Statistics
| Branch: | Revision:

root / block.c @ 67f0875e

History | View | Annotate | Download (50.3 kB)

1
/*
2
 * QEMU System Emulator block driver
3
 *
4
 * Copyright (c) 2003 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 "config-host.h"
25
#ifdef CONFIG_BSD
26
/* include native header before sys-queue.h */
27
#include <sys/queue.h>
28
#endif
29

    
30
#include "qemu-common.h"
31
#include "monitor.h"
32
#include "block_int.h"
33
#include "module.h"
34

    
35
#ifdef CONFIG_BSD
36
#include <sys/types.h>
37
#include <sys/stat.h>
38
#include <sys/ioctl.h>
39
#ifndef __DragonFly__
40
#include <sys/disk.h>
41
#endif
42
#endif
43

    
44
#ifdef _WIN32
45
#include <windows.h>
46
#endif
47

    
48
#define SECTOR_BITS 9
49
#define SECTOR_SIZE (1 << SECTOR_BITS)
50

    
51
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
52
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
53
        BlockDriverCompletionFunc *cb, void *opaque);
54
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
55
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
56
        BlockDriverCompletionFunc *cb, void *opaque);
57
static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
58
        BlockDriverCompletionFunc *cb, void *opaque);
59
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
60
                        uint8_t *buf, int nb_sectors);
61
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
62
                         const uint8_t *buf, int nb_sectors);
63

    
64
BlockDriverState *bdrv_first;
65

    
66
static BlockDriver *first_drv;
67

    
68
int path_is_absolute(const char *path)
69
{
70
    const char *p;
71
#ifdef _WIN32
72
    /* specific case for names like: "\\.\d:" */
73
    if (*path == '/' || *path == '\\')
74
        return 1;
75
#endif
76
    p = strchr(path, ':');
77
    if (p)
78
        p++;
79
    else
80
        p = path;
81
#ifdef _WIN32
82
    return (*p == '/' || *p == '\\');
83
#else
84
    return (*p == '/');
85
#endif
86
}
87

    
88
/* if filename is absolute, just copy it to dest. Otherwise, build a
89
   path to it by considering it is relative to base_path. URL are
90
   supported. */
91
void path_combine(char *dest, int dest_size,
92
                  const char *base_path,
93
                  const char *filename)
94
{
95
    const char *p, *p1;
96
    int len;
97

    
98
    if (dest_size <= 0)
99
        return;
100
    if (path_is_absolute(filename)) {
101
        pstrcpy(dest, dest_size, filename);
102
    } else {
103
        p = strchr(base_path, ':');
104
        if (p)
105
            p++;
106
        else
107
            p = base_path;
108
        p1 = strrchr(base_path, '/');
109
#ifdef _WIN32
110
        {
111
            const char *p2;
112
            p2 = strrchr(base_path, '\\');
113
            if (!p1 || p2 > p1)
114
                p1 = p2;
115
        }
116
#endif
117
        if (p1)
118
            p1++;
119
        else
120
            p1 = base_path;
121
        if (p1 > p)
122
            p = p1;
123
        len = p - base_path;
124
        if (len > dest_size - 1)
125
            len = dest_size - 1;
126
        memcpy(dest, base_path, len);
127
        dest[len] = '\0';
128
        pstrcat(dest, dest_size, filename);
129
    }
130
}
131

    
132
void bdrv_register(BlockDriver *bdrv)
133
{
134
    if (!bdrv->bdrv_aio_readv) {
135
        /* add AIO emulation layer */
136
        bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
137
        bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
138
    } else if (!bdrv->bdrv_read) {
139
        /* add synchronous IO emulation layer */
140
        bdrv->bdrv_read = bdrv_read_em;
141
        bdrv->bdrv_write = bdrv_write_em;
142
    }
143

    
144
    if (!bdrv->bdrv_aio_flush)
145
        bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
146

    
147
    bdrv->next = first_drv;
148
    first_drv = bdrv;
149
}
150

    
151
/* create a new block device (by default it is empty) */
152
BlockDriverState *bdrv_new(const char *device_name)
153
{
154
    BlockDriverState **pbs, *bs;
155

    
156
    bs = qemu_mallocz(sizeof(BlockDriverState));
157
    pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
158
    if (device_name[0] != '\0') {
159
        /* insert at the end */
160
        pbs = &bdrv_first;
161
        while (*pbs != NULL)
162
            pbs = &(*pbs)->next;
163
        *pbs = bs;
164
    }
165
    return bs;
166
}
167

    
168
BlockDriver *bdrv_find_format(const char *format_name)
169
{
170
    BlockDriver *drv1;
171
    for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
172
        if (!strcmp(drv1->format_name, format_name))
173
            return drv1;
174
    }
175
    return NULL;
176
}
177

    
178
int bdrv_create(BlockDriver *drv, const char* filename,
179
    QEMUOptionParameter *options)
180
{
181
    if (!drv->bdrv_create)
182
        return -ENOTSUP;
183

    
184
    return drv->bdrv_create(filename, options);
185
}
186

    
187
#ifdef _WIN32
188
void get_tmp_filename(char *filename, int size)
189
{
190
    char temp_dir[MAX_PATH];
191

    
192
    GetTempPath(MAX_PATH, temp_dir);
193
    GetTempFileName(temp_dir, "qem", 0, filename);
194
}
195
#else
196
void get_tmp_filename(char *filename, int size)
197
{
198
    int fd;
199
    const char *tmpdir;
200
    /* XXX: race condition possible */
201
    tmpdir = getenv("TMPDIR");
202
    if (!tmpdir)
203
        tmpdir = "/tmp";
204
    snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
205
    fd = mkstemp(filename);
206
    close(fd);
207
}
208
#endif
209

    
210
#ifdef _WIN32
211
static int is_windows_drive_prefix(const char *filename)
212
{
213
    return (((filename[0] >= 'a' && filename[0] <= 'z') ||
214
             (filename[0] >= 'A' && filename[0] <= 'Z')) &&
215
            filename[1] == ':');
216
}
217

    
218
int is_windows_drive(const char *filename)
219
{
220
    if (is_windows_drive_prefix(filename) &&
221
        filename[2] == '\0')
222
        return 1;
223
    if (strstart(filename, "\\\\.\\", NULL) ||
224
        strstart(filename, "//./", NULL))
225
        return 1;
226
    return 0;
227
}
228
#endif
229

    
230
static BlockDriver *find_protocol(const char *filename)
231
{
232
    BlockDriver *drv1;
233
    char protocol[128];
234
    int len;
235
    const char *p;
236

    
237
#ifdef _WIN32
238
    if (is_windows_drive(filename) ||
239
        is_windows_drive_prefix(filename))
240
        return bdrv_find_format("raw");
241
#endif
242
    p = strchr(filename, ':');
243
    if (!p)
244
        return bdrv_find_format("raw");
245
    len = p - filename;
246
    if (len > sizeof(protocol) - 1)
247
        len = sizeof(protocol) - 1;
248
    memcpy(protocol, filename, len);
249
    protocol[len] = '\0';
250
    for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
251
        if (drv1->protocol_name &&
252
            !strcmp(drv1->protocol_name, protocol))
253
            return drv1;
254
    }
255
    return NULL;
256
}
257

