Statistics
| Branch: | Revision:

root / block / qcow.c @ 77a5a000

History | View | Annotate | Download (30 kB)

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 "module.h"
27
#include <zlib.h>
28
#include "aes.h"
29

    
30
/**************************************************************/
31
/* QEMU COW block driver with compression and encryption support */
32

    
33
#define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
34
#define QCOW_VERSION 1
35

    
36
#define QCOW_CRYPT_NONE 0
37
#define QCOW_CRYPT_AES  1
38

    
39
#define QCOW_OFLAG_COMPRESSED (1LL << 63)
40

    
41
typedef struct QCowHeader {
42
    uint32_t magic;
43
    uint32_t version;
44
    uint64_t backing_file_offset;
45
    uint32_t backing_file_size;
46
    uint32_t mtime;
47
    uint64_t size; /* in bytes */
48
    uint8_t cluster_bits;
49
    uint8_t l2_bits;
50
    uint32_t crypt_method;
51
    uint64_t l1_table_offset;
52
} QCowHeader;
53

    
54
#define L2_CACHE_SIZE 16
55

    
56
typedef struct BDRVQcowState {
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(BlockDriverState *bs, uint64_t cluster_offset);
79

    
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

    
92
static int qcow_open(BlockDriverState *bs, int flags)
93
{
94
    BDRVQcowState *s = bs->opaque;
95
    int len, i, shift;
96
    QCowHeader header;
97

    
98
    if (bdrv_pread(bs->file, 0, &header, sizeof(header)) != sizeof(header))
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);
108

    
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;
134
    if (bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
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;
151

    
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;
157
        if (bdrv_pread(bs->file, header.backing_file_offset, bs->backing_file, len) != len)
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);
168
    return -1;
169
}
170

    
171
static int qcow_set_key(BlockDriverState *bs, const char *key)
172
{
173
    BDRVQcowState *s = bs->opaque;
174
    uint8_t keybuf[16];
175
    int len, i;
176

    
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 */
216
static 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;
230
        AES_cbc_encrypt(in_buf, out_buf, 512, key,
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 <
247
 * cluster_size
248
 *
249
 * return 0 if not allocated.
250
 */
251
static 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;
261

    
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 */
269
        l2_offset = bdrv_getlength(bs->file);
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);
275
        if (bdrv_pwrite_sync(bs->file,
276
                s->l1_table_offset + l1_index * sizeof(tmp),
277
                &tmp, sizeof(tmp)) < 0)
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);
303
    if (new_l2_table) {
304
        memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
305
        if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
306
                s->l2_size * sizeof(uint64_t)) < 0)
307
            return 0;
308
    } else {
309
        if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
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]);
318
    if (!cluster_offset ||
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 */
328
            if (decompress_cluster(bs, cluster_offset) < 0)
329
                return 0;
330
            cluster_offset = bdrv_getlength(bs->file);
331
            cluster_offset = (cluster_offset + s->cluster_size - 1) &
332
                ~(s->cluster_size - 1);
333
            /* write the cluster content */
334
            if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, s->cluster_size) !=
335
                s->cluster_size)
336
                return -1;
337
        } else {
338
            cluster_offset = bdrv_getlength(bs->file);
339
            if (allocate == 1) {
340
                /* round to cluster size */
341
                cluster_offset = (cluster_offset + s->cluster_size - 1) &
342
                    ~(s->cluster_size - 1);
343
                bdrv_truncate(bs->file, cluster_offset + s->cluster_size);
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);
357
                            if (bdrv_pwrite(bs->file, cluster_offset + i * 512,
358
                                            s->cluster_data, 512) != 512)
359
                                return -1;
360
                        }
361
                    }
362
                }
363
            } else if (allocate == 2) {
364
                cluster_offset |= QCOW_OFLAG_COMPRESSED |
365
                    (uint64_t)compressed_size << (63 - s->cluster_bits);
366
            }
367
        }
368
        /* update L2 table */
369
        tmp = cpu_to_be64(cluster_offset);
370
        l2_table[l2_index] = tmp;
371
        if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
372
                &tmp, sizeof(tmp)) < 0)
373
            return 0;
374
    }
375
    return cluster_offset;
376
}
377

    
378
static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
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

