]>
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 | ||
80c71a24 | 25 | #include "qemu/osdep.h" |
da34e65c | 26 | #include "qapi/error.h" |
f7d0fe02 | 27 | #include "qemu-common.h" |
737e150e | 28 | #include "block/block_int.h" |
f7d0fe02 | 29 | #include "block/qcow2.h" |
a40f1c2a | 30 | #include "qemu/range.h" |
58369e22 | 31 | #include "qemu/bswap.h" |
46b732cd | 32 | #include "qemu/cutils.h" |
f7d0fe02 | 33 | |
bb572aef | 34 | static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size); |
92dcb59f | 35 | static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, |
0e06528e | 36 | int64_t offset, int64_t length, uint64_t addend, |
2aabe7c7 | 37 | bool decrease, enum qcow2_discard_type type); |
f7d0fe02 | 38 | |
59c0cb78 HR |
39 | static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index); |
40 | static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index); | |
41 | static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index); | |
42 | static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index); | |
7453c96b | 43 | static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index); |
59c0cb78 HR |
44 | static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index); |
45 | static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index); | |
7453c96b | 46 | |
59c0cb78 HR |
47 | static void set_refcount_ro0(void *refcount_array, uint64_t index, |
48 | uint64_t value); | |
49 | static void set_refcount_ro1(void *refcount_array, uint64_t index, | |
50 | uint64_t value); | |
51 | static void set_refcount_ro2(void *refcount_array, uint64_t index, | |
52 | uint64_t value); | |
53 | static void set_refcount_ro3(void *refcount_array, uint64_t index, | |
54 | uint64_t value); | |
7453c96b HR |
55 | static void set_refcount_ro4(void *refcount_array, uint64_t index, |
56 | uint64_t value); | |
59c0cb78 HR |
57 | static void set_refcount_ro5(void *refcount_array, uint64_t index, |
58 | uint64_t value); | |
59 | static void set_refcount_ro6(void *refcount_array, uint64_t index, | |
60 | uint64_t value); | |
61 | ||
62 | ||
63 | static Qcow2GetRefcountFunc *const get_refcount_funcs[] = { | |
64 | &get_refcount_ro0, | |
65 | &get_refcount_ro1, | |
66 | &get_refcount_ro2, | |
67 | &get_refcount_ro3, | |
68 | &get_refcount_ro4, | |
69 | &get_refcount_ro5, | |
70 | &get_refcount_ro6 | |
71 | }; | |
72 | ||
73 | static Qcow2SetRefcountFunc *const set_refcount_funcs[] = { | |
74 | &set_refcount_ro0, | |
75 | &set_refcount_ro1, | |
76 | &set_refcount_ro2, | |
77 | &set_refcount_ro3, | |
78 | &set_refcount_ro4, | |
79 | &set_refcount_ro5, | |
80 | &set_refcount_ro6 | |
81 | }; | |
7453c96b | 82 | |
3b88e52b | 83 | |
f7d0fe02 KW |
84 | /*********************************************************/ |
85 | /* refcount handling */ | |
86 | ||
7061a078 AG |
87 | static void update_max_refcount_table_index(BDRVQcow2State *s) |
88 | { | |
89 | unsigned i = s->refcount_table_size - 1; | |
90 | while (i > 0 && (s->refcount_table[i] & REFT_OFFSET_MASK) == 0) { | |
91 | i--; | |
92 | } | |
93 | /* Set s->max_refcount_table_index to the index of the last used entry */ | |
94 | s->max_refcount_table_index = i; | |
95 | } | |
96 | ||
ed6ccf0f | 97 | int qcow2_refcount_init(BlockDriverState *bs) |
f7d0fe02 | 98 | { |
ff99129a | 99 | BDRVQcow2State *s = bs->opaque; |
5dab2fad KW |
100 | unsigned int refcount_table_size2, i; |
101 | int ret; | |
f7d0fe02 | 102 | |
59c0cb78 HR |
103 | assert(s->refcount_order >= 0 && s->refcount_order <= 6); |
104 | ||
105 | s->get_refcount = get_refcount_funcs[s->refcount_order]; | |
106 | s->set_refcount = set_refcount_funcs[s->refcount_order]; | |
7453c96b | 107 | |
5dab2fad | 108 | assert(s->refcount_table_size <= INT_MAX / sizeof(uint64_t)); |
f7d0fe02 | 109 | refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t); |
de82815d KW |
110 | s->refcount_table = g_try_malloc(refcount_table_size2); |
111 | ||
f7d0fe02 | 112 | if (s->refcount_table_size > 0) { |
de82815d | 113 | if (s->refcount_table == NULL) { |
8fcffa98 | 114 | ret = -ENOMEM; |
de82815d KW |
115 | goto fail; |
116 | } | |
66f82cee | 117 | BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD); |
cf2ab8fc | 118 | ret = bdrv_pread(bs->file, s->refcount_table_offset, |
f7d0fe02 | 119 | s->refcount_table, refcount_table_size2); |
8fcffa98 | 120 | if (ret < 0) { |
f7d0fe02 | 121 | goto fail; |
8fcffa98 | 122 | } |
f7d0fe02 KW |
123 | for(i = 0; i < s->refcount_table_size; i++) |
124 | be64_to_cpus(&s->refcount_table[i]); | |
7061a078 | 125 | update_max_refcount_table_index(s); |
f7d0fe02 KW |
126 | } |
127 | return 0; | |
128 | fail: | |
8fcffa98 | 129 | return ret; |
f7d0fe02 KW |
130 | } |
131 | ||
ed6ccf0f | 132 | void qcow2_refcount_close(BlockDriverState *bs) |
f7d0fe02 | 133 | { |
ff99129a | 134 | BDRVQcow2State *s = bs->opaque; |
7267c094 | 135 | g_free(s->refcount_table); |
f7d0fe02 KW |
136 | } |
137 | ||
138 | ||
59c0cb78 HR |
139 | static uint64_t get_refcount_ro0(const void *refcount_array, uint64_t index) |
140 | { | |
141 | return (((const uint8_t *)refcount_array)[index / 8] >> (index % 8)) & 0x1; | |
142 | } | |
143 | ||
144 | static void set_refcount_ro0(void *refcount_array, uint64_t index, | |
145 | uint64_t value) | |
146 | { | |
147 | assert(!(value >> 1)); | |
148 | ((uint8_t *)refcount_array)[index / 8] &= ~(0x1 << (index % 8)); | |
149 | ((uint8_t *)refcount_array)[index / 8] |= value << (index % 8); | |
150 | } | |
151 | ||
152 | static uint64_t get_refcount_ro1(const void *refcount_array, uint64_t index) | |
153 | { | |
154 | return (((const uint8_t *)refcount_array)[index / 4] >> (2 * (index % 4))) | |
155 | & 0x3; | |
156 | } | |
157 | ||
158 | static void set_refcount_ro1(void *refcount_array, uint64_t index, | |
159 | uint64_t value) | |
160 | { | |
161 | assert(!(value >> 2)); | |
162 | ((uint8_t *)refcount_array)[index / 4] &= ~(0x3 << (2 * (index % 4))); | |
163 | ((uint8_t *)refcount_array)[index / 4] |= value << (2 * (index % 4)); | |
164 | } | |
165 | ||
166 | static uint64_t get_refcount_ro2(const void *refcount_array, uint64_t index) | |
167 | { | |
168 | return (((const uint8_t *)refcount_array)[index / 2] >> (4 * (index % 2))) | |
169 | & 0xf; | |
170 | } | |
171 | ||
172 | static void set_refcount_ro2(void *refcount_array, uint64_t index, | |
173 | uint64_t value) | |
174 | { | |
175 | assert(!(value >> 4)); | |
176 | ((uint8_t *)refcount_array)[index / 2] &= ~(0xf << (4 * (index % 2))); | |
177 | ((uint8_t *)refcount_array)[index / 2] |= value << (4 * (index % 2)); | |
178 | } | |
179 | ||
180 | static uint64_t get_refcount_ro3(const void *refcount_array, uint64_t index) | |
181 | { | |
182 | return ((const uint8_t *)refcount_array)[index]; | |
183 | } | |
184 | ||
185 | static void set_refcount_ro3(void *refcount_array, uint64_t index, | |
186 | uint64_t value) | |
187 | { | |
188 | assert(!(value >> 8)); | |
189 | ((uint8_t *)refcount_array)[index] = value; | |
190 | } | |
191 | ||
7453c96b HR |
192 | static uint64_t get_refcount_ro4(const void *refcount_array, uint64_t index) |
193 | { | |
194 | return be16_to_cpu(((const uint16_t *)refcount_array)[index]); | |
195 | } | |
196 | ||
197 | static void set_refcount_ro4(void *refcount_array, uint64_t index, | |
198 | uint64_t value) | |
199 | { | |
200 | assert(!(value >> 16)); | |
201 | ((uint16_t *)refcount_array)[index] = cpu_to_be16(value); | |
202 | } | |
203 | ||
59c0cb78 HR |
204 | static uint64_t get_refcount_ro5(const void *refcount_array, uint64_t index) |
205 | { | |
206 | return be32_to_cpu(((const uint32_t *)refcount_array)[index]); | |
207 | } | |
208 | ||
209 | static void set_refcount_ro5(void *refcount_array, uint64_t index, | |
210 | uint64_t value) | |
211 | { | |
212 | assert(!(value >> 32)); | |
213 | ((uint32_t *)refcount_array)[index] = cpu_to_be32(value); | |
214 | } | |
215 | ||
216 | static uint64_t get_refcount_ro6(const void *refcount_array, uint64_t index) | |
217 | { | |
218 | return be64_to_cpu(((const uint64_t *)refcount_array)[index]); | |
219 | } | |
220 | ||
221 | static void set_refcount_ro6(void *refcount_array, uint64_t index, | |
222 | uint64_t value) | |
223 | { | |
224 | ((uint64_t *)refcount_array)[index] = cpu_to_be64(value); | |
225 | } | |
226 | ||
7453c96b | 227 | |
f7d0fe02 | 228 | static int load_refcount_block(BlockDriverState *bs, |
29c1a730 KW |
229 | int64_t refcount_block_offset, |
230 | void **refcount_block) | |
f7d0fe02 | 231 | { |
ff99129a | 232 | BDRVQcow2State *s = bs->opaque; |
3b88e52b | 233 | |
66f82cee | 234 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD); |
9be38598 EH |
235 | return qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, |
236 | refcount_block); | |
f7d0fe02 KW |
237 | } |
238 | ||
018faafd | 239 | /* |
7324c10f HR |
240 | * Retrieves the refcount of the cluster given by its index and stores it in |
241 | * *refcount. Returns 0 on success and -errno on failure. | |
018faafd | 242 | */ |
7324c10f | 243 | int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index, |
0e06528e | 244 | uint64_t *refcount) |
f7d0fe02 | 245 | { |
ff99129a | 246 | BDRVQcow2State *s = bs->opaque; |
db8a31d1 | 247 | uint64_t refcount_table_index, block_index; |
f7d0fe02 | 248 | int64_t refcount_block_offset; |
018faafd | 249 | int ret; |
7453c96b | 250 | void *refcount_block; |
f7d0fe02 | 251 | |
17bd5f47 | 252 | refcount_table_index = cluster_index >> s->refcount_block_bits; |
7324c10f HR |
253 | if (refcount_table_index >= s->refcount_table_size) { |
254 | *refcount = 0; | |
f7d0fe02 | 255 | return 0; |
7324c10f | 256 | } |
26d49c46 HR |
257 | refcount_block_offset = |
258 | s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK; | |
7324c10f HR |
259 | if (!refcount_block_offset) { |
260 | *refcount = 0; | |
f7d0fe02 | 261 | return 0; |
7324c10f | 262 | } |
29c1a730 | 263 | |
a97c67ee HR |
264 | if (offset_into_cluster(s, refcount_block_offset)) { |
265 | qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" PRIx64 | |
266 | " unaligned (reftable index: %#" PRIx64 ")", | |
267 | refcount_block_offset, refcount_table_index); | |
268 | return -EIO; | |
269 | } | |
270 | ||
29c1a730 | 271 | ret = qcow2_cache_get(bs, s->refcount_block_cache, refcount_block_offset, |
7453c96b | 272 | &refcount_block); |
29c1a730 KW |
273 | if (ret < 0) { |
274 | return ret; | |
f7d0fe02 | 275 | } |
29c1a730 | 276 | |
17bd5f47 | 277 | block_index = cluster_index & (s->refcount_block_size - 1); |
7453c96b | 278 | *refcount = s->get_refcount(refcount_block, block_index); |
29c1a730 | 279 | |
2013c3d4 | 280 | qcow2_cache_put(s->refcount_block_cache, &refcount_block); |
29c1a730 | 281 | |
7324c10f | 282 | return 0; |
f7d0fe02 KW |
283 | } |
284 | ||
92dcb59f | 285 | /* Checks if two offsets are described by the same refcount block */ |
ff99129a | 286 | static int in_same_refcount_block(BDRVQcow2State *s, uint64_t offset_a, |
92dcb59f KW |
287 | uint64_t offset_b) |
288 | { | |
17bd5f47 HR |
289 | uint64_t block_a = offset_a >> (s->cluster_bits + s->refcount_block_bits); |
290 | uint64_t block_b = offset_b >> (s->cluster_bits + s->refcount_block_bits); | |
92dcb59f KW |
291 | |
292 | return (block_a == block_b); | |
293 | } | |
294 | ||
295 | /* | |
296 | * Loads a refcount block. If it doesn't exist yet, it is allocated first | |
297 | * (including growing the refcount table if needed). | |
298 | * | |
29c1a730 | 299 | * Returns 0 on success or -errno in error case |
92dcb59f | 300 | */ |
29c1a730 | 301 | static int alloc_refcount_block(BlockDriverState *bs, |
7453c96b | 302 | int64_t cluster_index, void **refcount_block) |
f7d0fe02 | 303 | { |
ff99129a | 304 | BDRVQcow2State *s = bs->opaque; |
92dcb59f | 305 | unsigned int refcount_table_index; |
12cc30a8 | 306 | int64_t ret; |
92dcb59f | 307 | |
66f82cee | 308 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); |
8252278a | 309 | |
92dcb59f | 310 | /* Find the refcount block for the given cluster */ |
17bd5f47 | 311 | refcount_table_index = cluster_index >> s->refcount_block_bits; |
92dcb59f KW |
312 | |
313 | if (refcount_table_index < s->refcount_table_size) { | |
314 | ||
315 | uint64_t refcount_block_offset = | |
76dc9e0c | 316 | s->refcount_table[refcount_table_index] & REFT_OFFSET_MASK; |
92dcb59f KW |
317 | |
318 | /* If it's already there, we're done */ | |
319 | if (refcount_block_offset) { | |
a97c67ee HR |
320 | if (offset_into_cluster(s, refcount_block_offset)) { |
321 | qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" | |
322 | PRIx64 " unaligned (reftable index: " | |
323 | "%#x)", refcount_block_offset, | |
324 | refcount_table_index); | |
325 | return -EIO; | |
326 | } | |
327 | ||
29c1a730 | 328 | return load_refcount_block(bs, refcount_block_offset, |
7453c96b | 329 | refcount_block); |
92dcb59f KW |
330 | } |
331 | } | |
332 | ||
333 | /* | |
334 | * If we came here, we need to allocate something. Something is at least | |
335 | * a cluster for the new refcount block. It may also include a new refcount | |
336 | * table if the old refcount table is too small. | |
337 | * | |
338 | * Note that allocating clusters here needs some special care: | |
339 | * | |
340 | * - We can't use the normal qcow2_alloc_clusters(), it would try to | |
341 | * increase the refcount and very likely we would end up with an endless | |
342 | * recursion. Instead we must place the refcount blocks in a way that | |
343 | * they can describe them themselves. | |
344 | * | |
345 | * - We need to consider that at this point we are inside update_refcounts | |
b106ad91 KW |
346 | * and potentially doing an initial refcount increase. This means that |
347 | * some clusters have already been allocated by the caller, but their | |
348 | * refcount isn't accurate yet. If we allocate clusters for metadata, we | |
349 | * need to return -EAGAIN to signal the caller that it needs to restart | |
350 | * the search for free clusters. | |
92dcb59f KW |
351 | * |
352 | * - alloc_clusters_noref and qcow2_free_clusters may load a different | |
353 | * refcount block into the cache | |
354 | */ | |
355 | ||
29c1a730 KW |
356 | *refcount_block = NULL; |
357 | ||
358 | /* We write to the refcount table, so we might depend on L2 tables */ | |
9991923b SH |
359 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
360 | if (ret < 0) { | |
361 | return ret; | |
362 | } | |
92dcb59f KW |
363 | |
364 | /* Allocate the refcount block itself and mark it as used */ | |
2eaa8f63 KW |
365 | int64_t new_block = alloc_clusters_noref(bs, s->cluster_size); |
366 | if (new_block < 0) { | |
367 | return new_block; | |
368 | } | |
f7d0fe02 | 369 | |
6bf45d59 AG |
370 | /* If we're allocating the block at offset 0 then something is wrong */ |
371 | if (new_block == 0) { | |
372 | qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " | |
373 | "allocation of refcount block at offset 0"); | |
374 | return -EIO; | |
375 | } | |
376 | ||
f7d0fe02 | 377 | #ifdef DEBUG_ALLOC2 |
92dcb59f KW |
378 | fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64 |
379 | " at %" PRIx64 "\n", | |
380 | refcount_table_index, cluster_index << s->cluster_bits, new_block); | |
f7d0fe02 | 381 | #endif |
92dcb59f KW |
382 | |
383 | if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) { | |
25408c09 | 384 | /* Zero the new refcount block before updating it */ |
29c1a730 | 385 | ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, |
7453c96b | 386 | refcount_block); |
29c1a730 | 387 | if (ret < 0) { |
60c48a29 | 388 | goto fail; |
29c1a730 KW |
389 | } |
390 | ||
391 | memset(*refcount_block, 0, s->cluster_size); | |
25408c09 | 392 | |
92dcb59f KW |
393 | /* The block describes itself, need to update the cache */ |
394 | int block_index = (new_block >> s->cluster_bits) & | |
17bd5f47 | 395 | (s->refcount_block_size - 1); |
7453c96b | 396 | s->set_refcount(*refcount_block, block_index, 1); |
92dcb59f KW |
397 | } else { |
398 | /* Described somewhere else. This can recurse at most twice before we | |
399 | * arrive at a block that describes itself. */ | |
2aabe7c7 | 400 | ret = update_refcount(bs, new_block, s->cluster_size, 1, false, |
6cfcb9b8 | 401 | QCOW2_DISCARD_NEVER); |
92dcb59f | 402 | if (ret < 0) { |
60c48a29 | 403 | goto fail; |
92dcb59f | 404 | } |
25408c09 | 405 | |
9991923b SH |
406 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
407 | if (ret < 0) { | |
60c48a29 | 408 | goto fail; |
9991923b | 409 | } |
1c4c2814 | 410 | |
25408c09 KW |
411 | /* Initialize the new refcount block only after updating its refcount, |
