Statistics
| Branch: | Revision:

root / block / qcow2.c @ a9420734

History | View | Annotate | Download (44.8 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
#include "qemu-error.h"
31

    
32
/*
33
  Differences with QCOW:
34

35
  - Support for multiple incremental snapshots.
36
  - Memory management by reference counts.
37
  - Clusters which have a reference count of one have the bit
38
    QCOW_OFLAG_COPIED to optimize write performance.
39
  - Size of compressed clusters is stored in sectors to reduce bit usage
40
    in the cluster offsets.
41
  - Support for storing additional data (such as the VM state) in the
42
    snapshots.
43
  - If a backing store is used, the cluster size is not constrained
44
    (could be backported to QCOW).
45
  - L2 tables have always a size of one cluster.
46
*/
47

    
48

    
49
typedef struct {
50
    uint32_t magic;
51
    uint32_t len;
52
} QCowExtension;
53
#define  QCOW_EXT_MAGIC_END 0
54
#define  QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
55

    
56
static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
57
{
58
    const QCowHeader *cow_header = (const void *)buf;
59

    
60
    if (buf_size >= sizeof(QCowHeader) &&
61
        be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
62
        be32_to_cpu(cow_header->version) == QCOW_VERSION)
63
        return 100;
64
    else
65
        return 0;
66
}
67

    
68

    
69
/* 
70
 * read qcow2 extension and fill bs
71
 * start reading from start_offset
72
 * finish reading upon magic of value 0 or when end_offset reached
73
 * unknown magic is skipped (future extension this version knows nothing about)
74
 * return 0 upon success, non-0 otherwise
75
 */
76
static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
77
                                uint64_t end_offset)
78
{
79
    QCowExtension ext;
80
    uint64_t offset;
81

    
82
#ifdef DEBUG_EXT
83
    printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
84
#endif
85
    offset = start_offset;
86
    while (offset < end_offset) {
87

    
88
#ifdef DEBUG_EXT
89
        /* Sanity check */
90
        if (offset > s->cluster_size)
91
            printf("qcow_handle_extension: suspicious offset %lu\n", offset);
92

    
93
        printf("attemting to read extended header in offset %lu\n", offset);
94
#endif
95

    
96
        if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
97
            fprintf(stderr, "qcow_handle_extension: ERROR: "
98
                    "pread fail from offset %" PRIu64 "\n",
99
                    offset);
100
            return 1;
101
        }
102
        be32_to_cpus(&ext.magic);
103
        be32_to_cpus(&ext.len);
104
        offset += sizeof(ext);
105
#ifdef DEBUG_EXT
106
        printf("ext.magic = 0x%x\n", ext.magic);
107
#endif
108
        switch (ext.magic) {
109
        case QCOW_EXT_MAGIC_END:
110
            return 0;
111

    
112
        case QCOW_EXT_MAGIC_BACKING_FORMAT:
113
            if (ext.len >= sizeof(bs->backing_format)) {
114
                fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
115
                        " (>=%zu)\n",
116
                        ext.len, sizeof(bs->backing_format));
117
                return 2;
118
            }
119
            if (bdrv_pread(bs->file, offset , bs->backing_format,
120
                           ext.len) != ext.len)
121
                return 3;
122
            bs->backing_format[ext.len] = '\0';
123
#ifdef DEBUG_EXT
124
            printf("Qcow2: Got format extension %s\n", bs->backing_format);
125
#endif
126
            offset = ((offset + ext.len + 7) & ~7);
127
            break;
128

    
129
        default:
130
            /* unknown magic -- just skip it */
131
            offset = ((offset + ext.len + 7) & ~7);
132
            break;
133
        }
134
    }
135

    
136
    return 0;
137
}
138

    
139

    
140
static int qcow_open(BlockDriverState *bs, int flags)
141
{
142
    BDRVQcowState *s = bs->opaque;
143
    int len, i;
144
    QCowHeader header;
145
    uint64_t ext_end;
146

    
147
    if (bdrv_pread(bs->file, 0, &header, sizeof(header)) != sizeof(header))
148
        goto fail;
149
    be32_to_cpus(&header.magic);
150
    be32_to_cpus(&header.version);
151
    be64_to_cpus(&header.backing_file_offset);
152
    be32_to_cpus(&header.backing_file_size);
153
    be64_to_cpus(&header.size);
154
    be32_to_cpus(&header.cluster_bits);
155
    be32_to_cpus(&header.crypt_method);
156
    be64_to_cpus(&header.l1_table_offset);
157
    be32_to_cpus(&header.l1_size);
158
    be64_to_cpus(&header.refcount_table_offset);
159
    be32_to_cpus(&header.refcount_table_clusters);
160
    be64_to_cpus(&header.snapshots_offset);
161
    be32_to_cpus(&header.nb_snapshots);
162

    
163
    if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
164
        goto fail;
165
    if (header.cluster_bits < MIN_CLUSTER_BITS ||
166
        header.cluster_bits > MAX_CLUSTER_BITS)
167
        goto fail;
168
    if (header.crypt_method > QCOW_CRYPT_AES)
169
        goto fail;
170
    s->crypt_method_header = header.crypt_method;
171
    if (s->crypt_method_header)
172
        bs->encrypted = 1;
173
    s->cluster_bits = header.cluster_bits;
174
    s->cluster_size = 1 << s->cluster_bits;
175
    s->cluster_sectors = 1 << (s->cluster_bits - 9);
176
    s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
177
    s->l2_size = 1 << s->l2_bits;
178
    bs->total_sectors = header.size / 512;
179
    s->csize_shift = (62 - (s->cluster_bits - 8));
180
    s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
181
    s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
182
    s->refcount_table_offset = header.refcount_table_offset;
183
    s->refcount_table_size =
184
        header.refcount_table_clusters << (s->cluster_bits - 3);
185

    
186
    s->snapshots_offset = header.snapshots_offset;
187
    s->nb_snapshots = header.nb_snapshots;
188

    
189
    /* read the level 1 table */
190
    s->l1_size = header.l1_size;
191
    s->l1_vm_state_index = size_to_l1(s, header.size);
192
    /* the L1 table must contain at least enough entries to put
193
       header.size bytes */
194
    if (s->l1_size < s->l1_vm_state_index)
195
        goto fail;
196
    s->l1_table_offset = header.l1_table_offset;
197
    if (s->l1_size > 0) {
198
        s->l1_table = qemu_mallocz(
199
            align_offset(s->l1_size * sizeof(uint64_t), 512));
200
        if (bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
201
            s->l1_size * sizeof(uint64_t))
202
            goto fail;
203
        for(i = 0;i < s->l1_size; i++) {
204
            be64_to_cpus(&s->l1_table[i]);
205
        }
206
    }
207
    /* alloc L2 cache */
208
    s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
209
    s->cluster_cache = qemu_malloc(s->cluster_size);
210
    /* one more sector for decompressed data alignment */
211
    s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
212
                                  + 512);
213
    s->cluster_cache_offset = -1;
214

