Statistics
| Branch: | Revision:

root / block.c @ 79368c81

History | View | Annotate | Download (69.8 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
/* The device to use for VM snapshots */
67
static BlockDriverState *bs_snapshots;
68

    
69
/* If non-zero, use only whitelisted block drivers */
70
static int use_bdrv_whitelist;
71

    
72
int path_is_absolute(const char *path)
73
{
74
    const char *p;
75
#ifdef _WIN32
76
    /* specific case for names like: "\\.\d:" */
77
    if (*path == '/' || *path == '\\')
78
        return 1;
79
#endif
80
    p = strchr(path, ':');
81
    if (p)
82
        p++;
83
    else
84
        p = path;
85
#ifdef _WIN32
86
    return (*p == '/' || *p == '\\');
87
#else
88
    return (*p == '/');
89
#endif
90
}
91

    
92
/* if filename is absolute, just copy it to dest. Otherwise, build a
93
   path to it by considering it is relative to base_path. URL are
94
   supported. */
95
void path_combine(char *dest, int dest_size,
96
                  const char *base_path,
97
                  const char *filename)
98
{
99
    const char *p, *p1;
100
    int len;
101

    
102
    if (dest_size <= 0)
103
        return;
104
    if (path_is_absolute(filename)) {
105
        pstrcpy(dest, dest_size, filename);
106
    } else {
107
        p = strchr(base_path, ':');
108
        if (p)
109
            p++;
110
        else
111
            p = base_path;
112
        p1 = strrchr(base_path, '/');
113
#ifdef _WIN32
114
        {
115
            const char *p2;
116
            p2 = strrchr(base_path, '\\');
117
            if (!p1 || p2 > p1)
118
                p1 = p2;
119
        }
120
#endif
121
        if (p1)
122
            p1++;
123
        else
124
            p1 = base_path;
125
        if (p1 > p)
126
            p = p1;
127
        len = p - base_path;
128
        if (len > dest_size - 1)
129
            len = dest_size - 1;
130
        memcpy(dest, base_path, len);
131
        dest[len] = '\0';
132
        pstrcat(dest, dest_size, filename);
133
    }
134
}
135

    
136
void bdrv_register(BlockDriver *bdrv)
137
{
138
    if (!bdrv->bdrv_aio_readv) {
139
        /* add AIO emulation layer */
140
        bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
141
        bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
142
    } else if (!bdrv->bdrv_read) {
143
        /* add synchronous IO emulation layer */
144
        bdrv->bdrv_read = bdrv_read_em;
145
        bdrv->bdrv_write = bdrv_write_em;
146
    }
147

    
148
    if (!bdrv->bdrv_aio_flush)
149
        bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
150

    
151
    QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
152
}
153

    
154
/* create a new block device (by default it is empty) */
155
BlockDriverState *bdrv_new(const char *device_name)
156
{
157
    BlockDriverState *bs;
158

    
159
    bs = qemu_mallocz(sizeof(BlockDriverState));
160
    pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
161
    if (device_name[0] != '\0') {
162
        QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
163
    }
164
    return bs;
165
}
166

    
167
BlockDriver *bdrv_find_format(const char *format_name)
168
{
169
    BlockDriver *drv1;
170
    QLIST_FOREACH(drv1, &bdrv_drivers, list) {
171
        if (!strcmp(drv1->format_name, format_name)) {
172
            return drv1;
173
        }
174
    }
175
    return NULL;
176
}
177

    
178
static int bdrv_is_whitelisted(BlockDriver *drv)
179
{
180
    static const char *whitelist[] = {
181
        CONFIG_BDRV_WHITELIST
182
    };
183
    const char **p;
184

    
185
    if (!whitelist[0])
186
        return 1;               /* no whitelist, anything goes */
187

    
188
    for (p = whitelist; *p; p++) {
189
        if (!strcmp(drv->format_name, *p)) {
190
            return 1;
191
        }
192
    }
193
    return 0;
194
}
195

    
196
BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
197
{
198
    BlockDriver *drv = bdrv_find_format(format_name);
199
    return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
200
}
201

    
202
int bdrv_create(BlockDriver *drv, const char* filename,
203
    QEMUOptionParameter *options)
204
{
205
    if (!drv->bdrv_create)
206
        return -ENOTSUP;
207

    
208
    return drv->bdrv_create(filename, options);
209
}
210

    
211
int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
212
{
213
    BlockDriver *drv;
214

    
215
    drv = bdrv_find_protocol(filename);
216
    if (drv == NULL) {
217
        drv = bdrv_find_format("file");
218
    }
219

    
220
    return bdrv_create(drv, filename, options);
221
}
222

    
223
#ifdef _WIN32
224
void get_tmp_filename(char *filename, int size)
225
{
226
    char temp_dir[MAX_PATH];
227

    
228
    GetTempPath(MAX_PATH, temp_dir);
229
    GetTempFileName(temp_dir, "qem", 0, filename);
230
}
231
#else
232
void get_tmp_filename(char *filename, int size)
233
{
234
    int fd;
235
    const char *tmpdir;
236
    /* XXX: race condition possible */
237
    tmpdir = getenv("TMPDIR");
238
    if (!tmpdir)
239
        tmpdir = "/tmp";
240
    snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
241
    fd = mkstemp(filename);
242
    close(fd);
243
}
244
#endif
245

    
246
#ifdef _WIN32
247
static int is_windows_drive_prefix(const char *filename)
248
{
249
    return (((filename[0] >= 'a' && filename[0] <= 'z') ||
250
             (filename[0] >= 'A' && filename[0] <= 'Z')) &&
251
            filename[1] == ':');
252
}
253

    
254
int is_windows_drive(const char *filename)
255
{
256
    if (is_windows_drive_prefix(filename) &&
257
        filename[2] == '\0')
258
        return 1;
259
    if (strstart(filename, "\\\\.\\", NULL) ||
260
        strstart(filename, "//./", NULL))
261
        return 1;
262
    return 0;
263
}
264
#endif
265

    
266
/*
267
 * Detect host devices. By convention, /dev/cdrom[N] is always
268
 * recognized as a host CDROM.
269
 */
270
static BlockDriver *find_hdev_driver(const char *filename)
271
{
272
    int score_max = 0, score;
273
    BlockDriver *drv = NULL, *d;
274

    
275
    QLIST_FOREACH(d, &bdrv_drivers, list) {
276
        if (d->bdrv_probe_device) {
277
            score = d->bdrv_probe_device(filename);
278
            if (score > score_max) {
279
                score_max = score;
280
                drv = d;
281
            }
282
        }
283
    }
284

    
285
    return drv;
286
}
287

    
288
BlockDriver *bdrv_find_protocol(const char *filename)
289
{
290
    BlockDriver *drv1;
291
    char protocol[128];
292
    int len;
293
    const char *p;
294

    
295
    /* TODO Drivers without bdrv_file_open must be specified explicitly */
296

    
297
    /*
298
     * XXX(hch): we really should not let host device detection
299
     * override an explicit protocol specification, but moving this
300
     * later breaks access to device names with colons in them.
301
     * Thanks to the brain-dead persistent naming schemes on udev-
302
     * based Linux systems those actually are quite common.
303
     */
304
    drv1 = find_hdev_driver(filename);
305
    if (drv1) {
306
        return drv1;
307
    }
308

    
309
#ifdef _WIN32
310
     if (is_windows_drive(filename) ||
311
         is_windows_drive_prefix(filename))
312
         return bdrv_find_format("file");
313
#endif
314

    
315
    p = strchr(filename, ':');
316
    if (!p) {
317
        return bdrv_find_format("file");
318
    }
319
    len = p - filename;
320
    if (len > sizeof(protocol) - 1)
321
        len = sizeof(protocol) - 1;
322
    memcpy(protocol, filename, len);
323
    protocol[len] = '\0';
324
    QLIST_FOREACH(drv1, &bdrv_drivers, list) {
325
        if (drv1->protocol_name &&
326
            !strcmp(drv1->protocol_name, protocol)) {
327
            return drv1;
328
        }
329
    }
330
    return NULL;
331
}
332

    
333
static BlockDriver *find_image_format(const char *filename)
334
{
335
    int ret, score, score_max;
336
    BlockDriver *drv1, *drv;
337
    uint8_t buf[2048];
338
    BlockDriverState *bs;
339

    
340
    ret = bdrv_file_open(&bs, filename, 0);
341
    if (ret < 0)
342
        return NULL;
343

    
344
    /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
345
    if (bs->sg || !bdrv_is_inserted(bs)) {
346
        bdrv_delete(bs);
347
        return bdrv_find_format("raw");
348
    }
349

    
350
    ret = bdrv_pread(bs, 0, buf, sizeof(buf));
351
    bdrv_delete(bs);
352
    if (ret < 0) {
353
        return NULL;
354
    }
355

    
356
    score_max = 0;
357
    drv = NULL;
358
    QLIST_FOREACH(drv1, &bdrv_drivers, list) {
359
        if (drv1->bdrv_probe) {
360
            score = drv1->bdrv_probe(buf, ret, filename);
361
            if (score > score_max) {
362
                score_max = score;
363
                drv = drv1;
364
            }
365
        }
366
    }
367
    return drv;
368
}
369

    
370
/**
371
 * Set the current 'total_sectors' value
372
 */
373
static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
374
{
375
    BlockDriver *drv = bs->drv;
376

    
377
    /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
378
    if (bs->sg)
379
        return 0;
380

    
381
    /* query actual device if possible, otherwise just trust the hint */
382
    if (drv->bdrv_getlength) {
383
        int64_t length = drv->bdrv_getlength(bs);
384
        if (length < 0) {
385
            return length;
386
        }
387
        hint = length >> BDRV_SECTOR_BITS;
388
    }
389

    
390
    bs->total_sectors = hint;
391
    return 0;
392
}
393

    
394
/*
395
 * Common part for opening disk images and files
396
 */
397
static int bdrv_open_common(BlockDriverState *bs, const char *filename,
398
    int flags, BlockDriver *drv)
399
{
400
    int ret, open_flags;
401

