]>
Commit | Line | Data |
---|---|---|
ea2384d3 FB |
1 | /* |
2 | * Block driver for the QCOW format | |
5fafdf24 | 3 | * |
83f64091 | 4 | * Copyright (c) 2004-2006 Fabrice Bellard |
5fafdf24 | 5 | * |
ea2384d3 FB |
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 | */ | |
faf07963 | 24 | #include "qemu-common.h" |
ea2384d3 | 25 | #include "block_int.h" |
5efa9d5a | 26 | #include "module.h" |
28d34b82 | 27 | #include <zlib.h> |
ea2384d3 FB |
28 | #include "aes.h" |
29 | ||
30 | /**************************************************************/ | |
31 | /* QEMU COW block driver with compression and encryption support */ | |
32 | ||
33 | #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb) | |
34 | #define QCOW_VERSION 1 | |
35 | ||
36 | #define QCOW_CRYPT_NONE 0 | |
37 | #define QCOW_CRYPT_AES 1 | |
38 | ||
39 | #define QCOW_OFLAG_COMPRESSED (1LL << 63) | |
40 | ||
41 | typedef struct QCowHeader { | |
42 | uint32_t magic; | |
43 | uint32_t version; | |
44 | uint64_t backing_file_offset; | |
45 | uint32_t backing_file_size; | |
46 | uint32_t mtime; | |
47 | uint64_t size; /* in bytes */ | |
48 | uint8_t cluster_bits; | |
49 | uint8_t l2_bits; | |
50 | uint32_t crypt_method; | |
51 | uint64_t l1_table_offset; | |
52 | } QCowHeader; | |
53 | ||
54 | #define L2_CACHE_SIZE 16 | |
55 | ||
56 | typedef struct BDRVQcowState { | |
83f64091 | 57 | BlockDriverState *hd; |
ea2384d3 FB |
58 | int cluster_bits; |
59 | int cluster_size; | |
60 | int cluster_sectors; | |
61 | int l2_bits; | |
62 | int l2_size; | |
63 | int l1_size; | |
64 | uint64_t cluster_offset_mask; | |
65 | uint64_t l1_table_offset; | |
66 | uint64_t *l1_table; | |
67 | uint64_t *l2_cache; | |
68 | uint64_t l2_cache_offsets[L2_CACHE_SIZE]; | |
69 | uint32_t l2_cache_counts[L2_CACHE_SIZE]; | |
70 | uint8_t *cluster_cache; | |
71 | uint8_t *cluster_data; | |
72 | uint64_t cluster_cache_offset; | |
73 | uint32_t crypt_method; /* current crypt method, 0 if no key yet */ | |
74 | uint32_t crypt_method_header; | |
75 | AES_KEY aes_encrypt_key; | |
76 | AES_KEY aes_decrypt_key; | |
77 | } BDRVQcowState; | |
78 | ||
79 | static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset); | |
80 | ||
81 | static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename) | |
82 | { | |
83 | const QCowHeader *cow_header = (const void *)buf; | |
3b46e624 | 84 | |
712e7874 FB |
85 | if (buf_size >= sizeof(QCowHeader) && |
86 | be32_to_cpu(cow_header->magic) == QCOW_MAGIC && | |
5fafdf24 | 87 | be32_to_cpu(cow_header->version) == QCOW_VERSION) |
ea2384d3 FB |
88 | return 100; |
89 | else | |
90 | return 0; | |
91 | } | |
92 | ||
83f64091 | 93 | static int qcow_open(BlockDriverState *bs, const char *filename, int flags) |
ea2384d3 FB |
94 | { |
95 | BDRVQcowState *s = bs->opaque; | |
83f64091 | 96 | int len, i, shift, ret; |
ea2384d3 | 97 | QCowHeader header; |
83f64091 | 98 | |
b5eff355 | 99 | ret = bdrv_file_open(&s->hd, filename, flags); |
83f64091 FB |
100 | if (ret < 0) |
101 | return ret; | |
102 | if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header)) | |
ea2384d3 FB |
103 | goto fail; |
104 | be32_to_cpus(&header.magic); | |
105 | be32_to_cpus(&header.version); | |
106 | be64_to_cpus(&header.backing_file_offset); | |
107 | be32_to_cpus(&header.backing_file_size); | |
108 | be32_to_cpus(&header.mtime); | |
109 | be64_to_cpus(&header.size); | |
110 | be32_to_cpus(&header.crypt_method); | |
111 | be64_to_cpus(&header.l1_table_offset); | |
3b46e624 | 112 | |
ea2384d3 FB |
113 | if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION) |
114 | goto fail; | |
115 | if (header.size <= 1 || header.cluster_bits < 9) | |
116 | goto fail; | |
117 | if (header.crypt_method > QCOW_CRYPT_AES) | |
118 | goto fail; | |
119 | s->crypt_method_header = header.crypt_method; | |
120 | if (s->crypt_method_header) | |
121 | bs->encrypted = 1; | |
122 | s->cluster_bits = header.cluster_bits; | |
123 | s->cluster_size = 1 << s->cluster_bits; | |
124 | s->cluster_sectors = 1 << (s->cluster_bits - 9); | |
125 | s->l2_bits = header.l2_bits; | |
126 | s->l2_size = 1 << s->l2_bits; | |
127 | bs->total_sectors = header.size / 512; | |
128 | s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1; | |
129 | ||
130 | /* read the level 1 table */ | |
131 | shift = s->cluster_bits + s->l2_bits; | |