    
215
    if (qcow2_refcount_init(bs) < 0)
216
        goto fail;
217

    
218
    QLIST_INIT(&s->cluster_allocs);
219

    
220
    /* read qcow2 extensions */
221
    if (header.backing_file_offset)
222
        ext_end = header.backing_file_offset;
223
    else
224
        ext_end = s->cluster_size;
225
    if (qcow_read_extensions(bs, sizeof(header), ext_end))
226
        goto fail;
227

    
228
    /* read the backing file name */
229
    if (header.backing_file_offset != 0) {
230
        len = header.backing_file_size;
231
        if (len > 1023)
232
            len = 1023;
233
        if (bdrv_pread(bs->file, header.backing_file_offset, bs->backing_file, len) != len)
234
            goto fail;
235
        bs->backing_file[len] = '\0';
236
    }
237
    if (qcow2_read_snapshots(bs) < 0)
238
        goto fail;
239

    
240
#ifdef DEBUG_ALLOC
241
    qcow2_check_refcounts(bs);
242
#endif
243
    return 0;
244

    
245
 fail:
246
    qcow2_free_snapshots(bs);
247
    qcow2_refcount_close(bs);
248
    qemu_free(s->l1_table);
249
    qemu_free(s->l2_cache);
250
    qemu_free(s->cluster_cache);
251
    qemu_free(s->cluster_data);
252
    return -1;
253
}
254

    
255
static int qcow_set_key(BlockDriverState *bs, const char *key)
256
{
257
    BDRVQcowState *s = bs->opaque;
258
    uint8_t keybuf[16];
259
    int len, i;
260

    
261
    memset(keybuf, 0, 16);
262
    len = strlen(key);
263
    if (len > 16)
264
        len = 16;
265
    /* XXX: we could compress the chars to 7 bits to increase
266
       entropy */
267
    for(i = 0;i < len;i++) {
268
        keybuf[i] = key[i];
269
    }
270
    s->crypt_method = s->crypt_method_header;
271

    
272
    if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
273
        return -1;
274
    if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
275
        return -1;
276
#if 0
277
    /* test */
278
    {
279
        uint8_t in[16];
280
        uint8_t out[16];
281
        uint8_t tmp[16];
282
        for(i=0;i<16;i++)
283
            in[i] = i;
284
        AES_encrypt(in, tmp, &s->aes_encrypt_key);
285
        AES_decrypt(tmp, out, &s->aes_decrypt_key);
286
        for(i = 0; i < 16; i++)
287
            printf(" %02x", tmp[i]);
288
        printf("\n");
289
        for(i = 0; i < 16; i++)
290
            printf(" %02x", out[i]);
291
        printf("\n");
292
    }
293
#endif
294
    return 0;
295
}
296

    
297
static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
298
                             int nb_sectors, int *pnum)
299
{
300
    uint64_t cluster_offset;
301
    int ret;
302

    
303
    *pnum = nb_sectors;
304
    /* FIXME We can get errors here, but the bdrv_is_allocated interface can't
305
     * pass them on today */
306
    ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
307
    if (ret < 0) {
308
        *pnum = 0;
309
    }
310

    
311
    return (cluster_offset != 0);
312
}
313

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

    
326
    qemu_iovec_memset(qiov, 0, 512 * (nb_sectors - n1));
327

    
328
    return n1;
329
}
330

    
331
typedef struct QCowAIOCB {
332
    BlockDriverAIOCB common;
333
    int64_t sector_num;
334
    QEMUIOVector *qiov;
335
    int remaining_sectors;
336
    int cur_nr_sectors;        /* number of sectors in current iteration */
337
    uint64_t bytes_done;
338
    uint64_t cluster_offset;
339
    uint8_t *cluster_data;
340
    BlockDriverAIOCB *hd_aiocb;
341
    QEMUIOVector hd_qiov;
342
    QEMUBH *bh;
343
    QCowL2Meta l2meta;
344
    QLIST_ENTRY(QCowAIOCB) next_depend;
345
} QCowAIOCB;
346

    
347
static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
348
{
349
    QCowAIOCB *acb = container_of(blockacb, QCowAIOCB, common);
350
    if (acb->hd_aiocb)
351
        bdrv_aio_cancel(acb->hd_aiocb);
352
    qemu_aio_release(acb);
353
}
354

    
355
static AIOPool qcow_aio_pool = {
356
    .aiocb_size         = sizeof(QCowAIOCB),
357
    .cancel             = qcow_aio_cancel,
358
};
359

    
360
static void qcow_aio_read_cb(void *opaque, int ret);
361
static void qcow_aio_read_bh(void *opaque)
362
{
363
    QCowAIOCB *acb = opaque;
364
    qemu_bh_delete(acb->bh);
365
    acb->bh = NULL;
366
    qcow_aio_read_cb(opaque, 0);
367
}
368

    
369
static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
370
{
371
    if (acb->bh)
372
        return -EIO;
373

    
374
    acb->bh = qemu_bh_new(cb, acb);
375
    if (!acb->bh)
376
        return -EIO;
377

    
378
    qemu_bh_schedule(acb->bh);
379

    
380
    return 0;
381
}
382

    
383
static void qcow_aio_read_cb(void *opaque, int ret)
384
{
385
    QCowAIOCB *acb = opaque;
386
    BlockDriverState *bs = acb->common.bs;
387
    BDRVQcowState *s = bs->opaque;
388
    int index_in_cluster, n1;
389

    
390
    acb->hd_aiocb = NULL;
391
    if (ret < 0)
392
        goto done;
393

    
394
    /* post process the read buffer */
395
    if (!acb->cluster_offset) {
396
        /* nothing to do */
397
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
398
        /* nothing to do */
399
    } else {
400
        if (s->crypt_method) {
401
            qcow2_encrypt_sectors(s, acb->sector_num,  acb->cluster_data,
402
                acb->cluster_data, acb->cur_nr_sectors, 0, &s->aes_decrypt_key);
403
            qemu_iovec_reset(&acb->hd_qiov);
404
            qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done,
405
                acb->cur_nr_sectors * 512);
406
            qemu_iovec_from_buffer(&acb->hd_qiov, acb->cluster_data,
407
                512 * acb->cur_nr_sectors);
408
        }
409
    }
410