412 | * update_refcount uses the refcount cache itself */ | |
29c1a730 | 413 | ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, new_block, |
7453c96b | 414 | refcount_block); |
29c1a730 | 415 | if (ret < 0) { |
60c48a29 | 416 | goto fail; |
29c1a730 KW |
417 | } |
418 | ||
419 | memset(*refcount_block, 0, s->cluster_size); | |
92dcb59f KW |
420 | } |
421 | ||
422 | /* Now the new refcount block needs to be written to disk */ | |
66f82cee | 423 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE); |
2d135ee9 | 424 | qcow2_cache_entry_mark_dirty(s->refcount_block_cache, *refcount_block); |
29c1a730 | 425 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
92dcb59f | 426 | if (ret < 0) { |
60c48a29 | 427 | goto fail; |
92dcb59f KW |
428 | } |
429 | ||
430 | /* If the refcount table is big enough, just hook the block up there */ | |
431 | if (refcount_table_index < s->refcount_table_size) { | |
432 | uint64_t data64 = cpu_to_be64(new_block); | |
66f82cee | 433 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP); |
d9ca2ea2 | 434 | ret = bdrv_pwrite_sync(bs->file, |
92dcb59f KW |
435 | s->refcount_table_offset + refcount_table_index * sizeof(uint64_t), |
436 | &data64, sizeof(data64)); | |
437 | if (ret < 0) { | |
60c48a29 | 438 | goto fail; |
92dcb59f KW |
439 | } |
440 | ||
441 | s->refcount_table[refcount_table_index] = new_block; | |
7061a078 AG |
442 | /* If there's a hole in s->refcount_table then it can happen |
443 | * that refcount_table_index < s->max_refcount_table_index */ | |
444 | s->max_refcount_table_index = | |
445 | MAX(s->max_refcount_table_index, refcount_table_index); | |
b106ad91 KW |
446 | |
447 | /* The new refcount block may be where the caller intended to put its | |
448 | * data, so let it restart the search. */ | |
449 | return -EAGAIN; | |
29c1a730 KW |
450 | } |
451 | ||
2013c3d4 | 452 | qcow2_cache_put(s->refcount_block_cache, refcount_block); |
92dcb59f KW |
453 | |
454 | /* | |
455 | * If we come here, we need to grow the refcount table. Again, a new | |
456 | * refcount table needs some space and we can't simply allocate to avoid | |
457 | * endless recursion. | |
458 | * | |
459 | * Therefore let's grab new refcount blocks at the end of the image, which | |
460 | * will describe themselves and the new refcount table. This way we can | |
461 | * reference them only in the new table and do the switch to the new | |
462 | * refcount table at once without producing an inconsistent state in | |
463 | * between. | |
464 | */ | |
66f82cee | 465 | BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW); |
8252278a | 466 | |
14a58a4e HR |
467 | /* Calculate the number of refcount blocks needed so far; this will be the |
468 | * basis for calculating the index of the first cluster used for the | |
469 | * self-describing refcount structures which we are about to create. | |
470 | * | |
471 | * Because we reached this point, there cannot be any refcount entries for | |
472 | * cluster_index or higher indices yet. However, because new_block has been | |
473 | * allocated to describe that cluster (and it will assume this role later | |
474 | * on), we cannot use that index; also, new_block may actually have a higher | |
475 | * cluster index than cluster_index, so it needs to be taken into account | |
476 | * here (and 1 needs to be added to its value because that cluster is used). | |
477 | */ | |
478 | uint64_t blocks_used = DIV_ROUND_UP(MAX(cluster_index + 1, | |
479 | (new_block >> s->cluster_bits) + 1), | |
480 | s->refcount_block_size); | |
92dcb59f | 481 | |
12cc30a8 HR |
482 | /* Create the new refcount table and blocks */ |
483 | uint64_t meta_offset = (blocks_used * s->refcount_block_size) * | |
484 | s->cluster_size; | |
485 | ||
486 | ret = qcow2_refcount_area(bs, meta_offset, 0, false, | |
487 | refcount_table_index, new_block); | |
488 | if (ret < 0) { | |
489 | return ret; | |
2b5d5953 KW |
490 | } |
491 | ||
12cc30a8 HR |
492 | ret = load_refcount_block(bs, new_block, refcount_block); |
493 | if (ret < 0) { | |
494 | return ret; | |
495 | } | |
92dcb59f | 496 | |
12cc30a8 HR |
497 | /* If we were trying to do the initial refcount update for some cluster |
498 | * allocation, we might have used the same clusters to store newly | |
499 | * allocated metadata. Make the caller search some new space. */ | |
500 | return -EAGAIN; | |
92dcb59f | 501 | |
60c48a29 | 502 | fail: |
12cc30a8 | 503 | if (*refcount_block != NULL) { |
2013c3d4 | 504 | qcow2_cache_put(s->refcount_block_cache, refcount_block); |
12cc30a8 HR |
505 | } |
506 | return ret; | |
507 | } | |
92dcb59f | 508 | |
12cc30a8 HR |
509 | /* |
510 | * Starting at @start_offset, this function creates new self-covering refcount | |
511 | * structures: A new refcount table and refcount blocks which cover all of | |
512 | * themselves, and a number of @additional_clusters beyond their end. | |
513 | * @start_offset must be at the end of the image file, that is, there must be | |
514 | * only empty space beyond it. | |
515 | * If @exact_size is false, the refcount table will have 50 % more entries than | |
516 | * necessary so it will not need to grow again soon. | |
517 | * If @new_refblock_offset is not zero, it contains the offset of a refcount | |
518 | * block that should be entered into the new refcount table at index | |
519 | * @new_refblock_index. | |
520 | * | |
521 | * Returns: The offset after the new refcount structures (i.e. where the | |
522 | * @additional_clusters may be placed) on success, -errno on error. | |
523 | */ | |
772d1f97 HR |
524 | int64_t qcow2_refcount_area(BlockDriverState *bs, uint64_t start_offset, |
525 | uint64_t additional_clusters, bool exact_size, | |
526 | int new_refblock_index, | |
527 | uint64_t new_refblock_offset) | |
12cc30a8 HR |
528 | { |
529 | BDRVQcow2State *s = bs->opaque; | |
530 | uint64_t total_refblock_count_u64, additional_refblock_count; | |
531 | int total_refblock_count, table_size, area_reftable_index, table_clusters; | |
532 | int i; | |
533 | uint64_t table_offset, block_offset, end_offset; | |
534 | int ret; | |
535 | uint64_t *new_table; | |
92dcb59f | 536 | |
12cc30a8 | 537 | assert(!(start_offset % s->cluster_size)); |
de82815d | 538 | |
12cc30a8 HR |
539 | qcow2_refcount_metadata_size(start_offset / s->cluster_size + |
540 | additional_clusters, | |
541 | s->cluster_size, s->refcount_order, | |
542 | !exact_size, &total_refblock_count_u64); | |
543 | if (total_refblock_count_u64 > QCOW_MAX_REFTABLE_SIZE) { | |
544 | return -EFBIG; | |
545 | } | |
546 | total_refblock_count = total_refblock_count_u64; | |
547 | ||
548 | /* Index in the refcount table of the first refcount block to cover the area | |
549 | * of refcount structures we are about to create; we know that | |
550 | * @total_refblock_count can cover @start_offset, so this will definitely | |
551 | * fit into an int. */ | |
552 | area_reftable_index = (start_offset / s->cluster_size) / | |
553 | s->refcount_block_size; | |
554 | ||
555 | if (exact_size) { | |
556 | table_size = total_refblock_count; | |
557 | } else { | |
558 | table_size = total_refblock_count + | |
559 | DIV_ROUND_UP(total_refblock_count, 2); | |
560 | } | |
561 | /* The qcow2 file can only store the reftable size in number of clusters */ | |
562 | table_size = ROUND_UP(table_size, s->cluster_size / sizeof(uint64_t)); | |
563 | table_clusters = (table_size * sizeof(uint64_t)) / s->cluster_size; | |
564 | ||
565 | if (table_size > QCOW_MAX_REFTABLE_SIZE) { | |
566 | return -EFBIG; | |
567 | } | |
568 | ||
569 | new_table = g_try_new0(uint64_t, table_size); | |
570 | ||
571 | assert(table_size > 0); | |
572 | if (new_table == NULL) { | |
de82815d | 573 | ret = -ENOMEM; |
12cc30a8 | 574 | goto fail; |
de82815d | 575 | } |
92dcb59f | 576 | |
92dcb59f | 577 | /* Fill the new refcount table */ |
12cc30a8 HR |
578 | if (table_size > s->max_refcount_table_index) { |
579 | /* We're actually growing the reftable */ | |
580 | memcpy(new_table, s->refcount_table, | |
581 | (s->max_refcount_table_index + 1) * sizeof(uint64_t)); | |
582 | } else { | |
583 | /* Improbable case: We're shrinking the reftable. However, the caller | |
584 | * has assured us that there is only empty space beyond @start_offset, | |
585 | * so we can simply drop all of the refblocks that won't fit into the | |
586 | * new reftable. */ | |
587 | memcpy(new_table, s->refcount_table, table_size * sizeof(uint64_t)); | |
588 | } | |
92dcb59f | 589 | |
12cc30a8 HR |
590 | if (new_refblock_offset) { |
591 | assert(new_refblock_index < total_refblock_count); | |
592 | new_table[new_refblock_index] = new_refblock_offset; | |
593 | } | |
594 | ||
595 | /* Count how many new refblocks we have to create */ | |
596 | additional_refblock_count = 0; | |
597 | for (i = area_reftable_index; i < total_refblock_count; i++) { | |
598 | if (!new_table[i]) { | |
599 | additional_refblock_count++; | |
600 | } | |
92dcb59f KW |
601 | } |
602 | ||
12cc30a8 HR |
603 | table_offset = start_offset + additional_refblock_count * s->cluster_size; |
604 | end_offset = table_offset + table_clusters * s->cluster_size; | |
605 | ||
606 | /* Fill the refcount blocks, and create new ones, if necessary */ | |
607 | block_offset = start_offset; | |
608 | for (i = area_reftable_index; i < total_refblock_count; i++) { | |
609 | void *refblock_data; | |
610 | uint64_t first_offset_covered; | |
611 | ||
612 | /* Reuse an existing refblock if possible, create a new one otherwise */ | |
613 | if (new_table[i]) { | |
614 | ret = qcow2_cache_get(bs, s->refcount_block_cache, new_table[i], | |
615 | &refblock_data); | |
616 | if (ret < 0) { | |
617 | goto fail; | |
618 | } | |
619 | } else { | |
620 | ret = qcow2_cache_get_empty(bs, s->refcount_block_cache, | |
621 | block_offset, &refblock_data); | |
622 | if (ret < 0) { | |
623 | goto fail; | |
624 | } | |
625 | memset(refblock_data, 0, s->cluster_size); | |
2d135ee9 | 626 | qcow2_cache_entry_mark_dirty(s->refcount_block_cache, |
12cc30a8 HR |
627 | refblock_data); |
628 | ||
629 | new_table[i] = block_offset; | |
630 | block_offset += s->cluster_size; | |
631 | } | |
632 | ||
633 | /* First host offset covered by this refblock */ | |
634 | first_offset_covered = (uint64_t)i * s->refcount_block_size * | |
635 | s->cluster_size; | |
636 | if (first_offset_covered < end_offset) { | |
637 | int j, end_index; | |
638 | ||
639 | /* Set the refcount of all of the new refcount structures to 1 */ | |
640 | ||
641 | if (first_offset_covered < start_offset) { | |
642 | assert(i == area_reftable_index); | |
643 | j = (start_offset - first_offset_covered) / s->cluster_size; | |
644 | assert(j < s->refcount_block_size); | |
645 | } else { | |
646 | j = 0; | |
647 | } | |
648 | ||
649 | end_index = MIN((end_offset - first_offset_covered) / | |
650 | s->cluster_size, | |
651 | s->refcount_block_size); | |
652 | ||
653 | for (; j < end_index; j++) { | |
654 | /* The caller guaranteed us this space would be empty */ | |
655 | assert(s->get_refcount(refblock_data, j) == 0); | |
656 | s->set_refcount(refblock_data, j, 1); | |
657 | } | |
658 | ||
2d135ee9 | 659 | qcow2_cache_entry_mark_dirty(s->refcount_block_cache, |
12cc30a8 HR |
660 | refblock_data); |
661 | } | |
662 | ||
2013c3d4 | 663 | qcow2_cache_put(s->refcount_block_cache, &refblock_data); |
92dcb59f KW |
664 | } |
665 | ||
12cc30a8 HR |
666 | assert(block_offset == table_offset); |
667 | ||
92dcb59f | 668 | /* Write refcount blocks to disk */ |
66f82cee | 669 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS); |
12cc30a8 | 670 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
92dcb59f | 671 | if (ret < 0) { |
12cc30a8 | 672 | goto fail; |
92dcb59f KW |
673 | } |
674 | ||
675 | /* Write refcount table to disk */ | |
12cc30a8 | 676 | for (i = 0; i < total_refblock_count; i++) { |
92dcb59f KW |
677 | cpu_to_be64s(&new_table[i]); |
678 | } | |
679 | ||
66f82cee | 680 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE); |
d9ca2ea2 | 681 | ret = bdrv_pwrite_sync(bs->file, table_offset, new_table, |
92dcb59f KW |
682 | table_size * sizeof(uint64_t)); |
683 | if (ret < 0) { | |
12cc30a8 | 684 | goto fail; |
92dcb59f KW |
685 | } |
686 | ||
12cc30a8 | 687 | for (i = 0; i < total_refblock_count; i++) { |
87267753 | 688 | be64_to_cpus(&new_table[i]); |
92dcb59f | 689 | } |
f7d0fe02 | 690 | |
92dcb59f | 691 | /* Hook up the new refcount table in the qcow2 header */ |
95334230 JS |
692 | struct QEMU_PACKED { |
693 | uint64_t d64; | |
694 | uint32_t d32; | |
695 | } data; | |
f1f7a1dd PM |
696 | data.d64 = cpu_to_be64(table_offset); |
697 | data.d32 = cpu_to_be32(table_clusters); | |
66f82cee | 698 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE); |
d9ca2ea2 | 699 | ret = bdrv_pwrite_sync(bs->file, |
9a4f4c31 | 700 | offsetof(QCowHeader, refcount_table_offset), |
95334230 | 701 | &data, sizeof(data)); |
92dcb59f | 702 | if (ret < 0) { |
12cc30a8 | 703 | goto fail; |
f2b7c8b3 KW |
704 | } |
705 | ||
92dcb59f KW |
706 | /* And switch it in memory */ |
707 | uint64_t old_table_offset = s->refcount_table_offset; | |
708 | uint64_t old_table_size = s->refcount_table_size; | |
709 | ||
7267c094 | 710 | g_free(s->refcount_table); |
f7d0fe02 | 711 | s->refcount_table = new_table; |
92dcb59f | 712 | s->refcount_table_size = table_size; |
f7d0fe02 | 713 | s->refcount_table_offset = table_offset; |
7061a078 | 714 | update_max_refcount_table_index(s); |
f7d0fe02 | 715 | |
b106ad91 | 716 | /* Free old table. */ |
6cfcb9b8 KW |
717 | qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t), |
718 | QCOW2_DISCARD_OTHER); | |
f7d0fe02 | 719 | |
12cc30a8 | 720 | return end_offset; |
f7d0fe02 | 721 | |
12cc30a8 | 722 | fail: |
7267c094 | 723 | g_free(new_table); |
29c1a730 | 724 | return ret; |
9923e05e KW |
725 | } |
726 | ||
0b919fae KW |
727 | void qcow2_process_discards(BlockDriverState *bs, int ret) |
728 | { | |
ff99129a | 729 | BDRVQcow2State *s = bs->opaque; |
0b919fae KW |
730 | Qcow2DiscardRegion *d, *next; |
731 | ||
732 | QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) { | |
733 | QTAILQ_REMOVE(&s->discards, d, next); | |
734 | ||
735 | /* Discard is optional, ignore the return value */ | |
736 | if (ret >= 0) { | |
0c51a893 | 737 | bdrv_pdiscard(bs->file->bs, d->offset, d->bytes); |
0b919fae KW |
738 | } |
739 | ||
740 | g_free(d); | |
741 | } | |
742 | } | |
743 | ||
744 | static void update_refcount_discard(BlockDriverState *bs, | |
745 | uint64_t offset, uint64_t length) | |
746 | { | |
ff99129a | 747 | BDRVQcow2State *s = bs->opaque; |
0b919fae KW |
748 | Qcow2DiscardRegion *d, *p, *next; |
749 | ||
750 | QTAILQ_FOREACH(d, &s->discards, next) { | |
751 | uint64_t new_start = MIN(offset, d->offset); | |
752 | uint64_t new_end = MAX(offset + length, d->offset + d->bytes); | |
753 | ||
754 | if (new_end - new_start <= length + d->bytes) { | |
755 | /* There can't be any overlap, areas ending up here have no | |
756 | * references any more and therefore shouldn't get freed another | |
757 | * time. */ | |
758 | assert(d->bytes + length == new_end - new_start); | |
759 | d->offset = new_start; | |
760 | d->bytes = new_end - new_start; | |
761 | goto found; | |
762 | } | |
763 | } | |
764 | ||
765 | d = g_malloc(sizeof(*d)); | |
766 | *d = (Qcow2DiscardRegion) { | |
767 | .bs = bs, | |
768 | .offset = offset, | |
769 | .bytes = length, | |
770 | }; | |
771 | QTAILQ_INSERT_TAIL(&s->discards, d, next); | |
772 | ||
773 | found: | |
774 | /* Merge discard requests if they are adjacent now */ | |
775 | QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) { | |
776 | if (p == d | |
777 | || p->offset > d->offset + d->bytes | |
778 | || d->offset > p->offset + p->bytes) | |
779 | { | |
780 | continue; | |
781 | } | |
782 | ||
783 | /* Still no overlap possible */ | |
784 | assert(p->offset == d->offset + d->bytes | |
785 | || d->offset == p->offset + p->bytes); | |
786 | ||
787 | QTAILQ_REMOVE(&s->discards, p, next); | |
788 | d->offset = MIN(d->offset, p->offset); | |
789 | d->bytes += p->bytes; | |
d8bb71b6 | 790 | g_free(p); |
0b919fae KW |
791 | } |
792 | } | |
793 | ||
f7d0fe02 | 794 | /* XXX: cache several refcount block clusters ? */ |
2aabe7c7 HR |
795 | /* @addend is the absolute value of the addend; if @decrease is set, @addend |
796 | * will be subtracted from the current refcount, otherwise it will be added */ | |
db3a964f | 797 | static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs, |
2aabe7c7 HR |