132 | s->l1_size = (header.size + (1LL << shift) - 1) >> shift; | |
133 | ||
134 | s->l1_table_offset = header.l1_table_offset; | |
135 | s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t)); | |
136 | if (!s->l1_table) | |
137 | goto fail; | |
5fafdf24 | 138 | if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) != |
ea2384d3 FB |
139 | s->l1_size * sizeof(uint64_t)) |
140 | goto fail; | |
141 | for(i = 0;i < s->l1_size; i++) { | |
142 | be64_to_cpus(&s->l1_table[i]); | |
143 | } | |
144 | /* alloc L2 cache */ | |
145 | s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); | |
146 | if (!s->l2_cache) | |
147 | goto fail; | |
148 | s->cluster_cache = qemu_malloc(s->cluster_size); | |
149 | if (!s->cluster_cache) | |
150 | goto fail; | |
151 | s->cluster_data = qemu_malloc(s->cluster_size); | |
152 | if (!s->cluster_data) | |
153 | goto fail; | |
154 | s->cluster_cache_offset = -1; | |
3b46e624 | 155 | |
ea2384d3 FB |
156 | /* read the backing file name */ |
157 | if (header.backing_file_offset != 0) { | |
158 | len = header.backing_file_size; | |
159 | if (len > 1023) | |
160 | len = 1023; | |
83f64091 | 161 | if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len) |
ea2384d3 FB |
162 | goto fail; |
163 | bs->backing_file[len] = '\0'; | |
164 | } | |
165 | return 0; | |
166 | ||
167 | fail: | |
168 | qemu_free(s->l1_table); | |
169 | qemu_free(s->l2_cache); | |
170 | qemu_free(s->cluster_cache); | |
171 | qemu_free(s->cluster_data); | |
83f64091 | 172 | bdrv_delete(s->hd); |
ea2384d3 FB |
173 | return -1; |
174 | } | |
175 | ||
176 | static int qcow_set_key(BlockDriverState *bs, const char *key) | |
177 | { | |
178 | BDRVQcowState *s = bs->opaque; | |
179 | uint8_t keybuf[16]; | |
180 | int len, i; | |
3b46e624 | 181 | |
ea2384d3 FB |
182 | memset(keybuf, 0, 16); |
183 | len = strlen(key); | |
184 | if (len > 16) | |
185 | len = 16; | |
186 | /* XXX: we could compress the chars to 7 bits to increase | |
187 | entropy */ | |
188 | for(i = 0;i < len;i++) { | |
189 | keybuf[i] = key[i]; | |
190 | } | |
191 | s->crypt_method = s->crypt_method_header; | |
192 | ||
193 | if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0) | |
194 | return -1; | |
195 | if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0) | |
196 | return -1; | |
197 | #if 0 | |
198 | /* test */ | |
199 | { | |
200 | uint8_t in[16]; | |
201 | uint8_t out[16]; | |
202 | uint8_t tmp[16]; | |
203 | for(i=0;i<16;i++) | |
204 | in[i] = i; | |
205 | AES_encrypt(in, tmp, &s->aes_encrypt_key); | |
206 | AES_decrypt(tmp, out, &s->aes_decrypt_key); | |
207 | for(i = 0; i < 16; i++) | |
208 | printf(" %02x", tmp[i]); | |
209 | printf("\n"); | |
210 | for(i = 0; i < 16; i++) | |
211 | printf(" %02x", out[i]); | |
212 | printf("\n"); | |
213 | } | |
214 | #endif | |
215 | return 0; | |
216 | } | |
217 | ||
218 | /* The crypt function is compatible with the linux cryptoloop | |
219 | algorithm for < 4 GB images. NOTE: out_buf == in_buf is | |
220 | supported */ | |
221 | static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num, | |
222 | uint8_t *out_buf, const uint8_t *in_buf, | |
223 | int nb_sectors, int enc, | |
224 | const AES_KEY *key) | |
225 | { | |
226 | union { | |
227 | uint64_t ll[2]; | |
228 | uint8_t b[16]; | |
229 | } ivec; | |
230 | int i; | |
231 | ||
232 | for(i = 0; i < nb_sectors; i++) { | |
233 | ivec.ll[0] = cpu_to_le64(sector_num); | |
234 | ivec.ll[1] = 0; | |
5fafdf24 | 235 | AES_cbc_encrypt(in_buf, out_buf, 512, key, |
ea2384d3 FB |
236 | ivec.b, enc); |
237 | sector_num++; | |
238 | in_buf += 512; | |
239 | out_buf += 512; | |
240 | } | |
241 | } | |
242 | ||
243 | /* 'allocate' is: | |
244 | * | |
245 | * 0 to not allocate. | |
246 | * | |
247 | * 1 to allocate a normal cluster (for sector indexes 'n_start' to | |
248 | * 'n_end') | |
249 | * | |
250 | * 2 to allocate a compressed cluster of size | |
251 | * 'compressed_size'. 'compressed_size' must be > 0 and < | |
5fafdf24 | 252 | * cluster_size |
ea2384d3 FB |
253 | * |
254 | * return 0 if not allocated. | |
255 | */ | |
256 | static uint64_t get_cluster_offset(BlockDriverState *bs, | |
257 | uint64_t offset, int allocate, | |
258 | int compressed_size, | |
259 | int n_start, int n_end) | |
260 | { | |
261 | BDRVQcowState *s = bs->opaque; | |
