]> git.proxmox.com Git - qemu.git/blame - block-qcow.c
removed duplicated line
[qemu.git] / block-qcow.c
CommitLineData
ea2384d3
FB
1/*
2 * Block driver for the QCOW format
3 *
83f64091 4 * Copyright (c) 2004-2006 Fabrice Bellard
ea2384d3
FB
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 "vl.h"
25#include "block_int.h"
28d34b82 26#include <zlib.h>
ea2384d3
FB
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
40typedef 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
55typedef struct BDRVQcowState {
83f64091 56 BlockDriverState *hd;
ea2384d3
FB
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
78static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
79
80static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
81{
82 const QCowHeader *cow_header = (const void *)buf;
712e7874
FB
83
84 if (buf_size >= sizeof(QCowHeader) &&
85 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
ea2384d3
FB
86 be32_to_cpu(cow_header->version) == QCOW_VERSION)
87 return 100;
88 else
89 return 0;
90}
91
83f64091 92static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
ea2384d3
FB
93{
94 BDRVQcowState *s = bs->opaque;
83f64091 95 int len, i, shift, ret;
ea2384d3 96 QCowHeader header;
83f64091
FB
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))
ea2384d3
FB
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;
83f64091 137 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
ea2384d3
FB
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;
83f64091 160 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
ea2384d3
FB
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);
83f64091 171 bdrv_delete(s->hd);
ea2384d3
FB
172 return -1;
173}
174
175static 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 */
220static 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 */
255static 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 */
83f64091 273 l2_offset = bdrv_getlength(s->hd);
ea2384d3
FB
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);
83f64091
FB
279 if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
280 &tmp, sizeof(tmp)) != sizeof(tmp))
ea2384d3
FB
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);
ea2384d3
FB
306 if (new_l2_table) {
307 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
83f64091 308 if (bdrv_pwrite(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
ea2384d3
FB
309 s->l2_size * sizeof(uint64_t))
310 return 0;
311 } else {
83f64091 312 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
ea2384d3
FB
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;
83f64091 333 cluster_offset = bdrv_getlength(s->hd);
ea2384d3
FB
334 cluster_offset = (cluster_offset + s->cluster_size - 1) &
335 ~(s->cluster_size - 1);
336 /* write the cluster content */
83f64091 337 if (bdrv_pwrite(s->hd, cluster_offset, s->cluster_cache, s->cluster_size) !=
ea2384d3
FB
338 s->cluster_size)
339 return -1;
340 } else {
83f64091 341 cluster_offset = bdrv_getlength(s->hd);
ea2384d3
FB
342 if (allocate == 1) {
343 /* round to cluster size */
344 cluster_offset = (cluster_offset + s->cluster_size - 1) &
345 ~(s->cluster_size - 1);
83f64091 346 bdrv_truncate(s->hd, cluster_offset + s->cluster_size);
ea2384d3
FB
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;
c47c33b0 353 memset(s->cluster_data + 512, 0x00, 512);
ea2384d3
FB
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);
83f64091
FB
360 if (bdrv_pwrite(s->hd, cluster_offset + i * 512,
361 s->cluster_data, 512) != 512)
ea2384d3
FB
362 return -1;
363 }
364 }
365 }
366 } else {
367 cluster_offset |= QCOW_OFLAG_COMPRESSED |