798 | int64_t offset, |
799 | int64_t length, | |
0e06528e | 800 | uint64_t addend, |
2aabe7c7 HR |
801 | bool decrease, |
802 | enum qcow2_discard_type type) | |
f7d0fe02 | 803 | { |
ff99129a | 804 | BDRVQcow2State *s = bs->opaque; |
f7d0fe02 | 805 | int64_t start, last, cluster_offset; |
7453c96b | 806 | void *refcount_block = NULL; |
29c1a730 | 807 | int64_t old_table_index = -1; |
09508d13 | 808 | int ret; |
f7d0fe02 KW |
809 | |
810 | #ifdef DEBUG_ALLOC2 | |
2aabe7c7 | 811 | fprintf(stderr, "update_refcount: offset=%" PRId64 " size=%" PRId64 |
0e06528e | 812 | " addend=%s%" PRIu64 "\n", offset, length, decrease ? "-" : "", |
2aabe7c7 | 813 | addend); |
f7d0fe02 | 814 | #endif |
7322afe7 | 815 | if (length < 0) { |
f7d0fe02 | 816 | return -EINVAL; |
7322afe7 KW |
817 | } else if (length == 0) { |
818 | return 0; | |
819 | } | |
820 | ||
2aabe7c7 | 821 | if (decrease) { |
29c1a730 KW |
822 | qcow2_cache_set_dependency(bs, s->refcount_block_cache, |
823 | s->l2_table_cache); | |
824 | } | |
825 | ||
ac95acdb HT |
826 | start = start_of_cluster(s, offset); |
827 | last = start_of_cluster(s, offset + length - 1); | |
f7d0fe02 KW |
828 | for(cluster_offset = start; cluster_offset <= last; |
829 | cluster_offset += s->cluster_size) | |
830 | { | |
2aabe7c7 | 831 | int block_index; |
0e06528e | 832 | uint64_t refcount; |
f7d0fe02 | 833 | int64_t cluster_index = cluster_offset >> s->cluster_bits; |
17bd5f47 | 834 | int64_t table_index = cluster_index >> s->refcount_block_bits; |
f7d0fe02 | 835 | |
29c1a730 KW |
836 | /* Load the refcount block and allocate it if needed */ |
837 | if (table_index != old_table_index) { | |
838 | if (refcount_block) { | |
2013c3d4 | 839 | qcow2_cache_put(s->refcount_block_cache, &refcount_block); |
29c1a730 | 840 | } |
29c1a730 | 841 | ret = alloc_refcount_block(bs, cluster_index, &refcount_block); |
ed0df867 | 842 | if (ret < 0) { |
29c1a730 | 843 | goto fail; |
f7d0fe02 | 844 | } |
f7d0fe02 | 845 | } |
29c1a730 | 846 | old_table_index = table_index; |
f7d0fe02 | 847 | |
2d135ee9 | 848 | qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refcount_block); |
f7d0fe02 KW |
849 | |
850 | /* we can update the count and save it */ | |
17bd5f47 | 851 | block_index = cluster_index & (s->refcount_block_size - 1); |
f7d0fe02 | 852 | |
7453c96b | 853 | refcount = s->get_refcount(refcount_block, block_index); |
0e06528e HR |
854 | if (decrease ? (refcount - addend > refcount) |
855 | : (refcount + addend < refcount || | |
856 | refcount + addend > s->refcount_max)) | |
2aabe7c7 | 857 | { |
09508d13 KW |
858 | ret = -EINVAL; |
859 | goto fail; | |
860 | } | |
2aabe7c7 HR |
861 | if (decrease) { |
862 | refcount -= addend; | |
863 | } else { | |
864 | refcount += addend; | |
865 | } | |
f7d0fe02 KW |
866 | if (refcount == 0 && cluster_index < s->free_cluster_index) { |
867 | s->free_cluster_index = cluster_index; | |
868 | } | |
7453c96b | 869 | s->set_refcount(refcount_block, block_index, refcount); |
0b919fae | 870 | |
f71c08ea PB |
871 | if (refcount == 0) { |
872 | void *table; | |
873 | ||
6e6fa760 | 874 | table = qcow2_cache_is_table_offset(s->refcount_block_cache, |
f71c08ea PB |
875 | offset); |
876 | if (table != NULL) { | |
2013c3d4 | 877 | qcow2_cache_put(s->refcount_block_cache, &refcount_block); |
77aadd7b | 878 | qcow2_cache_discard(s->refcount_block_cache, table); |
f71c08ea PB |
879 | } |
880 | ||
6e6fa760 | 881 | table = qcow2_cache_is_table_offset(s->l2_table_cache, offset); |
f71c08ea | 882 | if (table != NULL) { |
77aadd7b | 883 | qcow2_cache_discard(s->l2_table_cache, table); |
f71c08ea PB |
884 | } |
885 | ||
886 | if (s->discard_passthrough[type]) { | |
887 | update_refcount_discard(bs, cluster_offset, s->cluster_size); | |
888 | } | |
67af674e | 889 | } |
f7d0fe02 KW |
890 | } |
891 | ||
09508d13 KW |
892 | ret = 0; |
893 | fail: | |
0b919fae KW |
894 | if (!s->cache_discards) { |
895 | qcow2_process_discards(bs, ret); | |
896 | } | |
897 | ||
f7d0fe02 | 898 | /* Write last changed block to disk */ |
29c1a730 | 899 | if (refcount_block) { |
2013c3d4 | 900 | qcow2_cache_put(s->refcount_block_cache, &refcount_block); |
f7d0fe02 KW |
901 | } |
902 | ||
09508d13 KW |
903 | /* |
904 | * Try do undo any updates if an error is returned (This may succeed in | |
905 | * some cases like ENOSPC for allocating a new refcount block) | |
906 | */ | |
907 | if (ret < 0) { | |
908 | int dummy; | |
2aabe7c7 HR |
909 | dummy = update_refcount(bs, offset, cluster_offset - offset, addend, |
910 | !decrease, QCOW2_DISCARD_NEVER); | |
83e3f76c | 911 | (void)dummy; |
09508d13 KW |
912 | } |
913 | ||
914 | return ret; | |
f7d0fe02 KW |
915 | } |
916 | ||
018faafd | 917 | /* |
44751917 | 918 | * Increases or decreases the refcount of a given cluster. |
018faafd | 919 | * |
2aabe7c7 HR |
920 | * @addend is the absolute value of the addend; if @decrease is set, @addend |
921 | * will be subtracted from the current refcount, otherwise it will be added. | |
922 | * | |
c6e9d8ae | 923 | * On success 0 is returned; on failure -errno is returned. |
018faafd | 924 | */ |
32b6444d HR |
925 | int qcow2_update_cluster_refcount(BlockDriverState *bs, |
926 | int64_t cluster_index, | |
0e06528e | 927 | uint64_t addend, bool decrease, |
32b6444d | 928 | enum qcow2_discard_type type) |
f7d0fe02 | 929 | { |
ff99129a | 930 | BDRVQcow2State *s = bs->opaque; |
f7d0fe02 KW |
931 | int ret; |
932 | ||
6cfcb9b8 | 933 | ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend, |
2aabe7c7 | 934 | decrease, type); |
f7d0fe02 KW |
935 | if (ret < 0) { |
936 | return ret; | |
937 | } | |
938 | ||
c6e9d8ae | 939 | return 0; |
f7d0fe02 KW |
940 | } |
941 | ||
942 | ||
943 | ||
944 | /*********************************************************/ | |
945 | /* cluster allocation functions */ | |
946 | ||
947 | ||
948 | ||
949 | /* return < 0 if error */ | |
bb572aef | 950 | static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size) |
f7d0fe02 | 951 | { |
ff99129a | 952 | BDRVQcow2State *s = bs->opaque; |
0e06528e | 953 | uint64_t i, nb_clusters, refcount; |
7324c10f | 954 | int ret; |
f7d0fe02 | 955 | |
ecbda7a2 KW |
956 | /* We can't allocate clusters if they may still be queued for discard. */ |
957 | if (s->cache_discards) { | |
958 | qcow2_process_discards(bs, 0); | |
959 | } | |
960 | ||
f7d0fe02 KW |
961 | nb_clusters = size_to_clusters(s, size); |
962 | retry: | |
963 | for(i = 0; i < nb_clusters; i++) { | |
bb572aef | 964 | uint64_t next_cluster_index = s->free_cluster_index++; |
7324c10f | 965 | ret = qcow2_get_refcount(bs, next_cluster_index, &refcount); |
2eaa8f63 | 966 | |
7324c10f HR |
967 | if (ret < 0) { |
968 | return ret; | |
2eaa8f63 | 969 | } else if (refcount != 0) { |
f7d0fe02 | 970 | goto retry; |
2eaa8f63 | 971 | } |
f7d0fe02 | 972 | } |
91f827dc HR |
973 | |
974 | /* Make sure that all offsets in the "allocated" range are representable | |
975 | * in an int64_t */ | |
65f33bc0 HR |
976 | if (s->free_cluster_index > 0 && |
977 | s->free_cluster_index - 1 > (INT64_MAX >> s->cluster_bits)) | |
978 | { | |
91f827dc HR |
979 | return -EFBIG; |
980 | } | |
981 | ||
f7d0fe02 | 982 | #ifdef DEBUG_ALLOC2 |
35ee5e39 | 983 | fprintf(stderr, "alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n", |
f7d0fe02 KW |
984 | size, |
985 | (s->free_cluster_index - nb_clusters) << s->cluster_bits); | |
986 | #endif | |
987 | return (s->free_cluster_index - nb_clusters) << s->cluster_bits; | |
988 | } | |
989 | ||
bb572aef | 990 | int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size) |
f7d0fe02 KW |
991 | { |
992 | int64_t offset; | |
db3a964f | 993 | int ret; |
f7d0fe02 | 994 | |
66f82cee | 995 | BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC); |
b106ad91 KW |
996 | do { |
997 | offset = alloc_clusters_noref(bs, size); | |
998 | if (offset < 0) { | |
999 | return offset; | |
1000 | } | |
1001 | ||
2aabe7c7 | 1002 | ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER); |
b106ad91 | 1003 | } while (ret == -EAGAIN); |
2eaa8f63 | 1004 | |
db3a964f KW |
1005 | if (ret < 0) { |
1006 | return ret; | |
1007 | } | |
1c4c2814 | 1008 | |
f7d0fe02 KW |
1009 | return offset; |
1010 | } | |
1011 | ||
b6d36def HR |
1012 | int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset, |
1013 | int64_t nb_clusters) | |
256900b1 | 1014 | { |
ff99129a | 1015 | BDRVQcow2State *s = bs->opaque; |
0e06528e | 1016 | uint64_t cluster_index, refcount; |
33304ec9 | 1017 | uint64_t i; |
7324c10f | 1018 | int ret; |
33304ec9 HT |
1019 | |
1020 | assert(nb_clusters >= 0); | |
1021 | if (nb_clusters == 0) { | |
1022 | return 0; | |
1023 | } | |
256900b1 | 1024 | |
b106ad91 KW |
1025 | do { |
1026 | /* Check how many clusters there are free */ | |
1027 | cluster_index = offset >> s->cluster_bits; | |
1028 | for(i = 0; i < nb_clusters; i++) { | |
7324c10f HR |
1029 | ret = qcow2_get_refcount(bs, cluster_index++, &refcount); |
1030 | if (ret < 0) { | |
1031 | return ret; | |
b106ad91 KW |
1032 | } else if (refcount != 0) { |
1033 | break; | |
1034 | } | |
256900b1 | 1035 | } |
256900b1 | 1036 | |
b106ad91 | 1037 | /* And then allocate them */ |
2aabe7c7 | 1038 | ret = update_refcount(bs, offset, i << s->cluster_bits, 1, false, |
b106ad91 KW |
1039 | QCOW2_DISCARD_NEVER); |
1040 | } while (ret == -EAGAIN); | |
f24423bd | 1041 | |
256900b1 KW |
1042 | if (ret < 0) { |
1043 | return ret; | |
1044 | } | |
1045 | ||
1046 | return i; | |
1047 | } | |
1048 | ||
f7d0fe02 KW |
1049 | /* only used to allocate compressed sectors. We try to allocate |
1050 | contiguous sectors. size must be <= cluster_size */ | |
ed6ccf0f | 1051 | int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size) |
f7d0fe02 | 1052 | { |
ff99129a | 1053 | BDRVQcow2State *s = bs->opaque; |
8c44dfbc HR |
1054 | int64_t offset; |
1055 | size_t free_in_cluster; | |
1056 | int ret; | |
f7d0fe02 | 1057 | |
66f82cee | 1058 | BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES); |
f7d0fe02 | 1059 | assert(size > 0 && size <= s->cluster_size); |
8c44dfbc HR |
1060 | assert(!s->free_byte_offset || offset_into_cluster(s, s->free_byte_offset)); |
1061 | ||
1062 | offset = s->free_byte_offset; | |
1063 | ||
1064 | if (offset) { | |
0e06528e | 1065 | uint64_t refcount; |
7324c10f HR |
1066 | ret = qcow2_get_refcount(bs, offset >> s->cluster_bits, &refcount); |
1067 | if (ret < 0) { | |
1068 | return ret; | |
5d757b56 | 1069 | } |
8c44dfbc | 1070 | |
346a53df | 1071 | if (refcount == s->refcount_max) { |
8c44dfbc | 1072 | offset = 0; |
5d757b56 | 1073 | } |
8c44dfbc HR |
1074 | } |
1075 | ||
1076 | free_in_cluster = s->cluster_size - offset_into_cluster(s, offset); | |
3e5feb62 JM |
1077 | do { |
1078 | if (!offset || free_in_cluster < size) { | |
1079 | int64_t new_cluster = alloc_clusters_noref(bs, s->cluster_size); | |
1080 | if (new_cluster < 0) { | |
1081 | return new_cluster; | |
1082 | } | |
8c44dfbc | 1083 | |
8aa34834 AG |
1084 | if (new_cluster == 0) { |
1085 | qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " | |
1086 | "allocation of compressed cluster " | |
1087 | "at offset 0"); | |
1088 | return -EIO; | |
1089 | } | |
1090 | ||
3e5feb62 JM |
1091 | if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) { |
1092 | offset = new_cluster; | |
2ac01520 HR |
1093 | free_in_cluster = s->cluster_size; |
1094 | } else { | |
1095 | free_in_cluster += s->cluster_size; | |
3e5feb62 | 1096 | } |
f7d0fe02 | 1097 | } |
29216ed1 | 1098 | |
3e5feb62 JM |
1099 | assert(offset); |
1100 | ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER); | |
2ac01520 HR |
1101 | if (ret < 0) { |
1102 | offset = 0; | |
1103 | } | |
3e5feb62 | 1104 | } while (ret == -EAGAIN); |
8c44dfbc HR |
1105 | if (ret < 0) { |
1106 | return ret; | |
1107 | } | |
1108 | ||
1109 | /* The cluster refcount was incremented; refcount blocks must be flushed | |
1110 | * before the caller's L2 table updates. */ | |
c1f5bafd | 1111 | qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache); |
8c44dfbc HR |
1112 | |
1113 | s->free_byte_offset = offset + size; | |
1114 | if (!offset_into_cluster(s, s->free_byte_offset)) { | |
1115 | s->free_byte_offset = 0; | |
1116 | } | |
1117 | ||
f7d0fe02 KW |
1118 | return offset; |
1119 | } | |
1120 | ||
ed6ccf0f | 1121 | void qcow2_free_clusters(BlockDriverState *bs, |
6cfcb9b8 KW |
1122 | int64_t offset, int64_t size, |
1123 | enum qcow2_discard_type type) | |
f7d0fe02 | 1124 | { |
db3a964f KW |
1125 | int ret; |
1126 | ||
66f82cee | 1127 | BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE); |
2aabe7c7 | 1128 | ret = update_refcount(bs, offset, size, 1, true, type); |
db3a964f KW |
1129 | if (ret < 0) { |
1130 | fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret)); | |
003fad6e | 1131 | /* TODO Remember the clusters to free them later and avoid leaking */ |
db3a964f | 1132 | } |
f7d0fe02 KW |
1133 | } |
1134 | ||
45aba42f | 1135 | /* |
c7a4c37a KW |
1136 | * Free a cluster using its L2 entry (handles clusters of all types, e.g. |
1137 | * normal cluster, compressed cluster, etc.) | |
45aba42f | 1138 | */ |
6cfcb9b8 KW |
1139 | void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry, |
1140 | int nb_clusters, enum qcow2_discard_type type) | |
45aba42f | 1141 | { |
ff99129a | 1142 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 1143 | |
c7a4c37a KW |
1144 | switch (qcow2_get_cluster_type(l2_entry)) { |
1145 | case QCOW2_CLUSTER_COMPRESSED: | |
1146 | { | |
1147 | int nb_csectors; | |
1148 | nb_csectors = ((l2_entry >> s->csize_shift) & | |
1149 | s->csize_mask) + 1; | |
1150 | qcow2_free_clusters(bs, | |
1151 | (l2_entry & s->cluster_offset_mask) & ~511, | |
6cfcb9b8 | 1152 | nb_csectors * 512, type); |
c7a4c37a KW |
1153 | } |
1154 | break; | |
1155 | case QCOW2_CLUSTER_NORMAL: | |
fdfab37d EB |
1156 | case QCOW2_CLUSTER_ZERO_ALLOC: |
1157 | if (offset_into_cluster(s, l2_entry & L2E_OFFSET_MASK)) { | |
1158 | qcow2_signal_corruption(bs, false, -1, -1, | |
1159 | "Cannot free unaligned cluster %#llx", | |
1160 | l2_entry & L2E_OFFSET_MASK); | |
1161 | } else { | |
1162 | qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK, | |
1163 | nb_clusters << s->cluster_bits, type); | |
8f730dd2 | 1164 | } |
c7a4c37a | 1165 | break; |
fdfab37d | 1166 | case QCOW2_CLUSTER_ZERO_PLAIN: |
c7a4c37a KW |
1167 | case QCOW2_CLUSTER_UNALLOCATED: |
1168 | break; | |
1169 | default: | |
1170 | abort(); | |
45aba42f | 1171 | } |
45aba42f KW |
1172 | } |
1173 | ||
8b220eb7 PB |
1174 | int coroutine_fn qcow2_write_caches(BlockDriverState *bs) |
1175 | { | |
1176 | BDRVQcow2State *s = bs->opaque; | |
1177 | int ret; | |
f7d0fe02 | 1178 | |
8b220eb7 PB |
1179 | ret = qcow2_cache_write(bs, s->l2_table_cache); |
1180 | if (ret < 0) { | |
1181 | return ret; | |
1182 | } | |
1183 | ||
1184 | if (qcow2_need_accurate_refcounts(s)) { | |
1185 | ret = qcow2_cache_write(bs, s->refcount_block_cache); | |
1186 | if (ret < 0) { | |
1187 | return ret; | |
1188 | } | |
1189 | } | |
1190 | ||
1191 | return 0; | |
1192 | } | |
1193 | ||
1194 | int coroutine_fn qcow2_flush_caches(BlockDriverState *bs) | |
1195 | { | |
1196 | int ret = qcow2_write_caches(bs); | |
1197 | if (ret < 0) { | |
1198 | return ret; | |
1199 | } | |
1200 | ||
1201 | return bdrv_flush(bs->file->bs); | |
1202 | } | |
f7d0fe02 KW |
1203 | |
1204 | /*********************************************************/ | |
1205 | /* snapshots and image creation */ | |
1206 | ||
1207 | ||
1208 | ||
f7d0fe02 | 1209 | /* update the refcounts of snapshots and the copied flag */ |
ed6ccf0f KW |
1210 | int qcow2_update_snapshot_refcount(BlockDriverState *bs, |
1211 | int64_t l1_table_offset, int l1_size, int addend) | |
f7d0fe02 | 1212 | { |
ff99129a | 1213 | BDRVQcow2State *s = bs->opaque; |
83ad165b | 1214 | uint64_t *l1_table, *l2_slice, l2_offset, entry, l1_size2, refcount; |
de82815d | 1215 | bool l1_allocated = false; |
b32cbae1 | 1216 | int64_t old_entry, old_l2_offset; |
83ad165b | 1217 | unsigned slice, slice_size2, n_slices; |
7324c10f | 1218 | int i, j, l1_modified = 0, nb_csectors; |
29c1a730 | 1219 | int ret; |
f7d0fe02 | 1220 | |
2aabe7c7 HR |
1221 | assert(addend >= -1 && addend <= 1); |
1222 | ||
83ad165b | 1223 | l2_slice = NULL; |
f7d0fe02 KW |
1224 | l1_table = NULL; |
1225 | l1_size2 = l1_size * sizeof(uint64_t); | |
83ad165b AG |
1226 | slice_size2 = s->l2_slice_size * sizeof(uint64_t); |
1227 | n_slices = s->cluster_size / slice_size2; | |
43a0cac4 | 1228 | |
0b919fae KW |
1229 | s->cache_discards = true; |
1230 | ||
43a0cac4 KW |