262 | int min_index, i, j, l1_index, l2_index; | |
263 | uint64_t l2_offset, *l2_table, cluster_offset, tmp; | |
264 | uint32_t min_count; | |
265 | int new_l2_table; | |
3b46e624 | 266 | |
ea2384d3 FB |
267 | l1_index = offset >> (s->l2_bits + s->cluster_bits); |
268 | l2_offset = s->l1_table[l1_index]; | |
269 | new_l2_table = 0; | |
270 | if (!l2_offset) { | |
271 | if (!allocate) | |
272 | return 0; | |
273 | /* allocate a new l2 entry */ | |
83f64091 | 274 | l2_offset = bdrv_getlength(s->hd); |
ea2384d3 FB |
275 | /* round to cluster size */ |
276 | l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1); | |
277 | /* update the L1 entry */ | |
278 | s->l1_table[l1_index] = l2_offset; | |
279 | tmp = cpu_to_be64(l2_offset); | |
5fafdf24 | 280 | if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp), |
83f64091 | 281 | &tmp, sizeof(tmp)) != sizeof(tmp)) |
ea2384d3 FB |
282 | return 0; |
283 | new_l2_table = 1; | |
284 | } | |
285 | for(i = 0; i < L2_CACHE_SIZE; i++) { | |
286 | if (l2_offset == s->l2_cache_offsets[i]) { | |
287 | /* increment the hit count */ | |
288 | if (++s->l2_cache_counts[i] == 0xffffffff) { | |
289 | for(j = 0; j < L2_CACHE_SIZE; j++) { | |
290 | s->l2_cache_counts[j] >>= 1; | |
291 | } | |
292 | } | |
293 | l2_table = s->l2_cache + (i << s->l2_bits); | |
294 | goto found; | |
295 | } | |
296 | } | |
297 | /* not found: load a new entry in the least used one */ | |
298 | min_index = 0; | |
299 | min_count = 0xffffffff; | |
300 | for(i = 0; i < L2_CACHE_SIZE; i++) { | |
301 | if (s->l2_cache_counts[i] < min_count) { | |
302 | min_count = s->l2_cache_counts[i]; | |
303 | min_index = i; | |
304 | } | |
305 | } | |
306 | l2_table = s->l2_cache + (min_index << s->l2_bits); | |
ea2384d3 FB |
307 | if (new_l2_table) { |
308 | memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); | |
83f64091 | 309 | if (bdrv_pwrite(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) != |
ea2384d3 FB |
310 | s->l2_size * sizeof(uint64_t)) |
311 | return 0; | |
312 | } else { | |
5fafdf24 | 313 | if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) != |
ea2384d3 FB |
314 | s->l2_size * sizeof(uint64_t)) |
315 | return 0; | |
316 | } | |
317 | s->l2_cache_offsets[min_index] = l2_offset; | |
318 | s->l2_cache_counts[min_index] = 1; | |
319 | found: | |
320 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
321 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
5fafdf24 | 322 | if (!cluster_offset || |
ea2384d3 FB |
323 | ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) { |
324 | if (!allocate) | |
325 | return 0; | |
326 | /* allocate a new cluster */ | |
327 | if ((cluster_offset & QCOW_OFLAG_COMPRESSED) && | |
328 | (n_end - n_start) < s->cluster_sectors) { | |
329 | /* if the cluster is already compressed, we must | |
330 | decompress it in the case it is not completely | |
331 | overwritten */ | |
332 | if (decompress_cluster(s, cluster_offset) < 0) | |
333 | return 0; | |
83f64091 | 334 | cluster_offset = bdrv_getlength(s->hd); |
5fafdf24 | 335 | cluster_offset = (cluster_offset + s->cluster_size - 1) & |
ea2384d3 FB |
336 | ~(s->cluster_size - 1); |
337 | /* write the cluster content */ | |
5fafdf24 | 338 | if (bdrv_pwrite(s->hd, cluster_offset, s->cluster_cache, s->cluster_size) != |
ea2384d3 FB |
339 | s->cluster_size) |
340 | return -1; | |
341 | } else { | |
83f64091 | 342 | cluster_offset = bdrv_getlength(s->hd); |
7f48fa1f AL |
343 | if (allocate == 1) { |
344 | /* round to cluster size */ | |
345 | cluster_offset = (cluster_offset + s->cluster_size - 1) & | |
346 | ~(s->cluster_size - 1); | |
347 | bdrv_truncate(s->hd, cluster_offset + s->cluster_size); | |
348 | /* if encrypted, we must initialize the cluster | |
349 | content which won't be written */ | |
350 | if (s->crypt_method && | |
351 | (n_end - n_start) < s->cluster_sectors) { | |
352 | uint64_t start_sect; | |
353 | start_sect = (offset & ~(s->cluster_size - 1)) >> 9; | |
354 | memset(s->cluster_data + 512, 0x00, 512); | |
355 | for(i = 0; i < s->cluster_sectors; i++) { | |
356 | if (i < n_start || i >= n_end) { | |
357 | encrypt_sectors(s, start_sect + i, | |
358 | s->cluster_data, | |
359 | s->cluster_data + 512, 1, 1, | |
360 | &s->aes_encrypt_key); | |
361 | if (bdrv_pwrite(s->hd, cluster_offset + i * 512, | |