    
394
static 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
}
420

    
421
static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
422
{
423
    BDRVQcowState *s = bs->opaque;
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);
431
        ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
432
        if (ret != csize)
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

    
443
#if 0
444

445
static int qcow_read(BlockDriverState *bs, int64_t sector_num,
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;
451

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) {
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
            }
467
        } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
468
            if (decompress_cluster(bs, cluster_offset) < 0)
469
                return -1;
470
            memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
471
        } else {
472
            ret = bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512);
473
            if (ret != n * 512)
474
                return -1;
475
            if (s->crypt_method) {
476
                encrypt_sectors(s, sector_num, buf, buf, n, 0,
477
                                &s->aes_decrypt_key);
478
            }
479
        }
480
        nb_sectors -= n;
481
        sector_num += n;
482
        buf += n * 512;
483
    }
484
    return 0;
485
}
486
#endif
487

    
488
typedef struct QCowAIOCB {
489
    BlockDriverAIOCB common;
490
    int64_t sector_num;
491
    QEMUIOVector *qiov;
492
    uint8_t *buf;
493
    void *orig_buf;
494
    int nb_sectors;
495
    int n;
496
    uint64_t cluster_offset;
497
    uint8_t *cluster_data;
498
    struct iovec hd_iov;
499
    QEMUIOVector hd_qiov;
500
    BlockDriverAIOCB *hd_aiocb;
501
} QCowAIOCB;
502

    
503
static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
504
{
505
    QCowAIOCB *acb = container_of(blockacb, QCowAIOCB, common);
506
    if (acb->hd_aiocb)
507
        bdrv_aio_cancel(acb->hd_aiocb);
508
    qemu_aio_release(acb);
509
}
510

    
511
static AIOPool qcow_aio_pool = {
512
    .aiocb_size         = sizeof(QCowAIOCB),
513
    .cancel             = qcow_aio_cancel,
514
};
515

    
516
static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
517
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
518
        BlockDriverCompletionFunc *cb, void *opaque, int is_write)
519
{
520
    QCowAIOCB *acb;
521

    
522
    acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
523
    if (!acb)
524
        return NULL;
525
    acb->hd_aiocb = NULL;
526
    acb->sector_num = sector_num;
527
    acb->qiov = qiov;
528
    if (qiov->niov > 1) {
529
        acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
530
        if (is_write)
531
            qemu_iovec_to_buffer(qiov, acb->buf);
532
    } else {
533
        acb->buf = (uint8_t *)qiov->iov->iov_base;
534
    }
535
    acb->nb_sectors = nb_sectors;
536
    acb->n = 0;
537
    acb->cluster_offset = 0;
538
    return acb;
539
}
540

    
541
static void qcow_aio_read_cb(void *opaque, int ret)
542
{
543
    QCowAIOCB *acb = opaque;
544
    BlockDriverState *bs = acb->common.bs;
545
    BDRVQcowState *s = bs->opaque;
546
    int index_in_cluster;
547

    
548
    acb->hd_aiocb = NULL;
549
    if (ret < 0)
550
        goto done;
551

    
552
 redo:
553
    /* post process the read buffer */
554
    if (!acb->cluster_offset) {
555
        /* nothing to do */
556
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
557
        /* nothing to do */
558
    } else {
559
        if (s->crypt_method) {
560
            encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
561
                            acb->n, 0,
562
                            &s->aes_decrypt_key);
563
        }
564
    }
565

    
566
    acb->nb_sectors -= acb->n;
567
    acb->sector_num += acb->n;
568
    acb->buf += acb->n * 512;
569

    
570
    if (acb->nb_sectors == 0) {
571
        /* request completed */
572
        ret = 0;
573
        goto done;
574
    }
575

    
576
    /* prepare next AIO request */
577
    acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9,
578
                                             0, 0, 0, 0);
579
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
580
    acb->n = s->cluster_sectors - index_in_cluster;
581
    if (acb->n > acb->nb_sectors)
582
        acb->n = acb->nb_sectors;
583