368 (uint64_t)compressed_size << (63 - s->cluster_bits);
369 }
370 }
371 /* update L2 table */
372 tmp = cpu_to_be64(cluster_offset);
373 l2_table[l2_index] = tmp;
83f64091
FB
374 if (bdrv_pwrite(s->hd,
375 l2_offset + l2_index * sizeof(tmp), &tmp, sizeof(tmp)) != sizeof(tmp))
ea2384d3
FB
376 return 0;
377 }
378 return cluster_offset;
379}
380
381static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
382 int nb_sectors, int *pnum)
383{
384 BDRVQcowState *s = bs->opaque;
385 int index_in_cluster, n;
386 uint64_t cluster_offset;
387
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);
395}
396
397static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
398 const uint8_t *buf, int buf_size)
399{
400 z_stream strm1, *strm = &strm1;
401 int ret, out_len;
402
403 memset(strm, 0, sizeof(*strm));
404
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;
409
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;
419 }
420 inflateEnd(strm);
421 return 0;
422}
423
424static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
425{
426 int ret, csize;
427 uint64_t coffset;
428
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);
83f64091 433 ret = bdrv_pread(s->hd, coffset, s->cluster_data, csize);
ea2384d3
FB
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;
439 }
440 s->cluster_cache_offset = coffset;
441 }
442 return 0;
443}
444
83f64091
FB
445#if 0
446
ea2384d3
FB
447static int qcow_read(BlockDriverState *bs, int64_t sector_num,
448 uint8_t *buf, int nb_sectors)
449{
450 BDRVQcowState *s = bs->opaque;
451 int ret, index_in_cluster, n;
452 uint64_t cluster_offset;
453
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) {
83f64091
FB
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);
468 }
ea2384d3
FB
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 {
83f64091 474 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
ea2384d3
FB
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);
480 }
481 }
482 nb_sectors -= n;
483 sector_num += n;
484 buf += n * 512;
485 }
486 return 0;
487}
83f64091 488#endif
ea2384d3
FB
489
490static int qcow_write(BlockDriverState *bs, int64_t sector_num,
491 const uint8_t *buf, int nb_sectors)
492{
493 BDRVQcowState *s = bs->opaque;
494 int ret, index_in_cluster, n;
495 uint64_t cluster_offset;
496
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;
ea2384d3
FB
507 if (s->crypt_method) {
508 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
509 &s->aes_encrypt_key);
83f64091
FB
510 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
511 s->cluster_data, n * 512);
ea2384d3 512 } else {
83f64091 513 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
ea2384d3
FB
514 }
515 if (ret != n * 512)
516 return -1;
517 nb_sectors -= n;
518 sector_num += n;
519 buf += n * 512;
520 }
521 s->cluster_cache_offset = -1; /* disable compressed cache */
522 return 0;
523}
524
83f64091
FB
525typedef struct {
526 int64_t sector_num;
527 uint8_t *buf;
528 int nb_sectors;
529 int n;
530 uint64_t cluster_offset;
531 uint8_t *cluster_data;
532 BlockDriverAIOCB *hd_aiocb;
533 BlockDriverAIOCB *backing_hd_aiocb;
534} QCowAIOCB;
535
536static void qcow_aio_delete(BlockDriverAIOCB *acb);
537
538static int qcow_aio_new(BlockDriverAIOCB *acb)
539{
540 BlockDriverState *bs = acb->bs;
541 BDRVQcowState *s = bs->opaque;
542 QCowAIOCB *acb1;
543 acb1 = qemu_mallocz(sizeof(QCowAIOCB));
544 if (!acb1)
545 return -1;
546 acb->opaque = acb1;
547 acb1->hd_aiocb = bdrv_aio_new(s->hd);
548 if (!acb1->hd_aiocb)
549 goto fail;
550 if (bs->backing_hd) {