    
411
    acb->remaining_sectors -= acb->cur_nr_sectors;
412
    acb->sector_num += acb->cur_nr_sectors;
413
    acb->bytes_done += acb->cur_nr_sectors * 512;
414

    
415
    if (acb->remaining_sectors == 0) {
416
        /* request completed */
417
        ret = 0;
418
        goto done;
419
    }
420

    
421
    /* prepare next AIO request */
422
    acb->cur_nr_sectors = acb->remaining_sectors;
423
    if (s->crypt_method) {
424
        acb->cur_nr_sectors = MIN(acb->cur_nr_sectors,
425
            QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
426
    }
427

    
428
    ret = qcow2_get_cluster_offset(bs, acb->sector_num << 9,
429
        &acb->cur_nr_sectors, &acb->cluster_offset);
430
    if (ret < 0) {
431
        goto done;
432
    }
433

    
434
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
435

    
436
    qemu_iovec_reset(&acb->hd_qiov);
437
    qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done,
438
        acb->cur_nr_sectors * 512);
439

    
440
    if (!acb->cluster_offset) {
441

    
442
        if (bs->backing_hd) {
443
            /* read from the base image */
444
            n1 = qcow2_backing_read1(bs->backing_hd, &acb->hd_qiov,
445
                acb->sector_num, acb->cur_nr_sectors);
446
            if (n1 > 0) {
447
                BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
448
                acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
449
                                    &acb->hd_qiov, acb->cur_nr_sectors,
450
                                    qcow_aio_read_cb, acb);
451
                if (acb->hd_aiocb == NULL)
452
                    goto done;
453
            } else {
454
                ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
455
                if (ret < 0)
456
                    goto done;
457
            }
458
        } else {
459
            /* Note: in this case, no need to wait */
460
            qemu_iovec_memset(&acb->hd_qiov, 0, 512 * acb->cur_nr_sectors);
461
            ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
462
            if (ret < 0)
463
                goto done;
464
        }
465
    } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
466
        /* add AIO support for compressed blocks ? */
467
        if (qcow2_decompress_cluster(bs, acb->cluster_offset) < 0)
468
            goto done;
469

    
470
        qemu_iovec_from_buffer(&acb->hd_qiov,
471
            s->cluster_cache + index_in_cluster * 512,
472
            512 * acb->cur_nr_sectors);
473

    
474
        ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
475
        if (ret < 0)
476
            goto done;
477
    } else {
478
        if ((acb->cluster_offset & 511) != 0) {
479
            ret = -EIO;
480
            goto done;
481
        }
482

    
483
        if (s->crypt_method) {
484
            /*
485
             * For encrypted images, read everything into a temporary
486
             * contiguous buffer on which the AES functions can work.
487
             */
488
            if (!acb->cluster_data) {
489
                acb->cluster_data =
490
                    qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
491
            }
492

    
493
            assert(acb->cur_nr_sectors <=
494
                QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
495
            qemu_iovec_reset(&acb->hd_qiov);
496
            qemu_iovec_add(&acb->hd_qiov, acb->cluster_data,
497
                512 * acb->cur_nr_sectors);
498
        }
499

    
500
        BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
501
        acb->hd_aiocb = bdrv_aio_readv(bs->file,
502
                            (acb->cluster_offset >> 9) + index_in_cluster,
503
                            &acb->hd_qiov, acb->cur_nr_sectors,
504
                            qcow_aio_read_cb, acb);
505
        if (acb->hd_aiocb == NULL) {
506
            ret = -EIO;
507
            goto done;
508
        }
509
    }
510

    
511
    return;
512
done:
513
    acb->common.cb(acb->common.opaque, ret);
514
    qemu_iovec_destroy(&acb->hd_qiov);
515
    qemu_aio_release(acb);
516
}
517

    
518
static 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

    
524
    acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
525
    if (!acb)
526
        return NULL;
527
    acb->hd_aiocb = NULL;
528
    acb->sector_num = sector_num;
529
    acb->qiov = qiov;
530

    
531
    qemu_iovec_init(&acb->hd_qiov, qiov->niov);
532

    
533
    acb->bytes_done = 0;
534
    acb->remaining_sectors = nb_sectors;
535
    acb->cur_nr_sectors = 0;
536
    acb->cluster_offset = 0;
537
    acb->l2meta.nb_clusters = 0;
538
    QLIST_INIT(&acb->l2meta.dependent_requests);
539
    return acb;
540
}
541

    
542
static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
543
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
544
        BlockDriverCompletionFunc *cb, void *opaque)
545
{
546
    QCowAIOCB *acb;
547

    
548
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
549
    if (!acb)
550
        return NULL;
551

    
552
    qcow_aio_read_cb(acb, 0);
553
    return &acb->common;
554
}
555

    
556
static void qcow_aio_write_cb(void *opaque, int ret);
557

    
558
static void run_dependent_requests(QCowL2Meta *m)
559
{
560
    QCowAIOCB *req;
561
    QCowAIOCB *next;
562

    
563
    /* Take the request off the list of running requests */
564
    if (m->nb_clusters != 0) {
565
        QLIST_REMOVE(m, next_in_flight);
566
    }
567

    
568
    /* Restart all dependent requests */
569
    QLIST_FOREACH_SAFE(req, &m->dependent_requests, next_depend, next) {
570
        qcow_aio_write_cb(req, 0);
571
    }
572

    
573
    /* Empty the list for the next part of the request */
574
    QLIST_INIT(&m->dependent_requests);
575
}
576

    
577
static void qcow_aio_write_cb(void *opaque, int ret)
578
{
579
    QCowAIOCB *acb = opaque;
580
    BlockDriverState *bs = acb->common.bs;
581
    BDRVQcowState *s = bs->opaque;
582
    int index_in_cluster;
583
    int n_end;
584

    
585
    acb->hd_aiocb = NULL;
586

    
587
    if (ret >= 0) {
588
        ret = qcow2_alloc_cluster_link_l2(bs, &acb->l2meta);
589
    }
590

    
591
    run_dependent_requests(&acb->l2meta);
592

    
593
    if (ret < 0)
594
        goto done;
595

    
596
    acb->remaining_sectors -= acb->cur_nr_sectors;
597
    acb->sector_num += acb->cur_nr_sectors;
598
    acb->bytes_done += acb->cur_nr_sectors * 512;
599

