | refcount = 0; | |
1036 | break; | |
1037 | ||
1038 | default: | |
1039 | abort(); | |
1040 | } | |
1041 | ||
1042 | if (refcount == 1) { | |
1043 | offset |= QCOW_OFLAG_COPIED; | |
1044 | } | |
1045 | if (offset != old_offset) { | |
1046 | if (addend > 0) { | |
1047 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
1048 | s->refcount_block_cache); | |
f7d0fe02 | 1049 | } |
8b81a7b6 HR |
1050 | l2_table[j] = cpu_to_be64(offset); |
1051 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); | |
f7d0fe02 KW |
1052 | } |
1053 | } | |
29c1a730 KW |
1054 | |
1055 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
1056 | if (ret < 0) { | |
1057 | goto fail; | |
f7d0fe02 KW |
1058 | } |
1059 | ||
29c1a730 | 1060 | |
f7d0fe02 | 1061 | if (addend != 0) { |
c6e9d8ae HR |
1062 | ret = qcow2_update_cluster_refcount(bs, l2_offset >> |
1063 | s->cluster_bits, | |
2aabe7c7 | 1064 | abs(addend), addend < 0, |
c6e9d8ae HR |
1065 | QCOW2_DISCARD_SNAPSHOT); |
1066 | if (ret < 0) { | |
1067 | goto fail; | |
1068 | } | |
f7d0fe02 | 1069 | } |
7324c10f HR |
1070 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1071 | &refcount); | |
1072 | if (ret < 0) { | |
018faafd KW |
1073 | goto fail; |
1074 | } else if (refcount == 1) { | |
f7d0fe02 KW |
1075 | l2_offset |= QCOW_OFLAG_COPIED; |
1076 | } | |
1077 | if (l2_offset != old_l2_offset) { | |
1078 | l1_table[i] = l2_offset; | |
1079 | l1_modified = 1; | |
1080 | } | |
1081 | } | |
1082 | } | |
93913dfd | 1083 | |
2154f24e | 1084 | ret = bdrv_flush(bs); |
93913dfd KW |
1085 | fail: |
1086 | if (l2_table) { | |
1087 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
1088 | } | |
1089 | ||
0b919fae KW |
1090 | s->cache_discards = false; |
1091 | qcow2_process_discards(bs, ret); | |
1092 | ||
43a0cac4 | 1093 | /* Update L1 only if it isn't deleted anyway (addend = -1) */ |
c2b6ff51 KW |
1094 | if (ret == 0 && addend >= 0 && l1_modified) { |
1095 | for (i = 0; i < l1_size; i++) { | |
f7d0fe02 | 1096 | cpu_to_be64s(&l1_table[i]); |
c2b6ff51 KW |
1097 | } |
1098 | ||
1099 | ret = bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table, l1_size2); | |
1100 | ||
1101 | for (i = 0; i < l1_size; i++) { | |
f7d0fe02 | 1102 | be64_to_cpus(&l1_table[i]); |
c2b6ff51 | 1103 | } |
f7d0fe02 KW |
1104 | } |
1105 | if (l1_allocated) | |
7267c094 | 1106 | g_free(l1_table); |
93913dfd | 1107 | return ret; |
f7d0fe02 KW |
1108 | } |
1109 | ||
1110 | ||
1111 | ||
1112 | ||
1113 | /*********************************************************/ | |
1114 | /* refcount checking functions */ | |
1115 | ||
1116 | ||
5fee192e HR |
1117 | static size_t refcount_array_byte_size(BDRVQcowState *s, uint64_t entries) |
1118 | { | |
1119 | /* This assertion holds because there is no way we can address more than | |
1120 | * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because | |
1121 | * offsets have to be representable in bytes); due to every cluster | |
1122 | * corresponding to one refcount entry, we are well below that limit */ | |
1123 | assert(entries < (UINT64_C(1) << (64 - 9))); | |
1124 | ||
1125 | /* Thanks to the assertion this will not overflow, because | |
1126 | * s->refcount_order < 7. | |
1127 | * (note: x << s->refcount_order == x * s->refcount_bits) */ | |
1128 | return DIV_ROUND_UP(entries << s->refcount_order, 8); | |
1129 | } | |
1130 | ||
1131 | /** | |
1132 | * Reallocates *array so that it can hold new_size entries. *size must contain | |
1133 | * the current number of entries in *array. If the reallocation fails, *array | |
1134 | * and *size will not be modified and -errno will be returned. If the | |
1135 | * reallocation is successful, *array will be set to the new buffer, *size | |
1136 | * will be set to new_size and 0 will be returned. The size of the reallocated | |
1137 | * refcount array buffer will be aligned to a cluster boundary, and the newly | |
1138 | * allocated area will be zeroed. | |
1139 | */ | |
7453c96b | 1140 | static int realloc_refcount_array(BDRVQcowState *s, void **array, |
5fee192e HR |
1141 | int64_t *size, int64_t new_size) |
1142 | { | |
1143 | size_t old_byte_size, new_byte_size; | |
7453c96b | 1144 | void *new_ptr; |
5fee192e HR |
1145 | |
1146 | /* Round to clusters so the array can be directly written to disk */ | |
1147 | old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size)) | |
1148 | * s->cluster_size; | |
1149 | new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size)) | |
1150 | * s->cluster_size; | |
1151 | ||
1152 | if (new_byte_size == old_byte_size) { | |
1153 | *size = new_size; | |
1154 | return 0; | |
1155 | } | |
1156 | ||
1157 | assert(new_byte_size > 0); | |
1158 | ||
1159 | new_ptr = g_try_realloc(*array, new_byte_size); | |
1160 | if (!new_ptr) { | |
1161 | return -ENOMEM; | |
1162 | } | |
1163 | ||
1164 | if (new_byte_size > old_byte_size) { | |
1165 | memset((void *)((uintptr_t)new_ptr + old_byte_size), 0, | |
1166 | new_byte_size - old_byte_size); | |
1167 | } | |
1168 | ||
1169 | *array = new_ptr; | |
1170 | *size = new_size; | |
1171 | ||
1172 | return 0; | |
1173 | } | |
f7d0fe02 KW |
1174 | |
1175 | /* | |
1176 | * Increases the refcount for a range of clusters in a given refcount table. | |
1177 | * This is used to construct a temporary refcount table out of L1 and L2 tables | |
1178 | * which can be compared the the refcount table saved in the image. | |
1179 | * | |
9ac228e0 | 1180 | * Modifies the number of errors in res. |
f7d0fe02 | 1181 | */ |
fef4d3d5 HR |
1182 | static int inc_refcounts(BlockDriverState *bs, |
1183 | BdrvCheckResult *res, | |
7453c96b | 1184 | void **refcount_table, |
641bb63c | 1185 | int64_t *refcount_table_size, |
fef4d3d5 | 1186 | int64_t offset, int64_t size) |
f7d0fe02 KW |
1187 | { |
1188 | BDRVQcowState *s = bs->opaque; | |
7453c96b | 1189 | uint64_t start, last, cluster_offset, k, refcount; |
5fee192e | 1190 | int ret; |
f7d0fe02 | 1191 | |
fef4d3d5 HR |
1192 | if (size <= 0) { |
1193 | return 0; | |
1194 | } | |
f7d0fe02 | 1195 | |
ac95acdb HT |
1196 | start = start_of_cluster(s, offset); |
1197 | last = start_of_cluster(s, offset + size - 1); | |
f7d0fe02 KW |
1198 | for(cluster_offset = start; cluster_offset <= last; |
1199 | cluster_offset += s->cluster_size) { | |
1200 | k = cluster_offset >> s->cluster_bits; | |
641bb63c | 1201 | if (k >= *refcount_table_size) { |
