Statistics
| Branch: | Revision:

root / block / qcow2.c @ 1e5b9d2f

History | View | Annotate | Download (33.7 kB)

1
/*
2
 * Block driver for the QCOW version 2 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
#include "block/qcow2.h"
30

    
31
/*
32
  Differences with QCOW:
33

34
  - Support for multiple incremental snapshots.
35
  - Memory management by reference counts.
36
  - Clusters which have a reference count of one have the bit
37
    QCOW_OFLAG_COPIED to optimize write performance.
38
  - Size of compressed clusters is stored in sectors to reduce bit usage
39
    in the cluster offsets.
40
  - Support for storing additional data (such as the VM state) in the
41
    snapshots.
42
  - If a backing store is used, the cluster size is not constrained
43
    (could be backported to QCOW).
44
  - L2 tables have always a size of one cluster.
45
*/
46

    
47

    
48
typedef struct {
49
    uint32_t magic;
50
    uint32_t len;
51
} QCowExtension;
52
#define  QCOW_EXT_MAGIC_END 0
53
#define  QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
54

    
55

    
56

    
57
static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
58
{
59
    const QCowHeader *cow_header = (const void *)buf;
60

    
61
    if (buf_size >= sizeof(QCowHeader) &&
62
        be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
63
        be32_to_cpu(cow_header->version) == QCOW_VERSION)
64
        return 100;
65
    else
66
        return 0;
67
}
68

    
69

    
70
/* 
71
 * read qcow2 extension and fill bs
72
 * start reading from start_offset
73
 * finish reading upon magic of value 0 or when end_offset reached
74
 * unknown magic is skipped (future extension this version knows nothing about)
75
 * return 0 upon success, non-0 otherwise
76
 */
77
static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
78
                                uint64_t end_offset)
79
{
80
    BDRVQcowState *s = bs->opaque;
81
    QCowExtension ext;
82
    uint64_t offset;
83

    
84
#ifdef DEBUG_EXT
85
    printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
86
#endif
87
    offset = start_offset;
88
    while (offset < end_offset) {
89

    
90
#ifdef DEBUG_EXT
91
        /* Sanity check */
92
        if (offset > s->cluster_size)
93
            printf("qcow_handle_extension: suspicious offset %lu\n", offset);
94

    
95
        printf("attemting to read extended header in offset %lu\n", offset);
96
#endif
97

    
98
        if (bdrv_pread(s->hd, offset, &ext, sizeof(ext)) != sizeof(ext)) {
99
            fprintf(stderr, "qcow_handle_extension: ERROR: pread fail from offset %llu\n",
100
                    (unsigned long long)offset);
101
            return 1;
102
        }
103
        be32_to_cpus(&ext.magic);
104
        be32_to_cpus(&ext.len);
105
        offset += sizeof(ext);
106
#ifdef DEBUG_EXT
107
        printf("ext.magic = 0x%x\n", ext.magic);
108
#endif
109
        switch (ext.magic) {
110
        case QCOW_EXT_MAGIC_END:
111
            return 0;
112

    
113
        case QCOW_EXT_MAGIC_BACKING_FORMAT:
114
            if (ext.len >= sizeof(bs->backing_format)) {
115
                fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
116
                        " (>=%zu)\n",
117
                        ext.len, sizeof(bs->backing_format));
118
                return 2;
119
            }
120
            if (bdrv_pread(s->hd, offset , bs->backing_format,
121
                           ext.len) != ext.len)
122
                return 3;
123
            bs->backing_format[ext.len] = '\0';
124
#ifdef DEBUG_EXT
125
            printf("Qcow2: Got format extension %s\n", bs->backing_format);
126
#endif
127
            offset += ((ext.len + 7) & ~7);
128
            break;
129

    
130
        default:
131
            /* unknown magic -- just skip it */
132
            offset += ((ext.len + 7) & ~7);
133
            break;
134
        }
135
    }
136

    
137
    return 0;
138
}
139

    
140

    
141
static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
142
{
143
    BDRVQcowState *s = bs->opaque;
144
    int len, i, shift, ret;
145
    QCowHeader header;
146
    uint64_t ext_end;
147

    
148
    ret = bdrv_file_open(&s->hd, filename, flags);
149
    if (ret < 0)
150
        return ret;
151
    if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
152
        goto fail;
153
    be32_to_cpus(&header.magic);
154
    be32_to_cpus(&header.version);
155
    be64_to_cpus(&header.backing_file_offset);
156
    be32_to_cpus(&header.backing_file_size);
157
    be64_to_cpus(&header.size);
158
    be32_to_cpus(&header.cluster_bits);
159
    be32_to_cpus(&header.crypt_method);
160
    be64_to_cpus(&header.l1_table_offset);
161
    be32_to_cpus(&header.l1_size);
162
    be64_to_cpus(&header.refcount_table_offset);
163
    be32_to_cpus(&header.refcount_table_clusters);
164
    be64_to_cpus(&header.snapshots_offset);
165
    be32_to_cpus(&header.nb_snapshots);
166