    
600
    if (acb->remaining_sectors == 0) {
601
        /* request completed */
602
        ret = 0;
603
        goto done;
604
    }
605

    
606
    index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
607
    n_end = index_in_cluster + acb->remaining_sectors;
608
    if (s->crypt_method &&
609
        n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
610
        n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
611

    
612
    ret = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
613
        index_in_cluster, n_end, &acb->cur_nr_sectors, &acb->l2meta);
614
    if (ret < 0) {
615
        goto done;
616
    }
617

    
618
    acb->cluster_offset = acb->l2meta.cluster_offset;
619

    
620
    /* Need to wait for another request? If so, we are done for now. */
621
    if (acb->l2meta.nb_clusters == 0 && acb->l2meta.depends_on != NULL) {
622
        QLIST_INSERT_HEAD(&acb->l2meta.depends_on->dependent_requests,
623
            acb, next_depend);
624
        return;
625
    }
626

    
627
    assert((acb->cluster_offset & 511) == 0);
628

    
629
    qemu_iovec_reset(&acb->hd_qiov);
630
    qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done,
631
        acb->cur_nr_sectors * 512);
632

    
633
    if (s->crypt_method) {
634
        if (!acb->cluster_data) {
635
            acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
636
                                             s->cluster_size);
637
        }
638

    
639
        assert(acb->hd_qiov.size <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
640
        qemu_iovec_to_buffer(&acb->hd_qiov, acb->cluster_data);
641

    
642
        qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data,
643
            acb->cluster_data, acb->cur_nr_sectors, 1, &s->aes_encrypt_key);
644

    
645
        qemu_iovec_reset(&acb->hd_qiov);
646
        qemu_iovec_add(&acb->hd_qiov, acb->cluster_data,
647
            acb->cur_nr_sectors * 512);
648
    }
649

    
650
    BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
651
    acb->hd_aiocb = bdrv_aio_writev(bs->file,
652
                                    (acb->cluster_offset >> 9) + index_in_cluster,
653
                                    &acb->hd_qiov, acb->cur_nr_sectors,
654
                                    qcow_aio_write_cb, acb);
655
    if (acb->hd_aiocb == NULL) {
656
        ret = -EIO;
657
        goto fail;
658
    }
659

    
660
    return;
661

    
662
fail:
663
    if (acb->l2meta.nb_clusters != 0) {
664
        QLIST_REMOVE(&acb->l2meta, next_in_flight);
665
    }
666
done:
667
    acb->common.cb(acb->common.opaque, ret);
668
    qemu_iovec_destroy(&acb->hd_qiov);
669
    qemu_aio_release(acb);
670
}
671

    
672
static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
673
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
674
        BlockDriverCompletionFunc *cb, void *opaque)
675
{
676
    BDRVQcowState *s = bs->opaque;
677
    QCowAIOCB *acb;
678

    
679
    s->cluster_cache_offset = -1; /* disable compressed cache */
680

    
681
    acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
682
    if (!acb)
683
        return NULL;
684

    
685
    qcow_aio_write_cb(acb, 0);
686
    return &acb->common;
687
}
688

    
689
static void qcow_close(BlockDriverState *bs)
690
{
691
    BDRVQcowState *s = bs->opaque;
692
    qemu_free(s->l1_table);
693
    qemu_free(s->l2_cache);
694
    qemu_free(s->cluster_cache);
695
    qemu_free(s->cluster_data);
696
    qcow2_refcount_close(bs);
697
}
698

    
699
/*
700
 * Updates the variable length parts of the qcow2 header, i.e. the backing file
701
 * name and all extensions. qcow2 was not designed to allow such changes, so if
702
 * we run out of space (we can only use the first cluster) this function may
703
 * fail.
704
 *
705
 * Returns 0 on success, -errno in error cases.
706
 */
707
static int qcow2_update_ext_header(BlockDriverState *bs,
708
    const char *backing_file, const char *backing_fmt)
709
{
710
    size_t backing_file_len = 0;
711
    size_t backing_fmt_len = 0;
712
    BDRVQcowState *s = bs->opaque;
713
    QCowExtension ext_backing_fmt = {0, 0};
714
    int ret;
715

    
716
    /* Backing file format doesn't make sense without a backing file */
717
    if (backing_fmt && !backing_file) {
718
        return -EINVAL;
719
    }
720

    
721
    /* Prepare the backing file format extension if needed */
722
    if (backing_fmt) {
723
        ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt));
724
        ext_backing_fmt.magic = cpu_to_be32(QCOW_EXT_MAGIC_BACKING_FORMAT);
725
        backing_fmt_len = ((sizeof(ext_backing_fmt)
726
            + strlen(backing_fmt) + 7) & ~7);
727
    }
728

    
729
    /* Check if we can fit the new header into the first cluster */
730
    if (backing_file) {
731
        backing_file_len = strlen(backing_file);
732
    }
733

    
734
    size_t header_size = sizeof(QCowHeader) + backing_file_len
735
        + backing_fmt_len;
736

    
737
    if (header_size > s->cluster_size) {
738
        return -ENOSPC;
739
    }
740

    
741
    /* Rewrite backing file name and qcow2 extensions */
742
    size_t ext_size = header_size - sizeof(QCowHeader);
743
    uint8_t buf[ext_size];
744
    size_t offset = 0;
745
    size_t backing_file_offset = 0;
746

    
747
    if (backing_file) {
748
        if (backing_fmt) {
749
            int padding = backing_fmt_len -
750
                (sizeof(ext_backing_fmt) + strlen(backing_fmt));
751

    
752
            memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt));
753
            offset += sizeof(ext_backing_fmt);
754

    
755
            memcpy(buf + offset, backing_fmt, strlen(backing_fmt));
756
            offset += strlen(backing_fmt);
757

    
758
            memset(buf + offset, 0, padding);
759
            offset += padding;
760
        }
761

    
762
        memcpy(buf + offset, backing_file, backing_file_len);
763
        backing_file_offset = sizeof(QCowHeader) + offset;
764
    }
765

    
766
    ret = bdrv_pwrite_sync(bs->file, sizeof(QCowHeader), buf, ext_size);
767
    if (ret < 0) {
768
        goto fail;
769
    }
770

    
771
    /* Update header fields */
772
    uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset);
773
    uint32_t be_backing_file_size = cpu_to_be32(backing_file_len);
774