1231 | /* WARNING: qcow2_snapshot_goto relies on this function not using the |
1232 | * l1_table_offset when it is the current s->l1_table_offset! Be careful | |
1233 | * when changing this! */ | |
f7d0fe02 | 1234 | if (l1_table_offset != s->l1_table_offset) { |
9e029689 | 1235 | l1_table = g_try_malloc0(ROUND_UP(l1_size2, 512)); |
de82815d KW |
1236 | if (l1_size2 && l1_table == NULL) { |
1237 | ret = -ENOMEM; | |
1238 | goto fail; | |
1239 | } | |
1240 | l1_allocated = true; | |
c2bc78b6 | 1241 | |
cf2ab8fc | 1242 | ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2); |
c2bc78b6 | 1243 | if (ret < 0) { |
f7d0fe02 | 1244 | goto fail; |
93913dfd KW |
1245 | } |
1246 | ||
b32cbae1 | 1247 | for (i = 0; i < l1_size; i++) { |
f7d0fe02 | 1248 | be64_to_cpus(&l1_table[i]); |
b32cbae1 | 1249 | } |
f7d0fe02 KW |
1250 | } else { |
1251 | assert(l1_size == s->l1_size); | |
1252 | l1_table = s->l1_table; | |
de82815d | 1253 | l1_allocated = false; |
f7d0fe02 KW |
1254 | } |
1255 | ||
b32cbae1 | 1256 | for (i = 0; i < l1_size; i++) { |
f7d0fe02 KW |
1257 | l2_offset = l1_table[i]; |
1258 | if (l2_offset) { | |
1259 | old_l2_offset = l2_offset; | |
8e37f681 | 1260 | l2_offset &= L1E_OFFSET_MASK; |
29c1a730 | 1261 | |
a97c67ee HR |
1262 | if (offset_into_cluster(s, l2_offset)) { |
1263 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
1264 | PRIx64 " unaligned (L1 index: %#x)", | |
1265 | l2_offset, i); | |
1266 | ret = -EIO; | |
1267 | goto fail; | |
1268 | } | |
1269 | ||
83ad165b | 1270 | for (slice = 0; slice < n_slices; slice++) { |
ca62dd5c | 1271 | ret = qcow2_cache_get(bs, s->l2_table_cache, |
83ad165b AG |
1272 | l2_offset + slice * slice_size2, |
1273 | (void **) &l2_slice); | |
ca62dd5c AG |
1274 | if (ret < 0) { |
1275 | goto fail; | |
1276 | } | |
29c1a730 | 1277 | |
83ad165b | 1278 | for (j = 0; j < s->l2_slice_size; j++) { |
ca62dd5c AG |
1279 | uint64_t cluster_index; |
1280 | uint64_t offset; | |
1281 | ||
83ad165b | 1282 | entry = be64_to_cpu(l2_slice[j]); |
ca62dd5c AG |
1283 | old_entry = entry; |
1284 | entry &= ~QCOW_OFLAG_COPIED; | |
1285 | offset = entry & L2E_OFFSET_MASK; | |
1286 | ||
1287 | switch (qcow2_get_cluster_type(entry)) { | |
1288 | case QCOW2_CLUSTER_COMPRESSED: | |
1289 | nb_csectors = ((entry >> s->csize_shift) & | |
1290 | s->csize_mask) + 1; | |
1291 | if (addend != 0) { | |
1292 | ret = update_refcount( | |
1293 | bs, (entry & s->cluster_offset_mask) & ~511, | |
2aabe7c7 | 1294 | nb_csectors * 512, abs(addend), addend < 0, |
6cfcb9b8 | 1295 | QCOW2_DISCARD_SNAPSHOT); |
ca62dd5c AG |
1296 | if (ret < 0) { |
1297 | goto fail; | |
1298 | } | |
1299 | } | |
1300 | /* compressed clusters are never modified */ | |
1301 | refcount = 2; | |
1302 | break; | |
1303 | ||
1304 | case QCOW2_CLUSTER_NORMAL: | |
1305 | case QCOW2_CLUSTER_ZERO_ALLOC: | |
1306 | if (offset_into_cluster(s, offset)) { | |
83ad165b AG |
1307 | /* Here l2_index means table (not slice) index */ |
1308 | int l2_index = slice * s->l2_slice_size + j; | |
ca62dd5c AG |
1309 | qcow2_signal_corruption( |
1310 | bs, true, -1, -1, "Cluster " | |
1311 | "allocation offset %#" PRIx64 | |
1312 | " unaligned (L2 offset: %#" | |
1313 | PRIx64 ", L2 index: %#x)", | |
83ad165b | 1314 | offset, l2_offset, l2_index); |
ca62dd5c | 1315 | ret = -EIO; |
a97c67ee HR |
1316 | goto fail; |
1317 | } | |
1318 | ||
ca62dd5c AG |
1319 | cluster_index = offset >> s->cluster_bits; |
1320 | assert(cluster_index); | |
1321 | if (addend != 0) { | |
1322 | ret = qcow2_update_cluster_refcount( | |
1323 | bs, cluster_index, abs(addend), addend < 0, | |
1324 | QCOW2_DISCARD_SNAPSHOT); | |
1325 | if (ret < 0) { | |
1326 | goto fail; | |
1327 | } | |
1328 | } | |
1329 | ||
1330 | ret = qcow2_get_refcount(bs, cluster_index, &refcount); | |
7324c10f | 1331 | if (ret < 0) { |
018faafd KW |
1332 | goto fail; |
1333 | } | |
ca62dd5c | 1334 | break; |
bbd995d8 | 1335 | |
ca62dd5c AG |
1336 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1337 | case QCOW2_CLUSTER_UNALLOCATED: | |
1338 | refcount = 0; | |
1339 | break; | |
8b81a7b6 | 1340 | |
ca62dd5c AG |
1341 | default: |
1342 | abort(); | |
1343 | } | |
8b81a7b6 | 1344 | |
ca62dd5c AG |
1345 | if (refcount == 1) { |
1346 | entry |= QCOW_OFLAG_COPIED; | |
1347 | } | |
1348 | if (entry != old_entry) { | |
1349 | if (addend > 0) { | |
1350 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
1351 | s->refcount_block_cache); | |
1352 | } | |
83ad165b | 1353 | l2_slice[j] = cpu_to_be64(entry); |
ca62dd5c | 1354 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, |
83ad165b | 1355 | l2_slice); |
f7d0fe02 KW |
1356 | } |
1357 | } | |
29c1a730 | 1358 | |
83ad165b | 1359 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
ca62dd5c | 1360 | } |
29c1a730 | 1361 | |
f7d0fe02 | 1362 | if (addend != 0) { |
c6e9d8ae HR |
1363 | ret = qcow2_update_cluster_refcount(bs, l2_offset >> |
1364 | s->cluster_bits, | |
2aabe7c7 | 1365 | abs(addend), addend < 0, |
c6e9d8ae HR |
1366 | QCOW2_DISCARD_SNAPSHOT); |
1367 | if (ret < 0) { | |
1368 | goto fail; | |
1369 | } | |
f7d0fe02 | 1370 | } |
7324c10f HR |
1371 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1372 | &refcount); | |
1373 | if (ret < 0) { | |
018faafd KW |
1374 | goto fail; |
1375 | } else if (refcount == 1) { | |
f7d0fe02 KW |
1376 | l2_offset |= QCOW_OFLAG_COPIED; |
1377 | } | |
1378 | if (l2_offset != old_l2_offset) { | |
1379 | l1_table[i] = l2_offset; | |
1380 | l1_modified = 1; | |
1381 | } | |
1382 | } | |
1383 | } | |
93913dfd | 1384 | |
2154f24e | 1385 | ret = bdrv_flush(bs); |
93913dfd | 1386 | fail: |
83ad165b AG |
1387 | if (l2_slice) { |
1388 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
93913dfd KW |
1389 | } |
1390 | ||
0b919fae KW |
1391 | s->cache_discards = false; |
1392 | qcow2_process_discards(bs, ret); | |
1393 | ||
43a0cac4 | 1394 | /* Update L1 only if it isn't deleted anyway (addend = -1) */ |
c2b6ff51 KW |
1395 | if (ret == 0 && addend >= 0 && l1_modified) { |
1396 | for (i = 0; i < l1_size; i++) { | |
f7d0fe02 | 1397 | cpu_to_be64s(&l1_table[i]); |
c2b6ff51 KW |
1398 | } |
1399 | ||
d9ca2ea2 | 1400 | ret = bdrv_pwrite_sync(bs->file, l1_table_offset, |
9a4f4c31 | 1401 | l1_table, l1_size2); |
c2b6ff51 KW |
1402 | |
1403 | for (i = 0; i < l1_size; i++) { | |
f7d0fe02 | 1404 | be64_to_cpus(&l1_table[i]); |
c2b6ff51 | 1405 | } |
f7d0fe02 KW |
1406 | } |
1407 | if (l1_allocated) | |
7267c094 | 1408 | g_free(l1_table); |
93913dfd | 1409 | return ret; |
f7d0fe02 KW |
1410 | } |
1411 | ||
1412 | ||
1413 | ||
1414 | ||
1415 | /*********************************************************/ | |
1416 | /* refcount checking functions */ | |
1417 | ||
1418 | ||
c2551b47 | 1419 | static uint64_t refcount_array_byte_size(BDRVQcow2State *s, uint64_t entries) |
5fee192e HR |
1420 | { |
1421 | /* This assertion holds because there is no way we can address more than | |
1422 | * 2^(64 - 9) clusters at once (with cluster size 512 = 2^9, and because | |
1423 | * offsets have to be representable in bytes); due to every cluster | |
1424 | * corresponding to one refcount entry, we are well below that limit */ | |
1425 | assert(entries < (UINT64_C(1) << (64 - 9))); | |
1426 | ||
1427 | /* Thanks to the assertion this will not overflow, because | |
1428 | * s->refcount_order < 7. | |
1429 | * (note: x << s->refcount_order == x * s->refcount_bits) */ | |
1430 | return DIV_ROUND_UP(entries << s->refcount_order, 8); | |
1431 | } | |
1432 | ||
1433 | /** | |
1434 | * Reallocates *array so that it can hold new_size entries. *size must contain | |
1435 | * the current number of entries in *array. If the reallocation fails, *array | |
1436 | * and *size will not be modified and -errno will be returned. If the | |
1437 | * reallocation is successful, *array will be set to the new buffer, *size | |
1438 | * will be set to new_size and 0 will be returned. The size of the reallocated | |
1439 | * refcount array buffer will be aligned to a cluster boundary, and the newly | |
1440 | * allocated area will be zeroed. | |
1441 | */ | |
ff99129a | 1442 | static int realloc_refcount_array(BDRVQcow2State *s, void **array, |
5fee192e HR |
1443 | int64_t *size, int64_t new_size) |
1444 | { | |
b6d36def | 1445 | int64_t old_byte_size, new_byte_size; |
7453c96b | 1446 | void *new_ptr; |
5fee192e HR |
1447 | |
1448 | /* Round to clusters so the array can be directly written to disk */ | |
1449 | old_byte_size = size_to_clusters(s, refcount_array_byte_size(s, *size)) | |
1450 | * s->cluster_size; | |
1451 | new_byte_size = size_to_clusters(s, refcount_array_byte_size(s, new_size)) | |
1452 | * s->cluster_size; | |
1453 | ||
1454 | if (new_byte_size == old_byte_size) { | |
1455 | *size = new_size; | |
1456 | return 0; | |
1457 | } | |
1458 | ||
1459 | assert(new_byte_size > 0); | |
1460 | ||
b6d36def HR |
1461 | if (new_byte_size > SIZE_MAX) { |
1462 | return -ENOMEM; | |
1463 | } | |
1464 | ||
5fee192e HR |
1465 | new_ptr = g_try_realloc(*array, new_byte_size); |
1466 | if (!new_ptr) { | |
1467 | return -ENOMEM; | |
1468 | } | |
1469 | ||
1470 | if (new_byte_size > old_byte_size) { | |
b6d36def | 1471 | memset((char *)new_ptr + old_byte_size, 0, |
5fee192e HR |
1472 | new_byte_size - old_byte_size); |
1473 | } | |
1474 | ||
1475 | *array = new_ptr; | |
1476 | *size = new_size; | |
1477 | ||
1478 | return 0; | |
1479 | } | |
f7d0fe02 KW |
1480 | |
1481 | /* | |
1482 | * Increases the refcount for a range of clusters in a given refcount table. | |
1483 | * This is used to construct a temporary refcount table out of L1 and L2 tables | |
b6af0975 | 1484 | * which can be compared to the refcount table saved in the image. |
f7d0fe02 | 1485 | * |
9ac228e0 | 1486 | * Modifies the number of errors in res. |
f7d0fe02 | 1487 | */ |
8a5bb1f1 VSO |
1488 | int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res, |
1489 | void **refcount_table, | |
1490 | int64_t *refcount_table_size, | |
1491 | int64_t offset, int64_t size) | |
f7d0fe02 | 1492 | { |
ff99129a | 1493 | BDRVQcow2State *s = bs->opaque; |
7453c96b | 1494 | uint64_t start, last, cluster_offset, k, refcount; |
5fee192e | 1495 | int ret; |
f7d0fe02 | 1496 | |
fef4d3d5 HR |
1497 | if (size <= 0) { |
1498 | return 0; | |
1499 | } | |
f7d0fe02 | 1500 | |
ac95acdb HT |
1501 | start = start_of_cluster(s, offset); |
1502 | last = start_of_cluster(s, offset + size - 1); | |
f7d0fe02 KW |
1503 | for(cluster_offset = start; cluster_offset <= last; |
1504 | cluster_offset += s->cluster_size) { | |
1505 | k = cluster_offset >> s->cluster_bits; | |
641bb63c | 1506 | if (k >= *refcount_table_size) { |
5fee192e HR |
1507 | ret = realloc_refcount_array(s, refcount_table, |
1508 | refcount_table_size, k + 1); | |
1509 | if (ret < 0) { | |
641bb63c | 1510 | res->check_errors++; |
5fee192e | 1511 | return ret; |
f7d0fe02 | 1512 | } |
641bb63c HR |
1513 | } |
1514 | ||
7453c96b HR |
1515 | refcount = s->get_refcount(*refcount_table, k); |
1516 | if (refcount == s->refcount_max) { | |
641bb63c HR |
1517 | fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64 |
1518 | "\n", cluster_offset); | |
03bb78ed HR |
1519 | fprintf(stderr, "Use qemu-img amend to increase the refcount entry " |
1520 | "width or qemu-img convert to create a clean copy if the " | |
1521 | "image cannot be opened for writing\n"); | |
641bb63c | 1522 | res->corruptions++; |
7453c96b | 1523 | continue; |
f7d0fe02 | 1524 | } |
7453c96b | 1525 | s->set_refcount(*refcount_table, k, refcount + 1); |
f7d0fe02 | 1526 | } |
fef4d3d5 HR |
1527 | |
1528 | return 0; | |
f7d0fe02 KW |
1529 | } |
1530 | ||
801f7044 SH |
1531 | /* Flags for check_refcounts_l1() and check_refcounts_l2() */ |
1532 | enum { | |
fba31bae | 1533 | CHECK_FRAG_INFO = 0x2, /* update BlockFragInfo counters */ |
801f7044 SH |
1534 | }; |
1535 | ||
f7d0fe02 KW |
1536 | /* |
1537 | * Increases the refcount in the given refcount table for the all clusters | |
1538 | * referenced in the L2 table. While doing so, performs some checks on L2 | |
1539 | * entries. | |
1540 | * | |
1541 | * Returns the number of errors found by the checks or -errno if an internal | |
1542 | * error occurred. | |
1543 | */ | |
9ac228e0 | 1544 | static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res, |
7453c96b HR |
1545 | void **refcount_table, |
1546 | int64_t *refcount_table_size, int64_t l2_offset, | |
ac5b787a | 1547 | int flags, BdrvCheckMode fix) |
f7d0fe02 | 1548 | { |
ff99129a | 1549 | BDRVQcow2State *s = bs->opaque; |
afdf0abe | 1550 | uint64_t *l2_table, l2_entry; |
fba31bae | 1551 | uint64_t next_contiguous_offset = 0; |
ad27390c | 1552 | int i, l2_size, nb_csectors, ret; |
f7d0fe02 KW |
1553 | |
1554 | /* Read L2 table from disk */ | |
1555 | l2_size = s->l2_size * sizeof(uint64_t); | |
7267c094 | 1556 | l2_table = g_malloc(l2_size); |
f7d0fe02 | 1557 | |
cf2ab8fc | 1558 | ret = bdrv_pread(bs->file, l2_offset, l2_table, l2_size); |
ad27390c HR |
1559 | if (ret < 0) { |
1560 | fprintf(stderr, "ERROR: I/O error in check_refcounts_l2\n"); | |
1561 | res->check_errors++; | |
f7d0fe02 | 1562 | goto fail; |
ad27390c | 1563 | } |
f7d0fe02 KW |
1564 | |
1565 | /* Do the actual checks */ | |
1566 | for(i = 0; i < s->l2_size; i++) { | |
afdf0abe KW |
1567 | l2_entry = be64_to_cpu(l2_table[i]); |
1568 | ||
1569 | switch (qcow2_get_cluster_type(l2_entry)) { | |
1570 | case QCOW2_CLUSTER_COMPRESSED: | |
1571 | /* Compressed clusters don't have QCOW_OFLAG_COPIED */ | |
1572 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
1573 | fprintf(stderr, "ERROR: cluster %" PRId64 ": " | |
1574 | "copied flag must never be set for compressed " | |
1575 | "clusters\n", l2_entry >> s->cluster_bits); | |
1576 | l2_entry &= ~QCOW_OFLAG_COPIED; | |
1577 | res->corruptions++; | |
1578 | } | |
f7d0fe02 | 1579 | |
afdf0abe KW |
1580 | /* Mark cluster as used */ |
1581 | nb_csectors = ((l2_entry >> s->csize_shift) & | |
1582 | s->csize_mask) + 1; | |
1583 | l2_entry &= s->cluster_offset_mask; | |
8a5bb1f1 VSO |
1584 | ret = qcow2_inc_refcounts_imrt(bs, res, |
1585 | refcount_table, refcount_table_size, | |
1586 | l2_entry & ~511, nb_csectors * 512); | |
fef4d3d5 HR |
1587 | if (ret < 0) { |
1588 | goto fail; | |
1589 | } | |
fba31bae SH |
1590 | |
1591 | if (flags & CHECK_FRAG_INFO) { | |
1592 | res->bfi.allocated_clusters++; | |
4db35162 | 1593 | res->bfi.compressed_clusters++; |
fba31bae SH |
1594 | |
1595 | /* Compressed clusters are fragmented by nature. Since they | |
1596 | * take up sub-sector space but we only have sector granularity | |
1597 | * I/O we need to re-read the same sectors even for adjacent | |
1598 | * compressed clusters. | |
1599 | */ | |
1600 | res->bfi.fragmented_clusters++; | |
1601 | } | |
afdf0abe | 1602 | break; |
f7d0fe02 | 1603 | |
fdfab37d | 1604 | case QCOW2_CLUSTER_ZERO_ALLOC: |
afdf0abe KW |
1605 | case QCOW2_CLUSTER_NORMAL: |
1606 | { | |
afdf0abe | 1607 | uint64_t offset = l2_entry & L2E_OFFSET_MASK; |
f7d0fe02 | 1608 | |
fba31bae SH |
1609 | if (flags & CHECK_FRAG_INFO) { |
1610 | res->bfi.allocated_clusters++; | |
1611 | if (next_contiguous_offset && | |
1612 | offset != next_contiguous_offset) { | |
1613 | res->bfi.fragmented_clusters++; | |
1614 | } | |
1615 | next_contiguous_offset = offset + s->cluster_size; | |
1616 | } | |
1617 | ||
ac5b787a HR |
1618 | /* Correct offsets are cluster aligned */ |
1619 | if (offset_into_cluster(s, offset)) { | |
1620 | if (qcow2_get_cluster_type(l2_entry) == | |
1621 | QCOW2_CLUSTER_ZERO_ALLOC) | |
1622 | { | |
1623 | fprintf(stderr, "%s offset=%" PRIx64 ": Preallocated zero " | |
1624 | "cluster is not properly aligned; L2 entry " | |
1625 | "corrupted.\n", | |
1626 | fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", | |
1627 | offset); | |
1628 | if (fix & BDRV_FIX_ERRORS) { | |
1629 | uint64_t l2e_offset = | |
1630 | l2_offset + (uint64_t)i * sizeof(uint64_t); | |
1631 | ||
1632 | l2_entry = QCOW_OFLAG_ZERO; | |
1633 | l2_table[i] = cpu_to_be64(l2_entry); | |
1634 | ret = qcow2_pre_write_overlap_check(bs, | |
1635 | QCOW2_OL_ACTIVE_L2 | QCOW2_OL_INACTIVE_L2, | |
1636 | l2e_offset, sizeof(uint64_t)); | |
1637 | if (ret < 0) { | |
1638 | fprintf(stderr, "ERROR: Overlap check failed\n"); | |
1639 | res->check_errors++; | |
1640 | /* Something is seriously wrong, so abort checking | |
1641 | * this L2 table */ | |
1642 | goto fail; | |
1643 | } | |
1644 | ||
1645 | ret = bdrv_pwrite_sync(bs->file, l2e_offset, | |
1646 | &l2_table[i], sizeof(uint64_t)); | |
1647 | if (ret < 0) { | |