    
402
    assert(drv != NULL);
403

    
404
    bs->file = NULL;
405
    bs->total_sectors = 0;
406
    bs->encrypted = 0;
407
    bs->valid_key = 0;
408
    bs->open_flags = flags;
409
    /* buffer_alignment defaulted to 512, drivers can change this value */
410
    bs->buffer_alignment = 512;
411

    
412
    pstrcpy(bs->filename, sizeof(bs->filename), filename);
413

    
414
    if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
415
        return -ENOTSUP;
416
    }
417

    
418
    bs->drv = drv;
419
    bs->opaque = qemu_mallocz(drv->instance_size);
420

    
421
    /*
422
     * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
423
     * write cache to the guest.  We do need the fdatasync to flush
424
     * out transactions for block allocations, and we maybe have a
425
     * volatile write cache in our backing device to deal with.
426
     */
427
    if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
428
        bs->enable_write_cache = 1;
429

    
430
    /*
431
     * Clear flags that are internal to the block layer before opening the
432
     * image.
433
     */
434
    open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
435

    
436
    /*
437
     * Snapshots should be writeable.
438
     */
439
    if (bs->is_temporary) {
440
        open_flags |= BDRV_O_RDWR;
441
    }
442

    
443
    /* Open the image, either directly or using a protocol */
444
    if (drv->bdrv_file_open) {
445
        ret = drv->bdrv_file_open(bs, filename, open_flags);
446
    } else {
447
        ret = bdrv_file_open(&bs->file, filename, open_flags);
448
        if (ret >= 0) {
449
            ret = drv->bdrv_open(bs, open_flags);
450
        }
451
    }
452

    
453
    if (ret < 0) {
454
        goto free_and_fail;
455
    }
456

    
457
    bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
458

    
459
    ret = refresh_total_sectors(bs, bs->total_sectors);
460
    if (ret < 0) {
461
        goto free_and_fail;
462
    }
463

    
464
#ifndef _WIN32
465
    if (bs->is_temporary) {
466
        unlink(filename);
467
    }
468
#endif
469
    return 0;
470

    
471
free_and_fail:
472
    if (bs->file) {
473
        bdrv_delete(bs->file);
474
        bs->file = NULL;
475
    }
476
    qemu_free(bs->opaque);
477
    bs->opaque = NULL;
478
    bs->drv = NULL;
479
    return ret;
480
}
481

    
482
/*
483
 * Opens a file using a protocol (file, host_device, nbd, ...)
484
 */
485
int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
486
{
487
    BlockDriverState *bs;
488
    BlockDriver *drv;
489
    int ret;
490

    
491
    drv = bdrv_find_protocol(filename);
492
    if (!drv) {
493
        return -ENOENT;
494
    }
495

    
496
    bs = bdrv_new("");
497
    ret = bdrv_open_common(bs, filename, flags, drv);
498
    if (ret < 0) {
499
        bdrv_delete(bs);
500
        return ret;
501
    }
502
    bs->growable = 1;
503
    *pbs = bs;
504
    return 0;
505
}
506

    
507
/*
508
 * Opens a disk image (raw, qcow2, vmdk, ...)
509
 */
510
int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
511
              BlockDriver *drv)
512
{
513
    int ret;
514
    int probed = 0;
515

    
516
    if (flags & BDRV_O_SNAPSHOT) {
517
        BlockDriverState *bs1;
518
        int64_t total_size;
519
        int is_protocol = 0;
520
        BlockDriver *bdrv_qcow2;
521
        QEMUOptionParameter *options;
522
        char tmp_filename[PATH_MAX];
523
        char backing_filename[PATH_MAX];
524

    
525
        /* if snapshot, we create a temporary backing file and open it
526
           instead of opening 'filename' directly */
527

    
528
        /* if there is a backing file, use it */
529
        bs1 = bdrv_new("");
530
        ret = bdrv_open(bs1, filename, 0, drv);
531
        if (ret < 0) {
532
            bdrv_delete(bs1);
533
            return ret;
534
        }
535
        total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
536

    
537
        if (bs1->drv && bs1->drv->protocol_name)
538
            is_protocol = 1;
539

    
540
        bdrv_delete(bs1);
541

    
542
        get_tmp_filename(tmp_filename, sizeof(tmp_filename));
543

    
544
        /* Real path is meaningless for protocols */
545
        if (is_protocol)
546
            snprintf(backing_filename, sizeof(backing_filename),
547
                     "%s", filename);
548
        else if (!realpath(filename, backing_filename))
549
            return -errno;
550

    
551
        bdrv_qcow2 = bdrv_find_format("qcow2");
552
        options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
553

    
554
        set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
555
        set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
556
        if (drv) {
557
            set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
558
                drv->format_name);
559
        }
560

    
561
        ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
562
        free_option_parameters(options);
563
        if (ret < 0) {
564
            return ret;
565
        }
566

    
567
        filename = tmp_filename;
568
        drv = bdrv_qcow2;
569
        bs->is_temporary = 1;
570
    }
571

    
572
    /* Find the right image format driver */
573
    if (!drv) {
574
        drv = find_image_format(filename);
575
        probed = 1;
576
    }
577

    
578
    if (!drv) {
579
        ret = -ENOENT;
580
        goto unlink_and_fail;
581
    }
582

    
583
    /* Open the image */
584
    ret = bdrv_open_common(bs, filename, flags, drv);
585
    if (ret < 0) {
586
        goto unlink_and_fail;
587
    }
588

    
589
    bs->probed = probed;
590

    
591
    /* If there is a backing file, use it */
592
    if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
593
        char backing_filename[PATH_MAX];
594
        int back_flags;
595
        BlockDriver *back_drv = NULL;
596

    
597
        bs->backing_hd = bdrv_new("");
598
        path_combine(backing_filename, sizeof(backing_filename),
599
                     filename, bs->backing_file);
600
        if (bs->backing_format[0] != '\0')
601
            back_drv = bdrv_find_format(bs->backing_format);
602

    
603
        /* backing files always opened read-only */
604
        back_flags =
605
            flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
606

    
607
        ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
608
        if (ret < 0) {
609
            bdrv_close(bs);
610
            return ret;
611
        }
612
        if (bs->is_temporary) {
613
            bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
614
        } else {
615
            /* base image inherits from "parent" */
616
            bs->backing_hd->keep_read_only = bs->keep_read_only;
617
        }
618
    }
619

    
620
    if (!bdrv_key_required(bs)) {
621
        /* call the change callback */
622
        bs->media_changed = 1;
623
        if (bs->change_cb)
624
            bs->change_cb(bs->change_opaque);
625
    }
626

    
627
    return 0;
628

    
629
unlink_and_fail:
630
    if (bs->is_temporary) {
631
        unlink(filename);
632
    }
633
    return ret;
634
}
635

    
636
void bdrv_close(BlockDriverState *bs)
637
{
638
    if (bs->drv) {
639
        if (bs == bs_snapshots) {
640
            bs_snapshots = NULL;
641
        }
642
        if (bs->backing_hd) {
643
            bdrv_delete(bs->backing_hd);
644
            bs->backing_hd = NULL;
645
        }
646
        bs->drv->bdrv_close(bs);
647
        qemu_free(bs->opaque);
648
#ifdef _WIN32
649
        if (bs->is_temporary) {
650
            unlink(bs->filename);
651
        }
652
#endif
653
        bs->opaque = NULL;
654
        bs->drv = NULL;
655

    
656
        if (bs->file != NULL) {
657
            bdrv_close(bs->file);
658
        }
659

    
660
        /* call the change callback */
661
        bs->media_changed = 1;
662
        if (bs->change_cb)
663
            bs->change_cb(bs->change_opaque);
664
    }
665
}
666

    
667
void bdrv_close_all(void)
668
{
669
    BlockDriverState *bs;
670

    
671
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
672
        bdrv_close(bs);
673
    }
674
}
675

    
676
void bdrv_delete(BlockDriverState *bs)
677
{
678
    assert(!bs->peer);
679

    
680
    /* remove from list, if necessary */
681
    if (bs->device_name[0] != '\0') {
682
        QTAILQ_REMOVE(&bdrv_states, bs, list);
683
    }
684

    
685
    bdrv_close(bs);
686
    if (bs->file != NULL) {
687
        bdrv_delete(bs->file);
688
    }
689

    
690
    assert(bs != bs_snapshots);
691
    qemu_free(bs);
692
}
693

    
694
int bdrv_attach(BlockDriverState *bs, DeviceState *qdev)
695
{
696
    if (bs->peer) {
697
        return -EBUSY;
698
    }
699
    bs->peer = qdev;
700
    return 0;
701
}
702

    
703
void bdrv_detach(BlockDriverState *bs, DeviceState *qdev)
704
{
705
    assert(bs->peer == qdev);
706
    bs->peer = NULL;
707
}
708

    
709
DeviceState *bdrv_get_attached(BlockDriverState *bs)
710
{
711
    return bs->peer;
712
}
713

    
714
/*
715
 * Run consistency checks on an image
716
 *
717
 * Returns 0 if the check could be completed (it doesn't mean that the image is
718
 * free of errors) or -errno when an internal error occured. The results of the
719
 * check are stored in res.
720
 */
721
int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
722
{
723
    if (bs->drv->bdrv_check == NULL) {
724
        return -ENOTSUP;
725
    }
726

    
727
    memset(res, 0, sizeof(*res));
728
    return bs->drv->bdrv_check(bs, res);
729
}
730

    
731
/* commit COW file into the raw image */
732
int bdrv_commit(BlockDriverState *bs)
733
{
734
    BlockDriver *drv = bs->drv;
735
    int64_t i, total_sectors;
736
    int n, j, ro, open_flags;
737
    int ret = 0, rw_ret = 0;
738
    unsigned char sector[BDRV_SECTOR_SIZE];
739
    char filename[1024];
740
    BlockDriverState *bs_rw, *bs_ro;
741