551 acb1->backing_hd_aiocb = bdrv_aio_new(bs->backing_hd);
552 if (!acb1->backing_hd_aiocb)
553 goto fail;
554 }
555 return 0;
556 fail:
557 qcow_aio_delete(acb);
558 return -1;
559}
560
561static void qcow_aio_read_cb(void *opaque, int ret)
562{
563 BlockDriverAIOCB *acb = opaque;
564 BlockDriverState *bs = acb->bs;
565 BDRVQcowState *s = bs->opaque;
566 QCowAIOCB *acb1 = acb->opaque;
567 int index_in_cluster;
568
569 if (ret < 0) {
570 fail:
571 acb->cb(acb->cb_opaque, ret);
572 return;
573 }
574
575 redo:
576 /* post process the read buffer */
577 if (!acb1->cluster_offset) {
578 /* nothing to do */
579 } else if (acb1->cluster_offset & QCOW_OFLAG_COMPRESSED) {
580 /* nothing to do */
581 } else {
582 if (s->crypt_method) {
583 encrypt_sectors(s, acb1->sector_num, acb1->buf, acb1->buf,
584 acb1->n, 0,
585 &s->aes_decrypt_key);
586 }
587 }
588
589 acb1->nb_sectors -= acb1->n;
590 acb1->sector_num += acb1->n;
591 acb1->buf += acb1->n * 512;
592
593 if (acb1->nb_sectors == 0) {
594 /* request completed */
595 acb->cb(acb->cb_opaque, 0);
596 return;
597 }
598
599 /* prepare next AIO request */
600 acb1->cluster_offset = get_cluster_offset(bs,
601 acb1->sector_num << 9,
602 0, 0, 0, 0);
603 index_in_cluster = acb1->sector_num & (s->cluster_sectors - 1);
604 acb1->n = s->cluster_sectors - index_in_cluster;
605 if (acb1->n > acb1->nb_sectors)
606 acb1->n = acb1->nb_sectors;
607
608 if (!acb1->cluster_offset) {
609 if (bs->backing_hd) {
610 /* read from the base image */
611 ret = bdrv_aio_read(acb1->backing_hd_aiocb, acb1->sector_num,
612 acb1->buf, acb1->n, qcow_aio_read_cb, acb);
613 if (ret < 0)
614 goto fail;
615 } else {
616 /* Note: in this case, no need to wait */
617 memset(acb1->buf, 0, 512 * acb1->n);
618 goto redo;
619 }
620 } else if (acb1->cluster_offset & QCOW_OFLAG_COMPRESSED) {
621 /* add AIO support for compressed blocks ? */
622 if (decompress_cluster(s, acb1->cluster_offset) < 0)
623 goto fail;
624 memcpy(acb1->buf,
625 s->cluster_cache + index_in_cluster * 512, 512 * acb1->n);
626 goto redo;
627 } else {
628 if ((acb1->cluster_offset & 511) != 0) {
629 ret = -EIO;
630 goto fail;
631 }
632 ret = bdrv_aio_read(acb1->hd_aiocb,
633 (acb1->cluster_offset >> 9) + index_in_cluster,
634 acb1->buf, acb1->n, qcow_aio_read_cb, acb);
635 if (ret < 0)
636 goto fail;
637 }
638}
639
640static int qcow_aio_read(BlockDriverAIOCB *acb, int64_t sector_num,
641 uint8_t *buf, int nb_sectors)
642{
643 QCowAIOCB *acb1 = acb->opaque;
644
645 acb1->sector_num = sector_num;
646 acb1->buf = buf;
647 acb1->nb_sectors = nb_sectors;
648 acb1->n = 0;
649 acb1->cluster_offset = 0;
650
651 qcow_aio_read_cb(acb, 0);
e7b81015 652 return 0;
83f64091
FB
653}
654
655static void qcow_aio_write_cb(void *opaque, int ret)
656{
657 BlockDriverAIOCB *acb = opaque;
658 BlockDriverState *bs = acb->bs;
659 BDRVQcowState *s = bs->opaque;
660 QCowAIOCB *acb1 = acb->opaque;
661 int index_in_cluster;
662 uint64_t cluster_offset;
663 const uint8_t *src_buf;
664
665 if (ret < 0) {
666 fail:
667 acb->cb(acb->cb_opaque, ret);
668 return;
669 }
670
671 acb1->nb_sectors -= acb1->n;
672 acb1->sector_num += acb1->n;
673 acb1->buf += acb1->n * 512;
674
675 if (acb1->nb_sectors == 0) {
676 /* request completed */
677 acb->cb(acb->cb_opaque, 0);
678 return;
679 }
680
681 index_in_cluster = acb1->sector_num & (s->cluster_sectors - 1);
682 acb1->n = s->cluster_sectors - index_in_cluster;
683 if (acb1->n > acb1->nb_sectors)
684 acb1->n = acb1->nb_sectors;
685 cluster_offset = get_cluster_offset(bs, acb1->sector_num << 9, 1, 0,
686 index_in_cluster,