    
167
    if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
168
        goto fail;
169
    if (header.size <= 1 ||
170
        header.cluster_bits < MIN_CLUSTER_BITS ||
171
        header.cluster_bits > MAX_CLUSTER_BITS)
172
        goto fail;
173
    if (header.crypt_method > QCOW_CRYPT_AES)
174
        goto fail;
175
    s->crypt_method_header = header.crypt_method;
176
    if (s->crypt_method_header)
177
        bs->encrypted = 1;
178
    s->cluster_bits = header.cluster_bits;
179
    s->cluster_size = 1 << s->cluster_bits;
180
    s->cluster_sectors = 1 << (s->cluster_bits - 9);
181
    s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
182
    s->l2_size = 1 << s->l2_bits;
183
    bs->total_sectors = header.size / 512;
184
    s->csize_shift = (62 - (s->cluster_bits - 8));
185
    s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
186
    s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
187
    s->refcount_table_offset = header.refcount_table_offset;
188
    s->refcount_table_size =
189
        header.refcount_table_clusters << (s->cluster_bits - 3);
190

    
191
    s->snapshots_offset = header.snapshots_offset;
192
    s->nb_snapshots = header.nb_snapshots;
193

    
194
    /* read the level 1 table */
195
    s->l1_size = header.l1_size;
196
    shift = s->cluster_bits + s->l2_bits;
197
    s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
198
    /* the L1 table must contain at least enough entries to put
199
       header.size bytes */
200
    if (s->l1_size < s->l1_vm_state_index)
201
        goto fail;
202
    s->l1_table_offset = header.l1_table_offset;
203
    s->l1_table = qemu_mallocz(
204
        align_offset(s->l1_size * sizeof(uint64_t), 512));
205
    if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
206
        s->l1_size * sizeof(uint64_t))
207
        goto fail;
208
    for(i = 0;i < s->l1_size; i++) {
209
        be64_to_cpus(&s->l1_table[i]);
210
    }
211
    /* alloc L2 cache */
212
    s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
213
    s->cluster_cache = qemu_malloc(s->cluster_size);
214
    /* one more sector for decompressed data alignment */
215
    s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
216
                                  + 512);
217
    s->cluster_cache_offset = -1;
218

    
219
    if (qcow2_refcount_init(bs) < 0)
220
        goto fail;
221

    
222
    QLIST_INIT(&s->cluster_allocs);
223

    
224
    /* read qcow2 extensions */
225
    if (header.backing_file_offset)
226
        ext_end = header.backing_file_offset;
227
    else
228
        ext_end = s->cluster_size;
229
    if (qcow_read_extensions(bs, sizeof(header), ext_end))
230
        goto fail;
231

    
232
    /* read the backing file name */
233
    if (header.backing_file_offset != 0) {
234
        len = header.backing_file_size;
235
        if (len > 1023)
236
            len = 1023;
237
        if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
238
            goto fail;
239
        bs->backing_file[len] = '\0';
240
    }
241
    if (qcow2_read_snapshots(bs) < 0)
242
        goto fail;
243

    
244
#ifdef DEBUG_ALLOC
245
    qcow2_check_refcounts(bs);
246
#endif
247
    return 0;
248

    
249
 fail:
250
    qcow2_free_snapshots(bs);
251
    qcow2_refcount_close(bs);
252
    qemu_free(s->l1_table);
253
    qemu_free(s->l2_cache);
254
    qemu_free(s->cluster_cache);
255
    qemu_free(s->cluster_data);
256
    bdrv_delete(s->hd);
257
    return -1;
258
}
259

    
260
static int qcow_set_key(BlockDriverState *bs, const char *key)
261
{
262
    BDRVQcowState *s = bs->opaque;
263
    uint8_t keybuf[16];
264
    int len, i;
265

    
266
    memset(keybuf, 0, 16);
267
    len = strlen(key);
268
    if (len > 16)
269
        len = 16;
270
    /* XXX: we could compress the chars to 7 bits to increase
271
       entropy */
272
    for(i = 0;i < len;i++) {
273
        keybuf[i] = key[i];
274
    }
275
    s->crypt_method = s->crypt_method_header;
276

    
277
    if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
278
        return -1;
279
    if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
280
        return -1;
281
#if 0
282
    /* test */
283
    {
284
        uint8_t in[16];
285
        uint8_t out[16];
286
        uint8_t tmp[16];
287
        for(i=0;i<16;i++)
288
            in[i] = i;
289
        AES_encrypt(in, tmp, &s->aes_encrypt_key);
290
        AES_decrypt(tmp, out, &s->aes_decrypt_key);
291
        for(i = 0; i < 16; i++)
292
            printf(" %02x", tmp[i]);
293
        printf("\n");
294
        for(i = 0; i < 16; i++)
295
            printf(" %02x", out[i]);
296
        printf("\n");
297
    }
298
#endif
299
    return 0;
300
}
301

