Statistics
| Branch: | Revision:

root / block / vdi.c @ 205ef796

History | View | Annotate | Download (29.2 kB)

1
/*
2
 * Block driver for the Virtual Disk Image (VDI) format
3
 *
4
 * Copyright (c) 2009 Stefan Weil
5
 *
6
 * This program is free software: you can redistribute it and/or modify
7
 * it under the terms of the GNU General Public License as published by
8
 * the Free Software Foundation, either version 2 of the License, or
9
 * (at your option) version 3 or any later version.
10
 *
11
 * This program is distributed in the hope that it will be useful,
12
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14
 * GNU General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU General Public License
17
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18
 *
19
 * Reference:
20
 * http://forums.virtualbox.org/viewtopic.php?t=8046
21
 *
22
 * This driver supports create / read / write operations on VDI images.
23
 *
24
 * Todo (see also TODO in code):
25
 *
26
 * Some features like snapshots are still missing.
27
 *
28
 * Deallocation of zero-filled blocks and shrinking images are missing, too
29
 * (might be added to common block layer).
30
 *
31
 * Allocation of blocks could be optimized (less writes to block map and
32
 * header).
33
 *
34
 * Read and write of adjacents blocks could be done in one operation
35
 * (current code uses one operation per block (1 MiB).
36
 *
37
 * The code is not thread safe (missing locks for changes in header and
38
 * block table, no problem with current QEMU).
39
 *
40
 * Hints:
41
 *
42
 * Blocks (VDI documentation) correspond to clusters (QEMU).
43
 * QEMU's backing files could be implemented using VDI snapshot files (TODO).
44
 * VDI snapshot files may also contain the complete machine state.
45
 * Maybe this machine state can be converted to QEMU PC machine snapshot data.
46
 *
47
 * The driver keeps a block cache (little endian entries) in memory.
48
 * For the standard block size (1 MiB), a 1 TiB disk will use 4 MiB RAM,
49
 * so this seems to be reasonable.
50
 */
51

    
52
#include "qemu-common.h"
53
#include "block_int.h"
54
#include "module.h"
55

    
56
#if defined(CONFIG_UUID)
57
#include <uuid/uuid.h>
58
#else
59
/* TODO: move uuid emulation to some central place in QEMU. */
60
#include "sysemu.h"     /* UUID_FMT */
61
typedef unsigned char uuid_t[16];
62
void uuid_generate(uuid_t out);
63
int uuid_is_null(const uuid_t uu);
64
void uuid_unparse(const uuid_t uu, char *out);
65
#endif
66

    
67
/* Code configuration options. */
68

    
69
/* Enable debug messages. */
70
//~ #define CONFIG_VDI_DEBUG
71

    
72
/* Support write operations on VDI images. */
73
#define CONFIG_VDI_WRITE
74

    
75
/* Support non-standard block (cluster) size. This is untested.
76
 * Maybe it will be needed for very large images.
77
 */
78
//~ #define CONFIG_VDI_BLOCK_SIZE
79

    
80
/* Support static (fixed, pre-allocated) images. */
81
#define CONFIG_VDI_STATIC_IMAGE
82

    
83
/* Command line option for static images. */
84
#define BLOCK_OPT_STATIC "static"
85

    
86
#define KiB     1024
87
#define MiB     (KiB * KiB)
88

    
89
#define SECTOR_SIZE 512
90

    
91
#if defined(CONFIG_VDI_DEBUG)
92
#define logout(fmt, ...) \
93
                fprintf(stderr, "vdi\t%-24s" fmt, __func__, ##__VA_ARGS__)
94
#else
95
#define logout(fmt, ...) ((void)0)
96
#endif
97

    
98
/* Image signature. */
99
#define VDI_SIGNATURE 0xbeda107f
100

    
101
/* Image version. */
102
#define VDI_VERSION_1_1 0x00010001
103

    
104
/* Image type. */
105
#define VDI_TYPE_DYNAMIC 1
106
#define VDI_TYPE_STATIC  2
107

    
108
/* Innotek / SUN images use these strings in header.text:
109
 * "<<< innotek VirtualBox Disk Image >>>\n"
110
 * "<<< Sun xVM VirtualBox Disk Image >>>\n"
111
 * "<<< Sun VirtualBox Disk Image >>>\n"
112
 * The value does not matter, so QEMU created images use a different text.
113
 */
114
#define VDI_TEXT "<<< QEMU VM Virtual Disk Image >>>\n"
115

    
116
/* Unallocated blocks use this index (no need to convert endianess). */
117
#define VDI_UNALLOCATED UINT32_MAX
118

    
119
#if !defined(CONFIG_UUID)
120
void uuid_generate(uuid_t out)
121
{
122
    memset(out, 0, sizeof(out));
123
}
124