5fee192e HR |
1202 | ret = realloc_refcount_array(s, refcount_table, |
1203 | refcount_table_size, k + 1); | |
1204 | if (ret < 0) { | |
641bb63c | 1205 | res->check_errors++; |
5fee192e | 1206 | return ret; |
f7d0fe02 | 1207 | } |
641bb63c HR |
1208 | } |
1209 | ||
7453c96b HR |
1210 | refcount = s->get_refcount(*refcount_table, k); |
1211 | if (refcount == s->refcount_max) { | |
641bb63c HR |
1212 | fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64 |
1213 | "\n", cluster_offset); | |
1214 | res->corruptions++; | |
7453c96b | 1215 | continue; |
f7d0fe02 | 1216 | } |
7453c96b | 1217 | s->set_refcount(*refcount_table, k, refcount + 1); |
f7d0fe02 | 1218 | } |
fef4d3d5 HR |
1219 | |
1220 | return 0; | |
f7d0fe02 KW |
1221 | } |
1222 | ||
801f7044 SH |
1223 | /* Flags for check_refcounts_l1() and check_refcounts_l2() */ |
1224 | enum { | |
fba31bae | 1225 | CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */ |
801f7044 SH |
1226 | }; |
1227 | ||
f7d0fe02 KW |
1228 | /* |
1229 | * Increases the refcount in the given refcount table for the all clusters | |
1230 | * referenced in the L2 table. While doing so, performs some checks on L2 | |
1231 | * entries. | |
1232 | * | |
1233 | * Returns the number of errors found by the checks or -errno if an internal | |
1234 | * error occurred. | |
1235 | */ | |
9ac228e0 | 1236 | static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res, |
7453c96b HR |
1237 | void **refcount_table, |
1238 | int64_t *refcount_table_size, int64_t l2_offset, | |
1239 | int flags) | |
f7d0fe02 KW |
1240 | { |
1241 | BDRVQcowState *s = bs->opaque; | |
afdf0abe | 1242 | uint64_t *l2_table, l2_entry; |
fba31bae | 1243 | uint64_t next_contiguous_offset = 0; |
ad27390c | 1244 | int i, l2_size, nb_csectors, ret; |
f7d0fe02 KW |
1245 | |
1246 | /* Read L2 table from disk */ | |
1247 | l2_size = s->l2_size * sizeof(uint64_t); | |
7267c094 | 1248 | l2_table = g_malloc(l2_size); |
f7d0fe02 | 1249 | |
ad27390c HR |
1250 | ret = bdrv_pread(bs->file, l2_offset, l2_table, l2_size); |
1251 | if (ret < 0) { | |
1252 | fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n"); | |
1253 | res->check_errors++; | |
f7d0fe02 | 1254 | goto fail; |
ad27390c | 1255 | } |
f7d0fe02 KW |
1256 | |
1257 | /* Do the actual checks */ | |
1258 | for(i = 0; i < s->l2_size; i++) { | |
afdf0abe KW |
1259 | l2_entry = be64_to_cpu(l2_table[i]); |
1260 | ||
1261 | switch (qcow2_get_cluster_type(l2_entry)) { | |
1262 | case QCOW2_CLUSTER_COMPRESSED: | |
1263 | /* Compressed clusters don't have QCOW_OFLAG_COPIED */ | |
1264 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
1265 | fprintf(stderr, "ERROR: cluster %" PRId64 ": " | |
1266 | "copied flag must never be set for compressed " | |
1267 | "clusters\n", l2_entry >> s->cluster_bits); | |
1268 | l2_entry &= ~QCOW_OFLAG_COPIED; | |
1269 | res->corruptions++; | |
1270 | } | |
f7d0fe02 | 1271 | |
afdf0abe KW |
1272 | /* Mark cluster as used */ |
1273 | nb_csectors = ((l2_entry >> s->csize_shift) & | |
1274 | s->csize_mask) + 1; | |
1275 | l2_entry &= s->cluster_offset_mask; | |
fef4d3d5 HR |
1276 | ret = inc_refcounts(bs, res, refcount_table, refcount_table_size, |
1277 | l2_entry & ~511, nb_csectors * 512); | |
1278 | if (ret < 0) { | |
1279 | goto fail; | |
1280 | } | |
fba31bae SH |
1281 | |
1282 | if (flags & CHECK_FRAG_INFO) { | |
1283 | res->bfi.allocated_clusters++; | |
4db35162 | 1284 | res->bfi.compressed_clusters++; |
fba31bae SH |
1285 | |
1286 | /* Compressed clusters are fragmented by nature. Since they | |
1287 | * take up sub-sector space but we only have sector granularity | |
1288 | * I/O we need to re-read the same sectors even for adjacent | |
1289 | * compressed clusters. | |
1290 | */ | |
1291 | res->bfi.fragmented_clusters++; | |
1292 | } | |
afdf0abe | 1293 | break; |
f7d0fe02 | 1294 | |
6377af48 KW |
1295 | case QCOW2_CLUSTER_ZERO: |
1296 | if ((l2_entry & L2E_OFFSET_MASK) == 0) { | |
1297 | break; | |
1298 | } | |
1299 | /* fall through */ | |
1300 | ||
afdf0abe KW |
1301 | case QCOW2_CLUSTER_NORMAL: |
1302 | { | |
afdf0abe | 1303 | uint64_t offset = l2_entry & L2E_OFFSET_MASK; |
f7d0fe02 | 1304 | |
fba31bae SH |
1305 | if (flags & CHECK_FRAG_INFO) { |
1306 | res->bfi.allocated_clusters++; | |
1307 | if (next_contiguous_offset && | |
1308 | offset != next_contiguous_offset) { | |
1309 | res->bfi.fragmented_clusters++; | |
1310 | } | |
1311 | next_contiguous_offset = offset + s->cluster_size; | |
1312 | } | |
1313 | ||
afdf0abe | 1314 | /* Mark cluster as used */ |
fef4d3d5 HR |
1315 | ret = inc_refcounts(bs, res, refcount_table, refcount_table_size, |
1316 | offset, s->cluster_size); | |
1317 | if (ret < 0) { | |
1318 | goto fail; | |
1319 | } | |
afdf0abe KW |
1320 | |
1321 | /* Correct offsets are cluster aligned */ | |
ac95acdb | 1322 | if (offset_into_cluster(s, offset)) { |
afdf0abe KW |
1323 | fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not " |
1324 | "properly aligned; L2 entry corrupted.\n", offset); | |
1325 | res->corruptions++; | |
1326 | } | |
1327 | break; | |
1328 | } | |
1329 | ||
1330 | case QCOW2_CLUSTER_UNALLOCATED: | |
1331 | break; | |
1332 | ||
1333 | default: | |
1334 | abort(); | |
f7d0fe02 KW |
1335 | } |
1336 | } | |
1337 | ||
7267c094 | 1338 | g_free(l2_table); |
9ac228e0 | 1339 | return 0; |
f7d0fe02 KW |
1340 | |
1341 | fail: | |
7267c094 | 1342 | g_free(l2_table); |
ad27390c | 1343 | return ret; |
f7d0fe02 KW |
1344 | } |
1345 | ||
1346 | /* | |
1347 | * Increases the refcount for the L1 table, its L2 tables and all referenced | |
1348 | * clusters in the given refcount table. While doing so, performs some checks | |
1349 | * on L1 and L2 entries. | |
1350 | * | |
1351 | * Returns the number of errors found by the checks or -errno if an internal | |
1352 | * error occurred. | |
1353 | */ | |
1354 | static int check_refcounts_l1(BlockDriverState *bs, | |
9ac228e0 | 1355 | BdrvCheckResult *res, |
7453c96b | 1356 | void **refcount_table, |
641bb63c | 1357 | int64_t *refcount_table_size, |
f7d0fe02 | 1358 | int64_t l1_table_offset, int l1_size, |
801f7044 | 1359 | int flags) |
f7d0fe02 KW |
1360 | { |
1361 | BDRVQcowState *s = bs->opaque; | |
fef4d3d5 | 1362 | uint64_t *l1_table = NULL, l2_offset, l1_size2; |
4f6ed88c | 1363 | int i, ret; |
f7d0fe02 KW |
1364 | |
