]> git.proxmox.com Git - qemu.git/blame - block/qcow.c
Merge remote-tracking branch 'kwolf/for-anthony' into staging
[qemu.git] / block / qcow.c
CommitLineData
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
41typedef 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
56typedef struct BDRVQcowState {
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
66f82cee 78static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
ea2384d3
FB
79
80static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
81{
82 const QCowHeader *cow_header = (const void *)buf;
3b46e624 83
712e7874
FB
84 if (buf_size >= sizeof(QCowHeader) &&
85 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
5fafdf24 86 be32_to_cpu(cow_header->version) == QCOW_VERSION)
ea2384d3
FB
87 return 100;
88 else
89 return 0;
90}
91
66f82cee 92static int qcow_open(BlockDriverState *bs, int flags)
ea2384d3
FB
93{
94 BDRVQcowState *s = bs->opaque;
66f82cee 95 int len, i, shift;
ea2384d3 96 QCowHeader header;
83f64091 97
66f82cee 98 if (bdrv_pread(bs->file, 0, &header, sizeof(header)) != sizeof(header))
ea2384d3
FB
99 goto fail;
100 be32_to_cpus(&header.magic);
101 be32_to_cpus(&header.version);
102 be64_to_cpus(&header.backing_file_offset);
103 be32_to_cpus(&header.backing_file_size);
104 be32_to_cpus(&header.mtime);
105 be64_to_cpus(&header.size);
106 be32_to_cpus(&header.crypt_method);
107 be64_to_cpus(&header.l1_table_offset);
3b46e624 108
ea2384d3
FB
109 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
110 goto fail;
111 if (header.size <= 1 || header.cluster_bits < 9)
112 goto fail;
113 if (header.crypt_method > QCOW_CRYPT_AES)
114 goto fail;
115 s->crypt_method_header = header.crypt_method;
116 if (s->crypt_method_header)
117 bs->encrypted = 1;
118 s->cluster_bits = header.cluster_bits;
119 s->cluster_size = 1 << s->cluster_bits;
120 s->cluster_sectors = 1 << (s->cluster_bits - 9);
121 s->l2_bits = header.l2_bits;
122 s->l2_size = 1 << s->l2_bits;
123 bs->total_sectors = header.size / 512;
124 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
125
126 /* read the level 1 table */
127 shift = s->cluster_bits + s->l2_bits;
128 s->l1_size = (header.size + (1LL << shift) - 1) >> shift;
129
130 s->l1_table_offset = header.l1_table_offset;
131 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
132 if (!s->l1_table)
133 goto fail;
66f82cee 134 if (bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
ea2384d3
FB
135 s->l1_size * sizeof(uint64_t))
136 goto fail;
137 for(i = 0;i < s->l1_size; i++) {
138 be64_to_cpus(&s->l1_table[i]);
139 }
140 /* alloc L2 cache */
141 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
142 if (!s->l2_cache)
143 goto fail;
144 s->cluster_cache = qemu_malloc(s->cluster_size);
145 if (!s->cluster_cache)
146 goto fail;
147 s->cluster_data = qemu_malloc(s->cluster_size);
148 if (!s->cluster_data)
149 goto fail;
150 s->cluster_cache_offset = -1;
3b46e624 151
ea2384d3
FB
152 /* read the backing file name */
153 if (header.backing_file_offset != 0) {
154 len = header.backing_file_size;
155 if (len > 1023)
156 len = 1023;
66f82cee 157 if (bdrv_pread(bs->file, header.backing_file_offset, bs->backing_file, len) != len)
ea2384d3
FB
158 goto fail;
159 bs->backing_file[len] = '\0';
160 }
161 return 0;
162
163 fail:
164 qemu_free(s->l1_table);
165 qemu_free(s->l2_cache);
166 qemu_free(s->cluster_cache);
167 qemu_free(s->cluster_data);
ea2384d3
FB
168 return -1;
169}
170
171static int qcow_set_key(BlockDriverState *bs, const char *key)
172{
173 BDRVQcowState *s = bs->opaque;
174 uint8_t keybuf[16];
175 int len, i;
3b46e624 176
ea2384d3
FB
177 memset(keybuf, 0, 16);
178 len = strlen(key);
179 if (len > 16)
180 len = 16;
181 /* XXX: we could compress the chars to 7 bits to increase