    
302
static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
303
                             int nb_sectors, int *pnum)
304
{
305
    uint64_t cluster_offset;
306

    
307
    *pnum = nb_sectors;
308
    cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, pnum);
309

    
310
    return (cluster_offset != 0);
311
}
312

    
313
/* handle reading after the end of the backing file */
314
int qcow2_backing_read1(BlockDriverState *bs,
315
                  int64_t sector_num, uint8_t *buf, int nb_sectors)
316
{
317
    int n1;
318
    if ((sector_num + nb_sectors) <= bs->total_sectors)
319
        return nb_sectors;
320
    if (sector_num >= bs->total_sectors)
321
        n1 = 0;
322
    else
323
        n1 = bs->total_sectors - sector_num;
324
    memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
325
    return n1;
326
}
327

    
328
typedef struct QCowAIOCB {
329
    BlockDriverAIOCB common;
330
    int64_t sector_num;
331
    QEMUIOVector *qiov;
332
    uint8_t *buf;
333
    void *orig_buf;
334
    int nb_sectors;
335
    int n;
336
    uint64_t cluster_offset;
337
    uint8_t *cluster_data;
338
    BlockDriverAIOCB *hd_aiocb;
339
    struct iovec hd_iov;
340
    QEMUIOVector hd_qiov;
341
    QEMUBH *bh;
342
    QCowL2Meta l2meta;
343
    QLIST_ENTRY(QCowAIOCB) next_depend;
344
} QCowAIOCB;
345

    
346
static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
347
{
348
    QCowAIOCB *acb = (QCowAIOCB *)blockacb;
349
    if (acb->hd_aiocb)
350
        bdrv_aio_cancel(acb->hd_aiocb);
351
    qemu_aio_release(acb);
352
}
353

    
354
static AIOPool qcow_aio_pool = {
355
    .aiocb_size         = sizeof(QCowAIOCB),
356
    .cancel             = qcow_aio_cancel,
357
};
358

    
359
static void qcow_aio_read_cb(void *opaque, int ret);
360
static void qcow_aio_read_bh(void *opaque)
361
{
362
    QCowAIOCB *acb = opaque;
363
    qemu_bh_delete(acb->bh);
364
    acb->bh = NULL;
365
    qcow_aio_read_cb(opaque, 0);
366
}
367

    
368
static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
369
{
370
    if (acb->bh)
371
        return -EIO;
372

    
373
    acb->bh = qemu_bh_new(cb, acb);
374
    if (!acb->bh)
375
        return -EIO;
376

    
377
    qemu_bh_schedule(acb->bh);
378

    
379
    return 0;
380
}
381

    
382
static void qcow_aio_read_cb(void *opaque, int ret)
383
{
384
    QCowAIOCB *acb = opaque;
385
    BlockDriverState *bs = acb->common.bs;
386
    BDRVQcowState *s = bs->opaque;
387
    int index_in_cluster, n1;
388

    
389
    acb->hd_aiocb = NULL;
390
    if (ret < 0)
391
        goto done;
392

    
393
    /* post process the read buffer */
394
    if (!acb->cluster_offset) {
395
        /* nothing to do */
396
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
397
        /* nothing to do */
398
    } else {
399
        if (s->crypt_method) {
400
            qcow2_encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
401
                            acb->n, 0,
402
                            &s->aes_decrypt_key);
403
        }
404
    }
405

    
406
    acb->nb_sectors -= acb->n;
407
    acb->sector_num += acb->n;
408
    acb->buf += acb->n * 512;
409

    
410
    if (acb->nb_sectors == 0) {
411
        /* request completed */
412
        ret = 0;
413
        goto done;
414
    }
415

    
416
    /* prepare next AIO request */
417
    acb->n = acb->nb_sectors;
418
    acb->cluster_offset =
419
        qcow2_get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
420
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
421

    
422
    if (!acb->cluster_offset) {
423
        if (bs->backing_hd) {
424
            /* read from the base image */
425
            n1 = qcow2_backing_read1(bs->backing_hd, acb->sector_num,
426
                               acb->buf, acb->n);
427
            if (n1 > 0) {
428
                acb->hd_iov.iov_base = (void *)acb->buf;
429
                acb->hd_iov.iov_len = acb->n * 512;
430
                qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
431
                acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
432
                                    &acb->hd_qiov, acb->n,
433
                                    qcow_aio_read_cb, acb);
434
                if (acb->hd_aiocb == NULL)
435
                    goto done;
436
            } else {
437
                ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
438
                if (ret < 0)
439
                    goto done;
440
            }
441
        } else {
442
            /* Note: in this case, no need to wait */
443
            memset(acb->buf, 0, 512 * acb->n);
444
            ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
445
            if (ret < 0)
446
                goto done;
447
        }
448
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
449
        /* add AIO support for compressed blocks ? */
450
        if (qcow2_decompress_cluster(s, acb->cluster_offset) < 0)
451
            goto done;
452
        memcpy(acb->buf,
453
               s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
454
        ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
455
        if (ret < 0)
456
            goto done;
457
    } else {
458
        if ((acb->cluster_offset & 511) != 0) {
459
            ret = -EIO;
460
            goto done;
461
        }
462

