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