182 entropy */
183 for(i = 0;i < len;i++) {
184 keybuf[i] = key[i];
185 }
186 s->crypt_method = s->crypt_method_header;
187
188 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
189 return -1;
190 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
191 return -1;
192#if 0
193 /* test */
194 {
195 uint8_t in[16];
196 uint8_t out[16];
197 uint8_t tmp[16];
198 for(i=0;i<16;i++)
199 in[i] = i;
200 AES_encrypt(in, tmp, &s->aes_encrypt_key);
201 AES_decrypt(tmp, out, &s->aes_decrypt_key);
202 for(i = 0; i < 16; i++)
203 printf(" %02x", tmp[i]);
204 printf("\n");
205 for(i = 0; i < 16; i++)
206 printf(" %02x", out[i]);
207 printf("\n");
208 }
209#endif
210 return 0;
211}
212
213/* The crypt function is compatible with the linux cryptoloop
214 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
215 supported */
216static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
217 uint8_t *out_buf, const uint8_t *in_buf,
218 int nb_sectors, int enc,
219 const AES_KEY *key)
220{
221 union {
222 uint64_t ll[2];
223 uint8_t b[16];
224 } ivec;
225 int i;
226
227 for(i = 0; i < nb_sectors; i++) {
228 ivec.ll[0] = cpu_to_le64(sector_num);
229 ivec.ll[1] = 0;
5fafdf24 230 AES_cbc_encrypt(in_buf, out_buf, 512, key,
ea2384d3
FB
231 ivec.b, enc);
232 sector_num++;
233 in_buf += 512;
234 out_buf += 512;
235 }
236}
237
238/* 'allocate' is:
239 *
240 * 0 to not allocate.
241 *
242 * 1 to allocate a normal cluster (for sector indexes 'n_start' to
243 * 'n_end')
244 *
245 * 2 to allocate a compressed cluster of size
246 * 'compressed_size'. 'compressed_size' must be > 0 and <
5fafdf24 247 * cluster_size
ea2384d3
FB
248 *
249 * return 0 if not allocated.
250 */
251static uint64_t get_cluster_offset(BlockDriverState *bs,
252 uint64_t offset, int allocate,
253 int compressed_size,
254 int n_start, int n_end)
255{
256 BDRVQcowState *s = bs->opaque;
257 int min_index, i, j, l1_index, l2_index;
258 uint64_t l2_offset, *l2_table, cluster_offset, tmp;
259 uint32_t min_count;
260 int new_l2_table;
3b46e624 261
ea2384d3
FB
262 l1_index = offset >> (s->l2_bits + s->cluster_bits);
263 l2_offset = s->l1_table[l1_index];
264 new_l2_table = 0;
265 if (!l2_offset) {
266 if (!allocate)
267 return 0;
268 /* allocate a new l2 entry */
66f82cee 269 l2_offset = bdrv_getlength(bs->file);
ea2384d3
FB
270 /* round to cluster size */
271 l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
272 /* update the L1 entry */
273 s->l1_table[l1_index] = l2_offset;
274 tmp = cpu_to_be64(l2_offset);
5e5557d9
KW
275 if (bdrv_pwrite_sync(bs->file,
276 s->l1_table_offset + l1_index * sizeof(tmp),
277 &tmp, sizeof(tmp)) < 0)
ea2384d3
FB
278 return 0;
279 new_l2_table = 1;
280 }
281 for(i = 0; i < L2_CACHE_SIZE; i++) {
282 if (l2_offset == s->l2_cache_offsets[i]) {
283 /* increment the hit count */
284 if (++s->l2_cache_counts[i] == 0xffffffff) {
285 for(j = 0; j < L2_CACHE_SIZE; j++) {
286 s->l2_cache_counts[j] >>= 1;
287 }
288 }
289 l2_table = s->l2_cache + (i << s->l2_bits);
290 goto found;
291 }
292 }
293 /* not found: load a new entry in the least used one */
294 min_index = 0;
295 min_count = 0xffffffff;
296 for(i = 0; i < L2_CACHE_SIZE; i++) {
297 if (s->l2_cache_counts[i] < min_count) {
298 min_count = s->l2_cache_counts[i];
299 min_index = i;
300 }
301 }
302 l2_table = s->l2_cache + (min_index << s->l2_bits);
ea2384d3
FB
303 if (new_l2_table) {
304 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
5e5557d9
KW
305 if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
306 s->l2_size * sizeof(uint64_t)) < 0)
ea2384d3
FB
307 return 0;
308 } else {
66f82cee 309 if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
ea2384d3
FB
310 s->l2_size * sizeof(uint64_t))
311 return 0;
