Statistics
| Branch: | Revision:

root / block.c @ 5ffbbc67

History | View | Annotate | Download (66.4 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 BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
54
        BlockDriverCompletionFunc *cb, void *opaque);
55
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
56
                        uint8_t *buf, int nb_sectors);
57
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
58
                         const uint8_t *buf, int nb_sectors);
59

    
60
static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
61
    QTAILQ_HEAD_INITIALIZER(bdrv_states);
62

    
63
static QLIST_HEAD(, BlockDriver) bdrv_drivers =
64
    QLIST_HEAD_INITIALIZER(bdrv_drivers);
65

    
66
/* If non-zero, use only whitelisted block drivers */
67
static int use_bdrv_whitelist;
68

    
69
int path_is_absolute(const char *path)
70
{
71
    const char *p;
72
#ifdef _WIN32
73
    /* specific case for names like: "\\.\d:" */
74
    if (*path == '/' || *path == '\\')
75
        return 1;
76
#endif
77
    p = strchr(path, ':');
78
    if (p)
79
        p++;
80
    else
81
        p = path;
82
#ifdef _WIN32
83
    return (*p == '/' || *p == '\\');
84
#else
85
    return (*p == '/');
86
#endif
87
}
88

    
89
/* if filename is absolute, just copy it to dest. Otherwise, build a
90
   path to it by considering it is relative to base_path. URL are
91
   supported. */
92
void path_combine(char *dest, int dest_size,
93
                  const char *base_path,
94
                  const char *filename)
95
{
96
    const char *p, *p1;
97
    int len;
98

    
99
    if (dest_size <= 0)
100
        return;
101
    if (path_is_absolute(filename)) {
102
        pstrcpy(dest, dest_size, filename);
103
    } else {
104
        p = strchr(base_path, ':');
105
        if (p)
106
            p++;
107
        else
108
            p = base_path;
109
        p1 = strrchr(base_path, '/');
110
#ifdef _WIN32
111
        {
112
            const char *p2;
113
            p2 = strrchr(base_path, '\\');
114
            if (!p1 || p2 > p1)
115
                p1 = p2;
116
        }
117
#endif
118
        if (p1)
119
            p1++;
120
        else
121
            p1 = base_path;
122
        if (p1 > p)
123
            p = p1;
124
        len = p - base_path;
125
        if (len > dest_size - 1)
126
            len = dest_size - 1;
127
        memcpy(dest, base_path, len);
128
        dest[len] = '\0';
129
        pstrcat(dest, dest_size, filename);
130
    }
131
}
132

    
133
void bdrv_register(BlockDriver *bdrv)
134
{
135
    if (!bdrv->bdrv_aio_readv) {
136
        /* add AIO emulation layer */
137
        bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
138
        bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
139
    } else if (!bdrv->bdrv_read) {
140
        /* add synchronous IO emulation layer */
141
        bdrv->bdrv_read = bdrv_read_em;
142
        bdrv->bdrv_write = bdrv_write_em;
143
    }
144

    
145
    if (!bdrv->bdrv_aio_flush)
146
        bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
147

    
148
    QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
149
}
150

    
151
/* create a new block device (by default it is empty) */
152
BlockDriverState *bdrv_new(const char *device_name)
153
{
154
    BlockDriverState *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
        QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
160
    }
161
    return bs;
162
}
163

    
164
BlockDriver *bdrv_find_format(const char *format_name)
165
{
166
    BlockDriver *drv1;
167
    QLIST_FOREACH(drv1, &bdrv_drivers, list) {
168
        if (!strcmp(drv1->format_name, format_name)) {
169
            return drv1;
170
        }
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
int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
209
{
210
    BlockDriver *drv;
211

    
212
    drv = bdrv_find_protocol(filename);
213
    if (drv == NULL) {
214
        drv = bdrv_find_format("file");
215
    }
216

    
217
    return bdrv_create(drv, filename, options);
218
}
219

    
220
#ifdef _WIN32
221
void get_tmp_filename(char *filename, int size)
222
{
223
    char temp_dir[MAX_PATH];
224

    
225
    GetTempPath(MAX_PATH, temp_dir);
226
    GetTempFileName(temp_dir, "qem", 0, filename);
227
}
228
#else
229
void get_tmp_filename(char *filename, int size)
230
{
231
    int fd;
232
    const char *tmpdir;
233
    /* XXX: race condition possible */
234
    tmpdir = getenv("TMPDIR");
235
    if (!tmpdir)
236
        tmpdir = "/tmp";
237
    snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
238
    fd = mkstemp(filename);
239
    close(fd);
240
}
241
#endif
242

    
243
#ifdef _WIN32
244
static int is_windows_drive_prefix(const char *filename)
245
{
246
    return (((filename[0] >= 'a' && filename[0] <= 'z') ||
247
             (filename[0] >= 'A' && filename[0] <= 'Z')) &&
248
            filename[1] == ':');
249
}
250

    
251
int is_windows_drive(const char *filename)
252
{
253
    if (is_windows_drive_prefix(filename) &&
254
        filename[2] == '\0')
255
        return 1;
256
    if (strstart(filename, "\\\\.\\", NULL) ||
257
        strstart(filename, "//./", NULL))
258
        return 1;
259
    return 0;
260
}
261
#endif
262

    
263
/*
264
 * Detect host devices. By convention, /dev/cdrom[N] is always
265
 * recognized as a host CDROM.
266
 */
267
static BlockDriver *find_hdev_driver(const char *filename)
268
{
269
    int score_max = 0, score;
270
    BlockDriver *drv = NULL, *d;
271

    
272
    QLIST_FOREACH(d, &bdrv_drivers, list) {
273
        if (d->bdrv_probe_device) {
274
            score = d->bdrv_probe_device(filename);
275
            if (score > score_max) {
276
                score_max = score;
277
                drv = d;
278
            }
279
        }
280
    }
281

    
282
    return drv;
283
}
284

    
285
BlockDriver *bdrv_find_protocol(const char *filename)
286
{
287
    BlockDriver *drv1;
288
    char protocol[128];
289
    int len;
290
    const char *p;
291
    int is_drive;
292

    
293
    /* TODO Drivers without bdrv_file_open must be specified explicitly */
294

    
295
#ifdef _WIN32
296
    is_drive = is_windows_drive(filename) ||
297
        is_windows_drive_prefix(filename);
298
#else
299
    is_drive = 0;
300
#endif
301
    p = strchr(filename, ':');
302
    if (!p || is_drive) {
303
        drv1 = find_hdev_driver(filename);
304
        if (!drv1) {
305
            drv1 = bdrv_find_format("file");
306
        }
307
        return drv1;
308
    }
309
    len = p - filename;
310
    if (len > sizeof(protocol) - 1)
311
        len = sizeof(protocol) - 1;
312
    memcpy(protocol, filename, len);
313
    protocol[len] = '\0';
314
    QLIST_FOREACH(drv1, &bdrv_drivers, list) {
315
        if (drv1->protocol_name &&
316
            !strcmp(drv1->protocol_name, protocol)) {
317
            return drv1;
318
        }
319
    }
320
    return NULL;
321
}
322

    
323
static BlockDriver *find_image_format(const char *filename)
324
{
325
    int ret, score, score_max;
326
    BlockDriver *drv1, *drv;
327
    uint8_t buf[2048];
328
    BlockDriverState *bs;
329

    
330
    ret = bdrv_file_open(&bs, filename, 0);
331
    if (ret < 0)
332
        return NULL;
333

    
334
    /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
335
    if (bs->sg || !bdrv_is_inserted(bs)) {
336
        bdrv_delete(bs);
337
        return bdrv_find_format("raw");
338
    }
339

    
340
    ret = bdrv_pread(bs, 0, buf, sizeof(buf));
341
    bdrv_delete(bs);
342
    if (ret < 0) {
343
        return NULL;
344
    }
345

    
346
    score_max = 0;
347
    drv = NULL;
348
    QLIST_FOREACH(drv1, &bdrv_drivers, list) {
349
        if (drv1->bdrv_probe) {
350
            score = drv1->bdrv_probe(buf, ret, filename);
351
            if (score > score_max) {
352
                score_max = score;
353
                drv = drv1;
354
            }
355
        }
356
    }
357
    return drv;
358
}
359

    
360
/**
361
 * Set the current 'total_sectors' value
362
 */
363
static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
364
{
365
    BlockDriver *drv = bs->drv;
366

    
367
    /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
368
    if (bs->sg)
369
        return 0;
370

    
371
    /* query actual device if possible, otherwise just trust the hint */
372
    if (drv->bdrv_getlength) {
373
        int64_t length = drv->bdrv_getlength(bs);
374
        if (length < 0) {
375
            return length;
376
        }
377
        hint = length >> BDRV_SECTOR_BITS;
378
    }
379

    
380
    bs->total_sectors = hint;
381
    return 0;
382
}
383

    
384
/*
385
 * Common part for opening disk images and files
386
 */
387
static int bdrv_open_common(BlockDriverState *bs, const char *filename,
388
    int flags, BlockDriver *drv)
389
{
390
    int ret, open_flags;
391

    
392
    assert(drv != NULL);
393

    
394
    bs->file = NULL;
395
    bs->total_sectors = 0;
396
    bs->is_temporary = 0;
397
    bs->encrypted = 0;
398
    bs->valid_key = 0;
399
    bs->open_flags = flags;
400
    /* buffer_alignment defaulted to 512, drivers can change this value */
401
    bs->buffer_alignment = 512;
402

