Statistics
| Branch: | Revision:

root / block.c @ f5edb014

History | View | Annotate | Download (58.9 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
#include "qemu-common.h"
26
#include "monitor.h"
27
#include "block_int.h"
28
#include "module.h"
29
#include "qemu-objects.h"
30

    
31
#ifdef CONFIG_BSD
32
#include <sys/types.h>
33
#include <sys/stat.h>
34
#include <sys/ioctl.h>
35
#include <sys/queue.h>
36
#ifndef __DragonFly__
37
#include <sys/disk.h>
38
#endif
39
#endif
40

    
41
#ifdef _WIN32
42
#include <windows.h>
43
#endif
44

    
45
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
46
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
47
        BlockDriverCompletionFunc *cb, void *opaque);
48
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
49
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
50
        BlockDriverCompletionFunc *cb, void *opaque);
51
static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
52
        BlockDriverCompletionFunc *cb, void *opaque);
53
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
54
                        uint8_t *buf, int nb_sectors);
55
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
56
                         const uint8_t *buf, int nb_sectors);
57

    
58
BlockDriverState *bdrv_first;
59

    
60
static BlockDriver *first_drv;
61

    
62
/* If non-zero, use only whitelisted block drivers */
63
static int use_bdrv_whitelist;
64

    
65
int path_is_absolute(const char *path)
66
{
67
    const char *p;
68
#ifdef _WIN32
69
    /* specific case for names like: "\\.\d:" */
70
    if (*path == '/' || *path == '\\')
71
        return 1;
72
#endif
73
    p = strchr(path, ':');
74
    if (p)
75
        p++;
76
    else
77
        p = path;
78
#ifdef _WIN32
79
    return (*p == '/' || *p == '\\');
80
#else
81
    return (*p == '/');
82
#endif
83
}
84

    
85
/* if filename is absolute, just copy it to dest. Otherwise, build a
86
   path to it by considering it is relative to base_path. URL are
87
   supported. */
88
void path_combine(char *dest, int dest_size,
89
                  const char *base_path,
90
                  const char *filename)
91
{
92
    const char *p, *p1;
93
    int len;
94

    
95
    if (dest_size <= 0)
96
        return;
97
    if (path_is_absolute(filename)) {
98
        pstrcpy(dest, dest_size, filename);
99
    } else {
100
        p = strchr(base_path, ':');
101
        if (p)
102
            p++;
103
        else
104
            p = base_path;
105
        p1 = strrchr(base_path, '/');
106
#ifdef _WIN32
107
        {
108
            const char *p2;
109
            p2 = strrchr(base_path, '\\');
110
            if (!p1 || p2 > p1)
111
                p1 = p2;
112
        }
113
#endif
114
        if (p1)
115
            p1++;
116
        else
117
            p1 = base_path;
118
        if (p1 > p)
119
            p = p1;
120
        len = p - base_path;
121
        if (len > dest_size - 1)
122
            len = dest_size - 1;
123
        memcpy(dest, base_path, len);
124
        dest[len] = '\0';
125
        pstrcat(dest, dest_size, filename);
126
    }
127
}
128

    
129
void bdrv_register(BlockDriver *bdrv)
130
{
131
    if (!bdrv->bdrv_aio_readv) {
132
        /* add AIO emulation layer */
133
        bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
134
        bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
135
    } else if (!bdrv->bdrv_read) {
136
        /* add synchronous IO emulation layer */
137
        bdrv->bdrv_read = bdrv_read_em;
138
        bdrv->bdrv_write = bdrv_write_em;
139
    }
140

    
141
    if (!bdrv->bdrv_aio_flush)
142
        bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
143

    
144
    bdrv->next = first_drv;
145
    first_drv = bdrv;
146
}
147

    
148
/* create a new block device (by default it is empty) */
149
BlockDriverState *bdrv_new(const char *device_name)
150
{
151
    BlockDriverState **pbs, *bs;
152

    
153
    bs = qemu_mallocz(sizeof(BlockDriverState));
154
    pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
155
    if (device_name[0] != '\0') {
156
        /* insert at the end */
157
        pbs = &bdrv_first;
158
        while (*pbs != NULL)
159
            pbs = &(*pbs)->next;
160
        *pbs = bs;
161
    }
162
    return bs;
163
}
164

    
165
BlockDriver *bdrv_find_format(const char *format_name)
166
{
167
    BlockDriver *drv1;
168
    for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
169
        if (!strcmp(drv1->format_name, format_name))
170
            return drv1;
171
    }
172
    return NULL;
173
}
174

    
175
static int bdrv_is_whitelisted(BlockDriver *drv)
176
{
177
    static const char *whitelist[] = {
178
        CONFIG_BDRV_WHITELIST
179
    };
180
    const char **p;
181

    
182
    if (!whitelist[0])
183
        return 1;               /* no whitelist, anything goes */
184

    
185
    for (p = whitelist; *p; p++) {
186
        if (!strcmp(drv->format_name, *p)) {
187
            return 1;
188
        }
189
    }
190
    return 0;
191
}
192

    
193
BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
194
{
195
    BlockDriver *drv = bdrv_find_format(format_name);
196
    return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
197
}
198

    
199
int bdrv_create(BlockDriver *drv, const char* filename,
200
    QEMUOptionParameter *options)
201
{
202
    if (!drv->bdrv_create)
203
        return -ENOTSUP;
204

    
205
    return drv->bdrv_create(filename, options);
206
}
207

    
208
#ifdef _WIN32
209
void get_tmp_filename(char *filename, int size)
210
{
211
    char temp_dir[MAX_PATH];
212

    
213
    GetTempPath(MAX_PATH, temp_dir);
214
    GetTempFileName(temp_dir, "qem", 0, filename);
215
}
216
#else
217
void get_tmp_filename(char *filename, int size)
218
{
219
    int fd;
220
    const char *tmpdir;
221
    /* XXX: race condition possible */
222
    tmpdir = getenv("TMPDIR");
223
    if (!tmpdir)
224
        tmpdir = "/tmp";
225
    snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
226
    fd = mkstemp(filename);
227
    close(fd);
228
}
229
#endif
230

    
231
#ifdef _WIN32
232
static int is_windows_drive_prefix(const char *filename)
233
{
234
    return (((filename[0] >= 'a' && filename[0] <= 'z') ||
235
             (filename[0] >= 'A' && filename[0] <= 'Z')) &&
236
            filename[1] == ':');
237
}
238

    
239
int is_windows_drive(const char *filename)
240
{
241
    if (is_windows_drive_prefix(filename) &&
242
        filename[2] == '\0')
243
        return 1;
244
    if (strstart(filename, "\\\\.\\", NULL) ||
245
        strstart(filename, "//./", NULL))
246
        return 1;
247
    return 0;
248
}
249
#endif
250

    
251
static BlockDriver *find_protocol(const char *filename)
252
{
253
    BlockDriver *drv1;
254
    char protocol[128];
255
    int len;
256
    const char *p;
257

    
258
#ifdef _WIN32
259
    if (is_windows_drive(filename) ||
260
        is_windows_drive_prefix(filename))
261
        return bdrv_find_format("raw");
262
#endif
263
    p = strchr(filename, ':');
264
    if (!p)
265
        return bdrv_find_format("raw");
266
    len = p - filename;
267
    if (len > sizeof(protocol) - 1)
268
        len = sizeof(protocol) - 1;
269
    memcpy(protocol, filename, len);
270
    protocol[len] = '\0';
271
    for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
272
        if (drv1->protocol_name &&
273
            !strcmp(drv1->protocol_name, protocol))
274
            return drv1;
275
    }
276
    return NULL;
277
}
278

    
279
/*
280
 * Detect host devices. By convention, /dev/cdrom[N] is always
281
 * recognized as a host CDROM.
282
 */
283
static BlockDriver *find_hdev_driver(const char *filename)
284
{
285
    int score_max = 0, score;
286
    BlockDriver *drv = NULL, *d;
287

    
288
    for (d = first_drv; d; d = d->next) {
289
        if (d->bdrv_probe_device) {
290
            score = d->bdrv_probe_device(filename);
291
            if (score > score_max) {
292
                score_max = score;
293
                drv = d;
294
            }
295
        }
296
    }
297

    
298
    return drv;
299
}
300

    
301
static BlockDriver *find_image_format(const char *filename)
302
{
303
    int ret, score, score_max;
304
    BlockDriver *drv1, *drv;
305
    uint8_t buf[2048];
306
    BlockDriverState *bs;
307