    
125
int uuid_is_null(const uuid_t uu)
126
{
127
    uuid_t null_uuid = { 0 };
128
    return memcmp(uu, null_uuid, sizeof(uu)) == 0;
129
}
130

    
131
void uuid_unparse(const uuid_t uu, char *out)
132
{
133
    snprintf(out, 37, UUID_FMT,
134
            uu[0], uu[1], uu[2], uu[3], uu[4], uu[5], uu[6], uu[7],
135
            uu[8], uu[9], uu[10], uu[11], uu[12], uu[13], uu[14], uu[15]);
136
}
137
#endif
138

    
139
typedef struct {
140
    BlockDriverAIOCB common;
141
    int64_t sector_num;
142
    QEMUIOVector *qiov;
143
    uint8_t *buf;
144
    /* Total number of sectors. */
145
    int nb_sectors;
146
    /* Number of sectors for current AIO. */
147
    int n_sectors;
148
    /* New allocated block map entry. */
149
    uint32_t bmap_first;
150
    uint32_t bmap_last;
151
    /* Buffer for new allocated block. */
152
    void *block_buffer;
153
    void *orig_buf;
154
    int header_modified;
155
    BlockDriverAIOCB *hd_aiocb;
156
    struct iovec hd_iov;
157
    QEMUIOVector hd_qiov;
158
    QEMUBH *bh;
159
} VdiAIOCB;
160

    
161
typedef struct {
162
    char text[0x40];
163
    uint32_t signature;
164
    uint32_t version;
165
    uint32_t header_size;
166
    uint32_t image_type;
167
    uint32_t image_flags;
168
    char description[256];
169
    uint32_t offset_bmap;
170
    uint32_t offset_data;
171
    uint32_t cylinders;         /* disk geometry, unused here */
172
    uint32_t heads;             /* disk geometry, unused here */
173
    uint32_t sectors;           /* disk geometry, unused here */
174
    uint32_t sector_size;
175
    uint32_t unused1;
176
    uint64_t disk_size;
177
    uint32_t block_size;
178
    uint32_t block_extra;       /* unused here */
179
    uint32_t blocks_in_image;
180
    uint32_t blocks_allocated;
181
    uuid_t uuid_image;
182
    uuid_t uuid_last_snap;
183
    uuid_t uuid_link;
184
    uuid_t uuid_parent;
185
    uint64_t unused2[7];
186
} VdiHeader;
187

    
188
typedef struct {
189
    BlockDriverState *hd;
190
    /* The block map entries are little endian (even in memory). */
191
    uint32_t *bmap;
192
    /* Size of block (bytes). */
193
    uint32_t block_size;
194
    /* Size of block (sectors). */
195
    uint32_t block_sectors;
196
    /* First sector of block map. */
197
    uint32_t bmap_sector;
198
    /* VDI header (converted to host endianess). */
199
    VdiHeader header;
200
} BDRVVdiState;
201

    
202
/* Change UUID from little endian (IPRT = VirtualBox format) to big endian
203
 * format (network byte order, standard, see RFC 4122) and vice versa.
204
 */
205
static void uuid_convert(uuid_t uuid)
206
{
207
    bswap32s((uint32_t *)&uuid[0]);
208
    bswap16s((uint16_t *)&uuid[4]);
209
    bswap16s((uint16_t *)&uuid[6]);
210
}
211

    
212
static void vdi_header_to_cpu(VdiHeader *header)
213
{
214
    le32_to_cpus(&header->signature);
215
    le32_to_cpus(&header->version);
216
    le32_to_cpus(&header->header_size);
217
    le32_to_cpus(&header->image_type);
218
    le32_to_cpus(&header->image_flags);
219
    le32_to_cpus(&header->offset_bmap);
220
    le32_to_cpus(&header->offset_data);
221
    le32_to_cpus(&header->cylinders);
222
    le32_to_cpus(&header->heads);
223
    le32_to_cpus(&header->sectors);
224
    le32_to_cpus(&header->sector_size);
225
    le64_to_cpus(&header->disk_size);
226
    le32_to_cpus(&header->block_size);
227
    le32_to_cpus(&header->block_extra);
228
    le32_to_cpus(&header->blocks_in_image);
229
    le32_to_cpus(&header->blocks_allocated);
230
    uuid_convert(header->uuid_image);
231
    uuid_convert(header->uuid_last_snap);
232
    uuid_convert(header->uuid_link);
233
    uuid_convert(header->uuid_parent);
234
}
235