1648 | fprintf(stderr, "ERROR: Failed to overwrite L2 " | |
1649 | "table entry: %s\n", strerror(-ret)); | |
1650 | res->check_errors++; | |
1651 | /* Do not abort, continue checking the rest of this | |
1652 | * L2 table's entries */ | |
1653 | } else { | |
1654 | res->corruptions_fixed++; | |
1655 | /* Skip marking the cluster as used | |
1656 | * (it is unused now) */ | |
1657 | continue; | |
1658 | } | |
1659 | } else { | |
1660 | res->corruptions++; | |
1661 | } | |
1662 | } else { | |
1663 | fprintf(stderr, "ERROR offset=%" PRIx64 ": Data cluster is " | |
1664 | "not properly aligned; L2 entry corrupted.\n", offset); | |
1665 | res->corruptions++; | |
1666 | } | |
1667 | } | |
1668 | ||
afdf0abe | 1669 | /* Mark cluster as used */ |
8a5bb1f1 VSO |
1670 | ret = qcow2_inc_refcounts_imrt(bs, res, |
1671 | refcount_table, refcount_table_size, | |
1672 | offset, s->cluster_size); | |
fef4d3d5 HR |
1673 | if (ret < 0) { |
1674 | goto fail; | |
1675 | } | |
afdf0abe KW |
1676 | break; |
1677 | } | |
1678 | ||
fdfab37d | 1679 | case QCOW2_CLUSTER_ZERO_PLAIN: |
afdf0abe KW |
1680 | case QCOW2_CLUSTER_UNALLOCATED: |
1681 | break; | |
1682 | ||
1683 | default: | |
1684 | abort(); | |
f7d0fe02 KW |
1685 | } |
1686 | } | |
1687 | ||
7267c094 | 1688 | g_free(l2_table); |
9ac228e0 | 1689 | return 0; |
f7d0fe02 KW |
1690 | |
1691 | fail: | |
7267c094 | 1692 | g_free(l2_table); |
ad27390c | 1693 | return ret; |
f7d0fe02 KW |
1694 | } |
1695 | ||
1696 | /* | |
1697 | * Increases the refcount for the L1 table, its L2 tables and all referenced | |
1698 | * clusters in the given refcount table. While doing so, performs some checks | |
1699 | * on L1 and L2 entries. | |
1700 | * | |
1701 | * Returns the number of errors found by the checks or -errno if an internal | |
1702 | * error occurred. | |
1703 | */ | |
1704 | static int check_refcounts_l1(BlockDriverState *bs, | |
9ac228e0 | 1705 | BdrvCheckResult *res, |
7453c96b | 1706 | void **refcount_table, |
641bb63c | 1707 | int64_t *refcount_table_size, |
f7d0fe02 | 1708 | int64_t l1_table_offset, int l1_size, |
ac5b787a | 1709 | int flags, BdrvCheckMode fix) |
f7d0fe02 | 1710 | { |
ff99129a | 1711 | BDRVQcow2State *s = bs->opaque; |
fef4d3d5 | 1712 | uint64_t *l1_table = NULL, l2_offset, l1_size2; |
4f6ed88c | 1713 | int i, ret; |
f7d0fe02 KW |
1714 | |
1715 | l1_size2 = l1_size * sizeof(uint64_t); | |
1716 | ||
1717 | /* Mark L1 table as used */ | |
8a5bb1f1 VSO |
1718 | ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, refcount_table_size, |
1719 | l1_table_offset, l1_size2); | |
fef4d3d5 HR |
1720 | if (ret < 0) { |
1721 | goto fail; | |
1722 | } | |
f7d0fe02 KW |
1723 | |
1724 | /* Read L1 table entries from disk */ | |
fef4d3d5 | 1725 | if (l1_size2 > 0) { |
de82815d KW |
1726 | l1_table = g_try_malloc(l1_size2); |
1727 | if (l1_table == NULL) { | |
1728 | ret = -ENOMEM; | |
ad27390c | 1729 | res->check_errors++; |
de82815d KW |
1730 | goto fail; |
1731 | } | |
cf2ab8fc | 1732 | ret = bdrv_pread(bs->file, l1_table_offset, l1_table, l1_size2); |
ad27390c HR |
1733 | if (ret < 0) { |
1734 | fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n"); | |
1735 | res->check_errors++; | |
702ef63f | 1736 | goto fail; |
ad27390c | 1737 | } |
702ef63f KW |
1738 | for(i = 0;i < l1_size; i++) |
1739 | be64_to_cpus(&l1_table[i]); | |
1740 | } | |
f7d0fe02 KW |
1741 | |
1742 | /* Do the actual checks */ | |
1743 | for(i = 0; i < l1_size; i++) { | |
1744 | l2_offset = l1_table[i]; | |
1745 | if (l2_offset) { | |
f7d0fe02 | 1746 | /* Mark L2 table as used */ |
afdf0abe | 1747 | l2_offset &= L1E_OFFSET_MASK; |
8a5bb1f1 VSO |
1748 | ret = qcow2_inc_refcounts_imrt(bs, res, |
1749 | refcount_table, refcount_table_size, | |
1750 | l2_offset, s->cluster_size); | |
fef4d3d5 HR |
1751 | if (ret < 0) { |
1752 | goto fail; | |
1753 | } | |
f7d0fe02 KW |
1754 | |
1755 | /* L2 tables are cluster aligned */ | |
ac95acdb | 1756 | if (offset_into_cluster(s, l2_offset)) { |
f7d0fe02 KW |
1757 | fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not " |
1758 | "cluster aligned; L1 entry corrupted\n", l2_offset); | |
9ac228e0 | 1759 | res->corruptions++; |
f7d0fe02 KW |
1760 | } |
1761 | ||
1762 | /* Process and check L2 entries */ | |
9ac228e0 | 1763 | ret = check_refcounts_l2(bs, res, refcount_table, |
ac5b787a HR |
1764 | refcount_table_size, l2_offset, flags, |
1765 | fix); | |
f7d0fe02 KW |
1766 | if (ret < 0) { |
1767 | goto fail; | |
1768 | } | |
f7d0fe02 KW |
1769 | } |
1770 | } | |
7267c094 | 1771 | g_free(l1_table); |
9ac228e0 | 1772 | return 0; |
f7d0fe02 KW |
1773 | |
1774 | fail: | |
7267c094 | 1775 | g_free(l1_table); |
ad27390c | 1776 | return ret; |
f7d0fe02 KW |
1777 | } |
1778 | ||
4f6ed88c HR |
1779 | /* |
1780 | * Checks the OFLAG_COPIED flag for all L1 and L2 entries. | |
1781 | * | |
1782 | * This function does not print an error message nor does it increment | |
44751917 HR |
1783 | * check_errors if qcow2_get_refcount fails (this is because such an error will |
1784 | * have been already detected and sufficiently signaled by the calling function | |
4f6ed88c HR |
1785 | * (qcow2_check_refcounts) by the time this function is called). |
1786 | */ | |
e23e400e HR |
1787 | static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res, |
1788 | BdrvCheckMode fix) | |
4f6ed88c | 1789 | { |
ff99129a | 1790 | BDRVQcow2State *s = bs->opaque; |
4f6ed88c HR |
1791 | uint64_t *l2_table = qemu_blockalign(bs, s->cluster_size); |
1792 | int ret; | |
0e06528e | 1793 | uint64_t refcount; |
4f6ed88c HR |
1794 | int i, j; |
1795 | ||
1796 | for (i = 0; i < s->l1_size; i++) { | |
1797 | uint64_t l1_entry = s->l1_table[i]; | |
1798 | uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK; | |
e23e400e | 1799 | bool l2_dirty = false; |
4f6ed88c HR |
1800 | |
1801 | if (!l2_offset) { | |
1802 | continue; | |
1803 | } | |
1804 | ||
7324c10f HR |
1805 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1806 | &refcount); | |
1807 | if (ret < 0) { | |
4f6ed88c HR |
1808 | /* don't print message nor increment check_errors */ |
1809 | continue; | |
1810 | } | |
1811 | if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) { | |
e23e400e | 1812 | fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_index=%d " |
0e06528e | 1813 | "l1_entry=%" PRIx64 " refcount=%" PRIu64 "\n", |
e23e400e HR |
1814 | fix & BDRV_FIX_ERRORS ? "Repairing" : |
1815 | "ERROR", | |
4f6ed88c | 1816 | i, l1_entry, refcount); |
e23e400e HR |
1817 | if (fix & BDRV_FIX_ERRORS) { |
1818 | s->l1_table[i] = refcount == 1 | |
1819 | ? l1_entry | QCOW_OFLAG_COPIED | |
1820 | : l1_entry & ~QCOW_OFLAG_COPIED; | |
1821 | ret = qcow2_write_l1_entry(bs, i); | |
1822 | if (ret < 0) { | |
1823 | res->check_errors++; | |
1824 | goto fail; | |
1825 | } | |
1826 | res->corruptions_fixed++; | |
1827 | } else { | |
1828 | res->corruptions++; | |
1829 | } | |
4f6ed88c HR |
1830 | } |
1831 | ||
cf2ab8fc | 1832 | ret = bdrv_pread(bs->file, l2_offset, l2_table, |
4f6ed88c HR |
1833 | s->l2_size * sizeof(uint64_t)); |
1834 | if (ret < 0) { | |
1835 | fprintf(stderr, "ERROR: Could not read L2 table: %s\n", | |
1836 | strerror(-ret)); | |
1837 | res->check_errors++; | |
1838 | goto fail; | |
1839 | } | |
1840 | ||
1841 | for (j = 0; j < s->l2_size; j++) { | |
1842 | uint64_t l2_entry = be64_to_cpu(l2_table[j]); | |
1843 | uint64_t data_offset = l2_entry & L2E_OFFSET_MASK; | |
3ef95218 | 1844 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
4f6ed88c | 1845 | |
fdfab37d EB |
1846 | if (cluster_type == QCOW2_CLUSTER_NORMAL || |
1847 | cluster_type == QCOW2_CLUSTER_ZERO_ALLOC) { | |
7324c10f HR |
1848 | ret = qcow2_get_refcount(bs, |
1849 | data_offset >> s->cluster_bits, | |
1850 | &refcount); | |
1851 | if (ret < 0) { | |
4f6ed88c HR |
1852 | /* don't print message nor increment check_errors */ |
1853 | continue; | |
1854 | } | |
1855 | if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) { | |
e23e400e | 1856 | fprintf(stderr, "%s OFLAG_COPIED data cluster: " |
0e06528e | 1857 | "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n", |
e23e400e HR |
1858 | fix & BDRV_FIX_ERRORS ? "Repairing" : |
1859 | "ERROR", | |
4f6ed88c | 1860 | l2_entry, refcount); |
e23e400e HR |
1861 | if (fix & BDRV_FIX_ERRORS) { |
1862 | l2_table[j] = cpu_to_be64(refcount == 1 | |
1863 | ? l2_entry | QCOW_OFLAG_COPIED | |
1864 | : l2_entry & ~QCOW_OFLAG_COPIED); | |
1865 | l2_dirty = true; | |
1866 | res->corruptions_fixed++; | |
1867 | } else { | |
1868 | res->corruptions++; | |
1869 | } | |
4f6ed88c HR |
1870 | } |
1871 | } | |
1872 | } | |
e23e400e HR |
1873 | |
1874 | if (l2_dirty) { | |
231bb267 HR |
1875 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2, |
1876 | l2_offset, s->cluster_size); | |
e23e400e HR |
1877 | if (ret < 0) { |
1878 | fprintf(stderr, "ERROR: Could not write L2 table; metadata " | |
1879 | "overlap check failed: %s\n", strerror(-ret)); | |
1880 | res->check_errors++; | |
1881 | goto fail; | |
1882 | } | |
1883 | ||
d9ca2ea2 | 1884 | ret = bdrv_pwrite(bs->file, l2_offset, l2_table, |
9a4f4c31 | 1885 | s->cluster_size); |
e23e400e HR |
1886 | if (ret < 0) { |
1887 | fprintf(stderr, "ERROR: Could not write L2 table: %s\n", | |
1888 | strerror(-ret)); | |
1889 | res->check_errors++; | |
1890 | goto fail; | |
1891 | } | |
1892 | } | |
4f6ed88c HR |
1893 | } |
1894 | ||
1895 | ret = 0; | |
1896 | ||
1897 | fail: | |
1898 | qemu_vfree(l2_table); | |
1899 | return ret; | |
1900 | } | |
1901 | ||
6ca56bf5 HR |
1902 | /* |
1903 | * Checks consistency of refblocks and accounts for each refblock in | |
1904 | * *refcount_table. | |
1905 | */ | |
1906 | static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res, | |
f307b255 | 1907 | BdrvCheckMode fix, bool *rebuild, |
7453c96b | 1908 | void **refcount_table, int64_t *nb_clusters) |
6ca56bf5 | 1909 | { |
ff99129a | 1910 | BDRVQcow2State *s = bs->opaque; |
001c158d | 1911 | int64_t i, size; |
fef4d3d5 | 1912 | int ret; |
6ca56bf5 | 1913 | |
f7d0fe02 | 1914 | for(i = 0; i < s->refcount_table_size; i++) { |
6882c8fa | 1915 | uint64_t offset, cluster; |
f7d0fe02 | 1916 | offset = s->refcount_table[i]; |
6882c8fa | 1917 | cluster = offset >> s->cluster_bits; |
746c3cb5 KW |
1918 | |
1919 | /* Refcount blocks are cluster aligned */ | |
ac95acdb | 1920 | if (offset_into_cluster(s, offset)) { |
166acf54 | 1921 | fprintf(stderr, "ERROR refcount block %" PRId64 " is not " |
746c3cb5 | 1922 | "cluster aligned; refcount table entry corrupted\n", i); |
9ac228e0 | 1923 | res->corruptions++; |
f307b255 | 1924 | *rebuild = true; |
6882c8fa KW |
1925 | continue; |
1926 | } | |
1927 | ||
6ca56bf5 | 1928 | if (cluster >= *nb_clusters) { |
001c158d HR |
1929 | fprintf(stderr, "%s refcount block %" PRId64 " is outside image\n", |
1930 | fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i); | |
1931 | ||
1932 | if (fix & BDRV_FIX_ERRORS) { | |
5fee192e | 1933 | int64_t new_nb_clusters; |
ed3d2ec9 | 1934 | Error *local_err = NULL; |
001c158d HR |
1935 | |
1936 | if (offset > INT64_MAX - s->cluster_size) { | |
1937 | ret = -EINVAL; | |
1938 | goto resize_fail; | |
1939 | } | |
1940 | ||
ed3d2ec9 | 1941 | ret = bdrv_truncate(bs->file, offset + s->cluster_size, |
7ea37c30 | 1942 | PREALLOC_MODE_OFF, &local_err); |
001c158d | 1943 | if (ret < 0) { |
ed3d2ec9 | 1944 | error_report_err(local_err); |
001c158d HR |
1945 | goto resize_fail; |
1946 | } | |
9a4f4c31 | 1947 | size = bdrv_getlength(bs->file->bs); |
001c158d HR |
1948 | if (size < 0) { |
1949 | ret = size; | |
1950 | goto resize_fail; | |
1951 | } | |
1952 | ||
5fee192e HR |
1953 | new_nb_clusters = size_to_clusters(s, size); |
1954 | assert(new_nb_clusters >= *nb_clusters); | |
001c158d | 1955 | |
5fee192e HR |
1956 | ret = realloc_refcount_array(s, refcount_table, |
1957 | nb_clusters, new_nb_clusters); | |
1958 | if (ret < 0) { | |
001c158d | 1959 | res->check_errors++; |
5fee192e | 1960 | return ret; |
001c158d | 1961 | } |
001c158d HR |
1962 | |
1963 | if (cluster >= *nb_clusters) { | |
1964 | ret = -EINVAL; | |
1965 | goto resize_fail; | |
1966 | } | |
1967 | ||
1968 | res->corruptions_fixed++; | |
8a5bb1f1 VSO |
1969 | ret = qcow2_inc_refcounts_imrt(bs, res, |
1970 | refcount_table, nb_clusters, | |
1971 | offset, s->cluster_size); | |
001c158d HR |
1972 | if (ret < 0) { |
1973 | return ret; | |
1974 | } | |
1975 | /* No need to check whether the refcount is now greater than 1: | |
1976 | * This area was just allocated and zeroed, so it can only be | |
8a5bb1f1 | 1977 | * exactly 1 after qcow2_inc_refcounts_imrt() */ |
001c158d HR |
1978 | continue; |
1979 | ||
1980 | resize_fail: | |
1981 | res->corruptions++; | |
f307b255 | 1982 | *rebuild = true; |
001c158d HR |
1983 | fprintf(stderr, "ERROR could not resize image: %s\n", |
1984 | strerror(-ret)); | |
1985 | } else { | |
1986 | res->corruptions++; | |
1987 | } | |
6882c8fa | 1988 | continue; |
746c3cb5 KW |
1989 | } |
1990 | ||
f7d0fe02 | 1991 | if (offset != 0) { |
8a5bb1f1 VSO |
1992 | ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, |
1993 | offset, s->cluster_size); | |
fef4d3d5 HR |
1994 | if (ret < 0) { |
1995 | return ret; | |
1996 | } | |
7453c96b | 1997 | if (s->get_refcount(*refcount_table, cluster) != 1) { |
f307b255 | 1998 | fprintf(stderr, "ERROR refcount block %" PRId64 |
7453c96b HR |
1999 | " refcount=%" PRIu64 "\n", i, |
2000 | s->get_refcount(*refcount_table, cluster)); | |
f307b255 HR |
2001 | res->corruptions++; |
2002 | *rebuild = true; | |
746c3cb5 | 2003 | } |
f7d0fe02 KW |
2004 | } |
2005 | } | |
2006 | ||
6ca56bf5 HR |
2007 | return 0; |
2008 | } | |
2009 | ||
057a3fe5 HR |
2010 | /* |
2011 | * Calculates an in-memory refcount table. | |
2012 | */ | |
2013 | static int calculate_refcounts(BlockDriverState *bs, BdrvCheckResult *res, | |
f307b255 | 2014 | BdrvCheckMode fix, bool *rebuild, |
7453c96b | 2015 | void **refcount_table, int64_t *nb_clusters) |
057a3fe5 | 2016 | { |
ff99129a | 2017 | BDRVQcow2State *s = bs->opaque; |
057a3fe5 HR |
2018 | int64_t i; |
2019 | QCowSnapshot *sn; | |
2020 | int ret; | |
2021 | ||
9696df21 | 2022 | if (!*refcount_table) { |
5fee192e HR |
2023 | int64_t old_size = 0; |
2024 | ret = realloc_refcount_array(s, refcount_table, | |
2025 | &old_size, *nb_clusters); | |
2026 | if (ret < 0) { | |
9696df21 | 2027 | res->check_errors++; |
5fee192e | 2028 | return ret; |
9696df21 | 2029 | } |
057a3fe5 HR |
2030 | } |
2031 | ||
2032 | /* header */ | |
8a5bb1f1 VSO |
2033 | ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, |
2034 | 0, s->cluster_size); | |
fef4d3d5 HR |
2035 | if (ret < 0) { |
2036 | return ret; | |
2037 | } | |
057a3fe5 HR |
2038 | |
2039 | /* current L1 table */ | |
641bb63c | 2040 | ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, |
ac5b787a HR |
2041 | s->l1_table_offset, s->l1_size, CHECK_FRAG_INFO, |
2042 | fix); | |
057a3fe5 HR |
2043 | if (ret < 0) { |
2044 | return ret; | |
2045 | } | |
2046 | ||
2047 | /* snapshots */ | |
2048 | for (i = 0; i < s->nb_snapshots; i++) { | |
2049 | sn = s->snapshots + i; | |
641bb63c | 2050 | ret = check_refcounts_l1(bs, res, refcount_table, nb_clusters, |
ac5b787a | 2051 | sn->l1_table_offset, sn->l1_size, 0, fix); |
057a3fe5 HR |
2052 | if (ret < 0) { |
2053 | return ret; | |
2054 | } | |
2055 | } | |
8a5bb1f1 VSO |
2056 | ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, |
2057 | s->snapshots_offset, s->snapshots_size); | |
fef4d3d5 HR |
2058 | if (ret < 0) { |
2059 | return ret; | |
2060 | } | |
057a3fe5 HR |
2061 | |
2062 | /* refcount data */ | |
8a5bb1f1 VSO |
2063 | ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, |
2064 | s->refcount_table_offset, | |
2065 | s->refcount_table_size * sizeof(uint64_t)); | |
fef4d3d5 HR |
2066 | if (ret < 0) { |
2067 | return ret; | |
2068 | } | |
057a3fe5 | 2069 | |
4652b8f3 DB |
2070 | /* encryption */ |
2071 | if (s->crypto_header.length) { | |
8a5bb1f1 VSO |
2072 | ret = qcow2_inc_refcounts_imrt(bs, res, refcount_table, nb_clusters, |
2073 | s->crypto_header.offset, | |
2074 | s->crypto_header.length); | |
4652b8f3 DB |
2075 | if (ret < 0) { |
2076 | return ret; | |
2077 | } | |
2078 | } | |
2079 | ||
88ddffae VSO |
2080 | /* bitmaps */ |
2081 | ret = qcow2_check_bitmaps_refcounts(bs, res, refcount_table, nb_clusters); | |
2082 | if (ret < 0) { | |
2083 | return ret; | |
2084 | } | |
2085 | ||
f307b255 | 2086 | return check_refblocks(bs, res, fix, rebuild, refcount_table, nb_clusters); |
057a3fe5 HR |
2087 | } |
2088 | ||
6ca56bf5 HR |
2089 | /* |