    
742
    if (!drv)
743
        return -ENOMEDIUM;
744
    
745
    if (!bs->backing_hd) {
746
        return -ENOTSUP;
747
    }
748

    
749
    if (bs->backing_hd->keep_read_only) {
750
        return -EACCES;
751
    }
752
    
753
    ro = bs->backing_hd->read_only;
754
    strncpy(filename, bs->backing_hd->filename, sizeof(filename));
755
    open_flags =  bs->backing_hd->open_flags;
756

    
757
    if (ro) {
758
        /* re-open as RW */
759
        bdrv_delete(bs->backing_hd);
760
        bs->backing_hd = NULL;
761
        bs_rw = bdrv_new("");
762
        rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR, drv);
763
        if (rw_ret < 0) {
764
            bdrv_delete(bs_rw);
765
            /* try to re-open read-only */
766
            bs_ro = bdrv_new("");
767
            ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
768
            if (ret < 0) {
769
                bdrv_delete(bs_ro);
770
                /* drive not functional anymore */
771
                bs->drv = NULL;
772
                return ret;
773
            }
774
            bs->backing_hd = bs_ro;
775
            return rw_ret;
776
        }
777
        bs->backing_hd = bs_rw;
778
    }
779

    
780
    total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
781
    for (i = 0; i < total_sectors;) {
782
        if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
783
            for(j = 0; j < n; j++) {
784
                if (bdrv_read(bs, i, sector, 1) != 0) {
785
                    ret = -EIO;
786
                    goto ro_cleanup;
787
                }
788

    
789
                if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
790
                    ret = -EIO;
791
                    goto ro_cleanup;
792
                }
793
                i++;
794
            }
795
        } else {
796
            i += n;
797
        }
798
    }
799

    
800
    if (drv->bdrv_make_empty) {
801
        ret = drv->bdrv_make_empty(bs);
802
        bdrv_flush(bs);
803
    }
804

    
805
    /*
806
     * Make sure all data we wrote to the backing device is actually
807
     * stable on disk.
808
     */
809
    if (bs->backing_hd)
810
        bdrv_flush(bs->backing_hd);
811

    
812
ro_cleanup:
813

    
814
    if (ro) {
815
        /* re-open as RO */
816
        bdrv_delete(bs->backing_hd);
817
        bs->backing_hd = NULL;
818
        bs_ro = bdrv_new("");
819
        ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
820
        if (ret < 0) {
821
            bdrv_delete(bs_ro);
822
            /* drive not functional anymore */
823
            bs->drv = NULL;
824
            return ret;
825
        }
826
        bs->backing_hd = bs_ro;
827
        bs->backing_hd->keep_read_only = 0;
828
    }
829

    
830
    return ret;
831
}
832

    
833
void bdrv_commit_all(void)
834
{
835
    BlockDriverState *bs;
836

    
837
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
838
        bdrv_commit(bs);
839
    }
840
}
841

    
842
/*
843
 * Return values:
844
 * 0        - success
845
 * -EINVAL  - backing format specified, but no file
846
 * -ENOSPC  - can't update the backing file because no space is left in the
847
 *            image file header
848
 * -ENOTSUP - format driver doesn't support changing the backing file
849
 */
850
int bdrv_change_backing_file(BlockDriverState *bs,
851
    const char *backing_file, const char *backing_fmt)
852
{
853
    BlockDriver *drv = bs->drv;
854

    
855
    if (drv->bdrv_change_backing_file != NULL) {
856
        return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
857
    } else {
858
        return -ENOTSUP;
859
    }
860
}
861

    
862
static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
863
                                   size_t size)
864
{
865
    int64_t len;
866

    
867
    if (!bdrv_is_inserted(bs))
868
        return -ENOMEDIUM;
869

    
870
    if (bs->growable)
871
        return 0;
872

    
873
    len = bdrv_getlength(bs);
874

    
875
    if (offset < 0)
876
        return -EIO;
877

    
878
    if ((offset > len) || (len - offset < size))
879
        return -EIO;
880

    
881
    return 0;
882
}
883

    
884
static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
885
                              int nb_sectors)
886
{
887
    return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
888
                                   nb_sectors * BDRV_SECTOR_SIZE);
889
}
890

    
891
/* return < 0 if error. See bdrv_write() for the return codes */
892
int bdrv_read(BlockDriverState *bs, int64_t sector_num,
893
              uint8_t *buf, int nb_sectors)
894
{
895
    BlockDriver *drv = bs->drv;
896

    
897
    if (!drv)
898
        return -ENOMEDIUM;
899
    if (bdrv_check_request(bs, sector_num, nb_sectors))
900
        return -EIO;
901

    
902
    return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
903
}
904

    
905
static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
906
                             int nb_sectors, int dirty)
907
{
908
    int64_t start, end;
909
    unsigned long val, idx, bit;
910

    
911
    start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
912
    end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
913

    
914
    for (; start <= end; start++) {
915
        idx = start / (sizeof(unsigned long) * 8);
916
        bit = start % (sizeof(unsigned long) * 8);
917
        val = bs->dirty_bitmap[idx];
918
        if (dirty) {
919
            if (!(val & (1 << bit))) {
920
                bs->dirty_count++;
921
                val |= 1 << bit;
922
            }
923
        } else {
924
            if (val & (1 << bit)) {
925
                bs->dirty_count--;
926
                val &= ~(1 << bit);
927
            }
928
        }
929
        bs->dirty_bitmap[idx] = val;
930
    }
931
}
932

    
933
/* Return < 0 if error. Important errors are:
934
  -EIO         generic I/O error (may happen for all errors)
935
  -ENOMEDIUM   No media inserted.
936
  -EINVAL      Invalid sector number or nb_sectors
937
  -EACCES      Trying to write a read-only device
938
*/
939
int bdrv_write(BlockDriverState *bs, int64_t sector_num,
940
               const uint8_t *buf, int nb_sectors)
941
{
942
    BlockDriver *drv = bs->drv;
943
    if (!bs->drv)
944
        return -ENOMEDIUM;
945
    if (bs->read_only)
946
        return -EACCES;
947
    if (bdrv_check_request(bs, sector_num, nb_sectors))
948
        return -EIO;
949

    
950
    if (bs->dirty_bitmap) {
951
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
952
    }
953

    
954
    if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
955
        bs->wr_highest_sector = sector_num + nb_sectors - 1;
956
    }
957

    
958
    return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
959
}
960

    
961
int bdrv_pread(BlockDriverState *bs, int64_t offset,
962
               void *buf, int count1)
963
{
964
    uint8_t tmp_buf[BDRV_SECTOR_SIZE];
965
    int len, nb_sectors, count;
966
    int64_t sector_num;
967
    int ret;
968

    
969
    count = count1;
970
    /* first read to align to sector start */
971
    len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
972
    if (len > count)
973
        len = count;
974
    sector_num = offset >> BDRV_SECTOR_BITS;
975
    if (len > 0) {
976
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
977
            return ret;
978
        memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
979
        count -= len;
980
        if (count == 0)
981
            return count1;
982
        sector_num++;
983
        buf += len;
984
    }
985

    
986
    /* read the sectors "in place" */
987
    nb_sectors = count >> BDRV_SECTOR_BITS;
988
    if (nb_sectors > 0) {
989
        if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
990
            return ret;
991
        sector_num += nb_sectors;
992
        len = nb_sectors << BDRV_SECTOR_BITS;
993
        buf += len;
994
        count -= len;
995
    }
996

    
997
    /* add data from the last sector */
998
    if (count > 0) {
999
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1000
            return ret;
1001
        memcpy(buf, tmp_buf, count);
1002
    }
1003
    return count1;
1004
}
1005

    
1006
int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1007
                const void *buf, int count1)
1008
{
1009
    uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1010
    int len, nb_sectors, count;
1011
    int64_t sector_num;
1012
    int ret;
1013

    
1014
    count = count1;
1015
    /* first write to align to sector start */
1016
    len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1017
    if (len > count)
1018
        len = count;
1019
    sector_num = offset >> BDRV_SECTOR_BITS;
1020
    if (len > 0) {
1021
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1022
            return ret;
1023
        memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1024
        if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1025
            return ret;
1026
        count -= len;
1027
        if (count == 0)
1028
            return count1;
1029
        sector_num++;
1030
        buf += len;
1031
    }
1032

    
1033
    /* write the sectors "in place" */
1034
    nb_sectors = count >> BDRV_SECTOR_BITS;
1035
    if (nb_sectors > 0) {
1036
        if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1037
            return ret;
1038
        sector_num += nb_sectors;
1039
        len = nb_sectors << BDRV_SECTOR_BITS;
1040
        buf += len;
1041
        count -= len;
1042
    }
1043

    
1044
    /* add data from the last sector */
1045
    if (count > 0) {
1046
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1047
            return ret;
1048
        memcpy(tmp_buf, buf, count);
1049
        if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1050
            return ret;
1051
    }
1052
    return count1;
1053
}
1054

    
1055
/*
1056
 * Writes to the file and ensures that no writes are reordered across this
1057
 * request (acts as a barrier)
1058
 *
1059
 * Returns 0 on success, -errno in error cases.
1060
 */
1061
int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1062
    const void *buf, int count)
1063
{
1064
    int ret;
1065

    
1066
    ret = bdrv_pwrite(bs, offset, buf, count);
1067
    if (ret < 0) {
1068
        return ret;
1069
    }
1070

    
1071
    /* No flush needed for cache=writethrough, it uses O_DSYNC */
1072
    if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1073
        bdrv_flush(bs);
1074
    }
1075

    
1076
    return 0;
1077
}
1078

    
1079
/*
1080
 * Writes to the file and ensures that no writes are reordered across this
1081
 * request (acts as a barrier)
1082
 *
1083
 * Returns 0 on success, -errno in error cases.
1084
 */
1085
int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1086
    const uint8_t *buf, int nb_sectors)
1087
{
1088
    return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1089
        buf, BDRV_SECTOR_SIZE * nb_sectors);
1090
}
1091