1365 | l1_size2 = l1_size * sizeof(uint64_t); | |
1366 | ||
1367 | /* Mark L1 table as used */ | |
fef4d3d5 HR |
1368 | ret = inc_refcounts(bs, res, refcount_table, refcount_table_size, |
1369 | l1_table_offset, l1_size2); | |
1370 | if (ret < 0) { | |
1371 | goto fail; | |
1372 | } | |
f7d0fe02 KW |
1373 | |
1374 | /* Read L1 table entries from disk */ | |
fef4d3d5 | 1375 | if (l1_size2 > 0) { |
de82815d KW |
1376 | l1_table = g_try_malloc(l1_size2); |
1377 | if (l1_table == NULL) { | |
1378 | ret = -ENOMEM; | |
ad27390c | 1379 | res->check_errors++; |
de82815d KW |
1380 | goto fail; |
1381 | } | |
ad27390c HR |
1382 | ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2); |
1383 | if (ret < 0) { | |
1384 | fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n"); | |
1385 | res->check_errors++; | |
702ef63f | 1386 | goto fail; |
ad27390c | 1387 | } |
702ef63f KW |
1388 | for(i = 0;i < l1_size; i++) |
1389 | be64_to_cpus(&l1_table[i]); | |
1390 | } | |
f7d0fe02 KW |
1391 | |
1392 | /* Do the actual checks */ | |
1393 | for(i = 0; i < l1_size; i++) { | |
1394 | l2_offset = l1_table[i]; | |
1395 | if (l2_offset) { | |
f7d0fe02 | 1396 | /* Mark L2 table as used */ |
afdf0abe | 1397 | l2_offset &= L1E_OFFSET_MASK; |
fef4d3d5 HR |
1398 | ret = inc_refcounts(bs, res, refcount_table, refcount_table_size, |
1399 | l2_offset, s->cluster_size); | |
1400 | if (ret < 0) { | |
1401 | goto fail; | |
1402 | } | |
f7d0fe02 KW |
1403 | |
1404 | /* L2 tables are cluster aligned */ | |
ac95acdb | 1405 | if (offset_into_cluster(s, l2_offset)) { |
f7d0fe02 KW |
1406 | fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not " |
1407 | "cluster aligned; L1 entry corrupted\n", l2_offset); | |
9ac228e0 | 1408 | res->corruptions++; |
f7d0fe02 KW |
1409 | } |
1410 | ||
1411 | /* Process and check L2 entries */ | |
9ac228e0 | 1412 | ret = check_refcounts_l2(bs, res, refcount_table, |
801f7044 | 1413 | refcount_table_size, l2_offset, flags); |
f7d0fe02 KW |
1414 | if (ret < 0) { |
1415 | goto fail; | |
1416 | } | |
f7d0fe02 KW |
1417 | } |
1418 | } | |
7267c094 | 1419 | g_free(l1_table); |
9ac228e0 | 1420 | return 0; |
f7d0fe02 KW |
1421 | |
1422 | fail: | |
7267c094 | 1423 | g_free(l1_table); |
ad27390c | 1424 | return ret; |
f7d0fe02 KW |
1425 | } |
1426 | ||
4f6ed88c HR |
1427 | /* |
1428 | * Checks the OFLAG_COPIED flag for all L1 and L2 entries. | |
1429 | * | |
1430 | * This function does not print an error message nor does it increment | |
44751917 HR |
1431 | * check_errors if qcow2_get_refcount fails (this is because such an error will |
1432 | * have been already detected and sufficiently signaled by the calling function | |
4f6ed88c HR |
1433 | * (qcow2_check_refcounts) by the time this function is called). |
1434 | */ | |
e23e400e HR |
1435 | static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res, |
1436 | BdrvCheckMode fix) | |
4f6ed88c HR |
1437 | { |
1438 | BDRVQcowState *s = bs->opaque; | |
1439 | uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size); | |
1440 | int ret; | |
0e06528e | 1441 | uint64_t refcount; |
4f6ed88c HR |
1442 | int i, j; |
1443 | ||
1444 | for (i = 0; i < s->l1_size; i++) { | |
1445 | uint64_t l1_entry = s->l1_table[i]; | |
1446 | uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK; | |
e23e400e | 1447 | bool l2_dirty = false; |
4f6ed88c HR |
1448 | |
1449 | if (!l2_offset) { | |
1450 | continue; | |
1451 | } | |
1452 | ||
7324c10f HR |
1453 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1454 | &refcount); | |
1455 | if (ret < 0) { | |
4f6ed88c HR |
1456 | /* don't print message nor increment check_errors */ |
1457 | continue; | |
1458 | } | |
1459 | if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) { | |
e23e400e | 1460 | fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d " |
0e06528e | 1461 | "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n", |
e23e400e HR |
1462 | fix & BDRV_FIX_ERRORS ? "Repairing" : |
1463 | "ERROR", | |
4f6ed88c | 1464 | i, l1_entry, refcount); |
e23e400e HR |
1465 | if (fix & BDRV_FIX_ERRORS) { |
1466 | s->l1_table[i] = refcount == 1 | |
1467 | ? l1_entry | QCOW_OFLAG_COPIED | |
1468 | : l1_entry & ~QCOW_OFLAG_COPIED; | |
1469 | ret = qcow2_write_l1_entry(bs, i); | |
1470 | if (ret < 0) { | |
1471 | res->check_errors++; | |
1472 | goto fail; | |
1473 | } | |
1474 | res->corruptions_fixed++; | |
1475 | } else { | |
1476 | res->corruptions++; | |
1477 | } | |
4f6ed88c HR |
1478 | } |
1479 | ||
1480 | ret = bdrv_pread(bs->file, l2_offset, l2_table, | |
1481 | s->l2_size * sizeof(uint64_t)); | |
1482 | if (ret < 0) { | |
1483 | fprintf(stderr, "ERROR: Could not read L2 table: %s\n", | |
1484 | strerror(-ret)); | |
1485 | res->check_errors++; | |
1486 | goto fail; | |
1487 | } | |
1488 | ||
1489 | for (j = 0; j < s->l2_size; j++) { | |
1490 | uint64_t l2_entry = be64_to_cpu(l2_table[j]); | |
1491 | uint64_t data_offset = l2_entry & L2E_OFFSET_MASK; | |
1492 | int cluster_type = qcow2_get_cluster_type(l2_entry); | |
1493 | ||
1494 | if ((cluster_type == QCOW2_CLUSTER_NORMAL) || | |
1495 | ((cluster_type == QCOW2_CLUSTER_ZERO) && (data_offset != 0))) { | |
7324c10f HR |
1496 | ret = qcow2_get_refcount(bs, |
1497 | data_offset >> s->cluster_bits, | |
1498 | &refcount); | |
1499 | if (ret < 0) { | |
4f6ed88c HR |
1500 | /* don't print message nor increment check_errors */ |
1501 | continue; | |
1502 | } | |
1503 | if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) { | |
e23e400e | 1504 | fprintf(stderr, "%s OFLAG_COPIED data cluster: " |
0e06528e | 1505 | "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n", |
e23e400e HR |
1506 | fix & BDRV_FIX_ERRORS ? "Repairing" : |
1507 | "ERROR", | |
4f6ed88c | 1508 | l2_entry, refcount); |
e23e400e HR |
1509 | if (fix & BDRV_FIX_ERRORS) { |
1510 | l2_table[j] = cpu_to_be64(refcount == 1 | |
1511 | ? l2_entry | QCOW_OFLAG_COPIED | |
1512 | : l2_entry & ~QCOW_OFLAG_COPIED); | |
1513 | l2_dirty = true; | |
1514 | res->corruptions_fixed++; | |
1515 | } else { | |
1516 | res->corruptions++; | |
1517 | } | |
4f6ed88c HR |
1518 | } |
1519 | } | |
1520 | } | |
e23e400e HR |
1521 | |
1522 | if (l2_dirty) { | |
231bb267 HR |
1523 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2, |
1524 | l2_offset, s->cluster_size); | |
e23e400e HR |
1525 | if (ret < 0) { |
1526 | fprintf(stderr, "ERROR: Could not write L2 table; metadata " | |