    
775
    ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_offset),
776
        &be_backing_file_offset, sizeof(uint64_t));
777
    if (ret < 0) {
778
        goto fail;
779
    }
780

    
781
    ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_size),
782
        &be_backing_file_size, sizeof(uint32_t));
783
    if (ret < 0) {
784
        goto fail;
785
    }
786

    
787
    ret = 0;
788
fail:
789
    return ret;
790
}
791

    
792
static int qcow2_change_backing_file(BlockDriverState *bs,
793
    const char *backing_file, const char *backing_fmt)
794
{
795
    return qcow2_update_ext_header(bs, backing_file, backing_fmt);
796
}
797

    
798
#if 0
799
static int get_bits_from_size(size_t size)
800
{
801
    int res = 0;
802

803
    if (size == 0) {
804
        return -1;
805
    }
806

807
    while (size != 1) {
808
        /* Not a power of two */
809
        if (size & 1) {
810
            return -1;
811
        }
812

813
        size >>= 1;
814
        res++;
815
    }
816

817
    return res;
818
}
819
#endif
820

    
821

    
822
static int preallocate(BlockDriverState *bs)
823
{
824
    uint64_t nb_sectors;
825
    uint64_t offset;
826
    int num;
827
    int ret;
828
    QCowL2Meta meta;
829

    
830
    nb_sectors = bdrv_getlength(bs) >> 9;
831
    offset = 0;
832
    QLIST_INIT(&meta.dependent_requests);
833
    meta.cluster_offset = 0;
834

    
835
    while (nb_sectors) {
836
        num = MIN(nb_sectors, INT_MAX >> 9);
837
        ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
838
        if (ret < 0) {
839
            return ret;
840
        }
841

    
842
        ret = qcow2_alloc_cluster_link_l2(bs, &meta);
843
        if (ret < 0) {
844
            qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
845
            return ret;
846
        }
847

    
848
        /* There are no dependent requests, but we need to remove our request
849
         * from the list of in-flight requests */
850
        run_dependent_requests(&meta);
851

    
852
        /* TODO Preallocate data if requested */
853

    
854
        nb_sectors -= num;
855
        offset += num << 9;
856
    }
857

    
858
    /*
859
     * It is expected that the image file is large enough to actually contain
860
     * all of the allocated clusters (otherwise we get failing reads after
861
     * EOF). Extend the image to the last allocated sector.
862
     */
863
    if (meta.cluster_offset != 0) {
864
        uint8_t buf[512];
865
        memset(buf, 0, 512);
866
        ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1);
867
        if (ret < 0) {
868
            return ret;
869
        }
870
    }
871

    
872
    return 0;
873
}
874

    
875
#if 0
876
static int qcow_create2(const char *filename, int64_t total_size,
877
                        const char *backing_file, const char *backing_format,
878
                        int flags, size_t cluster_size, int prealloc)
879
{
880

881
    int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
882
    int ref_clusters, reftable_clusters, backing_format_len = 0;
883
    int rounded_ext_bf_len = 0;
884
    QCowHeader header;
885
    uint64_t tmp, offset;
886
    uint64_t old_ref_clusters;
887
    QCowCreateState s1, *s = &s1;
888
    QCowExtension ext_bf = {0, 0};
889
    int ret;
890

891
    memset(s, 0, sizeof(*s));
892

893
    fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
894
    if (fd < 0)
895
        return -errno;
896
    memset(&header, 0, sizeof(header));
897
    header.magic = cpu_to_be32(QCOW_MAGIC);
898
    header.version = cpu_to_be32(QCOW_VERSION);
899
    header.size = cpu_to_be64(total_size * 512);
900
    header_size = sizeof(header);
901
    backing_filename_len = 0;
902
    if (backing_file) {
903
        if (backing_format) {
904
            ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
905
            backing_format_len = strlen(backing_format);
906
            ext_bf.len = backing_format_len;
907
            rounded_ext_bf_len = (sizeof(ext_bf) + ext_bf.len + 7) & ~7;
908
            header_size += rounded_ext_bf_len;
909
        }
910
        header.backing_file_offset = cpu_to_be64(header_size);
911
        backing_filename_len = strlen(backing_file);
912
        header.backing_file_size = cpu_to_be32(backing_filename_len);
913
        header_size += backing_filename_len;
914
    }
915

916
    /* Cluster size */
917
    s->cluster_bits = get_bits_from_size(cluster_size);
918
    if (s->cluster_bits < MIN_CLUSTER_BITS ||
919
        s->cluster_bits > MAX_CLUSTER_BITS)
920
    {
921
        fprintf(stderr, "Cluster size must be a power of two between "
922
            "%d and %dk\n",
923
            1 << MIN_CLUSTER_BITS,
924
            1 << (MAX_CLUSTER_BITS - 10));
925
        return -EINVAL;
926
    }
927
    s->cluster_size = 1 << s->cluster_bits;
928

929
    header.cluster_bits = cpu_to_be32(s->cluster_bits);
930
    header_size = (header_size + 7) & ~7;
931
    if (flags & BLOCK_FLAG_ENCRYPT) {
932
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
933
    } else {
934
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
935
    }
936
    l2_bits = s->cluster_bits - 3;
937
    shift = s->cluster_bits + l2_bits;
938
    l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
939
    offset = align_offset(header_size, s->cluster_size);
940
    s->l1_table_offset = offset;
941
    header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
942
    header.l1_size = cpu_to_be32(l1_size);
943
    offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
944

945
    /* count how many refcount blocks needed */
946

947
#define NUM_CLUSTERS(bytes) \
948
    (((bytes) + (s->cluster_size) - 1) / (s->cluster_size))
949

950
    ref_clusters = NUM_CLUSTERS(NUM_CLUSTERS(offset) * sizeof(uint16_t));
951

952
    do {
953
        uint64_t image_clusters;
954
        old_ref_clusters = ref_clusters;
955

956
        /* Number of clusters used for the refcount table */
957
        reftable_clusters = NUM_CLUSTERS(ref_clusters * sizeof(uint64_t));
958

959
        /* Number of clusters that the whole image will have */
960
        image_clusters = NUM_CLUSTERS(offset) + ref_clusters
961
            + reftable_clusters;
962

963
        /* Number of refcount blocks needed for the image */
964
        ref_clusters = NUM_CLUSTERS(image_clusters * sizeof(uint16_t));
965

966
    } while (ref_clusters != old_ref_clusters);
967

968
    s->refcount_table = qemu_mallocz(reftable_clusters * s->cluster_size);
969