    
258
/*
259
 * Detect host devices. By convention, /dev/cdrom[N] is always
260
 * recognized as a host CDROM.
261
 */
262
static BlockDriver *find_hdev_driver(const char *filename)
263
{
264
    int score_max = 0, score;
265
    BlockDriver *drv = NULL, *d;
266

    
267
    for (d = first_drv; d; d = d->next) {
268
        if (d->bdrv_probe_device) {
269
            score = d->bdrv_probe_device(filename);
270
            if (score > score_max) {
271
                score_max = score;
272
                drv = d;
273
            }
274
        }
275
    }
276

    
277
    return drv;
278
}
279

    
280
static BlockDriver *find_image_format(const char *filename)
281
{
282
    int ret, score, score_max;
283
    BlockDriver *drv1, *drv;
284
    uint8_t buf[2048];
285
    BlockDriverState *bs;
286

    
287
    drv = find_protocol(filename);
288
    /* no need to test disk image formats for vvfat */
289
    if (drv && strcmp(drv->format_name, "vvfat") == 0)
290
        return drv;
291

    
292
    ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
293
    if (ret < 0)
294
        return NULL;
295
    ret = bdrv_pread(bs, 0, buf, sizeof(buf));
296
    bdrv_delete(bs);
297
    if (ret < 0) {
298
        return NULL;
299
    }
300

    
301
    score_max = 0;
302
    for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
303
        if (drv1->bdrv_probe) {
304
            score = drv1->bdrv_probe(buf, ret, filename);
305
            if (score > score_max) {
306
                score_max = score;
307
                drv = drv1;
308
            }
309
        }
310
    }
311
    return drv;
312
}
313

    
314
int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
315
{
316
    BlockDriverState *bs;
317
    int ret;
318

    
319
    bs = bdrv_new("");
320
    ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
321
    if (ret < 0) {
322
        bdrv_delete(bs);
323
        return ret;
324
    }
325
    bs->growable = 1;
326
    *pbs = bs;
327
    return 0;
328
}
329

    
330
int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
331
{
332
    return bdrv_open2(bs, filename, flags, NULL);
333
}
334

    
335
int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
336
               BlockDriver *drv)
337
{
338
    int ret, open_flags;
339
    char tmp_filename[PATH_MAX];
340
    char backing_filename[PATH_MAX];
341

    
342
    bs->read_only = 0;
343
    bs->is_temporary = 0;
344
    bs->encrypted = 0;
345
    bs->valid_key = 0;
346
    /* buffer_alignment defaulted to 512, drivers can change this value */
347
    bs->buffer_alignment = 512;
348

    
349
    if (flags & BDRV_O_SNAPSHOT) {
350
        BlockDriverState *bs1;
351
        int64_t total_size;
352
        int is_protocol = 0;
353
        BlockDriver *bdrv_qcow2;
354
        QEMUOptionParameter *options;
355

    
356
        /* if snapshot, we create a temporary backing file and open it
357
           instead of opening 'filename' directly */
358

    
359
        /* if there is a backing file, use it */
360
        bs1 = bdrv_new("");
361
        ret = bdrv_open2(bs1, filename, 0, drv);
362
        if (ret < 0) {
363
            bdrv_delete(bs1);
364
            return ret;
365
        }
366
        total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
367

    
368
        if (bs1->drv && bs1->drv->protocol_name)
369
            is_protocol = 1;
370

    
371
        bdrv_delete(bs1);
372

    
373
        get_tmp_filename(tmp_filename, sizeof(tmp_filename));
374

    
375
        /* Real path is meaningless for protocols */
376
        if (is_protocol)
377
            snprintf(backing_filename, sizeof(backing_filename),
378
                     "%s", filename);
379
        else
380
            realpath(filename, backing_filename);
381

    
382
        bdrv_qcow2 = bdrv_find_format("qcow2");
383
        options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
384

    
385
        set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size * 512);
386
        set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
387
        if (drv) {
388
            set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
389
                drv->format_name);
390
        }
391

    
392
        ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
393
        if (ret < 0) {
394
            return ret;
395
        }
396

    
397
        filename = tmp_filename;
398
        drv = bdrv_qcow2;
399
        bs->is_temporary = 1;
400
    }
401

    
402
    pstrcpy(bs->filename, sizeof(bs->filename), filename);
403
    if (flags & BDRV_O_FILE) {
404
        drv = find_protocol(filename);
405
    } else if (!drv) {
406
        drv = find_hdev_driver(filename);
407
        if (!drv) {
408
            drv = find_image_format(filename);
409
        }
410
    }
411
    if (!drv) {
412
        ret = -ENOENT;
413
        goto unlink_and_fail;
414
    }
415
    bs->drv = drv;
416
    bs->opaque = qemu_mallocz(drv->instance_size);
417

    
418
    /*
419
     * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
420
     * write cache to the guest.  We do need the fdatasync to flush
421
     * out transactions for block allocations, and we maybe have a
422
     * volatile write cache in our backing device to deal with.
423
     */
424
    if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
425
        bs->enable_write_cache = 1;
426

    
427
    /* Note: for compatibility, we open disk image files as RDWR, and
428
       RDONLY as fallback */
429
    if (!(flags & BDRV_O_FILE))
430
        open_flags = BDRV_O_RDWR |
431
                (flags & (BDRV_O_CACHE_MASK|BDRV_O_NATIVE_AIO));
432
    else
433
        open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
434
    ret = drv->bdrv_open(bs, filename, open_flags);
435
    if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
436
        ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
437
        bs->read_only = 1;
438
    }
439
    if (ret < 0) {
440
        qemu_free(bs->opaque);
441
        bs->opaque = NULL;
442
        bs->drv = NULL;
443
    unlink_and_fail:
444
        if (bs->is_temporary)
445
            unlink(filename);
446
        return ret;
447
    }
448
    if (drv->bdrv_getlength) {
449
        bs->total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
450
    }
451
#ifndef _WIN32
452
    if (bs->is_temporary) {
453
        unlink(filename);
454
    }
455
#endif
456
    if (bs->backing_file[0] != '\0') {
457
        /* if there is a backing file, use it */
458
        BlockDriver *back_drv = NULL;
459
        bs->backing_hd = bdrv_new("");
460
        path_combine(backing_filename, sizeof(backing_filename),
461
                     filename, bs->backing_file);
462
        if (bs->backing_format[0] != '\0')
463
            back_drv = bdrv_find_format(bs->backing_format);
464
        ret = bdrv_open2(bs->backing_hd, backing_filename, open_flags,
465
                         back_drv);
466
        if (ret < 0) {
467
            bdrv_close(bs);
468
            return ret;
469
        }
470
    }
471

    
472
    if (!bdrv_key_required(bs)) {
473
        /* call the change callback */
474
        bs->media_changed = 1;
475
        if (bs->change_cb)
476
            bs->change_cb(bs->change_opaque);
477
    }
478
    return 0;
479
}
480

    
481
void bdrv_close(BlockDriverState *bs)
482
{
483
    if (bs->drv) {
484
        if (bs->backing_hd)
485
            bdrv_delete(bs->backing_hd);
486
        bs->drv->bdrv_close(bs);
487
        qemu_free(bs->opaque);
488
#ifdef _WIN32
489
        if (bs->is_temporary) {
490
            unlink(bs->filename);
491
        }
492
#endif
493
        bs->opaque = NULL;
494
        bs->drv = NULL;
495

