Merge branch 'master' of ssh://repo.or.cz/srv/git/qemu
[qemu/mmix.git] / block-qcow.c
blobfc6b809715510790e349e99ff4c6d8629528bf4e
1 /*
2 * Block driver for the QCOW format
4 * Copyright (c) 2004-2006 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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.
24 #include "qemu-common.h"
25 #include "block_int.h"
26 #include <zlib.h>
27 #include "aes.h"
29 /**************************************************************/
30 /* QEMU COW block driver with compression and encryption support */
32 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
33 #define QCOW_VERSION 1
35 #define QCOW_CRYPT_NONE 0
36 #define QCOW_CRYPT_AES 1
38 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
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;
53 #define L2_CACHE_SIZE 16
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;
78 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
80 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
82 const QCowHeader *cow_header = (const void *)buf;
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;
92 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
94 BDRVQcowState *s = bs->opaque;
95 int len, i, shift, ret;
96 QCowHeader header;
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);
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;
129 /* read the level 1 table */
130 shift = s->cluster_bits + s->l2_bits;
131 s->l1_size = (header.size + (1LL << shift) - 1) >> shift;
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]);
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;
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';
164 return 0;
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;
175 static int qcow_set_key(BlockDriverState *bs, const char *key)
177 BDRVQcowState *s = bs->opaque;
178 uint8_t keybuf[16];
179 int len, i;
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];
190 s->crypt_method = s->crypt_method_header;
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 */
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");
213 #endif
214 return 0;
217 /* The crypt function is compatible with the linux cryptoloop
218 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
219 supported */
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)
225 union {
226 uint64_t ll[2];
227 uint8_t b[16];
228 } ivec;
229 int i;
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;
242 /* 'allocate' is:
244 * 0 to not allocate.
246 * 1 to allocate a normal cluster (for sector indexes 'n_start' to
247 * 'n_end')
249 * 2 to allocate a compressed cluster of size
250 * 'compressed_size'. 'compressed_size' must be > 0 and <
251 * cluster_size
253 * return 0 if not allocated.
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)
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;
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;
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;
292 l2_table = s->l2_cache + (i << s->l2_bits);
293 goto found;
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;
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;
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 if (allocate == 1) {
343 /* round to cluster size */
344 cluster_offset = (cluster_offset + s->cluster_size - 1) &
345 ~(s->cluster_size - 1);
346 bdrv_truncate(s->hd, cluster_offset + s->cluster_size);
347 /* if encrypted, we must initialize the cluster
348 content which won't be written */
349 if (s->crypt_method &&
350 (n_end - n_start) < s->cluster_sectors) {
351 uint64_t start_sect;
352 start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
353 memset(s->cluster_data + 512, 0x00, 512);
354 for(i = 0; i < s->cluster_sectors; i++) {
355 if (i < n_start || i >= n_end) {
356 encrypt_sectors(s, start_sect + i,
357 s->cluster_data,