970
    s->refcount_table_offset = offset;
971
    header.refcount_table_offset = cpu_to_be64(offset);
972
    header.refcount_table_clusters = cpu_to_be32(reftable_clusters);
973
    offset += (reftable_clusters * s->cluster_size);
974
    s->refcount_block_offset = offset;
975

976
    for (i=0; i < ref_clusters; i++) {
977
        s->refcount_table[i] = cpu_to_be64(offset);
978
        offset += s->cluster_size;
979
    }
980

981
    s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
982

983
    /* update refcounts */
984
    qcow2_create_refcount_update(s, 0, header_size);
985
    qcow2_create_refcount_update(s, s->l1_table_offset,
986
        l1_size * sizeof(uint64_t));
987
    qcow2_create_refcount_update(s, s->refcount_table_offset,
988
        reftable_clusters * s->cluster_size);
989
    qcow2_create_refcount_update(s, s->refcount_block_offset,
990
        ref_clusters * s->cluster_size);
991

992
    /* write all the data */
993
    ret = qemu_write_full(fd, &header, sizeof(header));
994
    if (ret != sizeof(header)) {
995
        ret = -errno;
996
        goto exit;
997
    }
998
    if (backing_file) {
999
        if (backing_format_len) {
1000
            char zero[16];
1001
            int padding = rounded_ext_bf_len - (ext_bf.len + sizeof(ext_bf));
1002

1003
            memset(zero, 0, sizeof(zero));
1004
            cpu_to_be32s(&ext_bf.magic);
1005
            cpu_to_be32s(&ext_bf.len);
1006
            ret = qemu_write_full(fd, &ext_bf, sizeof(ext_bf));
1007
            if (ret != sizeof(ext_bf)) {
1008
                ret = -errno;
1009
                goto exit;
1010
            }
1011
            ret = qemu_write_full(fd, backing_format, backing_format_len);
1012
            if (ret != backing_format_len) {
1013
                ret = -errno;
1014
                goto exit;
1015
            }
1016
            if (padding > 0) {
1017
                ret = qemu_write_full(fd, zero, padding);
1018
                if (ret != padding) {
1019
                    ret = -errno;
1020
                    goto exit;
1021
                }
1022
            }
1023
        }
1024
        ret = qemu_write_full(fd, backing_file, backing_filename_len);
1025
        if (ret != backing_filename_len) {
1026
            ret = -errno;
1027
            goto exit;
1028
        }
1029
    }
1030
    lseek(fd, s->l1_table_offset, SEEK_SET);
1031
    tmp = 0;
1032
    for(i = 0;i < l1_size; i++) {
1033
        ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1034
        if (ret != sizeof(tmp)) {
1035
            ret = -errno;
1036
            goto exit;
1037
        }
1038
    }
1039
    lseek(fd, s->refcount_table_offset, SEEK_SET);
1040
    ret = qemu_write_full(fd, s->refcount_table,
1041
        reftable_clusters * s->cluster_size);
1042
    if (ret != reftable_clusters * s->cluster_size) {
1043
        ret = -errno;
1044
        goto exit;
1045
    }
1046

1047
    lseek(fd, s->refcount_block_offset, SEEK_SET);
1048
    ret = qemu_write_full(fd, s->refcount_block,
1049
                    ref_clusters * s->cluster_size);
1050
    if (ret != ref_clusters * s->cluster_size) {
1051
        ret = -errno;
1052
        goto exit;
1053
    }
1054

1055
    ret = 0;
1056
exit:
1057
    qemu_free(s->refcount_table);
1058
    qemu_free(s->refcount_block);
1059
    close(fd);
1060

1061
    /* Preallocate metadata */
1062
    if (ret == 0 && prealloc) {
1063
        BlockDriverState *bs;
1064
        BlockDriver *drv = bdrv_find_format("qcow2");
1065
        bs = bdrv_new("");
1066
        bdrv_open(bs, filename, BDRV_O_CACHE_WB | BDRV_O_RDWR, drv);
1067
        ret = preallocate(bs);
1068
        bdrv_close(bs);
1069
    }
1070

1071
    return ret;
1072
}
1073
#else
1074
static int qcow_create2(const char *filename, int64_t total_size,
1075
                        const char *backing_file, const char *backing_format,
1076
                        int flags, size_t cluster_size, int prealloc,
1077
                        QEMUOptionParameter *options)
1078
{
1079
    /* Calulate cluster_bits */
1080
    int cluster_bits;
1081
    cluster_bits = ffs(cluster_size) - 1;
1082
    if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
1083
        (1 << cluster_bits) != cluster_size)
1084
    {
1085
        error_report(
1086
            "Cluster size must be a power of two between %d and %dk\n",
1087
            1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
1088
        return -EINVAL;
1089
    }
1090

    
1091
    /*
1092
     * Open the image file and write a minimal qcow2 header.
1093
     *
1094
     * We keep things simple and start with a zero-sized image. We also
1095
     * do without refcount blocks or a L1 table for now. We'll fix the
1096
     * inconsistency later.
1097
     *
1098
     * We do need a refcount table because growing the refcount table means
1099
     * allocating two new refcount blocks - the seconds of which would be at
1100
     * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1101
     * size for any qcow2 image.
1102
     */
1103
    BlockDriverState* bs;
1104
    QCowHeader header;
1105
    uint8_t* refcount_table;
1106
    int ret;
1107

    
1108
    ret = bdrv_create_file(filename, options);
1109
    if (ret < 0) {
1110
        return ret;
1111
    }
1112

    
1113
    ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
1114
    if (ret < 0) {
1115
        return ret;
1116
    }
1117

    
1118
    /* Write the header */
1119
    memset(&header, 0, sizeof(header));
1120
    header.magic = cpu_to_be32(QCOW_MAGIC);
1121
    header.version = cpu_to_be32(QCOW_VERSION);
1122
    header.cluster_bits = cpu_to_be32(cluster_bits);
1123
    header.size = cpu_to_be64(0);
1124
    header.l1_table_offset = cpu_to_be64(0);
1125
    header.l1_size = cpu_to_be32(0);
1126
    header.refcount_table_offset = cpu_to_be64(cluster_size);
1127
    header.refcount_table_clusters = cpu_to_be32(1);
1128

    
1129
    if (flags & BLOCK_FLAG_ENCRYPT) {
1130
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1131
    } else {
1132
        header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1133
    }
1134

    
1135
    ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
1136
    if (ret < 0) {
1137
        goto out;
1138
    }
1139

