Statistics
| Branch: | Revision:

root / block.c @ 528f7663

History | View | Annotate | Download (69.7 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
    return bs->drv->bdrv_check(bs, res);
725
}
726

    
727
/* commit COW file into the raw image */
728
int bdrv_commit(BlockDriverState *bs)
729
{
730
    BlockDriver *drv = bs->drv;
731
    int64_t i, total_sectors;
732
    int n, j, ro, open_flags;
733
    int ret = 0, rw_ret = 0;
734
    unsigned char sector[BDRV_SECTOR_SIZE];
735
    char filename[1024];
736
    BlockDriverState *bs_rw, *bs_ro;
737

    
738
    if (!drv)
739
        return -ENOMEDIUM;
740
    
741
    if (!bs->backing_hd) {
742
        return -ENOTSUP;
743
    }
744

    
745
    if (bs->backing_hd->keep_read_only) {
746
        return -EACCES;
747
    }
748
    
749
    ro = bs->backing_hd->read_only;
750
    strncpy(filename, bs->backing_hd->filename, sizeof(filename));
751
    open_flags =  bs->backing_hd->open_flags;
752

    
753
    if (ro) {
754
        /* re-open as RW */
755
        bdrv_delete(bs->backing_hd);
756
        bs->backing_hd = NULL;
757
        bs_rw = bdrv_new("");
758
        rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR, drv);
759
        if (rw_ret < 0) {
760
            bdrv_delete(bs_rw);
761
            /* try to re-open read-only */
762
            bs_ro = bdrv_new("");
763
            ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
764
            if (ret < 0) {
765
                bdrv_delete(bs_ro);
766
                /* drive not functional anymore */
767
                bs->drv = NULL;
768
                return ret;
769
            }
770
            bs->backing_hd = bs_ro;
771
            return rw_ret;
772
        }
773
        bs->backing_hd = bs_rw;
774
    }
775

    
776
    total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
777
    for (i = 0; i < total_sectors;) {
778
        if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
779
            for(j = 0; j < n; j++) {
780
                if (bdrv_read(bs, i, sector, 1) != 0) {
781
                    ret = -EIO;
782
                    goto ro_cleanup;
783
                }
784

    
785
                if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
786
                    ret = -EIO;
787
                    goto ro_cleanup;
788
                }
789
                i++;
790
            }
791
        } else {
792
            i += n;
793
        }
794
    }
795

    
796
    if (drv->bdrv_make_empty) {
797
        ret = drv->bdrv_make_empty(bs);
798
        bdrv_flush(bs);
799
    }
800

    
801
    /*
802
     * Make sure all data we wrote to the backing device is actually
803
     * stable on disk.
804
     */
805
    if (bs->backing_hd)
806
        bdrv_flush(bs->backing_hd);
807

    
808
ro_cleanup:
809

    
810
    if (ro) {
811
        /* re-open as RO */
812
        bdrv_delete(bs->backing_hd);
813
        bs->backing_hd = NULL;
814
        bs_ro = bdrv_new("");
815
        ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
816
        if (ret < 0) {
817
            bdrv_delete(bs_ro);
818
            /* drive not functional anymore */
819
            bs->drv = NULL;
820
            return ret;
821
        }
822
        bs->backing_hd = bs_ro;
823
        bs->backing_hd->keep_read_only = 0;
824
    }
825

    
826
    return ret;
827
}
828

    
829
void bdrv_commit_all(void)
830
{
831
    BlockDriverState *bs;
832

    
833
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
834
        bdrv_commit(bs);
835
    }
836
}
837

    
838
/*
839
 * Return values:
840
 * 0        - success
841
 * -EINVAL  - backing format specified, but no file
842
 * -ENOSPC  - can't update the backing file because no space is left in the
843
 *            image file header
844
 * -ENOTSUP - format driver doesn't support changing the backing file
845
 */
846
int bdrv_change_backing_file(BlockDriverState *bs,
847
    const char *backing_file, const char *backing_fmt)
848
{
849
    BlockDriver *drv = bs->drv;
850

    
851
    if (drv->bdrv_change_backing_file != NULL) {
852
        return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
853
    } else {
854
        return -ENOTSUP;
855
    }
856
}
857

    
858
static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
859
                                   size_t size)
860
{
861
    int64_t len;
862

    
863
    if (!bdrv_is_inserted(bs))
864
        return -ENOMEDIUM;
865

    
866
    if (bs->growable)
867
        return 0;
868

    
869
    len = bdrv_getlength(bs);
870

    
871
    if (offset < 0)
872
        return -EIO;
873

    
874
    if ((offset > len) || (len - offset < size))
875
        return -EIO;
876

    
877
    return 0;
878
}
879

    
880
static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
881
                              int nb_sectors)
882
{
883
    return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
884
                                   nb_sectors * BDRV_SECTOR_SIZE);
885
}
886

    
887
/* return < 0 if error. See bdrv_write() for the return codes */
888
int bdrv_read(BlockDriverState *bs, int64_t sector_num,
889
              uint8_t *buf, int nb_sectors)
890
{
891
    BlockDriver *drv = bs->drv;
892

    
893
    if (!drv)
894
        return -ENOMEDIUM;
895
    if (bdrv_check_request(bs, sector_num, nb_sectors))
896
        return -EIO;
897

    
898
    return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
899
}
900

    
901
static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
902
                             int nb_sectors, int dirty)
903
{
904
    int64_t start, end;
905
    unsigned long val, idx, bit;
906