    
308
    drv = find_protocol(filename);
309
    /* no need to test disk image formats for vvfat */
310
    if (drv && strcmp(drv->format_name, "vvfat") == 0)
311
        return drv;
312

    
313
    ret = bdrv_file_open(&bs, filename, 0);
314
    if (ret < 0)
315
        return NULL;
316
    ret = bdrv_pread(bs, 0, buf, sizeof(buf));
317
    bdrv_delete(bs);
318
    if (ret < 0) {
319
        return NULL;
320
    }
321

    
322
    score_max = 0;
323
    for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
324
        if (drv1->bdrv_probe) {
325
            score = drv1->bdrv_probe(buf, ret, filename);
326
            if (score > score_max) {
327
                score_max = score;
328
                drv = drv1;
329
            }
330
        }
331
    }
332
    return drv;
333
}
334

    
335
int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
336
{
337
    BlockDriverState *bs;
338
    int ret;
339

    
340
    bs = bdrv_new("");
341
    ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
342
    if (ret < 0) {
343
        bdrv_delete(bs);
344
        return ret;
345
    }
346
    bs->growable = 1;
347
    *pbs = bs;
348
    return 0;
349
}
350

    
351
int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
352
{
353
    return bdrv_open2(bs, filename, flags, NULL);
354
}
355

    
356
int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
357
               BlockDriver *drv)
358
{
359
    int ret, open_flags;
360
    char tmp_filename[PATH_MAX];
361
    char backing_filename[PATH_MAX];
362

    
363
    bs->is_temporary = 0;
364
    bs->encrypted = 0;
365
    bs->valid_key = 0;
366
    /* buffer_alignment defaulted to 512, drivers can change this value */
367
    bs->buffer_alignment = 512;
368

    
369
    if (flags & BDRV_O_SNAPSHOT) {
370
        BlockDriverState *bs1;
371
        int64_t total_size;
372
        int is_protocol = 0;
373
        BlockDriver *bdrv_qcow2;
374
        QEMUOptionParameter *options;
375

    
376
        /* if snapshot, we create a temporary backing file and open it
377
           instead of opening 'filename' directly */
378

    
379
        /* if there is a backing file, use it */
380
        bs1 = bdrv_new("");
381
        ret = bdrv_open2(bs1, filename, 0, drv);
382
        if (ret < 0) {
383
            bdrv_delete(bs1);
384
            return ret;
385
        }
386
        total_size = bdrv_getlength(bs1) >> BDRV_SECTOR_BITS;
387

    
388
        if (bs1->drv && bs1->drv->protocol_name)
389
            is_protocol = 1;
390

    
391
        bdrv_delete(bs1);
392

    
393
        get_tmp_filename(tmp_filename, sizeof(tmp_filename));
394

    
395
        /* Real path is meaningless for protocols */
396
        if (is_protocol)
397
            snprintf(backing_filename, sizeof(backing_filename),
398
                     "%s", filename);
399
        else if (!realpath(filename, backing_filename))
400
            return -errno;
401

    
402
        bdrv_qcow2 = bdrv_find_format("qcow2");
403
        options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
404

    
405
        set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size * 512);
406
        set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
407
        if (drv) {
408
            set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
409
                drv->format_name);
410
        }
411

    
412
        ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
413
        if (ret < 0) {
414
            return ret;
415
        }
416

    
417
        filename = tmp_filename;
418
        drv = bdrv_qcow2;
419
        bs->is_temporary = 1;
420
    }
421

    
422
    pstrcpy(bs->filename, sizeof(bs->filename), filename);
423
    if (flags & BDRV_O_FILE) {
424
        drv = find_protocol(filename);
425
    } else if (!drv) {
426
        drv = find_hdev_driver(filename);
427
        if (!drv) {
428
            drv = find_image_format(filename);
429
        }
430
    }
431
    if (!drv) {
432
        ret = -ENOENT;
433
        goto unlink_and_fail;
434
    }
435
    bs->drv = drv;
436
    bs->opaque = qemu_mallocz(drv->instance_size);
437

    
438
    /*
439
     * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
440
     * write cache to the guest.  We do need the fdatasync to flush
441
     * out transactions for block allocations, and we maybe have a
442
     * volatile write cache in our backing device to deal with.
443
     */
444
    if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
445
        bs->enable_write_cache = 1;
446

    
447
    /* Note: for compatibility, we open disk image files as RDWR, and
448
       RDONLY as fallback */
449
    bs->read_only = (flags & BDRV_O_RDWR) == 0;
450
    if (!(flags & BDRV_O_FILE)) {
451
        open_flags = (flags & (BDRV_O_RDWR | BDRV_O_CACHE_MASK|BDRV_O_NATIVE_AIO));
452
        if (bs->is_temporary) { /* snapshot should be writeable */
453
            open_flags |= BDRV_O_RDWR;
454
        }
455
    } else {
456
        open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
457
    }
458
    if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
459
        ret = -ENOTSUP;
460
    } else {
461
        ret = drv->bdrv_open(bs, filename, open_flags);
462
        if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
463
            ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
464
            bs->read_only = 1;
465
        }
466
    }
467
    if (ret < 0) {
468
        qemu_free(bs->opaque);
469
        bs->opaque = NULL;
470
        bs->drv = NULL;
471
    unlink_and_fail:
472
        if (bs->is_temporary)
473
            unlink(filename);
474
        return ret;
475
    }
476
    if (drv->bdrv_getlength) {
477
        bs->total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
478
    }
479
#ifndef _WIN32
480
    if (bs->is_temporary) {
481
        unlink(filename);
482
    }
483
#endif
484
    if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
485
        /* if there is a backing file, use it */
486
        BlockDriver *back_drv = NULL;
487
        bs->backing_hd = bdrv_new("");
488
        path_combine(backing_filename, sizeof(backing_filename),
489
                     filename, bs->backing_file);
490
        if (bs->backing_format[0] != '\0')
491
            back_drv = bdrv_find_format(bs->backing_format);
492
        ret = bdrv_open2(bs->backing_hd, backing_filename, open_flags,
493
                         back_drv);
494
        bs->backing_hd->read_only =  (open_flags & BDRV_O_RDWR) == 0;
495
        if (ret < 0) {
496
            bdrv_close(bs);
497
            return ret;
498
        }
499
    }
500

    
501
    if (!bdrv_key_required(bs)) {
502
        /* call the change callback */
503
        bs->media_changed = 1;
504
        if (bs->change_cb)
505
            bs->change_cb(bs->change_opaque);
506
    }
507
    return 0;
508
}
509

    
510
void bdrv_close(BlockDriverState *bs)
511
{
512
    if (bs->drv) {
513
        if (bs->backing_hd)
514
            bdrv_delete(bs->backing_hd);
515
        bs->drv->bdrv_close(bs);
516
        qemu_free(bs->opaque);
517
#ifdef _WIN32
518
        if (bs->is_temporary) {
519
            unlink(bs->filename);
520
        }
521
#endif
522
        bs->opaque = NULL;
523
        bs->drv = NULL;
524

    
525
        /* call the change callback */
526
        bs->media_changed = 1;
527
        if (bs->change_cb)
528
            bs->change_cb(bs->change_opaque);
529
    }
530
}
531

    
532
void bdrv_delete(BlockDriverState *bs)
533
{
534
    BlockDriverState **pbs;
535

    
536
    pbs = &bdrv_first;
537
    while (*pbs != bs && *pbs != NULL)
538
        pbs = &(*pbs)->next;
539
    if (*pbs == bs)
540
        *pbs = bs->next;
541

    
542
    bdrv_close(bs);
543
    qemu_free(bs);
544
}
545

    
546
/*
547
 * Run consistency checks on an image
548
 *
549
 * Returns the number of errors or -errno when an internal error occurs
550
 */
551
int bdrv_check(BlockDriverState *bs)
552
{
553
    if (bs->drv->bdrv_check == NULL) {
554
        return -ENOTSUP;
555
    }
556

    
557
    return bs->drv->bdrv_check(bs);
558
}
559

    
560
/* commit COW file into the raw image */
561
int bdrv_commit(BlockDriverState *bs)
562
{
563
    BlockDriver *drv = bs->drv;
564
    int64_t i, total_sectors;
565
    int n, j;
566
    unsigned char sector[512];
567