    
496
        /* call the change callback */
497
        bs->media_changed = 1;
498
        if (bs->change_cb)
499
            bs->change_cb(bs->change_opaque);
500
    }
501
}
502

    
503
void bdrv_delete(BlockDriverState *bs)
504
{
505
    BlockDriverState **pbs;
506

    
507
    pbs = &bdrv_first;
508
    while (*pbs != bs && *pbs != NULL)
509
        pbs = &(*pbs)->next;
510
    if (*pbs == bs)
511
        *pbs = bs->next;
512

    
513
    bdrv_close(bs);
514
    qemu_free(bs);
515
}
516

    
517
/*
518
 * Run consistency checks on an image
519
 *
520
 * Returns the number of errors or -errno when an internal error occurs
521
 */
522
int bdrv_check(BlockDriverState *bs)
523
{
524
    if (bs->drv->bdrv_check == NULL) {
525
        return -ENOTSUP;
526
    }
527

    
528
    return bs->drv->bdrv_check(bs);
529
}
530

    
531
/* commit COW file into the raw image */
532
int bdrv_commit(BlockDriverState *bs)
533
{
534
    BlockDriver *drv = bs->drv;
535
    int64_t i, total_sectors;
536
    int n, j;
537
    unsigned char sector[512];
538

    
539
    if (!drv)
540
        return -ENOMEDIUM;
541

    
542
    if (bs->read_only) {
543
        return -EACCES;
544
    }
545

    
546
    if (!bs->backing_hd) {
547
        return -ENOTSUP;
548
    }
549

    
550
    total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
551
    for (i = 0; i < total_sectors;) {
552
        if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
553
            for(j = 0; j < n; j++) {
554
                if (bdrv_read(bs, i, sector, 1) != 0) {
555
                    return -EIO;
556
                }
557

    
558
                if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
559
                    return -EIO;
560
                }
561
                i++;
562
            }
563
        } else {
564
            i += n;
565
        }
566
    }
567

    
568
    if (drv->bdrv_make_empty)
569
        return drv->bdrv_make_empty(bs);
570

    
571
    return 0;
572
}
573

    
574
static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
575
                                   size_t size)
576
{
577
    int64_t len;
578

    
579
    if (!bdrv_is_inserted(bs))
580
        return -ENOMEDIUM;
581

    
582
    if (bs->growable)
583
        return 0;
584

    
585
    len = bdrv_getlength(bs);
586

    
587
    if (offset < 0)
588
        return -EIO;
589

    
590
    if ((offset > len) || (len - offset < size))
591
        return -EIO;
592

    
593
    return 0;
594
}
595

    
596
static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
597
                              int nb_sectors)
598
{
599
    return bdrv_check_byte_request(bs, sector_num * 512, nb_sectors * 512);
600
}
601

    
602
/* return < 0 if error. See bdrv_write() for the return codes */
603
int bdrv_read(BlockDriverState *bs, int64_t sector_num,
604
              uint8_t *buf, int nb_sectors)
605
{
606
    BlockDriver *drv = bs->drv;
607

    
608
    if (!drv)
609
        return -ENOMEDIUM;
610
    if (bdrv_check_request(bs, sector_num, nb_sectors))
611
        return -EIO;
612

    
613
    return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
614
}
615

    
616
/* Return < 0 if error. Important errors are:
617
  -EIO         generic I/O error (may happen for all errors)
618
  -ENOMEDIUM   No media inserted.
619
  -EINVAL      Invalid sector number or nb_sectors
620
  -EACCES      Trying to write a read-only device
621
*/
622
int bdrv_write(BlockDriverState *bs, int64_t sector_num,
623
               const uint8_t *buf, int nb_sectors)
624
{
625
    BlockDriver *drv = bs->drv;
626
    if (!bs->drv)
627
        return -ENOMEDIUM;
628
    if (bs->read_only)
629
        return -EACCES;
630
    if (bdrv_check_request(bs, sector_num, nb_sectors))
631
        return -EIO;
632

    
633
    return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
634
}
635

    
636
int bdrv_pread(BlockDriverState *bs, int64_t offset,
637
               void *buf, int count1)
638
{
639
    uint8_t tmp_buf[SECTOR_SIZE];
640
    int len, nb_sectors, count;
641
    int64_t sector_num;
642

    
643
    count = count1;
644
    /* first read to align to sector start */
645
    len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
646
    if (len > count)
647
        len = count;
648
    sector_num = offset >> SECTOR_BITS;
649
    if (len > 0) {
650
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
651
            return -EIO;
652
        memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
653
        count -= len;
654
        if (count == 0)
655
            return count1;
656
        sector_num++;
657
        buf += len;
658
    }
659

    
660
    /* read the sectors "in place" */
661
    nb_sectors = count >> SECTOR_BITS;
662
    if (nb_sectors > 0) {
663
        if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
664
            return -EIO;
665
        sector_num += nb_sectors;
666
        len = nb_sectors << SECTOR_BITS;
667
        buf += len;
668
        count -= len;
669
    }
670

    
671
    /* add data from the last sector */
672
    if (count > 0) {
673
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
674
            return -EIO;
675
        memcpy(buf, tmp_buf, count);
676
    }
677
    return count1;
678
}
679

    
680
int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
681
                const void *buf, int count1)
682
{
683
    uint8_t tmp_buf[SECTOR_SIZE];
684
    int len, nb_sectors, count;
685
    int64_t sector_num;
686

    
687
    count = count1;
688
    /* first write to align to sector start */
689
    len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
690
    if (len > count)
691
        len = count;
692
    sector_num = offset >> SECTOR_BITS;
693
    if (len > 0) {
694
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
695
            return -EIO;
696
        memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
697
        if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
698
            return -EIO;
699
        count -= len;
700
        if (count == 0)
701
            return count1;
702
        sector_num++;
703
        buf += len;
704
    }
705

    
706
    /* write the sectors "in place" */
707
    nb_sectors = count >> SECTOR_BITS;
708
    if (nb_sectors > 0) {
709
        if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
710
            return -EIO;
711
        sector_num += nb_sectors;
712
        len = nb_sectors << SECTOR_BITS;
713
        buf += len;
714
        count -= len;
715
    }
716

    
717
    /* add data from the last sector */
718
    if (count > 0) {
719
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
720
            return -EIO;
721
        memcpy(tmp_buf, buf, count);
722
        if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
723
            return -EIO;
724
    }
725
    return count1;
726
}
727

    
728
/**
729
 * Truncate file to 'offset' bytes (needed only for file protocols)
730
 */
731
int bdrv_truncate(BlockDriverState *bs, int64_t offset)
732
{
733
    BlockDriver *drv = bs->drv;
734
    if (!drv)
735
        return -ENOMEDIUM;
736
    if (!drv->bdrv_truncate)
737
        return -ENOTSUP;
738
    return drv->bdrv_truncate(bs, offset);
739
}
740

    
741
/**
742
 * Length of a file in bytes. Return < 0 if error or unknown.
743
 */
744
int64_t bdrv_getlength(BlockDriverState *bs)
745
{
746
    BlockDriver *drv = bs->drv;
747
    if (!drv)
748
        return -ENOMEDIUM;
749
    if (!drv->bdrv_getlength) {
750
        /* legacy mode */
751
        return bs->total_sectors * SECTOR_SIZE;
752
    }
753
    return drv->bdrv_getlength(bs);
754
}
755