    
1092
/**
1093
 * Truncate file to 'offset' bytes (needed only for file protocols)
1094
 */
1095
int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1096
{
1097
    BlockDriver *drv = bs->drv;
1098
    int ret;
1099
    if (!drv)
1100
        return -ENOMEDIUM;
1101
    if (!drv->bdrv_truncate)
1102
        return -ENOTSUP;
1103
    if (bs->read_only)
1104
        return -EACCES;
1105
    ret = drv->bdrv_truncate(bs, offset);
1106
    if (ret == 0) {
1107
        ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1108
    }
1109
    return ret;
1110
}
1111

    
1112
/**
1113
 * Length of a file in bytes. Return < 0 if error or unknown.
1114
 */
1115
int64_t bdrv_getlength(BlockDriverState *bs)
1116
{
1117
    BlockDriver *drv = bs->drv;
1118
    if (!drv)
1119
        return -ENOMEDIUM;
1120

    
1121
    /* Fixed size devices use the total_sectors value for speed instead of
1122
       issuing a length query (like lseek) on each call.  Also, legacy block
1123
       drivers don't provide a bdrv_getlength function and must use
1124
       total_sectors. */
1125
    if (!bs->growable || !drv->bdrv_getlength) {
1126
        return bs->total_sectors * BDRV_SECTOR_SIZE;
1127
    }
1128
    return drv->bdrv_getlength(bs);
1129
}
1130

    
1131
/* return 0 as number of sectors if no device present or error */
1132
void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1133
{
1134
    int64_t length;
1135
    length = bdrv_getlength(bs);
1136
    if (length < 0)
1137
        length = 0;
1138
    else
1139
        length = length >> BDRV_SECTOR_BITS;
1140
    *nb_sectors_ptr = length;
1141
}
1142

    
1143
struct partition {
1144
        uint8_t boot_ind;           /* 0x80 - active */
1145
        uint8_t head;               /* starting head */
1146
        uint8_t sector;             /* starting sector */
1147
        uint8_t cyl;                /* starting cylinder */
1148
        uint8_t sys_ind;            /* What partition type */
1149
        uint8_t end_head;           /* end head */
1150
        uint8_t end_sector;         /* end sector */
1151
        uint8_t end_cyl;            /* end cylinder */
1152
        uint32_t start_sect;        /* starting sector counting from 0 */
1153
        uint32_t nr_sects;          /* nr of sectors in partition */
1154
} __attribute__((packed));
1155

    
1156
/* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1157
static int guess_disk_lchs(BlockDriverState *bs,
1158
                           int *pcylinders, int *pheads, int *psectors)
1159
{
1160
    uint8_t buf[BDRV_SECTOR_SIZE];
1161
    int ret, i, heads, sectors, cylinders;
1162
    struct partition *p;
1163
    uint32_t nr_sects;
1164
    uint64_t nb_sectors;
1165

    
1166
    bdrv_get_geometry(bs, &nb_sectors);
1167

    
1168
    ret = bdrv_read(bs, 0, buf, 1);
1169
    if (ret < 0)
1170
        return -1;
1171
    /* test msdos magic */
1172
    if (buf[510] != 0x55 || buf[511] != 0xaa)
1173
        return -1;
1174
    for(i = 0; i < 4; i++) {
1175
        p = ((struct partition *)(buf + 0x1be)) + i;
1176
        nr_sects = le32_to_cpu(p->nr_sects);
1177
        if (nr_sects && p->end_head) {
1178
            /* We make the assumption that the partition terminates on
1179
               a cylinder boundary */
1180
            heads = p->end_head + 1;
1181
            sectors = p->end_sector & 63;
1182
            if (sectors == 0)
1183
                continue;
1184
            cylinders = nb_sectors / (heads * sectors);
1185
            if (cylinders < 1 || cylinders > 16383)
1186
                continue;
1187
            *pheads = heads;
1188
            *psectors = sectors;
1189
            *pcylinders = cylinders;
1190
#if 0
1191
            printf("guessed geometry: LCHS=%d %d %d\n",
1192
                   cylinders, heads, sectors);
1193
#endif
1194
            return 0;
1195
        }
1196
    }
1197
    return -1;
1198
}
1199

    
1200
void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1201
{
1202
    int translation, lba_detected = 0;
1203
    int cylinders, heads, secs;
1204
    uint64_t nb_sectors;
1205

    
1206
    /* if a geometry hint is available, use it */
1207
    bdrv_get_geometry(bs, &nb_sectors);
1208
    bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1209
    translation = bdrv_get_translation_hint(bs);
1210
    if (cylinders != 0) {
1211
        *pcyls = cylinders;
1212
        *pheads = heads;
1213
        *psecs = secs;
1214
    } else {
1215
        if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1216
            if (heads > 16) {
1217
                /* if heads > 16, it means that a BIOS LBA
1218
                   translation was active, so the default
1219
                   hardware geometry is OK */
1220
                lba_detected = 1;
1221
                goto default_geometry;
1222
            } else {
1223
                *pcyls = cylinders;
1224
                *pheads = heads;
1225
                *psecs = secs;
1226
                /* disable any translation to be in sync with
1227
                   the logical geometry */
1228
                if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1229
                    bdrv_set_translation_hint(bs,
1230
                                              BIOS_ATA_TRANSLATION_NONE);
1231
                }
1232
            }
1233
        } else {
1234
        default_geometry:
1235
            /* if no geometry, use a standard physical disk geometry */
1236
            cylinders = nb_sectors / (16 * 63);
1237

    
1238
            if (cylinders > 16383)
1239
                cylinders = 16383;
1240
            else if (cylinders < 2)
1241
                cylinders = 2;
1242
            *pcyls = cylinders;
1243
            *pheads = 16;
1244
            *psecs = 63;
1245
            if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1246
                if ((*pcyls * *pheads) <= 131072) {
1247
                    bdrv_set_translation_hint(bs,
1248
                                              BIOS_ATA_TRANSLATION_LARGE);
1249
                } else {
1250
                    bdrv_set_translation_hint(bs,
1251
                                              BIOS_ATA_TRANSLATION_LBA);
1252
                }
1253
            }
1254
        }
1255
        bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1256
    }
1257
}
1258

    
1259
void bdrv_set_geometry_hint(BlockDriverState *bs,
1260
                            int cyls, int heads, int secs)
1261
{
1262
    bs->cyls = cyls;
1263
    bs->heads = heads;
1264
    bs->secs = secs;
1265
}
1266

    
1267
void bdrv_set_type_hint(BlockDriverState *bs, int type)
1268
{
1269
    bs->type = type;
1270
    bs->removable = ((type == BDRV_TYPE_CDROM ||
1271
                      type == BDRV_TYPE_FLOPPY));
1272
}
1273

    
1274
void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1275
{
1276
    bs->translation = translation;
1277
}
1278

    
1279
void bdrv_get_geometry_hint(BlockDriverState *bs,
1280
                            int *pcyls, int *pheads, int *psecs)
1281
{
1282
    *pcyls = bs->cyls;
1283
    *pheads = bs->heads;
1284
    *psecs = bs->secs;
1285
}
1286

    
1287
int bdrv_get_type_hint(BlockDriverState *bs)
1288
{
1289
    return bs->type;
1290
}
1291

    
1292
int bdrv_get_translation_hint(BlockDriverState *bs)
1293
{
1294
    return bs->translation;
1295
}
1296

    
1297
void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1298
                       BlockErrorAction on_write_error)
1299
{
1300
    bs->on_read_error = on_read_error;
1301
    bs->on_write_error = on_write_error;
1302
}
1303

    
1304
BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1305
{
1306
    return is_read ? bs->on_read_error : bs->on_write_error;
1307
}
1308

    
1309
void bdrv_set_removable(BlockDriverState *bs, int removable)
1310
{
1311
    bs->removable = removable;
1312
    if (removable && bs == bs_snapshots) {
1313
        bs_snapshots = NULL;
1314
    }
1315
}
1316

    
1317
int bdrv_is_removable(BlockDriverState *bs)
1318
{
1319
    return bs->removable;
1320
}
1321

    
1322
int bdrv_is_read_only(BlockDriverState *bs)
1323
{
1324
    return bs->read_only;
1325
}
1326

    
1327
int bdrv_is_sg(BlockDriverState *bs)
1328
{
1329
    return bs->sg;
1330
}
1331

    
1332
int bdrv_enable_write_cache(BlockDriverState *bs)
1333
{
1334
    return bs->enable_write_cache;
1335
}
1336

    
1337
/* XXX: no longer used */
1338
void bdrv_set_change_cb(BlockDriverState *bs,
1339
                        void (*change_cb)(void *opaque), void *opaque)
1340
{
1341
    bs->change_cb = change_cb;
1342
    bs->change_opaque = opaque;
1343
}
1344

    
1345
int bdrv_is_encrypted(BlockDriverState *bs)
1346
{
1347
    if (bs->backing_hd && bs->backing_hd->encrypted)
1348
        return 1;
1349
    return bs->encrypted;
1350
}
1351

    
1352
int bdrv_key_required(BlockDriverState *bs)
1353
{
1354
    BlockDriverState *backing_hd = bs->backing_hd;
1355

    
1356
    if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1357
        return 1;
1358
    return (bs->encrypted && !bs->valid_key);
1359
}
1360

    
1361
int bdrv_set_key(BlockDriverState *bs, const char *key)
1362
{
1363
    int ret;
1364
    if (bs->backing_hd && bs->backing_hd->encrypted) {
1365
        ret = bdrv_set_key(bs->backing_hd, key);
1366
        if (ret < 0)
1367
            return ret;
1368
        if (!bs->encrypted)
1369
            return 0;
1370
    }
1371
    if (!bs->encrypted) {
1372
        return -EINVAL;
1373
    } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1374
        return -ENOMEDIUM;
1375
    }
1376
    ret = bs->drv->bdrv_set_key(bs, key);
1377
    if (ret < 0) {
1378
        bs->valid_key = 0;
1379
    } else if (!bs->valid_key) {
1380
        bs->valid_key = 1;
1381
        /* call the change callback now, we skipped it on open */
1382
        bs->media_changed = 1;
1383
        if (bs->change_cb)
1384
            bs->change_cb(bs->change_opaque);
1385
    }
1386
    return ret;
1387
}
1388