    
568
    if (!drv)
569
        return -ENOMEDIUM;
570

    
571
    if (bs->read_only) {
572
        return -EACCES;
573
    }
574

    
575
    if (!bs->backing_hd) {
576
        return -ENOTSUP;
577
    }
578

    
579
    total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
580
    for (i = 0; i < total_sectors;) {
581
        if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
582
            for(j = 0; j < n; j++) {
583
                if (bdrv_read(bs, i, sector, 1) != 0) {
584
                    return -EIO;
585
                }
586

    
587
                if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
588
                    return -EIO;
589
                }
590
                i++;
591
            }
592
        } else {
593
            i += n;
594
        }
595
    }
596

    
597
    if (drv->bdrv_make_empty)
598
        return drv->bdrv_make_empty(bs);
599

    
600
    /*
601
     * Make sure all data we wrote to the backing device is actually
602
     * stable on disk.
603
     */
604
    if (bs->backing_hd)
605
        bdrv_flush(bs->backing_hd);
606
    return 0;
607
}
608

    
609
/*
610
 * Return values:
611
 * 0        - success
612
 * -EINVAL  - backing format specified, but no file
613
 * -ENOSPC  - can't update the backing file because no space is left in the
614
 *            image file header
615
 * -ENOTSUP - format driver doesn't support changing the backing file
616
 */
617
int bdrv_change_backing_file(BlockDriverState *bs,
618
    const char *backing_file, const char *backing_fmt)
619
{
620
    BlockDriver *drv = bs->drv;
621

    
622
    if (drv->bdrv_change_backing_file != NULL) {
623
        return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
624
    } else {
625
        return -ENOTSUP;
626
    }
627
}
628

    
629
static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
630
                                   size_t size)
631
{
632
    int64_t len;
633

    
634
    if (!bdrv_is_inserted(bs))
635
        return -ENOMEDIUM;
636

    
637
    if (bs->growable)
638
        return 0;
639

    
640
    len = bdrv_getlength(bs);
641

    
642
    if (offset < 0)
643
        return -EIO;
644

    
645
    if ((offset > len) || (len - offset < size))
646
        return -EIO;
647

    
648
    return 0;
649
}
650

    
651
static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
652
                              int nb_sectors)
653
{
654
    return bdrv_check_byte_request(bs, sector_num * 512, nb_sectors * 512);
655
}
656

    
657
/* return < 0 if error. See bdrv_write() for the return codes */
658
int bdrv_read(BlockDriverState *bs, int64_t sector_num,
659
              uint8_t *buf, int nb_sectors)
660
{
661
    BlockDriver *drv = bs->drv;
662

    
663
    if (!drv)
664
        return -ENOMEDIUM;
665
    if (bdrv_check_request(bs, sector_num, nb_sectors))
666
        return -EIO;
667

    
668
    return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
669
}
670

    
671
static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
672
                             int nb_sectors, int dirty)
673
{
674
    int64_t start, end;
675
    unsigned long val, idx, bit;
676

    
677
    start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
678
    end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
679

    
680
    for (; start <= end; start++) {
681
        idx = start / (sizeof(unsigned long) * 8);
682
        bit = start % (sizeof(unsigned long) * 8);
683
        val = bs->dirty_bitmap[idx];
684
        if (dirty) {
685
            val |= 1 << bit;
686
        } else {
687
            val &= ~(1 << bit);
688
        }
689
        bs->dirty_bitmap[idx] = val;
690
    }
691
}
692

    
693
/* Return < 0 if error. Important errors are:
694
  -EIO         generic I/O error (may happen for all errors)
695
  -ENOMEDIUM   No media inserted.
696
  -EINVAL      Invalid sector number or nb_sectors
697
  -EACCES      Trying to write a read-only device
698
*/
699
int bdrv_write(BlockDriverState *bs, int64_t sector_num,
700
               const uint8_t *buf, int nb_sectors)
701
{
702
    BlockDriver *drv = bs->drv;
703
    if (!bs->drv)
704
        return -ENOMEDIUM;
705
    if (bs->read_only)
706
        return -EACCES;
707
    if (bdrv_check_request(bs, sector_num, nb_sectors))
708
        return -EIO;
709

    
710
    if (bs->dirty_bitmap) {
711
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
712
    }
713

    
714
    return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
715
}
716

    
717
int bdrv_pread(BlockDriverState *bs, int64_t offset,
718
               void *buf, int count1)
719
{
720
    uint8_t tmp_buf[BDRV_SECTOR_SIZE];
721
    int len, nb_sectors, count;
722
    int64_t sector_num;
723

    
724
    count = count1;
725
    /* first read to align to sector start */
726
    len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
727
    if (len > count)
728
        len = count;
729
    sector_num = offset >> BDRV_SECTOR_BITS;
730
    if (len > 0) {
731
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
732
            return -EIO;
733
        memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
734
        count -= len;
735
        if (count == 0)
736
            return count1;
737
        sector_num++;
738
        buf += len;
739
    }
740

    
741
    /* read the sectors "in place" */
742
    nb_sectors = count >> BDRV_SECTOR_BITS;
743
    if (nb_sectors > 0) {
744
        if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
745
            return -EIO;
746
        sector_num += nb_sectors;
747
        len = nb_sectors << BDRV_SECTOR_BITS;
748
        buf += len;
749
        count -= len;
750
    }
751

    
752
    /* add data from the last sector */
753
    if (count > 0) {
754
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
755
            return -EIO;
756
        memcpy(buf, tmp_buf, count);
757
    }
758
    return count1;
759
}
760

    
761
int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
762
                const void *buf, int count1)
763
{
764
    uint8_t tmp_buf[BDRV_SECTOR_SIZE];
765
    int len, nb_sectors, count;
766
    int64_t sector_num;
767

    
768
    count = count1;
769
    /* first write to align to sector start */
770
    len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
771
    if (len > count)
772
        len = count;
773
    sector_num = offset >> BDRV_SECTOR_BITS;
774
    if (len > 0) {
775
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
776
            return -EIO;
777
        memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
778
        if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
779
            return -EIO;
780
        count -= len;
781
        if (count == 0)
782
            return count1;
783
        sector_num++;
784
        buf += len;
785
    }
786

    
787
    /* write the sectors "in place" */
788
    nb_sectors = count >> BDRV_SECTOR_BITS;
789
    if (nb_sectors > 0) {
790
        if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
791
            return -EIO;
792
        sector_num += nb_sectors;
793
        len = nb_sectors << BDRV_SECTOR_BITS;
794
        buf += len;
795
        count -= len;
796
    }
797

    
798
    /* add data from the last sector */
799
    if (count > 0) {
800
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
801
            return -EIO;
802
        memcpy(tmp_buf, buf, count);
803
        if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
804
            return -EIO;
805
    }
806
    return count1;
807
}
808

    
809
/**
810
 * Truncate file to 'offset' bytes (needed only for file protocols)
811
 */
812
int bdrv_truncate(BlockDriverState *bs, int64_t offset)
813
{
814
    BlockDriver *drv = bs->drv;
815
    if (!drv)
816
        return -ENOMEDIUM;
817
    if (!drv->bdrv_truncate)
818
        return -ENOTSUP;
819
    if (bs->read_only)
820
        return -EACCES;
821
    return drv->bdrv_truncate(bs, offset);
822
}
823

    
824
/**
825
 * Length of a file in bytes. Return < 0 if error or unknown.
826
 */
827
int64_t bdrv_getlength(BlockDriverState *bs)
828
{
829
    BlockDriver *drv = bs->drv;
830
    if (!drv)
831
        return -ENOMEDIUM;
832
    if (!drv->bdrv_getlength) {
833
        /* legacy mode */
834
        return bs->total_sectors * BDRV_SECTOR_SIZE;
835
    }
836
    return drv->bdrv_getlength(bs);
837
}
838

    
839
/* return 0 as number of sectors if no device present or error */
840
void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
841
{
842
    int64_t length;
843
    length = bdrv_getlength(bs);
844
    if (length < 0)
845
        length = 0;
846
    else
847
        length = length >> BDRV_SECTOR_BITS;
848
    *nb_sectors_ptr = length;
849
}
850

    
851
struct partition {
852
        uint8_t boot_ind;           /* 0x80 - active */
853
        uint8_t head;               /* starting head */
854
        uint8_t sector;             /* starting sector */
855
        uint8_t cyl;                /* starting cylinder */
856
        uint8_t sys_ind;            /* What partition type */
857
        uint8_t end_head;           /* end head */
858
        uint8_t end_sector;         /* end sector */
859
        uint8_t end_cyl;            /* end cylinder */
860
        uint32_t start_sect;        /* starting sector counting from 0 */
861
        uint32_t nr_sects;          /* nr of sectors in partition */
862
} __attribute__((packed));
863