    
584
    if (!acb->cluster_offset) {
585
        if (bs->backing_hd) {
586
            /* read from the base image */
587
            acb->hd_iov.iov_base = (void *)acb->buf;
588
            acb->hd_iov.iov_len = acb->n * 512;
589
            qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
590
            acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
591
                &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
592
            if (acb->hd_aiocb == NULL) {
593
                ret = -EIO;
594
                goto done;
595
            }
596
        } else {
597
            /* Note: in this case, no need to wait */
598
            memset(acb->buf, 0, 512 * acb->n);
599
            goto redo;
600
        }
601
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
602
        /* add AIO support for compressed blocks ? */
603
        if (decompress_cluster(bs, acb->cluster_offset) < 0) {
604
            ret = -EIO;
605
            goto done;
606
        }
607
        memcpy(acb->buf,
608
               s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
609
        goto redo;
610
    } else {
611
        if ((acb->cluster_offset & 511) != 0) {
612
            ret = -EIO;
613
            goto done;
614
        }
615
        acb->hd_iov.iov_base = (void *)acb->buf;
616
        acb->hd_iov.iov_len = acb->n * 512;
617
        qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
618
        acb->hd_aiocb = bdrv_aio_readv(bs->file,
619
                            (acb->cluster_offset >> 9) + index_in_cluster,
620
                            &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
621
        if (acb->hd_aiocb == NULL) {
622
            ret = -EIO;
623
            goto done;
624
        }
625
    }
626

    
627
    return;
628

    
629
done:
630
    if (acb->qiov->niov > 1) {
631
        qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
632
        qemu_vfree(acb->orig_buf);
633
    }
634
    acb->common.cb(acb->common.opaque, ret);
635
    qemu_aio_release(acb);
636
}
637

    
638
static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
639
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
640
        BlockDriverCompletionFunc *cb, void *opaque)
641
{
642
    QCowAIOCB *acb;
643

    
644
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
645
    if (!acb)
646
        return NULL;
647

    
648
    qcow_aio_read_cb(acb, 0);
649
    return &acb->common;
650
}
651

    
652
static void qcow_aio_write_cb(void *opaque, int ret)
653
{
654
    QCowAIOCB *acb = opaque;
655
    BlockDriverState *bs = acb->common.bs;
656
    BDRVQcowState *s = bs->opaque;
657
    int index_in_cluster;
658
    uint64_t cluster_offset;
659
    const uint8_t *src_buf;
660

    
661
    acb->hd_aiocb = NULL;
662

    
663
    if (ret < 0)
664
        goto done;
665

    
666
    acb->nb_sectors -= acb->n;
667
    acb->sector_num += acb->n;
668
    acb->buf += acb->n * 512;
669

    
670
    if (acb->nb_sectors == 0) {
671
        /* request completed */
672
        ret = 0;
673
        goto done;
674
    }
675

    
676
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
677
    acb->n = s->cluster_sectors - index_in_cluster;
678
    if (acb->n > acb->nb_sectors)
679
        acb->n = acb->nb_sectors;
680
    cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, 1, 0,
681
                                        index_in_cluster,
682
                                        index_in_cluster + acb->n);
683
    if (!cluster_offset || (cluster_offset & 511) != 0) {
684
        ret = -EIO;
685
        goto done;
686
    }
687
    if (s->crypt_method) {
688
        if (!acb->cluster_data) {
689
            acb->cluster_data = qemu_mallocz(s->cluster_size);
690
            if (!acb->cluster_data) {
691
                ret = -ENOMEM;
692
                goto done;
693
            }
694
        }
695
        encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
696
                        acb->n, 1, &s->aes_encrypt_key);
697
        src_buf = acb->cluster_data;
698
    } else {
699
        src_buf = acb->buf;
700
    }
701

    
702
    acb->hd_iov.iov_base = (void *)src_buf;
703
    acb->hd_iov.iov_len = acb->n * 512;
704
    qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
705
    acb->hd_aiocb = bdrv_aio_writev(bs->file,
706
                                    (cluster_offset >> 9) + index_in_cluster,
707
                                    &acb->hd_qiov, acb->n,
708
                                    qcow_aio_write_cb, acb);
709
    if (acb->hd_aiocb == NULL) {
710
        ret = -EIO;
711
        goto done;
712
    }
713
    return;
714

    
715
done:
716
    if (acb->qiov->niov > 1)
717
        qemu_vfree(acb->orig_buf);
718
    acb->common.cb(acb->common.opaque, ret);
719
    qemu_aio_release(acb);
720
}
721