312 }
313 s->l2_cache_offsets[min_index] = l2_offset;
314 s->l2_cache_counts[min_index] = 1;
315 found:
316 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
317 cluster_offset = be64_to_cpu(l2_table[l2_index]);
5fafdf24 318 if (!cluster_offset ||
ea2384d3
FB
319 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
320 if (!allocate)
321 return 0;
322 /* allocate a new cluster */
323 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
324 (n_end - n_start) < s->cluster_sectors) {
325 /* if the cluster is already compressed, we must
326 decompress it in the case it is not completely
327 overwritten */
66f82cee 328 if (decompress_cluster(bs, cluster_offset) < 0)
ea2384d3 329 return 0;
66f82cee 330 cluster_offset = bdrv_getlength(bs->file);
5fafdf24 331 cluster_offset = (cluster_offset + s->cluster_size - 1) &
ea2384d3
FB
332 ~(s->cluster_size - 1);
333 /* write the cluster content */
66f82cee 334 if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, s->cluster_size) !=
ea2384d3
FB
335 s->cluster_size)
336 return -1;
337 } else {
66f82cee 338 cluster_offset = bdrv_getlength(bs->file);
7f48fa1f
AL
339 if (allocate == 1) {
340 /* round to cluster size */
341 cluster_offset = (cluster_offset + s->cluster_size - 1) &
342 ~(s->cluster_size - 1);
66f82cee 343 bdrv_truncate(bs->file, cluster_offset + s->cluster_size);
7f48fa1f
AL
344 /* if encrypted, we must initialize the cluster
345 content which won't be written */
346 if (s->crypt_method &&
347 (n_end - n_start) < s->cluster_sectors) {
348 uint64_t start_sect;
349 start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
350 memset(s->cluster_data + 512, 0x00, 512);
351 for(i = 0; i < s->cluster_sectors; i++) {
352 if (i < n_start || i >= n_end) {
353 encrypt_sectors(s, start_sect + i,
354 s->cluster_data,
355 s->cluster_data + 512, 1, 1,
356 &s->aes_encrypt_key);
66f82cee 357 if (bdrv_pwrite(bs->file, cluster_offset + i * 512,
7f48fa1f
AL
358 s->cluster_data, 512) != 512)
359 return -1;
360 }
ea2384d3
FB
361 }
362 }
7f48fa1f
AL
363 } else if (allocate == 2) {
364 cluster_offset |= QCOW_OFLAG_COMPRESSED |
365 (uint64_t)compressed_size << (63 - s->cluster_bits);
ea2384d3
FB
366 }
367 }
368 /* update L2 table */
369 tmp = cpu_to_be64(cluster_offset);
370 l2_table[l2_index] = tmp;
5e5557d9
KW
371 if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
372 &tmp, sizeof(tmp)) < 0)
ea2384d3
FB
373 return 0;
374 }
375 return cluster_offset;
376}
377
5fafdf24 378static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
ea2384d3
FB
379 int nb_sectors, int *pnum)
380{
381 BDRVQcowState *s = bs->opaque;
382 int index_in_cluster, n;
383 uint64_t cluster_offset;
384
385 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
386 index_in_cluster = sector_num & (s->cluster_sectors - 1);
387 n = s->cluster_sectors - index_in_cluster;
388 if (n > nb_sectors)
389 n = nb_sectors;
390 *pnum = n;
391 return (cluster_offset != 0);
392}
393
394static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
395 const uint8_t *buf, int buf_size)
396{
397 z_stream strm1, *strm = &strm1;
398 int ret, out_len;
399
400 memset(strm, 0, sizeof(*strm));
401
402 strm->next_in = (uint8_t *)buf;
403 strm->avail_in = buf_size;
404 strm->next_out = out_buf;
405 strm->avail_out = out_buf_size;
406
407 ret = inflateInit2(strm, -12);
408 if (ret != Z_OK)
409 return -1;
410 ret = inflate(strm, Z_FINISH);
411 out_len = strm->next_out - out_buf;
412 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
413 out_len != out_buf_size) {
414 inflateEnd(strm);
415 return -1;
416 }
417 inflateEnd(strm);
418 return 0;
419}
3b46e624 420
66f82cee 421static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
ea2384d3 422{
66f82cee 423 BDRVQcowState *s = bs->opaque;