    
236
static void vdi_header_to_le(VdiHeader *header)
237
{
238
    cpu_to_le32s(&header->signature);
239
    cpu_to_le32s(&header->version);
240
    cpu_to_le32s(&header->header_size);
241
    cpu_to_le32s(&header->image_type);
242
    cpu_to_le32s(&header->image_flags);
243
    cpu_to_le32s(&header->offset_bmap);
244
    cpu_to_le32s(&header->offset_data);
245
    cpu_to_le32s(&header->cylinders);
246
    cpu_to_le32s(&header->heads);
247
    cpu_to_le32s(&header->sectors);
248
    cpu_to_le32s(&header->sector_size);
249
    cpu_to_le64s(&header->disk_size);
250
    cpu_to_le32s(&header->block_size);
251
    cpu_to_le32s(&header->block_extra);
252
    cpu_to_le32s(&header->blocks_in_image);
253
    cpu_to_le32s(&header->blocks_allocated);
254
    cpu_to_le32s(&header->blocks_allocated);
255
    uuid_convert(header->uuid_image);
256
    uuid_convert(header->uuid_last_snap);
257
    uuid_convert(header->uuid_link);
258
    uuid_convert(header->uuid_parent);
259
}
260

    
261
#if defined(CONFIG_VDI_DEBUG)
262
static void vdi_header_print(VdiHeader *header)
263
{
264
    char uuid[37];
265
    logout("text        %s", header->text);
266
    logout("signature   0x%04x\n", header->signature);
267
    logout("header size 0x%04x\n", header->header_size);
268
    logout("image type  0x%04x\n", header->image_type);
269
    logout("image flags 0x%04x\n", header->image_flags);
270
    logout("description %s\n", header->description);
271
    logout("offset bmap 0x%04x\n", header->offset_bmap);
272
    logout("offset data 0x%04x\n", header->offset_data);
273
    logout("cylinders   0x%04x\n", header->cylinders);
274
    logout("heads       0x%04x\n", header->heads);
275
    logout("sectors     0x%04x\n", header->sectors);
276
    logout("sector size 0x%04x\n", header->sector_size);
277
    logout("image size  0x%" PRIx64 " B (%" PRIu64 " MiB)\n",
278
           header->disk_size, header->disk_size / MiB);
279
    logout("block size  0x%04x\n", header->block_size);
280
    logout("block extra 0x%04x\n", header->block_extra);
281
    logout("blocks tot. 0x%04x\n", header->blocks_in_image);
282
    logout("blocks all. 0x%04x\n", header->blocks_allocated);
283
    uuid_unparse(header->uuid_image, uuid);
284
    logout("uuid image  %s\n", uuid);
285
    uuid_unparse(header->uuid_last_snap, uuid);
286
    logout("uuid snap   %s\n", uuid);
287
    uuid_unparse(header->uuid_link, uuid);
288
    logout("uuid link   %s\n", uuid);
289
    uuid_unparse(header->uuid_parent, uuid);
290
    logout("uuid parent %s\n", uuid);
291
}
292
#endif
293

    
294
static int vdi_check(BlockDriverState *bs, BdrvCheckResult *res)
295
{
296
    /* TODO: additional checks possible. */
297
    BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
298
    uint32_t blocks_allocated = 0;
299
    uint32_t block;
300
    uint32_t *bmap;
301
    logout("\n");
302

    
303
    bmap = qemu_malloc(s->header.blocks_in_image * sizeof(uint32_t));
304
    memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t));
305

    
306
    /* Check block map and value of blocks_allocated. */
307
    for (block = 0; block < s->header.blocks_in_image; block++) {
308
        uint32_t bmap_entry = le32_to_cpu(s->bmap[block]);
309
        if (bmap_entry != VDI_UNALLOCATED) {
310
            if (bmap_entry < s->header.blocks_in_image) {
311
                blocks_allocated++;
312
                if (bmap[bmap_entry] == VDI_UNALLOCATED) {
313
                    bmap[bmap_entry] = bmap_entry;
314
                } else {
315
                    fprintf(stderr, "ERROR: block index %" PRIu32
316
                            " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry);
317
                    res->corruptions++;
318
                }
319
            } else {
320
                fprintf(stderr, "ERROR: block index %" PRIu32
321
                        " too large, is %" PRIu32 "\n", block, bmap_entry);
322
                res->corruptions++;
323
            }
324
        }
325
    }
326
    if (blocks_allocated != s->header.blocks_allocated) {
327
        fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32
328
               ", should be %" PRIu32 "\n",
329
               blocks_allocated, s->header.blocks_allocated);
330
        res->corruptions++;
331
    }
332

    
333
    qemu_free(bmap);
334

    
335
    return 0;
336
}
337

    
338
static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
339
{
340
    /* TODO: vdi_get_info would be needed for machine snapshots.
341
       vm_state_offset is still missing. */
342
    BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
343
    logout("\n");
344
    bdi->cluster_size = s->block_size;
345
    bdi->vm_state_offset = 0;
346
    return 0;
347
}
348