1527 | "overlap check failed: %s\n", strerror(-ret)); | |
1528 | res->check_errors++; | |
1529 | goto fail; | |
1530 | } | |
1531 | ||
1532 | ret = bdrv_pwrite(bs->file, l2_offset, l2_table, s->cluster_size); | |
1533 | if (ret < 0) { | |
1534 | fprintf(stderr, "ERROR: Could not write L2 table: %s\n", | |
1535 | strerror(-ret)); | |
1536 | res->check_errors++; | |
1537 | goto fail; | |
1538 | } | |
1539 | } | |
4f6ed88c HR |
1540 | } |
1541 | ||
1542 | ret = 0; | |
1543 | ||
1544 | fail: | |
1545 | qemu_vfree(l2_table); | |
1546 | return ret; | |
1547 | } | |
1548 | ||
6ca56bf5 HR |
1549 | /* |
1550 | * Checks consistency of refblocks and accounts for each refblock in | |
1551 | * *refcount_table. | |
1552 | */ | |
1553 | static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res, | |
f307b255 | 1554 | BdrvCheckMode fix, bool *rebuild, |
7453c96b | 1555 | void **refcount_table, int64_t *nb_clusters) |
6ca56bf5 HR |
1556 | { |
1557 | BDRVQcowState *s = bs->opaque; | |
001c158d | 1558 | int64_t i, size; |
fef4d3d5 | 1559 | int ret; |
6ca56bf5 | 1560 | |
f7d0fe02 | 1561 | for(i = 0; i < s->refcount_table_size; i++) { |
6882c8fa | 1562 | uint64_t offset, cluster; |
f7d0fe02 | 1563 | offset = s->refcount_table[i]; |
6882c8fa | 1564 | cluster = offset >> s->cluster_bits; |
746c3cb5 KW |
1565 | |
1566 | /* Refcount blocks are cluster aligned */ | |
ac95acdb | 1567 | if (offset_into_cluster(s, offset)) { |
166acf54 | 1568 | fprintf(stderr, "ERROR refcount block %" PRId64 " is not " |
746c3cb5 | 1569 | "cluster aligned; refcount table entry corrupted\n", i); |
9ac228e0 | 1570 | res->corruptions++; |
f307b255 | 1571 | *rebuild = true; |
6882c8fa KW |
1572 | continue; |
1573 | } | |
1574 | ||
6ca56bf5 | 1575 | if (cluster >= *nb_clusters) { |
001c158d HR |
1576 | fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n", |
1577 | fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i); | |
1578 | ||
1579 | if (fix & BDRV_FIX_ERRORS) { | |
5fee192e | 1580 | int64_t new_nb_clusters; |
001c158d HR |
1581 | |
1582 | if (offset > INT64_MAX - s->cluster_size) { | |
1583 | ret = -EINVAL; | |
1584 | goto resize_fail; | |
1585 | } | |
1586 | ||
1587 | ret = bdrv_truncate(bs->file, offset + s->cluster_size); | |
1588 | if (ret < 0) { | |
1589 | goto resize_fail; | |
1590 | } | |
1591 | size = bdrv_getlength(bs->file); | |
1592 | if (size < 0) { | |
1593 | ret = size; | |
1594 | goto resize_fail; | |
1595 | } | |
1596 | ||
5fee192e HR |
1597 | new_nb_clusters = size_to_clusters(s, size); |
1598 | assert(new_nb_clusters >= *nb_clusters); | |
001c158d | 1599 | |
5fee192e HR |
1600 | ret = realloc_refcount_array(s, refcount_table, |
1601 | nb_clusters, new_nb_clusters); | |
1602 | if (ret < 0) { | |
001c158d | 1603 | res->check_errors++; |
5fee192e | 1604 | return ret; |
001c158d | 1605 | } |
001c158d HR |
1606 | |
1607 | if (cluster >= *nb_clusters) { | |
1608 | ret = -EINVAL; | |
1609 | goto resize_fail; | |
1610 | } | |
1611 | ||
1612 | res->corruptions_fixed++; | |
1613 | ret = inc_refcounts(bs, res, refcount_table, nb_clusters, | |
1614 | offset, s->cluster_size); | |
1615 | if (ret < 0) { | |
1616 | return ret; | |
1617 | } | |
1618 | /* No need to check whether the refcount is now greater than 1: | |
1619 | * This area was just allocated and zeroed, so it can only be | |
1620 | * exactly 1 after inc_refcounts() */ | |
1621 | continue; | |
1622 | ||
1623 | resize_fail: | |
1624 | res->corruptions++; | |
f307b255 | 1625 | *rebuild = true; |
001c158d HR |
1626 | fprintf(stderr, "ERROR could not resize image: %s\n", |
1627 | strerror(-ret)); | |
1628 | } else { | |
1629 | res->corruptions++; | |
1630 | } | |
6882c8fa | 1631 | continue; |
746c3cb5 KW |
1632 | } |
1633 | ||
f7d0fe02 | 1634 | if (offset != 0) { |
641bb63c | 1635 | ret = inc_refcounts(bs, res, refcount_table, nb_clusters, |
fef4d3d5 HR |
1636 | offset, s->cluster_size); |
1637 | if (ret < 0) { | |
1638 | return ret; | |
1639 | } | |
7453c96b | 1640 | if (s->get_refcount(*refcount_table, cluster) != 1) { |
f307b255 | 1641 | fprintf(stderr, "ERROR refcount block %" PRId64 |
7453c96b HR |
1642 | " refcount=%" PRIu64 "\n", i, |
1643 | s->get_refcount(*refcount_table, cluster)); | |
f307b255 HR |
1644 | res->corruptions++; |
1645 | *rebuild = true; | |
746c3cb5 | 1646 | } |
f7d0fe02 KW |
1647 | } |
1648 | } | |
1649 | ||
6ca56bf5 HR |
1650 | return 0; |
1651 | } | |
1652 | ||
057a3fe5 HR |
1653 | /* |
1654 | * Calculates an in-memory refcount table. | |
1655 | */ | |
1656 | static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res, | |
f307b255 | 1657 | BdrvCheckMode fix, bool *rebuild, |
7453c96b | 1658 | void **refcount_table, int64_t *nb_clusters) |
057a3fe5 HR |
1659 | { |
1660 | BDRVQcowState *s = bs->opaque; | |
1661 | int64_t i; | |
1662 | QCowSnapshot *sn; | |
1663 | int ret; | |
1664 | ||
9696df21 | 1665 | if (!*refcount_table) { |
5fee192e HR |
1666 | int64_t old_size = 0; |
1667 | ret = realloc_refcount_array(s, refcount_table, | |
1668 | &old_size, *nb_clusters); | |
1669 | if (ret < 0) { | |
9696df21 | 1670 | res->check_errors++; |
5fee192e | 1671 | return ret; |
9696df21 | 1672 | } |
057a3fe5 HR |
1673 | } |
1674 | ||
1675 | /* header */ | |
641bb63c | 1676 | ret = inc_refcounts(bs, res, refcount_table, nb_clusters, |
fef4d3d5 HR |
1677 | 0, s->cluster_size); |
1678 | if (ret < 0) { | |
1679 | return ret; | |
1680 | } | |
057a3fe5 HR |
1681 | |
1682 | /* current L1 table */ | |
641bb63c | 1683 | ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, |
057a3fe5 HR |
1684 | s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO); |
1685 | if (ret < 0) { | |
1686 | return ret; | |
1687 | } | |
1688 | ||
1689 | /* snapshots */ | |
1690 | for (i = 0; i < s->nb_snapshots; i++) { | |
1691 | sn = s->snapshots + i; | |
641bb63c | 1692 | ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, |
fef4d3d5 | 1693 | sn->l1_table_offset, sn->l1_size, 0); |
057a3fe5 HR |
1694 | if (ret < 0) { |
1695 | return ret; | |
1696 | } | |
1697 | } | |
641bb63c | 1698 | ret = inc_refcounts(bs, res, refcount_table, nb_clusters, |
fef4d3d5 HR |
1699 | s->snapshots_offset, s->snapshots_size); |
1700 | if (ret < 0) { | |
1701 | return ret; | |
1702 | } | |
057a3fe5 HR |
1703 | |
1704 | /* refcount data */ | |
641bb63c | 1705 | ret = inc_refcounts(bs, res, refcount_table, nb_clusters, |