ea2384d3
FB
424 int ret, csize;
425 uint64_t coffset;
426
427 coffset = cluster_offset & s->cluster_offset_mask;
428 if (s->cluster_cache_offset != coffset) {
429 csize = cluster_offset >> (63 - s->cluster_bits);
430 csize &= (s->cluster_size - 1);
66f82cee 431 ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
5fafdf24 432 if (ret != csize)
ea2384d3
FB
433 return -1;
434 if (decompress_buffer(s->cluster_cache, s->cluster_size,
435 s->cluster_data, csize) < 0) {
436 return -1;
437 }
438 s->cluster_cache_offset = coffset;
439 }
440 return 0;
441}
442
83f64091
FB
443#if 0
444
5fafdf24 445static int qcow_read(BlockDriverState *bs, int64_t sector_num,
ea2384d3
FB
446 uint8_t *buf, int nb_sectors)
447{
448 BDRVQcowState *s = bs->opaque;
449 int ret, index_in_cluster, n;
450 uint64_t cluster_offset;
3b46e624 451
ea2384d3
FB
452 while (nb_sectors > 0) {
453 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
454 index_in_cluster = sector_num & (s->cluster_sectors - 1);
455 n = s->cluster_sectors - index_in_cluster;
456 if (n > nb_sectors)
457 n = nb_sectors;
458 if (!cluster_offset) {
83f64091
FB
459 if (bs->backing_hd) {
460 /* read from the base image */
461 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
462 if (ret < 0)
463 return -1;
464 } else {
465 memset(buf, 0, 512 * n);
466 }
ea2384d3 467 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
66f82cee 468 if (decompress_cluster(bs, cluster_offset) < 0)
ea2384d3
FB
469 return -1;
470 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
471 } else {
66f82cee 472 ret = bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512);
5fafdf24 473 if (ret != n * 512)
ea2384d3
FB
474 return -1;
475 if (s->crypt_method) {
5fafdf24 476 encrypt_sectors(s, sector_num, buf, buf, n, 0,
ea2384d3
FB
477 &s->aes_decrypt_key);
478 }
479 }
480 nb_sectors -= n;
481 sector_num += n;
482 buf += n * 512;
483 }
484 return 0;
485}
83f64091 486#endif
ea2384d3 487
ce1a14dc
PB
488typedef struct QCowAIOCB {
489 BlockDriverAIOCB common;
83f64091 490 int64_t sector_num;
f141eafe 491 QEMUIOVector *qiov;
83f64091 492 uint8_t *buf;
f141eafe 493 void *orig_buf;
83f64091
FB
494 int nb_sectors;
495 int n;
496 uint64_t cluster_offset;
5fafdf24 497 uint8_t *cluster_data;
c87c0672 498 struct iovec hd_iov;
b11a24de
KW
499 bool is_write;
500 QEMUBH *bh;
c87c0672 501 QEMUIOVector hd_qiov;
83f64091 502 BlockDriverAIOCB *hd_aiocb;
83f64091
FB
503} QCowAIOCB;
504
c16b5a2c
CH
505static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
506{
b666d239 507 QCowAIOCB *acb = container_of(blockacb, QCowAIOCB, common);
c16b5a2c
CH
508 if (acb->hd_aiocb)
509 bdrv_aio_cancel(acb->hd_aiocb);
510 qemu_aio_release(acb);
511}
512
513static AIOPool qcow_aio_pool = {
514 .aiocb_size = sizeof(QCowAIOCB),
515 .cancel = qcow_aio_cancel,
516};
ad53089b
CH
517
518static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
519 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
520 BlockDriverCompletionFunc *cb, void *opaque, int is_write)
521{
522 QCowAIOCB *acb;
523
c16b5a2c 524 acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
ad53089b
CH
525 if (!acb)
526 return NULL;
527 acb->hd_aiocb = NULL;
528 acb->sector_num = sector_num;
529 acb->qiov = qiov;
b11a24de
KW
530 acb->is_write = is_write;
531
ad53089b
CH
532 if (qiov->niov > 1) {
533 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
534 if (is_write)
535 qemu_iovec_to_buffer(qiov, acb->buf);
536 } else {
537 acb->buf = (uint8_t *)qiov->iov->iov_base;
538 }
539 acb->nb_sectors = nb_sectors;
540 acb->n = 0;
541 acb->cluster_offset = 0;
542 return acb;
543}
544
b11a24de
KW
545static void qcow_aio_read_cb(void *opaque, int ret);
546static void qcow_aio_write_cb(void *opaque, int ret);