362 | s->cluster_data, 512) != 512) | |
363 | return -1; | |
364 | } | |
ea2384d3 FB |
365 | } |
366 | } | |
7f48fa1f AL |
367 | } else if (allocate == 2) { |
368 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
369 | (uint64_t)compressed_size << (63 - s->cluster_bits); | |
ea2384d3 FB |
370 | } |
371 | } | |
372 | /* update L2 table */ | |
373 | tmp = cpu_to_be64(cluster_offset); | |
374 | l2_table[l2_index] = tmp; | |
5fafdf24 | 375 | if (bdrv_pwrite(s->hd, |
83f64091 | 376 | l2_offset + l2_index * sizeof(tmp), &tmp, sizeof(tmp)) != sizeof(tmp)) |
ea2384d3 FB |
377 | return 0; |
378 | } | |
379 | return cluster_offset; | |
380 | } | |
381 | ||
5fafdf24 | 382 | static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num, |
ea2384d3 FB |
383 | int nb_sectors, int *pnum) |
384 | { | |
385 | BDRVQcowState *s = bs->opaque; | |
386 | int index_in_cluster, n; | |
387 | uint64_t cluster_offset; | |
388 | ||
389 | cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0); | |
390 | index_in_cluster = sector_num & (s->cluster_sectors - 1); | |
391 | n = s->cluster_sectors - index_in_cluster; | |
392 | if (n > nb_sectors) | |
393 | n = nb_sectors; | |
394 | *pnum = n; | |
395 | return (cluster_offset != 0); | |
396 | } | |
397 | ||
398 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
399 | const uint8_t *buf, int buf_size) | |
400 | { | |
401 | z_stream strm1, *strm = &strm1; | |
402 | int ret, out_len; | |
403 | ||
404 | memset(strm, 0, sizeof(*strm)); | |
405 | ||
406 | strm->next_in = (uint8_t *)buf; | |
407 | strm->avail_in = buf_size; | |
408 | strm->next_out = out_buf; | |
409 | strm->avail_out = out_buf_size; | |
410 | ||
411 | ret = inflateInit2(strm, -12); | |
412 | if (ret != Z_OK) | |
413 | return -1; | |
414 | ret = inflate(strm, Z_FINISH); | |
415 | out_len = strm->next_out - out_buf; | |
416 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
417 | out_len != out_buf_size) { | |
418 | inflateEnd(strm); | |
419 | return -1; | |
420 | } | |
421 | inflateEnd(strm); | |
422 | return 0; | |
423 | } | |
3b46e624 | 424 | |
ea2384d3 FB |
425 | static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset) |
426 | { | |
427 | int ret, csize; | |
428 | uint64_t coffset; | |
429 | ||
430 | coffset = cluster_offset & s->cluster_offset_mask; | |
431 | if (s->cluster_cache_offset != coffset) { | |
432 | csize = cluster_offset >> (63 - s->cluster_bits); | |
433 | csize &= (s->cluster_size - 1); | |
83f64091 | 434 | ret = bdrv_pread(s->hd, coffset, s->cluster_data, csize); |
5fafdf24 | 435 | if (ret != csize) |
ea2384d3 FB |
436 | return -1; |
437 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
438 | s->cluster_data, csize) < 0) { | |
439 | return -1; | |
440 | } | |
441 | s->cluster_cache_offset = coffset; | |
442 | } | |
443 | return 0; | |
444 | } | |
445 | ||
83f64091 FB |
446 | #if 0 |
447 | ||
5fafdf24 | 448 | static int qcow_read(BlockDriverState *bs, int64_t sector_num, |
ea2384d3 FB |
449 | uint8_t *buf, int nb_sectors) |
450 | { | |
451 | BDRVQcowState *s = bs->opaque; | |
452 | int ret, index_in_cluster, n; | |
453 | uint64_t cluster_offset; | |
3b46e624 | 454 | |
ea2384d3 FB |
455 | while (nb_sectors > 0) { |
456 | cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0); | |
457 | index_in_cluster = sector_num & (s->cluster_sectors - 1); | |
458 | n = s->cluster_sectors - index_in_cluster; | |
459 | if (n > nb_sectors) | |
460 | n = nb_sectors; | |
461 | if (!cluster_offset) { | |
83f64091 FB |
462 | if (bs->backing_hd) { |
463 | /* read from the base image */ | |
464 | ret = bdrv_read(bs->backing_hd, sector_num, buf, n); | |
465 | if (ret < 0) | |
466 | return -1; | |
467 | } else { | |
468 | memset(buf, 0, 512 * n); | |
469 | } | |
ea2384d3 FB |
470 | } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) { |
471 | if (decompress_cluster(s, cluster_offset) < 0) | |
472 | return -1; | |
473 | memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n); | |
474 | } else { | |
83f64091 | 475 | ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512); |
5fafdf24 | 476 | if (ret != n * 512) |
ea2384d3 FB |
477 | return -1; |
478 | if (s->crypt_method) { | |
5fafdf24 | 479 | encrypt_sectors(s, sector_num, buf, buf, n, 0, |
ea2384d3 FB |
480 | &s->aes_decrypt_key); |
481 | } | |
482 | } | |
483 | nb_sectors -= n; | |
