]>
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 KW |
27 | #include "block/qcow2.h" |
28 | ||
29 | static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size); | |
92dcb59f | 30 | static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, |
f7d0fe02 | 31 | int64_t offset, int64_t length, |
6cfcb9b8 | 32 | int addend, enum qcow2_discard_type type); |
f7d0fe02 | 33 | |
3b88e52b | 34 | |
f7d0fe02 KW |
35 | /*********************************************************/ |
36 | /* refcount handling */ | |
37 | ||
ed6ccf0f | 38 | int qcow2_refcount_init(BlockDriverState *bs) |
f7d0fe02 KW |
39 | { |
40 | BDRVQcowState *s = bs->opaque; | |
41 | int ret, refcount_table_size2, i; | |
42 | ||
f7d0fe02 | 43 | refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t); |
7267c094 | 44 | s->refcount_table = g_malloc(refcount_table_size2); |
f7d0fe02 | 45 | if (s->refcount_table_size > 0) { |
66f82cee KW |
46 | BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD); |
47 | ret = bdrv_pread(bs->file, s->refcount_table_offset, | |
f7d0fe02 KW |
48 | s->refcount_table, refcount_table_size2); |
49 | if (ret != refcount_table_size2) | |
50 | goto fail; | |
51 | for(i = 0; i < s->refcount_table_size; i++) | |
52 | be64_to_cpus(&s->refcount_table[i]); | |
53 | } | |
54 | return 0; | |
55 | fail: | |
56 | return -ENOMEM; | |
57 | } | |
58 | ||
ed6ccf0f | 59 | void qcow2_refcount_close(BlockDriverState *bs) |
f7d0fe02 KW |
60 | { |
61 | BDRVQcowState *s = bs->opaque; | |
7267c094 | 62 | g_free(s->refcount_table); |
f7d0fe02 KW |
63 | } |
64 | ||
65 | ||
66 | static int load_refcount_block(BlockDriverState *bs, | |
29c1a730 KW |
67 | int64_t refcount_block_offset, |
68 | void **refcount_block) | |
f7d0fe02 KW |
69 | { |
70 | BDRVQcowState *s = bs->opaque; | |
71 | int ret; | |
3b88e52b | 72 | |
66f82cee | 73 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD); |
29c1a730 KW |
74 | ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, |
75 | refcount_block); | |
e14e8ba5 | 76 | |
29c1a730 | 77 | return ret; |
f7d0fe02 KW |
78 | } |
79 | ||
018faafd KW |
80 | /* |
81 | * Returns the refcount of the cluster given by its index. Any non-negative | |
82 | * return value is the refcount of the cluster, negative values are -errno | |
83 | * and indicate an error. | |
84 | */ | |
f7d0fe02 KW |
85 | static int get_refcount(BlockDriverState *bs, int64_t cluster_index) |
86 | { | |
87 | BDRVQcowState *s = bs->opaque; | |
88 | int refcount_table_index, block_index; | |
89 | int64_t refcount_block_offset; | |
018faafd | 90 | int ret; |
29c1a730 KW |
91 | uint16_t *refcount_block; |
92 | uint16_t refcount; | |
f7d0fe02 KW |
93 | |
94 | refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); | |
95 | if (refcount_table_index >= s->refcount_table_size) | |
96 | return 0; | |
97 | refcount_block_offset = s->refcount_table[refcount_table_index]; | |
98 | if (!refcount_block_offset) | |
99 | return 0; | |
29c1a730 KW |
100 | |
101 | ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, | |
102 | (void**) &refcount_block); | |
103 | if (ret < 0) { | |
104 | return ret; | |
f7d0fe02 | 105 | } |
29c1a730 | 106 | |
f7d0fe02 KW |
107 | block_index = cluster_index & |
108 | ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); | |
29c1a730 KW |
109 | refcount = be16_to_cpu(refcount_block[block_index]); |
110 | ||
111 | ret = qcow2_cache_put(bs, s->refcount_block_cache, | |
112 | (void**) &refcount_block); | |
113 | if (ret < 0) { | |
114 | return ret; | |
115 | } | |
116 | ||
117 | return refcount; | |
f7d0fe02 KW |
118 | } |
119 | ||
05121aed KW |
120 | /* |
121 | * Rounds the refcount table size up to avoid growing the table for each single | |
122 | * refcount block that is allocated. | |
123 | */ | |
124 | static unsigned int next_refcount_table_size(BDRVQcowState *s, | |
125 | unsigned int min_size) | |
126 | { | |
127 | unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1; | |
128 | unsigned int refcount_table_clusters = | |
129 | MAX(1, s->refcount_table_size >> (s->cluster_bits - 3)); | |
130 | ||
131 | while (min_clusters > refcount_table_clusters) { | |
132 | refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2; | |
133 | } | |
134 | ||
135 | return refcount_table_clusters << (s->cluster_bits - 3); | |
136 | } | |
137 | ||
92dcb59f KW |
138 | |
139 | /* Checks if two offsets are described by the same refcount block */ | |
140 | static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a, | |
141 | uint64_t offset_b) | |
142 | { | |
143 | uint64_t block_a = offset_a >> (2 * s->cluster_bits - REFCOUNT_SHIFT); | |
144 | uint64_t block_b = offset_b >> (2 * s->cluster_bits - REFCOUNT_SHIFT); | |
145 | ||
146 | return (block_a == block_b); | |
147 | } | |
148 | ||
149 | /* | |
150 | * Loads a refcount block. If it doesn't exist yet, it is allocated first | |
151 | * (including growing the refcount table if needed). | |
152 | * | |
29c1a730 | 153 | * Returns 0 on success or -errno in error case |
92dcb59f | 154 | */ |
29c1a730 KW |
155 | static int alloc_refcount_block(BlockDriverState *bs, |
156 | int64_t cluster_index, uint16_t **refcount_block) | |
f7d0fe02 KW |
157 | { |
158 | BDRVQcowState *s = bs->opaque; | |
92dcb59f KW |
159 | unsigned int refcount_table_index; |
160 | int ret; | |
161 | ||
66f82cee | 162 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); |
8252278a | 163 | |
92dcb59f KW |
164 | /* Find the refcount block for the given cluster */ |
165 | refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); | |
166 | ||
167 | if (refcount_table_index < s->refcount_table_size) { | |
168 | ||
169 | uint64_t refcount_block_offset = | |
76dc9e0c | 170 | s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK; |
92dcb59f KW |
171 | |
172 | /* If it's already there, we're done */ | |
173 | if (refcount_block_offset) { | |
29c1a730 KW |
174 | return load_refcount_block(bs, refcount_block_offset, |
175 | (void**) refcount_block); | |
92dcb59f KW |
176 | } |
177 | } | |
178 | ||
179 | /* | |
180 | * If we came here, we need to allocate something. Something is at least | |
181 | * a cluster for the new refcount block. It may also include a new refcount | |
182 | * table if the old refcount table is too small. | |
183 | * | |
184 | * Note that allocating clusters here needs some special care: | |
185 | * | |
186 | * - We can't use the normal qcow2_alloc_clusters(), it would try to | |
187 | * increase the refcount and very likely we would end up with an endless | |
188 | * recursion. Instead we must place the refcount blocks in a way that | |
189 | * they can describe them themselves. | |
