Statistics
| Branch: | Revision:

root / block.c @ e076f338

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

    
515
    if (flags & BDRV_O_SNAPSHOT) {
516
        BlockDriverState *bs1;
517
        int64_t total_size;
518
        int is_protocol = 0;
519
        BlockDriver *bdrv_qcow2;
520
        QEMUOptionParameter *options;
521
        char tmp_filename[PATH_MAX];
522
        char backing_filename[PATH_MAX];
523

    
524
        /* if snapshot, we create a temporary backing file and open it
525
           instead of opening 'filename' directly */
526

    
527
        /* if there is a backing file, use it */
528
        bs1 = bdrv_new("");
529
        ret = bdrv_open(bs1, filename, 0, drv);
530
        if (ret < 0) {
531
            bdrv_delete(bs1);
532
            return ret;
533
        }
534
        total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
535

    
536
        if (bs1->drv && bs1->drv->protocol_name)
537
            is_protocol = 1;
538

    
539
        bdrv_delete(bs1);
540

    
541
        get_tmp_filename(tmp_filename, sizeof(tmp_filename));
542

    
543
        /* Real path is meaningless for protocols */
544
        if (is_protocol)
545
            snprintf(backing_filename, sizeof(backing_filename),
546
                     "%s", filename);
547
        else if (!realpath(filename, backing_filename))
548
            return -errno;
549

    
550
        bdrv_qcow2 = bdrv_find_format("qcow2");
551
        options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
552

    
553
        set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
554
        set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
555
        if (drv) {
556
            set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
557
                drv->format_name);
558
        }
559

    
560
        ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
561
        free_option_parameters(options);
562
        if (ret < 0) {
563
            return ret;
564
        }
565

    
566
        filename = tmp_filename;
567
        drv = bdrv_qcow2;
568
        bs->is_temporary = 1;
569
    }
570

    
571
    /* Find the right image format driver */
572
    if (!drv) {
573
        drv = find_image_format(filename);
574
    }
575

    
576
    if (!drv) {
577
        ret = -ENOENT;
578
        goto unlink_and_fail;
579
    }
580

    
581
    /* Open the image */
582
    ret = bdrv_open_common(bs, filename, flags, drv);
583
    if (ret < 0) {
584
        goto unlink_and_fail;
585
    }
586

    
587
    /* If there is a backing file, use it */
588
    if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
589
        char backing_filename[PATH_MAX];
590
        int back_flags;
591
        BlockDriver *back_drv = NULL;
592

    
593
        bs->backing_hd = bdrv_new("");
594
        path_combine(backing_filename, sizeof(backing_filename),
595
                     filename, bs->backing_file);
596
        if (bs->backing_format[0] != '\0')
597
            back_drv = bdrv_find_format(bs->backing_format);
598

    
599
        /* backing files always opened read-only */
600
        back_flags =
601
            flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
602

    
603
        ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
604
        if (ret < 0) {
605
            bdrv_close(bs);
606
            return ret;
607
        }
608
        if (bs->is_temporary) {
609
            bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
610
        } else {
611
            /* base image inherits from "parent" */
612
            bs->backing_hd->keep_read_only = bs->keep_read_only;
613
        }
614
    }
615

    
616
    if (!bdrv_key_required(bs)) {
617
        /* call the change callback */
618
        bs->media_changed = 1;
619
        if (bs->change_cb)
620
            bs->change_cb(bs->change_opaque);
621
    }
622

    
623
    return 0;
624

    
625
unlink_and_fail:
626
    if (bs->is_temporary) {
627
        unlink(filename);
628
    }
629
    return ret;
630
}
631

    
632
void bdrv_close(BlockDriverState *bs)
633
{
634
    if (bs->drv) {
635
        if (bs == bs_snapshots) {
636
            bs_snapshots = NULL;
637
        }
638
        if (bs->backing_hd) {
639
            bdrv_delete(bs->backing_hd);
640
            bs->backing_hd = NULL;
641
        }
642
        bs->drv->bdrv_close(bs);
643
        qemu_free(bs->opaque);
644
#ifdef _WIN32
645
        if (bs->is_temporary) {
646
            unlink(bs->filename);
647
        }
648
#endif
649
        bs->opaque = NULL;
650
        bs->drv = NULL;
651

    
652
        if (bs->file != NULL) {
653
            bdrv_close(bs->file);
654
        }
655

    
656
        /* call the change callback */
657
        bs->media_changed = 1;
658
        if (bs->change_cb)
659
            bs->change_cb(bs->change_opaque);
660
    }
661
}
662

    
663
void bdrv_close_all(void)
664
{
665
    BlockDriverState *bs;
666

    
667
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
668
        bdrv_close(bs);
669
    }
670
}
671

    
672
void bdrv_delete(BlockDriverState *bs)
673
{
674
    assert(!bs->peer);
675

    
676
    /* remove from list, if necessary */
677
    if (bs->device_name[0] != '\0') {
678
        QTAILQ_REMOVE(&bdrv_states, bs, list);
679
    }
680

    
681
    bdrv_close(bs);
682
    if (bs->file != NULL) {
683
        bdrv_delete(bs->file);
684
    }
685

    
686
    assert(bs != bs_snapshots);
687
    qemu_free(bs);
688
}
689

    
690
int bdrv_attach(BlockDriverState *bs, DeviceState *qdev)
691
{
692
    if (bs->peer) {
693
        return -EBUSY;
694
    }
695
    bs->peer = qdev;
696
    return 0;
697
}
698

    
699
void bdrv_detach(BlockDriverState *bs, DeviceState *qdev)
700
{
701
    assert(bs->peer == qdev);
702
    bs->peer = NULL;
703
}
704

    
705
DeviceState *bdrv_get_attached(BlockDriverState *bs)
706
{
707
    return bs->peer;
708
}
709

    
710
/*
711
 * Run consistency checks on an image
712
 *
713
 * Returns 0 if the check could be completed (it doesn't mean that the image is
714
 * free of errors) or -errno when an internal error occured. The results of the
715
 * check are stored in res.
716
 */
717
int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
718
{
719
    if (bs->drv->bdrv_check == NULL) {
720
        return -ENOTSUP;
721
    }
722

    
723
    memset(res, 0, sizeof(*res));
724
    res->corruptions = bs->drv->bdrv_check(bs);
725
    return res->corruptions < 0 ? res->corruptions : 0;
726
}
727

    
728
/* commit COW file into the raw image */
729
int bdrv_commit(BlockDriverState *bs)
730
{
731
    BlockDriver *drv = bs->drv;
732
    int64_t i, total_sectors;
733
    int n, j, ro, open_flags;
734
    int ret = 0, rw_ret = 0;
735
    unsigned char sector[BDRV_SECTOR_SIZE];
736
    char filename[1024];
737
    BlockDriverState *bs_rw, *bs_ro;
738

    
739
    if (!drv)
740
        return -ENOMEDIUM;
741
    
742
    if (!bs->backing_hd) {
743
        return -ENOTSUP;
744
    }
745

    
746
    if (bs->backing_hd->keep_read_only) {
747
        return -EACCES;
748
    }
749
    
750
    ro = bs->backing_hd->read_only;
751
    strncpy(filename, bs->backing_hd->filename, sizeof(filename));
752
    open_flags =  bs->backing_hd->open_flags;
753

    
754
    if (ro) {
755
        /* re-open as RW */
756
        bdrv_delete(bs->backing_hd);
757
        bs->backing_hd = NULL;
758
        bs_rw = bdrv_new("");
759
        rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR, drv);
760
        if (rw_ret < 0) {
761
            bdrv_delete(bs_rw);
762
            /* try to re-open read-only */
763
            bs_ro = bdrv_new("");
764
            ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
765
            if (ret < 0) {
766
                bdrv_delete(bs_ro);
767
                /* drive not functional anymore */
768
                bs->drv = NULL;
769
                return ret;
770
            }
771
            bs->backing_hd = bs_ro;
772
            return rw_ret;
773
        }
774
        bs->backing_hd = bs_rw;
775
    }
776

    
777
    total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
778
    for (i = 0; i < total_sectors;) {
779
        if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
780
            for(j = 0; j < n; j++) {
781
                if (bdrv_read(bs, i, sector, 1) != 0) {
782
                    ret = -EIO;
783
                    goto ro_cleanup;
784
                }
785