fef4d3d5 HR |
1706 | s->refcount_table_offset, |
1707 | s->refcount_table_size * sizeof(uint64_t)); | |
1708 | if (ret < 0) { | |
1709 | return ret; | |
1710 | } | |
057a3fe5 | 1711 | |
f307b255 | 1712 | return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters); |
057a3fe5 HR |
1713 | } |
1714 | ||
6ca56bf5 HR |
1715 | /* |
1716 | * Compares the actual reference count for each cluster in the image against the | |
1717 | * refcount as reported by the refcount structures on-disk. | |
1718 | */ | |
1719 | static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res, | |
f307b255 HR |
1720 | BdrvCheckMode fix, bool *rebuild, |
1721 | int64_t *highest_cluster, | |
7453c96b | 1722 | void *refcount_table, int64_t nb_clusters) |
6ca56bf5 HR |
1723 | { |
1724 | BDRVQcowState *s = bs->opaque; | |
1725 | int64_t i; | |
0e06528e | 1726 | uint64_t refcount1, refcount2; |
7324c10f | 1727 | int ret; |
6ca56bf5 HR |
1728 | |
1729 | for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) { | |
7324c10f HR |
1730 | ret = qcow2_get_refcount(bs, i, &refcount1); |
1731 | if (ret < 0) { | |
166acf54 | 1732 | fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n", |
7324c10f | 1733 | i, strerror(-ret)); |
9ac228e0 | 1734 | res->check_errors++; |
f74550fd | 1735 | continue; |
018faafd KW |
1736 | } |
1737 | ||
7453c96b | 1738 | refcount2 = s->get_refcount(refcount_table, i); |
c6bb9ad1 FS |
1739 | |
1740 | if (refcount1 > 0 || refcount2 > 0) { | |
6ca56bf5 | 1741 | *highest_cluster = i; |
c6bb9ad1 FS |
1742 | } |
1743 | ||
f7d0fe02 | 1744 | if (refcount1 != refcount2) { |
166acf54 KW |
1745 | /* Check if we're allowed to fix the mismatch */ |
1746 | int *num_fixed = NULL; | |
f307b255 HR |
1747 | if (refcount1 == 0) { |
1748 | *rebuild = true; | |
1749 | } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) { | |
166acf54 KW |
1750 | num_fixed = &res->leaks_fixed; |
1751 | } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) { | |
1752 | num_fixed = &res->corruptions_fixed; | |
1753 | } | |
1754 | ||
0e06528e HR |
1755 | fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64 |
1756 | " reference=%" PRIu64 "\n", | |
166acf54 KW |
1757 | num_fixed != NULL ? "Repairing" : |
1758 | refcount1 < refcount2 ? "ERROR" : | |
1759 | "Leaked", | |
f7d0fe02 | 1760 | i, refcount1, refcount2); |
166acf54 KW |
1761 | |
1762 | if (num_fixed) { | |
1763 | ret = update_refcount(bs, i << s->cluster_bits, 1, | |
2aabe7c7 HR |
1764 | refcount_diff(refcount1, refcount2), |
1765 | refcount1 > refcount2, | |
6cfcb9b8 | 1766 | QCOW2_DISCARD_ALWAYS); |
166acf54 KW |
1767 | if (ret >= 0) { |
1768 | (*num_fixed)++; | |
1769 | continue; | |
1770 | } | |
1771 | } | |
1772 | ||
1773 | /* And if we couldn't, print an error */ | |
9ac228e0 KW |
1774 | if (refcount1 < refcount2) { |
1775 | res->corruptions++; | |
1776 | } else { | |
1777 | res->leaks++; | |
1778 | } | |
f7d0fe02 KW |
1779 | } |
1780 | } | |
6ca56bf5 HR |
1781 | } |
1782 | ||
c7c0681b HR |
1783 | /* |
1784 | * Allocates clusters using an in-memory refcount table (IMRT) in contrast to | |
1785 | * the on-disk refcount structures. | |
1786 | * | |
1787 | * On input, *first_free_cluster tells where to start looking, and need not | |
1788 | * actually be a free cluster; the returned offset will not be before that | |
1789 | * cluster. On output, *first_free_cluster points to the first gap found, even | |
1790 | * if that gap was too small to be used as the returned offset. | |
1791 | * | |
1792 | * Note that *first_free_cluster is a cluster index whereas the return value is | |
1793 | * an offset. | |
1794 | */ | |
1795 | static int64_t alloc_clusters_imrt(BlockDriverState *bs, | |
1796 | int cluster_count, | |
7453c96b | 1797 | void **refcount_table, |
c7c0681b HR |
1798 | int64_t *imrt_nb_clusters, |
1799 | int64_t *first_free_cluster) | |
1800 | { | |
1801 | BDRVQcowState *s = bs->opaque; | |
1802 | int64_t cluster = *first_free_cluster, i; | |
1803 | bool first_gap = true; | |
1804 | int contiguous_free_clusters; | |
5fee192e | 1805 | int ret; |
c7c0681b HR |
1806 | |
1807 | /* Starting at *first_free_cluster, find a range of at least cluster_count | |
1808 | * continuously free clusters */ | |
1809 | for (contiguous_free_clusters = 0; | |
1810 | cluster < *imrt_nb_clusters && | |
1811 | contiguous_free_clusters < cluster_count; | |
1812 | cluster++) | |
1813 | { | |
7453c96b | 1814 | if (!s->get_refcount(*refcount_table, cluster)) { |
c7c0681b HR |
1815 | contiguous_free_clusters++; |
1816 | if (first_gap) { | |
1817 | /* If this is the first free cluster found, update | |
1818 | * *first_free_cluster accordingly */ | |
1819 | *first_free_cluster = cluster; | |
1820 | first_gap = false; | |
1821 | } | |
1822 | } else if (contiguous_free_clusters) { | |
1823 | contiguous_free_clusters = 0; | |
1824 | } | |
1825 | } | |
1826 | ||
1827 | /* If contiguous_free_clusters is greater than zero, it contains the number | |
1828 | * of continuously free clusters until the current cluster; the first free | |
1829 | * cluster in the current "gap" is therefore | |
1830 | * cluster - contiguous_free_clusters */ | |
1831 | ||
1832 | /* If no such range could be found, grow the in-memory refcount table | |
1833 | * accordingly to append free clusters at the end of the image */ | |
1834 | if (contiguous_free_clusters < cluster_count) { | |
c7c0681b HR |
1835 | /* contiguous_free_clusters clusters are already empty at the image end; |
1836 | * we need cluster_count clusters; therefore, we have to allocate | |
1837 | * cluster_count - contiguous_free_clusters new clusters at the end of | |
1838 | * the image (which is the current value of cluster; note that cluster | |
1839 | * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond | |
1840 | * the image end) */ | |
5fee192e HR |
1841 | ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters, |
1842 | cluster + cluster_count | |
1843 | - contiguous_free_clusters); | |
1844 | if (ret < 0) { | |
1845 | return ret; | |
c7c0681b | 1846 | } |
c7c0681b HR |
1847 | } |
1848 | ||
1849 | /* Go back to the first free cluster */ | |
1850 | cluster -= contiguous_free_clusters; | |
1851 | for (i = 0; i < cluster_count; i++) { | |
7453c96b | 1852 | s->set_refcount(*refcount_table, cluster + i, 1); |
c7c0681b HR |