190 | * | |
191 | * - We need to consider that at this point we are inside update_refcounts | |
192 | * and doing the initial refcount increase. This means that some clusters | |
193 | * have already been allocated by the caller, but their refcount isn't | |
194 | * accurate yet. free_cluster_index tells us where this allocation ends | |
195 | * as long as we don't overwrite it by freeing clusters. | |
196 | * | |
197 | * - alloc_clusters_noref and qcow2_free_clusters may load a different | |
198 | * refcount block into the cache | |
199 | */ | |
200 | ||
29c1a730 KW |
201 | *refcount_block = NULL; |
202 | ||
203 | /* We write to the refcount table, so we might depend on L2 tables */ | |
9991923b SH |
204 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
205 | if (ret < 0) { | |
206 | return ret; | |
207 | } | |
92dcb59f KW |
208 | |
209 | /* Allocate the refcount block itself and mark it as used */ | |
2eaa8f63 KW |
210 | int64_t new_block = alloc_clusters_noref(bs, s->cluster_size); |
211 | if (new_block < 0) { | |
212 | return new_block; | |
213 | } | |
f7d0fe02 | 214 | |
f7d0fe02 | 215 | #ifdef DEBUG_ALLOC2 |
92dcb59f KW |
216 | fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64 |
217 | " at %" PRIx64 "\n", | |
218 | refcount_table_index, cluster_index << s->cluster_bits, new_block); | |
f7d0fe02 | 219 | #endif |
92dcb59f KW |
220 | |
221 | if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) { | |
25408c09 | 222 | /* Zero the new refcount block before updating it */ |
29c1a730 KW |
223 | ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, |
224 | (void**) refcount_block); | |
225 | if (ret < 0) { | |
226 | goto fail_block; | |
227 | } | |
228 | ||
229 | memset(*refcount_block, 0, s->cluster_size); | |
25408c09 | 230 | |
92dcb59f KW |
231 | /* The block describes itself, need to update the cache */ |
232 | int block_index = (new_block >> s->cluster_bits) & | |
233 | ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); | |
29c1a730 | 234 | (*refcount_block)[block_index] = cpu_to_be16(1); |
92dcb59f KW |
235 | } else { |
236 | /* Described somewhere else. This can recurse at most twice before we | |
237 | * arrive at a block that describes itself. */ | |
6cfcb9b8 KW |
238 | ret = update_refcount(bs, new_block, s->cluster_size, 1, |
239 | QCOW2_DISCARD_NEVER); | |
92dcb59f KW |
240 | if (ret < 0) { |
241 | goto fail_block; | |
242 | } | |
25408c09 | 243 | |
9991923b SH |
244 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
245 | if (ret < 0) { | |
246 | goto fail_block; | |
247 | } | |
1c4c2814 | 248 | |
25408c09 KW |
249 | /* Initialize the new refcount block only after updating its refcount, |
250 | * update_refcount uses the refcount cache itself */ | |
29c1a730 KW |
251 | ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, |
252 | (void**) refcount_block); | |
253 | if (ret < 0) { | |
254 | goto fail_block; | |
255 | } | |
256 | ||
257 | memset(*refcount_block, 0, s->cluster_size); | |
92dcb59f KW |
258 | } |
259 | ||
260 | /* Now the new refcount block needs to be written to disk */ | |
66f82cee | 261 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE); |
29c1a730 KW |
262 | qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block); |
263 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
92dcb59f KW |
264 | if (ret < 0) { |
265 | goto fail_block; | |
266 | } | |
267 | ||
268 | /* If the refcount table is big enough, just hook the block up there */ | |
269 | if (refcount_table_index < s->refcount_table_size) { | |
270 | uint64_t data64 = cpu_to_be64(new_block); | |
66f82cee | 271 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP); |
8b3b7206 | 272 | ret = bdrv_pwrite_sync(bs->file, |
92dcb59f KW |
273 | s->refcount_table_offset + refcount_table_index * sizeof(uint64_t), |
274 | &data64, sizeof(data64)); | |
275 | if (ret < 0) { | |
276 | goto fail_block; | |
277 | } | |
278 | ||
279 | s->refcount_table[refcount_table_index] = new_block; | |
29c1a730 KW |
280 | return 0; |
281 | } | |
282 | ||
283 | ret = qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block); | |
284 | if (ret < 0) { | |
285 | goto fail_block; | |
92dcb59f KW |
286 | } |
287 | ||
288 | /* | |
289 | * If we come here, we need to grow the refcount table. Again, a new | |
290 | * refcount table needs some space and we can't simply allocate to avoid | |
291 | * endless recursion. | |
292 | * | |
293 | * Therefore let's grab new refcount blocks at the end of the image, which | |
294 | * will describe themselves and the new refcount table. This way we can | |
295 | * reference them only in the new table and do the switch to the new | |
296 | * refcount table at once without producing an inconsistent state in | |
297 | * between. | |
298 | */ | |
66f82cee | 299 | BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW); |
8252278a | 300 | |
92dcb59f KW |
301 | /* Calculate the number of refcount blocks needed so far */ |
302 | uint64_t refcount_block_clusters = 1 << (s->cluster_bits - REFCOUNT_SHIFT); | |
303 | uint64_t blocks_used = (s->free_cluster_index + | |
304 | refcount_block_clusters - 1) / refcount_block_clusters; | |
305 | ||
306 | /* And now we need at least one block more for the new metadata */ | |
307 | uint64_t table_size = next_refcount_table_size(s, blocks_used + 1); | |
308 | uint64_t last_table_size; | |
309 | uint64_t blocks_clusters; | |
310 | do { | |
a3548077 KW |
311 | uint64_t table_clusters = |
312 | size_to_clusters(s, table_size * sizeof(uint64_t)); | |
92dcb59f KW |
313 | blocks_clusters = 1 + |
314 | ((table_clusters + refcount_block_clusters - 1) | |
315 | / refcount_block_clusters); | |
316 | uint64_t meta_clusters = table_clusters + blocks_clusters; | |
317 | ||
318 | last_table_size = table_size; | |
319 | table_size = next_refcount_table_size(s, blocks_used + | |
320 | ((meta_clusters + refcount_block_clusters - 1) | |
321 | / refcount_block_clusters)); | |
322 | ||
323 | } while (last_table_size != table_size); | |
324 | ||
325 | #ifdef DEBUG_ALLOC2 | |
326 | fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n", | |
327 | s->refcount_table_size, table_size); | |
328 | #endif | |
329 | ||
330 | /* Create the new refcount table and blocks */ | |
331 | uint64_t meta_offset = (blocks_used * refcount_block_clusters) * | |
332 | s->cluster_size; | |
333 | uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size; | |
7267c094 AL |
334 | uint16_t *new_blocks = g_malloc0(blocks_clusters * s->cluster_size); |
335 | uint64_t *new_table = g_malloc0(table_size * sizeof(uint64_t)); | |
92dcb59f KW |
336 | |