    
786
                if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
787
                    ret = -EIO;
788
                    goto ro_cleanup;
789
                }
790
                i++;
791
            }
792
        } else {
793
            i += n;
794
        }
795
    }
796

    
797
    if (drv->bdrv_make_empty) {
798
        ret = drv->bdrv_make_empty(bs);
799
        bdrv_flush(bs);
800
    }
801

    
802
    /*
803
     * Make sure all data we wrote to the backing device is actually
804
     * stable on disk.
805
     */
806
    if (bs->backing_hd)
807
        bdrv_flush(bs->backing_hd);
808

    
809
ro_cleanup:
810

    
811
    if (ro) {
812
        /* re-open as RO */
813
        bdrv_delete(bs->backing_hd);
814
        bs->backing_hd = NULL;
815
        bs_ro = bdrv_new("");
816
        ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
817
        if (ret < 0) {
818
            bdrv_delete(bs_ro);
819
            /* drive not functional anymore */
820
            bs->drv = NULL;
821
            return ret;
822
        }
823
        bs->backing_hd = bs_ro;
824
        bs->backing_hd->keep_read_only = 0;
825
    }
826

    
827
    return ret;
828
}
829

    
830
void bdrv_commit_all(void)
831
{
832
    BlockDriverState *bs;
833

    
834
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
835
        bdrv_commit(bs);
836
    }
837
}
838

    
839
/*
840
 * Return values:
841
 * 0        - success
842
 * -EINVAL  - backing format specified, but no file
843
 * -ENOSPC  - can't update the backing file because no space is left in the
844
 *            image file header
845
 * -ENOTSUP - format driver doesn't support changing the backing file
846
 */
847
int bdrv_change_backing_file(BlockDriverState *bs,
848
    const char *backing_file, const char *backing_fmt)
849
{
850
    BlockDriver *drv = bs->drv;
851

    
852
    if (drv->bdrv_change_backing_file != NULL) {
853
        return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
854
    } else {
855
        return -ENOTSUP;
856
    }
857
}
858

    
859
static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
860
                                   size_t size)
861
{
862
    int64_t len;
863

    
864
    if (!bdrv_is_inserted(bs))
865
        return -ENOMEDIUM;
866

    
867
    if (bs->growable)
868
        return 0;
869

    
870
    len = bdrv_getlength(bs);
871

    
872
    if (offset < 0)
873
        return -EIO;
874

    
875
    if ((offset > len) || (len - offset < size))
876
        return -EIO;
877

    
878
    return 0;
879
}
880

    
881
static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
882
                              int nb_sectors)
883
{
884
    return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
885
                                   nb_sectors * BDRV_SECTOR_SIZE);
886
}
887

    
888
/* return < 0 if error. See bdrv_write() for the return codes */
889
int bdrv_read(BlockDriverState *bs, int64_t sector_num,
890
              uint8_t *buf, int nb_sectors)
891
{
892
    BlockDriver *drv = bs->drv;
893

    
894
    if (!drv)
895
        return -ENOMEDIUM;
896
    if (bdrv_check_request(bs, sector_num, nb_sectors))
897
        return -EIO;
898

    
899
    return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
900
}
901

    
902
static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
903
                             int nb_sectors, int dirty)
904
{
905
    int64_t start, end;
906
    unsigned long val, idx, bit;
907

    
908
    start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
909
    end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
910

    
911
    for (; start <= end; start++) {
912
        idx = start / (sizeof(unsigned long) * 8);
913
        bit = start % (sizeof(unsigned long) * 8);
914
        val = bs->dirty_bitmap[idx];
915
        if (dirty) {
916
            if (!(val & (1 << bit))) {
917
                bs->dirty_count++;
918
                val |= 1 << bit;
919
            }
920
        } else {
921
            if (val & (1 << bit)) {
922
                bs->dirty_count--;
923
                val &= ~(1 << bit);
924
            }
925
        }
926
        bs->dirty_bitmap[idx] = val;
927
    }
928
}
929

    
930
/* Return < 0 if error. Important errors are:
931
  -EIO         generic I/O error (may happen for all errors)
932
  -ENOMEDIUM   No media inserted.
933
  -EINVAL      Invalid sector number or nb_sectors
934
  -EACCES      Trying to write a read-only device
935
*/
936
int bdrv_write(BlockDriverState *bs, int64_t sector_num,
937
               const uint8_t *buf, int nb_sectors)
938
{
939
    BlockDriver *drv = bs->drv;
940
    if (!bs->drv)
941
        return -ENOMEDIUM;
942
    if (bs->read_only)
943
        return -EACCES;
944
    if (bdrv_check_request(bs, sector_num, nb_sectors))
945
        return -EIO;
946

    
947
    if (bs->dirty_bitmap) {
948
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
949
    }
950

    
951
    if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
952
        bs->wr_highest_sector = sector_num + nb_sectors - 1;
953
    }
954

    
955
    return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
956
}
957

    
958
int bdrv_pread(BlockDriverState *bs, int64_t offset,
959
               void *buf, int count1)
960
{
961
    uint8_t tmp_buf[BDRV_SECTOR_SIZE];
962
    int len, nb_sectors, count;
963
    int64_t sector_num;
964
    int ret;
965

    
966
    count = count1;
967
    /* first read to align to sector start */
968
    len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
969
    if (len > count)
970
        len = count;
971
    sector_num = offset >> BDRV_SECTOR_BITS;
972
    if (len > 0) {
973
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
974
            return ret;
975
        memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
976
        count -= len;
977
        if (count == 0)
978
            return count1;
979
        sector_num++;
980
        buf += len;
981
    }
982

    
983
    /* read the sectors "in place" */
984
    nb_sectors = count >> BDRV_SECTOR_BITS;
985
    if (nb_sectors > 0) {
986
        if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
987
            return ret;
988
        sector_num += nb_sectors;
989
        len = nb_sectors << BDRV_SECTOR_BITS;
990
        buf += len;
991
        count -= len;
992
    }
993

    
994
    /* add data from the last sector */
995
    if (count > 0) {
996
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
997
            return ret;
998
        memcpy(buf, tmp_buf, count);
999
    }
1000
    return count1;
1001
}
1002

    
1003
int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1004
                const void *buf, int count1)
1005
{
1006
    uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1007
    int len, nb_sectors, count;
1008
    int64_t sector_num;
1009
    int ret;
1010

    
1011
    count = count1;
1012
    /* first write to align to sector start */
1013
    len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1014
    if (len > count)
1015
        len = count;
1016
    sector_num = offset >> BDRV_SECTOR_BITS;
1017
    if (len > 0) {
1018
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1019
            return ret;
1020
        memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1021
        if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1022
            return ret;
1023
        count -= len;
1024
        if (count == 0)
1025
            return count1;
1026
        sector_num++;
1027
        buf += len;
1028
    }
1029

    
1030
    /* write the sectors "in place" */
1031
    nb_sectors = count >> BDRV_SECTOR_BITS;
1032
    if (nb_sectors > 0) {
1033
        if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1034
            return ret;
1035
        sector_num += nb_sectors;
1036
        len = nb_sectors << BDRV_SECTOR_BITS;
1037
        buf += len;
1038
        count -= len;
1039
    }
1040

    
1041
    /* add data from the last sector */
1042
    if (count > 0) {
1043
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1044
            return ret;
1045
        memcpy(tmp_buf, buf, count);
1046
        if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1047
            return ret;
1048
    }
1049
    return count1;
1050
}
1051

    
1052
/*
1053
 * Writes to the file and ensures that no writes are reordered across this
1054
 * request (acts as a barrier)
1055
 *
1056
 * Returns 0 on success, -errno in error cases.
1057
 */
1058
int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1059
    const void *buf, int count)
1060
{
1061
    int ret;
1062

    
1063
    ret = bdrv_pwrite(bs, offset, buf, count);
1064
    if (ret < 0) {
1065
        return ret;
1066
    }
1067

    
1068
    /* No flush needed for cache=writethrough, it uses O_DSYNC */
1069
    if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1070
        bdrv_flush(bs);
1071
    }
1072

    
1073
    return 0;
1074
}
1075

    
1076
/*
1077
 * Writes to the file and ensures that no writes are reordered across this
1078
 * request (acts as a barrier)
1079
 *
1080
 * Returns 0 on success, -errno in error cases.
1081
 */
1082
int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1083
    const uint8_t *buf, int nb_sectors)
1084
{
1085
    return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1086
        buf, BDRV_SECTOR_SIZE * nb_sectors);
1087
}
1088