    
403
    pstrcpy(bs->filename, sizeof(bs->filename), filename);
404

    
405
    if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
406
        return -ENOTSUP;
407
    }
408

    
409
    bs->drv = drv;
410
    bs->opaque = qemu_mallocz(drv->instance_size);
411

    
412
    /*
413
     * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
414
     * write cache to the guest.  We do need the fdatasync to flush
415
     * out transactions for block allocations, and we maybe have a
416
     * volatile write cache in our backing device to deal with.
417
     */
418
    if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
419
        bs->enable_write_cache = 1;
420

    
421
    /*
422
     * Clear flags that are internal to the block layer before opening the
423
     * image.
424
     */
425
    open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
426

    
427
    /*
428
     * Snapshots should be writeable.
429
     */
430
    if (bs->is_temporary) {
431
        open_flags |= BDRV_O_RDWR;
432
    }
433

    
434
    /* Open the image, either directly or using a protocol */
435
    if (drv->bdrv_file_open) {
436
        ret = drv->bdrv_file_open(bs, filename, open_flags);
437
    } else {
438
        ret = bdrv_file_open(&bs->file, filename, open_flags);
439
        if (ret >= 0) {
440
            ret = drv->bdrv_open(bs, open_flags);
441
        }
442
    }
443

    
444
    if (ret < 0) {
445
        goto free_and_fail;
446
    }
447

    
448
    bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
449

    
450
    ret = refresh_total_sectors(bs, bs->total_sectors);
451
    if (ret < 0) {
452
        goto free_and_fail;
453
    }
454

    
455
#ifndef _WIN32
456
    if (bs->is_temporary) {
457
        unlink(filename);
458
    }
459
#endif
460
    return 0;
461

    
462
free_and_fail:
463
    if (bs->file) {
464
        bdrv_delete(bs->file);
465
        bs->file = NULL;
466
    }
467
    qemu_free(bs->opaque);
468
    bs->opaque = NULL;
469
    bs->drv = NULL;
470
    return ret;
471
}
472

    
473
/*
474
 * Opens a file using a protocol (file, host_device, nbd, ...)
475
 */
476
int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
477
{
478
    BlockDriverState *bs;
479
    BlockDriver *drv;
480
    int ret;
481

    
482
    drv = bdrv_find_protocol(filename);
483
    if (!drv) {
484
        return -ENOENT;
485
    }
486

    
487
    bs = bdrv_new("");
488
    ret = bdrv_open_common(bs, filename, flags, drv);
489
    if (ret < 0) {
490
        bdrv_delete(bs);
491
        return ret;
492
    }
493
    bs->growable = 1;
494
    *pbs = bs;
495
    return 0;
496
}
497

    
498
/*
499
 * Opens a disk image (raw, qcow2, vmdk, ...)
500
 */
501
int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
502
              BlockDriver *drv)
503
{
504
    int ret;
505

    
506
    if (flags & BDRV_O_SNAPSHOT) {
507
        BlockDriverState *bs1;
508
        int64_t total_size;
509
        int is_protocol = 0;
510
        BlockDriver *bdrv_qcow2;
511
        QEMUOptionParameter *options;
512
        char tmp_filename[PATH_MAX];
513
        char backing_filename[PATH_MAX];
514

    
515
        /* if snapshot, we create a temporary backing file and open it
516
           instead of opening 'filename' directly */
517

    
518
        /* if there is a backing file, use it */
519
        bs1 = bdrv_new("");
520
        ret = bdrv_open(bs1, filename, 0, drv);
521
        if (ret < 0) {
522
            bdrv_delete(bs1);
523
            return ret;
524
        }
525
        total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
526

    
527
        if (bs1->drv && bs1->drv->protocol_name)
528
            is_protocol = 1;
529

    
530
        bdrv_delete(bs1);
531

    
532
        get_tmp_filename(tmp_filename, sizeof(tmp_filename));
533

    
534
        /* Real path is meaningless for protocols */
535
        if (is_protocol)
536
            snprintf(backing_filename, sizeof(backing_filename),
537
                     "%s", filename);
538
        else if (!realpath(filename, backing_filename))
539
            return -errno;
540

    
541
        bdrv_qcow2 = bdrv_find_format("qcow2");
542
        options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
543

    
544
        set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
545
        set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
546
        if (drv) {
547
            set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
548
                drv->format_name);
549
        }
550

    
551
        ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
552
        free_option_parameters(options);
553
        if (ret < 0) {
554
            return ret;
555
        }
556

    
557
        filename = tmp_filename;
558
        drv = bdrv_qcow2;
559
        bs->is_temporary = 1;
560
    }
561

    
562
    /* Find the right image format driver */
563
    if (!drv) {
564
        drv = find_image_format(filename);
565
    }
566

    
567
    if (!drv) {
568
        ret = -ENOENT;
569
        goto unlink_and_fail;
570
    }
571

    
572
    /* Open the image */
573
    ret = bdrv_open_common(bs, filename, flags, drv);
574
    if (ret < 0) {
575
        goto unlink_and_fail;
576
    }
577

    
578
    /* If there is a backing file, use it */
579
    if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
580
        char backing_filename[PATH_MAX];
581
        int back_flags;
582
        BlockDriver *back_drv = NULL;
583

    
584
        bs->backing_hd = bdrv_new("");
585
        path_combine(backing_filename, sizeof(backing_filename),
586
                     filename, bs->backing_file);
587
        if (bs->backing_format[0] != '\0')
588
            back_drv = bdrv_find_format(bs->backing_format);
589

    
590
        /* backing files always opened read-only */
591
        back_flags =
592
            flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
593

    
594
        ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
595
        if (ret < 0) {
596
            bdrv_close(bs);
597
            return ret;
598
        }
599
        if (bs->is_temporary) {
600
            bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
601
        } else {
602
            /* base image inherits from "parent" */
603
            bs->backing_hd->keep_read_only = bs->keep_read_only;
604
        }
605
    }
606

    
607
    if (!bdrv_key_required(bs)) {
608
        /* call the change callback */
609
        bs->media_changed = 1;
610
        if (bs->change_cb)
611
            bs->change_cb(bs->change_opaque);
612
    }
613

    
614
    return 0;
615

    
616
unlink_and_fail:
617
    if (bs->is_temporary) {
618
        unlink(filename);
619
    }
620
    return ret;
621
}
622

    
623
void bdrv_close(BlockDriverState *bs)
624
{
625
    if (bs->drv) {
626
        if (bs->backing_hd) {
627
            bdrv_delete(bs->backing_hd);
628
            bs->backing_hd = NULL;
629
        }
630
        bs->drv->bdrv_close(bs);
631
        qemu_free(bs->opaque);
632
#ifdef _WIN32
633
        if (bs->is_temporary) {
634
            unlink(bs->filename);
635
        }
636
#endif
637
        bs->opaque = NULL;
638
        bs->drv = NULL;
639

    
640
        if (bs->file != NULL) {
641
            bdrv_close(bs->file);
642
        }
643

    
644
        /* call the change callback */
645
        bs->media_changed = 1;
646
        if (bs->change_cb)
647
            bs->change_cb(bs->change_opaque);
648
    }
649
}
650

    
651
void bdrv_close_all(void)
652
{
653
    BlockDriverState *bs;
654

    
655
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
656
        bdrv_close(bs);
657
    }
658
}
659

    
660
void bdrv_delete(BlockDriverState *bs)
661
{
662
    /* remove from list, if necessary */
663
    if (bs->device_name[0] != '\0') {
664
        QTAILQ_REMOVE(&bdrv_states, bs, list);
665
    }
666

    
667
    bdrv_close(bs);
668
    if (bs->file != NULL) {
669
        bdrv_delete(bs->file);
670
    }
671

    
672
    qemu_free(bs);
673
}
674

    
675
/*
676
 * Run consistency checks on an image
677
 *
678
 * Returns the number of errors or -errno when an internal error occurs
679
 */
680
int bdrv_check(BlockDriverState *bs)
681
{
682
    if (bs->drv->bdrv_check == NULL) {
683
        return -ENOTSUP;
684
    }
685

    
686
    return bs->drv->bdrv_check(bs);
687
}
688

    
689
/* commit COW file into the raw image */
690
int bdrv_commit(BlockDriverState *bs)
691
{
692
    BlockDriver *drv = bs->drv;
693
    int64_t i, total_sectors;
694
    int n, j, ro, open_flags;
695
    int ret = 0, rw_ret = 0;
696
    unsigned char sector[BDRV_SECTOR_SIZE];
697
    char filename[1024];
698
    BlockDriverState *bs_rw, *bs_ro;
699

    
700
    if (!drv)
701
        return -ENOMEDIUM;
702
    
703
    if (!bs->backing_hd) {
704
        return -ENOTSUP;
705
    }
706

    
707
    if (bs->backing_hd->keep_read_only) {
708
        return -EACCES;
709
    }
710
    
711
    ro = bs->backing_hd->read_only;
712
    strncpy(filename, bs->backing_hd->filename, sizeof(filename));
713
    open_flags =  bs->backing_hd->open_flags;
714