    
1140
    /* Write an empty refcount table */
1141
    refcount_table = qemu_mallocz(cluster_size);
1142
    ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size);
1143
    qemu_free(refcount_table);
1144

    
1145
    if (ret < 0) {
1146
        goto out;
1147
    }
1148

    
1149
    bdrv_close(bs);
1150

    
1151
    /*
1152
     * And now open the image and make it consistent first (i.e. increase the
1153
     * refcount of the cluster that is occupied by the header and the refcount
1154
     * table)
1155
     */
1156
    BlockDriver* drv = bdrv_find_format("qcow2");
1157
    assert(drv != NULL);
1158
    ret = bdrv_open(bs, filename, BDRV_O_RDWR | BDRV_O_NO_FLUSH, drv);
1159
    if (ret < 0) {
1160
        goto out;
1161
    }
1162

    
1163
    ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
1164
    if (ret < 0) {
1165
        goto out;
1166

    
1167
    } else if (ret != 0) {
1168
        error_report("Huh, first cluster in empty image is already in use?");
1169
        abort();
1170
    }
1171

    
1172
    /* Okay, now that we have a valid image, let's give it the right size */
1173
    ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
1174
    if (ret < 0) {
1175
        goto out;
1176
    }
1177

    
1178
    /* Want a backing file? There you go.*/
1179
    if (backing_file) {
1180
        ret = bdrv_change_backing_file(bs, backing_file, backing_format);
1181
        if (ret < 0) {
1182
            goto out;
1183
        }
1184
    }
1185

    
1186
    /* And if we're supposed to preallocate metadata, do that now */
1187
    if (prealloc) {
1188
        ret = preallocate(bs);
1189
        if (ret < 0) {
1190
            goto out;
1191
        }
1192
    }
1193

    
1194
    ret = 0;
1195
out:
1196
    bdrv_delete(bs);
1197
    return ret;
1198
}
1199
#endif
1200

    
1201
static int qcow_create(const char *filename, QEMUOptionParameter *options)
1202
{
1203
    const char *backing_file = NULL;
1204
    const char *backing_fmt = NULL;
1205
    uint64_t sectors = 0;
1206
    int flags = 0;
1207
    size_t cluster_size = 65536;
1208
    int prealloc = 0;
1209

    
1210
    /* Read out options */
1211
    while (options && options->name) {
1212
        if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1213
            sectors = options->value.n / 512;
1214
        } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1215
            backing_file = options->value.s;
1216
        } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1217
            backing_fmt = options->value.s;
1218
        } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1219
            flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1220
        } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1221
            if (options->value.n) {
1222
                cluster_size = options->value.n;
1223
            }
1224
        } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1225
            if (!options->value.s || !strcmp(options->value.s, "off")) {
1226
                prealloc = 0;
1227
            } else if (!strcmp(options->value.s, "metadata")) {
1228
                prealloc = 1;
1229
            } else {
1230
                fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1231
                    options->value.s);
1232
                return -EINVAL;
1233
            }
1234
        }
1235
        options++;
1236
    }
1237

    
1238
    if (backing_file && prealloc) {
1239
        fprintf(stderr, "Backing file and preallocation cannot be used at "
1240
            "the same time\n");
1241
        return -EINVAL;
1242
    }
1243

    
1244
    return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
1245
        cluster_size, prealloc, options);
1246
}
1247

    
1248
static int qcow_make_empty(BlockDriverState *bs)
1249
{
1250
#if 0
1251
    /* XXX: not correct */
1252
    BDRVQcowState *s = bs->opaque;
1253
    uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1254
    int ret;
1255

1256
    memset(s->l1_table, 0, l1_length);
1257
    if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1258
        return -1;
1259
    ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1260
    if (ret < 0)
1261
        return ret;
1262

1263
    l2_cache_reset(bs);
1264
#endif
1265
    return 0;
1266
}
1267

    
1268
static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1269
{
1270
    BDRVQcowState *s = bs->opaque;
1271
    int ret, new_l1_size;
1272

    
1273
    if (offset & 511) {
1274
        return -EINVAL;
1275
    }
1276

    
1277
    /* cannot proceed if image has snapshots */
1278
    if (s->nb_snapshots) {
1279
        return -ENOTSUP;
1280
    }
1281

    
1282
    /* shrinking is currently not supported */
1283
    if (offset < bs->total_sectors * 512) {
1284
        return -ENOTSUP;
1285
    }
1286

    
1287
    new_l1_size = size_to_l1(s, offset);
1288
    ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1289
    if (ret < 0) {
1290
        return ret;
1291
    }
1292

    
1293
    /* write updated header.size */
1294
    offset = cpu_to_be64(offset);
1295
    ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1296
                           &offset, sizeof(uint64_t));
1297
    if (ret < 0) {
1298
        return ret;
1299
    }
1300

    
1301
    s->l1_vm_state_index = new_l1_size;
1302
    return 0;
1303
}
1304

    
1305
/* XXX: put compressed sectors first, then all the cluster aligned
1306
   tables to avoid losing bytes in alignment */
1307
static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1308
                                 const uint8_t *buf, int nb_sectors)
1309
{
1310
    BDRVQcowState *s = bs->opaque;
1311
    z_stream strm;
1312
    int ret, out_len;
1313
    uint8_t *out_buf;
1314
    uint64_t cluster_offset;
1315

    
1316
    if (nb_sectors == 0) {
1317
        /* align end of file to a sector boundary to ease reading with
1318
           sector based I/Os */
1319
        cluster_offset = bdrv_getlength(bs->file);
1320
        cluster_offset = (cluster_offset + 511) & ~511;
1321
        bdrv_truncate(bs->file, cluster_offset);
1322
        return 0;
1323
    }
1324

    
1325
    if (nb_sectors != s->cluster_sectors)
1326
        return -EINVAL;
1327

    
1328
    out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1329

    
1330
    /* best compression, small window, no zlib header */
1331
    memset(&strm, 0, sizeof(strm));
1332
    ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1333
                       Z_DEFLATED, -12,
1334
                       9, Z_DEFAULT_STRATEGY);
1335
    if (ret != 0) {
1336
        qemu_free(out_buf);
1337
        return -1;
1338
    }
1339

    
1340
    strm.avail_in = s->cluster_size;
1341
    strm.next_in = (uint8_t *)buf;
1342
    strm.avail_out = s->cluster_size;
1343
    strm.next_out = out_buf;
1344

    
1345
    ret = deflate(&strm, Z_FINISH);
1346
    if (ret != Z_STREAM_END && ret != Z_OK) {
1347
        qemu_free(out_buf);
1348
        deflateEnd(&strm);
1349
        return -1;
1350
    }
1351
    out_len = strm.next_out - out_buf;
1352