    
463
        acb->hd_iov.iov_base = (void *)acb->buf;
464
        acb->hd_iov.iov_len = acb->n * 512;
465
        qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
466
        acb->hd_aiocb = bdrv_aio_readv(s->hd,
467
                            (acb->cluster_offset >> 9) + index_in_cluster,
468
                            &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
469
        if (acb->hd_aiocb == NULL)
470
            goto done;
471
    }
472

    
473
    return;
474
done:
475
    if (acb->qiov->niov > 1) {
476
        qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
477
        qemu_vfree(acb->orig_buf);
478
    }
479
    acb->common.cb(acb->common.opaque, ret);
480
    qemu_aio_release(acb);
481
}
482

    
483
static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
484
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
485
        BlockDriverCompletionFunc *cb, void *opaque, int is_write)
486
{
487
    QCowAIOCB *acb;
488

    
489
    acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
490
    if (!acb)
491
        return NULL;
492
    acb->hd_aiocb = NULL;
493
    acb->sector_num = sector_num;
494
    acb->qiov = qiov;
495
    if (qiov->niov > 1) {
496
        acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
497
        if (is_write)
498
            qemu_iovec_to_buffer(qiov, acb->buf);
499
    } else {
500
        acb->buf = (uint8_t *)qiov->iov->iov_base;
501
    }
502
    acb->nb_sectors = nb_sectors;
503
    acb->n = 0;
504
    acb->cluster_offset = 0;
505
    acb->l2meta.nb_clusters = 0;
506
    QLIST_INIT(&acb->l2meta.dependent_requests);
507
    return acb;
508
}
509

    
510
static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
511
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
512
        BlockDriverCompletionFunc *cb, void *opaque)
513
{
514
    QCowAIOCB *acb;
515

    
516
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
517
    if (!acb)
518
        return NULL;
519

    
520
    qcow_aio_read_cb(acb, 0);
521
    return &acb->common;
522
}
523

    
524
static void qcow_aio_write_cb(void *opaque, int ret);
525

    
526
static void run_dependent_requests(QCowL2Meta *m)
527
{
528
    QCowAIOCB *req;
529
    QCowAIOCB *next;
530

    
531
    /* Take the request off the list of running requests */
532
    if (m->nb_clusters != 0) {
533
        QLIST_REMOVE(m, next_in_flight);
534
    }
535

    
536
    /*
537
     * Restart all dependent requests.
538
     * Can't use QLIST_FOREACH here - the next link might not be the same
539
     * any more after the callback  (request could depend on a different
540
     * request now)
541
     */
542
    for (req = m->dependent_requests.lh_first; req != NULL; req = next) {
543
        next = req->next_depend.le_next;
544
        qcow_aio_write_cb(req, 0);
545
    }
546

    
547
    /* Empty the list for the next part of the request */
548
    QLIST_INIT(&m->dependent_requests);
549
}
550

    
551
static void qcow_aio_write_cb(void *opaque, int ret)
552
{
553
    QCowAIOCB *acb = opaque;
554
    BlockDriverState *bs = acb->common.bs;
555
    BDRVQcowState *s = bs->opaque;
556
    int index_in_cluster;
557
    const uint8_t *src_buf;
558
    int n_end;
559

    
560
    acb->hd_aiocb = NULL;
561

    
562
    if (ret >= 0) {
563
        ret = qcow2_alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta);
564
    }
565

    
566
    run_dependent_requests(&acb->l2meta);
567

    
568
    if (ret < 0)
569
        goto done;
570

    
571
    acb->nb_sectors -= acb->n;
572
    acb->sector_num += acb->n;
573
    acb->buf += acb->n * 512;
574

    
575
    if (acb->nb_sectors == 0) {
576
        /* request completed */
577
        ret = 0;
578
        goto done;
579
    }
580

    
581
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
582
    n_end = index_in_cluster + acb->nb_sectors;
583
    if (s->crypt_method &&
584
        n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
585
        n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
586

    
587
    acb->cluster_offset = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
588
                                          index_in_cluster,
589
                                          n_end, &acb->n, &acb->l2meta);
590