    
756
/* return 0 as number of sectors if no device present or error */
757
void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
758
{
759
    int64_t length;
760
    length = bdrv_getlength(bs);
761
    if (length < 0)
762
        length = 0;
763
    else
764
        length = length >> SECTOR_BITS;
765
    *nb_sectors_ptr = length;
766
}
767

    
768
struct partition {
769
        uint8_t boot_ind;           /* 0x80 - active */
770
        uint8_t head;               /* starting head */
771
        uint8_t sector;             /* starting sector */
772
        uint8_t cyl;                /* starting cylinder */
773
        uint8_t sys_ind;            /* What partition type */
774
        uint8_t end_head;           /* end head */
775
        uint8_t end_sector;         /* end sector */
776
        uint8_t end_cyl;            /* end cylinder */
777
        uint32_t start_sect;        /* starting sector counting from 0 */
778
        uint32_t nr_sects;          /* nr of sectors in partition */
779
} __attribute__((packed));
780

    
781
/* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
782
static int guess_disk_lchs(BlockDriverState *bs,
783
                           int *pcylinders, int *pheads, int *psectors)
784
{
785
    uint8_t buf[512];
786
    int ret, i, heads, sectors, cylinders;
787
    struct partition *p;
788
    uint32_t nr_sects;
789
    uint64_t nb_sectors;
790

    
791
    bdrv_get_geometry(bs, &nb_sectors);
792

    
793
    ret = bdrv_read(bs, 0, buf, 1);
794
    if (ret < 0)
795
        return -1;
796
    /* test msdos magic */
797
    if (buf[510] != 0x55 || buf[511] != 0xaa)
798
        return -1;
799
    for(i = 0; i < 4; i++) {
800
        p = ((struct partition *)(buf + 0x1be)) + i;
801
        nr_sects = le32_to_cpu(p->nr_sects);
802
        if (nr_sects && p->end_head) {
803
            /* We make the assumption that the partition terminates on
804
               a cylinder boundary */
805
            heads = p->end_head + 1;
806
            sectors = p->end_sector & 63;
807
            if (sectors == 0)
808
                continue;
809
            cylinders = nb_sectors / (heads * sectors);
810
            if (cylinders < 1 || cylinders > 16383)
811
                continue;
812
            *pheads = heads;
813
            *psectors = sectors;
814
            *pcylinders = cylinders;
815
#if 0
816
            printf("guessed geometry: LCHS=%d %d %d\n",
817
                   cylinders, heads, sectors);
818
#endif
819
            return 0;
820
        }
821
    }
822
    return -1;
823
}
824

    
825
void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
826
{
827
    int translation, lba_detected = 0;
828
    int cylinders, heads, secs;
829
    uint64_t nb_sectors;
830

    
831
    /* if a geometry hint is available, use it */
832
    bdrv_get_geometry(bs, &nb_sectors);
833
    bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
834
    translation = bdrv_get_translation_hint(bs);
835
    if (cylinders != 0) {
836
        *pcyls = cylinders;
837
        *pheads = heads;
838
        *psecs = secs;
839
    } else {
840
        if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
841
            if (heads > 16) {
842
                /* if heads > 16, it means that a BIOS LBA
843
                   translation was active, so the default
844
                   hardware geometry is OK */
845
                lba_detected = 1;
846
                goto default_geometry;
847
            } else {
848
                *pcyls = cylinders;
849
                *pheads = heads;
850
                *psecs = secs;
851
                /* disable any translation to be in sync with
852
                   the logical geometry */
853
                if (translation == BIOS_ATA_TRANSLATION_AUTO) {
854
                    bdrv_set_translation_hint(bs,
855
                                              BIOS_ATA_TRANSLATION_NONE);
856
                }
857
            }
858
        } else {
859
        default_geometry:
860
            /* if no geometry, use a standard physical disk geometry */
861
            cylinders = nb_sectors / (16 * 63);
862

    
863
            if (cylinders > 16383)
864
                cylinders = 16383;
865
            else if (cylinders < 2)
866
                cylinders = 2;
867
            *pcyls = cylinders;
868
            *pheads = 16;
869
            *psecs = 63;
870
            if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
871
                if ((*pcyls * *pheads) <= 131072) {
872
                    bdrv_set_translation_hint(bs,
873
                                              BIOS_ATA_TRANSLATION_LARGE);
874
                } else {
875
                    bdrv_set_translation_hint(bs,
876
                                              BIOS_ATA_TRANSLATION_LBA);
877
                }
878
            }
879
        }
880
        bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
881
    }
882
}
883

    
884
void bdrv_set_geometry_hint(BlockDriverState *bs,
885
                            int cyls, int heads, int secs)
886
{
887
    bs->cyls = cyls;
888
    bs->heads = heads;
889
    bs->secs = secs;
890
}
891

    
892
void bdrv_set_type_hint(BlockDriverState *bs, int type)
893
{
894
    bs->type = type;
895
    bs->removable = ((type == BDRV_TYPE_CDROM ||
896
                      type == BDRV_TYPE_FLOPPY));
897
}
898

    
899
void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
900
{
901
    bs->translation = translation;
902
}
903

    
904
void bdrv_get_geometry_hint(BlockDriverState *bs,
905
                            int *pcyls, int *pheads, int *psecs)
906
{
907
    *pcyls = bs->cyls;
908
    *pheads = bs->heads;
909
    *psecs = bs->secs;
910
}
911

    
912
int bdrv_get_type_hint(BlockDriverState *bs)
913
{
914
    return bs->type;
915
}
916

    
917
int bdrv_get_translation_hint(BlockDriverState *bs)
918
{
919
    return bs->translation;
920
}
921

    
922
int bdrv_is_removable(BlockDriverState *bs)
923
{
924
    return bs->removable;
925
}
926

    
927
int bdrv_is_read_only(BlockDriverState *bs)
928
{
929
    return bs->read_only;
930
}
931

    
932
int bdrv_is_sg(BlockDriverState *bs)
933
{
934
    return bs->sg;
935
}
936

    
937
int bdrv_enable_write_cache(BlockDriverState *bs)
938
{
939
    return bs->enable_write_cache;
940
}
941

    
942
/* XXX: no longer used */
943
void bdrv_set_change_cb(BlockDriverState *bs,
944
                        void (*change_cb)(void *opaque), void *opaque)
945
{
946
    bs->change_cb = change_cb;
947
    bs->change_opaque = opaque;
948
}
949

    
950
int bdrv_is_encrypted(BlockDriverState *bs)
951
{
952
    if (bs->backing_hd && bs->backing_hd->encrypted)
953
        return 1;
954
    return bs->encrypted;
955
}
956

    
957
int bdrv_key_required(BlockDriverState *bs)
958
{
959
    BlockDriverState *backing_hd = bs->backing_hd;
960

    
961
    if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
962
        return 1;
963
    return (bs->encrypted && !bs->valid_key);
964
}
965

    
966
int bdrv_set_key(BlockDriverState *bs, const char *key)
967
{
968
    int ret;
969
    if (bs->backing_hd && bs->backing_hd->encrypted) {
970
        ret = bdrv_set_key(bs->backing_hd, key);
971
        if (ret < 0)
972
            return ret;
973
        if (!bs->encrypted)
974
            return 0;
975
    }
976
    if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
977
        return -1;
978
    ret = bs->drv->bdrv_set_key(bs, key);
979
    if (ret < 0) {
980
        bs->valid_key = 0;
981
    } else if (!bs->valid_key) {
982
        bs->valid_key = 1;
983
        /* call the change callback now, we skipped it on open */
984
        bs->media_changed = 1;
985
        if (bs->change_cb)
986
            bs->change_cb(bs->change_opaque);
987
    }
988
    return ret;
989
}
990