337 | assert(meta_offset >= (s->free_cluster_index * s->cluster_size)); | |
338 | ||
339 | /* Fill the new refcount table */ | |
f7d0fe02 | 340 | memcpy(new_table, s->refcount_table, |
92dcb59f KW |
341 | s->refcount_table_size * sizeof(uint64_t)); |
342 | new_table[refcount_table_index] = new_block; | |
343 | ||
344 | int i; | |
345 | for (i = 0; i < blocks_clusters; i++) { | |
346 | new_table[blocks_used + i] = meta_offset + (i * s->cluster_size); | |
347 | } | |
348 | ||
349 | /* Fill the refcount blocks */ | |
350 | uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t)); | |
351 | int block = 0; | |
352 | for (i = 0; i < table_clusters + blocks_clusters; i++) { | |
353 | new_blocks[block++] = cpu_to_be16(1); | |
354 | } | |
355 | ||
356 | /* Write refcount blocks to disk */ | |
66f82cee | 357 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS); |
8b3b7206 | 358 | ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks, |
92dcb59f | 359 | blocks_clusters * s->cluster_size); |
7267c094 | 360 | g_free(new_blocks); |
92dcb59f KW |
361 | if (ret < 0) { |
362 | goto fail_table; | |
363 | } | |
364 | ||
365 | /* Write refcount table to disk */ | |
366 | for(i = 0; i < table_size; i++) { | |
367 | cpu_to_be64s(&new_table[i]); | |
368 | } | |
369 | ||
66f82cee | 370 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE); |
8b3b7206 | 371 | ret = bdrv_pwrite_sync(bs->file, table_offset, new_table, |
92dcb59f KW |
372 | table_size * sizeof(uint64_t)); |
373 | if (ret < 0) { | |
374 | goto fail_table; | |
375 | } | |
376 | ||
377 | for(i = 0; i < table_size; i++) { | |
87267753 | 378 | be64_to_cpus(&new_table[i]); |
92dcb59f | 379 | } |
f7d0fe02 | 380 | |
92dcb59f KW |
381 | /* Hook up the new refcount table in the qcow2 header */ |
382 | uint8_t data[12]; | |
f7d0fe02 | 383 | cpu_to_be64w((uint64_t*)data, table_offset); |
92dcb59f | 384 | cpu_to_be32w((uint32_t*)(data + 8), table_clusters); |
66f82cee | 385 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE); |
8b3b7206 | 386 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset), |
92dcb59f KW |
387 | data, sizeof(data)); |
388 | if (ret < 0) { | |
389 | goto fail_table; | |
f2b7c8b3 KW |
390 | } |
391 | ||
92dcb59f KW |
392 | /* And switch it in memory */ |
393 | uint64_t old_table_offset = s->refcount_table_offset; | |
394 | uint64_t old_table_size = s->refcount_table_size; | |
395 | ||
7267c094 | 396 | g_free(s->refcount_table); |
f7d0fe02 | 397 | s->refcount_table = new_table; |
92dcb59f | 398 | s->refcount_table_size = table_size; |
f7d0fe02 KW |
399 | s->refcount_table_offset = table_offset; |
400 | ||
92dcb59f KW |
401 | /* Free old table. Remember, we must not change free_cluster_index */ |
402 | uint64_t old_free_cluster_index = s->free_cluster_index; | |
6cfcb9b8 KW |
403 | qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t), |
404 | QCOW2_DISCARD_OTHER); | |
92dcb59f | 405 | s->free_cluster_index = old_free_cluster_index; |
f7d0fe02 | 406 | |
29c1a730 | 407 | ret = load_refcount_block(bs, new_block, (void**) refcount_block); |
92dcb59f | 408 | if (ret < 0) { |
29c1a730 | 409 | return ret; |
f7d0fe02 KW |
410 | } |
411 | ||
2795ecf6 | 412 | return 0; |
f7d0fe02 | 413 | |
92dcb59f | 414 | fail_table: |
7267c094 | 415 | g_free(new_table); |
92dcb59f | 416 | fail_block: |
29c1a730 KW |
417 | if (*refcount_block != NULL) { |
418 | qcow2_cache_put(bs, s->refcount_block_cache, (void**) refcount_block); | |
3b88e52b | 419 | } |
29c1a730 | 420 | return ret; |
9923e05e KW |
421 | } |
422 | ||
0b919fae KW |
423 | void qcow2_process_discards(BlockDriverState *bs, int ret) |
424 | { | |
425 | BDRVQcowState *s = bs->opaque; | |
426 | Qcow2DiscardRegion *d, *next; | |
427 | ||
428 | QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) { | |
429 | QTAILQ_REMOVE(&s->discards, d, next); | |
430 | ||
431 | /* Discard is optional, ignore the return value */ | |
432 | if (ret >= 0) { | |
433 | bdrv_discard(bs->file, | |
434 | d->offset >> BDRV_SECTOR_BITS, | |
435 | d->bytes >> BDRV_SECTOR_BITS); | |
436 | } | |
437 | ||
438 | g_free(d); | |
439 | } | |
440 | } | |
441 | ||
442 | static void update_refcount_discard(BlockDriverState *bs, | |
443 | uint64_t offset, uint64_t length) | |
444 | { | |
445 | BDRVQcowState *s = bs->opaque; | |
446 | Qcow2DiscardRegion *d, *p, *next; | |
447 | ||
448 | QTAILQ_FOREACH(d, &s->discards, next) { | |
449 | uint64_t new_start = MIN(offset, d->offset); | |
450 | uint64_t new_end = MAX(offset + length, d->offset + d->bytes); | |
451 | ||
452 | if (new_end - new_start <= length + d->bytes) { | |
453 | /* There can't be any overlap, areas ending up here have no | |
454 | * references any more and therefore shouldn't get freed another | |
455 | * time. */ | |
456 | assert(d->bytes + length == new_end - new_start); | |
457 | d->offset = new_start; | |
458 | d->bytes = new_end - new_start; | |
459 | goto found; | |
460 | } | |
461 | } | |
462 | ||
463 | d = g_malloc(sizeof(*d)); | |
464 | *d = (Qcow2DiscardRegion) { | |
465 | .bs = bs, | |
466 | .offset = offset, | |
467 | .bytes = length, | |
468 | }; | |
469 | QTAILQ_INSERT_TAIL(&s->discards, d, next); | |
470 | ||
471 | found: | |
472 | /* Merge discard requests if they are adjacent now */ | |
473 | QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) { | |
474 | if (p == d | |
475 | || p->offset > d->offset + d->bytes | |
476 | || d->offset > p->offset + p->bytes) | |
477 | { | |
478 | continue; | |
479 | } | |
480 | ||
481 | /* Still no overlap possible */ | |
482 | assert(p->offset == d->offset + d->bytes | |
483 | || d->offset == p->offset + p->bytes); | |
484 | ||
485 | QTAILQ_REMOVE(&s->discards, p, next); | |
486 | d->offset = MIN(d->offset, p->offset); | |
487 | d->bytes += p->bytes; | |
488 | } | |
489 | } | |
490 | ||
f7d0fe02 | 491 | /* XXX: cache several refcount block clusters ? */ |
db3a964f | 492 | static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, |
6cfcb9b8 | 493 | int64_t offset, int64_t length, int addend, enum qcow2_discard_type type) |
f7d0fe02 KW |
494 | { |
495 | BDRVQcowState *s = bs->opaque; | |
496 | int64_t start, last, cluster_offset; | |
29c1a730 KW |
497 | uint16_t *refcount_block = NULL; |
498 | int64_t old_table_index = -1; | |
09508d13 | 499 | int ret; |
f7d0fe02 KW |
500 | |
501 | #ifdef DEBUG_ALLOC2 | |
35ee5e39 | 502 | fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n", |
f7d0fe02 KW |
503 | offset, length, addend); |
504 | #endif | |
7322afe7 | 505 | if (length < 0) { |
f7d0fe02 | 506 | return -EINVAL; |
7322afe7 KW |
507 | } else if (length == 0) { |
508 | return 0; | |
509 | } | |
510 | ||
29c1a730 KW |