    
1089
/**
1090
 * Truncate file to 'offset' bytes (needed only for file protocols)
1091
 */
1092
int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1093
{
1094
    BlockDriver *drv = bs->drv;
1095
    int ret;
1096
    if (!drv)
1097
        return -ENOMEDIUM;
1098
    if (!drv->bdrv_truncate)
1099
        return -ENOTSUP;
1100
    if (bs->read_only)
1101
        return -EACCES;
1102
    ret = drv->bdrv_truncate(bs, offset);
1103
    if (ret == 0) {
1104
        ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1105
    }
1106
    return ret;
1107
}
1108

    
1109
/**
1110
 * Length of a file in bytes. Return < 0 if error or unknown.
1111
 */
1112
int64_t bdrv_getlength(BlockDriverState *bs)
1113
{
1114
    BlockDriver *drv = bs->drv;
1115
    if (!drv)
1116
        return -ENOMEDIUM;
1117

    
1118
    /* Fixed size devices use the total_sectors value for speed instead of
1119
       issuing a length query (like lseek) on each call.  Also, legacy block
1120
       drivers don't provide a bdrv_getlength function and must use
1121
       total_sectors. */
1122
    if (!bs->growable || !drv->bdrv_getlength) {
1123
        return bs->total_sectors * BDRV_SECTOR_SIZE;
1124
    }
1125
    return drv->bdrv_getlength(bs);
1126
}
1127

    
1128
/* return 0 as number of sectors if no device present or error */
1129
void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1130
{
1131
    int64_t length;
1132
    length = bdrv_getlength(bs);
1133
    if (length < 0)
1134
        length = 0;
1135
    else
1136
        length = length >> BDRV_SECTOR_BITS;
1137
    *nb_sectors_ptr = length;
1138
}
1139

    
1140
struct partition {
1141
        uint8_t boot_ind;           /* 0x80 - active */
1142
        uint8_t head;               /* starting head */
1143
        uint8_t sector;             /* starting sector */
1144
        uint8_t cyl;                /* starting cylinder */
1145
        uint8_t sys_ind;            /* What partition type */
1146
        uint8_t end_head;           /* end head */
1147
        uint8_t end_sector;         /* end sector */
1148
        uint8_t end_cyl;            /* end cylinder */
1149
        uint32_t start_sect;        /* starting sector counting from 0 */
1150
        uint32_t nr_sects;          /* nr of sectors in partition */
1151
} __attribute__((packed));
1152

    
1153
/* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1154
static int guess_disk_lchs(BlockDriverState *bs,
1155
                           int *pcylinders, int *pheads, int *psectors)
1156
{
1157
    uint8_t buf[BDRV_SECTOR_SIZE];
1158
    int ret, i, heads, sectors, cylinders;
1159
    struct partition *p;
1160
    uint32_t nr_sects;
1161
    uint64_t nb_sectors;
1162

    
1163
    bdrv_get_geometry(bs, &nb_sectors);
1164

    
1165
    ret = bdrv_read(bs, 0, buf, 1);
1166
    if (ret < 0)
1167
        return -1;
1168
    /* test msdos magic */
1169
    if (buf[510] != 0x55 || buf[511] != 0xaa)
1170
        return -1;
1171
    for(i = 0; i < 4; i++) {
1172
        p = ((struct partition *)(buf + 0x1be)) + i;
1173
        nr_sects = le32_to_cpu(p->nr_sects);
1174
        if (nr_sects && p->end_head) {
1175
            /* We make the assumption that the partition terminates on
1176
               a cylinder boundary */
1177
            heads = p->end_head + 1;
1178
            sectors = p->end_sector & 63;
1179
            if (sectors == 0)
1180
                continue;
1181
            cylinders = nb_sectors / (heads * sectors);
1182
            if (cylinders < 1 || cylinders > 16383)
1183
                continue;
1184
            *pheads = heads;
1185
            *psectors = sectors;
1186
            *pcylinders = cylinders;
1187
#if 0
1188
            printf("guessed geometry: LCHS=%d %d %d\n",
1189
                   cylinders, heads, sectors);
1190
#endif
1191
            return 0;
1192
        }
1193
    }
1194
    return -1;
1195
}
1196

    
1197
void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1198
{
1199
    int translation, lba_detected = 0;
1200
    int cylinders, heads, secs;
1201
    uint64_t nb_sectors;
1202

    
1203
    /* if a geometry hint is available, use it */
1204
    bdrv_get_geometry(bs, &nb_sectors);
1205
    bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1206
    translation = bdrv_get_translation_hint(bs);
1207
    if (cylinders != 0) {
1208
        *pcyls = cylinders;
1209
        *pheads = heads;
1210
        *psecs = secs;
1211
    } else {
1212
        if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1213
            if (heads > 16) {
1214
                /* if heads > 16, it means that a BIOS LBA
1215
                   translation was active, so the default
1216
                   hardware geometry is OK */
1217
                lba_detected = 1;
1218
                goto default_geometry;
1219
            } else {
1220
                *pcyls = cylinders;
1221
                *pheads = heads;
1222
                *psecs = secs;
1223
                /* disable any translation to be in sync with
1224
                   the logical geometry */
1225
                if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1226
                    bdrv_set_translation_hint(bs,
1227
                                              BIOS_ATA_TRANSLATION_NONE);
1228
                }
1229
            }
1230
        } else {
1231
        default_geometry:
1232
            /* if no geometry, use a standard physical disk geometry */
1233
            cylinders = nb_sectors / (16 * 63);
1234

    
1235
            if (cylinders > 16383)
1236
                cylinders = 16383;
1237
            else if (cylinders < 2)
1238
                cylinders = 2;
1239
            *pcyls = cylinders;
1240
            *pheads = 16;
1241
            *psecs = 63;
1242
            if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1243
                if ((*pcyls * *pheads) <= 131072) {
1244
                    bdrv_set_translation_hint(bs,
1245
                                              BIOS_ATA_TRANSLATION_LARGE);
1246
                } else {
1247
                    bdrv_set_translation_hint(bs,
1248
                                              BIOS_ATA_TRANSLATION_LBA);
1249
                }
1250
            }
1251
        }
1252
        bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1253
    }
1254
}
1255

    
1256
void bdrv_set_geometry_hint(BlockDriverState *bs,
1257
                            int cyls, int heads, int secs)
1258
{
1259
    bs->cyls = cyls;
1260
    bs->heads = heads;
1261
    bs->secs = secs;
1262
}
1263

    
1264
void bdrv_set_type_hint(BlockDriverState *bs, int type)
1265
{
1266
    bs->type = type;
1267
    bs->removable = ((type == BDRV_TYPE_CDROM ||
1268
                      type == BDRV_TYPE_FLOPPY));
1269
}
1270

    
1271
void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1272
{
1273
    bs->translation = translation;
1274
}
1275

    
1276
void bdrv_get_geometry_hint(BlockDriverState *bs,
1277
                            int *pcyls, int *pheads, int *psecs)
1278
{
1279
    *pcyls = bs->cyls;
1280
    *pheads = bs->heads;
1281
    *psecs = bs->secs;
1282
}
1283

    
1284
int bdrv_get_type_hint(BlockDriverState *bs)
1285
{
1286
    return bs->type;
1287
}
1288

    
1289
int bdrv_get_translation_hint(BlockDriverState *bs)
1290
{
1291
    return bs->translation;
1292
}
1293

    
1294
void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1295
                       BlockErrorAction on_write_error)
1296
{
1297
    bs->on_read_error = on_read_error;
1298
    bs->on_write_error = on_write_error;
1299
}
1300

    
1301
BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1302
{
1303
    return is_read ? bs->on_read_error : bs->on_write_error;
1304
}
1305

    
1306
void bdrv_set_removable(BlockDriverState *bs, int removable)
1307
{
1308
    bs->removable = removable;
1309
    if (removable && bs == bs_snapshots) {
1310
        bs_snapshots = NULL;
1311
    }
1312
}
1313

    
1314
int bdrv_is_removable(BlockDriverState *bs)
1315
{
1316
    return bs->removable;
1317
}
1318

    
1319
int bdrv_is_read_only(BlockDriverState *bs)
1320
{
1321
    return bs->read_only;
1322
}
1323