    
715
    if (ro) {
716
        /* re-open as RW */
717
        bdrv_delete(bs->backing_hd);
718
        bs->backing_hd = NULL;
719
        bs_rw = bdrv_new("");
720
        rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR, drv);
721
        if (rw_ret < 0) {
722
            bdrv_delete(bs_rw);
723
            /* try to re-open read-only */
724
            bs_ro = bdrv_new("");
725
            ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
726
            if (ret < 0) {
727
                bdrv_delete(bs_ro);
728
                /* drive not functional anymore */
729
                bs->drv = NULL;
730
                return ret;
731
            }
732
            bs->backing_hd = bs_ro;
733
            return rw_ret;
734
        }
735
        bs->backing_hd = bs_rw;
736
    }
737

    
738
    total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
739
    for (i = 0; i < total_sectors;) {
740
        if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
741
            for(j = 0; j < n; j++) {
742
                if (bdrv_read(bs, i, sector, 1) != 0) {
743
                    ret = -EIO;
744
                    goto ro_cleanup;
745
                }
746

    
747
                if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
748
                    ret = -EIO;
749
                    goto ro_cleanup;
750
                }
751
                i++;
752
            }
753
        } else {
754
            i += n;
755
        }
756
    }
757

    
758
    if (drv->bdrv_make_empty) {
759
        ret = drv->bdrv_make_empty(bs);
760
        bdrv_flush(bs);
761
    }
762

    
763
    /*
764
     * Make sure all data we wrote to the backing device is actually
765
     * stable on disk.
766
     */
767
    if (bs->backing_hd)
768
        bdrv_flush(bs->backing_hd);
769

    
770
ro_cleanup:
771

    
772
    if (ro) {
773
        /* re-open as RO */
774
        bdrv_delete(bs->backing_hd);
775
        bs->backing_hd = NULL;
776
        bs_ro = bdrv_new("");
777
        ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
778
        if (ret < 0) {
779
            bdrv_delete(bs_ro);
780
            /* drive not functional anymore */
781
            bs->drv = NULL;
782
            return ret;
783
        }
784
        bs->backing_hd = bs_ro;
785
        bs->backing_hd->keep_read_only = 0;
786
    }
787

    
788
    return ret;
789
}
790

    
791
void bdrv_commit_all(void)
792
{
793
    BlockDriverState *bs;
794

    
795
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
796
        bdrv_commit(bs);
797
    }
798
}
799

    
800
/*
801
 * Return values:
802
 * 0        - success
803
 * -EINVAL  - backing format specified, but no file
804
 * -ENOSPC  - can't update the backing file because no space is left in the
805
 *            image file header
806
 * -ENOTSUP - format driver doesn't support changing the backing file
807
 */
808
int bdrv_change_backing_file(BlockDriverState *bs,
809
    const char *backing_file, const char *backing_fmt)
810
{
811
    BlockDriver *drv = bs->drv;
812

    
813
    if (drv->bdrv_change_backing_file != NULL) {
814
        return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
815
    } else {
816
        return -ENOTSUP;
817
    }
818
}
819

    
820
static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
821
                                   size_t size)
822
{
823
    int64_t len;
824

    
825
    if (!bdrv_is_inserted(bs))
826
        return -ENOMEDIUM;
827

    
828
    if (bs->growable)
829
        return 0;
830

    
831
    len = bdrv_getlength(bs);
832

    
833
    if (offset < 0)
834
        return -EIO;
835

    
836
    if ((offset > len) || (len - offset < size))
837
        return -EIO;
838

    
839
    return 0;
840
}
841

    
842
static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
843
                              int nb_sectors)
844
{
845
    return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
846
                                   nb_sectors * BDRV_SECTOR_SIZE);
847
}
848

    
849
/* return < 0 if error. See bdrv_write() for the return codes */
850
int bdrv_read(BlockDriverState *bs, int64_t sector_num,
851
              uint8_t *buf, int nb_sectors)
852
{
853
    BlockDriver *drv = bs->drv;
854

    
855
    if (!drv)
856
        return -ENOMEDIUM;
857
    if (bdrv_check_request(bs, sector_num, nb_sectors))
858
        return -EIO;
859

    
860
    return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
861
}
862

    
863
static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
864
                             int nb_sectors, int dirty)
865
{
866
    int64_t start, end;
867
    unsigned long val, idx, bit;
868

    
869
    start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
870
    end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
871

    
872
    for (; start <= end; start++) {
873
        idx = start / (sizeof(unsigned long) * 8);
874
        bit = start % (sizeof(unsigned long) * 8);
875
        val = bs->dirty_bitmap[idx];
876
        if (dirty) {
877
            if (!(val & (1 << bit))) {
878
                bs->dirty_count++;
879
                val |= 1 << bit;
880
            }
881
        } else {
882
            if (val & (1 << bit)) {
883
                bs->dirty_count--;
884
                val &= ~(1 << bit);
885
            }
886
        }
887
        bs->dirty_bitmap[idx] = val;
888
    }
889
}
890

    
891
/* Return < 0 if error. Important errors are:
892
  -EIO         generic I/O error (may happen for all errors)
893
  -ENOMEDIUM   No media inserted.
894
  -EINVAL      Invalid sector number or nb_sectors
895
  -EACCES      Trying to write a read-only device
896
*/
897
int bdrv_write(BlockDriverState *bs, int64_t sector_num,
898
               const uint8_t *buf, int nb_sectors)
899
{
900
    BlockDriver *drv = bs->drv;
901
    if (!bs->drv)
902
        return -ENOMEDIUM;
903
    if (bs->read_only)
904
        return -EACCES;
905
    if (bdrv_check_request(bs, sector_num, nb_sectors))
906
        return -EIO;
907

    
908
    if (bs->dirty_bitmap) {
909
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
910
    }
911

    
912
    if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
913
        bs->wr_highest_sector = sector_num + nb_sectors - 1;
914
    }
915

    
916
    return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
917
}
918

    
919
int bdrv_pread(BlockDriverState *bs, int64_t offset,
920
               void *buf, int count1)
921
{
922
    uint8_t tmp_buf[BDRV_SECTOR_SIZE];
923
    int len, nb_sectors, count;
924
    int64_t sector_num;
925
    int ret;
926

    
927
    count = count1;
928
    /* first read to align to sector start */
929
    len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
930
    if (len > count)
931
        len = count;
932
    sector_num = offset >> BDRV_SECTOR_BITS;
933
    if (len > 0) {
934
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
935
            return ret;
936
        memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
937
        count -= len;
938
        if (count == 0)
939
            return count1;
940
        sector_num++;
941
        buf += len;
942
    }
943

    
944
    /* read the sectors "in place" */
945
    nb_sectors = count >> BDRV_SECTOR_BITS;
946
    if (nb_sectors > 0) {
947
        if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
948
            return ret;
949
        sector_num += nb_sectors;
950
        len = nb_sectors << BDRV_SECTOR_BITS;
951
        buf += len;
952
        count -= len;
953
    }
954

    
955
    /* add data from the last sector */
956
    if (count > 0) {
957
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
958
            return ret;
959
        memcpy(buf, tmp_buf, count);
960
    }
961
    return count1;
962
}
963

    
964
int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
965
                const void *buf, int count1)
966
{
967
    uint8_t tmp_buf[BDRV_SECTOR_SIZE];
968
    int len, nb_sectors, count;
969
    int64_t sector_num;
970
    int ret;
971

    
972
    count = count1;
973
    /* first write to align to sector start */
974
    len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
975
    if (len > count)
976
        len = count;
977
    sector_num = offset >> BDRV_SECTOR_BITS;
978
    if (len > 0) {
979
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
980
            return ret;
981
        memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
982
        if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
983
            return ret;
984
        count -= len;
985
        if (count == 0)
986
            return count1;
987
        sector_num++;
988
        buf += len;
989
    }
990

    
991
    /* write the sectors "in place" */
992
    nb_sectors = count >> BDRV_SECTOR_BITS;
993
    if (nb_sectors > 0) {
994
        if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
995
            return ret;
996
        sector_num += nb_sectors;
997
        len = nb_sectors << BDRV_SECTOR_BITS;
998
        buf += len;
999
        count -= len;
1000
    }
1001

    
1002
    /* add data from the last sector */
1003
    if (count > 0) {
1004
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1005
            return ret;
1006
        memcpy(tmp_buf, buf, count);
1007
        if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1008
            return ret;
1009
    }
1010
    return count1;
1011
}
1012

    
1013
/**
1014
 * Truncate file to 'offset' bytes (needed only for file protocols)
1015
 */
1016
int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1017
{
1018
    BlockDriver *drv = bs->drv;
1019
    int ret;
1020
    if (!drv)
1021
        return -ENOMEDIUM;
1022
    if (!drv->bdrv_truncate)
1023
        return -ENOTSUP;
1024
    if (bs->read_only)
1025
        return -EACCES;
1026
    ret = drv->bdrv_truncate(bs, offset);
1027
    if (ret == 0) {
1028
        ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1029
    }
1030
    return ret;
1031
}
1032

    
1033
/**
1034
 * Length of a file in bytes. Return < 0 if error or unknown.
1035
 */
1036
int64_t bdrv_getlength(BlockDriverState *bs)
1037
{
1038
    BlockDriver *drv = bs->drv;
1039
    if (!drv)
1040
        return -ENOMEDIUM;
1041

    
1042
    /* Fixed size devices use the total_sectors value for speed instead of
1043
       issuing a length query (like lseek) on each call.  Also, legacy block
1044
       drivers don't provide a bdrv_getlength function and must use
1045
       total_sectors. */
1046
    if (!bs->growable || !drv->bdrv_getlength) {
1047
        return bs->total_sectors * BDRV_SECTOR_SIZE;
1048
    }
1049
    return drv->bdrv_getlength(bs);
1050
}
1051