547
548static void qcow_aio_rw_bh(void *opaque)
549{
550 QCowAIOCB *acb = opaque;
551 qemu_bh_delete(acb->bh);
552 acb->bh = NULL;
553
554 if (acb->is_write) {
555 qcow_aio_write_cb(opaque, 0);
556 } else {
557 qcow_aio_read_cb(opaque, 0);
558 }
559}
560
561static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
562{
563 if (acb->bh) {
564 return -EIO;
565 }
566
567 acb->bh = qemu_bh_new(cb, acb);
568 if (!acb->bh) {
569 return -EIO;
570 }
571
572 qemu_bh_schedule(acb->bh);
573
574 return 0;
575}
576
83f64091
FB
577static void qcow_aio_read_cb(void *opaque, int ret)
578{
ce1a14dc
PB
579 QCowAIOCB *acb = opaque;
580 BlockDriverState *bs = acb->common.bs;
83f64091 581 BDRVQcowState *s = bs->opaque;
83f64091
FB
582 int index_in_cluster;
583
ce1a14dc 584 acb->hd_aiocb = NULL;
f141eafe
AL
585 if (ret < 0)
586 goto done;
83f64091
FB
587
588 redo:
589 /* post process the read buffer */
ce1a14dc 590 if (!acb->cluster_offset) {
83f64091 591 /* nothing to do */
ce1a14dc 592 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
83f64091
FB
593 /* nothing to do */
594 } else {
595 if (s->crypt_method) {
5fafdf24
TS
596 encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
597 acb->n, 0,
83f64091
FB
598 &s->aes_decrypt_key);
599 }
600 }
601
ce1a14dc
PB
602 acb->nb_sectors -= acb->n;
603 acb->sector_num += acb->n;
604 acb->buf += acb->n * 512;
83f64091 605
ce1a14dc 606 if (acb->nb_sectors == 0) {
83f64091 607 /* request completed */
f141eafe
AL
608 ret = 0;
609 goto done;
83f64091 610 }
3b46e624 611
83f64091 612 /* prepare next AIO request */
5fafdf24 613 acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9,
ce1a14dc
PB
614 0, 0, 0, 0);
615 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
616 acb->n = s->cluster_sectors - index_in_cluster;
617 if (acb->n > acb->nb_sectors)
618 acb->n = acb->nb_sectors;
619
620 if (!acb->cluster_offset) {
83f64091
FB
621 if (bs->backing_hd) {
622 /* read from the base image */
3f4cb3d3 623 acb->hd_iov.iov_base = (void *)acb->buf;
c87c0672
AL
624 acb->hd_iov.iov_len = acb->n * 512;
625 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
626 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
627 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
5614c188
SW
628 if (acb->hd_aiocb == NULL) {
629 ret = -EIO;
f141eafe 630 goto done;
5614c188 631 }
83f64091
FB
632 } else {
633 /* Note: in this case, no need to wait */
ce1a14dc 634 memset(acb->buf, 0, 512 * acb->n);
83f64091
FB
635 goto redo;
636 }
ce1a14dc 637 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
83f64091 638 /* add AIO support for compressed blocks ? */
5614c188
SW
639 if (decompress_cluster(bs, acb->cluster_offset) < 0) {
640 ret = -EIO;
f141eafe 641 goto done;
5614c188 642 }
5fafdf24 643 memcpy(acb->buf,
ce1a14dc 644 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
83f64091
FB
645 goto redo;
646 } else {
ce1a14dc 647 if ((acb->cluster_offset & 511) != 0) {
83f64091 648 ret = -EIO;
f141eafe 649 goto done;
83f64091 650 }
3f4cb3d3 651 acb->hd_iov.iov_base = (void *)acb->buf;
c87c0672
AL
652 acb->hd_iov.iov_len = acb->n * 512;
653 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
66f82cee 654 acb->hd_aiocb = bdrv_aio_readv(bs->file,
5fafdf24 655 (acb->cluster_offset >> 9) + index_in_cluster,
c87c0672 656 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
5614c188
SW
657 if (acb->hd_aiocb == NULL) {
658 ret = -EIO;
f141eafe 659 goto done;
5614c188 660 }
f141eafe
AL
661 }
662
663 return;
664
665done:
666 if (acb->qiov->niov > 1) {
667 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
668 qemu_vfree(acb->orig_buf);
83f64091 669 }
f141eafe
AL
670 acb->common.cb(acb->common.opaque, ret);
671 qemu_aio_release(acb);
83f64091
FB
672}
673