358 s->cluster_data + 512, 1, 1,
359 &s->aes_encrypt_key);
360 if (bdrv_pwrite(s->hd, cluster_offset + i * 512,
361 s->cluster_data, 512) != 512)
362 return -1;
366 } else if (allocate == 2) {
367 cluster_offset |= QCOW_OFLAG_COMPRESSED |
368 (uint64_t)compressed_size << (63 - s->cluster_bits);
371 /* update L2 table */
372 tmp = cpu_to_be64(cluster_offset);
373 l2_table[l2_index] = tmp;
374 if (bdrv_pwrite(s->hd,
375 l2_offset + l2_index * sizeof(tmp), &tmp, sizeof(tmp)) != sizeof(tmp))
376 return 0;
378 return cluster_offset;
381 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
382 int nb_sectors, int *pnum)
384 BDRVQcowState *s = bs->opaque;
385 int index_in_cluster, n;
386 uint64_t cluster_offset;
388 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
389 index_in_cluster = sector_num & (s->cluster_sectors - 1);
390 n = s->cluster_sectors - index_in_cluster;
391 if (n > nb_sectors)
392 n = nb_sectors;
393 *pnum = n;
394 return (cluster_offset != 0);
397 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
398 const uint8_t *buf, int buf_size)
400 z_stream strm1, *strm = &strm1;
401 int ret, out_len;
403 memset(strm, 0, sizeof(*strm));
405 strm->next_in = (uint8_t *)buf;
406 strm->avail_in = buf_size;
407 strm->next_out = out_buf;
408 strm->avail_out = out_buf_size;
410 ret = inflateInit2(strm, -12);
411 if (ret != Z_OK)
412 return -1;
413 ret = inflate(strm, Z_FINISH);
414 out_len = strm->next_out - out_buf;
415 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
416 out_len != out_buf_size) {
417 inflateEnd(strm);
418 return -1;
420 inflateEnd(strm);
421 return 0;
424 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
426 int ret, csize;
427 uint64_t coffset;
429 coffset = cluster_offset & s->cluster_offset_mask;
430 if (s->cluster_cache_offset != coffset) {
431 csize = cluster_offset >> (63 - s->cluster_bits);
432 csize &= (s->cluster_size - 1);
433 ret = bdrv_pread(s->hd, coffset, s->cluster_data, csize);
434 if (ret != csize)
435 return -1;
436 if (decompress_buffer(s->cluster_cache, s->cluster_size,
437 s->cluster_data, csize) < 0) {
438 return -1;
440 s->cluster_cache_offset = coffset;
442 return 0;
445 #if 0
447 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
448 uint8_t *buf, int nb_sectors)
450 BDRVQcowState *s = bs->opaque;
451 int ret, index_in_cluster, n;
452 uint64_t cluster_offset;
454 while (nb_sectors > 0) {
455 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
456 index_in_cluster = sector_num & (s->cluster_sectors - 1);
457 n = s->cluster_sectors - index_in_cluster;
458 if (n > nb_sectors)
459 n = nb_sectors;
460 if (!cluster_offset) {
461 if (bs->backing_hd) {
462 /* read from the base image */
463 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
464 if (ret < 0)
465 return -1;
466 } else {
467 memset(buf, 0, 512 * n);
469 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
470 if (decompress_cluster(s, cluster_offset) < 0)
471 return -1;
472 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
473 } else {
474 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
475 if (ret != n * 512)
476 return -1;
477 if (s->crypt_method) {
478 encrypt_sectors(s, sector_num, buf, buf, n, 0,
479 &s->aes_decrypt_key);
482 nb_sectors -= n;
483 sector_num += n;
484 buf += n * 512;
486 return 0;
488 #endif
490 static int qcow_write(BlockDriverState *bs, int64_t sector_num,
491 const uint8_t *buf, int nb_sectors)
493 BDRVQcowState *s = bs->opaque;
494 int ret, index_in_cluster, n;
495 uint64_t cluster_offset;
497 while (nb_sectors > 0) {
498 index_in_cluster = sector_num & (s->cluster_sectors - 1);
499 n = s->cluster_sectors - index_in_cluster;
500 if (n > nb_sectors)
501 n = nb_sectors;
502 cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
503 index_in_cluster,
504 index_in_cluster + n);
505 if (!cluster_offset)
506 return -1;