    
1389
void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1390
{
1391
    if (!bs->drv) {
1392
        buf[0] = '\0';
1393
    } else {
1394
        pstrcpy(buf, buf_size, bs->drv->format_name);
1395
    }
1396
}
1397

    
1398
void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1399
                         void *opaque)
1400
{
1401
    BlockDriver *drv;
1402

    
1403
    QLIST_FOREACH(drv, &bdrv_drivers, list) {
1404
        it(opaque, drv->format_name);
1405
    }
1406
}
1407

    
1408
BlockDriverState *bdrv_find(const char *name)
1409
{
1410
    BlockDriverState *bs;
1411

    
1412
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1413
        if (!strcmp(name, bs->device_name)) {
1414
            return bs;
1415
        }
1416
    }
1417
    return NULL;
1418
}
1419

    
1420
BlockDriverState *bdrv_next(BlockDriverState *bs)
1421
{
1422
    if (!bs) {
1423
        return QTAILQ_FIRST(&bdrv_states);
1424
    }
1425
    return QTAILQ_NEXT(bs, list);
1426
}
1427

    
1428
void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1429
{
1430
    BlockDriverState *bs;
1431

    
1432
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1433
        it(opaque, bs);
1434
    }
1435
}
1436

    
1437
const char *bdrv_get_device_name(BlockDriverState *bs)
1438
{
1439
    return bs->device_name;
1440
}
1441

    
1442
void bdrv_flush(BlockDriverState *bs)
1443
{
1444
    if (bs->open_flags & BDRV_O_NO_FLUSH) {
1445
        return;
1446
    }
1447

    
1448
    if (bs->drv && bs->drv->bdrv_flush)
1449
        bs->drv->bdrv_flush(bs);
1450
}
1451

    
1452
void bdrv_flush_all(void)
1453
{
1454
    BlockDriverState *bs;
1455

    
1456
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1457
        if (bs->drv && !bdrv_is_read_only(bs) &&
1458
            (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1459
            bdrv_flush(bs);
1460
        }
1461
    }
1462
}
1463

    
1464
int bdrv_has_zero_init(BlockDriverState *bs)
1465
{
1466
    assert(bs->drv);
1467

    
1468
    if (bs->drv->no_zero_init) {
1469
        return 0;
1470
    } else if (bs->file) {
1471
        return bdrv_has_zero_init(bs->file);
1472
    }
1473

    
1474
    return 1;
1475
}
1476

    
1477
/*
1478
 * Returns true iff the specified sector is present in the disk image. Drivers
1479
 * not implementing the functionality are assumed to not support backing files,
1480
 * hence all their sectors are reported as allocated.
1481
 *
1482
 * 'pnum' is set to the number of sectors (including and immediately following
1483
 * the specified sector) that are known to be in the same
1484
 * allocated/unallocated state.
1485
 *
1486
 * 'nb_sectors' is the max value 'pnum' should be set to.
1487
 */
1488
int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1489
        int *pnum)
1490
{
1491
    int64_t n;
1492
    if (!bs->drv->bdrv_is_allocated) {
1493
        if (sector_num >= bs->total_sectors) {
1494
            *pnum = 0;
1495
            return 0;
1496
        }
1497
        n = bs->total_sectors - sector_num;
1498
        *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1499
        return 1;
1500
    }
1501
    return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1502
}
1503

    
1504
void bdrv_mon_event(const BlockDriverState *bdrv,
1505
                    BlockMonEventAction action, int is_read)
1506
{
1507
    QObject *data;
1508
    const char *action_str;
1509

    
1510
    switch (action) {
1511
    case BDRV_ACTION_REPORT:
1512
        action_str = "report";
1513
        break;
1514
    case BDRV_ACTION_IGNORE:
1515
        action_str = "ignore";
1516
        break;
1517
    case BDRV_ACTION_STOP:
1518
        action_str = "stop";
1519
        break;
1520
    default:
1521
        abort();
1522
    }
1523

    
1524
    data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1525
                              bdrv->device_name,
1526
                              action_str,
1527
                              is_read ? "read" : "write");
1528
    monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1529

    
1530
    qobject_decref(data);
1531
}
1532

    
1533
static void bdrv_print_dict(QObject *obj, void *opaque)
1534
{
1535
    QDict *bs_dict;
1536
    Monitor *mon = opaque;
1537

    
1538
    bs_dict = qobject_to_qdict(obj);
1539

    
1540
    monitor_printf(mon, "%s: type=%s removable=%d",
1541
                        qdict_get_str(bs_dict, "device"),
1542
                        qdict_get_str(bs_dict, "type"),
1543
                        qdict_get_bool(bs_dict, "removable"));
1544

    
1545
    if (qdict_get_bool(bs_dict, "removable")) {
1546
        monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1547
    }
1548

    
1549
    if (qdict_haskey(bs_dict, "inserted")) {
1550
        QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1551

    
1552
        monitor_printf(mon, " file=");
1553
        monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1554
        if (qdict_haskey(qdict, "backing_file")) {
1555
            monitor_printf(mon, " backing_file=");
1556
            monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1557
        }
1558
        monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1559
                            qdict_get_bool(qdict, "ro"),
1560
                            qdict_get_str(qdict, "drv"),
1561
                            qdict_get_bool(qdict, "encrypted"));
1562
    } else {
1563
        monitor_printf(mon, " [not inserted]");
1564
    }
1565

    
1566
    monitor_printf(mon, "\n");
1567
}
1568

    
1569
void bdrv_info_print(Monitor *mon, const QObject *data)
1570
{
1571
    qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1572
}
1573

    
1574
void bdrv_info(Monitor *mon, QObject **ret_data)
1575
{
1576
    QList *bs_list;
1577
    BlockDriverState *bs;
1578

    
1579
    bs_list = qlist_new();
1580

    
1581
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1582
        QObject *bs_obj;
1583
        const char *type = "unknown";
1584

    
1585
        switch(bs->type) {
1586
        case BDRV_TYPE_HD:
1587
            type = "hd";
1588
            break;
1589
        case BDRV_TYPE_CDROM:
1590
            type = "cdrom";
1591
            break;
1592
        case BDRV_TYPE_FLOPPY:
1593
            type = "floppy";
1594
            break;
1595
        }
1596

    
1597
        bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1598
                                    "'removable': %i, 'locked': %i }",
1599
                                    bs->device_name, type, bs->removable,
1600
                                    bs->locked);
1601

    
1602
        if (bs->drv) {
1603
            QObject *obj;
1604
            QDict *bs_dict = qobject_to_qdict(bs_obj);
1605

    
1606
            obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1607
                                     "'encrypted': %i }",
1608
                                     bs->filename, bs->read_only,
1609
                                     bs->drv->format_name,
1610
                                     bdrv_is_encrypted(bs));
1611
            if (bs->backing_file[0] != '\0') {
1612
                QDict *qdict = qobject_to_qdict(obj);
1613
                qdict_put(qdict, "backing_file",
1614
                          qstring_from_str(bs->backing_file));
1615
            }
1616

    
1617
            qdict_put_obj(bs_dict, "inserted", obj);
1618
        }
1619
        qlist_append_obj(bs_list, bs_obj);
1620
    }
1621

    
1622
    *ret_data = QOBJECT(bs_list);
1623
}
1624

    
1625
static void bdrv_stats_iter(QObject *data, void *opaque)
1626
{
1627
    QDict *qdict;
1628
    Monitor *mon = opaque;
1629

    
1630
    qdict = qobject_to_qdict(data);
1631
    monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1632

    
1633
    qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1634
    monitor_printf(mon, " rd_bytes=%" PRId64
1635
                        " wr_bytes=%" PRId64
1636
                        " rd_operations=%" PRId64
1637
                        " wr_operations=%" PRId64
1638
                        "\n",
1639
                        qdict_get_int(qdict, "rd_bytes"),
1640
                        qdict_get_int(qdict, "wr_bytes"),
1641
                        qdict_get_int(qdict, "rd_operations"),
1642
                        qdict_get_int(qdict, "wr_operations"));
1643
}
1644

    
1645
void bdrv_stats_print(Monitor *mon, const QObject *data)
1646
{
1647
    qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1648
}
1649

    
1650
static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1651
{
1652
    QObject *res;
1653
    QDict *dict;
1654

    
1655
    res = qobject_from_jsonf("{ 'stats': {"
1656
                             "'rd_bytes': %" PRId64 ","
1657
                             "'wr_bytes': %" PRId64 ","
1658
                             "'rd_operations': %" PRId64 ","
1659
                             "'wr_operations': %" PRId64 ","
1660
                             "'wr_highest_offset': %" PRId64
1661
                             "} }",
1662
                             bs->rd_bytes, bs->wr_bytes,
1663
                             bs->rd_ops, bs->wr_ops,
1664
                             bs->wr_highest_sector *
1665
                             (uint64_t)BDRV_SECTOR_SIZE);
1666
    dict  = qobject_to_qdict(res);
1667

    
1668
    if (*bs->device_name) {
1669
        qdict_put(dict, "device", qstring_from_str(bs->device_name));
1670
    }
1671

    
1672
    if (bs->file) {
1673
        QObject *parent = bdrv_info_stats_bs(bs->file);
1674
        qdict_put_obj(dict, "parent", parent);
1675
    }
1676

    
1677
    return res;
1678
}
1679

    
1680
void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1681
{
1682
    QObject *obj;
1683
    QList *devices;
1684
    BlockDriverState *bs;
1685

    
1686
    devices = qlist_new();
1687

    
1688
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1689
        obj = bdrv_info_stats_bs(bs);
1690
        qlist_append_obj(devices, obj);
1691
    }
1692