    
349
static int vdi_make_empty(BlockDriverState *bs)
350
{
351
    /* TODO: missing code. */
352
    logout("\n");
353
    /* The return value for missing code must be 0, see block.c. */
354
    return 0;
355
}
356

    
357
static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename)
358
{
359
    const VdiHeader *header = (const VdiHeader *)buf;
360
    int result = 0;
361

    
362
    logout("\n");
363

    
364
    if (buf_size < sizeof(*header)) {
365
        /* Header too small, no VDI. */
366
    } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) {
367
        result = 100;
368
    }
369

    
370
    if (result == 0) {
371
        logout("no vdi image\n");
372
    } else {
373
        logout("%s", header->text);
374
    }
375

    
376
    return result;
377
}
378

    
379
static int vdi_open(BlockDriverState *bs, int flags)
380
{
381
    BDRVVdiState *s = bs->opaque;
382
    VdiHeader header;
383
    size_t bmap_size;
384

    
385
    logout("\n");
386

    
387
    if (bdrv_read(bs->file, 0, (uint8_t *)&header, 1) < 0) {
388
        goto fail;
389
    }
390

    
391
    vdi_header_to_cpu(&header);
392
#if defined(CONFIG_VDI_DEBUG)
393
    vdi_header_print(&header);
394
#endif
395

    
396
    if (header.disk_size % SECTOR_SIZE != 0) {
397
        /* 'VBoxManage convertfromraw' can create images with odd disk sizes.
398
           We accept them but round the disk size to the next multiple of
399
           SECTOR_SIZE. */
400
        logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size);
401
        header.disk_size += SECTOR_SIZE - 1;
402
        header.disk_size &= ~(SECTOR_SIZE - 1);
403
    }
404

    
405
    if (header.version != VDI_VERSION_1_1) {
406
        logout("unsupported version %u.%u\n",
407
               header.version >> 16, header.version & 0xffff);
408
        goto fail;
409
    } else if (header.offset_bmap % SECTOR_SIZE != 0) {
410
        /* We only support block maps which start on a sector boundary. */
411
        logout("unsupported block map offset 0x%x B\n", header.offset_bmap);
412
        goto fail;
413
    } else if (header.offset_data % SECTOR_SIZE != 0) {
414
        /* We only support data blocks which start on a sector boundary. */
415
        logout("unsupported data offset 0x%x B\n", header.offset_data);
416
        goto fail;
417
    } else if (header.sector_size != SECTOR_SIZE) {
418
        logout("unsupported sector size %u B\n", header.sector_size);
419
        goto fail;
420
    } else if (header.block_size != 1 * MiB) {
421
        logout("unsupported block size %u B\n", header.block_size);
422
        goto fail;
423
    } else if (header.disk_size >
424
               (uint64_t)header.blocks_in_image * header.block_size) {
425
        logout("unsupported disk size %" PRIu64 " B\n", header.disk_size);
426
        goto fail;
427
    } else if (!uuid_is_null(header.uuid_link)) {
428
        logout("link uuid != 0, unsupported\n");
429
        goto fail;
430
    } else if (!uuid_is_null(header.uuid_parent)) {
431
        logout("parent uuid != 0, unsupported\n");
432
        goto fail;
433
    }
434

    
435
    bs->total_sectors = header.disk_size / SECTOR_SIZE;
436

    
437
    s->block_size = header.block_size;
438
    s->block_sectors = header.block_size / SECTOR_SIZE;
439
    s->bmap_sector = header.offset_bmap / SECTOR_SIZE;
440
    s->header = header;
441

    
442
    bmap_size = header.blocks_in_image * sizeof(uint32_t);
443
    bmap_size = (bmap_size + SECTOR_SIZE - 1) / SECTOR_SIZE;
444
    if (bmap_size > 0) {
445
        s->bmap = qemu_malloc(bmap_size * SECTOR_SIZE);
446
    }
447
    if (bdrv_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap, bmap_size) < 0) {
448
        goto fail_free_bmap;
449
    }
450

    
451
    return 0;
452

    
453
 fail_free_bmap:
454
    qemu_free(s->bmap);
455

    
456
 fail:
457
    return -1;
458
}
459

    
460
static int vdi_is_allocated(BlockDriverState *bs, int64_t sector_num,
461
                             int nb_sectors, int *pnum)
462
{
463
    /* TODO: Check for too large sector_num (in bdrv_is_allocated or here). */
464
    BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
465
    size_t bmap_index = sector_num / s->block_sectors;
466
    size_t sector_in_block = sector_num % s->block_sectors;
467
    int n_sectors = s->block_sectors - sector_in_block;
468
    uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
469
    logout("%p, %" PRId64 ", %d, %p\n", bs, sector_num, nb_sectors, pnum);
470
    if (n_sectors > nb_sectors) {
471
        n_sectors = nb_sectors;
472
    }
473
    *pnum = n_sectors;
474
    return bmap_entry != VDI_UNALLOCATED;
475
}
476