484 | sector_num += n; | |
485 | buf += n * 512; | |
486 | } | |
487 | return 0; | |
488 | } | |
83f64091 | 489 | #endif |
ea2384d3 | 490 | |
ce1a14dc PB |
491 | typedef struct QCowAIOCB { |
492 | BlockDriverAIOCB common; | |
83f64091 | 493 | int64_t sector_num; |
f141eafe | 494 | QEMUIOVector *qiov; |
83f64091 | 495 | uint8_t *buf; |
f141eafe | 496 | void *orig_buf; |
83f64091 FB |
497 | int nb_sectors; |
498 | int n; | |
499 | uint64_t cluster_offset; | |
5fafdf24 | 500 | uint8_t *cluster_data; |
c87c0672 AL |
501 | struct iovec hd_iov; |
502 | QEMUIOVector hd_qiov; | |
83f64091 | 503 | BlockDriverAIOCB *hd_aiocb; |
83f64091 FB |
504 | } QCowAIOCB; |
505 | ||
83f64091 FB |
506 | static void qcow_aio_read_cb(void *opaque, int ret) |
507 | { | |
ce1a14dc PB |
508 | QCowAIOCB *acb = opaque; |
509 | BlockDriverState *bs = acb->common.bs; | |
83f64091 | 510 | BDRVQcowState *s = bs->opaque; |
83f64091 FB |
511 | int index_in_cluster; |
512 | ||
ce1a14dc | 513 | acb->hd_aiocb = NULL; |
f141eafe AL |
514 | if (ret < 0) |
515 | goto done; | |
83f64091 FB |
516 | |
517 | redo: | |
518 | /* post process the read buffer */ | |
ce1a14dc | 519 | if (!acb->cluster_offset) { |
83f64091 | 520 | /* nothing to do */ |
ce1a14dc | 521 | } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) { |
83f64091 FB |
522 | /* nothing to do */ |
523 | } else { | |
524 | if (s->crypt_method) { | |
5fafdf24 TS |
525 | encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf, |
526 | acb->n, 0, | |
83f64091 FB |
527 | &s->aes_decrypt_key); |
528 | } | |
529 | } | |
530 | ||
ce1a14dc PB |
531 | acb->nb_sectors -= acb->n; |
532 | acb->sector_num += acb->n; | |
533 | acb->buf += acb->n * 512; | |
83f64091 | 534 | |
ce1a14dc | 535 | if (acb->nb_sectors == 0) { |
83f64091 | 536 | /* request completed */ |
f141eafe AL |
537 | ret = 0; |
538 | goto done; | |
83f64091 | 539 | } |
3b46e624 | 540 | |
83f64091 | 541 | /* prepare next AIO request */ |
5fafdf24 | 542 | acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, |
ce1a14dc PB |
543 | 0, 0, 0, 0); |
544 | index_in_cluster = acb->sector_num & (s->cluster_sectors - 1); | |
545 | acb->n = s->cluster_sectors - index_in_cluster; | |
546 | if (acb->n > acb->nb_sectors) | |
547 | acb->n = acb->nb_sectors; | |
548 | ||
549 | if (!acb->cluster_offset) { | |
83f64091 FB |
550 | if (bs->backing_hd) { |
551 | /* read from the base image */ | |
3f4cb3d3 | 552 | acb->hd_iov.iov_base = (void *)acb->buf; |
c87c0672 AL |
553 | acb->hd_iov.iov_len = acb->n * 512; |
554 | qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); | |
555 | acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num, | |
556 | &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb); | |
ce1a14dc | 557 | if (acb->hd_aiocb == NULL) |
f141eafe | 558 | goto done; |
83f64091 FB |
559 | } else { |
560 | /* Note: in this case, no need to wait */ | |
ce1a14dc | 561 | memset(acb->buf, 0, 512 * acb->n); |
83f64091 FB |
562 | goto redo; |
563 | } | |
ce1a14dc | 564 | } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) { |
83f64091 | 565 | /* add AIO support for compressed blocks ? */ |
ce1a14dc | 566 | if (decompress_cluster(s, acb->cluster_offset) < 0) |
f141eafe | 567 | goto done; |
5fafdf24 | 568 | memcpy(acb->buf, |
ce1a14dc | 569 | s->cluster_cache + index_in_cluster * 512, 512 * acb->n); |
83f64091 FB |
570 | goto redo; |
571 | } else { | |
ce1a14dc | 572 | if ((acb->cluster_offset & 511) != 0) { |
83f64091 | 573 | ret = -EIO; |
f141eafe | 574 | goto done; |
83f64091 | 575 | } |
3f4cb3d3 | 576 | acb->hd_iov.iov_base = (void *)acb->buf; |
c87c0672 AL |
577 | acb->hd_iov.iov_len = acb->n * 512; |
578 | qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); | |
579 | acb->hd_aiocb = bdrv_aio_readv(s->hd, | |
5fafdf24 | 580 | (acb->cluster_offset >> 9) + index_in_cluster, |
c87c0672 | 581 | &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb); |
ce1a14dc | 582 | if (acb->hd_aiocb == NULL) |
f141eafe AL |
583 | goto done; |
584 | } | |
585 | ||
586 | return; | |
587 | ||
588 | done: | |
589 | if (acb->qiov->niov > 1) { | |
590 | qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size); | |
591 | qemu_vfree(acb->orig_buf); | |
83f64091 | 592 | } |
f141eafe AL |
593 | acb->common.cb(acb->common.opaque, ret); |
594 | qemu_aio_release(acb); | |