511 | if (addend < 0) { |
512 | qcow2_cache_set_dependency(bs, s->refcount_block_cache, | |
513 | s->l2_table_cache); | |
514 | } | |
515 | ||
f7d0fe02 KW |
516 | start = offset & ~(s->cluster_size - 1); |
517 | last = (offset + length - 1) & ~(s->cluster_size - 1); | |
518 | for(cluster_offset = start; cluster_offset <= last; | |
519 | cluster_offset += s->cluster_size) | |
520 | { | |
521 | int block_index, refcount; | |
522 | int64_t cluster_index = cluster_offset >> s->cluster_bits; | |
29c1a730 KW |
523 | int64_t table_index = |
524 | cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT); | |
f7d0fe02 | 525 | |
29c1a730 KW |
526 | /* Load the refcount block and allocate it if needed */ |
527 | if (table_index != old_table_index) { | |
528 | if (refcount_block) { | |
529 | ret = qcow2_cache_put(bs, s->refcount_block_cache, | |
530 | (void**) &refcount_block); | |
531 | if (ret < 0) { | |
532 | goto fail; | |
533 | } | |
534 | } | |
9923e05e | 535 | |
29c1a730 | 536 | ret = alloc_refcount_block(bs, cluster_index, &refcount_block); |
ed0df867 | 537 | if (ret < 0) { |
29c1a730 | 538 | goto fail; |
f7d0fe02 | 539 | } |
f7d0fe02 | 540 | } |
29c1a730 | 541 | old_table_index = table_index; |
f7d0fe02 | 542 | |
29c1a730 | 543 | qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block); |
f7d0fe02 KW |
544 | |
545 | /* we can update the count and save it */ | |
546 | block_index = cluster_index & | |
547 | ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1); | |
f7d0fe02 | 548 | |
29c1a730 | 549 | refcount = be16_to_cpu(refcount_block[block_index]); |
f7d0fe02 | 550 | refcount += addend; |
09508d13 KW |
551 | if (refcount < 0 || refcount > 0xffff) { |
552 | ret = -EINVAL; | |
553 | goto fail; | |
554 | } | |
f7d0fe02 KW |
555 | if (refcount == 0 && cluster_index < s->free_cluster_index) { |
556 | s->free_cluster_index = cluster_index; | |
557 | } | |
29c1a730 | 558 | refcount_block[block_index] = cpu_to_be16(refcount); |
0b919fae | 559 | |
67af674e | 560 | if (refcount == 0 && s->discard_passthrough[type]) { |
0b919fae | 561 | update_refcount_discard(bs, cluster_offset, s->cluster_size); |
67af674e | 562 | } |
f7d0fe02 KW |
563 | } |
564 | ||
09508d13 KW |
565 | ret = 0; |
566 | fail: | |
0b919fae KW |
567 | if (!s->cache_discards) { |
568 | qcow2_process_discards(bs, ret); | |
569 | } | |
570 | ||
f7d0fe02 | 571 | /* Write last changed block to disk */ |
29c1a730 | 572 | if (refcount_block) { |
ed0df867 | 573 | int wret; |
29c1a730 KW |
574 | wret = qcow2_cache_put(bs, s->refcount_block_cache, |
575 | (void**) &refcount_block); | |
ed0df867 KW |
576 | if (wret < 0) { |
577 | return ret < 0 ? ret : wret; | |
f7d0fe02 KW |
578 | } |
579 | } | |
580 | ||
09508d13 KW |
581 | /* |
582 | * Try do undo any updates if an error is returned (This may succeed in | |
583 | * some cases like ENOSPC for allocating a new refcount block) | |
584 | */ | |
585 | if (ret < 0) { | |
586 | int dummy; | |
6cfcb9b8 KW |
587 | dummy = update_refcount(bs, offset, cluster_offset - offset, -addend, |
588 | QCOW2_DISCARD_NEVER); | |
83e3f76c | 589 | (void)dummy; |
09508d13 KW |
590 | } |
591 | ||
592 | return ret; | |
f7d0fe02 KW |
593 | } |
594 | ||
018faafd KW |
595 | /* |
596 | * Increases or decreases the refcount of a given cluster by one. | |
597 | * addend must be 1 or -1. | |
598 | * | |
599 | * If the return value is non-negative, it is the new refcount of the cluster. | |
600 | * If it is negative, it is -errno and indicates an error. | |
601 | */ | |
f7d0fe02 KW |
602 | static int update_cluster_refcount(BlockDriverState *bs, |
603 | int64_t cluster_index, | |
6cfcb9b8 KW |
604 | int addend, |
605 | enum qcow2_discard_type type) | |
f7d0fe02 KW |
606 | { |
607 | BDRVQcowState *s = bs->opaque; | |
608 | int ret; | |
609 | ||
6cfcb9b8 KW |
610 | ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend, |
611 | type); | |
f7d0fe02 KW |
612 | if (ret < 0) { |
613 | return ret; | |
614 | } | |
615 | ||
616 | return get_refcount(bs, cluster_index); | |
617 | } | |
618 | ||
619 | ||
620 | ||
621 | /*********************************************************/ | |
622 | /* cluster allocation functions */ | |
623 | ||
624 | ||
625 | ||
626 | /* return < 0 if error */ | |
627 | static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size) | |
628 | { | |
629 | BDRVQcowState *s = bs->opaque; | |
2eaa8f63 | 630 | int i, nb_clusters, refcount; |
f7d0fe02 KW |
631 | |
632 | nb_clusters = size_to_clusters(s, size); | |
633 | retry: | |
634 | for(i = 0; i < nb_clusters; i++) { | |
508e0893 | 635 | int64_t next_cluster_index = s->free_cluster_index++; |
2eaa8f63 KW |
636 | refcount = get_refcount(bs, next_cluster_index); |
637 | ||
638 | if (refcount < 0) { | |
639 | return refcount; | |
640 | } else if (refcount != 0) { | |
f7d0fe02 | 641 | goto retry; |
2eaa8f63 | 642 | } |
f7d0fe02 KW |
643 | } |
644 | #ifdef DEBUG_ALLOC2 | |
35ee5e39 | 645 | fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n", |
f7d0fe02 KW |
646 | size, |
647 | (s->free_cluster_index - nb_clusters) << s->cluster_bits); | |
648 | #endif | |
649 | return (s->free_cluster_index - nb_clusters) << s->cluster_bits; | |
650 | } | |
651 | ||
ed6ccf0f | 652 | int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size) |
f7d0fe02 KW |
653 | { |
654 | int64_t offset; | |
db3a964f | 655 | int ret; |
f7d0fe02 | 656 | |
66f82cee | 657 | BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC); |
f7d0fe02 | 658 | offset = alloc_clusters_noref(bs, size); |
2eaa8f63 KW |
659 | if (offset < 0) { |
660 | return offset; | |
661 | } | |
662 | ||
6cfcb9b8 | 663 | ret = update_refcount(bs, offset, size, 1, QCOW2_DISCARD_NEVER); |
db3a964f KW |
664 | if (ret < 0) { |
665 | return ret; | |
666 | } | |
1c4c2814 | 667 | |
f7d0fe02 KW |
668 | return offset; |
669 | } | |
670 | ||
256900b1 KW |
671 | int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset, |
672 | int nb_clusters) | |
673 | { | |
674 | BDRVQcowState *s = bs->opaque; | |
675 | uint64_t cluster_index; | |
f24423bd | 676 | uint64_t old_free_cluster_index; |
256900b1 KW |
677 | int i, refcount, ret; |
678 | ||
679 | /* Check how many clusters there are free */ | |
680 | cluster_index = offset >> s->cluster_bits; | |
681 | for(i = 0; i < nb_clusters; i++) { | |
682 | refcount = get_refcount(bs, cluster_index++); | |
683 | ||
684 | if (refcount < 0) { | |
685 | return refcount; | |
686 | } else if (refcount != 0) { | |
687 | break; | |
688 | } | |
689 | } | |
690 | ||
691 | /* And then allocate them */ | |
f24423bd KW |
692 | old_free_cluster_index = s->free_cluster_index; |