    
1052
/* return 0 as number of sectors if no device present or error */
1053
void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1054
{
1055
    int64_t length;
1056
    length = bdrv_getlength(bs);
1057
    if (length < 0)
1058
        length = 0;
1059
    else
1060
        length = length >> BDRV_SECTOR_BITS;
1061
    *nb_sectors_ptr = length;
1062
}
1063

    
1064
struct partition {
1065
        uint8_t boot_ind;           /* 0x80 - active */
1066
        uint8_t head;               /* starting head */
1067
        uint8_t sector;             /* starting sector */
1068
        uint8_t cyl;                /* starting cylinder */
1069
        uint8_t sys_ind;            /* What partition type */
1070
        uint8_t end_head;           /* end head */
1071
        uint8_t end_sector;         /* end sector */
1072
        uint8_t end_cyl;            /* end cylinder */
1073
        uint32_t start_sect;        /* starting sector counting from 0 */
1074
        uint32_t nr_sects;          /* nr of sectors in partition */
1075
} __attribute__((packed));
1076

    
1077
/* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1078
static int guess_disk_lchs(BlockDriverState *bs,
1079
                           int *pcylinders, int *pheads, int *psectors)
1080
{
1081
    uint8_t buf[BDRV_SECTOR_SIZE];
1082
    int ret, i, heads, sectors, cylinders;
1083
    struct partition *p;
1084
    uint32_t nr_sects;
1085
    uint64_t nb_sectors;
1086

    
1087
    bdrv_get_geometry(bs, &nb_sectors);
1088

    
1089
    ret = bdrv_read(bs, 0, buf, 1);
1090
    if (ret < 0)
1091
        return -1;
1092
    /* test msdos magic */
1093
    if (buf[510] != 0x55 || buf[511] != 0xaa)
1094
        return -1;
1095
    for(i = 0; i < 4; i++) {
1096
        p = ((struct partition *)(buf + 0x1be)) + i;
1097
        nr_sects = le32_to_cpu(p->nr_sects);
1098
        if (nr_sects && p->end_head) {
1099
            /* We make the assumption that the partition terminates on
1100
               a cylinder boundary */
1101
            heads = p->end_head + 1;
1102
            sectors = p->end_sector & 63;
1103
            if (sectors == 0)
1104
                continue;
1105
            cylinders = nb_sectors / (heads * sectors);
1106
            if (cylinders < 1 || cylinders > 16383)
1107
                continue;
1108
            *pheads = heads;
1109
            *psectors = sectors;
1110
            *pcylinders = cylinders;
1111
#if 0
1112
            printf("guessed geometry: LCHS=%d %d %d\n",
1113
                   cylinders, heads, sectors);
1114
#endif
1115
            return 0;
1116
        }
1117
    }
1118
    return -1;
1119
}
1120

    
1121
void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1122
{
1123
    int translation, lba_detected = 0;
1124
    int cylinders, heads, secs;
1125
    uint64_t nb_sectors;
1126

    
1127
    /* if a geometry hint is available, use it */
1128
    bdrv_get_geometry(bs, &nb_sectors);
1129
    bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1130
    translation = bdrv_get_translation_hint(bs);
1131
    if (cylinders != 0) {
1132
        *pcyls = cylinders;
1133
        *pheads = heads;
1134
        *psecs = secs;
1135
    } else {
1136
        if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1137
            if (heads > 16) {
1138
                /* if heads > 16, it means that a BIOS LBA
1139
                   translation was active, so the default
1140
                   hardware geometry is OK */
1141
                lba_detected = 1;
1142
                goto default_geometry;
1143
            } else {
1144
                *pcyls = cylinders;
1145
                *pheads = heads;
1146
                *psecs = secs;
1147
                /* disable any translation to be in sync with
1148
                   the logical geometry */
1149
                if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1150
                    bdrv_set_translation_hint(bs,
1151
                                              BIOS_ATA_TRANSLATION_NONE);
1152
                }
1153
            }
1154
        } else {
1155
        default_geometry:
1156
            /* if no geometry, use a standard physical disk geometry */
1157
            cylinders = nb_sectors / (16 * 63);
1158

    
1159
            if (cylinders > 16383)
1160
                cylinders = 16383;
1161
            else if (cylinders < 2)
1162
                cylinders = 2;
1163
            *pcyls = cylinders;
1164
            *pheads = 16;
1165
            *psecs = 63;
1166
            if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1167
                if ((*pcyls * *pheads) <= 131072) {
1168
                    bdrv_set_translation_hint(bs,
1169
                                              BIOS_ATA_TRANSLATION_LARGE);
1170
                } else {
1171
                    bdrv_set_translation_hint(bs,
1172
                                              BIOS_ATA_TRANSLATION_LBA);
1173
                }
1174
            }
1175
        }
1176
        bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1177
    }
1178
}
1179

    
1180
void bdrv_set_geometry_hint(BlockDriverState *bs,
1181
                            int cyls, int heads, int secs)
1182
{
1183
    bs->cyls = cyls;
1184
    bs->heads = heads;
1185
    bs->secs = secs;
1186
}
1187

    
1188
void bdrv_set_type_hint(BlockDriverState *bs, int type)
1189
{
1190
    bs->type = type;
1191
    bs->removable = ((type == BDRV_TYPE_CDROM ||
1192
                      type == BDRV_TYPE_FLOPPY));
1193
}
1194

    
1195
void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1196
{
1197
    bs->translation = translation;
1198
}
1199

    
1200
void bdrv_get_geometry_hint(BlockDriverState *bs,
1201
                            int *pcyls, int *pheads, int *psecs)
1202
{
1203
    *pcyls = bs->cyls;
1204
    *pheads = bs->heads;
1205
    *psecs = bs->secs;
1206
}
1207

    
1208
int bdrv_get_type_hint(BlockDriverState *bs)
1209
{
1210
    return bs->type;
1211
}
1212

    
1213
int bdrv_get_translation_hint(BlockDriverState *bs)
1214
{
1215
    return bs->translation;
1216
}
1217

    
1218
void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1219
                       BlockErrorAction on_write_error)
1220
{
1221
    bs->on_read_error = on_read_error;
1222
    bs->on_write_error = on_write_error;
1223
}
1224

    
1225
BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1226
{
1227
    return is_read ? bs->on_read_error : bs->on_write_error;
1228
}
1229

    
1230
int bdrv_is_removable(BlockDriverState *bs)
1231
{
1232
    return bs->removable;
1233
}
1234

    
1235
int bdrv_is_read_only(BlockDriverState *bs)
1236
{
1237
    return bs->read_only;
1238
}
1239

    
1240
int bdrv_is_sg(BlockDriverState *bs)
1241
{
1242
    return bs->sg;
1243
}
1244

    
1245
int bdrv_enable_write_cache(BlockDriverState *bs)
1246
{
1247
    return bs->enable_write_cache;
1248
}
1249

    
1250
/* XXX: no longer used */
1251
void bdrv_set_change_cb(BlockDriverState *bs,
1252
                        void (*change_cb)(void *opaque), void *opaque)
1253
{
1254
    bs->change_cb = change_cb;
1255
    bs->change_opaque = opaque;
1256
}
1257

    
1258
int bdrv_is_encrypted(BlockDriverState *bs)
1259
{
1260
    if (bs->backing_hd && bs->backing_hd->encrypted)
1261
        return 1;
1262
    return bs->encrypted;
1263
}
1264

    
1265
int bdrv_key_required(BlockDriverState *bs)
1266
{
1267
    BlockDriverState *backing_hd = bs->backing_hd;
1268

    
1269
    if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1270
        return 1;
1271
    return (bs->encrypted && !bs->valid_key);
1272
}
1273

    
1274
int bdrv_set_key(BlockDriverState *bs, const char *key)
1275
{
1276
    int ret;
1277
    if (bs->backing_hd && bs->backing_hd->encrypted) {
1278
        ret = bdrv_set_key(bs->backing_hd, key);
1279
        if (ret < 0)
1280
            return ret;
1281
        if (!bs->encrypted)
1282
            return 0;
1283
    }
1284
    if (!bs->encrypted) {
1285
        return -EINVAL;
1286
    } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1287
        return -ENOMEDIUM;
1288
    }
1289
    ret = bs->drv->bdrv_set_key(bs, key);
1290
    if (ret < 0) {
1291
        bs->valid_key = 0;
1292
    } else if (!bs->valid_key) {
1293
        bs->valid_key = 1;
1294
        /* call the change callback now, we skipped it on open */
1295
        bs->media_changed = 1;
1296
        if (bs->change_cb)
1297
            bs->change_cb(bs->change_opaque);
1298
    }
1299
    return ret;
1300
}
1301

    
1302
void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1303
{
1304
    if (!bs->drv) {
1305
        buf[0] = '\0';
1306
    } else {
1307
        pstrcpy(buf, buf_size, bs->drv->format_name);
1308
    }
1309
}
1310

    
1311
void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1312
                         void *opaque)
1313
{
1314
    BlockDriver *drv;
1315