83f64091 FB |
595 | } |
596 | ||
f141eafe AL |
597 | static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs, |
598 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
ce1a14dc | 599 | BlockDriverCompletionFunc *cb, void *opaque) |
83f64091 | 600 | { |
ce1a14dc PB |
601 | QCowAIOCB *acb; |
602 | ||
603 | acb = qemu_aio_get(bs, cb, opaque); | |
604 | if (!acb) | |
605 | return NULL; | |
606 | acb->hd_aiocb = NULL; | |
607 | acb->sector_num = sector_num; | |
f141eafe AL |
608 | acb->qiov = qiov; |
609 | if (qiov->niov > 1) | |
e268ca52 | 610 | acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size); |
f141eafe | 611 | else |
3f4cb3d3 | 612 | acb->buf = (uint8_t *)qiov->iov->iov_base; |
ce1a14dc PB |
613 | acb->nb_sectors = nb_sectors; |
614 | acb->n = 0; | |
3b46e624 | 615 | acb->cluster_offset = 0; |
83f64091 FB |
616 | |
617 | qcow_aio_read_cb(acb, 0); | |
ce1a14dc | 618 | return &acb->common; |
83f64091 FB |
619 | } |
620 | ||
621 | static void qcow_aio_write_cb(void *opaque, int ret) | |
622 | { | |
ce1a14dc PB |
623 | QCowAIOCB *acb = opaque; |
624 | BlockDriverState *bs = acb->common.bs; | |
83f64091 | 625 | BDRVQcowState *s = bs->opaque; |
83f64091 FB |
626 | int index_in_cluster; |
627 | uint64_t cluster_offset; | |
628 | const uint8_t *src_buf; | |
ce1a14dc PB |
629 | |
630 | acb->hd_aiocb = NULL; | |
631 | ||
f141eafe AL |
632 | if (ret < 0) |
633 | goto done; | |
83f64091 | 634 | |
ce1a14dc PB |
635 | acb->nb_sectors -= acb->n; |
636 | acb->sector_num += acb->n; | |
637 | acb->buf += acb->n * 512; | |
83f64091 | 638 | |
ce1a14dc | 639 | if (acb->nb_sectors == 0) { |
83f64091 | 640 | /* request completed */ |
f141eafe AL |
641 | ret = 0; |
642 | goto done; | |
83f64091 | 643 | } |
3b46e624 | 644 | |
ce1a14dc PB |
645 | index_in_cluster = acb->sector_num & (s->cluster_sectors - 1); |
646 | acb->n = s->cluster_sectors - index_in_cluster; | |
647 | if (acb->n > acb->nb_sectors) | |
648 | acb->n = acb->nb_sectors; | |
5fafdf24 TS |
649 | cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, 1, 0, |
650 | index_in_cluster, | |
ce1a14dc | 651 | index_in_cluster + acb->n); |
83f64091 FB |
652 | if (!cluster_offset || (cluster_offset & 511) != 0) { |
653 | ret = -EIO; | |
f141eafe | 654 | goto done; |
83f64091 FB |
655 | } |
656 | if (s->crypt_method) { | |
ce1a14dc PB |
657 | if (!acb->cluster_data) { |
658 | acb->cluster_data = qemu_mallocz(s->cluster_size); | |
659 | if (!acb->cluster_data) { | |
83f64091 | 660 | ret = -ENOMEM; |
f141eafe | 661 | goto done; |
83f64091 FB |
662 | } |
663 | } | |
5fafdf24 | 664 | encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf, |
ce1a14dc PB |
665 | acb->n, 1, &s->aes_encrypt_key); |
666 | src_buf = acb->cluster_data; | |
83f64091 | 667 | } else { |
ce1a14dc | 668 | src_buf = acb->buf; |
83f64091 | 669 | } |
c87c0672 AL |
670 | |
671 | acb->hd_iov.iov_base = (void *)src_buf; | |
672 | acb->hd_iov.iov_len = acb->n * 512; | |
673 | qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); | |
674 | acb->hd_aiocb = bdrv_aio_writev(s->hd, | |
675 | (cluster_offset >> 9) + index_in_cluster, | |
676 | &acb->hd_qiov, acb->n, | |
677 | qcow_aio_write_cb, acb); | |
ce1a14dc | 678 | if (acb->hd_aiocb == NULL) |
f141eafe AL |
679 | goto done; |
680 | return; | |
681 | ||
682 | done: | |
683 | if (acb->qiov->niov > 1) | |
684 | qemu_vfree(acb->orig_buf); | |
685 | acb->common.cb(acb->common.opaque, ret); | |
686 | qemu_aio_release(acb); | |
83f64091 FB |
687 | } |
688 | ||
f141eafe AL |
689 | static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs, |
690 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
ce1a14dc | 691 | BlockDriverCompletionFunc *cb, void *opaque) |
83f64091 | 692 | { |
83f64091 | 693 | BDRVQcowState *s = bs->opaque; |
ce1a14dc | 694 | QCowAIOCB *acb; |
3b46e624 | 695 | |
83f64091 FB |
696 | s->cluster_cache_offset = -1; /* disable compressed cache */ |
697 | ||
ce1a14dc PB |
698 | acb = qemu_aio_get(bs, cb, opaque); |
699 | if (!acb) | |
700 | return NULL; | |
701 | acb->hd_aiocb = NULL; | |
702 | acb->sector_num = sector_num; | |
f141eafe AL |
703 | acb->qiov = qiov; |
704 | if (qiov->niov > 1) { | |
e268ca52 | 705 | acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size); |
f141eafe | 706 | qemu_iovec_to_buffer(qiov, acb->buf); |