    
1324
int bdrv_is_sg(BlockDriverState *bs)
1325
{
1326
    return bs->sg;
1327
}
1328

    
1329
int bdrv_enable_write_cache(BlockDriverState *bs)
1330
{
1331
    return bs->enable_write_cache;
1332
}
1333

    
1334
/* XXX: no longer used */
1335
void bdrv_set_change_cb(BlockDriverState *bs,
1336
                        void (*change_cb)(void *opaque), void *opaque)
1337
{
1338
    bs->change_cb = change_cb;
1339
    bs->change_opaque = opaque;
1340
}
1341

    
1342
int bdrv_is_encrypted(BlockDriverState *bs)
1343
{
1344
    if (bs->backing_hd && bs->backing_hd->encrypted)
1345
        return 1;
1346
    return bs->encrypted;
1347
}
1348

    
1349
int bdrv_key_required(BlockDriverState *bs)
1350
{
1351
    BlockDriverState *backing_hd = bs->backing_hd;
1352

    
1353
    if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1354
        return 1;
1355
    return (bs->encrypted && !bs->valid_key);
1356
}
1357

    
1358
int bdrv_set_key(BlockDriverState *bs, const char *key)
1359
{
1360
    int ret;
1361
    if (bs->backing_hd && bs->backing_hd->encrypted) {
1362
        ret = bdrv_set_key(bs->backing_hd, key);
1363
        if (ret < 0)
1364
            return ret;
1365
        if (!bs->encrypted)
1366
            return 0;
1367
    }
1368
    if (!bs->encrypted) {
1369
        return -EINVAL;
1370
    } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1371
        return -ENOMEDIUM;
1372
    }
1373
    ret = bs->drv->bdrv_set_key(bs, key);
1374
    if (ret < 0) {
1375
        bs->valid_key = 0;
1376
    } else if (!bs->valid_key) {
1377
        bs->valid_key = 1;
1378
        /* call the change callback now, we skipped it on open */
1379
        bs->media_changed = 1;
1380
        if (bs->change_cb)
1381
            bs->change_cb(bs->change_opaque);
1382
    }
1383
    return ret;
1384
}
1385

    
1386
void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1387
{
1388
    if (!bs->drv) {
1389
        buf[0] = '\0';
1390
    } else {
1391
        pstrcpy(buf, buf_size, bs->drv->format_name);
1392
    }
1393
}
1394

    
1395
void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1396
                         void *opaque)
1397
{
1398
    BlockDriver *drv;
1399

    
1400
    QLIST_FOREACH(drv, &bdrv_drivers, list) {
1401
        it(opaque, drv->format_name);
1402
    }
1403
}
1404

    
1405
BlockDriverState *bdrv_find(const char *name)
1406
{
1407
    BlockDriverState *bs;
1408

    
1409
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1410
        if (!strcmp(name, bs->device_name)) {
1411
            return bs;
1412
        }
1413
    }
1414
    return NULL;
1415
}
1416

    
1417
BlockDriverState *bdrv_next(BlockDriverState *bs)
1418
{
1419
    if (!bs) {
1420
        return QTAILQ_FIRST(&bdrv_states);
1421
    }
1422
    return QTAILQ_NEXT(bs, list);
1423
}
1424

    
1425
void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1426
{
1427
    BlockDriverState *bs;
1428

    
1429
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1430
        it(opaque, bs);
1431
    }
1432
}
1433

    
1434
const char *bdrv_get_device_name(BlockDriverState *bs)
1435
{
1436
    return bs->device_name;
1437
}
1438

    
1439
void bdrv_flush(BlockDriverState *bs)
1440
{
1441
    if (bs->open_flags & BDRV_O_NO_FLUSH) {
1442
        return;
1443
    }
1444

    
1445
    if (bs->drv && bs->drv->bdrv_flush)
1446
        bs->drv->bdrv_flush(bs);
1447
}
1448

    
1449
void bdrv_flush_all(void)
1450
{
1451
    BlockDriverState *bs;
1452

    
1453
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1454
        if (bs->drv && !bdrv_is_read_only(bs) &&
1455
            (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1456
            bdrv_flush(bs);
1457
        }
1458
    }
1459
}
1460

    
1461
int bdrv_has_zero_init(BlockDriverState *bs)
1462
{
1463
    assert(bs->drv);
1464

    
1465
    if (bs->drv->no_zero_init) {
1466
        return 0;
1467
    } else if (bs->file) {
1468
        return bdrv_has_zero_init(bs->file);
1469
    }
1470

    
1471
    return 1;
1472
}
1473

    
1474
/*
1475
 * Returns true iff the specified sector is present in the disk image. Drivers
1476
 * not implementing the functionality are assumed to not support backing files,
1477
 * hence all their sectors are reported as allocated.
1478
 *
1479
 * 'pnum' is set to the number of sectors (including and immediately following
1480
 * the specified sector) that are known to be in the same
1481
 * allocated/unallocated state.
1482
 *
1483
 * 'nb_sectors' is the max value 'pnum' should be set to.
1484
 */
1485
int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1486
        int *pnum)
1487
{
1488
    int64_t n;
1489
    if (!bs->drv->bdrv_is_allocated) {
1490
        if (sector_num >= bs->total_sectors) {
1491
            *pnum = 0;
1492
            return 0;
1493
        }
1494
        n = bs->total_sectors - sector_num;
1495
        *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1496
        return 1;
1497
    }
1498
    return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1499
}
1500

    
1501
void bdrv_mon_event(const BlockDriverState *bdrv,
1502
                    BlockMonEventAction action, int is_read)
1503
{
1504
    QObject *data;
1505
    const char *action_str;
1506

    
1507
    switch (action) {
1508
    case BDRV_ACTION_REPORT:
1509
        action_str = "report";
1510
        break;
1511
    case BDRV_ACTION_IGNORE:
1512
        action_str = "ignore";
1513
        break;
1514
    case BDRV_ACTION_STOP:
1515
        action_str = "stop";
1516
        break;
1517
    default:
1518
        abort();
1519
    }
1520

    
1521
    data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1522
                              bdrv->device_name,
1523
                              action_str,
1524
                              is_read ? "read" : "write");
1525
    monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1526

    
1527
    qobject_decref(data);
1528
}
1529

    
1530
static void bdrv_print_dict(QObject *obj, void *opaque)
1531
{
1532
    QDict *bs_dict;
1533
    Monitor *mon = opaque;
1534

    
1535
    bs_dict = qobject_to_qdict(obj);
1536

    
1537
    monitor_printf(mon, "%s: type=%s removable=%d",
1538
                        qdict_get_str(bs_dict, "device"),
1539
                        qdict_get_str(bs_dict, "type"),
1540
                        qdict_get_bool(bs_dict, "removable"));
1541

    
1542
    if (qdict_get_bool(bs_dict, "removable")) {
1543
        monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1544
    }
1545

    
1546
    if (qdict_haskey(bs_dict, "inserted")) {
1547
        QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1548

    
1549
        monitor_printf(mon, " file=");
1550
        monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1551
        if (qdict_haskey(qdict, "backing_file")) {
1552
            monitor_printf(mon, " backing_file=");
1553
            monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1554
        }
1555
        monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1556
                            qdict_get_bool(qdict, "ro"),
1557
                            qdict_get_str(qdict, "drv"),
1558
                            qdict_get_bool(qdict, "encrypted"));
1559
    } else {
1560
        monitor_printf(mon, " [not inserted]");
1561
    }
1562

    
1563
    monitor_printf(mon, "\n");
1564
}
1565

    
1566
void bdrv_info_print(Monitor *mon, const QObject *data)
1567
{
1568
    qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1569
}
1570

    
1571
void bdrv_info(Monitor *mon, QObject **ret_data)
1572
{
1573
    QList *bs_list;
1574
    BlockDriverState *bs;
1575

    
1576
    bs_list = qlist_new();
1577

    
1578
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1579
        QObject *bs_obj;
1580
        const char *type = "unknown";
1581

    
1582
        switch(bs->type) {
1583
        case BDRV_TYPE_HD:
1584
            type = "hd";
1585
            break;
1586
        case BDRV_TYPE_CDROM:
1587
            type = "cdrom";
1588
            break;
1589
        case BDRV_TYPE_FLOPPY:
1590
            type = "floppy";
1591
            break;
1592
        }
1593