507 if (s->crypt_method) {
508 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
509 &s->aes_encrypt_key);
510 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
511 s->cluster_data, n * 512);
512 } else {
513 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
515 if (ret != n * 512)
516 return -1;
517 nb_sectors -= n;
518 sector_num += n;
519 buf += n * 512;
521 s->cluster_cache_offset = -1; /* disable compressed cache */
522 return 0;
525 typedef struct QCowAIOCB {
526 BlockDriverAIOCB common;
527 int64_t sector_num;
528 QEMUIOVector *qiov;
529 uint8_t *buf;
530 void *orig_buf;
531 int nb_sectors;
532 int n;
533 uint64_t cluster_offset;
534 uint8_t *cluster_data;
535 struct iovec hd_iov;
536 QEMUIOVector hd_qiov;
537 BlockDriverAIOCB *hd_aiocb;
538 } QCowAIOCB;
540 static void qcow_aio_read_cb(void *opaque, int ret)
542 QCowAIOCB *acb = opaque;
543 BlockDriverState *bs = acb->common.bs;
544 BDRVQcowState *s = bs->opaque;
545 int index_in_cluster;
547 acb->hd_aiocb = NULL;
548 if (ret < 0)
549 goto done;
551 redo:
552 /* post process the read buffer */
553 if (!acb->cluster_offset) {
554 /* nothing to do */
555 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
556 /* nothing to do */
557 } else {
558 if (s->crypt_method) {
559 encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
560 acb->n, 0,
561 &s->aes_decrypt_key);
565 acb->nb_sectors -= acb->n;
566 acb->sector_num += acb->n;
567 acb->buf += acb->n * 512;
569 if (acb->nb_sectors == 0) {
570 /* request completed */
571 ret = 0;
572 goto done;
575 /* prepare next AIO request */
576 acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9,
577 0, 0, 0, 0);
578 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
579 acb->n = s->cluster_sectors - index_in_cluster;
580 if (acb->n > acb->nb_sectors)
581 acb->n = acb->nb_sectors;
583 if (!acb->cluster_offset) {
584 if (bs->backing_hd) {
585 /* read from the base image */
586 acb->hd_iov.iov_base = (void *)acb->buf;
587 acb->hd_iov.iov_len = acb->n * 512;
588 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
589 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
590 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
591 if (acb->hd_aiocb == NULL)
592 goto done;
593 } else {
594 /* Note: in this case, no need to wait */
595 memset(acb->buf, 0, 512 * acb->n);
596 goto redo;
598 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
599 /* add AIO support for compressed blocks ? */
600 if (decompress_cluster(s, acb->cluster_offset) < 0)
601 goto done;
602 memcpy(acb->buf,
603 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
604 goto redo;
605 } else {
606 if ((acb->cluster_offset & 511) != 0) {
607 ret = -EIO;
608 goto done;
610 acb->hd_iov.iov_base = (void *)acb->buf;
611 acb->hd_iov.iov_len = acb->n * 512;
612 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
613 acb->hd_aiocb = bdrv_aio_readv(s->hd,
614 (acb->cluster_offset >> 9) + index_in_cluster,
615 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
616 if (acb->hd_aiocb == NULL)
617 goto done;
620 return;
622 done:
623 if (acb->qiov->niov > 1) {
624 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
625 qemu_vfree(acb->orig_buf);
627 acb->common.cb(acb->common.opaque, ret);
628 qemu_aio_release(acb);
631 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
632 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
633 BlockDriverCompletionFunc *cb, void *opaque)
635 QCowAIOCB *acb;
637 acb = qemu_aio_get(bs, cb, opaque);
638 if (!acb)
639 return NULL;
640 acb->hd_aiocb = NULL;
641 acb->sector_num = sector_num;
642 acb->qiov = qiov;
643 if (qiov->niov > 1)
644 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
645 else
646 acb->buf = (uint8_t *)qiov->iov->iov_base;
647 acb->nb_sectors = nb_sectors;
648 acb->n = 0;
649 acb->cluster_offset = 0;
651 qcow_aio_read_cb(acb, 0);