    
477
static void vdi_aio_cancel(BlockDriverAIOCB *blockacb)
478
{
479
    /* TODO: This code is untested. How can I get it executed? */
480
    VdiAIOCB *acb = container_of(blockacb, VdiAIOCB, common);
481
    logout("\n");
482
    if (acb->hd_aiocb) {
483
        bdrv_aio_cancel(acb->hd_aiocb);
484
    }
485
    qemu_aio_release(acb);
486
}
487

    
488
static AIOPool vdi_aio_pool = {
489
    .aiocb_size = sizeof(VdiAIOCB),
490
    .cancel = vdi_aio_cancel,
491
};
492

    
493
static VdiAIOCB *vdi_aio_setup(BlockDriverState *bs, int64_t sector_num,
494
        QEMUIOVector *qiov, int nb_sectors,
495
        BlockDriverCompletionFunc *cb, void *opaque, int is_write)
496
{
497
    VdiAIOCB *acb;
498

    
499
    logout("%p, %" PRId64 ", %p, %d, %p, %p, %d\n",
500
           bs, sector_num, qiov, nb_sectors, cb, opaque, is_write);
501

    
502
    acb = qemu_aio_get(&vdi_aio_pool, bs, cb, opaque);
503
    if (acb) {
504
        acb->hd_aiocb = NULL;
505
        acb->sector_num = sector_num;
506
        acb->qiov = qiov;
507
        if (qiov->niov > 1) {
508
            acb->buf = qemu_blockalign(bs, qiov->size);
509
            acb->orig_buf = acb->buf;
510
            if (is_write) {
511
                qemu_iovec_to_buffer(qiov, acb->buf);
512
            }
513
        } else {
514
            acb->buf = (uint8_t *)qiov->iov->iov_base;
515
        }
516
        acb->nb_sectors = nb_sectors;
517
        acb->n_sectors = 0;
518
        acb->bmap_first = VDI_UNALLOCATED;
519
        acb->bmap_last = VDI_UNALLOCATED;
520
        acb->block_buffer = NULL;
521
        acb->header_modified = 0;
522
    }
523
    return acb;
524
}
525

    
526
static int vdi_schedule_bh(QEMUBHFunc *cb, VdiAIOCB *acb)
527
{
528
    logout("\n");
529

    
530
    if (acb->bh) {
531
        return -EIO;
532
    }
533

    
534
    acb->bh = qemu_bh_new(cb, acb);
535
    if (!acb->bh) {
536
        return -EIO;
537
    }
538

    
539
    qemu_bh_schedule(acb->bh);
540

    
541
    return 0;
542
}
543

    
544
static void vdi_aio_read_cb(void *opaque, int ret);
545

    
546
static void vdi_aio_read_bh(void *opaque)
547
{
548
    VdiAIOCB *acb = opaque;
549
    logout("\n");
550
    qemu_bh_delete(acb->bh);
551
    acb->bh = NULL;
552
    vdi_aio_read_cb(opaque, 0);
553
}
554

    
555
static void vdi_aio_read_cb(void *opaque, int ret)
556
{
557
    VdiAIOCB *acb = opaque;
558
    BlockDriverState *bs = acb->common.bs;
559
    BDRVVdiState *s = bs->opaque;
560
    uint32_t bmap_entry;
561
    uint32_t block_index;
562
    uint32_t sector_in_block;
563
    uint32_t n_sectors;
564

    
565
    logout("%u sectors read\n", acb->n_sectors);
566

    
567
    acb->hd_aiocb = NULL;
568

    
569
    if (ret < 0) {
570
        goto done;
571
    }
572

    
573
    acb->nb_sectors -= acb->n_sectors;
574

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

    
581
    acb->sector_num += acb->n_sectors;
582
    acb->buf += acb->n_sectors * SECTOR_SIZE;
583

    
584
    block_index = acb->sector_num / s->block_sectors;
585
    sector_in_block = acb->sector_num % s->block_sectors;
586
    n_sectors = s->block_sectors - sector_in_block;
587
    if (n_sectors > acb->nb_sectors) {
588
        n_sectors = acb->nb_sectors;
589
    }
590

    
591
    logout("will read %u sectors starting at sector %" PRIu64 "\n",
592
           n_sectors, acb->sector_num);
593