693 | s->free_cluster_index = cluster_index + i; | |
694 | ||
6cfcb9b8 KW |
695 | ret = update_refcount(bs, offset, i << s->cluster_bits, 1, |
696 | QCOW2_DISCARD_NEVER); | |
256900b1 KW |
697 | if (ret < 0) { |
698 | return ret; | |
699 | } | |
700 | ||
f24423bd KW |
701 | s->free_cluster_index = old_free_cluster_index; |
702 | ||
256900b1 KW |
703 | return i; |
704 | } | |
705 | ||
f7d0fe02 KW |
706 | /* only used to allocate compressed sectors. We try to allocate |
707 | contiguous sectors. size must be <= cluster_size */ | |
ed6ccf0f | 708 | int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size) |
f7d0fe02 KW |
709 | { |
710 | BDRVQcowState *s = bs->opaque; | |
711 | int64_t offset, cluster_offset; | |
712 | int free_in_cluster; | |
713 | ||
66f82cee | 714 | BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES); |
f7d0fe02 KW |
715 | assert(size > 0 && size <= s->cluster_size); |
716 | if (s->free_byte_offset == 0) { | |
206e6d85 SH |
717 | offset = qcow2_alloc_clusters(bs, s->cluster_size); |
718 | if (offset < 0) { | |
719 | return offset; | |
5d757b56 | 720 | } |
206e6d85 | 721 | s->free_byte_offset = offset; |
f7d0fe02 KW |
722 | } |
723 | redo: | |
724 | free_in_cluster = s->cluster_size - | |
725 | (s->free_byte_offset & (s->cluster_size - 1)); | |
726 | if (size <= free_in_cluster) { | |
727 | /* enough space in current cluster */ | |
728 | offset = s->free_byte_offset; | |
729 | s->free_byte_offset += size; | |
730 | free_in_cluster -= size; | |
731 | if (free_in_cluster == 0) | |
732 | s->free_byte_offset = 0; | |
733 | if ((offset & (s->cluster_size - 1)) != 0) | |
6cfcb9b8 KW |
734 | update_cluster_refcount(bs, offset >> s->cluster_bits, 1, |
735 | QCOW2_DISCARD_NEVER); | |
f7d0fe02 | 736 | } else { |
ed6ccf0f | 737 | offset = qcow2_alloc_clusters(bs, s->cluster_size); |
5d757b56 KW |
738 | if (offset < 0) { |
739 | return offset; | |
740 | } | |
f7d0fe02 KW |
741 | cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1); |
742 | if ((cluster_offset + s->cluster_size) == offset) { | |
743 | /* we are lucky: contiguous data */ | |
744 | offset = s->free_byte_offset; | |
6cfcb9b8 KW |
745 | update_cluster_refcount(bs, offset >> s->cluster_bits, 1, |
746 | QCOW2_DISCARD_NEVER); | |
f7d0fe02 KW |
747 | s->free_byte_offset += size; |
748 | } else { | |
749 | s->free_byte_offset = offset; | |
750 | goto redo; | |
751 | } | |
752 | } | |
29216ed1 | 753 | |
c1f5bafd SH |
754 | /* The cluster refcount was incremented, either by qcow2_alloc_clusters() |
755 | * or explicitly by update_cluster_refcount(). Refcount blocks must be | |
756 | * flushed before the caller's L2 table updates. | |
757 | */ | |
758 | qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache); | |
f7d0fe02 KW |
759 | return offset; |
760 | } | |
761 | ||
ed6ccf0f | 762 | void qcow2_free_clusters(BlockDriverState *bs, |
6cfcb9b8 KW |
763 | int64_t offset, int64_t size, |
764 | enum qcow2_discard_type type) | |
f7d0fe02 | 765 | { |
db3a964f KW |
766 | int ret; |
767 | ||
66f82cee | 768 | BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE); |
6cfcb9b8 | 769 | ret = update_refcount(bs, offset, size, -1, type); |
db3a964f KW |
770 | if (ret < 0) { |
771 | fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret)); | |
003fad6e | 772 | /* TODO Remember the clusters to free them later and avoid leaking */ |
db3a964f | 773 | } |
f7d0fe02 KW |
774 | } |
775 | ||
45aba42f | 776 | /* |
c7a4c37a KW |
777 | * Free a cluster using its L2 entry (handles clusters of all types, e.g. |
778 | * normal cluster, compressed cluster, etc.) | |
45aba42f | 779 | */ |
6cfcb9b8 KW |
780 | void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry, |
781 | int nb_clusters, enum qcow2_discard_type type) | |
45aba42f KW |
782 | { |
783 | BDRVQcowState *s = bs->opaque; | |
784 | ||
c7a4c37a KW |
785 | switch (qcow2_get_cluster_type(l2_entry)) { |
786 | case QCOW2_CLUSTER_COMPRESSED: | |
787 | { | |
788 | int nb_csectors; | |
789 | nb_csectors = ((l2_entry >> s->csize_shift) & | |
790 | s->csize_mask) + 1; | |
791 | qcow2_free_clusters(bs, | |
792 | (l2_entry & s->cluster_offset_mask) & ~511, | |
6cfcb9b8 | 793 | nb_csectors * 512, type); |
c7a4c37a KW |
794 | } |
795 | break; | |
796 | case QCOW2_CLUSTER_NORMAL: | |
797 | qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK, | |
6cfcb9b8 | 798 | nb_clusters << s->cluster_bits, type); |
c7a4c37a KW |
799 | break; |
800 | case QCOW2_CLUSTER_UNALLOCATED: | |
6377af48 | 801 | case QCOW2_CLUSTER_ZERO: |
c7a4c37a KW |
802 | break; |
803 | default: | |
804 | abort(); | |
45aba42f | 805 | } |
45aba42f KW |
806 | } |
807 | ||
f7d0fe02 KW |
808 | |
809 | ||
810 | /*********************************************************/ | |
811 | /* snapshots and image creation */ | |
812 | ||
813 | ||
814 | ||
f7d0fe02 | 815 | /* update the refcounts of snapshots and the copied flag */ |
ed6ccf0f KW |
816 | int qcow2_update_snapshot_refcount(BlockDriverState *bs, |
817 | int64_t l1_table_offset, int l1_size, int addend) | |
f7d0fe02 KW |
818 | { |
819 | BDRVQcowState *s = bs->opaque; | |
820 | uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated; | |
821 | int64_t old_offset, old_l2_offset; | |
93913dfd | 822 | int i, j, l1_modified = 0, nb_csectors, refcount; |
29c1a730 | 823 | int ret; |
f7d0fe02 KW |
824 | |
825 | l2_table = NULL; | |
826 | l1_table = NULL; | |
827 | l1_size2 = l1_size * sizeof(uint64_t); | |
43a0cac4 | 828 | |
0b919fae KW |
829 | s->cache_discards = true; |
830 | ||
43a0cac4 KW |
831 | /* WARNING: qcow2_snapshot_goto relies on this function not using the |
832 | * l1_table_offset when it is the current s->l1_table_offset! Be careful | |
833 | * when changing this! */ | |
f7d0fe02 | 834 | if (l1_table_offset != s->l1_table_offset) { |
6528499f | 835 | l1_table = g_malloc0(align_offset(l1_size2, 512)); |
f7d0fe02 | 836 | l1_allocated = 1; |
c2bc78b6 KW |
837 | |
838 | ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2); | |
839 | if (ret < 0) { | |
f7d0fe02 | 840 | goto fail; |
93913dfd KW |
841 | } |
842 | ||
f7d0fe02 KW |
843 | for(i = 0;i < l1_size; i++) |
844 | be64_to_cpus(&l1_table[i]); | |
845 | } else { | |
846 | assert(l1_size == s->l1_size); | |
847 | l1_table = s->l1_table; | |
848 | l1_allocated = 0; | |
849 | } | |
850 | ||
f7d0fe02 KW |
851 | for(i = 0; i < l1_size; i++) { |
852 | l2_offset = l1_table[i]; | |
853 | if (l2_offset) { | |
854 | old_l2_offset = l2_offset; | |
8e37f681 | 855 | l2_offset &= L1E_OFFSET_MASK; |
29c1a730 KW |
856 | |
857 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, | |
858 | (void**) &l2_table); | |
859 | if (ret < 0) { | |