    
864
/* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
865
static int guess_disk_lchs(BlockDriverState *bs,
866
                           int *pcylinders, int *pheads, int *psectors)
867
{
868
    uint8_t buf[512];
869
    int ret, i, heads, sectors, cylinders;
870
    struct partition *p;
871
    uint32_t nr_sects;
872
    uint64_t nb_sectors;
873

    
874
    bdrv_get_geometry(bs, &nb_sectors);
875

    
876
    ret = bdrv_read(bs, 0, buf, 1);
877
    if (ret < 0)
878
        return -1;
879
    /* test msdos magic */
880
    if (buf[510] != 0x55 || buf[511] != 0xaa)
881
        return -1;
882
    for(i = 0; i < 4; i++) {
883
        p = ((struct partition *)(buf + 0x1be)) + i;
884
        nr_sects = le32_to_cpu(p->nr_sects);
885
        if (nr_sects && p->end_head) {
886
            /* We make the assumption that the partition terminates on
887
               a cylinder boundary */
888
            heads = p->end_head + 1;
889
            sectors = p->end_sector & 63;
890
            if (sectors == 0)
891
                continue;
892
            cylinders = nb_sectors / (heads * sectors);
893
            if (cylinders < 1 || cylinders > 16383)
894
                continue;
895
            *pheads = heads;
896
            *psectors = sectors;
897
            *pcylinders = cylinders;
898
#if 0
899
            printf("guessed geometry: LCHS=%d %d %d\n",
900
                   cylinders, heads, sectors);
901
#endif
902
            return 0;
903
        }
904
    }
905
    return -1;
906
}
907

    
908
void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
909
{
910
    int translation, lba_detected = 0;
911
    int cylinders, heads, secs;
912
    uint64_t nb_sectors;
913

    
914
    /* if a geometry hint is available, use it */
915
    bdrv_get_geometry(bs, &nb_sectors);
916
    bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
917
    translation = bdrv_get_translation_hint(bs);
918
    if (cylinders != 0) {
919
        *pcyls = cylinders;
920
        *pheads = heads;
921
        *psecs = secs;
922
    } else {
923
        if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
924
            if (heads > 16) {
925
                /* if heads > 16, it means that a BIOS LBA
926
                   translation was active, so the default
927
                   hardware geometry is OK */
928
                lba_detected = 1;
929
                goto default_geometry;
930
            } else {
931
                *pcyls = cylinders;
932
                *pheads = heads;
933
                *psecs = secs;
934
                /* disable any translation to be in sync with
935
                   the logical geometry */
936
                if (translation == BIOS_ATA_TRANSLATION_AUTO) {
937
                    bdrv_set_translation_hint(bs,
938
                                              BIOS_ATA_TRANSLATION_NONE);
939
                }
940
            }
941
        } else {
942
        default_geometry:
943
            /* if no geometry, use a standard physical disk geometry */
944
            cylinders = nb_sectors / (16 * 63);
945

    
946
            if (cylinders > 16383)
947
                cylinders = 16383;
948
            else if (cylinders < 2)
949
                cylinders = 2;
950
            *pcyls = cylinders;
951
            *pheads = 16;
952
            *psecs = 63;
953
            if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
954
                if ((*pcyls * *pheads) <= 131072) {
955
                    bdrv_set_translation_hint(bs,
956
                                              BIOS_ATA_TRANSLATION_LARGE);
957
                } else {
958
                    bdrv_set_translation_hint(bs,
959
                                              BIOS_ATA_TRANSLATION_LBA);
960
                }
961
            }
962
        }
963
        bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
964
    }
965
}
966

    
967
void bdrv_set_geometry_hint(BlockDriverState *bs,
968
                            int cyls, int heads, int secs)
969
{
970
    bs->cyls = cyls;
971
    bs->heads = heads;
972
    bs->secs = secs;
973
}
974

    
975
void bdrv_set_type_hint(BlockDriverState *bs, int type)
976
{
977
    bs->type = type;
978
    bs->removable = ((type == BDRV_TYPE_CDROM ||
979
                      type == BDRV_TYPE_FLOPPY));
980
}
981

    
982
void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
983
{
984
    bs->translation = translation;
985
}
986

    
987
void bdrv_get_geometry_hint(BlockDriverState *bs,
988
                            int *pcyls, int *pheads, int *psecs)
989
{
990
    *pcyls = bs->cyls;
991
    *pheads = bs->heads;
992
    *psecs = bs->secs;
993
}
994

    
995
int bdrv_get_type_hint(BlockDriverState *bs)
996
{
997
    return bs->type;
998
}
999

    
1000
int bdrv_get_translation_hint(BlockDriverState *bs)
1001
{
1002
    return bs->translation;
1003
}
1004

    
1005
int bdrv_is_removable(BlockDriverState *bs)
1006
{
1007
    return bs->removable;
1008
}
1009

    
1010
int bdrv_is_read_only(BlockDriverState *bs)
1011
{
1012
    return bs->read_only;
1013
}
1014

    
1015
int bdrv_set_read_only(BlockDriverState *bs, int read_only)
1016
{
1017
    int ret = bs->read_only;
1018
    bs->read_only = read_only;
1019
    return ret;
1020
}
1021

    
1022
int bdrv_is_sg(BlockDriverState *bs)
1023
{
1024
    return bs->sg;
1025
}
1026

    
1027
int bdrv_enable_write_cache(BlockDriverState *bs)
1028
{
1029
    return bs->enable_write_cache;
1030
}
1031

    
1032
/* XXX: no longer used */
1033
void bdrv_set_change_cb(BlockDriverState *bs,
1034
                        void (*change_cb)(void *opaque), void *opaque)
1035
{
1036
    bs->change_cb = change_cb;
1037
    bs->change_opaque = opaque;
1038
}
1039

    
1040
int bdrv_is_encrypted(BlockDriverState *bs)
1041
{
1042
    if (bs->backing_hd && bs->backing_hd->encrypted)
1043
        return 1;
1044
    return bs->encrypted;
1045
}
1046

    
1047
int bdrv_key_required(BlockDriverState *bs)
1048
{
1049
    BlockDriverState *backing_hd = bs->backing_hd;
1050

    
1051
    if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1052
        return 1;
1053
    return (bs->encrypted && !bs->valid_key);
1054
}
1055

    
1056
int bdrv_set_key(BlockDriverState *bs, const char *key)
1057
{
1058
    int ret;
1059
    if (bs->backing_hd && bs->backing_hd->encrypted) {
1060
        ret = bdrv_set_key(bs->backing_hd, key);
1061
        if (ret < 0)
1062
            return ret;
1063
        if (!bs->encrypted)
1064
            return 0;
1065
    }
1066
    if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
1067
        return -1;
1068
    ret = bs->drv->bdrv_set_key(bs, key);
1069
    if (ret < 0) {
1070
        bs->valid_key = 0;
1071
    } else if (!bs->valid_key) {
1072
        bs->valid_key = 1;
1073
        /* call the change callback now, we skipped it on open */
1074
        bs->media_changed = 1;
1075
        if (bs->change_cb)
1076
            bs->change_cb(bs->change_opaque);
1077
    }
1078
    return ret;
1079
}
1080

    
1081
void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1082
{
1083
    if (!bs->drv) {
1084
        buf[0] = '\0';
1085
    } else {
1086
        pstrcpy(buf, buf_size, bs->drv->format_name);
1087
    }
1088
}
1089

    
1090
void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1091
                         void *opaque)
1092
{
1093
    BlockDriver *drv;
1094

    
1095
    for (drv = first_drv; drv != NULL; drv = drv->next) {
1096
        it(opaque, drv->format_name);
1097
    }
1098
}
1099

    
1100
BlockDriverState *bdrv_find(const char *name)
1101
{
1102
    BlockDriverState *bs;
1103

    
1104
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1105
        if (!strcmp(name, bs->device_name))
1106
            return bs;
1107
    }
1108
    return NULL;
1109
}
1110

    
1111
void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1112
{
1113
    BlockDriverState *bs;
1114

    
1115
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1116
        it(opaque, bs);
1117
    }
1118
}
1119

    
1120
const char *bdrv_get_device_name(BlockDriverState *bs)
1121
{
1122
    return bs->device_name;
1123
}
1124

    
1125
void bdrv_flush(BlockDriverState *bs)
1126
{
1127
    if (bs->drv && bs->drv->bdrv_flush)
1128
        bs->drv->bdrv_flush(bs);
1129
}
1130