1853 | } |
1854 | ||
1855 | return cluster << s->cluster_bits; | |
1856 | } | |
1857 | ||
1858 | /* | |
1859 | * Creates a new refcount structure based solely on the in-memory information | |
1860 | * given through *refcount_table. All necessary allocations will be reflected | |
1861 | * in that array. | |
1862 | * | |
1863 | * On success, the old refcount structure is leaked (it will be covered by the | |
1864 | * new refcount structure). | |
1865 | */ | |
1866 | static int rebuild_refcount_structure(BlockDriverState *bs, | |
1867 | BdrvCheckResult *res, | |
7453c96b | 1868 | void **refcount_table, |
c7c0681b HR |
1869 | int64_t *nb_clusters) |
1870 | { | |
1871 | BDRVQcowState *s = bs->opaque; | |
1872 | int64_t first_free_cluster = 0, reftable_offset = -1, cluster = 0; | |
1873 | int64_t refblock_offset, refblock_start, refblock_index; | |
1874 | uint32_t reftable_size = 0; | |
1875 | uint64_t *on_disk_reftable = NULL; | |
7453c96b HR |
1876 | void *on_disk_refblock; |
1877 | int ret = 0; | |
c7c0681b HR |
1878 | struct { |
1879 | uint64_t reftable_offset; | |
1880 | uint32_t reftable_clusters; | |
1881 | } QEMU_PACKED reftable_offset_and_clusters; | |
1882 | ||
1883 | qcow2_cache_empty(bs, s->refcount_block_cache); | |
1884 | ||
1885 | write_refblocks: | |
1886 | for (; cluster < *nb_clusters; cluster++) { | |
7453c96b | 1887 | if (!s->get_refcount(*refcount_table, cluster)) { |
c7c0681b HR |
1888 | continue; |
1889 | } | |
1890 | ||
1891 | refblock_index = cluster >> s->refcount_block_bits; | |
1892 | refblock_start = refblock_index << s->refcount_block_bits; | |
1893 | ||
1894 | /* Don't allocate a cluster in a refblock already written to disk */ | |
1895 | if (first_free_cluster < refblock_start) { | |
1896 | first_free_cluster = refblock_start; | |
1897 | } | |
1898 | refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table, | |
1899 | nb_clusters, &first_free_cluster); | |
1900 | if (refblock_offset < 0) { | |
1901 | fprintf(stderr, "ERROR allocating refblock: %s\n", | |
1902 | strerror(-refblock_offset)); | |
1903 | res->check_errors++; | |
1904 | ret = refblock_offset; | |
1905 | goto fail; | |
1906 | } | |
1907 | ||
1908 | if (reftable_size <= refblock_index) { | |
1909 | uint32_t old_reftable_size = reftable_size; | |
1910 | uint64_t *new_on_disk_reftable; | |
1911 | ||
1912 | reftable_size = ROUND_UP((refblock_index + 1) * sizeof(uint64_t), | |
1913 | s->cluster_size) / sizeof(uint64_t); | |
1914 | new_on_disk_reftable = g_try_realloc(on_disk_reftable, | |
1915 | reftable_size * | |
1916 | sizeof(uint64_t)); | |
1917 | if (!new_on_disk_reftable) { | |
1918 | res->check_errors++; | |
1919 | ret = -ENOMEM; | |
1920 | goto fail; | |
1921 | } | |
1922 | on_disk_reftable = new_on_disk_reftable; | |
1923 | ||
1924 | memset(on_disk_reftable + old_reftable_size, 0, | |
1925 | (reftable_size - old_reftable_size) * sizeof(uint64_t)); | |
1926 | ||
1927 | /* The offset we have for the reftable is now no longer valid; | |
1928 | * this will leak that range, but we can easily fix that by running | |
1929 | * a leak-fixing check after this rebuild operation */ | |
1930 | reftable_offset = -1; | |
1931 | } | |
1932 | on_disk_reftable[refblock_index] = refblock_offset; | |
1933 | ||
1934 | /* If this is apparently the last refblock (for now), try to squeeze the | |
1935 | * reftable in */ | |
1936 | if (refblock_index == (*nb_clusters - 1) >> s->refcount_block_bits && | |
1937 | reftable_offset < 0) | |
1938 | { | |
1939 | uint64_t reftable_clusters = size_to_clusters(s, reftable_size * | |
1940 | sizeof(uint64_t)); | |
1941 | reftable_offset = alloc_clusters_imrt(bs, reftable_clusters, | |
1942 | refcount_table, nb_clusters, | |
1943 | &first_free_cluster); | |
1944 | if (reftable_offset < 0) { | |
1945 | fprintf(stderr, "ERROR allocating reftable: %s\n", | |
1946 | strerror(-reftable_offset)); | |
1947 | res->check_errors++; | |
1948 | ret = reftable_offset; | |
1949 | goto fail; | |
1950 | } | |
1951 | } | |
1952 | ||
1953 | ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset, | |
1954 | s->cluster_size); | |
1955 | if (ret < 0) { | |
1956 | fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret)); | |
1957 | goto fail; | |
1958 | } | |
1959 | ||
7453c96b HR |
1960 | /* The size of *refcount_table is always cluster-aligned, therefore the |
1961 | * write operation will not overflow */ | |
1962 | on_disk_refblock = (void *)((char *) *refcount_table + | |
1963 | refblock_index * s->cluster_size); | |
c7c0681b HR |
1964 | |
1965 | ret = bdrv_write(bs->file, refblock_offset / BDRV_SECTOR_SIZE, | |
7453c96b | 1966 | on_disk_refblock, s->cluster_sectors); |
c7c0681b HR |
1967 | if (ret < 0) { |
1968 | fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret)); | |
1969 | goto fail; | |
1970 | } | |
1971 | ||
1972 | /* Go to the end of this refblock */ | |
1973 | cluster = refblock_start + s->refcount_block_size - 1; | |
1974 | } | |
1975 | ||
1976 | if (reftable_offset < 0) { | |
1977 | uint64_t post_refblock_start, reftable_clusters; | |
1978 | ||
1979 | post_refblock_start = ROUND_UP(*nb_clusters, s->refcount_block_size); | |
1980 | reftable_clusters = size_to_clusters(s, | |
1981 | reftable_size * sizeof(uint64_t)); | |
1982 | /* Not pretty but simple */ | |
1983 | if (first_free_cluster < post_refblock_start) { | |
1984 | first_free_cluster = post_refblock_start; | |
1985 | } | |
1986 | reftable_offset = alloc_clusters_imrt(bs, reftable_clusters, | |
1987 | refcount_table, nb_clusters, | |
1988 | &first_free_cluster); | |
1989 | if (reftable_offset < 0) { | |
1990 | fprintf(stderr, "ERROR allocating reftable: %s\n", | |
1991 | strerror(-reftable_offset)); | |
1992 | res->check_errors++; | |
1993 | ret = reftable_offset; | |
1994 | goto fail; | |
1995 | } | |
1996 | ||
1997 | goto write_refblocks; | |
1998 | } | |
1999 | ||
2000 | assert(on_disk_reftable); | |
2001 | ||
2002 | for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) { | |
2003 | cpu_to_be64s(&on_disk_reftable[refblock_index]); | |
2004 | } | |
2005 | ||
2006 | ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset, | |
2007 | reftable_size * sizeof(uint64_t)); | |
2008 | if (ret < 0) { | |
2009 | fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret)); | |
2010 | goto fail; | |
2011 | } | |
2012 | ||