    
1594
        bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1595
                                    "'removable': %i, 'locked': %i }",
1596
                                    bs->device_name, type, bs->removable,
1597
                                    bs->locked);
1598

    
1599
        if (bs->drv) {
1600
            QObject *obj;
1601
            QDict *bs_dict = qobject_to_qdict(bs_obj);
1602

    
1603
            obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1604
                                     "'encrypted': %i }",
1605
                                     bs->filename, bs->read_only,
1606
                                     bs->drv->format_name,
1607
                                     bdrv_is_encrypted(bs));
1608
            if (bs->backing_file[0] != '\0') {
1609
                QDict *qdict = qobject_to_qdict(obj);
1610
                qdict_put(qdict, "backing_file",
1611
                          qstring_from_str(bs->backing_file));
1612
            }
1613

    
1614
            qdict_put_obj(bs_dict, "inserted", obj);
1615
        }
1616
        qlist_append_obj(bs_list, bs_obj);
1617
    }
1618

    
1619
    *ret_data = QOBJECT(bs_list);
1620
}
1621

    
1622
static void bdrv_stats_iter(QObject *data, void *opaque)
1623
{
1624
    QDict *qdict;
1625
    Monitor *mon = opaque;
1626

    
1627
    qdict = qobject_to_qdict(data);
1628
    monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1629

    
1630
    qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1631
    monitor_printf(mon, " rd_bytes=%" PRId64
1632
                        " wr_bytes=%" PRId64
1633
                        " rd_operations=%" PRId64
1634
                        " wr_operations=%" PRId64
1635
                        "\n",
1636
                        qdict_get_int(qdict, "rd_bytes"),
1637
                        qdict_get_int(qdict, "wr_bytes"),
1638
                        qdict_get_int(qdict, "rd_operations"),
1639
                        qdict_get_int(qdict, "wr_operations"));
1640
}
1641

    
1642
void bdrv_stats_print(Monitor *mon, const QObject *data)
1643
{
1644
    qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1645
}
1646

    
1647
static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1648
{
1649
    QObject *res;
1650
    QDict *dict;
1651

    
1652
    res = qobject_from_jsonf("{ 'stats': {"
1653
                             "'rd_bytes': %" PRId64 ","
1654
                             "'wr_bytes': %" PRId64 ","
1655
                             "'rd_operations': %" PRId64 ","
1656
                             "'wr_operations': %" PRId64 ","
1657
                             "'wr_highest_offset': %" PRId64
1658
                             "} }",
1659
                             bs->rd_bytes, bs->wr_bytes,
1660
                             bs->rd_ops, bs->wr_ops,
1661
                             bs->wr_highest_sector *
1662
                             (uint64_t)BDRV_SECTOR_SIZE);
1663
    dict  = qobject_to_qdict(res);
1664

    
1665
    if (*bs->device_name) {
1666
        qdict_put(dict, "device", qstring_from_str(bs->device_name));
1667
    }
1668

    
1669
    if (bs->file) {
1670
        QObject *parent = bdrv_info_stats_bs(bs->file);
1671
        qdict_put_obj(dict, "parent", parent);
1672
    }
1673

    
1674
    return res;
1675
}
1676

    
1677
void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1678
{
1679
    QObject *obj;
1680
    QList *devices;
1681
    BlockDriverState *bs;
1682

    
1683
    devices = qlist_new();
1684

    
1685
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1686
        obj = bdrv_info_stats_bs(bs);
1687
        qlist_append_obj(devices, obj);
1688
    }
1689

    
1690
    *ret_data = QOBJECT(devices);
1691
}
1692

    
1693
const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1694
{
1695
    if (bs->backing_hd && bs->backing_hd->encrypted)
1696
        return bs->backing_file;
1697
    else if (bs->encrypted)
1698
        return bs->filename;
1699
    else
1700
        return NULL;
1701
}
1702

    
1703
void bdrv_get_backing_filename(BlockDriverState *bs,
1704
                               char *filename, int filename_size)
1705
{
1706
    if (!bs->backing_file) {
1707
        pstrcpy(filename, filename_size, "");
1708
    } else {
1709
        pstrcpy(filename, filename_size, bs->backing_file);
1710
    }
1711
}
1712

    
1713
int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1714
                          const uint8_t *buf, int nb_sectors)
1715
{
1716
    BlockDriver *drv = bs->drv;
1717
    if (!drv)
1718
        return -ENOMEDIUM;
1719
    if (!drv->bdrv_write_compressed)
1720
        return -ENOTSUP;
1721
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1722
        return -EIO;
1723

    
1724
    if (bs->dirty_bitmap) {
1725
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1726
    }
1727

    
1728
    return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1729
}
1730

    
1731
int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1732
{
1733
    BlockDriver *drv = bs->drv;
1734
    if (!drv)
1735
        return -ENOMEDIUM;
1736
    if (!drv->bdrv_get_info)
1737
        return -ENOTSUP;
1738
    memset(bdi, 0, sizeof(*bdi));
1739
    return drv->bdrv_get_info(bs, bdi);
1740
}
1741

    
1742
int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1743
                      int64_t pos, int size)
1744
{
1745
    BlockDriver *drv = bs->drv;
1746
    if (!drv)
1747
        return -ENOMEDIUM;
1748
    if (drv->bdrv_save_vmstate)
1749
        return drv->bdrv_save_vmstate(bs, buf, pos, size);
1750
    if (bs->file)
1751
        return bdrv_save_vmstate(bs->file, buf, pos, size);
1752
    return -ENOTSUP;
1753
}
1754

    
1755
int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1756
                      int64_t pos, int size)
1757
{
1758
    BlockDriver *drv = bs->drv;
1759
    if (!drv)
1760
        return -ENOMEDIUM;
1761
    if (drv->bdrv_load_vmstate)
1762
        return drv->bdrv_load_vmstate(bs, buf, pos, size);
1763
    if (bs->file)
1764
        return bdrv_load_vmstate(bs->file, buf, pos, size);
1765
    return -ENOTSUP;
1766
}
1767

    
1768
void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1769
{
1770
    BlockDriver *drv = bs->drv;
1771

    
1772
    if (!drv || !drv->bdrv_debug_event) {
1773
        return;
1774
    }
1775

    
1776
    return drv->bdrv_debug_event(bs, event);
1777

    
1778
}
1779

    
1780
/**************************************************************/
1781
/* handling of snapshots */
1782

    
1783
int bdrv_can_snapshot(BlockDriverState *bs)
1784
{
1785
    BlockDriver *drv = bs->drv;
1786
    if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1787
        return 0;
1788
    }
1789

    
1790
    if (!drv->bdrv_snapshot_create) {
1791
        if (bs->file != NULL) {
1792
            return bdrv_can_snapshot(bs->file);
1793
        }
1794
        return 0;
1795
    }
1796

    
1797
    return 1;
1798
}
1799

    
1800
BlockDriverState *bdrv_snapshots(void)
1801
{
1802
    BlockDriverState *bs;
1803

    
1804
    if (bs_snapshots) {
1805
        return bs_snapshots;
1806
    }
1807

    
1808
    bs = NULL;
1809
    while ((bs = bdrv_next(bs))) {
1810
        if (bdrv_can_snapshot(bs)) {
1811
            bs_snapshots = bs;
1812
            return bs;
1813
        }
1814
    }
1815
    return NULL;
1816
}
1817

    
1818
int bdrv_snapshot_create(BlockDriverState *bs,
1819
                         QEMUSnapshotInfo *sn_info)
1820
{
1821
    BlockDriver *drv = bs->drv;
1822
    if (!drv)
1823
        return -ENOMEDIUM;
1824
    if (drv->bdrv_snapshot_create)
1825
        return drv->bdrv_snapshot_create(bs, sn_info);
1826
    if (bs->file)
1827
        return bdrv_snapshot_create(bs->file, sn_info);
1828
    return -ENOTSUP;
1829
}
1830

    
1831
int bdrv_snapshot_goto(BlockDriverState *bs,
1832
                       const char *snapshot_id)
1833
{
1834
    BlockDriver *drv = bs->drv;
1835
    int ret, open_ret;
1836

    
1837
    if (!drv)
1838
        return -ENOMEDIUM;
1839
    if (drv->bdrv_snapshot_goto)
1840
        return drv->bdrv_snapshot_goto(bs, snapshot_id);
1841