2090 | * Compares the actual reference count for each cluster in the image against the | |
2091 | * refcount as reported by the refcount structures on-disk. | |
2092 | */ | |
2093 | static void compare_refcounts(BlockDriverState *bs, BdrvCheckResult *res, | |
f307b255 HR |
2094 | BdrvCheckMode fix, bool *rebuild, |
2095 | int64_t *highest_cluster, | |
7453c96b | 2096 | void *refcount_table, int64_t nb_clusters) |
6ca56bf5 | 2097 | { |
ff99129a | 2098 | BDRVQcow2State *s = bs->opaque; |
6ca56bf5 | 2099 | int64_t i; |
0e06528e | 2100 | uint64_t refcount1, refcount2; |
7324c10f | 2101 | int ret; |
6ca56bf5 HR |
2102 | |
2103 | for (i = 0, *highest_cluster = 0; i < nb_clusters; i++) { | |
7324c10f HR |
2104 | ret = qcow2_get_refcount(bs, i, &refcount1); |
2105 | if (ret < 0) { | |
166acf54 | 2106 | fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n", |
7324c10f | 2107 | i, strerror(-ret)); |
9ac228e0 | 2108 | res->check_errors++; |
f74550fd | 2109 | continue; |
018faafd KW |
2110 | } |
2111 | ||
7453c96b | 2112 | refcount2 = s->get_refcount(refcount_table, i); |
c6bb9ad1 FS |
2113 | |
2114 | if (refcount1 > 0 || refcount2 > 0) { | |
6ca56bf5 | 2115 | *highest_cluster = i; |
c6bb9ad1 FS |
2116 | } |
2117 | ||
f7d0fe02 | 2118 | if (refcount1 != refcount2) { |
166acf54 KW |
2119 | /* Check if we're allowed to fix the mismatch */ |
2120 | int *num_fixed = NULL; | |
f307b255 HR |
2121 | if (refcount1 == 0) { |
2122 | *rebuild = true; | |
2123 | } else if (refcount1 > refcount2 && (fix & BDRV_FIX_LEAKS)) { | |
166acf54 KW |
2124 | num_fixed = &res->leaks_fixed; |
2125 | } else if (refcount1 < refcount2 && (fix & BDRV_FIX_ERRORS)) { | |
2126 | num_fixed = &res->corruptions_fixed; | |
2127 | } | |
2128 | ||
0e06528e HR |
2129 | fprintf(stderr, "%s cluster %" PRId64 " refcount=%" PRIu64 |
2130 | " reference=%" PRIu64 "\n", | |
166acf54 KW |
2131 | num_fixed != NULL ? "Repairing" : |
2132 | refcount1 < refcount2 ? "ERROR" : | |
2133 | "Leaked", | |
f7d0fe02 | 2134 | i, refcount1, refcount2); |
166acf54 KW |
2135 | |
2136 | if (num_fixed) { | |
2137 | ret = update_refcount(bs, i << s->cluster_bits, 1, | |
2aabe7c7 HR |
2138 | refcount_diff(refcount1, refcount2), |
2139 | refcount1 > refcount2, | |
6cfcb9b8 | 2140 | QCOW2_DISCARD_ALWAYS); |
166acf54 KW |
2141 | if (ret >= 0) { |
2142 | (*num_fixed)++; | |
2143 | continue; | |
2144 | } | |
2145 | } | |
2146 | ||
2147 | /* And if we couldn't, print an error */ | |
9ac228e0 KW |
2148 | if (refcount1 < refcount2) { |
2149 | res->corruptions++; | |
2150 | } else { | |
2151 | res->leaks++; | |
2152 | } | |
f7d0fe02 KW |
2153 | } |
2154 | } | |
6ca56bf5 HR |
2155 | } |
2156 | ||
c7c0681b HR |
2157 | /* |
2158 | * Allocates clusters using an in-memory refcount table (IMRT) in contrast to | |
2159 | * the on-disk refcount structures. | |
2160 | * | |
2161 | * On input, *first_free_cluster tells where to start looking, and need not | |
2162 | * actually be a free cluster; the returned offset will not be before that | |
2163 | * cluster. On output, *first_free_cluster points to the first gap found, even | |
2164 | * if that gap was too small to be used as the returned offset. | |
2165 | * | |
2166 | * Note that *first_free_cluster is a cluster index whereas the return value is | |
2167 | * an offset. | |
2168 | */ | |
2169 | static int64_t alloc_clusters_imrt(BlockDriverState *bs, | |
2170 | int cluster_count, | |
7453c96b | 2171 | void **refcount_table, |
c7c0681b HR |
2172 | int64_t *imrt_nb_clusters, |
2173 | int64_t *first_free_cluster) | |
2174 | { | |
ff99129a | 2175 | BDRVQcow2State *s = bs->opaque; |
c7c0681b HR |
2176 | int64_t cluster = *first_free_cluster, i; |
2177 | bool first_gap = true; | |
2178 | int contiguous_free_clusters; | |
5fee192e | 2179 | int ret; |
c7c0681b HR |
2180 | |
2181 | /* Starting at *first_free_cluster, find a range of at least cluster_count | |
2182 | * continuously free clusters */ | |
2183 | for (contiguous_free_clusters = 0; | |
2184 | cluster < *imrt_nb_clusters && | |
2185 | contiguous_free_clusters < cluster_count; | |
2186 | cluster++) | |
2187 | { | |
7453c96b | 2188 | if (!s->get_refcount(*refcount_table, cluster)) { |
c7c0681b HR |
2189 | contiguous_free_clusters++; |
2190 | if (first_gap) { | |
2191 | /* If this is the first free cluster found, update | |
2192 | * *first_free_cluster accordingly */ | |
2193 | *first_free_cluster = cluster; | |
2194 | first_gap = false; | |
2195 | } | |
2196 | } else if (contiguous_free_clusters) { | |
2197 | contiguous_free_clusters = 0; | |
2198 | } | |
2199 | } | |
2200 | ||
2201 | /* If contiguous_free_clusters is greater than zero, it contains the number | |
2202 | * of continuously free clusters until the current cluster; the first free | |
2203 | * cluster in the current "gap" is therefore | |
2204 | * cluster - contiguous_free_clusters */ | |
2205 | ||
2206 | /* If no such range could be found, grow the in-memory refcount table | |
2207 | * accordingly to append free clusters at the end of the image */ | |
2208 | if (contiguous_free_clusters < cluster_count) { | |
c7c0681b HR |
2209 | /* contiguous_free_clusters clusters are already empty at the image end; |
2210 | * we need cluster_count clusters; therefore, we have to allocate | |
2211 | * cluster_count - contiguous_free_clusters new clusters at the end of | |
2212 | * the image (which is the current value of cluster; note that cluster | |
2213 | * may exceed old_imrt_nb_clusters if *first_free_cluster pointed beyond | |
2214 | * the image end) */ | |
5fee192e HR |
2215 | ret = realloc_refcount_array(s, refcount_table, imrt_nb_clusters, |
2216 | cluster + cluster_count | |
2217 | - contiguous_free_clusters); | |
2218 | if (ret < 0) { | |
2219 | return ret; | |
c7c0681b | 2220 | } |
c7c0681b HR |
2221 | } |
2222 | ||
2223 | /* Go back to the first free cluster */ | |
2224 | cluster -= contiguous_free_clusters; | |
2225 | for (i = 0; i < cluster_count; i++) { | |
7453c96b | 2226 | s->set_refcount(*refcount_table, cluster + i, 1); |
c7c0681b HR |
2227 | } |
2228 | ||
2229 | return cluster << s->cluster_bits; | |
2230 | } | |
2231 | ||
2232 | /* | |
2233 | * Creates a new refcount structure based solely on the in-memory information | |
2234 | * given through *refcount_table. All necessary allocations will be reflected | |
2235 | * in that array. | |
2236 | * | |
2237 | * On success, the old refcount structure is leaked (it will be covered by the | |
2238 | * new refcount structure). | |
2239 | */ | |
2240 | static int rebuild_refcount_structure(BlockDriverState *bs, | |
2241 | BdrvCheckResult *res, | |
7453c96b | 2242 | void **refcount_table, |
c7c0681b HR |
2243 | int64_t *nb_clusters) |
2244 | { | |
ff99129a | 2245 | BDRVQcow2State *s = bs->opaque; |
c7c0681b HR |
2246 | int64_t first_free_cluster = 0, reftable_offset = -1, cluster = 0; |
2247 | int64_t refblock_offset, refblock_start, refblock_index; | |
2248 | uint32_t reftable_size = 0; | |
2249 | uint64_t *on_disk_reftable = NULL; | |
7453c96b HR |
2250 | void *on_disk_refblock; |
2251 | int ret = 0; | |
c7c0681b HR |
2252 | struct { |
2253 | uint64_t reftable_offset; | |
2254 | uint32_t reftable_clusters; | |
2255 | } QEMU_PACKED reftable_offset_and_clusters; | |
2256 | ||
2257 | qcow2_cache_empty(bs, s->refcount_block_cache); | |
2258 | ||
2259 | write_refblocks: | |
2260 | for (; cluster < *nb_clusters; cluster++) { | |
7453c96b | 2261 | if (!s->get_refcount(*refcount_table, cluster)) { |
c7c0681b HR |
2262 | continue; |
2263 | } | |
2264 | ||
2265 | refblock_index = cluster >> s->refcount_block_bits; | |
2266 | refblock_start = refblock_index << s->refcount_block_bits; | |
2267 | ||
2268 | /* Don't allocate a cluster in a refblock already written to disk */ | |
2269 | if (first_free_cluster < refblock_start) { | |
2270 | first_free_cluster = refblock_start; | |
2271 | } | |
2272 | refblock_offset = alloc_clusters_imrt(bs, 1, refcount_table, | |
2273 | nb_clusters, &first_free_cluster); | |
2274 | if (refblock_offset < 0) { | |
2275 | fprintf(stderr, "ERROR allocating refblock: %s\n", | |
2276 | strerror(-refblock_offset)); | |
2277 | res->check_errors++; | |
2278 | ret = refblock_offset; | |
2279 | goto fail; | |
2280 | } | |
2281 | ||
2282 | if (reftable_size <= refblock_index) { | |
2283 | uint32_t old_reftable_size = reftable_size; | |
2284 | uint64_t *new_on_disk_reftable; | |
2285 | ||
2286 | reftable_size = ROUND_UP((refblock_index + 1) * sizeof(uint64_t), | |
2287 | s->cluster_size) / sizeof(uint64_t); | |
2288 | new_on_disk_reftable = g_try_realloc(on_disk_reftable, | |
2289 | reftable_size * | |
2290 | sizeof(uint64_t)); | |
2291 | if (!new_on_disk_reftable) { | |
2292 | res->check_errors++; | |
2293 | ret = -ENOMEM; | |
2294 | goto fail; | |
2295 | } | |
2296 | on_disk_reftable = new_on_disk_reftable; | |
2297 | ||
2298 | memset(on_disk_reftable + old_reftable_size, 0, | |
2299 | (reftable_size - old_reftable_size) * sizeof(uint64_t)); | |
2300 | ||
2301 | /* The offset we have for the reftable is now no longer valid; | |
2302 | * this will leak that range, but we can easily fix that by running | |
2303 | * a leak-fixing check after this rebuild operation */ | |
2304 | reftable_offset = -1; | |
f80ac75d PMD |
2305 | } else { |
2306 | assert(on_disk_reftable); | |
c7c0681b HR |
2307 | } |
2308 | on_disk_reftable[refblock_index] = refblock_offset; | |
2309 | ||
2310 | /* If this is apparently the last refblock (for now), try to squeeze the | |
2311 | * reftable in */ | |
2312 | if (refblock_index == (*nb_clusters - 1) >> s->refcount_block_bits && | |
2313 | reftable_offset < 0) | |
2314 | { | |
2315 | uint64_t reftable_clusters = size_to_clusters(s, reftable_size * | |
2316 | sizeof(uint64_t)); | |
2317 | reftable_offset = alloc_clusters_imrt(bs, reftable_clusters, | |
2318 | refcount_table, nb_clusters, | |
2319 | &first_free_cluster); | |
2320 | if (reftable_offset < 0) { | |
2321 | fprintf(stderr, "ERROR allocating reftable: %s\n", | |
2322 | strerror(-reftable_offset)); | |
2323 | res->check_errors++; | |
2324 | ret = reftable_offset; | |
2325 | goto fail; | |
2326 | } | |
2327 | } | |
2328 | ||
2329 | ret = qcow2_pre_write_overlap_check(bs, 0, refblock_offset, | |
2330 | s->cluster_size); | |
2331 | if (ret < 0) { | |
2332 | fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret)); | |
2333 | goto fail; | |
2334 | } | |
2335 | ||
7453c96b HR |
2336 | /* The size of *refcount_table is always cluster-aligned, therefore the |
2337 | * write operation will not overflow */ | |
2338 | on_disk_refblock = (void *)((char *) *refcount_table + | |
2339 | refblock_index * s->cluster_size); | |
c7c0681b | 2340 | |
18d51c4b | 2341 | ret = bdrv_write(bs->file, refblock_offset / BDRV_SECTOR_SIZE, |
7453c96b | 2342 | on_disk_refblock, s->cluster_sectors); |
c7c0681b HR |
2343 | if (ret < 0) { |
2344 | fprintf(stderr, "ERROR writing refblock: %s\n", strerror(-ret)); | |
2345 | goto fail; | |
2346 | } | |
2347 | ||
2348 | /* Go to the end of this refblock */ | |
2349 | cluster = refblock_start + s->refcount_block_size - 1; | |
2350 | } | |
2351 | ||
2352 | if (reftable_offset < 0) { | |
2353 | uint64_t post_refblock_start, reftable_clusters; | |
2354 | ||
2355 | post_refblock_start = ROUND_UP(*nb_clusters, s->refcount_block_size); | |
2356 | reftable_clusters = size_to_clusters(s, | |
2357 | reftable_size * sizeof(uint64_t)); | |
2358 | /* Not pretty but simple */ | |
2359 | if (first_free_cluster < post_refblock_start) { | |
2360 | first_free_cluster = post_refblock_start; | |
2361 | } | |
2362 | reftable_offset = alloc_clusters_imrt(bs, reftable_clusters, | |
2363 | refcount_table, nb_clusters, | |
2364 | &first_free_cluster); | |
2365 | if (reftable_offset < 0) { | |
2366 | fprintf(stderr, "ERROR allocating reftable: %s\n", | |
2367 | strerror(-reftable_offset)); | |
2368 | res->check_errors++; | |
2369 | ret = reftable_offset; | |
2370 | goto fail; | |
2371 | } | |
2372 | ||
2373 | goto write_refblocks; | |
2374 | } | |
2375 | ||
c7c0681b HR |
2376 | for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) { |
2377 | cpu_to_be64s(&on_disk_reftable[refblock_index]); | |
2378 | } | |
2379 | ||
2380 | ret = qcow2_pre_write_overlap_check(bs, 0, reftable_offset, | |
2381 | reftable_size * sizeof(uint64_t)); | |
2382 | if (ret < 0) { | |
2383 | fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret)); | |
2384 | goto fail; | |
2385 | } | |
2386 | ||
2387 | assert(reftable_size < INT_MAX / sizeof(uint64_t)); | |
d9ca2ea2 | 2388 | ret = bdrv_pwrite(bs->file, reftable_offset, on_disk_reftable, |
c7c0681b HR |
2389 | reftable_size * sizeof(uint64_t)); |
2390 | if (ret < 0) { | |
2391 | fprintf(stderr, "ERROR writing reftable: %s\n", strerror(-ret)); | |
2392 | goto fail; | |
2393 | } | |
2394 | ||
2395 | /* Enter new reftable into the image header */ | |
f1f7a1dd PM |
2396 | reftable_offset_and_clusters.reftable_offset = cpu_to_be64(reftable_offset); |
2397 | reftable_offset_and_clusters.reftable_clusters = | |
2398 | cpu_to_be32(size_to_clusters(s, reftable_size * sizeof(uint64_t))); | |
d9ca2ea2 KW |
2399 | ret = bdrv_pwrite_sync(bs->file, |
2400 | offsetof(QCowHeader, refcount_table_offset), | |
c7c0681b HR |
2401 | &reftable_offset_and_clusters, |
2402 | sizeof(reftable_offset_and_clusters)); | |
2403 | if (ret < 0) { | |
2404 | fprintf(stderr, "ERROR setting reftable: %s\n", strerror(-ret)); | |
2405 | goto fail; | |
2406 | } | |
2407 | ||
2408 | for (refblock_index = 0; refblock_index < reftable_size; refblock_index++) { | |
2409 | be64_to_cpus(&on_disk_reftable[refblock_index]); | |
2410 | } | |
2411 | s->refcount_table = on_disk_reftable; | |
2412 | s->refcount_table_offset = reftable_offset; | |
2413 | s->refcount_table_size = reftable_size; | |
7061a078 | 2414 | update_max_refcount_table_index(s); |
c7c0681b HR |
2415 | |
2416 | return 0; | |
2417 | ||
2418 | fail: | |
2419 | g_free(on_disk_reftable); | |
2420 | return ret; | |
2421 | } | |
2422 | ||
6ca56bf5 HR |
2423 | /* |
2424 | * Checks an image for refcount consistency. | |
2425 | * | |
2426 | * Returns 0 if no errors are found, the number of errors in case the image is | |
2427 | * detected as corrupted, and -errno when an internal error occurred. | |
2428 | */ | |
2429 | int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res, | |
2430 | BdrvCheckMode fix) | |
2431 | { | |
ff99129a | 2432 | BDRVQcow2State *s = bs->opaque; |
c7c0681b | 2433 | BdrvCheckResult pre_compare_res; |
6ca56bf5 | 2434 | int64_t size, highest_cluster, nb_clusters; |
7453c96b | 2435 | void *refcount_table = NULL; |
f307b255 | 2436 | bool rebuild = false; |
6ca56bf5 HR |
2437 | int ret; |
2438 | ||
9a4f4c31 | 2439 | size = bdrv_getlength(bs->file->bs); |
6ca56bf5 HR |
2440 | if (size < 0) { |
2441 | res->check_errors++; | |
2442 | return size; | |
2443 | } | |
2444 | ||
2445 | nb_clusters = size_to_clusters(s, size); | |
2446 | if (nb_clusters > INT_MAX) { | |
2447 | res->check_errors++; | |
2448 | return -EFBIG; | |
2449 | } | |
2450 | ||
2451 | res->bfi.total_clusters = | |
2452 | size_to_clusters(s, bs->total_sectors * BDRV_SECTOR_SIZE); | |
2453 | ||
f307b255 HR |
2454 | ret = calculate_refcounts(bs, res, fix, &rebuild, &refcount_table, |
2455 | &nb_clusters); | |
6ca56bf5 HR |
2456 | if (ret < 0) { |
2457 | goto fail; | |
2458 | } | |
2459 | ||
c7c0681b HR |
2460 | /* In case we don't need to rebuild the refcount structure (but want to fix |
2461 | * something), this function is immediately called again, in which case the | |
2462 | * result should be ignored */ | |
2463 | pre_compare_res = *res; | |
2464 | compare_refcounts(bs, res, 0, &rebuild, &highest_cluster, refcount_table, | |
6ca56bf5 | 2465 | nb_clusters); |
f7d0fe02 | 2466 | |
c7c0681b | 2467 | if (rebuild && (fix & BDRV_FIX_ERRORS)) { |
791230d8 HR |
2468 | BdrvCheckResult old_res = *res; |
2469 | int fresh_leaks = 0; | |
2470 | ||
c7c0681b HR |
2471 | fprintf(stderr, "Rebuilding refcount structure\n"); |
2472 | ret = rebuild_refcount_structure(bs, res, &refcount_table, | |
2473 | &nb_clusters); | |
2474 | if (ret < 0) { | |
2475 | goto fail; | |
2476 | } | |
791230d8 HR |
2477 | |
2478 | res->corruptions = 0; | |
2479 | res->leaks = 0; | |
2480 | ||
2481 | /* Because the old reftable has been exchanged for a new one the | |
2482 | * references have to be recalculated */ | |
2483 | rebuild = false; | |
7453c96b | 2484 | memset(refcount_table, 0, refcount_array_byte_size(s, nb_clusters)); |
791230d8 HR |