    
591
    /* Need to wait for another request? If so, we are done for now. */
592
    if (!acb->cluster_offset && acb->l2meta.depends_on != NULL) {
593
        QLIST_INSERT_HEAD(&acb->l2meta.depends_on->dependent_requests,
594
            acb, next_depend);
595
        return;
596
    }
597

    
598
    if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
599
        ret = -EIO;
600
        goto done;
601
    }
602
    if (s->crypt_method) {
603
        if (!acb->cluster_data) {
604
            acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
605
                                             s->cluster_size);
606
        }
607
        qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
608
                        acb->n, 1, &s->aes_encrypt_key);
609
        src_buf = acb->cluster_data;
610
    } else {
611
        src_buf = acb->buf;
612
    }
613
    acb->hd_iov.iov_base = (void *)src_buf;
614
    acb->hd_iov.iov_len = acb->n * 512;
615
    qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
616
    acb->hd_aiocb = bdrv_aio_writev(s->hd,
617
                                    (acb->cluster_offset >> 9) + index_in_cluster,
618
                                    &acb->hd_qiov, acb->n,
619
                                    qcow_aio_write_cb, acb);
620
    if (acb->hd_aiocb == NULL)
621
        goto done;
622

    
623
    return;
624

    
625
done:
626
    if (acb->qiov->niov > 1)
627
        qemu_vfree(acb->orig_buf);
628
    acb->common.cb(acb->common.opaque, ret);
629
    qemu_aio_release(acb);
630
}
631

    
632
static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
633
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
634
        BlockDriverCompletionFunc *cb, void *opaque)
635
{
636
    BDRVQcowState *s = bs->opaque;
637
    QCowAIOCB *acb;
638

    
639
    s->cluster_cache_offset = -1; /* disable compressed cache */
640

    
641
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
642
    if (!acb)
643
        return NULL;
644

    
645
    qcow_aio_write_cb(acb, 0);
646
    return &acb->common;
647
}
648

    
649
static void qcow_close(BlockDriverState *bs)
650
{
651
    BDRVQcowState *s = bs->opaque;
652
    qemu_free(s->l1_table);
653
    qemu_free(s->l2_cache);
654
    qemu_free(s->cluster_cache);
655
    qemu_free(s->cluster_data);
656
    qcow2_refcount_close(bs);
657
    bdrv_delete(s->hd);
658
}
659

    
660
static int get_bits_from_size(size_t size)
661
{
662
    int res = 0;
663

    
664
    if (size == 0) {
665
        return -1;
666
    }
667

    
668
    while (size != 1) {
669
        /* Not a power of two */
670
        if (size & 1) {
671
            return -1;
672
        }
673

    
674
        size >>= 1;
675
        res++;
676
    }
677

    
678
    return res;
679
}
680

    
681

    
682
static int preallocate(BlockDriverState *bs)
683
{
684
    BDRVQcowState *s = bs->opaque;
685
    uint64_t cluster_offset = 0;
686
    uint64_t nb_sectors;
687
    uint64_t offset;
688
    int num;
689
    QCowL2Meta meta;
690

    
691
    nb_sectors = bdrv_getlength(bs) >> 9;
692
    offset = 0;
693
    QLIST_INIT(&meta.dependent_requests);
694

    
695
    while (nb_sectors) {
696
        num = MIN(nb_sectors, INT_MAX >> 9);
697
        cluster_offset = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num,
698
            &meta);
699

    
700
        if (cluster_offset == 0) {
701
            return -1;
702
        }
703

    
704
        if (qcow2_alloc_cluster_link_l2(bs, cluster_offset, &meta) < 0) {
705
            qcow2_free_any_clusters(bs, cluster_offset, meta.nb_clusters);
706
            return -1;
707
        }
708

    
709
        /* There are no dependent requests, but we need to remove our request
710
         * from the list of in-flight requests */
711
        run_dependent_requests(&meta);
712

    
713
        /* TODO Preallocate data if requested */
714

    
715
        nb_sectors -= num;
716
        offset += num << 9;
717
    }
718

    
719
    /*
720
     * It is expected that the image file is large enough to actually contain
721
     * all of the allocated clusters (otherwise we get failing reads after
722
     * EOF). Extend the image to the last allocated sector.
723
     */
724
    if (cluster_offset != 0) {
725
        uint8_t buf[512];
726
        memset(buf, 0, 512);
727
        bdrv_write(s->hd, (cluster_offset >> 9) + num - 1, buf, 1);
728
    }
729

    
730
    return 0;
731
}
732

    
733
static int qcow_create2(const char *filename, int64_t total_size,
734
                        const char *backing_file, const char *backing_format,
735
                        int flags, size_t cluster_size, int prealloc)
736
{
737

    
738
    int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
739
    int ref_clusters, backing_format_len = 0;
740
    QCowHeader header;
741
    uint64_t tmp, offset;
742
    QCowCreateState s1, *s = &s1;
743
    QCowExtension ext_bf = {0, 0};
744