3f4cb3d3 BS |
707 | } else { |
708 | acb->buf = (uint8_t *)qiov->iov->iov_base; | |
709 | } | |
ce1a14dc PB |
710 | acb->nb_sectors = nb_sectors; |
711 | acb->n = 0; | |
3b46e624 | 712 | |
83f64091 | 713 | qcow_aio_write_cb(acb, 0); |
ce1a14dc | 714 | return &acb->common; |
83f64091 FB |
715 | } |
716 | ||
ce1a14dc | 717 | static void qcow_aio_cancel(BlockDriverAIOCB *blockacb) |
83f64091 | 718 | { |
ce1a14dc PB |
719 | QCowAIOCB *acb = (QCowAIOCB *)blockacb; |
720 | if (acb->hd_aiocb) | |
721 | bdrv_aio_cancel(acb->hd_aiocb); | |
722 | qemu_aio_release(acb); | |
83f64091 FB |
723 | } |
724 | ||
e2731add | 725 | static void qcow_close(BlockDriverState *bs) |
ea2384d3 FB |
726 | { |
727 | BDRVQcowState *s = bs->opaque; | |
728 | qemu_free(s->l1_table); | |
729 | qemu_free(s->l2_cache); | |
730 | qemu_free(s->cluster_cache); | |
731 | qemu_free(s->cluster_data); | |
83f64091 | 732 | bdrv_delete(s->hd); |
ea2384d3 FB |
733 | } |
734 | ||
0e7e1989 | 735 | static int qcow_create(const char *filename, QEMUOptionParameter *options) |
ea2384d3 FB |
736 | { |
737 | int fd, header_size, backing_filename_len, l1_size, i, shift; | |
738 | QCowHeader header; | |
ea2384d3 | 739 | uint64_t tmp; |
0e7e1989 KW |
740 | int64_t total_size = 0; |
741 | const char *backing_file = NULL; | |
742 | int flags = 0; | |
743 | ||
744 | /* Read out options */ | |
745 | while (options && options->name) { | |
746 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
747 | total_size = options->value.n / 512; | |
748 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { | |
749 | backing_file = options->value.s; | |
750 | } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) { | |
751 | flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0; | |
752 | } | |
753 | options++; | |
754 | } | |
ea2384d3 | 755 | |
83f64091 | 756 | fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644); |
ea2384d3 FB |
757 | if (fd < 0) |
758 | return -1; | |
759 | memset(&header, 0, sizeof(header)); | |
760 | header.magic = cpu_to_be32(QCOW_MAGIC); | |
761 | header.version = cpu_to_be32(QCOW_VERSION); | |
762 | header.size = cpu_to_be64(total_size * 512); | |
763 | header_size = sizeof(header); | |
764 | backing_filename_len = 0; | |
765 | if (backing_file) { | |
7852e5da AJ |
766 | if (strcmp(backing_file, "fat:")) { |
767 | header.backing_file_offset = cpu_to_be64(header_size); | |
768 | backing_filename_len = strlen(backing_file); | |
769 | header.backing_file_size = cpu_to_be32(backing_filename_len); | |
770 | header_size += backing_filename_len; | |
771 | } else { | |
772 | /* special backing file for vvfat */ | |
773 | backing_file = NULL; | |
774 | } | |
ea2384d3 FB |
775 | header.cluster_bits = 9; /* 512 byte cluster to avoid copying |
776 | unmodifyed sectors */ | |
777 | header.l2_bits = 12; /* 32 KB L2 tables */ | |
778 | } else { | |
779 | header.cluster_bits = 12; /* 4 KB clusters */ | |
780 | header.l2_bits = 9; /* 4 KB L2 tables */ | |
781 | } | |
782 | header_size = (header_size + 7) & ~7; | |
783 | shift = header.cluster_bits + header.l2_bits; | |
784 | l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift; | |
785 | ||
786 | header.l1_table_offset = cpu_to_be64(header_size); | |
ec36ba14 | 787 | if (flags & BLOCK_FLAG_ENCRYPT) { |
ea2384d3 FB |
788 | header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); |
789 | } else { | |
790 | header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); | |
791 | } | |
3b46e624 | 792 | |
ea2384d3 FB |
793 | /* write all the data */ |
794 | write(fd, &header, sizeof(header)); | |
795 | if (backing_file) { | |
83f64091 | 796 | write(fd, backing_file, backing_filename_len); |
ea2384d3 FB |
797 | } |
798 | lseek(fd, header_size, SEEK_SET); | |
799 | tmp = 0; | |
800 | for(i = 0;i < l1_size; i++) { | |
801 | write(fd, &tmp, sizeof(tmp)); | |
802 | } | |
803 | close(fd); | |
804 | return 0; | |
805 | } | |
806 | ||
c47c33b0 | 807 | static int qcow_make_empty(BlockDriverState *bs) |
95389c86 FB |
808 | { |
809 | BDRVQcowState *s = bs->opaque; | |
810 | uint32_t l1_length = s->l1_size * sizeof(uint64_t); | |
83f64091 | 811 | int ret; |
95389c86 FB |
812 | |
813 | memset(s->l1_table, 0, l1_length); | |
83f64091 | 814 | if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0) |
95389c86 | 815 | return -1; |