687 index_in_cluster + acb1->n);
688 if (!cluster_offset || (cluster_offset & 511) != 0) {
689 ret = -EIO;
690 goto fail;
691 }
692 if (s->crypt_method) {
693 if (!acb1->cluster_data) {
694 acb1->cluster_data = qemu_mallocz(s->cluster_size);
695 if (!acb1->cluster_data) {
696 ret = -ENOMEM;
697 goto fail;
698 }
699 }
700 encrypt_sectors(s, acb1->sector_num, acb1->cluster_data, acb1->buf,
701 acb1->n, 1, &s->aes_encrypt_key);
702 src_buf = acb1->cluster_data;
703 } else {
704 src_buf = acb1->buf;
705 }
706 ret = bdrv_aio_write(acb1->hd_aiocb,
707 (cluster_offset >> 9) + index_in_cluster,
708 src_buf, acb1->n,
709 qcow_aio_write_cb, acb);
710 if (ret < 0)
711 goto fail;
712}
713
714static int qcow_aio_write(BlockDriverAIOCB *acb, int64_t sector_num,
715 const uint8_t *buf, int nb_sectors)
716{
717 QCowAIOCB *acb1 = acb->opaque;
718 BlockDriverState *bs = acb->bs;
719 BDRVQcowState *s = bs->opaque;
720
721 s->cluster_cache_offset = -1; /* disable compressed cache */
722
723 acb1->sector_num = sector_num;
724 acb1->buf = (uint8_t *)buf;
725 acb1->nb_sectors = nb_sectors;
726 acb1->n = 0;
727
728 qcow_aio_write_cb(acb, 0);
e7b81015 729 return 0;
83f64091
FB
730}
731
732static void qcow_aio_cancel(BlockDriverAIOCB *acb)
733{
734 QCowAIOCB *acb1 = acb->opaque;
735 if (acb1->hd_aiocb)
736 bdrv_aio_cancel(acb1->hd_aiocb);
737 if (acb1->backing_hd_aiocb)
738 bdrv_aio_cancel(acb1->backing_hd_aiocb);
739}
740
741static void qcow_aio_delete(BlockDriverAIOCB *acb)
742{
743 QCowAIOCB *acb1 = acb->opaque;
744 if (acb1->hd_aiocb)
745 bdrv_aio_delete(acb1->hd_aiocb);
746 if (acb1->backing_hd_aiocb)
747 bdrv_aio_delete(acb1->backing_hd_aiocb);
748 qemu_free(acb1->cluster_data);
749 qemu_free(acb1);
750}
751
e2731add 752static void qcow_close(BlockDriverState *bs)
ea2384d3
FB
753{
754 BDRVQcowState *s = bs->opaque;
755 qemu_free(s->l1_table);
756 qemu_free(s->l2_cache);
757 qemu_free(s->cluster_cache);
758 qemu_free(s->cluster_data);
83f64091 759 bdrv_delete(s->hd);
ea2384d3
FB
760}
761
762static int qcow_create(const char *filename, int64_t total_size,
763 const char *backing_file, int flags)
764{
765 int fd, header_size, backing_filename_len, l1_size, i, shift;
766 QCowHeader header;
ea2384d3 767 uint64_t tmp;
ea2384d3 768
83f64091 769 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
ea2384d3
FB
770 if (fd < 0)
771 return -1;
772 memset(&header, 0, sizeof(header));
773 header.magic = cpu_to_be32(QCOW_MAGIC);
774 header.version = cpu_to_be32(QCOW_VERSION);
775 header.size = cpu_to_be64(total_size * 512);
776 header_size = sizeof(header);
777 backing_filename_len = 0;
778 if (backing_file) {
83f64091
FB
779 header.backing_file_offset = cpu_to_be64(header_size);
780 backing_filename_len = strlen(backing_file);
781 header.backing_file_size = cpu_to_be32(backing_filename_len);
782 header_size += backing_filename_len;
783 header.mtime = cpu_to_be32(0);
ea2384d3
FB
784 header.cluster_bits = 9; /* 512 byte cluster to avoid copying
785 unmodifyed sectors */
786 header.l2_bits = 12; /* 32 KB L2 tables */
787 } else {
788 header.cluster_bits = 12; /* 4 KB clusters */
789 header.l2_bits = 9; /* 4 KB L2 tables */
790 }
791 header_size = (header_size + 7) & ~7;
792 shift = header.cluster_bits + header.l2_bits;
793 l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
794
795 header.l1_table_offset = cpu_to_be64(header_size);
796 if (flags) {
797 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
798 } else {
799 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
800 }
801
802 /* write all the data */
803 write(fd, &header, sizeof(header));