f7d0fe02 | 860 | goto fail; |
29c1a730 KW |
861 | } |
862 | ||
f7d0fe02 KW |
863 | for(j = 0; j < s->l2_size; j++) { |
864 | offset = be64_to_cpu(l2_table[j]); | |
865 | if (offset != 0) { | |
866 | old_offset = offset; | |
867 | offset &= ~QCOW_OFLAG_COPIED; | |
868 | if (offset & QCOW_OFLAG_COMPRESSED) { | |
869 | nb_csectors = ((offset >> s->csize_shift) & | |
870 | s->csize_mask) + 1; | |
db3a964f KW |
871 | if (addend != 0) { |
872 | int ret; | |
873 | ret = update_refcount(bs, | |
874 | (offset & s->cluster_offset_mask) & ~511, | |
6cfcb9b8 KW |
875 | nb_csectors * 512, addend, |
876 | QCOW2_DISCARD_SNAPSHOT); | |
db3a964f KW |
877 | if (ret < 0) { |
878 | goto fail; | |
879 | } | |
880 | } | |
f7d0fe02 KW |
881 | /* compressed clusters are never modified */ |
882 | refcount = 2; | |
883 | } else { | |
8e37f681 | 884 | uint64_t cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits; |
f7d0fe02 | 885 | if (addend != 0) { |
6cfcb9b8 KW |
886 | refcount = update_cluster_refcount(bs, cluster_index, addend, |
887 | QCOW2_DISCARD_SNAPSHOT); | |
f7d0fe02 | 888 | } else { |
8e37f681 | 889 | refcount = get_refcount(bs, cluster_index); |
f7d0fe02 | 890 | } |
018faafd KW |
891 | |
892 | if (refcount < 0) { | |
c2bc78b6 | 893 | ret = refcount; |
018faafd KW |
894 | goto fail; |
895 | } | |
f7d0fe02 KW |
896 | } |
897 | ||
898 | if (refcount == 1) { | |
899 | offset |= QCOW_OFLAG_COPIED; | |
900 | } | |
901 | if (offset != old_offset) { | |
29c1a730 KW |
902 | if (addend > 0) { |
903 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
904 | s->refcount_block_cache); | |
905 | } | |
f7d0fe02 | 906 | l2_table[j] = cpu_to_be64(offset); |
29c1a730 | 907 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
f7d0fe02 KW |
908 | } |
909 | } | |
910 | } | |
29c1a730 KW |
911 | |
912 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
913 | if (ret < 0) { | |
914 | goto fail; | |
f7d0fe02 KW |
915 | } |
916 | ||
29c1a730 | 917 | |
f7d0fe02 | 918 | if (addend != 0) { |
6cfcb9b8 KW |
919 | refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend, |
920 | QCOW2_DISCARD_SNAPSHOT); | |
f7d0fe02 KW |
921 | } else { |
922 | refcount = get_refcount(bs, l2_offset >> s->cluster_bits); | |
923 | } | |
018faafd | 924 | if (refcount < 0) { |
c2bc78b6 | 925 | ret = refcount; |
018faafd KW |
926 | goto fail; |
927 | } else if (refcount == 1) { | |
f7d0fe02 KW |
928 | l2_offset |= QCOW_OFLAG_COPIED; |
929 | } | |
930 | if (l2_offset != old_l2_offset) { | |
931 | l1_table[i] = l2_offset; | |
932 | l1_modified = 1; | |
933 | } | |
934 | } | |
935 | } | |
93913dfd | 936 | |
2154f24e | 937 | ret = bdrv_flush(bs); |
93913dfd KW |
938 | fail: |
939 | if (l2_table) { | |
940 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
941 | } | |
942 | ||
0b919fae KW |
943 | s->cache_discards = false; |
944 | qcow2_process_discards(bs, ret); | |
945 | ||
43a0cac4 | 946 | /* Update L1 only if it isn't deleted anyway (addend = -1) */ |
c2b6ff51 KW |
947 | if (ret == 0 && addend >= 0 && l1_modified) { |
948 | for (i = 0; i < l1_size; i++) { | |
f7d0fe02 | 949 | cpu_to_be64s(&l1_table[i]); |
c2b6ff51 KW |
950 | } |
951 | ||
952 | ret = bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table, l1_size2); | |
953 | ||
954 | for (i = 0; i < l1_size; i++) { | |
f7d0fe02 | 955 | be64_to_cpus(&l1_table[i]); |
c2b6ff51 | 956 | } |
f7d0fe02 KW |
957 | } |
958 | if (l1_allocated) | |
7267c094 | 959 | g_free(l1_table); |
93913dfd | 960 | return ret; |
f7d0fe02 KW |
961 | } |
962 | ||
963 | ||
964 | ||
965 | ||
966 | /*********************************************************/ | |
967 | /* refcount checking functions */ | |
968 | ||
969 | ||
970 | ||
971 | /* | |
972 | * Increases the refcount for a range of clusters in a given refcount table. | |
973 | * This is used to construct a temporary refcount table out of L1 and L2 tables | |
974 | * which can be compared the the refcount table saved in the image. | |
975 | * | |
9ac228e0 | 976 | * Modifies the number of errors in res. |
f7d0fe02 | 977 | */ |
9ac228e0 KW |
978 | static void inc_refcounts(BlockDriverState *bs, |
979 | BdrvCheckResult *res, | |
f7d0fe02 KW |
980 | uint16_t *refcount_table, |
981 | int refcount_table_size, | |
982 | int64_t offset, int64_t size) | |
983 | { | |
984 | BDRVQcowState *s = bs->opaque; | |
985 | int64_t start, last, cluster_offset; | |
986 | int k; | |
f7d0fe02 KW |
987 | |
988 | if (size <= 0) | |
9ac228e0 | 989 | return; |
f7d0fe02 KW |
990 | |
991 | start = offset & ~(s->cluster_size - 1); | |
992 | last = (offset + size - 1) & ~(s->cluster_size - 1); | |
993 | for(cluster_offset = start; cluster_offset <= last; | |
994 | cluster_offset += s->cluster_size) { | |
995 | k = cluster_offset >> s->cluster_bits; | |
9ac228e0 | 996 | if (k < 0) { |
f7d0fe02 KW |
997 | fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n", |
998 | cluster_offset); | |
9ac228e0 KW |
999 | res->corruptions++; |
1000 | } else if (k >= refcount_table_size) { | |
1001 | fprintf(stderr, "Warning: cluster offset=0x%" PRIx64 " is after " | |
1002 | "the end of the image file, can't properly check refcounts.\n", | |
1003 | cluster_offset); | |
1004 | res->check_errors++; | |
f7d0fe02 KW |
1005 | } else { |
1006 | if (++refcount_table[k] == 0) { | |
1007 | fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64 | |
1008 | "\n", cluster_offset); | |
9ac228e0 | 1009 | res->corruptions++; |
f7d0fe02 KW |
1010 | } |
1011 | } | |
1012 | } | |
f7d0fe02 KW |
1013 | } |
1014 | ||
801f7044 SH |
1015 | /* Flags for check_refcounts_l1() and check_refcounts_l2() */ |
1016 | enum { | |
1017 | CHECK_OFLAG_COPIED = 0x1, /* check QCOW_OFLAG_COPIED matches refcount */ | |
fba31bae | 1018 | CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */ |
801f7044 SH |
1019 | }; |
1020 | ||
f7d0fe02 KW |
1021 | /* |
1022 | * Increases the refcount in the given refcount table for the all clusters | |
1023 | * referenced in the L2 table. While doing so, performs some checks on L2 | |
1024 | * entries. | |
1025 | * | |
1026 | * Returns the number of errors found by the checks or -errno if an internal | |
1027 | * error occurred. | |
1028 | */ | |
9ac228e0 | 1029 | static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res, |
f7d0fe02 | 1030 | uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset, |
801f7044 | 1031 | int flags) |
f7d0fe02 KW |
1032 | { |
1033 | BDRVQcowState *s = bs->opaque; | |
afdf0abe | 1034 | uint64_t *l2_table, l2_entry; |
fba31bae | 1035 | uint64_t next_contiguous_offset = 0; |
f7d0fe02 | 1036 | int i, l2_size, nb_csectors, refcount; |
f7d0fe02 KW |
1037 | |