    
745

    
746
    memset(s, 0, sizeof(*s));
747

    
748
    fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
749
    if (fd < 0)
750
        return -1;
751
    memset(&header, 0, sizeof(header));
752
    header.magic = cpu_to_be32(QCOW_MAGIC);
753
    header.version = cpu_to_be32(QCOW_VERSION);
754
    header.size = cpu_to_be64(total_size * 512);
755
    header_size = sizeof(header);
756
    backing_filename_len = 0;
757
    if (backing_file) {
758
        if (backing_format) {
759
            ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
760
            backing_format_len = strlen(backing_format);
761
            ext_bf.len = (backing_format_len + 7) & ~7;
762
            header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7);
763
        }
764
        header.backing_file_offset = cpu_to_be64(header_size);
765
        backing_filename_len = strlen(backing_file);
766
        header.backing_file_size = cpu_to_be32(backing_filename_len);
767
        header_size += backing_filename_len;
768
    }
769

    
770
    /* Cluster size */
771
    s->cluster_bits = get_bits_from_size(cluster_size);
772
    if (s->cluster_bits < MIN_CLUSTER_BITS ||
773
        s->cluster_bits > MAX_CLUSTER_BITS)
774
    {
775
        fprintf(stderr, "Cluster size must be a power of two between "
776
            "%d and %dk\n",
777
            1 << MIN_CLUSTER_BITS,
778
            1 << (MAX_CLUSTER_BITS - 10));
779
        return -EINVAL;
780
    }
781
    s->cluster_size = 1 << s->cluster_bits;
782

    
783
    header.cluster_bits = cpu_to_be32(s->cluster_bits);
784
    header_size = (header_size + 7) & ~7;
785
    if (flags & BLOCK_FLAG_ENCRYPT) {
786
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
787
    } else {
788
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
789
    }
790
    l2_bits = s->cluster_bits - 3;
791
    shift = s->cluster_bits + l2_bits;
792
    l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
793
    offset = align_offset(header_size, s->cluster_size);
794
    s->l1_table_offset = offset;
795
    header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
796
    header.l1_size = cpu_to_be32(l1_size);
797
    offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
798

    
799
    s->refcount_table = qemu_mallocz(s->cluster_size);
800

    
801
    s->refcount_table_offset = offset;
802
    header.refcount_table_offset = cpu_to_be64(offset);
803
    header.refcount_table_clusters = cpu_to_be32(1);
804
    offset += s->cluster_size;
805
    s->refcount_block_offset = offset;
806

    
807
    /* count how many refcount blocks needed */
808
    tmp = offset >> s->cluster_bits;
809
    ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
810
    for (i=0; i < ref_clusters; i++) {
811
        s->refcount_table[i] = cpu_to_be64(offset);
812
        offset += s->cluster_size;
813
    }
814

    
815
    s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
816

    
817
    /* update refcounts */
818
    qcow2_create_refcount_update(s, 0, header_size);
819
    qcow2_create_refcount_update(s, s->l1_table_offset,
820
        l1_size * sizeof(uint64_t));
821
    qcow2_create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
822
    qcow2_create_refcount_update(s, s->refcount_block_offset,
823
        ref_clusters * s->cluster_size);
824

    
825
    /* write all the data */
826
    write(fd, &header, sizeof(header));
827
    if (backing_file) {
828
        if (backing_format_len) {
829
            char zero[16];
830
            int d = ext_bf.len - backing_format_len;
831

    
832
            memset(zero, 0, sizeof(zero));
833
            cpu_to_be32s(&ext_bf.magic);
834
            cpu_to_be32s(&ext_bf.len);
835
            write(fd, &ext_bf, sizeof(ext_bf));
836
            write(fd, backing_format, backing_format_len);
837
            if (d>0) {
838
                write(fd, zero, d);
839
            }
840
        }
841
        write(fd, backing_file, backing_filename_len);
842
    }
843
    lseek(fd, s->l1_table_offset, SEEK_SET);
844
    tmp = 0;
845
    for(i = 0;i < l1_size; i++) {
846
        write(fd, &tmp, sizeof(tmp));
847
    }
848
    lseek(fd, s->refcount_table_offset, SEEK_SET);
849
    write(fd, s->refcount_table, s->cluster_size);
850

    
851
    lseek(fd, s->refcount_block_offset, SEEK_SET);
852
    write(fd, s->refcount_block, ref_clusters * s->cluster_size);
853

    
854
    qemu_free(s->refcount_table);
855
    qemu_free(s->refcount_block);
856
    close(fd);
857

    
858
    /* Preallocate metadata */
859
    if (prealloc) {
860
        BlockDriverState *bs;
861
        bs = bdrv_new("");
862
        bdrv_open(bs, filename, BDRV_O_CACHE_WB);
863
        preallocate(bs);
864
        bdrv_close(bs);
865
    }
866