    
594
    /* prepare next AIO request */
595
    acb->n_sectors = n_sectors;
596
    bmap_entry = le32_to_cpu(s->bmap[block_index]);
597
    if (bmap_entry == VDI_UNALLOCATED) {
598
        /* Block not allocated, return zeros, no need to wait. */
599
        memset(acb->buf, 0, n_sectors * SECTOR_SIZE);
600
        ret = vdi_schedule_bh(vdi_aio_read_bh, acb);
601
        if (ret < 0) {
602
            goto done;
603
        }
604
    } else {
605
        uint64_t offset = s->header.offset_data / SECTOR_SIZE +
606
                          (uint64_t)bmap_entry * s->block_sectors +
607
                          sector_in_block;
608
        acb->hd_iov.iov_base = (void *)acb->buf;
609
        acb->hd_iov.iov_len = n_sectors * SECTOR_SIZE;
610
        qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
611
        acb->hd_aiocb = bdrv_aio_readv(bs->file, offset, &acb->hd_qiov,
612
                                       n_sectors, vdi_aio_read_cb, acb);
613
        if (acb->hd_aiocb == NULL) {
614
            goto done;
615
        }
616
    }
617
    return;
618
done:
619
    if (acb->qiov->niov > 1) {
620
        qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
621
        qemu_vfree(acb->orig_buf);
622
    }
623
    acb->common.cb(acb->common.opaque, ret);
624
    qemu_aio_release(acb);
625
}
626

    
627
static BlockDriverAIOCB *vdi_aio_readv(BlockDriverState *bs,
628
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
629
        BlockDriverCompletionFunc *cb, void *opaque)
630
{
631
    VdiAIOCB *acb;
632
    logout("\n");
633
    acb = vdi_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
634
    if (!acb) {
635
        return NULL;
636
    }
637
    vdi_aio_read_cb(acb, 0);
638
    return &acb->common;
639
}
640

    
641
static void vdi_aio_write_cb(void *opaque, int ret)
642
{
643
    VdiAIOCB *acb = opaque;
644
    BlockDriverState *bs = acb->common.bs;
645
    BDRVVdiState *s = bs->opaque;
646
    uint32_t bmap_entry;
647
    uint32_t block_index;
648
    uint32_t sector_in_block;
649
    uint32_t n_sectors;
650

    
651
    acb->hd_aiocb = NULL;
652

    
653
    if (ret < 0) {
654
        goto done;
655
    }
656

    
657
    acb->nb_sectors -= acb->n_sectors;
658
    acb->sector_num += acb->n_sectors;
659
    acb->buf += acb->n_sectors * SECTOR_SIZE;
660

    
661
    if (acb->nb_sectors == 0) {
662
        logout("finished data write\n");
663
        acb->n_sectors = 0;
664
        if (acb->header_modified) {
665
            VdiHeader *header = acb->block_buffer;
666
            logout("now writing modified header\n");
667
            assert(acb->bmap_first != VDI_UNALLOCATED);
668
            *header = s->header;
669
            vdi_header_to_le(header);
670
            acb->header_modified = 0;
671
            acb->hd_iov.iov_base = acb->block_buffer;
672
            acb->hd_iov.iov_len = SECTOR_SIZE;
673
            qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
674
            acb->hd_aiocb = bdrv_aio_writev(bs->file, 0, &acb->hd_qiov, 1,
675
                                            vdi_aio_write_cb, acb);
676
            if (acb->hd_aiocb == NULL) {
677
                goto done;
678
            }
679
            return;
680
        } else if (acb->bmap_first != VDI_UNALLOCATED) {
681
            /* One or more new blocks were allocated. */
682
            uint64_t offset;
683
            uint32_t bmap_first;
684
            uint32_t bmap_last;
685
            qemu_free(acb->block_buffer);
686
            acb->block_buffer = NULL;
687
            bmap_first = acb->bmap_first;
688
            bmap_last = acb->bmap_last;
689
            logout("now writing modified block map entry %u...%u\n",
690
                   bmap_first, bmap_last);
691
            /* Write modified sectors from block map. */
692
            bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
693
            bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
694
            n_sectors = bmap_last - bmap_first + 1;
695
            offset = s->bmap_sector + bmap_first;
696
            acb->bmap_first = VDI_UNALLOCATED;
697
            acb->hd_iov.iov_base = (void *)((uint8_t *)&s->bmap[0] +
698
                                            bmap_first * SECTOR_SIZE);
699
            acb->hd_iov.iov_len = n_sectors * SECTOR_SIZE;
700
            qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
701
            logout("will write %u block map sectors starting from entry %u\n",
702
                   n_sectors, bmap_first);
703
            acb->hd_aiocb = bdrv_aio_writev(bs->file, offset, &acb->hd_qiov,
704
                                            n_sectors, vdi_aio_write_cb, acb);
705
            if (acb->hd_aiocb == NULL) {
706
                goto done;
707
            }
708
            return;
709
        }
710
        ret = 0;
711
        goto done;
712
    }
713