    
991
void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
992
{
993
    if (!bs->drv) {
994
        buf[0] = '\0';
995
    } else {
996
        pstrcpy(buf, buf_size, bs->drv->format_name);
997
    }
998
}
999

    
1000
void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1001
                         void *opaque)
1002
{
1003
    BlockDriver *drv;
1004

    
1005
    for (drv = first_drv; drv != NULL; drv = drv->next) {
1006
        it(opaque, drv->format_name);
1007
    }
1008
}
1009

    
1010
BlockDriverState *bdrv_find(const char *name)
1011
{
1012
    BlockDriverState *bs;
1013

    
1014
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1015
        if (!strcmp(name, bs->device_name))
1016
            return bs;
1017
    }
1018
    return NULL;
1019
}
1020

    
1021
void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1022
{
1023
    BlockDriverState *bs;
1024

    
1025
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1026
        it(opaque, bs);
1027
    }
1028
}
1029

    
1030
const char *bdrv_get_device_name(BlockDriverState *bs)
1031
{
1032
    return bs->device_name;
1033
}
1034

    
1035
void bdrv_flush(BlockDriverState *bs)
1036
{
1037
    if (!bs->drv)
1038
        return;
1039
    if (bs->drv->bdrv_flush)
1040
        bs->drv->bdrv_flush(bs);
1041
    if (bs->backing_hd)
1042
        bdrv_flush(bs->backing_hd);
1043
}
1044

    
1045
void bdrv_flush_all(void)
1046
{
1047
    BlockDriverState *bs;
1048

    
1049
    for (bs = bdrv_first; bs != NULL; bs = bs->next)
1050
        if (bs->drv && !bdrv_is_read_only(bs) && 
1051
            (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1052
            bdrv_flush(bs);
1053
}
1054

    
1055
/*
1056
 * Returns true iff the specified sector is present in the disk image. Drivers
1057
 * not implementing the functionality are assumed to not support backing files,
1058
 * hence all their sectors are reported as allocated.
1059
 *
1060
 * 'pnum' is set to the number of sectors (including and immediately following
1061
 * the specified sector) that are known to be in the same
1062
 * allocated/unallocated state.
1063
 *
1064
 * 'nb_sectors' is the max value 'pnum' should be set to.
1065
 */
1066
int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1067
        int *pnum)
1068
{
1069
    int64_t n;
1070
    if (!bs->drv->bdrv_is_allocated) {
1071
        if (sector_num >= bs->total_sectors) {
1072
            *pnum = 0;
1073
            return 0;
1074
        }
1075
        n = bs->total_sectors - sector_num;
1076
        *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1077
        return 1;
1078
    }
1079
    return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1080
}
1081

    
1082
void bdrv_info(Monitor *mon)
1083
{
1084
    BlockDriverState *bs;
1085

    
1086
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1087
        monitor_printf(mon, "%s:", bs->device_name);
1088
        monitor_printf(mon, " type=");
1089
        switch(bs->type) {
1090
        case BDRV_TYPE_HD:
1091
            monitor_printf(mon, "hd");
1092
            break;
1093
        case BDRV_TYPE_CDROM:
1094
            monitor_printf(mon, "cdrom");
1095
            break;
1096
        case BDRV_TYPE_FLOPPY:
1097
            monitor_printf(mon, "floppy");
1098
            break;
1099
        }
1100
        monitor_printf(mon, " removable=%d", bs->removable);
1101
        if (bs->removable) {
1102
            monitor_printf(mon, " locked=%d", bs->locked);
1103
        }
1104
        if (bs->drv) {
1105
            monitor_printf(mon, " file=");
1106
            monitor_print_filename(mon, bs->filename);
1107
            if (bs->backing_file[0] != '\0') {
1108
                monitor_printf(mon, " backing_file=");
1109
                monitor_print_filename(mon, bs->backing_file);
1110
            }
1111
            monitor_printf(mon, " ro=%d", bs->read_only);
1112
            monitor_printf(mon, " drv=%s", bs->drv->format_name);
1113
            monitor_printf(mon, " encrypted=%d", bdrv_is_encrypted(bs));
1114
        } else {
1115
            monitor_printf(mon, " [not inserted]");
1116
        }
1117
        monitor_printf(mon, "\n");
1118
    }
1119
}
1120

    
1121
/* The "info blockstats" command. */
1122
void bdrv_info_stats(Monitor *mon)
1123
{
1124
    BlockDriverState *bs;
1125

    
1126
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1127
        monitor_printf(mon, "%s:"
1128
                       " rd_bytes=%" PRIu64
1129
                       " wr_bytes=%" PRIu64
1130
                       " rd_operations=%" PRIu64
1131
                       " wr_operations=%" PRIu64
1132
                       "\n",
1133
                       bs->device_name,
1134
                       bs->rd_bytes, bs->wr_bytes,
1135
                       bs->rd_ops, bs->wr_ops);
1136
    }
1137
}
1138

    
1139
const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1140
{
1141
    if (bs->backing_hd && bs->backing_hd->encrypted)
1142
        return bs->backing_file;
1143
    else if (bs->encrypted)
1144
        return bs->filename;
1145
    else
1146
        return NULL;
1147
}
1148

    
1149
void bdrv_get_backing_filename(BlockDriverState *bs,
1150
                               char *filename, int filename_size)
1151
{
1152
    if (!bs->backing_hd) {
1153
        pstrcpy(filename, filename_size, "");
1154
    } else {
1155
        pstrcpy(filename, filename_size, bs->backing_file);
1156
    }
1157
}
1158

    
1159
int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1160
                          const uint8_t *buf, int nb_sectors)
1161
{
1162
    BlockDriver *drv = bs->drv;
1163
    if (!drv)
1164
        return -ENOMEDIUM;
1165
    if (!drv->bdrv_write_compressed)
1166
        return -ENOTSUP;
1167
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1168
        return -EIO;
1169
    return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1170
}
1171

    
1172
int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1173
{
1174
    BlockDriver *drv = bs->drv;
1175
    if (!drv)
1176
        return -ENOMEDIUM;
1177
    if (!drv->bdrv_get_info)
1178
        return -ENOTSUP;
1179
    memset(bdi, 0, sizeof(*bdi));
1180
    return drv->bdrv_get_info(bs, bdi);
1181
}
1182

    
1183
int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1184
                      int64_t pos, int size)
1185
{
1186
    BlockDriver *drv = bs->drv;
1187
    if (!drv)
1188
        return -ENOMEDIUM;
1189
    if (!drv->bdrv_save_vmstate)
1190
        return -ENOTSUP;
1191
    return drv->bdrv_save_vmstate(bs, buf, pos, size);
1192
}
1193

    
1194
int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1195
                      int64_t pos, int size)
1196
{
1197
    BlockDriver *drv = bs->drv;
1198
    if (!drv)
1199
        return -ENOMEDIUM;
1200
    if (!drv->bdrv_load_vmstate)
1201
        return -ENOTSUP;
1202
    return drv->bdrv_load_vmstate(bs, buf, pos, size);
1203
}
1204