2013 | assert(reftable_size < INT_MAX / sizeof(uint64_t)); | |
2014 | ret = bdrv_pwrite(bs->file, reftable_offset, on_disk_reftable, | |
2015 | reftable_size * sizeof(uint64_t)); | |
2016 | if (ret < 0) { | |
2017 | fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret)); | |
2018 | goto fail; | |
2019 | } | |
2020 | ||
2021 | /* Enter new reftable into the image header */ | |
2022 | cpu_to_be64w(&reftable_offset_and_clusters.reftable_offset, | |
2023 | reftable_offset); | |
2024 | cpu_to_be32w(&reftable_offset_and_clusters.reftable_clusters, | |
2025 | size_to_clusters(s, reftable_size * sizeof(uint64_t))); | |
2026 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, | |
2027 | refcount_table_offset), | |
2028 | &reftable_offset_and_clusters, | |
2029 | sizeof(reftable_offset_and_clusters)); | |
2030 | if (ret < 0) { | |
2031 | fprintf(stderr, "ERROR setting reftable: %s\n", strerror(-ret)); | |
2032 | goto fail; | |
2033 | } | |
2034 | ||
2035 | for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) { | |
2036 | be64_to_cpus(&on_disk_reftable[refblock_index]); | |
2037 | } | |
2038 | s->refcount_table = on_disk_reftable; | |
2039 | s->refcount_table_offset = reftable_offset; | |
2040 | s->refcount_table_size = reftable_size; | |
2041 | ||
2042 | return 0; | |
2043 | ||
2044 | fail: | |
2045 | g_free(on_disk_reftable); | |
2046 | return ret; | |
2047 | } | |
2048 | ||
6ca56bf5 HR |
2049 | /* |
2050 | * Checks an image for refcount consistency. | |
2051 | * | |
2052 | * Returns 0 if no errors are found, the number of errors in case the image is | |
2053 | * detected as corrupted, and -errno when an internal error occurred. | |
2054 | */ | |
2055 | int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res, | |
2056 | BdrvCheckMode fix) | |
2057 | { | |
2058 | BDRVQcowState *s = bs->opaque; | |
c7c0681b | 2059 | BdrvCheckResult pre_compare_res; |
6ca56bf5 | 2060 | int64_t size, highest_cluster, nb_clusters; |
7453c96b | 2061 | void *refcount_table = NULL; |
f307b255 | 2062 | bool rebuild = false; |
6ca56bf5 HR |
2063 | int ret; |
2064 | ||
2065 | size = bdrv_getlength(bs->file); | |
2066 | if (size < 0) { | |
2067 | res->check_errors++; | |
2068 | return size; | |
2069 | } | |
2070 | ||
2071 | nb_clusters = size_to_clusters(s, size); | |
2072 | if (nb_clusters > INT_MAX) { | |
2073 | res->check_errors++; | |
2074 | return -EFBIG; | |
2075 | } | |
2076 | ||
2077 | res->bfi.total_clusters = | |
2078 | size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE); | |
2079 | ||
f307b255 HR |
2080 | ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table, |
2081 | &nb_clusters); | |
6ca56bf5 HR |
2082 | if (ret < 0) { |
2083 | goto fail; | |
2084 | } | |
2085 | ||
c7c0681b HR |
2086 | /* In case we don't need to rebuild the refcount structure (but want to fix |
2087 | * something), this function is immediately called again, in which case the | |
2088 | * result should be ignored */ | |
2089 | pre_compare_res = *res; | |
2090 | compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table, | |
6ca56bf5 | 2091 | nb_clusters); |
f7d0fe02 | 2092 | |
c7c0681b | 2093 | if (rebuild && (fix & BDRV_FIX_ERRORS)) { |
791230d8 HR |
2094 | BdrvCheckResult old_res = *res; |
2095 | int fresh_leaks = 0; | |
2096 | ||
c7c0681b HR |
2097 | fprintf(stderr, "Rebuilding refcount structure\n"); |
2098 | ret = rebuild_refcount_structure(bs, res, &refcount_table, | |
2099 | &nb_clusters); | |
2100 | if (ret < 0) { | |
2101 | goto fail; | |
2102 | } | |
791230d8 HR |
2103 | |
2104 | res->corruptions = 0; | |
2105 | res->leaks = 0; | |
2106 | ||
2107 | /* Because the old reftable has been exchanged for a new one the | |
2108 | * references have to be recalculated */ | |
2109 | rebuild = false; | |
7453c96b | 2110 | memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters)); |
791230d8 HR |
2111 | ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table, |
2112 | &nb_clusters); | |
2113 | if (ret < 0) { | |
2114 | goto fail; | |
2115 | } | |
2116 | ||
2117 | if (fix & BDRV_FIX_LEAKS) { | |
2118 | /* The old refcount structures are now leaked, fix it; the result | |
2119 | * can be ignored, aside from leaks which were introduced by | |
2120 | * rebuild_refcount_structure() that could not be fixed */ | |
2121 | BdrvCheckResult saved_res = *res; | |
2122 | *res = (BdrvCheckResult){ 0 }; | |
2123 | ||
2124 | compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild, | |
2125 | &highest_cluster, refcount_table, nb_clusters); | |
2126 | if (rebuild) { | |
2127 | fprintf(stderr, "ERROR rebuilt refcount structure is still " | |
2128 | "broken\n"); | |
2129 | } | |
2130 | ||
2131 | /* Any leaks accounted for here were introduced by | |
2132 | * rebuild_refcount_structure() because that function has created a | |
2133 | * new refcount structure from scratch */ | |
2134 | fresh_leaks = res->leaks; | |
2135 | *res = saved_res; | |
2136 | } | |
2137 | ||
2138 | if (res->corruptions < old_res.corruptions) { | |
2139 | res->corruptions_fixed += old_res.corruptions - res->corruptions; | |
2140 | } | |
2141 | if (res->leaks < old_res.leaks) { | |
2142 | res->leaks_fixed += old_res.leaks - res->leaks; | |
2143 | } | |
2144 | res->leaks += fresh_leaks; | |
c7c0681b HR |
2145 | } else if (fix) { |
2146 | if (rebuild) { | |
2147 | fprintf(stderr, "ERROR need to rebuild refcount structures\n"); | |
2148 | res->check_errors++; | |
2149 | ret = -EIO; | |
2150 | goto fail; | |
2151 | } | |
2152 | ||
2153 | if (res->leaks || res->corruptions) { | |
2154 | *res = pre_compare_res; | |
2155 | compare_refcounts(bs, res, fix, &rebuild, &highest_cluster, | |
2156 | refcount_table, nb_clusters); | |
2157 | } | |
f307b255 HR |
2158 | } |
2159 | ||
4f6ed88c | 2160 | /* check OFLAG_COPIED */ |
e23e400e | 2161 | ret = check_oflag_copied(bs, res, fix); |
4f6ed88c HR |
2162 | if (ret < 0) { |
2163 | goto fail; | |
2164 | } | |
2165 | ||
c6bb9ad1 | 2166 | res->image_end_offset = (highest_cluster + 1) * s->cluster_size; |
80fa3341 KW |
2167 | ret = 0; |
2168 | ||
2169 | fail: | |
7267c094 | 2170 | g_free(refcount_table); |
f7d0fe02 | 2171 | |
80fa3341 | 2172 | return ret; |
f7d0fe02 KW |
2173 | } |
2174 | ||
a40f1c2a HR |
2175 | #define overlaps_with(ofs, sz) \ |
2176 | ranges_overlap(offset, size, ofs, sz) | |
2177 | ||
2178 | /* | |