    
867
    return 0;
868
}
869

    
870
static int qcow_create(const char *filename, QEMUOptionParameter *options)
871
{
872
    const char *backing_file = NULL;
873
    const char *backing_fmt = NULL;
874
    uint64_t sectors = 0;
875
    int flags = 0;
876
    size_t cluster_size = 65536;
877
    int prealloc = 0;
878

    
879
    /* Read out options */
880
    while (options && options->name) {
881
        if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
882
            sectors = options->value.n / 512;
883
        } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
884
            backing_file = options->value.s;
885
        } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
886
            backing_fmt = options->value.s;
887
        } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
888
            flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
889
        } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
890
            if (options->value.n) {
891
                cluster_size = options->value.n;
892
            }
893
        } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
894
            if (!options->value.s || !strcmp(options->value.s, "off")) {
895
                prealloc = 0;
896
            } else if (!strcmp(options->value.s, "metadata")) {
897
                prealloc = 1;
898
            } else {
899
                fprintf(stderr, "Invalid preallocation mode: '%s'\n",
900
                    options->value.s);
901
                return -EINVAL;
902
            }
903
        }
904
        options++;
905
    }
906

    
907
    if (backing_file && prealloc) {
908
        fprintf(stderr, "Backing file and preallocation cannot be used at "
909
            "the same time\n");
910
        return -EINVAL;
911
    }
912

    
913
    return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
914
        cluster_size, prealloc);
915
}
916

    
917
static int qcow_make_empty(BlockDriverState *bs)
918
{
919
#if 0
920
    /* XXX: not correct */
921
    BDRVQcowState *s = bs->opaque;
922
    uint32_t l1_length = s->l1_size * sizeof(uint64_t);
923
    int ret;
924

925
    memset(s->l1_table, 0, l1_length);
926
    if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
927
        return -1;
928
    ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
929
    if (ret < 0)
930
        return ret;
931

932
    l2_cache_reset(bs);
933
#endif
934
    return 0;
935
}
936

    
937
/* XXX: put compressed sectors first, then all the cluster aligned
938
   tables to avoid losing bytes in alignment */
939
static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
940
                                 const uint8_t *buf, int nb_sectors)
941
{
942
    BDRVQcowState *s = bs->opaque;
943
    z_stream strm;
944
    int ret, out_len;
945
    uint8_t *out_buf;
946
    uint64_t cluster_offset;
947

    
948
    if (nb_sectors == 0) {
949
        /* align end of file to a sector boundary to ease reading with
950
           sector based I/Os */
951
        cluster_offset = bdrv_getlength(s->hd);
952
        cluster_offset = (cluster_offset + 511) & ~511;
953
        bdrv_truncate(s->hd, cluster_offset);
954
        return 0;
955
    }
956

    
957
    if (nb_sectors != s->cluster_sectors)
958
        return -EINVAL;
959

    
960
    out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
961

    
962
    /* best compression, small window, no zlib header */
963
    memset(&strm, 0, sizeof(strm));
964
    ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
965
                       Z_DEFLATED, -12,
966
                       9, Z_DEFAULT_STRATEGY);
967
    if (ret != 0) {
968
        qemu_free(out_buf);
969
        return -1;
970
    }
971

    
972
    strm.avail_in = s->cluster_size;
973
    strm.next_in = (uint8_t *)buf;
974
    strm.avail_out = s->cluster_size;
975
    strm.next_out = out_buf;
976

    
977
    ret = deflate(&strm, Z_FINISH);
978
    if (ret != Z_STREAM_END && ret != Z_OK) {
979
        qemu_free(out_buf);
980
        deflateEnd(&strm);
981
        return -1;
982
    }
983
    out_len = strm.next_out - out_buf;
984

    
985
    deflateEnd(&strm);
986

    
987
    if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
988
        /* could not compress: write normal cluster */
989
        bdrv_write(bs, sector_num, buf, s->cluster_sectors);
990
    } else {
991
        cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
992
            sector_num << 9, out_len);
993
        if (!cluster_offset)
994
            return -1;
995
        cluster_offset &= s->cluster_offset_mask;
996
        if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
997
            qemu_free(out_buf);
998
            return -1;
999
        }
1000
    }
1001

    
1002
    qemu_free(out_buf);
1003
    return 0;
1004
}
1005

    
1006
static void qcow_flush(BlockDriverState *bs)
1007
{
1008
    BDRVQcowState *s = bs->opaque;
1009
    bdrv_flush(s->hd);
1010
}
1011

    
1012
static int64_t qcow_vm_state_offset(BDRVQcowState *s)
1013
{
1014
        return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1015
}
1016