    
1205
/**************************************************************/
1206
/* handling of snapshots */
1207

    
1208
int bdrv_snapshot_create(BlockDriverState *bs,
1209
                         QEMUSnapshotInfo *sn_info)
1210
{
1211
    BlockDriver *drv = bs->drv;
1212
    if (!drv)
1213
        return -ENOMEDIUM;
1214
    if (!drv->bdrv_snapshot_create)
1215
        return -ENOTSUP;
1216
    return drv->bdrv_snapshot_create(bs, sn_info);
1217
}
1218

    
1219
int bdrv_snapshot_goto(BlockDriverState *bs,
1220
                       const char *snapshot_id)
1221
{
1222
    BlockDriver *drv = bs->drv;
1223
    if (!drv)
1224
        return -ENOMEDIUM;
1225
    if (!drv->bdrv_snapshot_goto)
1226
        return -ENOTSUP;
1227
    return drv->bdrv_snapshot_goto(bs, snapshot_id);
1228
}
1229

    
1230
int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1231
{
1232
    BlockDriver *drv = bs->drv;
1233
    if (!drv)
1234
        return -ENOMEDIUM;
1235
    if (!drv->bdrv_snapshot_delete)
1236
        return -ENOTSUP;
1237
    return drv->bdrv_snapshot_delete(bs, snapshot_id);
1238
}
1239

    
1240
int bdrv_snapshot_list(BlockDriverState *bs,
1241
                       QEMUSnapshotInfo **psn_info)
1242
{
1243
    BlockDriver *drv = bs->drv;
1244
    if (!drv)
1245
        return -ENOMEDIUM;
1246
    if (!drv->bdrv_snapshot_list)
1247
        return -ENOTSUP;
1248
    return drv->bdrv_snapshot_list(bs, psn_info);
1249
}
1250

    
1251
#define NB_SUFFIXES 4
1252

    
1253
char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1254
{
1255
    static const char suffixes[NB_SUFFIXES] = "KMGT";
1256
    int64_t base;
1257
    int i;
1258

    
1259
    if (size <= 999) {
1260
        snprintf(buf, buf_size, "%" PRId64, size);
1261
    } else {
1262
        base = 1024;
1263
        for(i = 0; i < NB_SUFFIXES; i++) {
1264
            if (size < (10 * base)) {
1265
                snprintf(buf, buf_size, "%0.1f%c",
1266
                         (double)size / base,
1267
                         suffixes[i]);
1268
                break;
1269
            } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1270
                snprintf(buf, buf_size, "%" PRId64 "%c",
1271
                         ((size + (base >> 1)) / base),
1272
                         suffixes[i]);
1273
                break;
1274
            }
1275
            base = base * 1024;
1276
        }
1277
    }
1278
    return buf;
1279
}
1280

    
1281
char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1282
{
1283
    char buf1[128], date_buf[128], clock_buf[128];
1284
#ifdef _WIN32
1285
    struct tm *ptm;
1286
#else
1287
    struct tm tm;
1288
#endif
1289
    time_t ti;
1290
    int64_t secs;
1291

    
1292
    if (!sn) {
1293
        snprintf(buf, buf_size,
1294
                 "%-10s%-20s%7s%20s%15s",
1295
                 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1296
    } else {
1297
        ti = sn->date_sec;
1298
#ifdef _WIN32
1299
        ptm = localtime(&ti);
1300
        strftime(date_buf, sizeof(date_buf),
1301
                 "%Y-%m-%d %H:%M:%S", ptm);
1302
#else
1303
        localtime_r(&ti, &tm);
1304
        strftime(date_buf, sizeof(date_buf),
1305
                 "%Y-%m-%d %H:%M:%S", &tm);
1306
#endif
1307
        secs = sn->vm_clock_nsec / 1000000000;
1308
        snprintf(clock_buf, sizeof(clock_buf),
1309
                 "%02d:%02d:%02d.%03d",
1310
                 (int)(secs / 3600),
1311
                 (int)((secs / 60) % 60),
1312
                 (int)(secs % 60),
1313
                 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1314
        snprintf(buf, buf_size,
1315
                 "%-10s%-20s%7s%20s%15s",
1316
                 sn->id_str, sn->name,
1317
                 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1318
                 date_buf,
1319
                 clock_buf);
1320
    }
1321
    return buf;
1322
}
1323

    
1324

    
1325
/**************************************************************/
1326
/* async I/Os */
1327

    
1328
BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1329
                                 QEMUIOVector *qiov, int nb_sectors,
1330
                                 BlockDriverCompletionFunc *cb, void *opaque)
1331
{
1332
    BlockDriver *drv = bs->drv;
1333
    BlockDriverAIOCB *ret;
1334

    
1335
    if (!drv)
1336
        return NULL;
1337
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1338
        return NULL;
1339

    
1340
    ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1341
                              cb, opaque);
1342

    
1343
    if (ret) {
1344
        /* Update stats even though technically transfer has not happened. */
1345
        bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1346
        bs->rd_ops ++;
1347
    }
1348

    
1349
    return ret;
1350
}
1351

    
1352
BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1353
                                  QEMUIOVector *qiov, int nb_sectors,
1354
                                  BlockDriverCompletionFunc *cb, void *opaque)
1355
{
1356
    BlockDriver *drv = bs->drv;
1357
    BlockDriverAIOCB *ret;
1358

    
1359
    if (!drv)
1360
        return NULL;
1361
    if (bs->read_only)
1362
        return NULL;
1363
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1364
        return NULL;
1365

    
1366
    ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
1367
                               cb, opaque);
1368

    
1369
    if (ret) {
1370
        /* Update stats even though technically transfer has not happened. */
1371
        bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1372
        bs->wr_ops ++;
1373
    }
1374

    
1375
    return ret;
1376
}
1377

    
1378

    
1379
typedef struct MultiwriteCB {
1380
    int error;
1381
    int num_requests;
1382
    int num_callbacks;
1383
    struct {
1384
        BlockDriverCompletionFunc *cb;
1385
        void *opaque;
1386
        QEMUIOVector *free_qiov;
1387
        void *free_buf;
1388
    } callbacks[];
1389
} MultiwriteCB;
1390

    
1391
static void multiwrite_user_cb(MultiwriteCB *mcb)
1392
{
1393
    int i;
1394

    
1395
    for (i = 0; i < mcb->num_callbacks; i++) {
1396
        mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
1397
        qemu_free(mcb->callbacks[i].free_qiov);
1398
        qemu_free(mcb->callbacks[i].free_buf);
1399
    }
1400
}
1401

    
1402
static void multiwrite_cb(void *opaque, int ret)
1403
{
1404
    MultiwriteCB *mcb = opaque;
1405

    
1406
    if (ret < 0) {
1407
        mcb->error = ret;
1408
        multiwrite_user_cb(mcb);
1409
    }
1410

    
1411
    mcb->num_requests--;
1412
    if (mcb->num_requests == 0) {
1413
        if (mcb->error == 0) {
1414
            multiwrite_user_cb(mcb);
1415
        }
1416
        qemu_free(mcb);
1417
    }
1418
}
1419

    
1420
static int multiwrite_req_compare(const void *a, const void *b)
1421
{
1422
    return (((BlockRequest*) a)->sector - ((BlockRequest*) b)->sector);
1423
}
1424