    
1131
void bdrv_flush_all(void)
1132
{
1133
    BlockDriverState *bs;
1134

    
1135
    for (bs = bdrv_first; bs != NULL; bs = bs->next)
1136
        if (bs->drv && !bdrv_is_read_only(bs) && 
1137
            (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1138
            bdrv_flush(bs);
1139
}
1140

    
1141
/*
1142
 * Returns true iff the specified sector is present in the disk image. Drivers
1143
 * not implementing the functionality are assumed to not support backing files,
1144
 * hence all their sectors are reported as allocated.
1145
 *
1146
 * 'pnum' is set to the number of sectors (including and immediately following
1147
 * the specified sector) that are known to be in the same
1148
 * allocated/unallocated state.
1149
 *
1150
 * 'nb_sectors' is the max value 'pnum' should be set to.
1151
 */
1152
int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1153
        int *pnum)
1154
{
1155
    int64_t n;
1156
    if (!bs->drv->bdrv_is_allocated) {
1157
        if (sector_num >= bs->total_sectors) {
1158
            *pnum = 0;
1159
            return 0;
1160
        }
1161
        n = bs->total_sectors - sector_num;
1162
        *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1163
        return 1;
1164
    }
1165
    return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1166
}
1167

    
1168
static void bdrv_print_dict(QObject *obj, void *opaque)
1169
{
1170
    QDict *bs_dict;
1171
    Monitor *mon = opaque;
1172

    
1173
    bs_dict = qobject_to_qdict(obj);
1174

    
1175
    monitor_printf(mon, "%s: type=%s removable=%d",
1176
                        qdict_get_str(bs_dict, "device"),
1177
                        qdict_get_str(bs_dict, "type"),
1178
                        qdict_get_bool(bs_dict, "removable"));
1179

    
1180
    if (qdict_get_bool(bs_dict, "removable")) {
1181
        monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1182
    }
1183

    
1184
    if (qdict_haskey(bs_dict, "inserted")) {
1185
        QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1186

    
1187
        monitor_printf(mon, " file=");
1188
        monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1189
        if (qdict_haskey(qdict, "backing_file")) {
1190
            monitor_printf(mon, " backing_file=");
1191
            monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1192
        }
1193
        monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1194
                            qdict_get_bool(qdict, "ro"),
1195
                            qdict_get_str(qdict, "drv"),
1196
                            qdict_get_bool(qdict, "encrypted"));
1197
    } else {
1198
        monitor_printf(mon, " [not inserted]");
1199
    }
1200

    
1201
    monitor_printf(mon, "\n");
1202
}
1203

    
1204
void bdrv_info_print(Monitor *mon, const QObject *data)
1205
{
1206
    qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1207
}
1208

    
1209
/**
1210
 * bdrv_info(): Block devices information
1211
 *
1212
 * Each block device information is stored in a QDict and the
1213
 * returned QObject is a QList of all devices.
1214
 *
1215
 * The QDict contains the following:
1216
 *
1217
 * - "device": device name
1218
 * - "type": device type
1219
 * - "removable": true if the device is removable, false otherwise
1220
 * - "locked": true if the device is locked, false otherwise
1221
 * - "inserted": only present if the device is inserted, it is a QDict
1222
 *    containing the following:
1223
 *          - "file": device file name
1224
 *          - "ro": true if read-only, false otherwise
1225
 *          - "drv": driver format name
1226
 *          - "backing_file": backing file name if one is used
1227
 *          - "encrypted": true if encrypted, false otherwise
1228
 *
1229
 * Example:
1230
 *
1231
 * [ { "device": "ide0-hd0", "type": "hd", "removable": false, "locked": false,
1232
 *     "inserted": { "file": "/tmp/foobar", "ro": false, "drv": "qcow2" } },
1233
 *   { "device": "floppy0", "type": "floppy", "removable": true,
1234
 *     "locked": false } ]
1235
 */
1236
void bdrv_info(Monitor *mon, QObject **ret_data)
1237
{
1238
    QList *bs_list;
1239
    BlockDriverState *bs;
1240

    
1241
    bs_list = qlist_new();
1242

    
1243
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1244
        QObject *bs_obj;
1245
        const char *type = "unknown";
1246

    
1247
        switch(bs->type) {
1248
        case BDRV_TYPE_HD:
1249
            type = "hd";
1250
            break;
1251
        case BDRV_TYPE_CDROM:
1252
            type = "cdrom";
1253
            break;
1254
        case BDRV_TYPE_FLOPPY:
1255
            type = "floppy";
1256
            break;
1257
        }
1258

    
1259
        bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1260
                                    "'removable': %i, 'locked': %i }",
1261
                                    bs->device_name, type, bs->removable,
1262
                                    bs->locked);
1263
        assert(bs_obj != NULL);
1264

    
1265
        if (bs->drv) {
1266
            QObject *obj;
1267
            QDict *bs_dict = qobject_to_qdict(bs_obj);
1268

    
1269
            obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1270
                                     "'encrypted': %i }",
1271
                                     bs->filename, bs->read_only,
1272
                                     bs->drv->format_name,
1273
                                     bdrv_is_encrypted(bs));
1274
            assert(obj != NULL);
1275
            if (bs->backing_file[0] != '\0') {
1276
                QDict *qdict = qobject_to_qdict(obj);
1277
                qdict_put(qdict, "backing_file",
1278
                          qstring_from_str(bs->backing_file));
1279
            }
1280

    
1281
            qdict_put_obj(bs_dict, "inserted", obj);
1282
        }
1283
        qlist_append_obj(bs_list, bs_obj);
1284
    }
1285

    
1286
    *ret_data = QOBJECT(bs_list);
1287
}
1288

    
1289
static void bdrv_stats_iter(QObject *data, void *opaque)
1290
{
1291
    QDict *qdict;
1292
    Monitor *mon = opaque;
1293

    
1294
    qdict = qobject_to_qdict(data);
1295
    monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1296

    
1297
    qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1298
    monitor_printf(mon, " rd_bytes=%" PRId64
1299
                        " wr_bytes=%" PRId64
1300
                        " rd_operations=%" PRId64
1301
                        " wr_operations=%" PRId64
1302
                        "\n",
1303
                        qdict_get_int(qdict, "rd_bytes"),
1304
                        qdict_get_int(qdict, "wr_bytes"),
1305
                        qdict_get_int(qdict, "rd_operations"),
1306
                        qdict_get_int(qdict, "wr_operations"));
1307
}
1308

    
1309
void bdrv_stats_print(Monitor *mon, const QObject *data)
1310
{
1311
    qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1312
}
1313

    
1314
/**
1315
 * bdrv_info_stats(): show block device statistics
1316
 *
1317
 * Each device statistic information is stored in a QDict and
1318
 * the returned QObject is a QList of all devices.
1319
 *
1320
 * The QDict contains the following:
1321
 *
1322
 * - "device": device name
1323
 * - "stats": A QDict with the statistics information, it contains:
1324
 *     - "rd_bytes": bytes read
1325
 *     - "wr_bytes": bytes written
1326
 *     - "rd_operations": read operations
1327
 *     - "wr_operations": write operations
1328
 * 
1329
 * Example:
1330
 *
1331
 * [ { "device": "ide0-hd0",
1332
 *               "stats": { "rd_bytes": 512,
1333
 *                          "wr_bytes": 0,
1334
 *                          "rd_operations": 1,
1335
 *                          "wr_operations": 0 } },
1336
 *   { "device": "ide1-cd0",
1337
 *               "stats": { "rd_bytes": 0,
1338
 *                          "wr_bytes": 0,
1339
 *                          "rd_operations": 0,
1340
 *                          "wr_operations": 0 } } ]
1341
 */
1342
void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1343
{
1344
    QObject *obj;
1345
    QList *devices;
1346
    BlockDriverState *bs;
1347

    
1348
    devices = qlist_new();
1349

    
1350
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1351
        obj = qobject_from_jsonf("{ 'device': %s, 'stats': {"
1352
                                 "'rd_bytes': %" PRId64 ","
1353
                                 "'wr_bytes': %" PRId64 ","
1354
                                 "'rd_operations': %" PRId64 ","
1355
                                 "'wr_operations': %" PRId64
1356
                                 "} }",
1357
                                 bs->device_name,
1358
                                 bs->rd_bytes, bs->wr_bytes,
1359
                                 bs->rd_ops, bs->wr_ops);
1360
        assert(obj != NULL);
1361
        qlist_append_obj(devices, obj);
1362
    }
1363