    
1316
    QLIST_FOREACH(drv, &bdrv_drivers, list) {
1317
        it(opaque, drv->format_name);
1318
    }
1319
}
1320

    
1321
BlockDriverState *bdrv_find(const char *name)
1322
{
1323
    BlockDriverState *bs;
1324

    
1325
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1326
        if (!strcmp(name, bs->device_name)) {
1327
            return bs;
1328
        }
1329
    }
1330
    return NULL;
1331
}
1332

    
1333
BlockDriverState *bdrv_next(BlockDriverState *bs)
1334
{
1335
    if (!bs) {
1336
        return QTAILQ_FIRST(&bdrv_states);
1337
    }
1338
    return QTAILQ_NEXT(bs, list);
1339
}
1340

    
1341
void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1342
{
1343
    BlockDriverState *bs;
1344

    
1345
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1346
        it(opaque, bs);
1347
    }
1348
}
1349

    
1350
const char *bdrv_get_device_name(BlockDriverState *bs)
1351
{
1352
    return bs->device_name;
1353
}
1354

    
1355
void bdrv_flush(BlockDriverState *bs)
1356
{
1357
    if (bs->open_flags & BDRV_O_NO_FLUSH) {
1358
        return;
1359
    }
1360

    
1361
    if (bs->drv && bs->drv->bdrv_flush)
1362
        bs->drv->bdrv_flush(bs);
1363
}
1364

    
1365
void bdrv_flush_all(void)
1366
{
1367
    BlockDriverState *bs;
1368

    
1369
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1370
        if (bs->drv && !bdrv_is_read_only(bs) &&
1371
            (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1372
            bdrv_flush(bs);
1373
        }
1374
    }
1375
}
1376

    
1377
int bdrv_has_zero_init(BlockDriverState *bs)
1378
{
1379
    assert(bs->drv);
1380

    
1381
    if (bs->drv->no_zero_init) {
1382
        return 0;
1383
    } else if (bs->file) {
1384
        return bdrv_has_zero_init(bs->file);
1385
    }
1386

    
1387
    return 1;
1388
}
1389

    
1390
/*
1391
 * Returns true iff the specified sector is present in the disk image. Drivers
1392
 * not implementing the functionality are assumed to not support backing files,
1393
 * hence all their sectors are reported as allocated.
1394
 *
1395
 * 'pnum' is set to the number of sectors (including and immediately following
1396
 * the specified sector) that are known to be in the same
1397
 * allocated/unallocated state.
1398
 *
1399
 * 'nb_sectors' is the max value 'pnum' should be set to.
1400
 */
1401
int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1402
        int *pnum)
1403
{
1404
    int64_t n;
1405
    if (!bs->drv->bdrv_is_allocated) {
1406
        if (sector_num >= bs->total_sectors) {
1407
            *pnum = 0;
1408
            return 0;
1409
        }
1410
        n = bs->total_sectors - sector_num;
1411
        *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1412
        return 1;
1413
    }
1414
    return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1415
}
1416

    
1417
void bdrv_mon_event(const BlockDriverState *bdrv,
1418
                    BlockMonEventAction action, int is_read)
1419
{
1420
    QObject *data;
1421
    const char *action_str;
1422

    
1423
    switch (action) {
1424
    case BDRV_ACTION_REPORT:
1425
        action_str = "report";
1426
        break;
1427
    case BDRV_ACTION_IGNORE:
1428
        action_str = "ignore";
1429
        break;
1430
    case BDRV_ACTION_STOP:
1431
        action_str = "stop";
1432
        break;
1433
    default:
1434
        abort();
1435
    }
1436

    
1437
    data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1438
                              bdrv->device_name,
1439
                              action_str,
1440
                              is_read ? "read" : "write");
1441
    monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1442

    
1443
    qobject_decref(data);
1444
}
1445

    
1446
static void bdrv_print_dict(QObject *obj, void *opaque)
1447
{
1448
    QDict *bs_dict;
1449
    Monitor *mon = opaque;
1450

    
1451
    bs_dict = qobject_to_qdict(obj);
1452

    
1453
    monitor_printf(mon, "%s: type=%s removable=%d",
1454
                        qdict_get_str(bs_dict, "device"),
1455
                        qdict_get_str(bs_dict, "type"),
1456
                        qdict_get_bool(bs_dict, "removable"));
1457

    
1458
    if (qdict_get_bool(bs_dict, "removable")) {
1459
        monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1460
    }
1461

    
1462
    if (qdict_haskey(bs_dict, "inserted")) {
1463
        QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1464

    
1465
        monitor_printf(mon, " file=");
1466
        monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1467
        if (qdict_haskey(qdict, "backing_file")) {
1468
            monitor_printf(mon, " backing_file=");
1469
            monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1470
        }
1471
        monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1472
                            qdict_get_bool(qdict, "ro"),
1473
                            qdict_get_str(qdict, "drv"),
1474
                            qdict_get_bool(qdict, "encrypted"));
1475
    } else {
1476
        monitor_printf(mon, " [not inserted]");
1477
    }
1478

    
1479
    monitor_printf(mon, "\n");
1480
}
1481

    
1482
void bdrv_info_print(Monitor *mon, const QObject *data)
1483
{
1484
    qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1485
}
1486

    
1487
void bdrv_info(Monitor *mon, QObject **ret_data)
1488
{
1489
    QList *bs_list;
1490
    BlockDriverState *bs;
1491

    
1492
    bs_list = qlist_new();
1493

    
1494
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1495
        QObject *bs_obj;
1496
        const char *type = "unknown";
1497

    
1498
        switch(bs->type) {
1499
        case BDRV_TYPE_HD:
1500
            type = "hd";
1501
            break;
1502
        case BDRV_TYPE_CDROM:
1503
            type = "cdrom";
1504
            break;
1505
        case BDRV_TYPE_FLOPPY:
1506
            type = "floppy";
1507
            break;
1508
        }
1509

    
1510
        bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1511
                                    "'removable': %i, 'locked': %i }",
1512
                                    bs->device_name, type, bs->removable,
1513
                                    bs->locked);
1514

    
1515
        if (bs->drv) {
1516
            QObject *obj;
1517
            QDict *bs_dict = qobject_to_qdict(bs_obj);
1518

    
1519
            obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1520
                                     "'encrypted': %i }",
1521
                                     bs->filename, bs->read_only,
1522
                                     bs->drv->format_name,
1523
                                     bdrv_is_encrypted(bs));
1524
            if (bs->backing_file[0] != '\0') {
1525
                QDict *qdict = qobject_to_qdict(obj);
1526
                qdict_put(qdict, "backing_file",
1527
                          qstring_from_str(bs->backing_file));
1528
            }
1529

    
1530
            qdict_put_obj(bs_dict, "inserted", obj);
1531
        }
1532
        qlist_append_obj(bs_list, bs_obj);
1533
    }
1534

    
1535
    *ret_data = QOBJECT(bs_list);
1536
}
1537

    
1538
static void bdrv_stats_iter(QObject *data, void *opaque)
1539
{
1540
    QDict *qdict;
1541
    Monitor *mon = opaque;
1542

    
1543
    qdict = qobject_to_qdict(data);
1544
    monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1545

    
1546
    qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1547
    monitor_printf(mon, " rd_bytes=%" PRId64
1548
                        " wr_bytes=%" PRId64
1549
                        " rd_operations=%" PRId64
1550
                        " wr_operations=%" PRId64
1551
                        "\n",
1552
                        qdict_get_int(qdict, "rd_bytes"),
1553
                        qdict_get_int(qdict, "wr_bytes"),
1554
                        qdict_get_int(qdict, "rd_operations"),
1555
                        qdict_get_int(qdict, "wr_operations"));
1556
}
1557

    
1558
void bdrv_stats_print(Monitor *mon, const QObject *data)
1559
{
1560
    qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1561
}
1562

    
1563
static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1564
{
1565
    QObject *res;
1566
    QDict *dict;
1567

    
1568
    res = qobject_from_jsonf("{ 'stats': {"
1569
                             "'rd_bytes': %" PRId64 ","
1570
                             "'wr_bytes': %" PRId64 ","
1571
                             "'rd_operations': %" PRId64 ","
1572
                             "'wr_operations': %" PRId64 ","
1573
                             "'wr_highest_offset': %" PRId64
1574
                             "} }",
1575
                             bs->rd_bytes, bs->wr_bytes,
1576
                             bs->rd_ops, bs->wr_ops,
1577
                             bs->wr_highest_sector *
1578
                             (uint64_t)BDRV_SECTOR_SIZE);
1579
    dict  = qobject_to_qdict(res);
1580

    
1581
    if (*bs->device_name) {
1582
        qdict_put(dict, "device", qstring_from_str(bs->device_name));
1583
    }
1584

    
1585
    if (bs->file) {
1586
        QObject *parent = bdrv_info_stats_bs(bs->file);
1587
        qdict_put_obj(dict, "parent", parent);
1588
    }
1589

    
1590
    return res;
1591
}
1592

    
1593
void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1594
{
1595
    QObject *obj;
1596
    QList *devices;
1597
    BlockDriverState *bs;
1598

    
1599
    devices = qlist_new();
1600

    
1601
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1602
        obj = bdrv_info_stats_bs(bs);
1603
        qlist_append_obj(devices, obj);
1604
    }
1605

    
1606
    *ret_data = QOBJECT(devices);
1607
}
1608

    
1609
const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1610
{
1611
    if (bs->backing_hd && bs->backing_hd->encrypted)
1612
        return bs->backing_file;
1613
    else if (bs->encrypted)
1614
        return bs->filename;
1615
    else
1616
        return NULL;
1617
}
1618