    
1425
/*
1426
 * Takes a bunch of requests and tries to merge them. Returns the number of
1427
 * requests that remain after merging.
1428
 */
1429
static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
1430
    int num_reqs, MultiwriteCB *mcb)
1431
{
1432
    int i, outidx;
1433

    
1434
    // Sort requests by start sector
1435
    qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
1436

    
1437
    // Check if adjacent requests touch the same clusters. If so, combine them,
1438
    // filling up gaps with zero sectors.
1439
    outidx = 0;
1440
    for (i = 1; i < num_reqs; i++) {
1441
        int merge = 0;
1442
        int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
1443

    
1444
        // This handles the cases that are valid for all block drivers, namely
1445
        // exactly sequential writes and overlapping writes.
1446
        if (reqs[i].sector <= oldreq_last) {
1447
            merge = 1;
1448
        }
1449

    
1450
        // The block driver may decide that it makes sense to combine requests
1451
        // even if there is a gap of some sectors between them. In this case,
1452
        // the gap is filled with zeros (therefore only applicable for yet
1453
        // unused space in format like qcow2).
1454
        if (!merge && bs->drv->bdrv_merge_requests) {
1455
            merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
1456
        }
1457

    
1458
        if (merge) {
1459
            size_t size;
1460
            QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
1461
            qemu_iovec_init(qiov,
1462
                reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
1463

    
1464
            // Add the first request to the merged one. If the requests are
1465
            // overlapping, drop the last sectors of the first request.
1466
            size = (reqs[i].sector - reqs[outidx].sector) << 9;
1467
            qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
1468

    
1469
            // We might need to add some zeros between the two requests
1470
            if (reqs[i].sector > oldreq_last) {
1471
                size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
1472
                uint8_t *buf = qemu_blockalign(bs, zero_bytes);
1473
                memset(buf, 0, zero_bytes);
1474
                qemu_iovec_add(qiov, buf, zero_bytes);
1475
                mcb->callbacks[i].free_buf = buf;
1476
            }
1477

    
1478
            // Add the second request
1479
            qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
1480

    
1481
            reqs[outidx].nb_sectors += reqs[i].nb_sectors;
1482
            reqs[outidx].qiov = qiov;
1483

    
1484
            mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
1485
        } else {
1486
            outidx++;
1487
            reqs[outidx].sector     = reqs[i].sector;
1488
            reqs[outidx].nb_sectors = reqs[i].nb_sectors;
1489
            reqs[outidx].qiov       = reqs[i].qiov;
1490
        }
1491
    }
1492

    
1493
    return outidx + 1;
1494
}
1495

    
1496
/*
1497
 * Submit multiple AIO write requests at once.
1498
 *
1499
 * On success, the function returns 0 and all requests in the reqs array have
1500
 * been submitted. In error case this function returns -1, and any of the
1501
 * requests may or may not be submitted yet. In particular, this means that the
1502
 * callback will be called for some of the requests, for others it won't. The
1503
 * caller must check the error field of the BlockRequest to wait for the right
1504
 * callbacks (if error != 0, no callback will be called).
1505
 *
1506
 * The implementation may modify the contents of the reqs array, e.g. to merge
1507
 * requests. However, the fields opaque and error are left unmodified as they
1508
 * are used to signal failure for a single request to the caller.
1509
 */
1510
int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
1511
{
1512
    BlockDriverAIOCB *acb;
1513
    MultiwriteCB *mcb;
1514
    int i;
1515

    
1516
    if (num_reqs == 0) {
1517
        return 0;
1518
    }
1519

    
1520
    // Create MultiwriteCB structure
1521
    mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
1522
    mcb->num_requests = 0;
1523
    mcb->num_callbacks = num_reqs;
1524

    
1525
    for (i = 0; i < num_reqs; i++) {
1526
        mcb->callbacks[i].cb = reqs[i].cb;
1527
        mcb->callbacks[i].opaque = reqs[i].opaque;
1528
    }
1529

    
1530
    // Check for mergable requests
1531
    num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
1532

    
1533
    // Run the aio requests
1534
    for (i = 0; i < num_reqs; i++) {
1535
        acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
1536
            reqs[i].nb_sectors, multiwrite_cb, mcb);
1537

    
1538
        if (acb == NULL) {
1539
            // We can only fail the whole thing if no request has been
1540
            // submitted yet. Otherwise we'll wait for the submitted AIOs to
1541
            // complete and report the error in the callback.
1542
            if (mcb->num_requests == 0) {
1543
                reqs[i].error = EIO;
1544
                goto fail;
1545
            } else {
1546
                mcb->error = EIO;
1547
                break;
1548
            }
1549
        } else {
1550
            mcb->num_requests++;
1551
        }
1552
    }
1553

    
1554
    return 0;
1555

    
1556
fail:
1557
    free(mcb);
1558
    return -1;
1559
}
1560

    
1561
BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
1562
        BlockDriverCompletionFunc *cb, void *opaque)
1563
{
1564
    BlockDriver *drv = bs->drv;
1565

    
1566
    if (!drv)
1567
        return NULL;
1568

    
1569
    /*
1570
     * Note that unlike bdrv_flush the driver is reponsible for flushing a
1571
     * backing image if it exists.
1572
     */
1573
    return drv->bdrv_aio_flush(bs, cb, opaque);
1574
}
1575

    
1576
void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1577
{
1578
    acb->pool->cancel(acb);
1579
}
1580

    
1581

    
1582
/**************************************************************/
1583
/* async block device emulation */
1584

    
1585
typedef struct BlockDriverAIOCBSync {
1586
    BlockDriverAIOCB common;
1587
    QEMUBH *bh;
1588
    int ret;
1589
    /* vector translation state */
1590
    QEMUIOVector *qiov;
1591
    uint8_t *bounce;
1592
    int is_write;
1593
} BlockDriverAIOCBSync;
1594

    
1595
static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1596
{
1597
    BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1598
    qemu_bh_delete(acb->bh);
1599
    acb->bh = NULL;
1600
    qemu_aio_release(acb);
1601
}
1602

    
1603
static AIOPool bdrv_em_aio_pool = {
1604
    .aiocb_size         = sizeof(BlockDriverAIOCBSync),
1605
    .cancel             = bdrv_aio_cancel_em,
1606
};
1607

    
1608
static void bdrv_aio_bh_cb(void *opaque)
1609
{
1610
    BlockDriverAIOCBSync *acb = opaque;
1611

    
1612
    if (!acb->is_write)
1613
        qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
1614
    qemu_vfree(acb->bounce);
1615
    acb->common.cb(acb->common.opaque, acb->ret);
1616
    qemu_bh_delete(acb->bh);
1617
    acb->bh = NULL;
1618
    qemu_aio_release(acb);
1619
}
1620

    
1621
static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
1622
                                            int64_t sector_num,
1623
                                            QEMUIOVector *qiov,
1624
                                            int nb_sectors,
1625
                                            BlockDriverCompletionFunc *cb,
1626
                                            void *opaque,
1627
                                            int is_write)
1628

    
1629
{
1630
    BlockDriverAIOCBSync *acb;
1631

    
1632
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
1633
    acb->is_write = is_write;
1634
    acb->qiov = qiov;
1635
    acb->bounce = qemu_blockalign(bs, qiov->size);
1636

    
1637
    if (!acb->bh)
1638
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1639