    
714
    logout("%u sectors written\n", acb->n_sectors);
715

    
716
    block_index = acb->sector_num / s->block_sectors;
717
    sector_in_block = acb->sector_num % s->block_sectors;
718
    n_sectors = s->block_sectors - sector_in_block;
719
    if (n_sectors > acb->nb_sectors) {
720
        n_sectors = acb->nb_sectors;
721
    }
722

    
723
    logout("will write %u sectors starting at sector %" PRIu64 "\n",
724
           n_sectors, acb->sector_num);
725

    
726
    /* prepare next AIO request */
727
    acb->n_sectors = n_sectors;
728
    bmap_entry = le32_to_cpu(s->bmap[block_index]);
729
    if (bmap_entry == VDI_UNALLOCATED) {
730
        /* Allocate new block and write to it. */
731
        uint64_t offset;
732
        uint8_t *block;
733
        bmap_entry = s->header.blocks_allocated;
734
        s->bmap[block_index] = cpu_to_le32(bmap_entry);
735
        s->header.blocks_allocated++;
736
        offset = s->header.offset_data / SECTOR_SIZE +
737
                 (uint64_t)bmap_entry * s->block_sectors;
738
        block = acb->block_buffer;
739
        if (block == NULL) {
740
            block = qemu_mallocz(s->block_size);
741
            acb->block_buffer = block;
742
            acb->bmap_first = block_index;
743
            assert(!acb->header_modified);
744
            acb->header_modified = 1;
745
        }
746
        acb->bmap_last = block_index;
747
        memcpy(block + sector_in_block * SECTOR_SIZE,
748
               acb->buf, n_sectors * SECTOR_SIZE);
749
        acb->hd_iov.iov_base = (void *)block;
750
        acb->hd_iov.iov_len = s->block_size;
751
        qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
752
        acb->hd_aiocb = bdrv_aio_writev(bs->file, offset,
753
                                        &acb->hd_qiov, s->block_sectors,
754
                                        vdi_aio_write_cb, acb);
755
        if (acb->hd_aiocb == NULL) {
756
            goto done;
757
        }
758
    } else {
759
        uint64_t offset = s->header.offset_data / SECTOR_SIZE +
760
                          (uint64_t)bmap_entry * s->block_sectors +
761
                          sector_in_block;
762
        acb->hd_iov.iov_base = (void *)acb->buf;
763
        acb->hd_iov.iov_len = n_sectors * SECTOR_SIZE;
764
        qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
765
        acb->hd_aiocb = bdrv_aio_writev(bs->file, offset, &acb->hd_qiov,
766
                                        n_sectors, vdi_aio_write_cb, acb);
767
        if (acb->hd_aiocb == NULL) {
768
            goto done;
769
        }
770
    }
771

    
772
    return;
773

    
774
done:
775
    if (acb->qiov->niov > 1) {
776
        qemu_vfree(acb->orig_buf);
777
    }
778
    acb->common.cb(acb->common.opaque, ret);
779
    qemu_aio_release(acb);
780
}
781

    
782
static BlockDriverAIOCB *vdi_aio_writev(BlockDriverState *bs,
783
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
784
        BlockDriverCompletionFunc *cb, void *opaque)
785
{
786
    VdiAIOCB *acb;
787
    logout("\n");
788
    acb = vdi_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
789
    if (!acb) {
790
        return NULL;
791
    }
792
    vdi_aio_write_cb(acb, 0);
793
    return &acb->common;
794
}
795

    
796
static int vdi_create(const char *filename, QEMUOptionParameter *options)
797
{
798
    int fd;
799
    int result = 0;
800
    uint64_t bytes = 0;
801
    uint32_t blocks;
802
    size_t block_size = 1 * MiB;
803
    uint32_t image_type = VDI_TYPE_DYNAMIC;
804
    VdiHeader header;
805
    size_t i;
806
    size_t bmap_size;
807
    uint32_t *bmap;
808

    
809
    logout("\n");
810

    
811
    /* Read out options. */
812
    while (options && options->name) {
813
        if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
814
            bytes = options->value.n;
815
#if defined(CONFIG_VDI_BLOCK_SIZE)
816
        } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
817
            if (options->value.n) {
818
                /* TODO: Additional checks (SECTOR_SIZE * 2^n, ...). */
819
                block_size = options->value.n;
820
            }
821
#endif
822
#if defined(CONFIG_VDI_STATIC_IMAGE)
823
        } else if (!strcmp(options->name, BLOCK_OPT_STATIC)) {
824
            if (options->value.n) {
825
                image_type = VDI_TYPE_STATIC;
826
            }
827
#endif
828
        }
829
        options++;
830
    }
831