    
1364
    *ret_data = QOBJECT(devices);
1365
}
1366

    
1367
const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1368
{
1369
    if (bs->backing_hd && bs->backing_hd->encrypted)
1370
        return bs->backing_file;
1371
    else if (bs->encrypted)
1372
        return bs->filename;
1373
    else
1374
        return NULL;
1375
}
1376

    
1377
void bdrv_get_backing_filename(BlockDriverState *bs,
1378
                               char *filename, int filename_size)
1379
{
1380
    if (!bs->backing_file) {
1381
        pstrcpy(filename, filename_size, "");
1382
    } else {
1383
        pstrcpy(filename, filename_size, bs->backing_file);
1384
    }
1385
}
1386

    
1387
int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1388
                          const uint8_t *buf, int nb_sectors)
1389
{
1390
    BlockDriver *drv = bs->drv;
1391
    if (!drv)
1392
        return -ENOMEDIUM;
1393
    if (!drv->bdrv_write_compressed)
1394
        return -ENOTSUP;
1395
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1396
        return -EIO;
1397

    
1398
    if (bs->dirty_bitmap) {
1399
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1400
    }
1401

    
1402
    return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1403
}
1404

    
1405
int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1406
{
1407
    BlockDriver *drv = bs->drv;
1408
    if (!drv)
1409
        return -ENOMEDIUM;
1410
    if (!drv->bdrv_get_info)
1411
        return -ENOTSUP;
1412
    memset(bdi, 0, sizeof(*bdi));
1413
    return drv->bdrv_get_info(bs, bdi);
1414
}
1415

    
1416
int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1417
                      int64_t pos, int size)
1418
{
1419
    BlockDriver *drv = bs->drv;
1420
    if (!drv)
1421
        return -ENOMEDIUM;
1422
    if (!drv->bdrv_save_vmstate)
1423
        return -ENOTSUP;
1424
    return drv->bdrv_save_vmstate(bs, buf, pos, size);
1425
}
1426

    
1427
int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1428
                      int64_t pos, int size)
1429
{
1430
    BlockDriver *drv = bs->drv;
1431
    if (!drv)
1432
        return -ENOMEDIUM;
1433
    if (!drv->bdrv_load_vmstate)
1434
        return -ENOTSUP;
1435
    return drv->bdrv_load_vmstate(bs, buf, pos, size);
1436
}
1437

    
1438
/**************************************************************/
1439
/* handling of snapshots */
1440

    
1441
int bdrv_snapshot_create(BlockDriverState *bs,
1442
                         QEMUSnapshotInfo *sn_info)
1443
{
1444
    BlockDriver *drv = bs->drv;
1445
    if (!drv)
1446
        return -ENOMEDIUM;
1447
    if (!drv->bdrv_snapshot_create)
1448
        return -ENOTSUP;
1449
    return drv->bdrv_snapshot_create(bs, sn_info);
1450
}
1451

    
1452
int bdrv_snapshot_goto(BlockDriverState *bs,
1453
                       const char *snapshot_id)
1454
{
1455
    BlockDriver *drv = bs->drv;
1456
    if (!drv)
1457
        return -ENOMEDIUM;
1458
    if (!drv->bdrv_snapshot_goto)
1459
        return -ENOTSUP;
1460
    return drv->bdrv_snapshot_goto(bs, snapshot_id);
1461
}
1462

    
1463
int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1464
{
1465
    BlockDriver *drv = bs->drv;
1466
    if (!drv)
1467
        return -ENOMEDIUM;
1468
    if (!drv->bdrv_snapshot_delete)
1469
        return -ENOTSUP;
1470
    return drv->bdrv_snapshot_delete(bs, snapshot_id);
1471
}
1472

    
1473
int bdrv_snapshot_list(BlockDriverState *bs,
1474
                       QEMUSnapshotInfo **psn_info)
1475
{
1476
    BlockDriver *drv = bs->drv;
1477
    if (!drv)
1478
        return -ENOMEDIUM;
1479
    if (!drv->bdrv_snapshot_list)
1480
        return -ENOTSUP;
1481
    return drv->bdrv_snapshot_list(bs, psn_info);
1482
}
1483

    
1484
#define NB_SUFFIXES 4
1485

    
1486
char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1487
{
1488
    static const char suffixes[NB_SUFFIXES] = "KMGT";
1489
    int64_t base;
1490
    int i;
1491

    
1492
    if (size <= 999) {
1493
        snprintf(buf, buf_size, "%" PRId64, size);
1494
    } else {
1495
        base = 1024;
1496
        for(i = 0; i < NB_SUFFIXES; i++) {
1497
            if (size < (10 * base)) {
1498
                snprintf(buf, buf_size, "%0.1f%c",
1499
                         (double)size / base,
1500
                         suffixes[i]);
1501
                break;
1502
            } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1503
                snprintf(buf, buf_size, "%" PRId64 "%c",
1504
                         ((size + (base >> 1)) / base),
1505
                         suffixes[i]);
1506
                break;
1507
            }
1508
            base = base * 1024;
1509
        }
1510
    }
1511
    return buf;
1512
}
1513

    
1514
char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1515
{
1516
    char buf1[128], date_buf[128], clock_buf[128];
1517
#ifdef _WIN32
1518
    struct tm *ptm;
1519
#else
1520
    struct tm tm;
1521
#endif
1522
    time_t ti;
1523
    int64_t secs;
1524

    
1525
    if (!sn) {
1526
        snprintf(buf, buf_size,
1527
                 "%-10s%-20s%7s%20s%15s",
1528
                 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1529
    } else {
1530
        ti = sn->date_sec;
1531
#ifdef _WIN32
1532
        ptm = localtime(&ti);
1533
        strftime(date_buf, sizeof(date_buf),
1534
                 "%Y-%m-%d %H:%M:%S", ptm);
1535
#else
1536
        localtime_r(&ti, &tm);
1537
        strftime(date_buf, sizeof(date_buf),
1538
                 "%Y-%m-%d %H:%M:%S", &tm);
1539
#endif
1540
        secs = sn->vm_clock_nsec / 1000000000;
1541
        snprintf(clock_buf, sizeof(clock_buf),
1542
                 "%02d:%02d:%02d.%03d",
1543
                 (int)(secs / 3600),
1544
                 (int)((secs / 60) % 60),
1545
                 (int)(secs % 60),
1546
                 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1547
        snprintf(buf, buf_size,
1548
                 "%-10s%-20s%7s%20s%15s",
1549
                 sn->id_str, sn->name,
1550
                 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1551
                 date_buf,
1552
                 clock_buf);
1553
    }
1554
    return buf;
1555
}
1556

    
1557

    
1558
/**************************************************************/
1559
/* async I/Os */
1560

    
1561
BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1562
                                 QEMUIOVector *qiov, int nb_sectors,
1563
                                 BlockDriverCompletionFunc *cb, void *opaque)
1564
{
1565
    BlockDriver *drv = bs->drv;
1566
    BlockDriverAIOCB *ret;
1567

    
1568
    if (!drv)
1569
        return NULL;
1570
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1571
        return NULL;
1572

    
1573
    ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1574
                              cb, opaque);
1575

    
1576
    if (ret) {
1577
        /* Update stats even though technically transfer has not happened. */
1578
        bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1579
        bs->rd_ops ++;
1580
    }
1581

    
1582
    return ret;
1583
}
1584

    
1585
BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1586
                                  QEMUIOVector *qiov, int nb_sectors,
1587
                                  BlockDriverCompletionFunc *cb, void *opaque)
1588
{
1589
    BlockDriver *drv = bs->drv;
1590
    BlockDriverAIOCB *ret;
1591

    
1592
    if (!drv)
1593
        return NULL;
1594
    if (bs->read_only)
1595
        return NULL;
1596
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1597
        return NULL;
1598

    
1599
    if (bs->dirty_bitmap) {
1600
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1601
    }
1602

    
1603
    ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
1604
                               cb, opaque);
1605

    
1606
    if (ret) {
1607
        /* Update stats even though technically transfer has not happened. */
1608
        bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1609
        bs->wr_ops ++;
1610
    }
1611

    
1612
    return ret;
1613
}
1614

    
1615

    
1616
typedef struct MultiwriteCB {
1617
    int error;
1618
    int num_requests;
1619
    int num_callbacks;
1620
    struct {
1621
        BlockDriverCompletionFunc *cb;
1622
        void *opaque;
1623
        QEMUIOVector *free_qiov;
1624
        void *free_buf;
1625
    } callbacks[];
1626
} MultiwriteCB;
1627