f141eafe
AL
674static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
675 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
ce1a14dc 676 BlockDriverCompletionFunc *cb, void *opaque)
83f64091 677{
ce1a14dc 678 QCowAIOCB *acb;
b11a24de 679 int ret;
ce1a14dc 680
ad53089b 681 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
ce1a14dc
PB
682 if (!acb)
683 return NULL;
83f64091 684
b11a24de
KW
685 ret = qcow_schedule_bh(qcow_aio_rw_bh, acb);
686 if (ret < 0) {
687 if (acb->qiov->niov > 1) {
688 qemu_vfree(acb->orig_buf);
689 }
690 qemu_aio_release(acb);
691 return NULL;
692 }
693
ce1a14dc 694 return &acb->common;
83f64091
FB
695}
696
697static void qcow_aio_write_cb(void *opaque, int ret)
698{
ce1a14dc
PB
699 QCowAIOCB *acb = opaque;
700 BlockDriverState *bs = acb->common.bs;
83f64091 701 BDRVQcowState *s = bs->opaque;
83f64091
FB
702 int index_in_cluster;
703 uint64_t cluster_offset;
704 const uint8_t *src_buf;
ce1a14dc
PB
705
706 acb->hd_aiocb = NULL;
707
f141eafe
AL
708 if (ret < 0)
709 goto done;
83f64091 710
ce1a14dc
PB
711 acb->nb_sectors -= acb->n;
712 acb->sector_num += acb->n;
713 acb->buf += acb->n * 512;
83f64091 714
ce1a14dc 715 if (acb->nb_sectors == 0) {
83f64091 716 /* request completed */
f141eafe
AL
717 ret = 0;
718 goto done;
83f64091 719 }
3b46e624 720
ce1a14dc
PB
721 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
722 acb->n = s->cluster_sectors - index_in_cluster;
723 if (acb->n > acb->nb_sectors)
724 acb->n = acb->nb_sectors;
5fafdf24
TS
725 cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, 1, 0,
726 index_in_cluster,
ce1a14dc 727 index_in_cluster + acb->n);
83f64091
FB
728 if (!cluster_offset || (cluster_offset & 511) != 0) {
729 ret = -EIO;
f141eafe 730 goto done;
83f64091
FB
731 }
732 if (s->crypt_method) {
ce1a14dc
PB
733 if (!acb->cluster_data) {
734 acb->cluster_data = qemu_mallocz(s->cluster_size);
735 if (!acb->cluster_data) {
83f64091 736 ret = -ENOMEM;
f141eafe 737 goto done;
83f64091
FB
738 }
739 }
5fafdf24 740 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
ce1a14dc
PB
741 acb->n, 1, &s->aes_encrypt_key);
742 src_buf = acb->cluster_data;
83f64091 743 } else {
ce1a14dc 744 src_buf = acb->buf;
83f64091 745 }
c87c0672
AL
746
747 acb->hd_iov.iov_base = (void *)src_buf;
748 acb->hd_iov.iov_len = acb->n * 512;
749 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
66f82cee 750 acb->hd_aiocb = bdrv_aio_writev(bs->file,
c87c0672
AL
751 (cluster_offset >> 9) + index_in_cluster,
752 &acb->hd_qiov, acb->n,
753 qcow_aio_write_cb, acb);
5614c188
SW
754 if (acb->hd_aiocb == NULL) {
755 ret = -EIO;
f141eafe 756 goto done;
5614c188 757 }
f141eafe
AL
758 return;
759
760done:
761 if (acb->qiov->niov > 1)
762 qemu_vfree(acb->orig_buf);
763 acb->common.cb(acb->common.opaque, ret);
764 qemu_aio_release(acb);
83f64091
FB
765}
766
f141eafe
AL
767static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
768 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
ce1a14dc 769 BlockDriverCompletionFunc *cb, void *opaque)
83f64091 770{
83f64091 771 BDRVQcowState *s = bs->opaque;
ce1a14dc 772 QCowAIOCB *acb;
b11a24de 773 int ret;
3b46e624 774
83f64091
FB
775 s->cluster_cache_offset = -1; /* disable compressed cache */
776
c53ffce9 777 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
ce1a14dc
PB
778 if (!acb)
779 return NULL;
ad53089b 780
3b46e624 781
b11a24de
KW
782 ret = qcow_schedule_bh(qcow_aio_rw_bh, acb);
783 if (ret < 0) {
784 if (acb->qiov->niov > 1) {
785 qemu_vfree(acb->orig_buf);
786 }
787 qemu_aio_release(acb);
788 return NULL;
789 }
790
ce1a14dc 791 return &acb->common;
83f64091
FB
792}
793
e2731add 794static void qcow_close(BlockDriverState *bs)
ea2384d3
FB
795{
796 BDRVQcowState *s = bs->opaque;