    
1353
    deflateEnd(&strm);
1354

    
1355
    if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1356
        /* could not compress: write normal cluster */
1357
        bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1358
    } else {
1359
        cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1360
            sector_num << 9, out_len);
1361
        if (!cluster_offset)
1362
            return -1;
1363
        cluster_offset &= s->cluster_offset_mask;
1364
        BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1365
        if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) {
1366
            qemu_free(out_buf);
1367
            return -1;
1368
        }
1369
    }
1370

    
1371
    qemu_free(out_buf);
1372
    return 0;
1373
}
1374

    
1375
static void qcow_flush(BlockDriverState *bs)
1376
{
1377
    bdrv_flush(bs->file);
1378
}
1379

    
1380
static BlockDriverAIOCB *qcow_aio_flush(BlockDriverState *bs,
1381
         BlockDriverCompletionFunc *cb, void *opaque)
1382
{
1383
    return bdrv_aio_flush(bs->file, cb, opaque);
1384
}
1385

    
1386
static int64_t qcow_vm_state_offset(BDRVQcowState *s)
1387
{
1388
        return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1389
}
1390

    
1391
static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1392
{
1393
    BDRVQcowState *s = bs->opaque;
1394
    bdi->cluster_size = s->cluster_size;
1395
    bdi->vm_state_offset = qcow_vm_state_offset(s);
1396
    return 0;
1397
}
1398

    
1399

    
1400
static int qcow_check(BlockDriverState *bs, BdrvCheckResult *result)
1401
{
1402
    return qcow2_check_refcounts(bs, result);
1403
}
1404

    
1405
#if 0
1406
static void dump_refcounts(BlockDriverState *bs)
1407
{
1408
    BDRVQcowState *s = bs->opaque;
1409
    int64_t nb_clusters, k, k1, size;
1410
    int refcount;
1411

1412
    size = bdrv_getlength(bs->file);
1413
    nb_clusters = size_to_clusters(s, size);
1414
    for(k = 0; k < nb_clusters;) {
1415
        k1 = k;
1416
        refcount = get_refcount(bs, k);
1417
        k++;
1418
        while (k < nb_clusters && get_refcount(bs, k) == refcount)
1419
            k++;
1420
        printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1421
               k - k1);
1422
    }
1423
}
1424
#endif
1425

    
1426
static int qcow_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1427
                           int64_t pos, int size)
1428
{
1429
    BDRVQcowState *s = bs->opaque;
1430
    int growable = bs->growable;
1431
    int ret;
1432

    
1433
    BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1434
    bs->growable = 1;
1435
    ret = bdrv_pwrite(bs, qcow_vm_state_offset(s) + pos, buf, size);
1436
    bs->growable = growable;
1437

    
1438
    return ret;
1439
}
1440

    
1441
static int qcow_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1442
                           int64_t pos, int size)
1443
{
1444
    BDRVQcowState *s = bs->opaque;
1445
    int growable = bs->growable;
1446
    int ret;
1447

    
1448
    BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1449
    bs->growable = 1;
1450
    ret = bdrv_pread(bs, qcow_vm_state_offset(s) + pos, buf, size);
1451
    bs->growable = growable;
1452

    
1453
    return ret;
1454
}
1455

    
1456
static QEMUOptionParameter qcow_create_options[] = {
1457
    {
1458
        .name = BLOCK_OPT_SIZE,
1459
        .type = OPT_SIZE,
1460
        .help = "Virtual disk size"
1461
    },
1462
    {
1463
        .name = BLOCK_OPT_BACKING_FILE,
1464
        .type = OPT_STRING,
1465
        .help = "File name of a base image"
1466
    },
1467
    {
1468
        .name = BLOCK_OPT_BACKING_FMT,
1469
        .type = OPT_STRING,
1470
        .help = "Image format of the base image"
1471
    },
1472
    {
1473
        .name = BLOCK_OPT_ENCRYPT,
1474
        .type = OPT_FLAG,
1475
        .help = "Encrypt the image"
1476
    },
1477
    {
1478
        .name = BLOCK_OPT_CLUSTER_SIZE,
1479
        .type = OPT_SIZE,
1480
        .help = "qcow2 cluster size"
1481
    },
1482
    {
1483
        .name = BLOCK_OPT_PREALLOC,
1484
        .type = OPT_STRING,
1485
        .help = "Preallocation mode (allowed values: off, metadata)"
1486
    },
1487
    { NULL }
1488
};
1489

    
1490
static BlockDriver bdrv_qcow2 = {
1491
    .format_name        = "qcow2",
1492
    .instance_size        = sizeof(BDRVQcowState),
1493
    .bdrv_probe                = qcow_probe,
1494
    .bdrv_open                = qcow_open,
1495
    .bdrv_close                = qcow_close,
1496
    .bdrv_create        = qcow_create,
1497
    .bdrv_flush                = qcow_flush,
1498
    .bdrv_is_allocated        = qcow_is_allocated,
1499
    .bdrv_set_key        = qcow_set_key,
1500
    .bdrv_make_empty        = qcow_make_empty,
1501

    
1502
    .bdrv_aio_readv        = qcow_aio_readv,
1503
    .bdrv_aio_writev        = qcow_aio_writev,
1504
    .bdrv_aio_flush        = qcow_aio_flush,
1505

    
1506
    .bdrv_truncate          = qcow2_truncate,
1507
    .bdrv_write_compressed  = qcow_write_compressed,
1508

    
1509
    .bdrv_snapshot_create   = qcow2_snapshot_create,
1510
    .bdrv_snapshot_goto     = qcow2_snapshot_goto,
1511
    .bdrv_snapshot_delete   = qcow2_snapshot_delete,
1512
    .bdrv_snapshot_list     = qcow2_snapshot_list,
1513
    .bdrv_get_info        = qcow_get_info,
1514

    
1515
    .bdrv_save_vmstate    = qcow_save_vmstate,
1516
    .bdrv_load_vmstate    = qcow_load_vmstate,
1517

    
1518
    .bdrv_change_backing_file   = qcow2_change_backing_file,
1519

    
1520
    .create_options = qcow_create_options,
1521
    .bdrv_check = qcow_check,
1522
};
1523

    
1524
static void bdrv_qcow2_init(void)
1525
{
1526
    bdrv_register(&bdrv_qcow2);
1527
}
1528

    
1529
block_init(bdrv_qcow2_init);