652 return &acb->common;
655 static void qcow_aio_write_cb(void *opaque, int ret)
657 QCowAIOCB *acb = opaque;
658 BlockDriverState *bs = acb->common.bs;
659 BDRVQcowState *s = bs->opaque;
660 int index_in_cluster;
661 uint64_t cluster_offset;
662 const uint8_t *src_buf;
664 acb->hd_aiocb = NULL;
666 if (ret < 0)
667 goto done;
669 acb->nb_sectors -= acb->n;
670 acb->sector_num += acb->n;
671 acb->buf += acb->n * 512;
673 if (acb->nb_sectors == 0) {
674 /* request completed */
675 ret = 0;
676 goto done;
679 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
680 acb->n = s->cluster_sectors - index_in_cluster;
681 if (acb->n > acb->nb_sectors)
682 acb->n = acb->nb_sectors;
683 cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, 1, 0,
684 index_in_cluster,
685 index_in_cluster + acb->n);
686 if (!cluster_offset || (cluster_offset & 511) != 0) {
687 ret = -EIO;
688 goto done;
690 if (s->crypt_method) {
691 if (!acb->cluster_data) {
692 acb->cluster_data = qemu_mallocz(s->cluster_size);
693 if (!acb->cluster_data) {
694 ret = -ENOMEM;
695 goto done;
698 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
699 acb->n, 1, &s->aes_encrypt_key);
700 src_buf = acb->cluster_data;
701 } else {
702 src_buf = acb->buf;
705 acb->hd_iov.iov_base = (void *)src_buf;
706 acb->hd_iov.iov_len = acb->n * 512;
707 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
708 acb->hd_aiocb = bdrv_aio_writev(s->hd,
709 (cluster_offset >> 9) + index_in_cluster,
710 &acb->hd_qiov, acb->n,
711 qcow_aio_write_cb, acb);
712 if (acb->hd_aiocb == NULL)
713 goto done;
714 return;
716 done:
717 if (acb->qiov->niov > 1)
718 qemu_vfree(acb->orig_buf);
719 acb->common.cb(acb->common.opaque, ret);
720 qemu_aio_release(acb);
723 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
724 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
725 BlockDriverCompletionFunc *cb, void *opaque)
727 BDRVQcowState *s = bs->opaque;
728 QCowAIOCB *acb;
730 s->cluster_cache_offset = -1; /* disable compressed cache */
732 acb = qemu_aio_get(bs, cb, opaque);
733 if (!acb)
734 return NULL;
735 acb->hd_aiocb = NULL;
736 acb->sector_num = sector_num;
737 acb->qiov = qiov;
738 if (qiov->niov > 1) {
739 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
740 qemu_iovec_to_buffer(qiov, acb->buf);
741 } else {
742 acb->buf = (uint8_t *)qiov->iov->iov_base;
744 acb->nb_sectors = nb_sectors;
745 acb->n = 0;
747 qcow_aio_write_cb(acb, 0);
748 return &acb->common;
751 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
753 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
754 if (acb->hd_aiocb)
755 bdrv_aio_cancel(acb->hd_aiocb);
756 qemu_aio_release(acb);
759 static void qcow_close(BlockDriverState *bs)
761 BDRVQcowState *s = bs->opaque;
762 qemu_free(s->l1_table);
763 qemu_free(s->l2_cache);
764 qemu_free(s->cluster_cache);
765 qemu_free(s->cluster_data);
766 bdrv_delete(s->hd);
769 static int qcow_create(const char *filename, int64_t total_size,
770 const char *backing_file, int flags)
772 int fd, header_size, backing_filename_len, l1_size, i, shift;
773 QCowHeader header;
774 uint64_t tmp;
776 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
777 if (fd < 0)
778 return -1;
779 memset(&header, 0, sizeof(header));
780 header.magic = cpu_to_be32(QCOW_MAGIC);
781 header.version = cpu_to_be32(QCOW_VERSION);
782 header.size = cpu_to_be64(total_size * 512);
783 header_size = sizeof(header);
784 backing_filename_len = 0;
785 if (backing_file) {
786 if (strcmp(backing_file, "fat:")) {
787 header.backing_file_offset = cpu_to_be64(header_size);
788 backing_filename_len = strlen(backing_file);
789 header.backing_file_size = cpu_to_be32(backing_filename_len);
790 header_size += backing_filename_len;
791 } else {
792 /* special backing file for vvfat */
793 backing_file = NULL;
795 header.cluster_bits = 9; /* 512 byte cluster to avoid copying