1038 | /* Read L2 table from disk */ | |
1039 | l2_size = s->l2_size * sizeof(uint64_t); | |
7267c094 | 1040 | l2_table = g_malloc(l2_size); |
f7d0fe02 | 1041 | |
66f82cee | 1042 | if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size) |
f7d0fe02 KW |
1043 | goto fail; |
1044 | ||
1045 | /* Do the actual checks */ | |
1046 | for(i = 0; i < s->l2_size; i++) { | |
afdf0abe KW |
1047 | l2_entry = be64_to_cpu(l2_table[i]); |
1048 | ||
1049 | switch (qcow2_get_cluster_type(l2_entry)) { | |
1050 | case QCOW2_CLUSTER_COMPRESSED: | |
1051 | /* Compressed clusters don't have QCOW_OFLAG_COPIED */ | |
1052 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
1053 | fprintf(stderr, "ERROR: cluster %" PRId64 ": " | |
1054 | "copied flag must never be set for compressed " | |
1055 | "clusters\n", l2_entry >> s->cluster_bits); | |
1056 | l2_entry &= ~QCOW_OFLAG_COPIED; | |
1057 | res->corruptions++; | |
1058 | } | |
f7d0fe02 | 1059 | |
afdf0abe KW |
1060 | /* Mark cluster as used */ |
1061 | nb_csectors = ((l2_entry >> s->csize_shift) & | |
1062 | s->csize_mask) + 1; | |
1063 | l2_entry &= s->cluster_offset_mask; | |
1064 | inc_refcounts(bs, res, refcount_table, refcount_table_size, | |
1065 | l2_entry & ~511, nb_csectors * 512); | |
fba31bae SH |
1066 | |
1067 | if (flags & CHECK_FRAG_INFO) { | |
1068 | res->bfi.allocated_clusters++; | |
4db35162 | 1069 | res->bfi.compressed_clusters++; |
fba31bae SH |
1070 | |
1071 | /* Compressed clusters are fragmented by nature. Since they | |
1072 | * take up sub-sector space but we only have sector granularity | |
1073 | * I/O we need to re-read the same sectors even for adjacent | |
1074 | * compressed clusters. | |
1075 | */ | |
1076 | res->bfi.fragmented_clusters++; | |
1077 | } | |
afdf0abe | 1078 | break; |
f7d0fe02 | 1079 | |
6377af48 KW |
1080 | case QCOW2_CLUSTER_ZERO: |
1081 | if ((l2_entry & L2E_OFFSET_MASK) == 0) { | |
1082 | break; | |
1083 | } | |
1084 | /* fall through */ | |
1085 | ||
afdf0abe KW |
1086 | case QCOW2_CLUSTER_NORMAL: |
1087 | { | |
1088 | /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */ | |
1089 | uint64_t offset = l2_entry & L2E_OFFSET_MASK; | |
f7d0fe02 | 1090 | |
801f7044 | 1091 | if (flags & CHECK_OFLAG_COPIED) { |
afdf0abe KW |
1092 | refcount = get_refcount(bs, offset >> s->cluster_bits); |
1093 | if (refcount < 0) { | |
1094 | fprintf(stderr, "Can't get refcount for offset %" | |
1095 | PRIx64 ": %s\n", l2_entry, strerror(-refcount)); | |
1096 | goto fail; | |
1097 | } | |
1098 | if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) { | |
1099 | fprintf(stderr, "ERROR OFLAG_COPIED: offset=%" | |
1100 | PRIx64 " refcount=%d\n", l2_entry, refcount); | |
9ac228e0 | 1101 | res->corruptions++; |
f7d0fe02 KW |
1102 | } |
1103 | } | |
afdf0abe | 1104 | |
fba31bae SH |
1105 | if (flags & CHECK_FRAG_INFO) { |
1106 | res->bfi.allocated_clusters++; | |
1107 | if (next_contiguous_offset && | |
1108 | offset != next_contiguous_offset) { | |
1109 | res->bfi.fragmented_clusters++; | |
1110 | } | |
1111 | next_contiguous_offset = offset + s->cluster_size; | |
1112 | } | |
1113 | ||
afdf0abe KW |
1114 | /* Mark cluster as used */ |
1115 | inc_refcounts(bs, res, refcount_table,refcount_table_size, | |
1116 | offset, s->cluster_size); | |
1117 | ||
1118 | /* Correct offsets are cluster aligned */ | |
1119 | if (offset & (s->cluster_size - 1)) { | |
1120 | fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not " | |
1121 | "properly aligned; L2 entry corrupted.\n", offset); | |
1122 | res->corruptions++; | |
1123 | } | |
1124 | break; | |
1125 | } | |
1126 | ||
1127 | case QCOW2_CLUSTER_UNALLOCATED: | |
1128 | break; | |
1129 | ||
1130 | default: | |
1131 | abort(); | |
f7d0fe02 KW |
1132 | } |
1133 | } | |
1134 | ||
7267c094 | 1135 | g_free(l2_table); |
9ac228e0 | 1136 | return 0; |
f7d0fe02 KW |
1137 | |
1138 | fail: | |
9ac228e0 | 1139 | fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n"); |
7267c094 | 1140 | g_free(l2_table); |
f7d0fe02 KW |
1141 | return -EIO; |
1142 | } | |
1143 | ||
1144 | /* | |
1145 | * Increases the refcount for the L1 table, its L2 tables and all referenced | |
1146 | * clusters in the given refcount table. While doing so, performs some checks | |
1147 | * on L1 and L2 entries. | |
1148 | * | |
1149 | * Returns the number of errors found by the checks or -errno if an internal | |
1150 | * error occurred. | |
1151 | */ | |
1152 | static int check_refcounts_l1(BlockDriverState *bs, | |
9ac228e0 | 1153 | BdrvCheckResult *res, |
f7d0fe02 KW |
1154 | uint16_t *refcount_table, |
1155 | int refcount_table_size, | |
1156 | int64_t l1_table_offset, int l1_size, | |
801f7044 | 1157 | int flags) |
f7d0fe02 KW |
1158 | { |
1159 | BDRVQcowState *s = bs->opaque; | |
1160 | uint64_t *l1_table, l2_offset, l1_size2; | |
1161 | int i, refcount, ret; | |
f7d0fe02 KW |
1162 | |
1163 | l1_size2 = l1_size * sizeof(uint64_t); | |
1164 | ||
1165 | /* Mark L1 table as used */ | |
9ac228e0 KW |
1166 | inc_refcounts(bs, res, refcount_table, refcount_table_size, |
1167 | l1_table_offset, l1_size2); | |
f7d0fe02 KW |
1168 | |
1169 | /* Read L1 table entries from disk */ | |
702ef63f KW |
1170 | if (l1_size2 == 0) { |
1171 | l1_table = NULL; | |
1172 | } else { | |
7267c094 | 1173 | l1_table = g_malloc(l1_size2); |
66f82cee | 1174 | if (bdrv_pread(bs->file, l1_table_offset, |
702ef63f KW |
1175 | l1_table, l1_size2) != l1_size2) |
1176 | goto fail; | |
1177 | for(i = 0;i < l1_size; i++) | |
1178 | be64_to_cpus(&l1_table[i]); | |
1179 | } | |
f7d0fe02 KW |
1180 | |
1181 | /* Do the actual checks */ | |
1182 | for(i = 0; i < l1_size; i++) { | |
1183 | l2_offset = l1_table[i]; | |
1184 | if (l2_offset) { | |
1185 | /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */ | |
801f7044 | 1186 | if (flags & CHECK_OFLAG_COPIED) { |
f7d0fe02 KW |
1187 | refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) |
1188 | >> s->cluster_bits); | |
018faafd KW |
1189 | if (refcount < 0) { |
1190 | fprintf(stderr, "Can't get refcount for l2_offset %" | |
1191 | PRIx64 ": %s\n", l2_offset, strerror(-refcount)); | |
9ac228e0 | 1192 | goto fail; |
018faafd | 1193 | } |
f7d0fe02 KW |
1194 | if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) { |
1195 | fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64 | |
1196 | " refcount=%d\n", l2_offset, refcount); | |
9ac228e0 | 1197 | res->corruptions++; |
f7d0fe02 KW |
1198 | } |
1199 | } | |
1200 | ||
1201 | /* Mark L2 table as used */ | |
afdf0abe | 1202 | l2_offset &= L1E_OFFSET_MASK; |
9ac228e0 KW |
1203 | inc_refcounts(bs, res, refcount_table, refcount_table_size, |
1204 | l2_offset, s->cluster_size); | |