    
1693
    *ret_data = QOBJECT(devices);
1694
}
1695

    
1696
const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1697
{
1698
    if (bs->backing_hd && bs->backing_hd->encrypted)
1699
        return bs->backing_file;
1700
    else if (bs->encrypted)
1701
        return bs->filename;
1702
    else
1703
        return NULL;
1704
}
1705

    
1706
void bdrv_get_backing_filename(BlockDriverState *bs,
1707
                               char *filename, int filename_size)
1708
{
1709
    if (!bs->backing_file) {
1710
        pstrcpy(filename, filename_size, "");
1711
    } else {
1712
        pstrcpy(filename, filename_size, bs->backing_file);
1713
    }
1714
}
1715

    
1716
int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1717
                          const uint8_t *buf, int nb_sectors)
1718
{
1719
    BlockDriver *drv = bs->drv;
1720
    if (!drv)
1721
        return -ENOMEDIUM;
1722
    if (!drv->bdrv_write_compressed)
1723
        return -ENOTSUP;
1724
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1725
        return -EIO;
1726

    
1727
    if (bs->dirty_bitmap) {
1728
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1729
    }
1730

    
1731
    return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1732
}
1733

    
1734
int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1735
{
1736
    BlockDriver *drv = bs->drv;
1737
    if (!drv)
1738
        return -ENOMEDIUM;
1739
    if (!drv->bdrv_get_info)
1740
        return -ENOTSUP;
1741
    memset(bdi, 0, sizeof(*bdi));
1742
    return drv->bdrv_get_info(bs, bdi);
1743
}
1744

    
1745
int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1746
                      int64_t pos, int size)
1747
{
1748
    BlockDriver *drv = bs->drv;
1749
    if (!drv)
1750
        return -ENOMEDIUM;
1751
    if (drv->bdrv_save_vmstate)
1752
        return drv->bdrv_save_vmstate(bs, buf, pos, size);
1753
    if (bs->file)
1754
        return bdrv_save_vmstate(bs->file, buf, pos, size);
1755
    return -ENOTSUP;
1756
}
1757

    
1758
int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1759
                      int64_t pos, int size)
1760
{
1761
    BlockDriver *drv = bs->drv;
1762
    if (!drv)
1763
        return -ENOMEDIUM;
1764
    if (drv->bdrv_load_vmstate)
1765
        return drv->bdrv_load_vmstate(bs, buf, pos, size);
1766
    if (bs->file)
1767
        return bdrv_load_vmstate(bs->file, buf, pos, size);
1768
    return -ENOTSUP;
1769
}
1770

    
1771
void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1772
{
1773
    BlockDriver *drv = bs->drv;
1774

    
1775
    if (!drv || !drv->bdrv_debug_event) {
1776
        return;
1777
    }
1778

    
1779
    return drv->bdrv_debug_event(bs, event);
1780

    
1781
}
1782

    
1783
/**************************************************************/
1784
/* handling of snapshots */
1785

    
1786
int bdrv_can_snapshot(BlockDriverState *bs)
1787
{
1788
    BlockDriver *drv = bs->drv;
1789
    if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1790
        return 0;
1791
    }
1792

    
1793
    if (!drv->bdrv_snapshot_create) {
1794
        if (bs->file != NULL) {
1795
            return bdrv_can_snapshot(bs->file);
1796
        }
1797
        return 0;
1798
    }
1799

    
1800
    return 1;
1801
}
1802

    
1803
BlockDriverState *bdrv_snapshots(void)
1804
{
1805
    BlockDriverState *bs;
1806

    
1807
    if (bs_snapshots) {
1808
        return bs_snapshots;
1809
    }
1810

    
1811
    bs = NULL;
1812
    while ((bs = bdrv_next(bs))) {
1813
        if (bdrv_can_snapshot(bs)) {
1814
            bs_snapshots = bs;
1815
            return bs;
1816
        }
1817
    }
1818
    return NULL;
1819
}
1820

    
1821
int bdrv_snapshot_create(BlockDriverState *bs,
1822
                         QEMUSnapshotInfo *sn_info)
1823
{
1824
    BlockDriver *drv = bs->drv;
1825
    if (!drv)
1826
        return -ENOMEDIUM;
1827
    if (drv->bdrv_snapshot_create)
1828
        return drv->bdrv_snapshot_create(bs, sn_info);
1829
    if (bs->file)
1830
        return bdrv_snapshot_create(bs->file, sn_info);
1831
    return -ENOTSUP;
1832
}
1833

    
1834
int bdrv_snapshot_goto(BlockDriverState *bs,
1835
                       const char *snapshot_id)
1836
{
1837
    BlockDriver *drv = bs->drv;
1838
    int ret, open_ret;
1839

    
1840
    if (!drv)
1841
        return -ENOMEDIUM;
1842
    if (drv->bdrv_snapshot_goto)
1843
        return drv->bdrv_snapshot_goto(bs, snapshot_id);
1844

    
1845
    if (bs->file) {
1846
        drv->bdrv_close(bs);
1847
        ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1848
        open_ret = drv->bdrv_open(bs, bs->open_flags);
1849
        if (open_ret < 0) {
1850
            bdrv_delete(bs->file);
1851
            bs->drv = NULL;
1852
            return open_ret;
1853
        }
1854
        return ret;
1855
    }
1856

    
1857
    return -ENOTSUP;
1858
}
1859

    
1860
int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1861
{
1862
    BlockDriver *drv = bs->drv;
1863
    if (!drv)
1864
        return -ENOMEDIUM;
1865
    if (drv->bdrv_snapshot_delete)
1866
        return drv->bdrv_snapshot_delete(bs, snapshot_id);
1867
    if (bs->file)
1868
        return bdrv_snapshot_delete(bs->file, snapshot_id);
1869
    return -ENOTSUP;
1870
}
1871

    
1872
int bdrv_snapshot_list(BlockDriverState *bs,
1873
                       QEMUSnapshotInfo **psn_info)
1874
{
1875
    BlockDriver *drv = bs->drv;
1876
    if (!drv)
1877
        return -ENOMEDIUM;
1878
    if (drv->bdrv_snapshot_list)
1879
        return drv->bdrv_snapshot_list(bs, psn_info);
1880
    if (bs->file)
1881
        return bdrv_snapshot_list(bs->file, psn_info);
1882
    return -ENOTSUP;
1883
}
1884

    
1885
#define NB_SUFFIXES 4
1886

    
1887
char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1888
{
1889
    static const char suffixes[NB_SUFFIXES] = "KMGT";
1890
    int64_t base;
1891
    int i;
1892

    
1893
    if (size <= 999) {
1894
        snprintf(buf, buf_size, "%" PRId64, size);
1895
    } else {
1896
        base = 1024;
1897
        for(i = 0; i < NB_SUFFIXES; i++) {
1898
            if (size < (10 * base)) {
1899
                snprintf(buf, buf_size, "%0.1f%c",
1900
                         (double)size / base,
1901
                         suffixes[i]);
1902
                break;
1903
            } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1904
                snprintf(buf, buf_size, "%" PRId64 "%c",
1905
                         ((size + (base >> 1)) / base),
1906
                         suffixes[i]);
1907
                break;
1908
            }
1909
            base = base * 1024;
1910
        }
1911
    }
1912
    return buf;
1913
}
1914

    
1915
char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1916
{
1917
    char buf1[128], date_buf[128], clock_buf[128];
1918
#ifdef _WIN32
1919
    struct tm *ptm;
1920
#else
1921
    struct tm tm;
1922
#endif
1923
    time_t ti;
1924
    int64_t secs;
1925

    
1926
    if (!sn) {
1927
        snprintf(buf, buf_size,
1928
                 "%-10s%-20s%7s%20s%15s",
1929
                 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1930
    } else {
1931
        ti = sn->date_sec;
1932
#ifdef _WIN32
1933
        ptm = localtime(&ti);
1934
        strftime(date_buf, sizeof(date_buf),
1935
                 "%Y-%m-%d %H:%M:%S", ptm);
1936
#else
1937
        localtime_r(&ti, &tm);
1938
        strftime(date_buf, sizeof(date_buf),
1939
                 "%Y-%m-%d %H:%M:%S", &tm);
1940
#endif
1941
        secs = sn->vm_clock_nsec / 1000000000;
1942
        snprintf(clock_buf, sizeof(clock_buf),
1943
                 "%02d:%02d:%02d.%03d",
1944
                 (int)(secs / 3600),
1945
                 (int)((secs / 60) % 60),
1946
                 (int)(secs % 60),
1947
                 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1948
        snprintf(buf, buf_size,
1949
                 "%-10s%-20s%7s%20s%15s",
1950
                 sn->id_str, sn->name,
1951
                 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1952
                 date_buf,
1953
                 clock_buf);
1954
    }
1955
    return buf;
1956
}
1957

    
1958

    
1959
/**************************************************************/
1960
/* async I/Os */
1961

    
1962
BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1963
                                 QEMUIOVector *qiov, int nb_sectors,
1964
                                 BlockDriverCompletionFunc *cb, void *opaque)
1965
{
1966
    BlockDriver *drv = bs->drv;
1967
    BlockDriverAIOCB *ret;
1968

    
1969
    if (!drv)
1970
        return NULL;
1971
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1972
        return NULL;
1973

    
1974
    ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1975
                              cb, opaque);
1976

    
1977
    if (ret) {
1978
        /* Update stats even though technically transfer has not happened. */
1979
        bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1980
        bs->rd_ops ++;
1981
    }
1982

    
1983
    return ret;
1984
}
1985

    
1986
BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1987
                                  QEMUIOVector *qiov, int nb_sectors,
1988
                                  BlockDriverCompletionFunc *cb, void *opaque)
1989
{
1990
    BlockDriver *drv = bs->drv;
1991
    BlockDriverAIOCB *ret;
1992

    
1993
    if (!drv)
1994
        return NULL;
1995
    if (bs->read_only)
1996
        return NULL;
1997
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1998
        return NULL;
1999