    
1619
void bdrv_get_backing_filename(BlockDriverState *bs,
1620
                               char *filename, int filename_size)
1621
{
1622
    if (!bs->backing_file) {
1623
        pstrcpy(filename, filename_size, "");
1624
    } else {
1625
        pstrcpy(filename, filename_size, bs->backing_file);
1626
    }
1627
}
1628

    
1629
int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1630
                          const uint8_t *buf, int nb_sectors)
1631
{
1632
    BlockDriver *drv = bs->drv;
1633
    if (!drv)
1634
        return -ENOMEDIUM;
1635
    if (!drv->bdrv_write_compressed)
1636
        return -ENOTSUP;
1637
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1638
        return -EIO;
1639

    
1640
    if (bs->dirty_bitmap) {
1641
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1642
    }
1643

    
1644
    return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1645
}
1646

    
1647
int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1648
{
1649
    BlockDriver *drv = bs->drv;
1650
    if (!drv)
1651
        return -ENOMEDIUM;
1652
    if (!drv->bdrv_get_info)
1653
        return -ENOTSUP;
1654
    memset(bdi, 0, sizeof(*bdi));
1655
    return drv->bdrv_get_info(bs, bdi);
1656
}
1657

    
1658
int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1659
                      int64_t pos, int size)
1660
{
1661
    BlockDriver *drv = bs->drv;
1662
    if (!drv)
1663
        return -ENOMEDIUM;
1664
    if (drv->bdrv_save_vmstate)
1665
        return drv->bdrv_save_vmstate(bs, buf, pos, size);
1666
    if (bs->file)
1667
        return bdrv_save_vmstate(bs->file, buf, pos, size);
1668
    return -ENOTSUP;
1669
}
1670

    
1671
int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1672
                      int64_t pos, int size)
1673
{
1674
    BlockDriver *drv = bs->drv;
1675
    if (!drv)
1676
        return -ENOMEDIUM;
1677
    if (drv->bdrv_load_vmstate)
1678
        return drv->bdrv_load_vmstate(bs, buf, pos, size);
1679
    if (bs->file)
1680
        return bdrv_load_vmstate(bs->file, buf, pos, size);
1681
    return -ENOTSUP;
1682
}
1683

    
1684
void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1685
{
1686
    BlockDriver *drv = bs->drv;
1687

    
1688
    if (!drv || !drv->bdrv_debug_event) {
1689
        return;
1690
    }
1691

    
1692
    return drv->bdrv_debug_event(bs, event);
1693

    
1694
}
1695

    
1696
/**************************************************************/
1697
/* handling of snapshots */
1698

    
1699
int bdrv_can_snapshot(BlockDriverState *bs)
1700
{
1701
    BlockDriver *drv = bs->drv;
1702
    if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1703
        return 0;
1704
    }
1705

    
1706
    if (!drv->bdrv_snapshot_create) {
1707
        if (bs->file != NULL) {
1708
            return bdrv_can_snapshot(bs->file);
1709
        }
1710
        return 0;
1711
    }
1712

    
1713
    return 1;
1714
}
1715

    
1716
int bdrv_snapshot_create(BlockDriverState *bs,
1717
                         QEMUSnapshotInfo *sn_info)
1718
{
1719
    BlockDriver *drv = bs->drv;
1720
    if (!drv)
1721
        return -ENOMEDIUM;
1722
    if (drv->bdrv_snapshot_create)
1723
        return drv->bdrv_snapshot_create(bs, sn_info);
1724
    if (bs->file)
1725
        return bdrv_snapshot_create(bs->file, sn_info);
1726
    return -ENOTSUP;
1727
}
1728

    
1729
int bdrv_snapshot_goto(BlockDriverState *bs,
1730
                       const char *snapshot_id)
1731
{
1732
    BlockDriver *drv = bs->drv;
1733
    int ret, open_ret;
1734

    
1735
    if (!drv)
1736
        return -ENOMEDIUM;
1737
    if (drv->bdrv_snapshot_goto)
1738
        return drv->bdrv_snapshot_goto(bs, snapshot_id);
1739

    
1740
    if (bs->file) {
1741
        drv->bdrv_close(bs);
1742
        ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1743
        open_ret = drv->bdrv_open(bs, bs->open_flags);
1744
        if (open_ret < 0) {
1745
            bdrv_delete(bs->file);
1746
            bs->drv = NULL;
1747
            return open_ret;
1748
        }
1749
        return ret;
1750
    }
1751

    
1752
    return -ENOTSUP;
1753
}
1754

    
1755
int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1756
{
1757
    BlockDriver *drv = bs->drv;
1758
    if (!drv)
1759
        return -ENOMEDIUM;
1760
    if (drv->bdrv_snapshot_delete)
1761
        return drv->bdrv_snapshot_delete(bs, snapshot_id);
1762
    if (bs->file)
1763
        return bdrv_snapshot_delete(bs->file, snapshot_id);
1764
    return -ENOTSUP;
1765
}
1766

    
1767
int bdrv_snapshot_list(BlockDriverState *bs,
1768
                       QEMUSnapshotInfo **psn_info)
1769
{
1770
    BlockDriver *drv = bs->drv;
1771
    if (!drv)
1772
        return -ENOMEDIUM;
1773
    if (drv->bdrv_snapshot_list)
1774
        return drv->bdrv_snapshot_list(bs, psn_info);
1775
    if (bs->file)
1776
        return bdrv_snapshot_list(bs->file, psn_info);
1777
    return -ENOTSUP;
1778
}
1779

    
1780
#define NB_SUFFIXES 4
1781

    
1782
char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1783
{
1784
    static const char suffixes[NB_SUFFIXES] = "KMGT";
1785
    int64_t base;
1786
    int i;
1787

    
1788
    if (size <= 999) {
1789
        snprintf(buf, buf_size, "%" PRId64, size);
1790
    } else {
1791
        base = 1024;
1792
        for(i = 0; i < NB_SUFFIXES; i++) {
1793
            if (size < (10 * base)) {
1794
                snprintf(buf, buf_size, "%0.1f%c",
1795
                         (double)size / base,
1796
                         suffixes[i]);
1797
                break;
1798
            } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1799
                snprintf(buf, buf_size, "%" PRId64 "%c",
1800
                         ((size + (base >> 1)) / base),
1801
                         suffixes[i]);
1802
                break;
1803
            }
1804
            base = base * 1024;
1805
        }
1806
    }
1807
    return buf;
1808
}
1809

    
1810
char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1811
{
1812
    char buf1[128], date_buf[128], clock_buf[128];
1813
#ifdef _WIN32
1814
    struct tm *ptm;
1815
#else
1816
    struct tm tm;
1817
#endif
1818
    time_t ti;
1819
    int64_t secs;
1820

    
1821
    if (!sn) {
1822
        snprintf(buf, buf_size,
1823
                 "%-10s%-20s%7s%20s%15s",
1824
                 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1825
    } else {
1826
        ti = sn->date_sec;
1827
#ifdef _WIN32
1828
        ptm = localtime(&ti);
1829
        strftime(date_buf, sizeof(date_buf),
1830
                 "%Y-%m-%d %H:%M:%S", ptm);
1831
#else
1832
        localtime_r(&ti, &tm);
1833
        strftime(date_buf, sizeof(date_buf),
1834
                 "%Y-%m-%d %H:%M:%S", &tm);
1835
#endif
1836
        secs = sn->vm_clock_nsec / 1000000000;
1837
        snprintf(clock_buf, sizeof(clock_buf),
1838
                 "%02d:%02d:%02d.%03d",
1839
                 (int)(secs / 3600),
1840
                 (int)((secs / 60) % 60),
1841
                 (int)(secs % 60),
1842
                 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1843
        snprintf(buf, buf_size,
1844
                 "%-10s%-20s%7s%20s%15s",
1845
                 sn->id_str, sn->name,
1846
                 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1847
                 date_buf,
1848
                 clock_buf);
1849
    }
1850
    return buf;
1851
}
1852

    
1853

    
1854
/**************************************************************/
1855
/* async I/Os */
1856

    
1857
BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1858
                                 QEMUIOVector *qiov, int nb_sectors,
1859
                                 BlockDriverCompletionFunc *cb, void *opaque)
1860
{
1861
    BlockDriver *drv = bs->drv;
1862
    BlockDriverAIOCB *ret;
1863

    
1864
    if (!drv)
1865
        return NULL;
1866
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1867
        return NULL;
1868

    
1869
    ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1870
                              cb, opaque);
1871

    
1872
    if (ret) {
1873
        /* Update stats even though technically transfer has not happened. */
1874
        bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1875
        bs->rd_ops ++;
1876
    }
1877

    
1878
    return ret;
1879
}
1880

    
1881
BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1882
                                  QEMUIOVector *qiov, int nb_sectors,
1883
                                  BlockDriverCompletionFunc *cb, void *opaque)
1884
{
1885
    BlockDriver *drv = bs->drv;
1886
    BlockDriverAIOCB *ret;
1887

    
1888
    if (!drv)
1889
        return NULL;
1890
    if (bs->read_only)
1891
        return NULL;
1892
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1893
        return NULL;
1894

    
1895
    if (bs->dirty_bitmap) {
1896
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1897
    }
1898

    
1899
    ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
1900
                               cb, opaque);
1901