797 qemu_free(s->l1_table);
798 qemu_free(s->l2_cache);
799 qemu_free(s->cluster_cache);
800 qemu_free(s->cluster_data);
ea2384d3
FB
801}
802
0e7e1989 803static int qcow_create(const char *filename, QEMUOptionParameter *options)
ea2384d3
FB
804{
805 int fd, header_size, backing_filename_len, l1_size, i, shift;
806 QCowHeader header;
ea2384d3 807 uint64_t tmp;
0e7e1989
KW
808 int64_t total_size = 0;
809 const char *backing_file = NULL;
810 int flags = 0;
3e1a8134 811 int ret;
0e7e1989
KW
812
813 /* Read out options */
814 while (options && options->name) {
815 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
816 total_size = options->value.n / 512;
817 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
818 backing_file = options->value.s;
819 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
820 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
821 }
822 options++;
823 }
ea2384d3 824
83f64091 825 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
ea2384d3 826 if (fd < 0)
98c2b2f4 827 return -errno;
ea2384d3
FB
828 memset(&header, 0, sizeof(header));
829 header.magic = cpu_to_be32(QCOW_MAGIC);
830 header.version = cpu_to_be32(QCOW_VERSION);
831 header.size = cpu_to_be64(total_size * 512);
832 header_size = sizeof(header);
833 backing_filename_len = 0;
834 if (backing_file) {
7852e5da
AJ
835 if (strcmp(backing_file, "fat:")) {
836 header.backing_file_offset = cpu_to_be64(header_size);
837 backing_filename_len = strlen(backing_file);
838 header.backing_file_size = cpu_to_be32(backing_filename_len);
839 header_size += backing_filename_len;
840 } else {
841 /* special backing file for vvfat */
842 backing_file = NULL;
843 }
ea2384d3
FB
844 header.cluster_bits = 9; /* 512 byte cluster to avoid copying
845 unmodifyed sectors */
846 header.l2_bits = 12; /* 32 KB L2 tables */
847 } else {
848 header.cluster_bits = 12; /* 4 KB clusters */
849 header.l2_bits = 9; /* 4 KB L2 tables */
850 }
851 header_size = (header_size + 7) & ~7;
852 shift = header.cluster_bits + header.l2_bits;
853 l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
854
855 header.l1_table_offset = cpu_to_be64(header_size);
ec36ba14 856 if (flags & BLOCK_FLAG_ENCRYPT) {
ea2384d3
FB
857 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
858 } else {
859 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
860 }
3b46e624 861
ea2384d3 862 /* write all the data */
3e1a8134
KS
863 ret = qemu_write_full(fd, &header, sizeof(header));
864 if (ret != sizeof(header)) {
98c2b2f4 865 ret = -errno;
3e1a8134
KS
866 goto exit;
867 }
868
ea2384d3 869 if (backing_file) {
3e1a8134
KS
870 ret = qemu_write_full(fd, backing_file, backing_filename_len);
871 if (ret != backing_filename_len) {
98c2b2f4 872 ret = -errno;
3e1a8134
KS
873 goto exit;
874 }
875
ea2384d3
FB
876 }
877 lseek(fd, header_size, SEEK_SET);
878 tmp = 0;
879 for(i = 0;i < l1_size; i++) {
3e1a8134
KS
880 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
881 if (ret != sizeof(tmp)) {
98c2b2f4 882 ret = -errno;
3e1a8134
KS
883 goto exit;
884 }
ea2384d3 885 }
3e1a8134
KS
886
887 ret = 0;
888exit:
ea2384d3 889 close(fd);
3e1a8134 890 return ret;
ea2384d3
FB
891}
892
c47c33b0 893static int qcow_make_empty(BlockDriverState *bs)
95389c86
FB
894{
895 BDRVQcowState *s = bs->opaque;
896 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
83f64091 897 int ret;
95389c86
FB
898
899 memset(s->l1_table, 0, l1_length);
5e5557d9
KW
900 if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
901 l1_length) < 0)
902 return -1;
66f82cee 903 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
83f64091
FB
904 if (ret < 0)
905 return ret;
95389c86
FB
906
907 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
908 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