    
2000
    if (bs->dirty_bitmap) {
2001
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2002
    }
2003

    
2004
    ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2005
                               cb, opaque);
2006

    
2007
    if (ret) {
2008
        /* Update stats even though technically transfer has not happened. */
2009
        bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2010
        bs->wr_ops ++;
2011
        if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2012
            bs->wr_highest_sector = sector_num + nb_sectors - 1;
2013
        }
2014
    }
2015

    
2016
    return ret;
2017
}
2018

    
2019

    
2020
typedef struct MultiwriteCB {
2021
    int error;
2022
    int num_requests;
2023
    int num_callbacks;
2024
    struct {
2025
        BlockDriverCompletionFunc *cb;
2026
        void *opaque;
2027
        QEMUIOVector *free_qiov;
2028
        void *free_buf;
2029
    } callbacks[];
2030
} MultiwriteCB;
2031

    
2032
static void multiwrite_user_cb(MultiwriteCB *mcb)
2033
{
2034
    int i;
2035

    
2036
    for (i = 0; i < mcb->num_callbacks; i++) {
2037
        mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2038
        if (mcb->callbacks[i].free_qiov) {
2039
            qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2040
        }
2041
        qemu_free(mcb->callbacks[i].free_qiov);
2042
        qemu_vfree(mcb->callbacks[i].free_buf);
2043
    }
2044
}
2045

    
2046
static void multiwrite_cb(void *opaque, int ret)
2047
{
2048
    MultiwriteCB *mcb = opaque;
2049

    
2050
    if (ret < 0 && !mcb->error) {
2051
        mcb->error = ret;
2052
    }
2053

    
2054
    mcb->num_requests--;
2055
    if (mcb->num_requests == 0) {
2056
        multiwrite_user_cb(mcb);
2057
        qemu_free(mcb);
2058
    }
2059
}
2060

    
2061
static int multiwrite_req_compare(const void *a, const void *b)
2062
{
2063
    const BlockRequest *req1 = a, *req2 = b;
2064

    
2065
    /*
2066
     * Note that we can't simply subtract req2->sector from req1->sector
2067
     * here as that could overflow the return value.
2068
     */
2069
    if (req1->sector > req2->sector) {
2070
        return 1;
2071
    } else if (req1->sector < req2->sector) {
2072
        return -1;
2073
    } else {
2074
        return 0;
2075
    }
2076
}
2077

    
2078
/*
2079
 * Takes a bunch of requests and tries to merge them. Returns the number of
2080
 * requests that remain after merging.
2081
 */
2082
static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2083
    int num_reqs, MultiwriteCB *mcb)
2084
{
2085
    int i, outidx;
2086

    
2087
    // Sort requests by start sector
2088
    qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2089

    
2090
    // Check if adjacent requests touch the same clusters. If so, combine them,
2091
    // filling up gaps with zero sectors.
2092
    outidx = 0;
2093
    for (i = 1; i < num_reqs; i++) {
2094
        int merge = 0;
2095
        int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2096

    
2097
        // This handles the cases that are valid for all block drivers, namely
2098
        // exactly sequential writes and overlapping writes.
2099
        if (reqs[i].sector <= oldreq_last) {
2100
            merge = 1;
2101
        }
2102

    
2103
        // The block driver may decide that it makes sense to combine requests
2104
        // even if there is a gap of some sectors between them. In this case,
2105
        // the gap is filled with zeros (therefore only applicable for yet
2106
        // unused space in format like qcow2).
2107
        if (!merge && bs->drv->bdrv_merge_requests) {
2108
            merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2109
        }
2110

    
2111
        if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2112
            merge = 0;
2113
        }
2114

    
2115
        if (merge) {
2116
            size_t size;
2117
            QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2118
            qemu_iovec_init(qiov,
2119
                reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2120

    
2121
            // Add the first request to the merged one. If the requests are
2122
            // overlapping, drop the last sectors of the first request.
2123
            size = (reqs[i].sector - reqs[outidx].sector) << 9;
2124
            qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2125

    
2126
            // We might need to add some zeros between the two requests
2127
            if (reqs[i].sector > oldreq_last) {
2128
                size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2129
                uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2130
                memset(buf, 0, zero_bytes);
2131
                qemu_iovec_add(qiov, buf, zero_bytes);
2132
                mcb->callbacks[i].free_buf = buf;
2133
            }
2134

    
2135
            // Add the second request
2136
            qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2137

    
2138
            reqs[outidx].nb_sectors = qiov->size >> 9;
2139
            reqs[outidx].qiov = qiov;
2140

    
2141
            mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2142
        } else {
2143
            outidx++;
2144
            reqs[outidx].sector     = reqs[i].sector;
2145
            reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2146
            reqs[outidx].qiov       = reqs[i].qiov;
2147
        }
2148
    }
2149

    
2150
    return outidx + 1;
2151
}
2152

    
2153
/*
2154
 * Submit multiple AIO write requests at once.
2155
 *
2156
 * On success, the function returns 0 and all requests in the reqs array have
2157
 * been submitted. In error case this function returns -1, and any of the
2158
 * requests may or may not be submitted yet. In particular, this means that the
2159
 * callback will be called for some of the requests, for others it won't. The
2160
 * caller must check the error field of the BlockRequest to wait for the right
2161
 * callbacks (if error != 0, no callback will be called).
2162
 *
2163
 * The implementation may modify the contents of the reqs array, e.g. to merge
2164
 * requests. However, the fields opaque and error are left unmodified as they
2165
 * are used to signal failure for a single request to the caller.
2166
 */
2167
int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2168
{
2169
    BlockDriverAIOCB *acb;
2170
    MultiwriteCB *mcb;
2171
    int i;
2172

    
2173
    if (num_reqs == 0) {
2174
        return 0;
2175
    }
2176

    
2177
    // Create MultiwriteCB structure
2178
    mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2179
    mcb->num_requests = 0;
2180
    mcb->num_callbacks = num_reqs;
2181

    
2182
    for (i = 0; i < num_reqs; i++) {
2183
        mcb->callbacks[i].cb = reqs[i].cb;
2184
        mcb->callbacks[i].opaque = reqs[i].opaque;
2185
    }
2186

    
2187
    // Check for mergable requests
2188
    num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2189

    
2190
    /*
2191
     * Run the aio requests. As soon as one request can't be submitted
2192
     * successfully, fail all requests that are not yet submitted (we must
2193
     * return failure for all requests anyway)
2194
     *
2195
     * num_requests cannot be set to the right value immediately: If
2196
     * bdrv_aio_writev fails for some request, num_requests would be too high
2197
     * and therefore multiwrite_cb() would never recognize the multiwrite
2198
     * request as completed. We also cannot use the loop variable i to set it
2199
     * when the first request fails because the callback may already have been
2200
     * called for previously submitted requests. Thus, num_requests must be
2201
     * incremented for each request that is submitted.
2202
     *
2203
     * The problem that callbacks may be called early also means that we need
2204
     * to take care that num_requests doesn't become 0 before all requests are
2205
     * submitted - multiwrite_cb() would consider the multiwrite request
2206
     * completed. A dummy request that is "completed" by a manual call to
2207
     * multiwrite_cb() takes care of this.
2208
     */
2209
    mcb->num_requests = 1;
2210

    
2211
    for (i = 0; i < num_reqs; i++) {
2212
        mcb->num_requests++;
2213
        acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2214
            reqs[i].nb_sectors, multiwrite_cb, mcb);
2215

    
2216
        if (acb == NULL) {
2217
            // We can only fail the whole thing if no request has been
2218
            // submitted yet. Otherwise we'll wait for the submitted AIOs to
2219
            // complete and report the error in the callback.
2220
            if (i == 0) {
2221
                goto fail;
2222
            } else {
2223
                multiwrite_cb(mcb, -EIO);
2224
                break;
2225
            }
2226
        }
2227
    }
2228

    
2229
    /* Complete the dummy request */
2230
    multiwrite_cb(mcb, 0);
2231

    
2232
    return 0;
2233

    
2234
fail:
2235
    for (i = 0; i < mcb->num_callbacks; i++) {
2236
        reqs[i].error = -EIO;
2237
    }
2238
    qemu_free(mcb);
2239
    return -1;
2240
}
2241

    
2242
BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2243
        BlockDriverCompletionFunc *cb, void *opaque)
2244
{
2245
    BlockDriver *drv = bs->drv;
2246

    
2247
    if (bs->open_flags & BDRV_O_NO_FLUSH) {
2248
        return bdrv_aio_noop_em(bs, cb, opaque);
2249
    }
2250

    
2251
    if (!drv)
2252
        return NULL;
2253
    return drv->bdrv_aio_flush(bs, cb, opaque);
2254
}
2255

    
2256
void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2257
{
2258
    acb->pool->cancel(acb);
2259
}
2260

    
2261

    
2262
/**************************************************************/
2263
/* async block device emulation */
2264

    
2265
typedef struct BlockDriverAIOCBSync {
2266
    BlockDriverAIOCB common;
2267
    QEMUBH *bh;
2268
    int ret;
2269
    /* vector translation state */
2270
    QEMUIOVector *qiov;
2271
    uint8_t *bounce;
2272
    int is_write;
2273
} BlockDriverAIOCBSync;
2274

    
2275
static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2276
{
2277
    BlockDriverAIOCBSync *acb =
2278
        container_of(blockacb, BlockDriverAIOCBSync, common);
2279
    qemu_bh_delete(acb->bh);
2280
    acb->bh = NULL;
2281
    qemu_aio_release(acb);
2282
}
2283

    
2284
static AIOPool bdrv_em_aio_pool = {
2285
    .aiocb_size         = sizeof(BlockDriverAIOCBSync),
2286
    .cancel             = bdrv_aio_cancel_em,
2287
};
2288

    
2289
static void bdrv_aio_bh_cb(void *opaque)
2290
{
2291
    BlockDriverAIOCBSync *acb = opaque;
2292