    
1902
    if (ret) {
1903
        /* Update stats even though technically transfer has not happened. */
1904
        bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1905
        bs->wr_ops ++;
1906
        if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1907
            bs->wr_highest_sector = sector_num + nb_sectors - 1;
1908
        }
1909
    }
1910

    
1911
    return ret;
1912
}
1913

    
1914

    
1915
typedef struct MultiwriteCB {
1916
    int error;
1917
    int num_requests;
1918
    int num_callbacks;
1919
    struct {
1920
        BlockDriverCompletionFunc *cb;
1921
        void *opaque;
1922
        QEMUIOVector *free_qiov;
1923
        void *free_buf;
1924
    } callbacks[];
1925
} MultiwriteCB;
1926

    
1927
static void multiwrite_user_cb(MultiwriteCB *mcb)
1928
{
1929
    int i;
1930

    
1931
    for (i = 0; i < mcb->num_callbacks; i++) {
1932
        mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
1933
        if (mcb->callbacks[i].free_qiov) {
1934
            qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
1935
        }
1936
        qemu_free(mcb->callbacks[i].free_qiov);
1937
        qemu_vfree(mcb->callbacks[i].free_buf);
1938
    }
1939
}
1940

    
1941
static void multiwrite_cb(void *opaque, int ret)
1942
{
1943
    MultiwriteCB *mcb = opaque;
1944

    
1945
    if (ret < 0 && !mcb->error) {
1946
        mcb->error = ret;
1947
        multiwrite_user_cb(mcb);
1948
    }
1949

    
1950
    mcb->num_requests--;
1951
    if (mcb->num_requests == 0) {
1952
        if (mcb->error == 0) {
1953
            multiwrite_user_cb(mcb);
1954
        }
1955
        qemu_free(mcb);
1956
    }
1957
}
1958

    
1959
static int multiwrite_req_compare(const void *a, const void *b)
1960
{
1961
    const BlockRequest *req1 = a, *req2 = b;
1962

    
1963
    /*
1964
     * Note that we can't simply subtract req2->sector from req1->sector
1965
     * here as that could overflow the return value.
1966
     */
1967
    if (req1->sector > req2->sector) {
1968
        return 1;
1969
    } else if (req1->sector < req2->sector) {
1970
        return -1;
1971
    } else {
1972
        return 0;
1973
    }
1974
}
1975

    
1976
/*
1977
 * Takes a bunch of requests and tries to merge them. Returns the number of
1978
 * requests that remain after merging.
1979
 */
1980
static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
1981
    int num_reqs, MultiwriteCB *mcb)
1982
{
1983
    int i, outidx;
1984

    
1985
    // Sort requests by start sector
1986
    qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
1987

    
1988
    // Check if adjacent requests touch the same clusters. If so, combine them,
1989
    // filling up gaps with zero sectors.
1990
    outidx = 0;
1991
    for (i = 1; i < num_reqs; i++) {
1992
        int merge = 0;
1993
        int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
1994

    
1995
        // This handles the cases that are valid for all block drivers, namely
1996
        // exactly sequential writes and overlapping writes.
1997
        if (reqs[i].sector <= oldreq_last) {
1998
            merge = 1;
1999
        }
2000

    
2001
        // The block driver may decide that it makes sense to combine requests
2002
        // even if there is a gap of some sectors between them. In this case,
2003
        // the gap is filled with zeros (therefore only applicable for yet
2004
        // unused space in format like qcow2).
2005
        if (!merge && bs->drv->bdrv_merge_requests) {
2006
            merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2007
        }
2008

    
2009
        if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2010
            merge = 0;
2011
        }
2012

    
2013
        if (merge) {
2014
            size_t size;
2015
            QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2016
            qemu_iovec_init(qiov,
2017
                reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2018

    
2019
            // Add the first request to the merged one. If the requests are
2020
            // overlapping, drop the last sectors of the first request.
2021
            size = (reqs[i].sector - reqs[outidx].sector) << 9;
2022
            qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2023

    
2024
            // We might need to add some zeros between the two requests
2025
            if (reqs[i].sector > oldreq_last) {
2026
                size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2027
                uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2028
                memset(buf, 0, zero_bytes);
2029
                qemu_iovec_add(qiov, buf, zero_bytes);
2030
                mcb->callbacks[i].free_buf = buf;
2031
            }
2032

    
2033
            // Add the second request
2034
            qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2035

    
2036
            reqs[outidx].nb_sectors = qiov->size >> 9;
2037
            reqs[outidx].qiov = qiov;
2038

    
2039
            mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2040
        } else {
2041
            outidx++;
2042
            reqs[outidx].sector     = reqs[i].sector;
2043
            reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2044
            reqs[outidx].qiov       = reqs[i].qiov;
2045
        }
2046
    }
2047

    
2048
    return outidx + 1;
2049
}
2050

    
2051
/*
2052
 * Submit multiple AIO write requests at once.
2053
 *
2054
 * On success, the function returns 0 and all requests in the reqs array have
2055
 * been submitted. In error case this function returns -1, and any of the
2056
 * requests may or may not be submitted yet. In particular, this means that the
2057
 * callback will be called for some of the requests, for others it won't. The
2058
 * caller must check the error field of the BlockRequest to wait for the right
2059
 * callbacks (if error != 0, no callback will be called).
2060
 *
2061
 * The implementation may modify the contents of the reqs array, e.g. to merge
2062
 * requests. However, the fields opaque and error are left unmodified as they
2063
 * are used to signal failure for a single request to the caller.
2064
 */
2065
int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2066
{
2067
    BlockDriverAIOCB *acb;
2068
    MultiwriteCB *mcb;
2069
    int i;
2070

    
2071
    if (num_reqs == 0) {
2072
        return 0;
2073
    }
2074

    
2075
    // Create MultiwriteCB structure
2076
    mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2077
    mcb->num_requests = 0;
2078
    mcb->num_callbacks = num_reqs;
2079

    
2080
    for (i = 0; i < num_reqs; i++) {
2081
        mcb->callbacks[i].cb = reqs[i].cb;
2082
        mcb->callbacks[i].opaque = reqs[i].opaque;
2083
    }
2084

    
2085
    // Check for mergable requests
2086
    num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2087

    
2088
    // Run the aio requests
2089
    for (i = 0; i < num_reqs; i++) {
2090
        acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2091
            reqs[i].nb_sectors, multiwrite_cb, mcb);
2092

    
2093
        if (acb == NULL) {
2094
            // We can only fail the whole thing if no request has been
2095
            // submitted yet. Otherwise we'll wait for the submitted AIOs to
2096
            // complete and report the error in the callback.
2097
            if (mcb->num_requests == 0) {
2098
                reqs[i].error = -EIO;
2099
                goto fail;
2100
            } else {
2101
                mcb->num_requests++;
2102
                multiwrite_cb(mcb, -EIO);
2103
                break;
2104
            }
2105
        } else {
2106
            mcb->num_requests++;
2107
        }
2108
    }
2109

    
2110
    return 0;
2111

    
2112
fail:
2113
    qemu_free(mcb);
2114
    return -1;
2115
}
2116

    
2117
BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2118
        BlockDriverCompletionFunc *cb, void *opaque)
2119
{
2120
    BlockDriver *drv = bs->drv;
2121

    
2122
    if (bs->open_flags & BDRV_O_NO_FLUSH) {
2123
        return bdrv_aio_noop_em(bs, cb, opaque);
2124
    }
2125

    
2126
    if (!drv)
2127
        return NULL;
2128
    return drv->bdrv_aio_flush(bs, cb, opaque);
2129
}
2130

    
2131
void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2132
{
2133
    acb->pool->cancel(acb);
2134
}
2135

    
2136

    
2137
/**************************************************************/
2138
/* async block device emulation */
2139

    
2140
typedef struct BlockDriverAIOCBSync {
2141
    BlockDriverAIOCB common;
2142
    QEMUBH *bh;
2143
    int ret;
2144
    /* vector translation state */
2145
    QEMUIOVector *qiov;
2146
    uint8_t *bounce;
2147
    int is_write;
2148
} BlockDriverAIOCBSync;
2149

    
2150
static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2151
{
2152
    BlockDriverAIOCBSync *acb =
2153
        container_of(blockacb, BlockDriverAIOCBSync, common);
2154
    qemu_bh_delete(acb->bh);
2155
    acb->bh = NULL;
2156
    qemu_aio_release(acb);
2157
}
2158

    
2159
static AIOPool bdrv_em_aio_pool = {
2160
    .aiocb_size         = sizeof(BlockDriverAIOCBSync),
2161
    .cancel             = bdrv_aio_cancel_em,
2162
};
2163

    
2164
static void bdrv_aio_bh_cb(void *opaque)
2165
{
2166
    BlockDriverAIOCBSync *acb = opaque;
2167

    
2168
    if (!acb->is_write)
2169
        qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2170
    qemu_vfree(acb->bounce);
2171
    acb->common.cb(acb->common.opaque, acb->ret);
2172
    qemu_bh_delete(acb->bh);
2173
    acb->bh = NULL;
2174
    qemu_aio_release(acb);
2175
}
2176

    
2177
static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2178
                                            int64_t sector_num,
2179
                                            QEMUIOVector *qiov,
2180
                                            int nb_sectors,
2181
                                            BlockDriverCompletionFunc *cb,
2182
                                            void *opaque,
2183
                                            int is_write)
2184