2485 | ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table, |
2486 | &nb_clusters); | |
2487 | if (ret < 0) { | |
2488 | goto fail; | |
2489 | } | |
2490 | ||
2491 | if (fix & BDRV_FIX_LEAKS) { | |
2492 | /* The old refcount structures are now leaked, fix it; the result | |
2493 | * can be ignored, aside from leaks which were introduced by | |
2494 | * rebuild_refcount_structure() that could not be fixed */ | |
2495 | BdrvCheckResult saved_res = *res; | |
2496 | *res = (BdrvCheckResult){ 0 }; | |
2497 | ||
2498 | compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild, | |
2499 | &highest_cluster, refcount_table, nb_clusters); | |
2500 | if (rebuild) { | |
2501 | fprintf(stderr, "ERROR rebuilt refcount structure is still " | |
2502 | "broken\n"); | |
2503 | } | |
2504 | ||
2505 | /* Any leaks accounted for here were introduced by | |
2506 | * rebuild_refcount_structure() because that function has created a | |
2507 | * new refcount structure from scratch */ | |
2508 | fresh_leaks = res->leaks; | |
2509 | *res = saved_res; | |
2510 | } | |
2511 | ||
2512 | if (res->corruptions < old_res.corruptions) { | |
2513 | res->corruptions_fixed += old_res.corruptions - res->corruptions; | |
2514 | } | |
2515 | if (res->leaks < old_res.leaks) { | |
2516 | res->leaks_fixed += old_res.leaks - res->leaks; | |
2517 | } | |
2518 | res->leaks += fresh_leaks; | |
c7c0681b HR |
2519 | } else if (fix) { |
2520 | if (rebuild) { | |
2521 | fprintf(stderr, "ERROR need to rebuild refcount structures\n"); | |
2522 | res->check_errors++; | |
2523 | ret = -EIO; | |
2524 | goto fail; | |
2525 | } | |
2526 | ||
2527 | if (res->leaks || res->corruptions) { | |
2528 | *res = pre_compare_res; | |
2529 | compare_refcounts(bs, res, fix, &rebuild, &highest_cluster, | |
2530 | refcount_table, nb_clusters); | |
2531 | } | |
f307b255 HR |
2532 | } |
2533 | ||
4f6ed88c | 2534 | /* check OFLAG_COPIED */ |
e23e400e | 2535 | ret = check_oflag_copied(bs, res, fix); |
4f6ed88c HR |
2536 | if (ret < 0) { |
2537 | goto fail; | |
2538 | } | |
2539 | ||
c6bb9ad1 | 2540 | res->image_end_offset = (highest_cluster + 1) * s->cluster_size; |
80fa3341 KW |
2541 | ret = 0; |
2542 | ||
2543 | fail: | |
7267c094 | 2544 | g_free(refcount_table); |
f7d0fe02 | 2545 | |
80fa3341 | 2546 | return ret; |
f7d0fe02 KW |
2547 | } |
2548 | ||
a40f1c2a HR |
2549 | #define overlaps_with(ofs, sz) \ |
2550 | ranges_overlap(offset, size, ofs, sz) | |
2551 | ||
2552 | /* | |
2553 | * Checks if the given offset into the image file is actually free to use by | |
2554 | * looking for overlaps with important metadata sections (L1/L2 tables etc.), | |
2555 | * i.e. a sanity check without relying on the refcount tables. | |
2556 | * | |
231bb267 HR |
2557 | * The ign parameter specifies what checks not to perform (being a bitmask of |
2558 | * QCow2MetadataOverlap values), i.e., what sections to ignore. | |
a40f1c2a HR |
2559 | * |
2560 | * Returns: | |
2561 | * - 0 if writing to this offset will not affect the mentioned metadata | |
2562 | * - a positive QCow2MetadataOverlap value indicating one overlapping section | |
2563 | * - a negative value (-errno) indicating an error while performing a check, | |
2564 | * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2 | |
2565 | */ | |
231bb267 | 2566 | int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset, |
a40f1c2a HR |
2567 | int64_t size) |
2568 | { | |
ff99129a | 2569 | BDRVQcow2State *s = bs->opaque; |
3e355390 | 2570 | int chk = s->overlap_check & ~ign; |
a40f1c2a HR |
2571 | int i, j; |
2572 | ||
2573 | if (!size) { | |
2574 | return 0; | |
2575 | } | |
2576 | ||
2577 | if (chk & QCOW2_OL_MAIN_HEADER) { | |
2578 | if (offset < s->cluster_size) { | |
2579 | return QCOW2_OL_MAIN_HEADER; | |
2580 | } | |
2581 | } | |
2582 | ||
2583 | /* align range to test to cluster boundaries */ | |
9e029689 | 2584 | size = ROUND_UP(offset_into_cluster(s, offset) + size, s->cluster_size); |
a40f1c2a HR |
2585 | offset = start_of_cluster(s, offset); |
2586 | ||
2587 | if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) { | |
2588 | if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) { | |
2589 | return QCOW2_OL_ACTIVE_L1; | |
2590 | } | |
2591 | } | |
2592 | ||
2593 | if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) { | |
2594 | if (overlaps_with(s->refcount_table_offset, | |
2595 | s->refcount_table_size * sizeof(uint64_t))) { | |
2596 | return QCOW2_OL_REFCOUNT_TABLE; | |
2597 | } | |
2598 | } | |
2599 | ||
2600 | if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) { | |
2601 | if (overlaps_with(s->snapshots_offset, s->snapshots_size)) { | |
2602 | return QCOW2_OL_SNAPSHOT_TABLE; | |
2603 | } | |
2604 | } | |
2605 | ||
2606 | if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) { | |
2607 | for (i = 0; i < s->nb_snapshots; i++) { | |
2608 | if (s->snapshots[i].l1_size && | |
2609 | overlaps_with(s->snapshots[i].l1_table_offset, | |
2610 | s->snapshots[i].l1_size * sizeof(uint64_t))) { | |
2611 | return QCOW2_OL_INACTIVE_L1; | |
2612 | } | |
2613 | } | |
2614 | } | |
2615 | ||
2616 | if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) { | |
2617 | for (i = 0; i < s->l1_size; i++) { | |
2618 | if ((s->l1_table[i] & L1E_OFFSET_MASK) && | |
2619 | overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK, | |
2620 | s->cluster_size)) { | |
2621 | return QCOW2_OL_ACTIVE_L2; | |
2622 | } | |
2623 | } | |
2624 | } | |
2625 | ||
2626 | if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) { | |
7061a078 AG |
2627 | unsigned last_entry = s->max_refcount_table_index; |
2628 | assert(last_entry < s->refcount_table_size); | |
2629 | assert(last_entry + 1 == s->refcount_table_size || | |
2630 | (s->refcount_table[last_entry + 1] & REFT_OFFSET_MASK) == 0); | |
2631 | for (i = 0; i <= last_entry; i++) { | |
a40f1c2a HR |
2632 | if ((s->refcount_table[i] & REFT_OFFSET_MASK) && |
2633 | overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK, | |
2634 | s->cluster_size)) { | |
2635 | return QCOW2_OL_REFCOUNT_BLOCK; | |
2636 | } | |
2637 | } | |
2638 | } | |
2639 | ||
2640 | if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) { | |
2641 | for (i = 0; i < s->nb_snapshots; i++) { | |
2642 | uint64_t l1_ofs = s->snapshots[i].l1_table_offset; | |
2643 | uint32_t l1_sz = s->snapshots[i].l1_size; | |
998b959c | 2644 | uint64_t l1_sz2 = l1_sz * sizeof(uint64_t); |
de82815d | 2645 | uint64_t *l1 = g_try_malloc(l1_sz2); |
a40f1c2a HR |
2646 | int ret; |
2647 | ||
de82815d KW |
2648 | if (l1_sz2 && l1 == NULL) { |
2649 | return -ENOMEM; | |
2650 | } | |
2651 | ||
cf2ab8fc | 2652 | ret = bdrv_pread(bs->file, l1_ofs, l1, l1_sz2); |
a40f1c2a HR |
2653 | if (ret < 0) { |
2654 | g_free(l1); | |
2655 | return ret; | |
2656 | } | |
2657 | ||
2658 | for (j = 0; j < l1_sz; j++) { | |
1e242b55 HR |
2659 | uint64_t l2_ofs = be64_to_cpu(l1[j]) & L1E_OFFSET_MASK; |
2660 | if (l2_ofs && overlaps_with(l2_ofs, s->cluster_size)) { | |
a40f1c2a HR |
2661 | g_free(l1); |
2662 | return QCOW2_OL_INACTIVE_L2; | |
2663 | } | |
2664 | } | |
2665 | ||
2666 | g_free(l1); | |
2667 | } | |
2668 | } | |
2669 | ||
2670 | return 0; | |
2671 | } | |
2672 | ||
2673 | static const char *metadata_ol_names[] = { | |
2674 | [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header", | |
2675 | [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table", | |
2676 | [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table", | |
2677 | [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table", | |
2678 | [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block", | |
2679 | [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table", | |
2680 | [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table", | |
2681 | [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table", | |
2682 | }; | |
2683 | ||
2684 | /* | |
2685 | * First performs a check for metadata overlaps (through | |
2686 | * qcow2_check_metadata_overlap); if that fails with a negative value (error | |
2687 | * while performing a check), that value is returned. If an impending overlap | |
2688 | * is detected, the BDS will be made unusable, the qcow2 file marked corrupt | |
2689 | * and -EIO returned. | |
2690 | * | |
2691 | * Returns 0 if there were neither overlaps nor errors while checking for | |
2692 | * overlaps; or a negative value (-errno) on error. | |
2693 | */ | |
231bb267 | 2694 | int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset, |
a40f1c2a HR |
2695 | int64_t size) |
2696 | { | |
231bb267 | 2697 | int ret = qcow2_check_metadata_overlap(bs, ign, offset, size); |
a40f1c2a HR |
2698 | |
2699 | if (ret < 0) { | |
2700 | return ret; | |
2701 | } else if (ret > 0) { | |
786a4ea8 | 2702 | int metadata_ol_bitnr = ctz32(ret); |
a40f1c2a HR |
2703 | assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR); |
2704 | ||
adb43552 HR |
2705 | qcow2_signal_corruption(bs, true, offset, size, "Preventing invalid " |
2706 | "write on metadata (overlaps with %s)", | |
2707 | metadata_ol_names[metadata_ol_bitnr]); | |
a40f1c2a HR |
2708 | return -EIO; |
2709 | } | |
2710 | ||
2711 | return 0; | |
2712 | } | |
791c9a00 HR |
2713 | |
2714 | /* A pointer to a function of this type is given to walk_over_reftable(). That | |
2715 | * function will create refblocks and pass them to a RefblockFinishOp once they | |
2716 | * are completed (@refblock). @refblock_empty is set if the refblock is | |
2717 | * completely empty. | |
2718 | * | |
2719 | * Along with the refblock, a corresponding reftable entry is passed, in the | |
2720 | * reftable @reftable (which may be reallocated) at @reftable_index. | |
2721 | * | |
2722 | * @allocated should be set to true if a new cluster has been allocated. | |
2723 | */ | |
2724 | typedef int (RefblockFinishOp)(BlockDriverState *bs, uint64_t **reftable, | |
2725 | uint64_t reftable_index, uint64_t *reftable_size, | |
2726 | void *refblock, bool refblock_empty, | |
2727 | bool *allocated, Error **errp); | |
2728 | ||
2729 | /** | |
2730 | * This "operation" for walk_over_reftable() allocates the refblock on disk (if | |
2731 | * it is not empty) and inserts its offset into the new reftable. The size of | |
2732 | * this new reftable is increased as required. | |
2733 | */ | |
2734 | static int alloc_refblock(BlockDriverState *bs, uint64_t **reftable, | |
2735 | uint64_t reftable_index, uint64_t *reftable_size, | |
2736 | void *refblock, bool refblock_empty, bool *allocated, | |
2737 | Error **errp) | |
2738 | { | |
2739 | BDRVQcow2State *s = bs->opaque; | |
2740 | int64_t offset; | |
2741 | ||
2742 | if (!refblock_empty && reftable_index >= *reftable_size) { | |
2743 | uint64_t *new_reftable; | |
2744 | uint64_t new_reftable_size; | |
2745 | ||
2746 | new_reftable_size = ROUND_UP(reftable_index + 1, | |
2747 | s->cluster_size / sizeof(uint64_t)); | |
2748 | if (new_reftable_size > QCOW_MAX_REFTABLE_SIZE / sizeof(uint64_t)) { | |
2749 | error_setg(errp, | |
2750 | "This operation would make the refcount table grow " | |
2751 | "beyond the maximum size supported by QEMU, aborting"); | |
2752 | return -ENOTSUP; | |
2753 | } | |
2754 | ||
2755 | new_reftable = g_try_realloc(*reftable, new_reftable_size * | |
2756 | sizeof(uint64_t)); | |
2757 | if (!new_reftable) { | |
2758 | error_setg(errp, "Failed to increase reftable buffer size"); | |
2759 | return -ENOMEM; | |
2760 | } | |
2761 | ||
2762 | memset(new_reftable + *reftable_size, 0, | |
2763 | (new_reftable_size - *reftable_size) * sizeof(uint64_t)); | |
2764 | ||
2765 | *reftable = new_reftable; | |
2766 | *reftable_size = new_reftable_size; | |
2767 | } | |
2768 | ||
2769 | if (!refblock_empty && !(*reftable)[reftable_index]) { | |
2770 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
2771 | if (offset < 0) { | |
2772 | error_setg_errno(errp, -offset, "Failed to allocate refblock"); | |
2773 | return offset; | |
2774 | } | |
2775 | (*reftable)[reftable_index] = offset; | |
2776 | *allocated = true; | |
2777 | } | |
2778 | ||
2779 | return 0; | |
2780 | } | |
2781 | ||
2782 | /** | |
2783 | * This "operation" for walk_over_reftable() writes the refblock to disk at the | |
2784 | * offset specified by the new reftable's entry. It does not modify the new | |
2785 | * reftable or change any refcounts. | |
2786 | */ | |
2787 | static int flush_refblock(BlockDriverState *bs, uint64_t **reftable, | |
2788 | uint64_t reftable_index, uint64_t *reftable_size, | |
2789 | void *refblock, bool refblock_empty, bool *allocated, | |
2790 | Error **errp) | |
2791 | { | |
2792 | BDRVQcow2State *s = bs->opaque; | |
2793 | int64_t offset; | |
2794 | int ret; | |
2795 | ||
2796 | if (reftable_index < *reftable_size && (*reftable)[reftable_index]) { | |
2797 | offset = (*reftable)[reftable_index]; | |
2798 | ||
2799 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size); | |
2800 | if (ret < 0) { | |
2801 | error_setg_errno(errp, -ret, "Overlap check failed"); | |
2802 | return ret; | |
2803 | } | |
2804 | ||
d9ca2ea2 | 2805 | ret = bdrv_pwrite(bs->file, offset, refblock, s->cluster_size); |
791c9a00 HR |
2806 | if (ret < 0) { |
2807 | error_setg_errno(errp, -ret, "Failed to write refblock"); | |
2808 | return ret; | |
2809 | } | |
2810 | } else { | |
2811 | assert(refblock_empty); | |
2812 | } | |
2813 | ||
2814 | return 0; | |
2815 | } | |
2816 | ||
2817 | /** | |
2818 | * This function walks over the existing reftable and every referenced refblock; | |
2819 | * if @new_set_refcount is non-NULL, it is called for every refcount entry to | |
2820 | * create an equal new entry in the passed @new_refblock. Once that | |
2821 | * @new_refblock is completely filled, @operation will be called. | |
2822 | * | |
2823 | * @status_cb and @cb_opaque are used for the amend operation's status callback. | |
2824 | * @index is the index of the walk_over_reftable() calls and @total is the total | |
2825 | * number of walk_over_reftable() calls per amend operation. Both are used for | |
2826 | * calculating the parameters for the status callback. | |
2827 | * | |
2828 | * @allocated is set to true if a new cluster has been allocated. | |
2829 | */ | |
2830 | static int walk_over_reftable(BlockDriverState *bs, uint64_t **new_reftable, | |
2831 | uint64_t *new_reftable_index, | |
2832 | uint64_t *new_reftable_size, | |
2833 | void *new_refblock, int new_refblock_size, | |
2834 | int new_refcount_bits, | |
2835 | RefblockFinishOp *operation, bool *allocated, | |
2836 | Qcow2SetRefcountFunc *new_set_refcount, | |
2837 | BlockDriverAmendStatusCB *status_cb, | |
2838 | void *cb_opaque, int index, int total, | |
2839 | Error **errp) | |
2840 | { | |
2841 | BDRVQcow2State *s = bs->opaque; | |
2842 | uint64_t reftable_index; | |
2843 | bool new_refblock_empty = true; | |
2844 | int refblock_index; | |
2845 | int new_refblock_index = 0; | |
2846 | int ret; | |
2847 | ||
2848 | for (reftable_index = 0; reftable_index < s->refcount_table_size; | |
2849 | reftable_index++) | |
2850 | { | |
2851 | uint64_t refblock_offset = s->refcount_table[reftable_index] | |
2852 | & REFT_OFFSET_MASK; | |
2853 | ||
2854 | status_cb(bs, (uint64_t)index * s->refcount_table_size + reftable_index, | |
2855 | (uint64_t)total * s->refcount_table_size, cb_opaque); | |
2856 | ||
2857 | if (refblock_offset) { | |
2858 | void *refblock; | |
2859 | ||
2860 | if (offset_into_cluster(s, refblock_offset)) { | |
2861 | qcow2_signal_corruption(bs, true, -1, -1, "Refblock offset %#" | |
2862 | PRIx64 " unaligned (reftable index: %#" | |
2863 | PRIx64 ")", refblock_offset, | |
2864 | reftable_index); | |
2865 | error_setg(errp, | |
2866 | "Image is corrupt (unaligned refblock offset)"); | |
2867 | return -EIO; | |
2868 | } | |
2869 | ||
2870 | ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offset, | |
2871 | &refblock); | |
2872 | if (ret < 0) { | |
2873 | error_setg_errno(errp, -ret, "Failed to retrieve refblock"); | |
2874 | return ret; | |
2875 | } | |
2876 | ||
2877 | for (refblock_index = 0; refblock_index < s->refcount_block_size; | |
2878 | refblock_index++) | |
2879 | { | |
2880 | uint64_t refcount; | |
2881 | ||
2882 | if (new_refblock_index >= new_refblock_size) { | |
2883 | /* new_refblock is now complete */ | |
2884 | ret = operation(bs, new_reftable, *new_reftable_index, | |
2885 | new_reftable_size, new_refblock, | |
2886 | new_refblock_empty, allocated, errp); | |
2887 | if (ret < 0) { | |
2013c3d4 | 2888 | qcow2_cache_put(s->refcount_block_cache, &refblock); |
791c9a00 HR |
2889 | return ret; |