    
1628
static void multiwrite_user_cb(MultiwriteCB *mcb)
1629
{
1630
    int i;
1631

    
1632
    for (i = 0; i < mcb->num_callbacks; i++) {
1633
        mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
1634
        qemu_free(mcb->callbacks[i].free_qiov);
1635
        qemu_free(mcb->callbacks[i].free_buf);
1636
    }
1637
}
1638

    
1639
static void multiwrite_cb(void *opaque, int ret)
1640
{
1641
    MultiwriteCB *mcb = opaque;
1642

    
1643
    if (ret < 0) {
1644
        mcb->error = ret;
1645
        multiwrite_user_cb(mcb);
1646
    }
1647

    
1648
    mcb->num_requests--;
1649
    if (mcb->num_requests == 0) {
1650
        if (mcb->error == 0) {
1651
            multiwrite_user_cb(mcb);
1652
        }
1653
        qemu_free(mcb);
1654
    }
1655
}
1656

    
1657
static int multiwrite_req_compare(const void *a, const void *b)
1658
{
1659
    return (((BlockRequest*) a)->sector - ((BlockRequest*) b)->sector);
1660
}
1661

    
1662
/*
1663
 * Takes a bunch of requests and tries to merge them. Returns the number of
1664
 * requests that remain after merging.
1665
 */
1666
static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
1667
    int num_reqs, MultiwriteCB *mcb)
1668
{
1669
    int i, outidx;
1670

    
1671
    // Sort requests by start sector
1672
    qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
1673

    
1674
    // Check if adjacent requests touch the same clusters. If so, combine them,
1675
    // filling up gaps with zero sectors.
1676
    outidx = 0;
1677
    for (i = 1; i < num_reqs; i++) {
1678
        int merge = 0;
1679
        int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
1680

    
1681
        // This handles the cases that are valid for all block drivers, namely
1682
        // exactly sequential writes and overlapping writes.
1683
        if (reqs[i].sector <= oldreq_last) {
1684
            merge = 1;
1685
        }
1686

    
1687
        // The block driver may decide that it makes sense to combine requests
1688
        // even if there is a gap of some sectors between them. In this case,
1689
        // the gap is filled with zeros (therefore only applicable for yet
1690
        // unused space in format like qcow2).
1691
        if (!merge && bs->drv->bdrv_merge_requests) {
1692
            merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
1693
        }
1694

    
1695
        if (merge) {
1696
            size_t size;
1697
            QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
1698
            qemu_iovec_init(qiov,
1699
                reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
1700

    
1701
            // Add the first request to the merged one. If the requests are
1702
            // overlapping, drop the last sectors of the first request.
1703
            size = (reqs[i].sector - reqs[outidx].sector) << 9;
1704
            qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
1705

    
1706
            // We might need to add some zeros between the two requests
1707
            if (reqs[i].sector > oldreq_last) {
1708
                size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
1709
                uint8_t *buf = qemu_blockalign(bs, zero_bytes);
1710
                memset(buf, 0, zero_bytes);
1711
                qemu_iovec_add(qiov, buf, zero_bytes);
1712
                mcb->callbacks[i].free_buf = buf;
1713
            }
1714

    
1715
            // Add the second request
1716
            qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
1717

    
1718
            reqs[outidx].nb_sectors += reqs[i].nb_sectors;
1719
            reqs[outidx].qiov = qiov;
1720

    
1721
            mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
1722
        } else {
1723
            outidx++;
1724
            reqs[outidx].sector     = reqs[i].sector;
1725
            reqs[outidx].nb_sectors = reqs[i].nb_sectors;
1726
            reqs[outidx].qiov       = reqs[i].qiov;
1727
        }
1728
    }
1729

    
1730
    return outidx + 1;
1731
}
1732

    
1733
/*
1734
 * Submit multiple AIO write requests at once.
1735
 *
1736
 * On success, the function returns 0 and all requests in the reqs array have
1737
 * been submitted. In error case this function returns -1, and any of the
1738
 * requests may or may not be submitted yet. In particular, this means that the
1739
 * callback will be called for some of the requests, for others it won't. The
1740
 * caller must check the error field of the BlockRequest to wait for the right
1741
 * callbacks (if error != 0, no callback will be called).
1742
 *
1743
 * The implementation may modify the contents of the reqs array, e.g. to merge
1744
 * requests. However, the fields opaque and error are left unmodified as they
1745
 * are used to signal failure for a single request to the caller.
1746
 */
1747
int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
1748
{
1749
    BlockDriverAIOCB *acb;
1750
    MultiwriteCB *mcb;
1751
    int i;
1752

    
1753
    if (num_reqs == 0) {
1754
        return 0;
1755
    }
1756

    
1757
    // Create MultiwriteCB structure
1758
    mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
1759
    mcb->num_requests = 0;
1760
    mcb->num_callbacks = num_reqs;
1761

    
1762
    for (i = 0; i < num_reqs; i++) {
1763
        mcb->callbacks[i].cb = reqs[i].cb;
1764
        mcb->callbacks[i].opaque = reqs[i].opaque;
1765
    }
1766

    
1767
    // Check for mergable requests
1768
    num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
1769

    
1770
    // Run the aio requests
1771
    for (i = 0; i < num_reqs; i++) {
1772
        acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
1773
            reqs[i].nb_sectors, multiwrite_cb, mcb);
1774

    
1775
        if (acb == NULL) {
1776
            // We can only fail the whole thing if no request has been
1777
            // submitted yet. Otherwise we'll wait for the submitted AIOs to
1778
            // complete and report the error in the callback.
1779
            if (mcb->num_requests == 0) {
1780
                reqs[i].error = EIO;
1781
                goto fail;
1782
            } else {
1783
                mcb->error = EIO;
1784
                break;
1785
            }
1786
        } else {
1787
            mcb->num_requests++;
1788
        }
1789
    }
1790

    
1791
    return 0;
1792

    
1793
fail:
1794
    free(mcb);
1795
    return -1;
1796
}
1797

    
1798
BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
1799
        BlockDriverCompletionFunc *cb, void *opaque)
1800
{
1801
    BlockDriver *drv = bs->drv;
1802

    
1803
    if (!drv)
1804
        return NULL;
1805
    return drv->bdrv_aio_flush(bs, cb, opaque);
1806
}
1807

    
1808
void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1809
{
1810
    acb->pool->cancel(acb);
1811
}
1812

    
1813

    
1814
/**************************************************************/
1815
/* async block device emulation */
1816

    
1817
typedef struct BlockDriverAIOCBSync {
1818
    BlockDriverAIOCB common;
1819
    QEMUBH *bh;
1820
    int ret;
1821
    /* vector translation state */
1822
    QEMUIOVector *qiov;
1823
    uint8_t *bounce;
1824
    int is_write;
1825
} BlockDriverAIOCBSync;
1826

    
1827
static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1828
{
1829
    BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1830
    qemu_bh_delete(acb->bh);
1831
    acb->bh = NULL;
1832
    qemu_aio_release(acb);
1833
}
1834

    
1835
static AIOPool bdrv_em_aio_pool = {
1836
    .aiocb_size         = sizeof(BlockDriverAIOCBSync),
1837
    .cancel             = bdrv_aio_cancel_em,
1838
};
1839

    
1840
static void bdrv_aio_bh_cb(void *opaque)
1841
{
1842
    BlockDriverAIOCBSync *acb = opaque;
1843

    
1844
    if (!acb->is_write)
1845
        qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
1846
    qemu_vfree(acb->bounce);
1847
    acb->common.cb(acb->common.opaque, acb->ret);
1848
    qemu_bh_delete(acb->bh);
1849
    acb->bh = NULL;
1850
    qemu_aio_release(acb);
1851
}
1852

    
1853
static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
1854
                                            int64_t sector_num,
1855
                                            QEMUIOVector *qiov,
1856
                                            int nb_sectors,
1857
                                            BlockDriverCompletionFunc *cb,
1858
                                            void *opaque,
1859
                                            int is_write)
1860

    
1861
{
1862
    BlockDriverAIOCBSync *acb;
1863

    
1864
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
1865
    acb->is_write = is_write;
1866
    acb->qiov = qiov;
1867
    acb->bounce = qemu_blockalign(bs, qiov->size);
1868

    
1869
    if (!acb->bh)
1870
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1871

    
1872
    if (is_write) {
1873
        qemu_iovec_to_buffer(acb->qiov, acb->bounce);
1874
        acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
1875
    } else {
1876
        acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
1877
    }
1878