804 if (backing_file) {
83f64091 805 write(fd, backing_file, backing_filename_len);
ea2384d3
FB
806 }
807 lseek(fd, header_size, SEEK_SET);
808 tmp = 0;
809 for(i = 0;i < l1_size; i++) {
810 write(fd, &tmp, sizeof(tmp));
811 }
812 close(fd);
813 return 0;
814}
815
c47c33b0 816static int qcow_make_empty(BlockDriverState *bs)
95389c86
FB
817{
818 BDRVQcowState *s = bs->opaque;
819 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
83f64091 820 int ret;
95389c86
FB
821
822 memset(s->l1_table, 0, l1_length);
83f64091 823 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
95389c86 824 return -1;
83f64091
FB
825 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
826 if (ret < 0)
827 return ret;
95389c86
FB
828
829 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
830 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
831 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
832
833 return 0;
834}
835
ea2384d3
FB
836/* XXX: put compressed sectors first, then all the cluster aligned
837 tables to avoid losing bytes in alignment */
c47c33b0
FB
838static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
839 const uint8_t *buf, int nb_sectors)
ea2384d3
FB
840{
841 BDRVQcowState *s = bs->opaque;
842 z_stream strm;
843 int ret, out_len;
844 uint8_t *out_buf;
845 uint64_t cluster_offset;
846
c47c33b0
FB
847 if (nb_sectors != s->cluster_sectors)
848 return -EINVAL;
ea2384d3
FB
849
850 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
851 if (!out_buf)
852 return -1;
853
854 /* best compression, small window, no zlib header */
855 memset(&strm, 0, sizeof(strm));
856 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
857 Z_DEFLATED, -12,
858 9, Z_DEFAULT_STRATEGY);
859 if (ret != 0) {
860 qemu_free(out_buf);
861 return -1;
862 }
863
864 strm.avail_in = s->cluster_size;
865 strm.next_in = (uint8_t *)buf;
866 strm.avail_out = s->cluster_size;
867 strm.next_out = out_buf;
868
869 ret = deflate(&strm, Z_FINISH);
870 if (ret != Z_STREAM_END && ret != Z_OK) {
871 qemu_free(out_buf);
872 deflateEnd(&strm);
873 return -1;
874 }
875 out_len = strm.next_out - out_buf;
876
877 deflateEnd(&strm);
878
879 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
880 /* could not compress: write normal cluster */
881 qcow_write(bs, sector_num, buf, s->cluster_sectors);
882 } else {
883 cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
884 out_len, 0, 0);
885 cluster_offset &= s->cluster_offset_mask;
83f64091 886 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
ea2384d3
FB
887 qemu_free(out_buf);
888 return -1;
889 }
890 }
891
892 qemu_free(out_buf);
893 return 0;
894}
895
7a6cba61
PB
896static void qcow_flush(BlockDriverState *bs)
897{
898 BDRVQcowState *s = bs->opaque;
83f64091 899 bdrv_flush(s->hd);
7a6cba61
PB
900}
901
c47c33b0
FB
902static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
903{
904 BDRVQcowState *s = bs->opaque;
905 bdi->cluster_size = s->cluster_size;
906 return 0;
907}
908
ea2384d3
FB
909BlockDriver bdrv_qcow = {
910 "qcow",
911 sizeof(BDRVQcowState),
912 qcow_probe,
913 qcow_open,
83f64091
FB
914 NULL,
915 NULL,
ea2384d3
FB
916 qcow_close,
917 qcow_create,
7a6cba61 918 qcow_flush,
ea2384d3
FB
919 qcow_is_allocated,
920 qcow_set_key,
83f64091
FB
921 qcow_make_empty,
922
923 .bdrv_aio_new = qcow_aio_new,
924 .bdrv_aio_read = qcow_aio_read,
925 .bdrv_aio_write = qcow_aio_write,
926 .bdrv_aio_cancel = qcow_aio_cancel,
927 .bdrv_aio_delete = qcow_aio_delete,
c47c33b0
FB
928 .bdrv_write_compressed = qcow_write_compressed,
929 .bdrv_get_info = qcow_get_info,
ea2384d3 930};