2890 | } | |
2891 | ||
2892 | (*new_reftable_index)++; | |
2893 | new_refblock_index = 0; | |
2894 | new_refblock_empty = true; | |
2895 | } | |
2896 | ||
2897 | refcount = s->get_refcount(refblock, refblock_index); | |
2898 | if (new_refcount_bits < 64 && refcount >> new_refcount_bits) { | |
2899 | uint64_t offset; | |
2900 | ||
2013c3d4 | 2901 | qcow2_cache_put(s->refcount_block_cache, &refblock); |
791c9a00 HR |
2902 | |
2903 | offset = ((reftable_index << s->refcount_block_bits) | |
2904 | + refblock_index) << s->cluster_bits; | |
2905 | ||
2906 | error_setg(errp, "Cannot decrease refcount entry width to " | |
2907 | "%i bits: Cluster at offset %#" PRIx64 " has a " | |
2908 | "refcount of %" PRIu64, new_refcount_bits, | |
2909 | offset, refcount); | |
2910 | return -EINVAL; | |
2911 | } | |
2912 | ||
2913 | if (new_set_refcount) { | |
2914 | new_set_refcount(new_refblock, new_refblock_index++, | |
2915 | refcount); | |
2916 | } else { | |
2917 | new_refblock_index++; | |
2918 | } | |
2919 | new_refblock_empty = new_refblock_empty && refcount == 0; | |
2920 | } | |
2921 | ||
2013c3d4 | 2922 | qcow2_cache_put(s->refcount_block_cache, &refblock); |
791c9a00 HR |
2923 | } else { |
2924 | /* No refblock means every refcount is 0 */ | |
2925 | for (refblock_index = 0; refblock_index < s->refcount_block_size; | |
2926 | refblock_index++) | |
2927 | { | |
2928 | if (new_refblock_index >= new_refblock_size) { | |
2929 | /* new_refblock is now complete */ | |
2930 | ret = operation(bs, new_reftable, *new_reftable_index, | |
2931 | new_reftable_size, new_refblock, | |
2932 | new_refblock_empty, allocated, errp); | |
2933 | if (ret < 0) { | |
2934 | return ret; | |
2935 | } | |
2936 | ||
2937 | (*new_reftable_index)++; | |
2938 | new_refblock_index = 0; | |
2939 | new_refblock_empty = true; | |
2940 | } | |
2941 | ||
2942 | if (new_set_refcount) { | |
2943 | new_set_refcount(new_refblock, new_refblock_index++, 0); | |
2944 | } else { | |
2945 | new_refblock_index++; | |
2946 | } | |
2947 | } | |
2948 | } | |
2949 | } | |
2950 | ||
2951 | if (new_refblock_index > 0) { | |
2952 | /* Complete the potentially existing partially filled final refblock */ | |
2953 | if (new_set_refcount) { | |
2954 | for (; new_refblock_index < new_refblock_size; | |
2955 | new_refblock_index++) | |
2956 | { | |
2957 | new_set_refcount(new_refblock, new_refblock_index, 0); | |
2958 | } | |
2959 | } | |
2960 | ||
2961 | ret = operation(bs, new_reftable, *new_reftable_index, | |
2962 | new_reftable_size, new_refblock, new_refblock_empty, | |
2963 | allocated, errp); | |
2964 | if (ret < 0) { | |
2965 | return ret; | |
2966 | } | |
2967 | ||
2968 | (*new_reftable_index)++; | |
2969 | } | |
2970 | ||
2971 | status_cb(bs, (uint64_t)(index + 1) * s->refcount_table_size, | |
2972 | (uint64_t)total * s->refcount_table_size, cb_opaque); | |
2973 | ||
2974 | return 0; | |
2975 | } | |
2976 | ||
2977 | int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order, | |
2978 | BlockDriverAmendStatusCB *status_cb, | |
2979 | void *cb_opaque, Error **errp) | |
2980 | { | |
2981 | BDRVQcow2State *s = bs->opaque; | |
2982 | Qcow2GetRefcountFunc *new_get_refcount; | |
2983 | Qcow2SetRefcountFunc *new_set_refcount; | |
2984 | void *new_refblock = qemu_blockalign(bs->file->bs, s->cluster_size); | |
2985 | uint64_t *new_reftable = NULL, new_reftable_size = 0; | |
2986 | uint64_t *old_reftable, old_reftable_size, old_reftable_offset; | |
2987 | uint64_t new_reftable_index = 0; | |
2988 | uint64_t i; | |
2989 | int64_t new_reftable_offset = 0, allocated_reftable_size = 0; | |
2990 | int new_refblock_size, new_refcount_bits = 1 << refcount_order; | |
2991 | int old_refcount_order; | |
2992 | int walk_index = 0; | |
2993 | int ret; | |
2994 | bool new_allocation; | |
2995 | ||
2996 | assert(s->qcow_version >= 3); | |
2997 | assert(refcount_order >= 0 && refcount_order <= 6); | |
2998 | ||
2999 | /* see qcow2_open() */ | |
3000 | new_refblock_size = 1 << (s->cluster_bits - (refcount_order - 3)); | |
3001 | ||
3002 | new_get_refcount = get_refcount_funcs[refcount_order]; | |
3003 | new_set_refcount = set_refcount_funcs[refcount_order]; | |
3004 | ||
3005 | ||
3006 | do { | |
3007 | int total_walks; | |
3008 | ||
3009 | new_allocation = false; | |
3010 | ||
3011 | /* At least we have to do this walk and the one which writes the | |
3012 | * refblocks; also, at least we have to do this loop here at least | |
3013 | * twice (normally), first to do the allocations, and second to | |
3014 | * determine that everything is correctly allocated, this then makes | |
3015 | * three walks in total */ | |
3016 | total_walks = MAX(walk_index + 2, 3); | |
3017 | ||
3018 | /* First, allocate the structures so they are present in the refcount | |
3019 | * structures */ | |
3020 | ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index, | |
3021 | &new_reftable_size, NULL, new_refblock_size, | |
3022 | new_refcount_bits, &alloc_refblock, | |
3023 | &new_allocation, NULL, status_cb, cb_opaque, | |
3024 | walk_index++, total_walks, errp); | |
3025 | if (ret < 0) { | |
3026 | goto done; | |
3027 | } | |
3028 | ||
3029 | new_reftable_index = 0; | |
3030 | ||
3031 | if (new_allocation) { | |
3032 | if (new_reftable_offset) { | |
3033 | qcow2_free_clusters(bs, new_reftable_offset, | |
3034 | allocated_reftable_size * sizeof(uint64_t), | |
3035 | QCOW2_DISCARD_NEVER); | |
3036 | } | |
3037 | ||
3038 | new_reftable_offset = qcow2_alloc_clusters(bs, new_reftable_size * | |
3039 | sizeof(uint64_t)); | |
3040 | if (new_reftable_offset < 0) { | |
3041 | error_setg_errno(errp, -new_reftable_offset, | |
3042 | "Failed to allocate the new reftable"); | |
3043 | ret = new_reftable_offset; | |
3044 | goto done; | |
3045 | } | |
3046 | allocated_reftable_size = new_reftable_size; | |
3047 | } | |
3048 | } while (new_allocation); | |
3049 | ||
3050 | /* Second, write the new refblocks */ | |
3051 | ret = walk_over_reftable(bs, &new_reftable, &new_reftable_index, | |
3052 | &new_reftable_size, new_refblock, | |
3053 | new_refblock_size, new_refcount_bits, | |
3054 | &flush_refblock, &new_allocation, new_set_refcount, | |
3055 | status_cb, cb_opaque, walk_index, walk_index + 1, | |
3056 | errp); | |
3057 | if (ret < 0) { | |
3058 | goto done; | |
3059 | } | |
3060 | assert(!new_allocation); | |
3061 | ||
3062 | ||
3063 | /* Write the new reftable */ | |
3064 | ret = qcow2_pre_write_overlap_check(bs, 0, new_reftable_offset, | |
3065 | new_reftable_size * sizeof(uint64_t)); | |
3066 | if (ret < 0) { | |
3067 | error_setg_errno(errp, -ret, "Overlap check failed"); | |
3068 | goto done; | |
3069 | } | |
3070 | ||
3071 | for (i = 0; i < new_reftable_size; i++) { | |
3072 | cpu_to_be64s(&new_reftable[i]); | |
3073 | } | |
3074 | ||
d9ca2ea2 | 3075 | ret = bdrv_pwrite(bs->file, new_reftable_offset, new_reftable, |
791c9a00 HR |
3076 | new_reftable_size * sizeof(uint64_t)); |
3077 | ||
3078 | for (i = 0; i < new_reftable_size; i++) { | |
3079 | be64_to_cpus(&new_reftable[i]); | |
3080 | } | |
3081 | ||
3082 | if (ret < 0) { | |
3083 | error_setg_errno(errp, -ret, "Failed to write the new reftable"); | |
3084 | goto done; | |
3085 | } | |
3086 | ||
3087 | ||
3088 | /* Empty the refcount cache */ | |
3089 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
3090 | if (ret < 0) { | |
3091 | error_setg_errno(errp, -ret, "Failed to flush the refblock cache"); | |
3092 | goto done; | |
3093 | } | |
3094 | ||
3095 | /* Update the image header to point to the new reftable; this only updates | |
3096 | * the fields which are relevant to qcow2_update_header(); other fields | |
3097 | * such as s->refcount_table or s->refcount_bits stay stale for now | |
3098 | * (because we have to restore everything if qcow2_update_header() fails) */ | |
3099 | old_refcount_order = s->refcount_order; | |
3100 | old_reftable_size = s->refcount_table_size; | |
3101 | old_reftable_offset = s->refcount_table_offset; | |
3102 | ||
3103 | s->refcount_order = refcount_order; | |
3104 | s->refcount_table_size = new_reftable_size; | |
3105 | s->refcount_table_offset = new_reftable_offset; | |
3106 | ||
3107 | ret = qcow2_update_header(bs); | |
3108 | if (ret < 0) { | |
3109 | s->refcount_order = old_refcount_order; | |
3110 | s->refcount_table_size = old_reftable_size; | |
3111 | s->refcount_table_offset = old_reftable_offset; | |
3112 | error_setg_errno(errp, -ret, "Failed to update the qcow2 header"); | |
3113 | goto done; | |
3114 | } | |
3115 | ||
3116 | /* Now update the rest of the in-memory information */ | |
3117 | old_reftable = s->refcount_table; | |
3118 | s->refcount_table = new_reftable; | |
7061a078 | 3119 | update_max_refcount_table_index(s); |
791c9a00 HR |
3120 | |
3121 | s->refcount_bits = 1 << refcount_order; | |
3122 | s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); | |
3123 | s->refcount_max += s->refcount_max - 1; | |
3124 | ||
3125 | s->refcount_block_bits = s->cluster_bits - (refcount_order - 3); | |
3126 | s->refcount_block_size = 1 << s->refcount_block_bits; | |
3127 | ||
3128 | s->get_refcount = new_get_refcount; | |
3129 | s->set_refcount = new_set_refcount; | |
3130 | ||
3131 | /* For cleaning up all old refblocks and the old reftable below the "done" | |
3132 | * label */ | |
3133 | new_reftable = old_reftable; | |
3134 | new_reftable_size = old_reftable_size; | |
3135 | new_reftable_offset = old_reftable_offset; | |
3136 | ||
3137 | done: | |
3138 | if (new_reftable) { | |
3139 | /* On success, new_reftable actually points to the old reftable (and | |
3140 | * new_reftable_size is the old reftable's size); but that is just | |
3141 | * fine */ | |
3142 | for (i = 0; i < new_reftable_size; i++) { | |
3143 | uint64_t offset = new_reftable[i] & REFT_OFFSET_MASK; | |
3144 | if (offset) { | |
3145 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
3146 | QCOW2_DISCARD_OTHER); | |
3147 | } | |
3148 | } | |
3149 | g_free(new_reftable); | |
3150 | ||
3151 | if (new_reftable_offset > 0) { | |
3152 | qcow2_free_clusters(bs, new_reftable_offset, | |
3153 | new_reftable_size * sizeof(uint64_t), | |
3154 | QCOW2_DISCARD_OTHER); | |
3155 | } | |
3156 | } | |
3157 | ||
3158 | qemu_vfree(new_refblock); | |
3159 | return ret; | |
3160 | } | |
46b732cd | 3161 | |
23482f8a HR |
3162 | static int64_t get_refblock_offset(BlockDriverState *bs, uint64_t offset) |
3163 | { | |
3164 | BDRVQcow2State *s = bs->opaque; | |
3165 | uint32_t index = offset_to_reftable_index(s, offset); | |
3166 | int64_t covering_refblock_offset = 0; | |
3167 | ||
3168 | if (index < s->refcount_table_size) { | |
3169 | covering_refblock_offset = s->refcount_table[index] & REFT_OFFSET_MASK; | |
3170 | } | |
3171 | if (!covering_refblock_offset) { | |
3172 | qcow2_signal_corruption(bs, true, -1, -1, "Refblock at %#" PRIx64 " is " | |
3173 | "not covered by the refcount structures", | |
3174 | offset); | |
3175 | return -EIO; | |
3176 | } | |
3177 | ||
3178 | return covering_refblock_offset; | |
3179 | } | |
3180 | ||
46b732cd PB |
3181 | static int qcow2_discard_refcount_block(BlockDriverState *bs, |
3182 | uint64_t discard_block_offs) | |
3183 | { | |
3184 | BDRVQcow2State *s = bs->opaque; | |
23482f8a | 3185 | int64_t refblock_offs; |
46b732cd PB |
3186 | uint64_t cluster_index = discard_block_offs >> s->cluster_bits; |
3187 | uint32_t block_index = cluster_index & (s->refcount_block_size - 1); | |
3188 | void *refblock; | |
3189 | int ret; | |
3190 | ||
23482f8a HR |
3191 | refblock_offs = get_refblock_offset(bs, discard_block_offs); |
3192 | if (refblock_offs < 0) { | |
3193 | return refblock_offs; | |
3194 | } | |
3195 | ||
46b732cd PB |
3196 | assert(discard_block_offs != 0); |
3197 | ||
3198 | ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs, | |
3199 | &refblock); | |
3200 | if (ret < 0) { | |
3201 | return ret; | |
3202 | } | |
3203 | ||
3204 | if (s->get_refcount(refblock, block_index) != 1) { | |
3205 | qcow2_signal_corruption(bs, true, -1, -1, "Invalid refcount:" | |
3206 | " refblock offset %#" PRIx64 | |
3207 | ", reftable index %u" | |
3208 | ", block offset %#" PRIx64 | |
3209 | ", refcount %#" PRIx64, | |
3210 | refblock_offs, | |
3211 | offset_to_reftable_index(s, discard_block_offs), | |
3212 | discard_block_offs, | |
3213 | s->get_refcount(refblock, block_index)); | |
2013c3d4 | 3214 | qcow2_cache_put(s->refcount_block_cache, &refblock); |
46b732cd PB |
3215 | return -EINVAL; |
3216 | } | |
3217 | s->set_refcount(refblock, block_index, 0); | |
3218 | ||
2d135ee9 | 3219 | qcow2_cache_entry_mark_dirty(s->refcount_block_cache, refblock); |
46b732cd | 3220 | |
2013c3d4 | 3221 | qcow2_cache_put(s->refcount_block_cache, &refblock); |
46b732cd PB |
3222 | |
3223 | if (cluster_index < s->free_cluster_index) { | |
3224 | s->free_cluster_index = cluster_index; | |
3225 | } | |
3226 | ||
6e6fa760 | 3227 | refblock = qcow2_cache_is_table_offset(s->refcount_block_cache, |
46b732cd PB |
3228 | discard_block_offs); |
3229 | if (refblock) { | |
3230 | /* discard refblock from the cache if refblock is cached */ | |
77aadd7b | 3231 | qcow2_cache_discard(s->refcount_block_cache, refblock); |
46b732cd PB |
3232 | } |
3233 | update_refcount_discard(bs, discard_block_offs, s->cluster_size); | |
3234 | ||
3235 | return 0; | |
3236 | } | |
3237 | ||
3238 | int qcow2_shrink_reftable(BlockDriverState *bs) | |
3239 | { | |
3240 | BDRVQcow2State *s = bs->opaque; | |
3241 | uint64_t *reftable_tmp = | |
3242 | g_malloc(s->refcount_table_size * sizeof(uint64_t)); | |
3243 | int i, ret; | |
3244 | ||
3245 | for (i = 0; i < s->refcount_table_size; i++) { | |
3246 | int64_t refblock_offs = s->refcount_table[i] & REFT_OFFSET_MASK; | |
3247 | void *refblock; | |
3248 | bool unused_block; | |
3249 | ||
3250 | if (refblock_offs == 0) { | |
3251 | reftable_tmp[i] = 0; | |
3252 | continue; | |
3253 | } | |
3254 | ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs, | |
3255 | &refblock); | |
3256 | if (ret < 0) { | |
3257 | goto out; | |
3258 | } | |
3259 | ||
3260 | /* the refblock has own reference */ | |
3261 | if (i == offset_to_reftable_index(s, refblock_offs)) { | |
3262 | uint64_t block_index = (refblock_offs >> s->cluster_bits) & | |
3263 | (s->refcount_block_size - 1); | |
3264 | uint64_t refcount = s->get_refcount(refblock, block_index); | |
3265 | ||
3266 | s->set_refcount(refblock, block_index, 0); | |
3267 | ||
3268 | unused_block = buffer_is_zero(refblock, s->cluster_size); | |
3269 | ||
3270 | s->set_refcount(refblock, block_index, refcount); | |
3271 | } else { | |
3272 | unused_block = buffer_is_zero(refblock, s->cluster_size); | |
3273 | } | |
2013c3d4 | 3274 | qcow2_cache_put(s->refcount_block_cache, &refblock); |
46b732cd PB |
3275 | |
3276 | reftable_tmp[i] = unused_block ? 0 : cpu_to_be64(s->refcount_table[i]); | |
3277 | } | |
3278 | ||
3279 | ret = bdrv_pwrite_sync(bs->file, s->refcount_table_offset, reftable_tmp, | |
3280 | s->refcount_table_size * sizeof(uint64_t)); | |
3281 | /* | |
3282 | * If the write in the reftable failed the image may contain a partially | |
3283 | * overwritten reftable. In this case it would be better to clear the | |
3284 | * reftable in memory to avoid possible image corruption. | |
3285 | */ | |
3286 | for (i = 0; i < s->refcount_table_size; i++) { | |
3287 | if (s->refcount_table[i] && !reftable_tmp[i]) { | |
3288 | if (ret == 0) { | |
3289 | ret = qcow2_discard_refcount_block(bs, s->refcount_table[i] & | |
3290 | REFT_OFFSET_MASK); | |
3291 | } | |
3292 | s->refcount_table[i] = 0; | |
3293 | } | |
3294 | } | |
3295 | ||
3296 | if (!s->cache_discards) { | |
3297 | qcow2_process_discards(bs, ret); | |
3298 | } | |
3299 | ||
3300 | out: | |
3301 | g_free(reftable_tmp); | |
3302 | return ret; | |
3303 | } | |
163bc39d PB |
3304 | |
3305 | int64_t qcow2_get_last_cluster(BlockDriverState *bs, int64_t size) | |
3306 | { | |
3307 | BDRVQcow2State *s = bs->opaque; | |
3308 | int64_t i; | |
3309 | ||
3310 | for (i = size_to_clusters(s, size) - 1; i >= 0; i--) { | |
3311 | uint64_t refcount; | |
3312 | int ret = qcow2_get_refcount(bs, i, &refcount); | |
3313 | if (ret < 0) { | |
3314 | fprintf(stderr, "Can't get refcount for cluster %" PRId64 ": %s\n", | |
3315 | i, strerror(-ret)); | |
3316 | return ret; | |
3317 | } | |
3318 | if (refcount > 0) { | |
3319 | return i; | |
3320 | } | |
3321 | } | |
3322 | qcow2_signal_corruption(bs, true, -1, -1, | |
3323 | "There are no references in the refcount table."); | |
3324 | return -EIO; | |
3325 | } |