796 unmodifyed sectors */
797 header.l2_bits = 12; /* 32 KB L2 tables */
798 } else {
799 header.cluster_bits = 12; /* 4 KB clusters */
800 header.l2_bits = 9; /* 4 KB L2 tables */
802 header_size = (header_size + 7) & ~7;
803 shift = header.cluster_bits + header.l2_bits;
804 l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
806 header.l1_table_offset = cpu_to_be64(header_size);
807 if (flags & BLOCK_FLAG_ENCRYPT) {
808 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
809 } else {
810 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
813 /* write all the data */
814 write(fd, &header, sizeof(header));
815 if (backing_file) {
816 write(fd, backing_file, backing_filename_len);
818 lseek(fd, header_size, SEEK_SET);
819 tmp = 0;
820 for(i = 0;i < l1_size; i++) {
821 write(fd, &tmp, sizeof(tmp));
823 close(fd);
824 return 0;
827 static int qcow_make_empty(BlockDriverState *bs)
829 BDRVQcowState *s = bs->opaque;
830 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
831 int ret;
833 memset(s->l1_table, 0, l1_length);
834 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
835 return -1;
836 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
837 if (ret < 0)
838 return ret;
840 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
841 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
842 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
844 return 0;
847 /* XXX: put compressed sectors first, then all the cluster aligned
848 tables to avoid losing bytes in alignment */
849 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
850 const uint8_t *buf, int nb_sectors)
852 BDRVQcowState *s = bs->opaque;
853 z_stream strm;
854 int ret, out_len;
855 uint8_t *out_buf;
856 uint64_t cluster_offset;
858 if (nb_sectors != s->cluster_sectors)
859 return -EINVAL;
861 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
862 if (!out_buf)
863 return -1;
865 /* best compression, small window, no zlib header */
866 memset(&strm, 0, sizeof(strm));
867 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
868 Z_DEFLATED, -12,
869 9, Z_DEFAULT_STRATEGY);
870 if (ret != 0) {
871 qemu_free(out_buf);
872 return -1;
875 strm.avail_in = s->cluster_size;
876 strm.next_in = (uint8_t *)buf;
877 strm.avail_out = s->cluster_size;
878 strm.next_out = out_buf;
880 ret = deflate(&strm, Z_FINISH);
881 if (ret != Z_STREAM_END && ret != Z_OK) {
882 qemu_free(out_buf);
883 deflateEnd(&strm);
884 return -1;
886 out_len = strm.next_out - out_buf;
888 deflateEnd(&strm);
890 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
891 /* could not compress: write normal cluster */
892 qcow_write(bs, sector_num, buf, s->cluster_sectors);
893 } else {
894 cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
895 out_len, 0, 0);
896 cluster_offset &= s->cluster_offset_mask;
897 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
898 qemu_free(out_buf);
899 return -1;
903 qemu_free(out_buf);
904 return 0;
907 static void qcow_flush(BlockDriverState *bs)
909 BDRVQcowState *s = bs->opaque;
910 bdrv_flush(s->hd);
913 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
915 BDRVQcowState *s = bs->opaque;
916 bdi->cluster_size = s->cluster_size;
917 return 0;
920 BlockDriver bdrv_qcow = {
921 .format_name = "qcow",
922 .instance_size = sizeof(BDRVQcowState),
923 .bdrv_probe = qcow_probe,
924 .bdrv_open = qcow_open,
925 .bdrv_close = qcow_close,
926 .bdrv_create = qcow_create,
927 .bdrv_flush = qcow_flush,
928 .bdrv_is_allocated = qcow_is_allocated,
929 .bdrv_set_key = qcow_set_key,
930 .bdrv_make_empty = qcow_make_empty,
931 .bdrv_aio_readv = qcow_aio_readv,
932 .bdrv_aio_writev = qcow_aio_writev,
933 .bdrv_aio_cancel = qcow_aio_cancel,
934 .aiocb_size = sizeof(QCowAIOCB),
935 .bdrv_write_compressed = qcow_write_compressed,
936 .bdrv_get_info = qcow_get_info,