    
722
static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
723
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
724
        BlockDriverCompletionFunc *cb, void *opaque)
725
{
726
    BDRVQcowState *s = bs->opaque;
727
    QCowAIOCB *acb;
728

    
729
    s->cluster_cache_offset = -1; /* disable compressed cache */
730

    
731
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
732
    if (!acb)
733
        return NULL;
734

    
735

    
736
    qcow_aio_write_cb(acb, 0);
737
    return &acb->common;
738
}
739

    
740
static void qcow_close(BlockDriverState *bs)
741
{
742
    BDRVQcowState *s = bs->opaque;
743
    qemu_free(s->l1_table);
744
    qemu_free(s->l2_cache);
745
    qemu_free(s->cluster_cache);
746
    qemu_free(s->cluster_data);
747
}
748

    
749
static int qcow_create(const char *filename, QEMUOptionParameter *options)
750
{
751
    int fd, header_size, backing_filename_len, l1_size, i, shift;
752
    QCowHeader header;
753
    uint64_t tmp;
754
    int64_t total_size = 0;
755
    const char *backing_file = NULL;
756
    int flags = 0;
757
    int ret;
758

    
759
    /* Read out options */
760
    while (options && options->name) {
761
        if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
762
            total_size = options->value.n / 512;
763
        } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
764
            backing_file = options->value.s;
765
        } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
766
            flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
767
        }
768
        options++;
769
    }
770

    
771
    fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
772
    if (fd < 0)
773
        return -errno;
774
    memset(&header, 0, sizeof(header));
775
    header.magic = cpu_to_be32(QCOW_MAGIC);
776
    header.version = cpu_to_be32(QCOW_VERSION);
777
    header.size = cpu_to_be64(total_size * 512);
778
    header_size = sizeof(header);
779
    backing_filename_len = 0;
780
    if (backing_file) {
781
        if (strcmp(backing_file, "fat:")) {
782
            header.backing_file_offset = cpu_to_be64(header_size);
783
            backing_filename_len = strlen(backing_file);
784
            header.backing_file_size = cpu_to_be32(backing_filename_len);
785
            header_size += backing_filename_len;
786
        } else {
787
            /* special backing file for vvfat */
788
            backing_file = NULL;
789
        }
790
        header.cluster_bits = 9; /* 512 byte cluster to avoid copying
791
                                    unmodifyed sectors */
792
        header.l2_bits = 12; /* 32 KB L2 tables */
793
    } else {
794
        header.cluster_bits = 12; /* 4 KB clusters */
795
        header.l2_bits = 9; /* 4 KB L2 tables */
796
    }
797
    header_size = (header_size + 7) & ~7;
798
    shift = header.cluster_bits + header.l2_bits;
799
    l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
800

    
801
    header.l1_table_offset = cpu_to_be64(header_size);
802
    if (flags & BLOCK_FLAG_ENCRYPT) {
803
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
804
    } else {
805
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
806
    }
807

    
808
    /* write all the data */
809
    ret = qemu_write_full(fd, &header, sizeof(header));
810
    if (ret != sizeof(header)) {
811
        ret = -errno;
812
        goto exit;
813
    }
814

    
815
    if (backing_file) {
816
        ret = qemu_write_full(fd, backing_file, backing_filename_len);
817
        if (ret != backing_filename_len) {
818
            ret = -errno;
819
            goto exit;
820
        }
821

    
822
    }
823
    lseek(fd, header_size, SEEK_SET);
824
    tmp = 0;
825
    for(i = 0;i < l1_size; i++) {
826
        ret = qemu_write_full(fd, &tmp, sizeof(tmp));
827
        if (ret != sizeof(tmp)) {
828
            ret = -errno;
829
            goto exit;
830
        }
831
    }
832

    
833
    ret = 0;
834
exit:
835
    close(fd);
836
    return ret;
837
}
838

    
839
static int qcow_make_empty(BlockDriverState *bs)
840
{
841
    BDRVQcowState *s = bs->opaque;
842
    uint32_t l1_length = s->l1_size * sizeof(uint64_t);
843
    int ret;
844

    
845
    memset(s->l1_table, 0, l1_length);
846
    if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
847
            l1_length) < 0)
848
        return -1;
849
    ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
850
    if (ret < 0)
851
        return ret;
852

    
853
    memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
854
    memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
855
    memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
856