2179 | * Checks if the given offset into the image file is actually free to use by | |
2180 | * looking for overlaps with important metadata sections (L1/L2 tables etc.), | |
2181 | * i.e. a sanity check without relying on the refcount tables. | |
2182 | * | |
231bb267 HR |
2183 | * The ign parameter specifies what checks not to perform (being a bitmask of |
2184 | * QCow2MetadataOverlap values), i.e., what sections to ignore. | |
a40f1c2a HR |
2185 | * |
2186 | * Returns: | |
2187 | * - 0 if writing to this offset will not affect the mentioned metadata | |
2188 | * - a positive QCow2MetadataOverlap value indicating one overlapping section | |
2189 | * - a negative value (-errno) indicating an error while performing a check, | |
2190 | * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2 | |
2191 | */ | |
231bb267 | 2192 | int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset, |
a40f1c2a HR |
2193 | int64_t size) |
2194 | { | |
2195 | BDRVQcowState *s = bs->opaque; | |
3e355390 | 2196 | int chk = s->overlap_check & ~ign; |
a40f1c2a HR |
2197 | int i, j; |
2198 | ||
2199 | if (!size) { | |
2200 | return 0; | |
2201 | } | |
2202 | ||
2203 | if (chk & QCOW2_OL_MAIN_HEADER) { | |
2204 | if (offset < s->cluster_size) { | |
2205 | return QCOW2_OL_MAIN_HEADER; | |
2206 | } | |
2207 | } | |
2208 | ||
2209 | /* align range to test to cluster boundaries */ | |
2210 | size = align_offset(offset_into_cluster(s, offset) + size, s->cluster_size); | |
2211 | offset = start_of_cluster(s, offset); | |
2212 | ||
2213 | if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) { | |
2214 | if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) { | |
2215 | return QCOW2_OL_ACTIVE_L1; | |
2216 | } | |
2217 | } | |
2218 | ||
2219 | if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) { | |
2220 | if (overlaps_with(s->refcount_table_offset, | |
2221 | s->refcount_table_size * sizeof(uint64_t))) { | |
2222 | return QCOW2_OL_REFCOUNT_TABLE; | |
2223 | } | |
2224 | } | |
2225 | ||
2226 | if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) { | |
2227 | if (overlaps_with(s->snapshots_offset, s->snapshots_size)) { | |
2228 | return QCOW2_OL_SNAPSHOT_TABLE; | |
2229 | } | |
2230 | } | |
2231 | ||
2232 | if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) { | |
2233 | for (i = 0; i < s->nb_snapshots; i++) { | |
2234 | if (s->snapshots[i].l1_size && | |
2235 | overlaps_with(s->snapshots[i].l1_table_offset, | |
2236 | s->snapshots[i].l1_size * sizeof(uint64_t))) { | |
2237 | return QCOW2_OL_INACTIVE_L1; | |
2238 | } | |
2239 | } | |
2240 | } | |
2241 | ||
2242 | if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) { | |
2243 | for (i = 0; i < s->l1_size; i++) { | |
2244 | if ((s->l1_table[i] & L1E_OFFSET_MASK) && | |
2245 | overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK, | |
2246 | s->cluster_size)) { | |
2247 | return QCOW2_OL_ACTIVE_L2; | |
2248 | } | |
2249 | } | |
2250 | } | |
2251 | ||
2252 | if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) { | |
2253 | for (i = 0; i < s->refcount_table_size; i++) { | |
2254 | if ((s->refcount_table[i] & REFT_OFFSET_MASK) && | |
2255 | overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK, | |
2256 | s->cluster_size)) { | |
2257 | return QCOW2_OL_REFCOUNT_BLOCK; | |
2258 | } | |
2259 | } | |
2260 | } | |
2261 | ||
2262 | if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) { | |
2263 | for (i = 0; i < s->nb_snapshots; i++) { | |
2264 | uint64_t l1_ofs = s->snapshots[i].l1_table_offset; | |
2265 | uint32_t l1_sz = s->snapshots[i].l1_size; | |
998b959c | 2266 | uint64_t l1_sz2 = l1_sz * sizeof(uint64_t); |
de82815d | 2267 | uint64_t *l1 = g_try_malloc(l1_sz2); |
a40f1c2a HR |
2268 | int ret; |
2269 | ||
de82815d KW |
2270 | if (l1_sz2 && l1 == NULL) { |
2271 | return -ENOMEM; | |
2272 | } | |
2273 | ||
998b959c | 2274 | ret = bdrv_pread(bs->file, l1_ofs, l1, l1_sz2); |
a40f1c2a HR |
2275 | if (ret < 0) { |
2276 | g_free(l1); | |
2277 | return ret; | |
2278 | } | |
2279 | ||
2280 | for (j = 0; j < l1_sz; j++) { | |
1e242b55 HR |
2281 | uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK; |
2282 | if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) { | |
a40f1c2a HR |
2283 | g_free(l1); |
2284 | return QCOW2_OL_INACTIVE_L2; | |
2285 | } | |
2286 | } | |
2287 | ||
2288 | g_free(l1); | |
2289 | } | |
2290 | } | |
2291 | ||
2292 | return 0; | |
2293 | } | |
2294 | ||
2295 | static const char *metadata_ol_names[] = { | |
2296 | [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header", | |
2297 | [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table", | |
2298 | [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table", | |
2299 | [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table", | |
2300 | [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block", | |
2301 | [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table", | |
2302 | [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table", | |
2303 | [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table", | |
2304 | }; | |
2305 | ||
2306 | /* | |
2307 | * First performs a check for metadata overlaps (through | |
2308 | * qcow2_check_metadata_overlap); if that fails with a negative value (error | |
2309 | * while performing a check), that value is returned. If an impending overlap | |
2310 | * is detected, the BDS will be made unusable, the qcow2 file marked corrupt | |
2311 | * and -EIO returned. | |
2312 | * | |
2313 | * Returns 0 if there were neither overlaps nor errors while checking for | |
2314 | * overlaps; or a negative value (-errno) on error. | |
2315 | */ | |
231bb267 | 2316 | int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset, |
a40f1c2a HR |
2317 | int64_t size) |
2318 | { | |
231bb267 | 2319 | int ret = qcow2_check_metadata_overlap(bs, ign, offset, size); |
a40f1c2a HR |
2320 | |
2321 | if (ret < 0) { | |
2322 | return ret; | |
2323 | } else if (ret > 0) { | |
2324 | int metadata_ol_bitnr = ffs(ret) - 1; | |
a40f1c2a HR |
2325 | assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR); |
2326 | ||
adb43552 HR |
2327 | qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid " |
2328 | "write on metadata (overlaps with %s)", | |
2329 | metadata_ol_names[metadata_ol_bitnr]); | |
a40f1c2a HR |
2330 | return -EIO; |
2331 | } | |
2332 | ||
2333 | return 0; | |
2334 | } |