    
832
    fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
833
              0644);
834
    if (fd < 0) {
835
        return -errno;
836
    }
837

    
838
    /* We need enough blocks to store the given disk size,
839
       so always round up. */
840
    blocks = (bytes + block_size - 1) / block_size;
841

    
842
    bmap_size = blocks * sizeof(uint32_t);
843
    bmap_size = ((bmap_size + SECTOR_SIZE - 1) & ~(SECTOR_SIZE -1));
844

    
845
    memset(&header, 0, sizeof(header));
846
    pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
847
    header.signature = VDI_SIGNATURE;
848
    header.version = VDI_VERSION_1_1;
849
    header.header_size = 0x180;
850
    header.image_type = image_type;
851
    header.offset_bmap = 0x200;
852
    header.offset_data = 0x200 + bmap_size;
853
    header.sector_size = SECTOR_SIZE;
854
    header.disk_size = bytes;
855
    header.block_size = block_size;
856
    header.blocks_in_image = blocks;
857
    if (image_type == VDI_TYPE_STATIC) {
858
        header.blocks_allocated = blocks;
859
    }
860
    uuid_generate(header.uuid_image);
861
    uuid_generate(header.uuid_last_snap);
862
    /* There is no need to set header.uuid_link or header.uuid_parent here. */
863
#if defined(CONFIG_VDI_DEBUG)
864
    vdi_header_print(&header);
865
#endif
866
    vdi_header_to_le(&header);
867
    if (write(fd, &header, sizeof(header)) < 0) {
868
        result = -errno;
869
    }
870

    
871
    bmap = NULL;
872
    if (bmap_size > 0) {
873
        bmap = (uint32_t *)qemu_mallocz(bmap_size);
874
    }
875
    for (i = 0; i < blocks; i++) {
876
        if (image_type == VDI_TYPE_STATIC) {
877
            bmap[i] = i;
878
        } else {
879
            bmap[i] = VDI_UNALLOCATED;
880
        }
881
    }
882
    if (write(fd, bmap, bmap_size) < 0) {
883
        result = -errno;
884
    }
885
    qemu_free(bmap);
886
    if (image_type == VDI_TYPE_STATIC) {
887
        if (ftruncate(fd, sizeof(header) + bmap_size + blocks * block_size)) {
888
            result = -errno;
889
        }
890
    }
891

    
892
    if (close(fd) < 0) {
893
        result = -errno;
894
    }
895

    
896
    return result;
897
}
898

    
899
static void vdi_close(BlockDriverState *bs)
900
{
901
}
902

    
903
static int vdi_flush(BlockDriverState *bs)
904
{
905
    logout("\n");
906
    return bdrv_flush(bs->file);
907
}
908

    
909

    
910
static QEMUOptionParameter vdi_create_options[] = {
911
    {
912
        .name = BLOCK_OPT_SIZE,
913
        .type = OPT_SIZE,
914
        .help = "Virtual disk size"
915
    },
916
#if defined(CONFIG_VDI_BLOCK_SIZE)
917
    {
918
        .name = BLOCK_OPT_CLUSTER_SIZE,
919
        .type = OPT_SIZE,
920
        .help = "VDI cluster (block) size"
921
    },
922
#endif
923
#if defined(CONFIG_VDI_STATIC_IMAGE)
924
    {
925
        .name = BLOCK_OPT_STATIC,
926
        .type = OPT_FLAG,
927
        .help = "VDI static (pre-allocated) image"
928
    },
929
#endif
930
    /* TODO: An additional option to set UUID values might be useful. */
931
    { NULL }
932
};
933

    
934
static BlockDriver bdrv_vdi = {
935
    .format_name = "vdi",
936
    .instance_size = sizeof(BDRVVdiState),
937
    .bdrv_probe = vdi_probe,
938
    .bdrv_open = vdi_open,
939
    .bdrv_close = vdi_close,
940
    .bdrv_create = vdi_create,
941
    .bdrv_flush = vdi_flush,
942
    .bdrv_is_allocated = vdi_is_allocated,
943
    .bdrv_make_empty = vdi_make_empty,
944

    
945
    .bdrv_aio_readv = vdi_aio_readv,
946
#if defined(CONFIG_VDI_WRITE)
947
    .bdrv_aio_writev = vdi_aio_writev,
948
#endif
949

    
950
    .bdrv_get_info = vdi_get_info,
951

    
952
    .create_options = vdi_create_options,
953
    .bdrv_check = vdi_check,
954
};
955

    
956
static void bdrv_vdi_init(void)
957
{
958
    logout("\n");
959
    bdrv_register(&bdrv_vdi);
960
}
961

    
962
block_init(bdrv_vdi_init);