    
1640
    if (is_write) {
1641
        qemu_iovec_to_buffer(acb->qiov, acb->bounce);
1642
        acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
1643
    } else {
1644
        acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
1645
    }
1646

    
1647
    qemu_bh_schedule(acb->bh);
1648

    
1649
    return &acb->common;
1650
}
1651

    
1652
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
1653
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1654
        BlockDriverCompletionFunc *cb, void *opaque)
1655
{
1656
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
1657
}
1658

    
1659
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
1660
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1661
        BlockDriverCompletionFunc *cb, void *opaque)
1662
{
1663
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
1664
}
1665

    
1666
static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
1667
        BlockDriverCompletionFunc *cb, void *opaque)
1668
{
1669
    BlockDriverAIOCBSync *acb;
1670

    
1671
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
1672
    acb->is_write = 1; /* don't bounce in the completion hadler */
1673
    acb->qiov = NULL;
1674
    acb->bounce = NULL;
1675
    acb->ret = 0;
1676

    
1677
    if (!acb->bh)
1678
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1679

    
1680
    bdrv_flush(bs);
1681
    qemu_bh_schedule(acb->bh);
1682
    return &acb->common;
1683
}
1684

    
1685
/**************************************************************/
1686
/* sync block device emulation */
1687

    
1688
static void bdrv_rw_em_cb(void *opaque, int ret)
1689
{
1690
    *(int *)opaque = ret;
1691
}
1692

    
1693
#define NOT_DONE 0x7fffffff
1694

    
1695
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1696
                        uint8_t *buf, int nb_sectors)
1697
{
1698
    int async_ret;
1699
    BlockDriverAIOCB *acb;
1700
    struct iovec iov;
1701
    QEMUIOVector qiov;
1702

    
1703
    async_ret = NOT_DONE;
1704
    iov.iov_base = (void *)buf;
1705
    iov.iov_len = nb_sectors * 512;
1706
    qemu_iovec_init_external(&qiov, &iov, 1);
1707
    acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
1708
        bdrv_rw_em_cb, &async_ret);
1709
    if (acb == NULL)
1710
        return -1;
1711

    
1712
    while (async_ret == NOT_DONE) {
1713
        qemu_aio_wait();
1714
    }
1715

    
1716
    return async_ret;
1717
}
1718

    
1719
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1720
                         const uint8_t *buf, int nb_sectors)
1721
{
1722
    int async_ret;
1723
    BlockDriverAIOCB *acb;
1724
    struct iovec iov;
1725
    QEMUIOVector qiov;
1726

    
1727
    async_ret = NOT_DONE;
1728
    iov.iov_base = (void *)buf;
1729
    iov.iov_len = nb_sectors * 512;
1730
    qemu_iovec_init_external(&qiov, &iov, 1);
1731
    acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
1732
        bdrv_rw_em_cb, &async_ret);
1733
    if (acb == NULL)
1734
        return -1;
1735
    while (async_ret == NOT_DONE) {
1736
        qemu_aio_wait();
1737
    }
1738
    return async_ret;
1739
}
1740

    
1741
void bdrv_init(void)
1742
{
1743
    module_call_init(MODULE_INIT_BLOCK);
1744
}
1745

    
1746
void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
1747
                   BlockDriverCompletionFunc *cb, void *opaque)
1748
{
1749
    BlockDriverAIOCB *acb;
1750

    
1751
    if (pool->free_aiocb) {
1752
        acb = pool->free_aiocb;
1753
        pool->free_aiocb = acb->next;
1754
    } else {
1755
        acb = qemu_mallocz(pool->aiocb_size);
1756
        acb->pool = pool;
1757
    }
1758
    acb->bs = bs;
1759
    acb->cb = cb;
1760
    acb->opaque = opaque;
1761
    return acb;
1762
}
1763

    
1764
void qemu_aio_release(void *p)
1765
{
1766
    BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
1767
    AIOPool *pool = acb->pool;
1768
    acb->next = pool->free_aiocb;
1769
    pool->free_aiocb = acb;
1770
}
1771

    
1772
/**************************************************************/
1773
/* removable device support */
1774

    
1775
/**
1776
 * Return TRUE if the media is present
1777
 */
1778
int bdrv_is_inserted(BlockDriverState *bs)
1779
{
1780
    BlockDriver *drv = bs->drv;
1781
    int ret;
1782
    if (!drv)
1783
        return 0;
1784
    if (!drv->bdrv_is_inserted)
1785
        return 1;
1786
    ret = drv->bdrv_is_inserted(bs);
1787
    return ret;
1788
}
1789

    
1790
/**
1791
 * Return TRUE if the media changed since the last call to this
1792
 * function. It is currently only used for floppy disks
1793
 */
1794
int bdrv_media_changed(BlockDriverState *bs)
1795
{
1796
    BlockDriver *drv = bs->drv;
1797
    int ret;
1798

    
1799
    if (!drv || !drv->bdrv_media_changed)
1800
        ret = -ENOTSUP;
1801
    else
1802
        ret = drv->bdrv_media_changed(bs);
1803
    if (ret == -ENOTSUP)
1804
        ret = bs->media_changed;
1805
    bs->media_changed = 0;
1806
    return ret;
1807
}
1808

    
1809
/**
1810
 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1811
 */
1812
int bdrv_eject(BlockDriverState *bs, int eject_flag)
1813
{
1814
    BlockDriver *drv = bs->drv;
1815
    int ret;
1816

    
1817
    if (bs->locked) {
1818
        return -EBUSY;
1819
    }
1820

    
1821
    if (!drv || !drv->bdrv_eject) {
1822
        ret = -ENOTSUP;
1823
    } else {
1824
        ret = drv->bdrv_eject(bs, eject_flag);
1825
    }
1826
    if (ret == -ENOTSUP) {
1827
        if (eject_flag)
1828
            bdrv_close(bs);
1829
        ret = 0;
1830
    }
1831

    
1832
    return ret;
1833
}
1834

    
1835
int bdrv_is_locked(BlockDriverState *bs)
1836
{
1837
    return bs->locked;
1838
}
1839

    
1840
/**
1841
 * Lock or unlock the media (if it is locked, the user won't be able
1842
 * to eject it manually).
1843
 */
1844
void bdrv_set_locked(BlockDriverState *bs, int locked)
1845
{
1846
    BlockDriver *drv = bs->drv;
1847

    
1848
    bs->locked = locked;
1849
    if (drv && drv->bdrv_set_locked) {
1850
        drv->bdrv_set_locked(bs, locked);
1851
    }
1852
}
1853

    
1854
/* needed for generic scsi interface */
1855

    
1856
int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1857
{
1858
    BlockDriver *drv = bs->drv;
1859

    
1860
    if (drv && drv->bdrv_ioctl)
1861
        return drv->bdrv_ioctl(bs, req, buf);
1862
    return -ENOTSUP;
1863
}
1864

    
1865
BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
1866
        unsigned long int req, void *buf,
1867
        BlockDriverCompletionFunc *cb, void *opaque)
1868
{
1869
    BlockDriver *drv = bs->drv;
1870

    
1871
    if (drv && drv->bdrv_aio_ioctl)
1872
        return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
1873
    return NULL;
1874
}
1875

    
1876
void *qemu_blockalign(BlockDriverState *bs, size_t size)
1877
{
1878
    return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
1879
}