    
1842
    if (bs->file) {
1843
        drv->bdrv_close(bs);
1844
        ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1845
        open_ret = drv->bdrv_open(bs, bs->open_flags);
1846
        if (open_ret < 0) {
1847
            bdrv_delete(bs->file);
1848
            bs->drv = NULL;
1849
            return open_ret;
1850
        }
1851
        return ret;
1852
    }
1853

    
1854
    return -ENOTSUP;
1855
}
1856

    
1857
int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1858
{
1859
    BlockDriver *drv = bs->drv;
1860
    if (!drv)
1861
        return -ENOMEDIUM;
1862
    if (drv->bdrv_snapshot_delete)
1863
        return drv->bdrv_snapshot_delete(bs, snapshot_id);
1864
    if (bs->file)
1865
        return bdrv_snapshot_delete(bs->file, snapshot_id);
1866
    return -ENOTSUP;
1867
}
1868

    
1869
int bdrv_snapshot_list(BlockDriverState *bs,
1870
                       QEMUSnapshotInfo **psn_info)
1871
{
1872
    BlockDriver *drv = bs->drv;
1873
    if (!drv)
1874
        return -ENOMEDIUM;
1875
    if (drv->bdrv_snapshot_list)
1876
        return drv->bdrv_snapshot_list(bs, psn_info);
1877
    if (bs->file)
1878
        return bdrv_snapshot_list(bs->file, psn_info);
1879
    return -ENOTSUP;
1880
}
1881

    
1882
#define NB_SUFFIXES 4
1883

    
1884
char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1885
{
1886
    static const char suffixes[NB_SUFFIXES] = "KMGT";
1887
    int64_t base;
1888
    int i;
1889

    
1890
    if (size <= 999) {
1891
        snprintf(buf, buf_size, "%" PRId64, size);
1892
    } else {
1893
        base = 1024;
1894
        for(i = 0; i < NB_SUFFIXES; i++) {
1895
            if (size < (10 * base)) {
1896
                snprintf(buf, buf_size, "%0.1f%c",
1897
                         (double)size / base,
1898
                         suffixes[i]);
1899
                break;
1900
            } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1901
                snprintf(buf, buf_size, "%" PRId64 "%c",
1902
                         ((size + (base >> 1)) / base),
1903
                         suffixes[i]);
1904
                break;
1905
            }
1906
            base = base * 1024;
1907
        }
1908
    }
1909
    return buf;
1910
}
1911

    
1912
char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1913
{
1914
    char buf1[128], date_buf[128], clock_buf[128];
1915
#ifdef _WIN32
1916
    struct tm *ptm;
1917
#else
1918
    struct tm tm;
1919
#endif
1920
    time_t ti;
1921
    int64_t secs;
1922

    
1923
    if (!sn) {
1924
        snprintf(buf, buf_size,
1925
                 "%-10s%-20s%7s%20s%15s",
1926
                 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1927
    } else {
1928
        ti = sn->date_sec;
1929
#ifdef _WIN32
1930
        ptm = localtime(&ti);
1931
        strftime(date_buf, sizeof(date_buf),
1932
                 "%Y-%m-%d %H:%M:%S", ptm);
1933
#else
1934
        localtime_r(&ti, &tm);
1935
        strftime(date_buf, sizeof(date_buf),
1936
                 "%Y-%m-%d %H:%M:%S", &tm);
1937
#endif
1938
        secs = sn->vm_clock_nsec / 1000000000;
1939
        snprintf(clock_buf, sizeof(clock_buf),
1940
                 "%02d:%02d:%02d.%03d",
1941
                 (int)(secs / 3600),
1942
                 (int)((secs / 60) % 60),
1943
                 (int)(secs % 60),
1944
                 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1945
        snprintf(buf, buf_size,
1946
                 "%-10s%-20s%7s%20s%15s",
1947
                 sn->id_str, sn->name,
1948
                 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1949
                 date_buf,
1950
                 clock_buf);
1951
    }
1952
    return buf;
1953
}
1954

    
1955

    
1956
/**************************************************************/
1957
/* async I/Os */
1958

    
1959
BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1960
                                 QEMUIOVector *qiov, int nb_sectors,
1961
                                 BlockDriverCompletionFunc *cb, void *opaque)
1962
{
1963
    BlockDriver *drv = bs->drv;
1964
    BlockDriverAIOCB *ret;
1965

    
1966
    if (!drv)
1967
        return NULL;
1968
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1969
        return NULL;
1970

    
1971
    ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1972
                              cb, opaque);
1973

    
1974
    if (ret) {
1975
        /* Update stats even though technically transfer has not happened. */
1976
        bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1977
        bs->rd_ops ++;
1978
    }
1979

    
1980
    return ret;
1981
}
1982

    
1983
BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1984
                                  QEMUIOVector *qiov, int nb_sectors,
1985
                                  BlockDriverCompletionFunc *cb, void *opaque)
1986
{
1987
    BlockDriver *drv = bs->drv;
1988
    BlockDriverAIOCB *ret;
1989

    
1990
    if (!drv)
1991
        return NULL;
1992
    if (bs->read_only)
1993
        return NULL;
1994
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1995
        return NULL;
1996

    
1997
    if (bs->dirty_bitmap) {
1998
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1999
    }
2000

    
2001
    ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2002
                               cb, opaque);
2003

    
2004
    if (ret) {
2005
        /* Update stats even though technically transfer has not happened. */
2006
        bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2007
        bs->wr_ops ++;
2008
        if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2009
            bs->wr_highest_sector = sector_num + nb_sectors - 1;
2010
        }
2011
    }
2012

    
2013
    return ret;
2014
}
2015

    
2016

    
2017
typedef struct MultiwriteCB {
2018
    int error;
2019
    int num_requests;
2020
    int num_callbacks;
2021
    struct {
2022
        BlockDriverCompletionFunc *cb;
2023
        void *opaque;
2024
        QEMUIOVector *free_qiov;
2025
        void *free_buf;
2026
    } callbacks[];
2027
} MultiwriteCB;
2028

    
2029
static void multiwrite_user_cb(MultiwriteCB *mcb)
2030
{
2031
    int i;
2032

    
2033
    for (i = 0; i < mcb->num_callbacks; i++) {
2034
        mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2035
        if (mcb->callbacks[i].free_qiov) {
2036
            qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2037
        }
2038
        qemu_free(mcb->callbacks[i].free_qiov);
2039
        qemu_vfree(mcb->callbacks[i].free_buf);
2040
    }
2041
}
2042

    
2043
static void multiwrite_cb(void *opaque, int ret)
2044
{
2045
    MultiwriteCB *mcb = opaque;
2046

    
2047
    if (ret < 0 && !mcb->error) {
2048
        mcb->error = ret;
2049
    }
2050

    
2051
    mcb->num_requests--;
2052
    if (mcb->num_requests == 0) {
2053
        multiwrite_user_cb(mcb);
2054
        qemu_free(mcb);
2055
    }
2056
}
2057

    
2058
static int multiwrite_req_compare(const void *a, const void *b)
2059
{
2060
    const BlockRequest *req1 = a, *req2 = b;
2061

    
2062
    /*
2063
     * Note that we can't simply subtract req2->sector from req1->sector
2064
     * here as that could overflow the return value.
2065
     */
2066
    if (req1->sector > req2->sector) {
2067
        return 1;
2068
    } else if (req1->sector < req2->sector) {
2069
        return -1;
2070
    } else {
2071
        return 0;
2072
    }
2073
}
2074

    
2075
/*
2076
 * Takes a bunch of requests and tries to merge them. Returns the number of
2077
 * requests that remain after merging.
2078
 */
2079
static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2080
    int num_reqs, MultiwriteCB *mcb)
2081
{
2082
    int i, outidx;
2083

    
2084
    // Sort requests by start sector
2085
    qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2086

    
2087
    // Check if adjacent requests touch the same clusters. If so, combine them,
2088
    // filling up gaps with zero sectors.
2089
    outidx = 0;
2090
    for (i = 1; i < num_reqs; i++) {
2091
        int merge = 0;
2092
        int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2093

    
2094
        // This handles the cases that are valid for all block drivers, namely
2095
        // exactly sequential writes and overlapping writes.
2096
        if (reqs[i].sector <= oldreq_last) {
2097
            merge = 1;
2098
        }
2099