83f64091 FB |
816 | ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length); |
817 | if (ret < 0) | |
818 | return ret; | |
95389c86 FB |
819 | |
820 | memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); | |
821 | memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t)); | |
822 | memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t)); | |
823 | ||
824 | return 0; | |
825 | } | |
826 | ||
ea2384d3 FB |
827 | /* XXX: put compressed sectors first, then all the cluster aligned |
828 | tables to avoid losing bytes in alignment */ | |
5fafdf24 | 829 | static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num, |
c47c33b0 | 830 | const uint8_t *buf, int nb_sectors) |
ea2384d3 FB |
831 | { |
832 | BDRVQcowState *s = bs->opaque; | |
833 | z_stream strm; | |
834 | int ret, out_len; | |
835 | uint8_t *out_buf; | |
836 | uint64_t cluster_offset; | |
837 | ||
c47c33b0 FB |
838 | if (nb_sectors != s->cluster_sectors) |
839 | return -EINVAL; | |
ea2384d3 FB |
840 | |
841 | out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); | |
842 | if (!out_buf) | |
843 | return -1; | |
844 | ||
845 | /* best compression, small window, no zlib header */ | |
846 | memset(&strm, 0, sizeof(strm)); | |
847 | ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, | |
5fafdf24 | 848 | Z_DEFLATED, -12, |
ea2384d3 FB |
849 | 9, Z_DEFAULT_STRATEGY); |
850 | if (ret != 0) { | |
851 | qemu_free(out_buf); | |
852 | return -1; | |
853 | } | |
854 | ||
855 | strm.avail_in = s->cluster_size; | |
856 | strm.next_in = (uint8_t *)buf; | |
857 | strm.avail_out = s->cluster_size; | |
858 | strm.next_out = out_buf; | |
859 | ||
860 | ret = deflate(&strm, Z_FINISH); | |
861 | if (ret != Z_STREAM_END && ret != Z_OK) { | |
862 | qemu_free(out_buf); | |
863 | deflateEnd(&strm); | |
864 | return -1; | |
865 | } | |
866 | out_len = strm.next_out - out_buf; | |
867 | ||
868 | deflateEnd(&strm); | |
869 | ||
870 | if (ret != Z_STREAM_END || out_len >= s->cluster_size) { | |
871 | /* could not compress: write normal cluster */ | |
ade40677 | 872 | bdrv_write(bs, sector_num, buf, s->cluster_sectors); |
ea2384d3 | 873 | } else { |
5fafdf24 | 874 | cluster_offset = get_cluster_offset(bs, sector_num << 9, 2, |
ea2384d3 FB |
875 | out_len, 0, 0); |
876 | cluster_offset &= s->cluster_offset_mask; | |
83f64091 | 877 | if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) { |
ea2384d3 FB |
878 | qemu_free(out_buf); |
879 | return -1; | |
880 | } | |
881 | } | |
3b46e624 | 882 | |
ea2384d3 FB |
883 | qemu_free(out_buf); |
884 | return 0; | |
885 | } | |
886 | ||
7a6cba61 PB |
887 | static void qcow_flush(BlockDriverState *bs) |
888 | { | |
889 | BDRVQcowState *s = bs->opaque; | |
83f64091 | 890 | bdrv_flush(s->hd); |
7a6cba61 PB |
891 | } |
892 | ||
c47c33b0 FB |
893 | static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
894 | { | |
895 | BDRVQcowState *s = bs->opaque; | |
896 | bdi->cluster_size = s->cluster_size; | |
897 | return 0; | |
898 | } | |
899 | ||
0e7e1989 KW |
900 | |
901 | static QEMUOptionParameter qcow_create_options[] = { | |
902 | { BLOCK_OPT_SIZE, OPT_SIZE }, | |
903 | { BLOCK_OPT_BACKING_FILE, OPT_STRING }, | |
904 | { BLOCK_OPT_ENCRYPT, OPT_FLAG }, | |
905 | { NULL } | |
906 | }; | |
907 | ||
5efa9d5a | 908 | static BlockDriver bdrv_qcow = { |
e60f469c AJ |
909 | .format_name = "qcow", |
910 | .instance_size = sizeof(BDRVQcowState), | |
911 | .bdrv_probe = qcow_probe, | |
912 | .bdrv_open = qcow_open, | |
913 | .bdrv_close = qcow_close, | |
914 | .bdrv_create = qcow_create, | |
915 | .bdrv_flush = qcow_flush, | |
916 | .bdrv_is_allocated = qcow_is_allocated, | |
917 | .bdrv_set_key = qcow_set_key, | |
918 | .bdrv_make_empty = qcow_make_empty, | |
f141eafe AL |
919 | .bdrv_aio_readv = qcow_aio_readv, |
920 | .bdrv_aio_writev = qcow_aio_writev, | |
e60f469c AJ |
921 | .bdrv_aio_cancel = qcow_aio_cancel, |
922 | .aiocb_size = sizeof(QCowAIOCB), | |
c47c33b0 | 923 | .bdrv_write_compressed = qcow_write_compressed, |
e60f469c | 924 | .bdrv_get_info = qcow_get_info, |
0e7e1989 KW |
925 | |
926 | .create_options = qcow_create_options, | |
ea2384d3 | 927 | }; |
5efa9d5a AL |
928 | |
929 | static void bdrv_qcow_init(void) | |
930 | { | |
931 | bdrv_register(&bdrv_qcow); | |
932 | } | |
933 | ||
934 | block_init(bdrv_qcow_init); |