    
857
    return 0;
858
}
859

    
860
/* XXX: put compressed sectors first, then all the cluster aligned
861
   tables to avoid losing bytes in alignment */
862
static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
863
                                 const uint8_t *buf, int nb_sectors)
864
{
865
    BDRVQcowState *s = bs->opaque;
866
    z_stream strm;
867
    int ret, out_len;
868
    uint8_t *out_buf;
869
    uint64_t cluster_offset;
870

    
871
    if (nb_sectors != s->cluster_sectors)
872
        return -EINVAL;
873

    
874
    out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
875
    if (!out_buf)
876
        return -1;
877

    
878
    /* best compression, small window, no zlib header */
879
    memset(&strm, 0, sizeof(strm));
880
    ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
881
                       Z_DEFLATED, -12,
882
                       9, Z_DEFAULT_STRATEGY);
883
    if (ret != 0) {
884
        qemu_free(out_buf);
885
        return -1;
886
    }
887

    
888
    strm.avail_in = s->cluster_size;
889
    strm.next_in = (uint8_t *)buf;
890
    strm.avail_out = s->cluster_size;
891
    strm.next_out = out_buf;
892

    
893
    ret = deflate(&strm, Z_FINISH);
894
    if (ret != Z_STREAM_END && ret != Z_OK) {
895
        qemu_free(out_buf);
896
        deflateEnd(&strm);
897
        return -1;
898
    }
899
    out_len = strm.next_out - out_buf;
900

    
901
    deflateEnd(&strm);
902

    
903
    if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
904
        /* could not compress: write normal cluster */
905
        bdrv_write(bs, sector_num, buf, s->cluster_sectors);
906
    } else {
907
        cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
908
                                            out_len, 0, 0);
909
        cluster_offset &= s->cluster_offset_mask;
910
        if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) {
911
            qemu_free(out_buf);
912
            return -1;
913
        }
914
    }
915

    
916
    qemu_free(out_buf);
917
    return 0;
918
}
919

    
920
static int qcow_flush(BlockDriverState *bs)
921
{
922
    return bdrv_flush(bs->file);
923
}
924

    
925
static BlockDriverAIOCB *qcow_aio_flush(BlockDriverState *bs,
926
        BlockDriverCompletionFunc *cb, void *opaque)
927
{
928
    return bdrv_aio_flush(bs->file, cb, opaque);
929
}
930

    
931
static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
932
{
933
    BDRVQcowState *s = bs->opaque;
934
    bdi->cluster_size = s->cluster_size;
935
    return 0;
936
}
937

    
938

    
939
static QEMUOptionParameter qcow_create_options[] = {
940
    {
941
        .name = BLOCK_OPT_SIZE,
942
        .type = OPT_SIZE,
943
        .help = "Virtual disk size"
944
    },
945
    {
946
        .name = BLOCK_OPT_BACKING_FILE,
947
        .type = OPT_STRING,
948
        .help = "File name of a base image"
949
    },
950
    {
951
        .name = BLOCK_OPT_ENCRYPT,
952
        .type = OPT_FLAG,
953
        .help = "Encrypt the image"
954
    },
955
    { NULL }
956
};
957

    
958
static BlockDriver bdrv_qcow = {
959
    .format_name        = "qcow",
960
    .instance_size        = sizeof(BDRVQcowState),
961
    .bdrv_probe                = qcow_probe,
962
    .bdrv_open                = qcow_open,
963
    .bdrv_close                = qcow_close,
964
    .bdrv_create        = qcow_create,
965
    .bdrv_flush                = qcow_flush,
966
    .bdrv_is_allocated        = qcow_is_allocated,
967
    .bdrv_set_key        = qcow_set_key,
968
    .bdrv_make_empty        = qcow_make_empty,
969
    .bdrv_aio_readv        = qcow_aio_readv,
970
    .bdrv_aio_writev        = qcow_aio_writev,
971
    .bdrv_aio_flush        = qcow_aio_flush,
972
    .bdrv_write_compressed = qcow_write_compressed,
973
    .bdrv_get_info        = qcow_get_info,
974

    
975
    .create_options = qcow_create_options,
976
};
977

    
978
static void bdrv_qcow_init(void)
979
{
980
    bdrv_register(&bdrv_qcow);
981
}
982

    
983
block_init(bdrv_qcow_init);