    
1879
    qemu_bh_schedule(acb->bh);
1880

    
1881
    return &acb->common;
1882
}
1883

    
1884
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
1885
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1886
        BlockDriverCompletionFunc *cb, void *opaque)
1887
{
1888
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
1889
}
1890

    
1891
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
1892
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1893
        BlockDriverCompletionFunc *cb, void *opaque)
1894
{
1895
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
1896
}
1897

    
1898
static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
1899
        BlockDriverCompletionFunc *cb, void *opaque)
1900
{
1901
    BlockDriverAIOCBSync *acb;
1902

    
1903
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
1904
    acb->is_write = 1; /* don't bounce in the completion hadler */
1905
    acb->qiov = NULL;
1906
    acb->bounce = NULL;
1907
    acb->ret = 0;
1908

    
1909
    if (!acb->bh)
1910
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1911

    
1912
    bdrv_flush(bs);
1913
    qemu_bh_schedule(acb->bh);
1914
    return &acb->common;
1915
}
1916

    
1917
/**************************************************************/
1918
/* sync block device emulation */
1919

    
1920
static void bdrv_rw_em_cb(void *opaque, int ret)
1921
{
1922
    *(int *)opaque = ret;
1923
}
1924

    
1925
#define NOT_DONE 0x7fffffff
1926

    
1927
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1928
                        uint8_t *buf, int nb_sectors)
1929
{
1930
    int async_ret;
1931
    BlockDriverAIOCB *acb;
1932
    struct iovec iov;
1933
    QEMUIOVector qiov;
1934

    
1935
    async_context_push();
1936

    
1937
    async_ret = NOT_DONE;
1938
    iov.iov_base = (void *)buf;
1939
    iov.iov_len = nb_sectors * 512;
1940
    qemu_iovec_init_external(&qiov, &iov, 1);
1941
    acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
1942
        bdrv_rw_em_cb, &async_ret);
1943
    if (acb == NULL) {
1944
        async_ret = -1;
1945
        goto fail;
1946
    }
1947

    
1948
    while (async_ret == NOT_DONE) {
1949
        qemu_aio_wait();
1950
    }
1951

    
1952

    
1953
fail:
1954
    async_context_pop();
1955
    return async_ret;
1956
}
1957

    
1958
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1959
                         const uint8_t *buf, int nb_sectors)
1960
{
1961
    int async_ret;
1962
    BlockDriverAIOCB *acb;
1963
    struct iovec iov;
1964
    QEMUIOVector qiov;
1965

    
1966
    async_context_push();
1967

    
1968
    async_ret = NOT_DONE;
1969
    iov.iov_base = (void *)buf;
1970
    iov.iov_len = nb_sectors * 512;
1971
    qemu_iovec_init_external(&qiov, &iov, 1);
1972
    acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
1973
        bdrv_rw_em_cb, &async_ret);
1974
    if (acb == NULL) {
1975
        async_ret = -1;
1976
        goto fail;
1977
    }
1978
    while (async_ret == NOT_DONE) {
1979
        qemu_aio_wait();
1980
    }
1981

    
1982
fail:
1983
    async_context_pop();
1984
    return async_ret;
1985
}
1986

    
1987
void bdrv_init(void)
1988
{
1989
    module_call_init(MODULE_INIT_BLOCK);
1990
}
1991

    
1992
void bdrv_init_with_whitelist(void)
1993
{
1994
    use_bdrv_whitelist = 1;
1995
    bdrv_init();
1996
}
1997

    
1998
void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
1999
                   BlockDriverCompletionFunc *cb, void *opaque)
2000
{
2001
    BlockDriverAIOCB *acb;
2002

    
2003
    if (pool->free_aiocb) {
2004
        acb = pool->free_aiocb;
2005
        pool->free_aiocb = acb->next;
2006
    } else {
2007
        acb = qemu_mallocz(pool->aiocb_size);
2008
        acb->pool = pool;
2009
    }
2010
    acb->bs = bs;
2011
    acb->cb = cb;
2012
    acb->opaque = opaque;
2013
    return acb;
2014
}
2015

    
2016
void qemu_aio_release(void *p)
2017
{
2018
    BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2019
    AIOPool *pool = acb->pool;
2020
    acb->next = pool->free_aiocb;
2021
    pool->free_aiocb = acb;
2022
}
2023

    
2024
/**************************************************************/
2025
/* removable device support */
2026

    
2027
/**
2028
 * Return TRUE if the media is present
2029
 */
2030
int bdrv_is_inserted(BlockDriverState *bs)
2031
{
2032
    BlockDriver *drv = bs->drv;
2033
    int ret;
2034
    if (!drv)
2035
        return 0;
2036
    if (!drv->bdrv_is_inserted)
2037
        return 1;
2038
    ret = drv->bdrv_is_inserted(bs);
2039
    return ret;
2040
}
2041

    
2042
/**
2043
 * Return TRUE if the media changed since the last call to this
2044
 * function. It is currently only used for floppy disks
2045
 */
2046
int bdrv_media_changed(BlockDriverState *bs)
2047
{
2048
    BlockDriver *drv = bs->drv;
2049
    int ret;
2050

    
2051
    if (!drv || !drv->bdrv_media_changed)
2052
        ret = -ENOTSUP;
2053
    else
2054
        ret = drv->bdrv_media_changed(bs);
2055
    if (ret == -ENOTSUP)
2056
        ret = bs->media_changed;
2057
    bs->media_changed = 0;
2058
    return ret;
2059
}
2060

    
2061
/**
2062
 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2063
 */
2064
int bdrv_eject(BlockDriverState *bs, int eject_flag)
2065
{
2066
    BlockDriver *drv = bs->drv;
2067
    int ret;
2068

    
2069
    if (bs->locked) {
2070
        return -EBUSY;
2071
    }
2072

    
2073
    if (!drv || !drv->bdrv_eject) {
2074
        ret = -ENOTSUP;
2075
    } else {
2076
        ret = drv->bdrv_eject(bs, eject_flag);
2077
    }
2078
    if (ret == -ENOTSUP) {
2079
        if (eject_flag)
2080
            bdrv_close(bs);
2081
        ret = 0;
2082
    }
2083

    
2084
    return ret;
2085
}
2086

    
2087
int bdrv_is_locked(BlockDriverState *bs)
2088
{
2089
    return bs->locked;
2090
}
2091

    
2092
/**
2093
 * Lock or unlock the media (if it is locked, the user won't be able
2094
 * to eject it manually).
2095
 */
2096
void bdrv_set_locked(BlockDriverState *bs, int locked)
2097
{
2098
    BlockDriver *drv = bs->drv;
2099

    
2100
    bs->locked = locked;
2101
    if (drv && drv->bdrv_set_locked) {
2102
        drv->bdrv_set_locked(bs, locked);
2103
    }
2104
}
2105

    
2106
/* needed for generic scsi interface */
2107

    
2108
int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2109
{
2110
    BlockDriver *drv = bs->drv;
2111

    
2112
    if (drv && drv->bdrv_ioctl)
2113
        return drv->bdrv_ioctl(bs, req, buf);
2114
    return -ENOTSUP;
2115
}
2116

    
2117
BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2118
        unsigned long int req, void *buf,
2119
        BlockDriverCompletionFunc *cb, void *opaque)
2120
{
2121
    BlockDriver *drv = bs->drv;
2122

    
2123
    if (drv && drv->bdrv_aio_ioctl)
2124
        return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2125
    return NULL;
2126
}
2127

    
2128

    
2129

    
2130
void *qemu_blockalign(BlockDriverState *bs, size_t size)
2131
{
2132
    return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2133
}
2134

    
2135
void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2136
{
2137
    int64_t bitmap_size;
2138

    
2139
    if (enable) {
2140
        if (!bs->dirty_bitmap) {
2141
            bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2142
                    BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2143
            bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2144

    
2145
            bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2146
        }
2147
    } else {
2148
        if (bs->dirty_bitmap) {
2149
            qemu_free(bs->dirty_bitmap);
2150
            bs->dirty_bitmap = NULL;
2151
        }
2152
    }
2153
}
2154

    
2155
int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2156
{
2157
    int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2158

    
2159
    if (bs->dirty_bitmap &&
2160
        (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2161
        return bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2162
            (1 << (chunk % (sizeof(unsigned long) * 8)));
2163
    } else {
2164
        return 0;
2165
    }
2166
}
2167

    
2168
void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2169
                      int nr_sectors)
2170
{
2171
    set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2172
}