    
2100
        // The block driver may decide that it makes sense to combine requests
2101
        // even if there is a gap of some sectors between them. In this case,
2102
        // the gap is filled with zeros (therefore only applicable for yet
2103
        // unused space in format like qcow2).
2104
        if (!merge && bs->drv->bdrv_merge_requests) {
2105
            merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2106
        }
2107

    
2108
        if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2109
            merge = 0;
2110
        }
2111

    
2112
        if (merge) {
2113
            size_t size;
2114
            QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2115
            qemu_iovec_init(qiov,
2116
                reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2117

    
2118
            // Add the first request to the merged one. If the requests are
2119
            // overlapping, drop the last sectors of the first request.
2120
            size = (reqs[i].sector - reqs[outidx].sector) << 9;
2121
            qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2122

    
2123
            // We might need to add some zeros between the two requests
2124
            if (reqs[i].sector > oldreq_last) {
2125
                size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2126
                uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2127
                memset(buf, 0, zero_bytes);
2128
                qemu_iovec_add(qiov, buf, zero_bytes);
2129
                mcb->callbacks[i].free_buf = buf;
2130
            }
2131

    
2132
            // Add the second request
2133
            qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2134

    
2135
            reqs[outidx].nb_sectors = qiov->size >> 9;
2136
            reqs[outidx].qiov = qiov;
2137

    
2138
            mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2139
        } else {
2140
            outidx++;
2141
            reqs[outidx].sector     = reqs[i].sector;
2142
            reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2143
            reqs[outidx].qiov       = reqs[i].qiov;
2144
        }
2145
    }
2146

    
2147
    return outidx + 1;
2148
}
2149

    
2150
/*
2151
 * Submit multiple AIO write requests at once.
2152
 *
2153
 * On success, the function returns 0 and all requests in the reqs array have
2154
 * been submitted. In error case this function returns -1, and any of the
2155
 * requests may or may not be submitted yet. In particular, this means that the
2156
 * callback will be called for some of the requests, for others it won't. The
2157
 * caller must check the error field of the BlockRequest to wait for the right
2158
 * callbacks (if error != 0, no callback will be called).
2159
 *
2160
 * The implementation may modify the contents of the reqs array, e.g. to merge
2161
 * requests. However, the fields opaque and error are left unmodified as they
2162
 * are used to signal failure for a single request to the caller.
2163
 */
2164
int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2165
{
2166
    BlockDriverAIOCB *acb;
2167
    MultiwriteCB *mcb;
2168
    int i;
2169

    
2170
    if (num_reqs == 0) {
2171
        return 0;
2172
    }
2173

    
2174
    // Create MultiwriteCB structure
2175
    mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2176
    mcb->num_requests = 0;
2177
    mcb->num_callbacks = num_reqs;
2178

    
2179
    for (i = 0; i < num_reqs; i++) {
2180
        mcb->callbacks[i].cb = reqs[i].cb;
2181
        mcb->callbacks[i].opaque = reqs[i].opaque;
2182
    }
2183

    
2184
    // Check for mergable requests
2185
    num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2186

    
2187
    /*
2188
     * Run the aio requests. As soon as one request can't be submitted
2189
     * successfully, fail all requests that are not yet submitted (we must
2190
     * return failure for all requests anyway)
2191
     *
2192
     * num_requests cannot be set to the right value immediately: If
2193
     * bdrv_aio_writev fails for some request, num_requests would be too high
2194
     * and therefore multiwrite_cb() would never recognize the multiwrite
2195
     * request as completed. We also cannot use the loop variable i to set it
2196
     * when the first request fails because the callback may already have been
2197
     * called for previously submitted requests. Thus, num_requests must be
2198
     * incremented for each request that is submitted.
2199
     *
2200
     * The problem that callbacks may be called early also means that we need
2201
     * to take care that num_requests doesn't become 0 before all requests are
2202
     * submitted - multiwrite_cb() would consider the multiwrite request
2203
     * completed. A dummy request that is "completed" by a manual call to
2204
     * multiwrite_cb() takes care of this.
2205
     */
2206
    mcb->num_requests = 1;
2207

    
2208
    for (i = 0; i < num_reqs; i++) {
2209
        mcb->num_requests++;
2210
        acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2211
            reqs[i].nb_sectors, multiwrite_cb, mcb);
2212

    
2213
        if (acb == NULL) {
2214
            // We can only fail the whole thing if no request has been
2215
            // submitted yet. Otherwise we'll wait for the submitted AIOs to
2216
            // complete and report the error in the callback.
2217
            if (i == 0) {
2218
                goto fail;
2219
            } else {
2220
                multiwrite_cb(mcb, -EIO);
2221
                break;
2222
            }
2223
        }
2224
    }
2225

    
2226
    /* Complete the dummy request */
2227
    multiwrite_cb(mcb, 0);
2228

    
2229
    return 0;
2230

    
2231
fail:
2232
    for (i = 0; i < mcb->num_callbacks; i++) {
2233
        reqs[i].error = -EIO;
2234
    }
2235
    qemu_free(mcb);
2236
    return -1;
2237
}
2238

    
2239
BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2240
        BlockDriverCompletionFunc *cb, void *opaque)
2241
{
2242
    BlockDriver *drv = bs->drv;
2243

    
2244
    if (bs->open_flags & BDRV_O_NO_FLUSH) {
2245
        return bdrv_aio_noop_em(bs, cb, opaque);
2246
    }
2247

    
2248
    if (!drv)
2249
        return NULL;
2250
    return drv->bdrv_aio_flush(bs, cb, opaque);
2251
}
2252

    
2253
void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2254
{
2255
    acb->pool->cancel(acb);
2256
}
2257

    
2258

    
2259
/**************************************************************/
2260
/* async block device emulation */
2261

    
2262
typedef struct BlockDriverAIOCBSync {
2263
    BlockDriverAIOCB common;
2264
    QEMUBH *bh;
2265
    int ret;
2266
    /* vector translation state */
2267
    QEMUIOVector *qiov;
2268
    uint8_t *bounce;
2269
    int is_write;
2270
} BlockDriverAIOCBSync;
2271

    
2272
static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2273
{
2274
    BlockDriverAIOCBSync *acb =
2275
        container_of(blockacb, BlockDriverAIOCBSync, common);
2276
    qemu_bh_delete(acb->bh);
2277
    acb->bh = NULL;
2278
    qemu_aio_release(acb);
2279
}
2280

    
2281
static AIOPool bdrv_em_aio_pool = {
2282
    .aiocb_size         = sizeof(BlockDriverAIOCBSync),
2283
    .cancel             = bdrv_aio_cancel_em,
2284
};
2285

    
2286
static void bdrv_aio_bh_cb(void *opaque)
2287
{
2288
    BlockDriverAIOCBSync *acb = opaque;
2289

    
2290
    if (!acb->is_write)
2291
        qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2292
    qemu_vfree(acb->bounce);
2293
    acb->common.cb(acb->common.opaque, acb->ret);
2294
    qemu_bh_delete(acb->bh);
2295
    acb->bh = NULL;
2296
    qemu_aio_release(acb);
2297
}
2298

    
2299
static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2300
                                            int64_t sector_num,
2301
                                            QEMUIOVector *qiov,
2302
                                            int nb_sectors,
2303
                                            BlockDriverCompletionFunc *cb,
2304
                                            void *opaque,
2305
                                            int is_write)
2306

    
2307
{
2308
    BlockDriverAIOCBSync *acb;
2309

    
2310
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2311
    acb->is_write = is_write;
2312
    acb->qiov = qiov;
2313
    acb->bounce = qemu_blockalign(bs, qiov->size);
2314

    
2315
    if (!acb->bh)
2316
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2317

    
2318
    if (is_write) {
2319
        qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2320
        acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2321
    } else {
2322
        acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2323
    }
2324

    
2325
    qemu_bh_schedule(acb->bh);
2326

    
2327
    return &acb->common;
2328
}
2329

    
2330
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2331
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2332
        BlockDriverCompletionFunc *cb, void *opaque)
2333
{
2334
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2335
}
2336

    
2337
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2338
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2339
        BlockDriverCompletionFunc *cb, void *opaque)
2340
{
2341
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2342
}
2343

    
2344
static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2345
        BlockDriverCompletionFunc *cb, void *opaque)
2346
{
2347
    BlockDriverAIOCBSync *acb;
2348

    
2349
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2350
    acb->is_write = 1; /* don't bounce in the completion hadler */
2351
    acb->qiov = NULL;
2352
    acb->bounce = NULL;
2353
    acb->ret = 0;
2354