909 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
910
911 return 0;
912}
913
ea2384d3
FB
914/* XXX: put compressed sectors first, then all the cluster aligned
915 tables to avoid losing bytes in alignment */
5fafdf24 916static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
c47c33b0 917 const uint8_t *buf, int nb_sectors)
ea2384d3
FB
918{
919 BDRVQcowState *s = bs->opaque;
920 z_stream strm;
921 int ret, out_len;
922 uint8_t *out_buf;
923 uint64_t cluster_offset;
924
c47c33b0
FB
925 if (nb_sectors != s->cluster_sectors)
926 return -EINVAL;
ea2384d3
FB
927
928 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
929 if (!out_buf)
930 return -1;
931
932 /* best compression, small window, no zlib header */
933 memset(&strm, 0, sizeof(strm));
934 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
5fafdf24 935 Z_DEFLATED, -12,
ea2384d3
FB
936 9, Z_DEFAULT_STRATEGY);
937 if (ret != 0) {
938 qemu_free(out_buf);
939 return -1;
940 }
941
942 strm.avail_in = s->cluster_size;
943 strm.next_in = (uint8_t *)buf;
944 strm.avail_out = s->cluster_size;
945 strm.next_out = out_buf;
946
947 ret = deflate(&strm, Z_FINISH);
948 if (ret != Z_STREAM_END && ret != Z_OK) {
949 qemu_free(out_buf);
950 deflateEnd(&strm);
951 return -1;
952 }
953 out_len = strm.next_out - out_buf;
954
955 deflateEnd(&strm);
956
957 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
958 /* could not compress: write normal cluster */
ade40677 959 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
ea2384d3 960 } else {
5fafdf24 961 cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
ea2384d3
FB
962 out_len, 0, 0);
963 cluster_offset &= s->cluster_offset_mask;
66f82cee 964 if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) {
ea2384d3
FB
965 qemu_free(out_buf);
966 return -1;
967 }
968 }
3b46e624 969
ea2384d3
FB
970 qemu_free(out_buf);
971 return 0;
972}
973
205ef796 974static int qcow_flush(BlockDriverState *bs)
7a6cba61 975{
205ef796 976 return bdrv_flush(bs->file);
7a6cba61
PB
977}
978
f8012c13
KW
979static BlockDriverAIOCB *qcow_aio_flush(BlockDriverState *bs,
980 BlockDriverCompletionFunc *cb, void *opaque)
981{
66f82cee 982 return bdrv_aio_flush(bs->file, cb, opaque);
f8012c13
KW
983}
984
c47c33b0
FB
985static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
986{
987 BDRVQcowState *s = bs->opaque;
988 bdi->cluster_size = s->cluster_size;
989 return 0;
990}
991
0e7e1989
KW
992
993static QEMUOptionParameter qcow_create_options[] = {
db08adf5
KW
994 {
995 .name = BLOCK_OPT_SIZE,
996 .type = OPT_SIZE,
997 .help = "Virtual disk size"
998 },
999 {
1000 .name = BLOCK_OPT_BACKING_FILE,
1001 .type = OPT_STRING,
1002 .help = "File name of a base image"
1003 },
1004 {
1005 .name = BLOCK_OPT_ENCRYPT,
1006 .type = OPT_FLAG,
1007 .help = "Encrypt the image"
1008 },
0e7e1989
KW
1009 { NULL }
1010};
1011
5efa9d5a 1012static BlockDriver bdrv_qcow = {
e60f469c
AJ
1013 .format_name = "qcow",
1014 .instance_size = sizeof(BDRVQcowState),
1015 .bdrv_probe = qcow_probe,
1016 .bdrv_open = qcow_open,
1017 .bdrv_close = qcow_close,
1018 .bdrv_create = qcow_create,
1019 .bdrv_flush = qcow_flush,
1020 .bdrv_is_allocated = qcow_is_allocated,
1021 .bdrv_set_key = qcow_set_key,
1022 .bdrv_make_empty = qcow_make_empty,
f141eafe
AL
1023 .bdrv_aio_readv = qcow_aio_readv,
1024 .bdrv_aio_writev = qcow_aio_writev,
f8012c13 1025 .bdrv_aio_flush = qcow_aio_flush,
c47c33b0 1026 .bdrv_write_compressed = qcow_write_compressed,
e60f469c 1027 .bdrv_get_info = qcow_get_info,
0e7e1989
KW
1028
1029 .create_options = qcow_create_options,
ea2384d3 1030};
5efa9d5a
AL
1031
1032static void bdrv_qcow_init(void)
1033{
1034 bdrv_register(&bdrv_qcow);
1035}
1036
1037block_init(bdrv_qcow_init);