    
1017
static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1018
{
1019
    BDRVQcowState *s = bs->opaque;
1020
    bdi->cluster_size = s->cluster_size;
1021
    bdi->vm_state_offset = qcow_vm_state_offset(s);
1022
    return 0;
1023
}
1024

    
1025

    
1026
static int qcow_check(BlockDriverState *bs)
1027
{
1028
    return qcow2_check_refcounts(bs);
1029
}
1030

    
1031
#if 0
1032
static void dump_refcounts(BlockDriverState *bs)
1033
{
1034
    BDRVQcowState *s = bs->opaque;
1035
    int64_t nb_clusters, k, k1, size;
1036
    int refcount;
1037

1038
    size = bdrv_getlength(s->hd);
1039
    nb_clusters = size_to_clusters(s, size);
1040
    for(k = 0; k < nb_clusters;) {
1041
        k1 = k;
1042
        refcount = get_refcount(bs, k);
1043
        k++;
1044
        while (k < nb_clusters && get_refcount(bs, k) == refcount)
1045
            k++;
1046
        printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
1047
    }
1048
}
1049
#endif
1050

    
1051
static int qcow_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1052
                           int64_t pos, int size)
1053
{
1054
    BDRVQcowState *s = bs->opaque;
1055
    int growable = bs->growable;
1056

    
1057
    bs->growable = 1;
1058
    bdrv_pwrite(bs, qcow_vm_state_offset(s) + pos, buf, size);
1059
    bs->growable = growable;
1060

    
1061
    return size;
1062
}
1063

    
1064
static int qcow_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1065
                           int64_t pos, int size)
1066
{
1067
    BDRVQcowState *s = bs->opaque;
1068
    int growable = bs->growable;
1069
    int ret;
1070

    
1071
    bs->growable = 1;
1072
    ret = bdrv_pread(bs, qcow_vm_state_offset(s) + pos, buf, size);
1073
    bs->growable = growable;
1074

    
1075
    return ret;
1076
}
1077

    
1078
static QEMUOptionParameter qcow_create_options[] = {
1079
    {
1080
        .name = BLOCK_OPT_SIZE,
1081
        .type = OPT_SIZE,
1082
        .help = "Virtual disk size"
1083
    },
1084
    {
1085
        .name = BLOCK_OPT_BACKING_FILE,
1086
        .type = OPT_STRING,
1087
        .help = "File name of a base image"
1088
    },
1089
    {
1090
        .name = BLOCK_OPT_BACKING_FMT,
1091
        .type = OPT_STRING,
1092
        .help = "Image format of the base image"
1093
    },
1094
    {
1095
        .name = BLOCK_OPT_ENCRYPT,
1096
        .type = OPT_FLAG,
1097
        .help = "Encrypt the image"
1098
    },
1099
    {
1100
        .name = BLOCK_OPT_CLUSTER_SIZE,
1101
        .type = OPT_SIZE,
1102
        .help = "qcow2 cluster size"
1103
    },
1104
    {
1105
        .name = BLOCK_OPT_PREALLOC,
1106
        .type = OPT_STRING,
1107
        .help = "Preallocation mode (allowed values: off, metadata)"
1108
    },
1109
    { NULL }
1110
};
1111

    
1112
static BlockDriver bdrv_qcow2 = {
1113
    .format_name        = "qcow2",
1114
    .instance_size        = sizeof(BDRVQcowState),
1115
    .bdrv_probe                = qcow_probe,
1116
    .bdrv_open                = qcow_open,
1117
    .bdrv_close                = qcow_close,
1118
    .bdrv_create        = qcow_create,
1119
    .bdrv_flush                = qcow_flush,
1120
    .bdrv_is_allocated        = qcow_is_allocated,
1121
    .bdrv_set_key        = qcow_set_key,
1122
    .bdrv_make_empty        = qcow_make_empty,
1123

    
1124
    .bdrv_aio_readv        = qcow_aio_readv,
1125
    .bdrv_aio_writev        = qcow_aio_writev,
1126
    .bdrv_write_compressed = qcow_write_compressed,
1127

    
1128
    .bdrv_snapshot_create   = qcow2_snapshot_create,
1129
    .bdrv_snapshot_goto     = qcow2_snapshot_goto,
1130
    .bdrv_snapshot_delete   = qcow2_snapshot_delete,
1131
    .bdrv_snapshot_list     = qcow2_snapshot_list,
1132
    .bdrv_get_info        = qcow_get_info,
1133

    
1134
    .bdrv_save_vmstate    = qcow_save_vmstate,
1135
    .bdrv_load_vmstate    = qcow_load_vmstate,
1136

    
1137
    .create_options = qcow_create_options,
1138
    .bdrv_check = qcow_check,
1139
};
1140

    
1141
static void bdrv_qcow2_init(void)
1142
{
1143
    bdrv_register(&bdrv_qcow2);
1144
}
1145

    
1146
block_init(bdrv_qcow2_init);