    
2185
{
2186
    BlockDriverAIOCBSync *acb;
2187

    
2188
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2189
    acb->is_write = is_write;
2190
    acb->qiov = qiov;
2191
    acb->bounce = qemu_blockalign(bs, qiov->size);
2192

    
2193
    if (!acb->bh)
2194
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2195

    
2196
    if (is_write) {
2197
        qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2198
        acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2199
    } else {
2200
        acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2201
    }
2202

    
2203
    qemu_bh_schedule(acb->bh);
2204

    
2205
    return &acb->common;
2206
}
2207

    
2208
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2209
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2210
        BlockDriverCompletionFunc *cb, void *opaque)
2211
{
2212
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2213
}
2214

    
2215
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2216
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2217
        BlockDriverCompletionFunc *cb, void *opaque)
2218
{
2219
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2220
}
2221

    
2222
static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2223
        BlockDriverCompletionFunc *cb, void *opaque)
2224
{
2225
    BlockDriverAIOCBSync *acb;
2226

    
2227
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2228
    acb->is_write = 1; /* don't bounce in the completion hadler */
2229
    acb->qiov = NULL;
2230
    acb->bounce = NULL;
2231
    acb->ret = 0;
2232

    
2233
    if (!acb->bh)
2234
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2235

    
2236
    bdrv_flush(bs);
2237
    qemu_bh_schedule(acb->bh);
2238
    return &acb->common;
2239
}
2240

    
2241
static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2242
        BlockDriverCompletionFunc *cb, void *opaque)
2243
{
2244
    BlockDriverAIOCBSync *acb;
2245

    
2246
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2247
    acb->is_write = 1; /* don't bounce in the completion handler */
2248
    acb->qiov = NULL;
2249
    acb->bounce = NULL;
2250
    acb->ret = 0;
2251

    
2252
    if (!acb->bh) {
2253
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2254
    }
2255

    
2256
    qemu_bh_schedule(acb->bh);
2257
    return &acb->common;
2258
}
2259

    
2260
/**************************************************************/
2261
/* sync block device emulation */
2262

    
2263
static void bdrv_rw_em_cb(void *opaque, int ret)
2264
{
2265
    *(int *)opaque = ret;
2266
}
2267

    
2268
#define NOT_DONE 0x7fffffff
2269

    
2270
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2271
                        uint8_t *buf, int nb_sectors)
2272
{
2273
    int async_ret;
2274
    BlockDriverAIOCB *acb;
2275
    struct iovec iov;
2276
    QEMUIOVector qiov;
2277

    
2278
    async_context_push();
2279

    
2280
    async_ret = NOT_DONE;
2281
    iov.iov_base = (void *)buf;
2282
    iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2283
    qemu_iovec_init_external(&qiov, &iov, 1);
2284
    acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2285
        bdrv_rw_em_cb, &async_ret);
2286
    if (acb == NULL) {
2287
        async_ret = -1;
2288
        goto fail;
2289
    }
2290

    
2291
    while (async_ret == NOT_DONE) {
2292
        qemu_aio_wait();
2293
    }
2294

    
2295

    
2296
fail:
2297
    async_context_pop();
2298
    return async_ret;
2299
}
2300

    
2301
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2302
                         const uint8_t *buf, int nb_sectors)
2303
{
2304
    int async_ret;
2305
    BlockDriverAIOCB *acb;
2306
    struct iovec iov;
2307
    QEMUIOVector qiov;
2308

    
2309
    async_context_push();
2310

    
2311
    async_ret = NOT_DONE;
2312
    iov.iov_base = (void *)buf;
2313
    iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2314
    qemu_iovec_init_external(&qiov, &iov, 1);
2315
    acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2316
        bdrv_rw_em_cb, &async_ret);
2317
    if (acb == NULL) {
2318
        async_ret = -1;
2319
        goto fail;
2320
    }
2321
    while (async_ret == NOT_DONE) {
2322
        qemu_aio_wait();
2323
    }
2324

    
2325
fail:
2326
    async_context_pop();
2327
    return async_ret;
2328
}
2329

    
2330
void bdrv_init(void)
2331
{
2332
    module_call_init(MODULE_INIT_BLOCK);
2333
}
2334

    
2335
void bdrv_init_with_whitelist(void)
2336
{
2337
    use_bdrv_whitelist = 1;
2338
    bdrv_init();
2339
}
2340

    
2341
void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2342
                   BlockDriverCompletionFunc *cb, void *opaque)
2343
{
2344
    BlockDriverAIOCB *acb;
2345

    
2346
    if (pool->free_aiocb) {
2347
        acb = pool->free_aiocb;
2348
        pool->free_aiocb = acb->next;
2349
    } else {
2350
        acb = qemu_mallocz(pool->aiocb_size);
2351
        acb->pool = pool;
2352
    }
2353
    acb->bs = bs;
2354
    acb->cb = cb;
2355
    acb->opaque = opaque;
2356
    return acb;
2357
}
2358

    
2359
void qemu_aio_release(void *p)
2360
{
2361
    BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2362
    AIOPool *pool = acb->pool;
2363
    acb->next = pool->free_aiocb;
2364
    pool->free_aiocb = acb;
2365
}
2366

    
2367
/**************************************************************/
2368
/* removable device support */
2369

    
2370
/**
2371
 * Return TRUE if the media is present
2372
 */
2373
int bdrv_is_inserted(BlockDriverState *bs)
2374
{
2375
    BlockDriver *drv = bs->drv;
2376
    int ret;
2377
    if (!drv)
2378
        return 0;
2379
    if (!drv->bdrv_is_inserted)
2380
        return 1;
2381
    ret = drv->bdrv_is_inserted(bs);
2382
    return ret;
2383
}
2384

    
2385
/**
2386
 * Return TRUE if the media changed since the last call to this
2387
 * function. It is currently only used for floppy disks
2388
 */
2389
int bdrv_media_changed(BlockDriverState *bs)
2390
{
2391
    BlockDriver *drv = bs->drv;
2392
    int ret;
2393

    
2394
    if (!drv || !drv->bdrv_media_changed)
2395
        ret = -ENOTSUP;
2396
    else
2397
        ret = drv->bdrv_media_changed(bs);
2398
    if (ret == -ENOTSUP)
2399
        ret = bs->media_changed;
2400
    bs->media_changed = 0;
2401
    return ret;
2402
}
2403

    
2404
/**
2405
 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2406
 */
2407
int bdrv_eject(BlockDriverState *bs, int eject_flag)
2408
{
2409
    BlockDriver *drv = bs->drv;
2410
    int ret;
2411

    
2412
    if (bs->locked) {
2413
        return -EBUSY;
2414
    }
2415

    
2416
    if (!drv || !drv->bdrv_eject) {
2417
        ret = -ENOTSUP;
2418
    } else {
2419
        ret = drv->bdrv_eject(bs, eject_flag);
2420
    }
2421
    if (ret == -ENOTSUP) {
2422
        if (eject_flag)
2423
            bdrv_close(bs);
2424
        ret = 0;
2425
    }
2426

    
2427
    return ret;
2428
}
2429

    
2430
int bdrv_is_locked(BlockDriverState *bs)
2431
{
2432
    return bs->locked;
2433
}
2434

    
2435
/**
2436
 * Lock or unlock the media (if it is locked, the user won't be able
2437
 * to eject it manually).
2438
 */
2439
void bdrv_set_locked(BlockDriverState *bs, int locked)
2440
{
2441
    BlockDriver *drv = bs->drv;
2442

    
2443
    bs->locked = locked;
2444
    if (drv && drv->bdrv_set_locked) {
2445
        drv->bdrv_set_locked(bs, locked);
2446
    }
2447
}
2448

    
2449
/* needed for generic scsi interface */
2450

    
2451
int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2452
{
2453
    BlockDriver *drv = bs->drv;
2454

    
2455
    if (drv && drv->bdrv_ioctl)
2456
        return drv->bdrv_ioctl(bs, req, buf);
2457
    return -ENOTSUP;
2458
}
2459

    
2460
BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2461
        unsigned long int req, void *buf,
2462
        BlockDriverCompletionFunc *cb, void *opaque)
2463
{
2464
    BlockDriver *drv = bs->drv;
2465

    
2466
    if (drv && drv->bdrv_aio_ioctl)
2467
        return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2468
    return NULL;
2469
}
2470

    
2471

    
2472

    
2473
void *qemu_blockalign(BlockDriverState *bs, size_t size)
2474
{
2475
    return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2476
}
2477

    
2478
void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2479
{
2480
    int64_t bitmap_size;
2481

    
2482
    bs->dirty_count = 0;
2483
    if (enable) {
2484
        if (!bs->dirty_bitmap) {
2485
            bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2486
                    BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2487
            bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2488

    
2489
            bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2490
        }
2491
    } else {
2492
        if (bs->dirty_bitmap) {
2493
            qemu_free(bs->dirty_bitmap);
2494
            bs->dirty_bitmap = NULL;
2495
        }
2496
    }
2497
}
2498

    
2499
int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2500
{
2501
    int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2502

    
2503
    if (bs->dirty_bitmap &&
2504
        (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2505
        return bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2506
            (1 << (chunk % (sizeof(unsigned long) * 8)));
2507
    } else {
2508
        return 0;
2509
    }
2510
}
2511

    
2512
void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2513
                      int nr_sectors)
2514
{
2515
    set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2516
}
2517

    
2518
int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2519
{
2520
    return bs->dirty_count;
2521
}