    
2293
    if (!acb->is_write)
2294
        qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2295
    qemu_vfree(acb->bounce);
2296
    acb->common.cb(acb->common.opaque, acb->ret);
2297
    qemu_bh_delete(acb->bh);
2298
    acb->bh = NULL;
2299
    qemu_aio_release(acb);
2300
}
2301

    
2302
static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2303
                                            int64_t sector_num,
2304
                                            QEMUIOVector *qiov,
2305
                                            int nb_sectors,
2306
                                            BlockDriverCompletionFunc *cb,
2307
                                            void *opaque,
2308
                                            int is_write)
2309

    
2310
{
2311
    BlockDriverAIOCBSync *acb;
2312

    
2313
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2314
    acb->is_write = is_write;
2315
    acb->qiov = qiov;
2316
    acb->bounce = qemu_blockalign(bs, qiov->size);
2317

    
2318
    if (!acb->bh)
2319
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2320

    
2321
    if (is_write) {
2322
        qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2323
        acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2324
    } else {
2325
        acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2326
    }
2327

    
2328
    qemu_bh_schedule(acb->bh);
2329

    
2330
    return &acb->common;
2331
}
2332

    
2333
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2334
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2335
        BlockDriverCompletionFunc *cb, void *opaque)
2336
{
2337
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2338
}
2339

    
2340
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2341
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2342
        BlockDriverCompletionFunc *cb, void *opaque)
2343
{
2344
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2345
}
2346

    
2347
static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2348
        BlockDriverCompletionFunc *cb, void *opaque)
2349
{
2350
    BlockDriverAIOCBSync *acb;
2351

    
2352
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2353
    acb->is_write = 1; /* don't bounce in the completion hadler */
2354
    acb->qiov = NULL;
2355
    acb->bounce = NULL;
2356
    acb->ret = 0;
2357

    
2358
    if (!acb->bh)
2359
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2360

    
2361
    bdrv_flush(bs);
2362
    qemu_bh_schedule(acb->bh);
2363
    return &acb->common;
2364
}
2365

    
2366
static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2367
        BlockDriverCompletionFunc *cb, void *opaque)
2368
{
2369
    BlockDriverAIOCBSync *acb;
2370

    
2371
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2372
    acb->is_write = 1; /* don't bounce in the completion handler */
2373
    acb->qiov = NULL;
2374
    acb->bounce = NULL;
2375
    acb->ret = 0;
2376

    
2377
    if (!acb->bh) {
2378
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2379
    }
2380

    
2381
    qemu_bh_schedule(acb->bh);
2382
    return &acb->common;
2383
}
2384

    
2385
/**************************************************************/
2386
/* sync block device emulation */
2387

    
2388
static void bdrv_rw_em_cb(void *opaque, int ret)
2389
{
2390
    *(int *)opaque = ret;
2391
}
2392

    
2393
#define NOT_DONE 0x7fffffff
2394

    
2395
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2396
                        uint8_t *buf, int nb_sectors)
2397
{
2398
    int async_ret;
2399
    BlockDriverAIOCB *acb;
2400
    struct iovec iov;
2401
    QEMUIOVector qiov;
2402

    
2403
    async_context_push();
2404

    
2405
    async_ret = NOT_DONE;
2406
    iov.iov_base = (void *)buf;
2407
    iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2408
    qemu_iovec_init_external(&qiov, &iov, 1);
2409
    acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2410
        bdrv_rw_em_cb, &async_ret);
2411
    if (acb == NULL) {
2412
        async_ret = -1;
2413
        goto fail;
2414
    }
2415

    
2416
    while (async_ret == NOT_DONE) {
2417
        qemu_aio_wait();
2418
    }
2419

    
2420

    
2421
fail:
2422
    async_context_pop();
2423
    return async_ret;
2424
}
2425

    
2426
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2427
                         const uint8_t *buf, int nb_sectors)
2428
{
2429
    int async_ret;
2430
    BlockDriverAIOCB *acb;
2431
    struct iovec iov;
2432
    QEMUIOVector qiov;
2433

    
2434
    async_context_push();
2435

    
2436
    async_ret = NOT_DONE;
2437
    iov.iov_base = (void *)buf;
2438
    iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2439
    qemu_iovec_init_external(&qiov, &iov, 1);
2440
    acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2441
        bdrv_rw_em_cb, &async_ret);
2442
    if (acb == NULL) {
2443
        async_ret = -1;
2444
        goto fail;
2445
    }
2446
    while (async_ret == NOT_DONE) {
2447
        qemu_aio_wait();
2448
    }
2449

    
2450
fail:
2451
    async_context_pop();
2452
    return async_ret;
2453
}
2454

    
2455
void bdrv_init(void)
2456
{
2457
    module_call_init(MODULE_INIT_BLOCK);
2458
}
2459

    
2460
void bdrv_init_with_whitelist(void)
2461
{
2462
    use_bdrv_whitelist = 1;
2463
    bdrv_init();
2464
}
2465

    
2466
void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2467
                   BlockDriverCompletionFunc *cb, void *opaque)
2468
{
2469
    BlockDriverAIOCB *acb;
2470

    
2471
    if (pool->free_aiocb) {
2472
        acb = pool->free_aiocb;
2473
        pool->free_aiocb = acb->next;
2474
    } else {
2475
        acb = qemu_mallocz(pool->aiocb_size);
2476
        acb->pool = pool;
2477
    }
2478
    acb->bs = bs;
2479
    acb->cb = cb;
2480
    acb->opaque = opaque;
2481
    return acb;
2482
}
2483

    
2484
void qemu_aio_release(void *p)
2485
{
2486
    BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2487
    AIOPool *pool = acb->pool;
2488
    acb->next = pool->free_aiocb;
2489
    pool->free_aiocb = acb;
2490
}
2491

    
2492
/**************************************************************/
2493
/* removable device support */
2494

    
2495
/**
2496
 * Return TRUE if the media is present
2497
 */
2498
int bdrv_is_inserted(BlockDriverState *bs)
2499
{
2500
    BlockDriver *drv = bs->drv;
2501
    int ret;
2502
    if (!drv)
2503
        return 0;
2504
    if (!drv->bdrv_is_inserted)
2505
        return 1;
2506
    ret = drv->bdrv_is_inserted(bs);
2507
    return ret;
2508
}
2509

    
2510
/**
2511
 * Return TRUE if the media changed since the last call to this
2512
 * function. It is currently only used for floppy disks
2513
 */
2514
int bdrv_media_changed(BlockDriverState *bs)
2515
{
2516
    BlockDriver *drv = bs->drv;
2517
    int ret;
2518

    
2519
    if (!drv || !drv->bdrv_media_changed)
2520
        ret = -ENOTSUP;
2521
    else
2522
        ret = drv->bdrv_media_changed(bs);
2523
    if (ret == -ENOTSUP)
2524
        ret = bs->media_changed;
2525
    bs->media_changed = 0;
2526
    return ret;
2527
}
2528

    
2529
/**
2530
 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2531
 */
2532
int bdrv_eject(BlockDriverState *bs, int eject_flag)
2533
{
2534
    BlockDriver *drv = bs->drv;
2535
    int ret;
2536

    
2537
    if (bs->locked) {
2538
        return -EBUSY;
2539
    }
2540

    
2541
    if (!drv || !drv->bdrv_eject) {
2542
        ret = -ENOTSUP;
2543
    } else {
2544
        ret = drv->bdrv_eject(bs, eject_flag);
2545
    }
2546
    if (ret == -ENOTSUP) {
2547
        if (eject_flag)
2548
            bdrv_close(bs);
2549
        ret = 0;
2550
    }
2551

    
2552
    return ret;
2553
}
2554

    
2555
int bdrv_is_locked(BlockDriverState *bs)
2556
{
2557
    return bs->locked;
2558
}
2559

    
2560
/**
2561
 * Lock or unlock the media (if it is locked, the user won't be able
2562
 * to eject it manually).
2563
 */
2564
void bdrv_set_locked(BlockDriverState *bs, int locked)
2565
{
2566
    BlockDriver *drv = bs->drv;
2567

    
2568
    bs->locked = locked;
2569
    if (drv && drv->bdrv_set_locked) {
2570
        drv->bdrv_set_locked(bs, locked);
2571
    }
2572
}
2573

    
2574
/* needed for generic scsi interface */
2575

    
2576
int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2577
{
2578
    BlockDriver *drv = bs->drv;
2579

    
2580
    if (drv && drv->bdrv_ioctl)
2581
        return drv->bdrv_ioctl(bs, req, buf);
2582
    return -ENOTSUP;
2583
}
2584

    
2585
BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2586
        unsigned long int req, void *buf,
2587
        BlockDriverCompletionFunc *cb, void *opaque)
2588
{
2589
    BlockDriver *drv = bs->drv;
2590

    
2591
    if (drv && drv->bdrv_aio_ioctl)
2592
        return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2593
    return NULL;
2594
}
2595

    
2596

    
2597

    
2598
void *qemu_blockalign(BlockDriverState *bs, size_t size)
2599
{
2600
    return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2601
}
2602

    
2603
void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2604
{
2605
    int64_t bitmap_size;
2606

    
2607
    bs->dirty_count = 0;
2608
    if (enable) {
2609
        if (!bs->dirty_bitmap) {
2610
            bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2611
                    BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2612
            bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2613

    
2614
            bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2615
        }
2616
    } else {
2617
        if (bs->dirty_bitmap) {
2618
            qemu_free(bs->dirty_bitmap);
2619
            bs->dirty_bitmap = NULL;
2620
        }
2621
    }
2622
}
2623

    
2624
int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2625
{
2626
    int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2627

    
2628
    if (bs->dirty_bitmap &&
2629
        (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2630
        return bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2631
            (1 << (chunk % (sizeof(unsigned long) * 8)));
2632
    } else {
2633
        return 0;
2634
    }
2635
}
2636

    
2637
void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2638
                      int nr_sectors)
2639
{
2640
    set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2641
}
2642

    
2643
int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2644
{
2645
    return bs->dirty_count;
2646
}