f7d0fe02 KW |
1205 | |
1206 | /* L2 tables are cluster aligned */ | |
1207 | if (l2_offset & (s->cluster_size - 1)) { | |
1208 | fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not " | |
1209 | "cluster aligned; L1 entry corrupted\n", l2_offset); | |
9ac228e0 | 1210 | res->corruptions++; |
f7d0fe02 KW |
1211 | } |
1212 | ||
1213 | /* Process and check L2 entries */ | |
9ac228e0 | 1214 | ret = check_refcounts_l2(bs, res, refcount_table, |
801f7044 | 1215 | refcount_table_size, l2_offset, flags); |
f7d0fe02 KW |
1216 | if (ret < 0) { |
1217 | goto fail; | |
1218 | } | |
f7d0fe02 KW |
1219 | } |
1220 | } | |
7267c094 | 1221 | g_free(l1_table); |
9ac228e0 | 1222 | return 0; |
f7d0fe02 KW |
1223 | |
1224 | fail: | |
1225 | fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n"); | |
9ac228e0 | 1226 | res->check_errors++; |
7267c094 | 1227 | g_free(l1_table); |
f7d0fe02 KW |
1228 | return -EIO; |
1229 | } | |
1230 | ||
1231 | /* | |
1232 | * Checks an image for refcount consistency. | |
1233 | * | |
1234 | * Returns 0 if no errors are found, the number of errors in case the image is | |
a1c7273b | 1235 | * detected as corrupted, and -errno when an internal error occurred. |
f7d0fe02 | 1236 | */ |
166acf54 KW |
1237 | int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res, |
1238 | BdrvCheckMode fix) | |
f7d0fe02 KW |
1239 | { |
1240 | BDRVQcowState *s = bs->opaque; | |
c6bb9ad1 | 1241 | int64_t size, i, highest_cluster; |
166acf54 | 1242 | int nb_clusters, refcount1, refcount2; |
f7d0fe02 KW |
1243 | QCowSnapshot *sn; |
1244 | uint16_t *refcount_table; | |
9ac228e0 | 1245 | int ret; |
f7d0fe02 | 1246 | |
66f82cee | 1247 | size = bdrv_getlength(bs->file); |
f7d0fe02 | 1248 | nb_clusters = size_to_clusters(s, size); |
7267c094 | 1249 | refcount_table = g_malloc0(nb_clusters * sizeof(uint16_t)); |
f7d0fe02 | 1250 | |
c349ca4b KW |
1251 | res->bfi.total_clusters = |
1252 | size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE); | |
1253 | ||
f7d0fe02 | 1254 | /* header */ |
9ac228e0 KW |
1255 | inc_refcounts(bs, res, refcount_table, nb_clusters, |
1256 | 0, s->cluster_size); | |
f7d0fe02 KW |
1257 | |
1258 | /* current L1 table */ | |
9ac228e0 | 1259 | ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, |
801f7044 | 1260 | s->l1_table_offset, s->l1_size, |
fba31bae | 1261 | CHECK_OFLAG_COPIED | CHECK_FRAG_INFO); |
f7d0fe02 | 1262 | if (ret < 0) { |
80fa3341 | 1263 | goto fail; |
f7d0fe02 | 1264 | } |
f7d0fe02 KW |
1265 | |
1266 | /* snapshots */ | |
1267 | for(i = 0; i < s->nb_snapshots; i++) { | |
1268 | sn = s->snapshots + i; | |
9ac228e0 KW |
1269 | ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, |
1270 | sn->l1_table_offset, sn->l1_size, 0); | |
1271 | if (ret < 0) { | |
80fa3341 | 1272 | goto fail; |
9ac228e0 | 1273 | } |
f7d0fe02 | 1274 | } |
9ac228e0 KW |
1275 | inc_refcounts(bs, res, refcount_table, nb_clusters, |
1276 | s->snapshots_offset, s->snapshots_size); | |
f7d0fe02 KW |
1277 | |
1278 | /* refcount data */ | |
9ac228e0 KW |
1279 | inc_refcounts(bs, res, refcount_table, nb_clusters, |
1280 | s->refcount_table_offset, | |
1281 | s->refcount_table_size * sizeof(uint64_t)); | |
1282 | ||
f7d0fe02 | 1283 | for(i = 0; i < s->refcount_table_size; i++) { |
6882c8fa | 1284 | uint64_t offset, cluster; |
f7d0fe02 | 1285 | offset = s->refcount_table[i]; |
6882c8fa | 1286 | cluster = offset >> s->cluster_bits; |
746c3cb5 KW |
1287 | |
1288 | /* Refcount blocks are cluster aligned */ | |
1289 | if (offset & (s->cluster_size - 1)) { | |
166acf54 | 1290 | fprintf(stderr, "ERROR refcount block %" PRId64 " is not " |
746c3cb5 | 1291 | "cluster aligned; refcount table entry corrupted\n", i); |
9ac228e0 | 1292 | res->corruptions++; |
6882c8fa KW |
1293 | continue; |
1294 | } | |
1295 | ||
1296 | if (cluster >= nb_clusters) { | |
166acf54 KW |
1297 | fprintf(stderr, "ERROR refcount block %" PRId64 |
1298 | " is outside image\n", i); | |
9ac228e0 | 1299 | res->corruptions++; |
6882c8fa | 1300 | continue; |
746c3cb5 KW |
1301 | } |
1302 | ||
f7d0fe02 | 1303 | if (offset != 0) { |
9ac228e0 KW |
1304 | inc_refcounts(bs, res, refcount_table, nb_clusters, |
1305 | offset, s->cluster_size); | |
6882c8fa | 1306 | if (refcount_table[cluster] != 1) { |
166acf54 KW |
1307 | fprintf(stderr, "ERROR refcount block %" PRId64 |
1308 | " refcount=%d\n", | |
6882c8fa | 1309 | i, refcount_table[cluster]); |
9ac228e0 | 1310 | res->corruptions++; |
746c3cb5 | 1311 | } |
f7d0fe02 KW |
1312 | } |
1313 | } | |
1314 | ||
1315 | /* compare ref counts */ | |
c6bb9ad1 | 1316 | for (i = 0, highest_cluster = 0; i < nb_clusters; i++) { |
f7d0fe02 | 1317 | refcount1 = get_refcount(bs, i); |
018faafd | 1318 | if (refcount1 < 0) { |
166acf54 | 1319 | fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n", |
018faafd | 1320 | i, strerror(-refcount1)); |
9ac228e0 | 1321 | res->check_errors++; |
f74550fd | 1322 | continue; |
018faafd KW |
1323 | } |
1324 | ||
f7d0fe02 | 1325 | refcount2 = refcount_table[i]; |
c6bb9ad1 FS |
1326 | |
1327 | if (refcount1 > 0 || refcount2 > 0) { | |
1328 | highest_cluster = i; | |
1329 | } | |
1330 | ||
f7d0fe02 | 1331 | if (refcount1 != refcount2) { |
166acf54 KW |
1332 | |
1333 | /* Check if we're allowed to fix the mismatch */ | |
1334 | int *num_fixed = NULL; | |
1335 | if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) { | |
1336 | num_fixed = &res->leaks_fixed; | |
1337 | } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) { | |
1338 | num_fixed = &res->corruptions_fixed; | |
1339 | } | |
1340 | ||
1341 | fprintf(stderr, "%s cluster %" PRId64 " refcount=%d reference=%d\n", | |
1342 | num_fixed != NULL ? "Repairing" : | |
1343 | refcount1 < refcount2 ? "ERROR" : | |
1344 | "Leaked", | |
f7d0fe02 | 1345 | i, refcount1, refcount2); |
166acf54 KW |
1346 | |
1347 | if (num_fixed) { | |
1348 | ret = update_refcount(bs, i << s->cluster_bits, 1, | |
6cfcb9b8 KW |
1349 | refcount2 - refcount1, |
1350 | QCOW2_DISCARD_ALWAYS); | |
166acf54 KW |
1351 | if (ret >= 0) { |
1352 | (*num_fixed)++; | |
1353 | continue; | |
1354 | } | |
1355 | } | |
1356 | ||
1357 | /* And if we couldn't, print an error */ | |
9ac228e0 KW |
1358 | if (refcount1 < refcount2) { |
1359 | res->corruptions++; | |
1360 | } else { | |
1361 | res->leaks++; | |
1362 | } | |
f7d0fe02 KW |
1363 | } |
1364 | } | |
1365 | ||
c6bb9ad1 | 1366 | res->image_end_offset = (highest_cluster + 1) * s->cluster_size; |
80fa3341 KW |
1367 | ret = 0; |
1368 | ||
1369 | fail: | |
7267c094 | 1370 | g_free(refcount_table); |
f7d0fe02 | 1371 | |
80fa3341 | 1372 | return ret; |
f7d0fe02 KW |
1373 | } |
1374 |