    
2355
    if (!acb->bh)
2356
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2357

    
2358
    bdrv_flush(bs);
2359
    qemu_bh_schedule(acb->bh);
2360
    return &acb->common;
2361
}
2362

    
2363
static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2364
        BlockDriverCompletionFunc *cb, void *opaque)
2365
{
2366
    BlockDriverAIOCBSync *acb;
2367

    
2368
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2369
    acb->is_write = 1; /* don't bounce in the completion handler */
2370
    acb->qiov = NULL;
2371
    acb->bounce = NULL;
2372
    acb->ret = 0;
2373

    
2374
    if (!acb->bh) {
2375
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2376
    }
2377

    
2378
    qemu_bh_schedule(acb->bh);
2379
    return &acb->common;
2380
}
2381

    
2382
/**************************************************************/
2383
/* sync block device emulation */
2384

    
2385
static void bdrv_rw_em_cb(void *opaque, int ret)
2386
{
2387
    *(int *)opaque = ret;
2388
}
2389

    
2390
#define NOT_DONE 0x7fffffff
2391

    
2392
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2393
                        uint8_t *buf, int nb_sectors)
2394
{
2395
    int async_ret;
2396
    BlockDriverAIOCB *acb;
2397
    struct iovec iov;
2398
    QEMUIOVector qiov;
2399

    
2400
    async_context_push();
2401

    
2402
    async_ret = NOT_DONE;
2403
    iov.iov_base = (void *)buf;
2404
    iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2405
    qemu_iovec_init_external(&qiov, &iov, 1);
2406
    acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2407
        bdrv_rw_em_cb, &async_ret);
2408
    if (acb == NULL) {
2409
        async_ret = -1;
2410
        goto fail;
2411
    }
2412

    
2413
    while (async_ret == NOT_DONE) {
2414
        qemu_aio_wait();
2415
    }
2416

    
2417

    
2418
fail:
2419
    async_context_pop();
2420
    return async_ret;
2421
}
2422

    
2423
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2424
                         const uint8_t *buf, int nb_sectors)
2425
{
2426
    int async_ret;
2427
    BlockDriverAIOCB *acb;
2428
    struct iovec iov;
2429
    QEMUIOVector qiov;
2430

    
2431
    async_context_push();
2432

    
2433
    async_ret = NOT_DONE;
2434
    iov.iov_base = (void *)buf;
2435
    iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2436
    qemu_iovec_init_external(&qiov, &iov, 1);
2437
    acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2438
        bdrv_rw_em_cb, &async_ret);
2439
    if (acb == NULL) {
2440
        async_ret = -1;
2441
        goto fail;
2442
    }
2443
    while (async_ret == NOT_DONE) {
2444
        qemu_aio_wait();
2445
    }
2446

    
2447
fail:
2448
    async_context_pop();
2449
    return async_ret;
2450
}
2451

    
2452
void bdrv_init(void)
2453
{
2454
    module_call_init(MODULE_INIT_BLOCK);
2455
}
2456

    
2457
void bdrv_init_with_whitelist(void)
2458
{
2459
    use_bdrv_whitelist = 1;
2460
    bdrv_init();
2461
}
2462

    
2463
void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2464
                   BlockDriverCompletionFunc *cb, void *opaque)
2465
{
2466
    BlockDriverAIOCB *acb;
2467

    
2468
    if (pool->free_aiocb) {
2469
        acb = pool->free_aiocb;
2470
        pool->free_aiocb = acb->next;
2471
    } else {
2472
        acb = qemu_mallocz(pool->aiocb_size);
2473
        acb->pool = pool;
2474
    }
2475
    acb->bs = bs;
2476
    acb->cb = cb;
2477
    acb->opaque = opaque;
2478
    return acb;
2479
}
2480

    
2481
void qemu_aio_release(void *p)
2482
{
2483
    BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2484
    AIOPool *pool = acb->pool;
2485
    acb->next = pool->free_aiocb;
2486
    pool->free_aiocb = acb;
2487
}
2488

    
2489
/**************************************************************/
2490
/* removable device support */
2491

    
2492
/**
2493
 * Return TRUE if the media is present
2494
 */
2495
int bdrv_is_inserted(BlockDriverState *bs)
2496
{
2497
    BlockDriver *drv = bs->drv;
2498
    int ret;
2499
    if (!drv)
2500
        return 0;
2501
    if (!drv->bdrv_is_inserted)
2502
        return 1;
2503
    ret = drv->bdrv_is_inserted(bs);
2504
    return ret;
2505
}
2506

    
2507
/**
2508
 * Return TRUE if the media changed since the last call to this
2509
 * function. It is currently only used for floppy disks
2510
 */
2511
int bdrv_media_changed(BlockDriverState *bs)
2512
{
2513
    BlockDriver *drv = bs->drv;
2514
    int ret;
2515

    
2516
    if (!drv || !drv->bdrv_media_changed)
2517
        ret = -ENOTSUP;
2518
    else
2519
        ret = drv->bdrv_media_changed(bs);
2520
    if (ret == -ENOTSUP)
2521
        ret = bs->media_changed;
2522
    bs->media_changed = 0;
2523
    return ret;
2524
}
2525

    
2526
/**
2527
 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2528
 */
2529
int bdrv_eject(BlockDriverState *bs, int eject_flag)
2530
{
2531
    BlockDriver *drv = bs->drv;
2532
    int ret;
2533

    
2534
    if (bs->locked) {
2535
        return -EBUSY;
2536
    }
2537

    
2538
    if (!drv || !drv->bdrv_eject) {
2539
        ret = -ENOTSUP;
2540
    } else {
2541
        ret = drv->bdrv_eject(bs, eject_flag);
2542
    }
2543
    if (ret == -ENOTSUP) {
2544
        if (eject_flag)
2545
            bdrv_close(bs);
2546
        ret = 0;
2547
    }
2548

    
2549
    return ret;
2550
}
2551

    
2552
int bdrv_is_locked(BlockDriverState *bs)
2553
{
2554
    return bs->locked;
2555
}
2556

    
2557
/**
2558
 * Lock or unlock the media (if it is locked, the user won't be able
2559
 * to eject it manually).
2560
 */
2561
void bdrv_set_locked(BlockDriverState *bs, int locked)
2562
{
2563
    BlockDriver *drv = bs->drv;
2564

    
2565
    bs->locked = locked;
2566
    if (drv && drv->bdrv_set_locked) {
2567
        drv->bdrv_set_locked(bs, locked);
2568
    }
2569
}
2570

    
2571
/* needed for generic scsi interface */
2572

    
2573
int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2574
{
2575
    BlockDriver *drv = bs->drv;
2576

    
2577
    if (drv && drv->bdrv_ioctl)
2578
        return drv->bdrv_ioctl(bs, req, buf);
2579
    return -ENOTSUP;
2580
}
2581

    
2582
BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2583
        unsigned long int req, void *buf,
2584
        BlockDriverCompletionFunc *cb, void *opaque)
2585
{
2586
    BlockDriver *drv = bs->drv;
2587

    
2588
    if (drv && drv->bdrv_aio_ioctl)
2589
        return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2590
    return NULL;
2591
}
2592

    
2593

    
2594

    
2595
void *qemu_blockalign(BlockDriverState *bs, size_t size)
2596
{
2597
    return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2598
}
2599

    
2600
void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2601
{
2602
    int64_t bitmap_size;
2603

    
2604
    bs->dirty_count = 0;
2605
    if (enable) {
2606
        if (!bs->dirty_bitmap) {
2607
            bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2608
                    BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2609
            bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2610

    
2611
            bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2612
        }
2613
    } else {
2614
        if (bs->dirty_bitmap) {
2615
            qemu_free(bs->dirty_bitmap);
2616
            bs->dirty_bitmap = NULL;
2617
        }
2618
    }
2619
}
2620

    
2621
int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2622
{
2623
    int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2624

    
2625
    if (bs->dirty_bitmap &&
2626
        (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2627
        return bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2628
            (1 << (chunk % (sizeof(unsigned long) * 8)));
2629
    } else {
2630
        return 0;
2631
    }
2632
}
2633

    
2634
void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2635
                      int nr_sectors)
2636
{
2637
    set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2638
}
2639

    
2640
int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2641
{
2642
    return bs->dirty_count;
2643
}