    
907
    start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
908
    end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
909

    
910
    for (; start <= end; start++) {
911
        idx = start / (sizeof(unsigned long) * 8);
912
        bit = start % (sizeof(unsigned long) * 8);
913
        val = bs->dirty_bitmap[idx];
914
        if (dirty) {
915
            if (!(val & (1 << bit))) {
916
                bs->dirty_count++;
917
                val |= 1 << bit;
918
            }
919
        } else {
920
            if (val & (1 << bit)) {
921
                bs->dirty_count--;
922
                val &= ~(1 << bit);
923
            }
924
        }
925
        bs->dirty_bitmap[idx] = val;
926
    }
927
}
928

    
929
/* Return < 0 if error. Important errors are:
930
  -EIO         generic I/O error (may happen for all errors)
931
  -ENOMEDIUM   No media inserted.
932
  -EINVAL      Invalid sector number or nb_sectors
933
  -EACCES      Trying to write a read-only device
934
*/
935
int bdrv_write(BlockDriverState *bs, int64_t sector_num,
936
               const uint8_t *buf, int nb_sectors)
937
{
938
    BlockDriver *drv = bs->drv;
939
    if (!bs->drv)
940
        return -ENOMEDIUM;
941
    if (bs->read_only)
942
        return -EACCES;
943
    if (bdrv_check_request(bs, sector_num, nb_sectors))
944
        return -EIO;
945

    
946
    if (bs->dirty_bitmap) {
947
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
948
    }
949

    
950
    if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
951
        bs->wr_highest_sector = sector_num + nb_sectors - 1;
952
    }
953

    
954
    return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
955
}
956

    
957
int bdrv_pread(BlockDriverState *bs, int64_t offset,
958
               void *buf, int count1)
959
{
960
    uint8_t tmp_buf[BDRV_SECTOR_SIZE];
961
    int len, nb_sectors, count;
962
    int64_t sector_num;
963
    int ret;
964

    
965
    count = count1;
966
    /* first read to align to sector start */
967
    len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
968
    if (len > count)
969
        len = count;
970
    sector_num = offset >> BDRV_SECTOR_BITS;
971
    if (len > 0) {
972
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
973
            return ret;
974
        memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
975
        count -= len;
976
        if (count == 0)
977
            return count1;
978
        sector_num++;
979
        buf += len;
980
    }
981

    
982
    /* read the sectors "in place" */
983
    nb_sectors = count >> BDRV_SECTOR_BITS;
984
    if (nb_sectors > 0) {
985
        if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
986
            return ret;
987
        sector_num += nb_sectors;
988
        len = nb_sectors << BDRV_SECTOR_BITS;
989
        buf += len;
990
        count -= len;
991
    }
992

    
993
    /* add data from the last sector */
994
    if (count > 0) {
995
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
996
            return ret;
997
        memcpy(buf, tmp_buf, count);
998
    }
999
    return count1;
1000
}
1001

    
1002
int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1003
                const void *buf, int count1)
1004
{
1005
    uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1006
    int len, nb_sectors, count;
1007
    int64_t sector_num;
1008
    int ret;
1009

    
1010
    count = count1;
1011
    /* first write to align to sector start */
1012
    len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1013
    if (len > count)
1014
        len = count;
1015
    sector_num = offset >> BDRV_SECTOR_BITS;
1016
    if (len > 0) {
1017
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1018
            return ret;
1019
        memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1020
        if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1021
            return ret;
1022
        count -= len;
1023
        if (count == 0)
1024
            return count1;
1025
        sector_num++;
1026
        buf += len;
1027
    }
1028

    
1029
    /* write the sectors "in place" */
1030
    nb_sectors = count >> BDRV_SECTOR_BITS;
1031
    if (nb_sectors > 0) {
1032
        if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1033
            return ret;
1034
        sector_num += nb_sectors;
1035
        len = nb_sectors << BDRV_SECTOR_BITS;
1036
        buf += len;
1037
        count -= len;
1038
    }
1039

    
1040
    /* add data from the last sector */
1041
    if (count > 0) {
1042
        if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1043
            return ret;
1044
        memcpy(tmp_buf, buf, count);
1045
        if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1046
            return ret;
1047
    }
1048
    return count1;
1049
}
1050

    
1051
/*
1052
 * Writes to the file and ensures that no writes are reordered across this
1053
 * request (acts as a barrier)
1054
 *
1055
 * Returns 0 on success, -errno in error cases.
1056
 */
1057
int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1058
    const void *buf, int count)
1059
{
1060
    int ret;
1061

    
1062
    ret = bdrv_pwrite(bs, offset, buf, count);
1063
    if (ret < 0) {
1064
        return ret;
1065
    }
1066

    
1067
    /* No flush needed for cache=writethrough, it uses O_DSYNC */
1068
    if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1069
        bdrv_flush(bs);
1070
    }
1071

    
1072
    return 0;
1073
}
1074

    
1075
/*
1076
 * Writes to the file and ensures that no writes are reordered across this
1077
 * request (acts as a barrier)
1078
 *
1079
 * Returns 0 on success, -errno in error cases.
1080
 */
1081
int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1082
    const uint8_t *buf, int nb_sectors)
1083
{
1084
    return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1085
        buf, BDRV_SECTOR_SIZE * nb_sectors);
1086
}
1087

    
1088
/**
1089
 * Truncate file to 'offset' bytes (needed only for file protocols)
1090
 */
1091
int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1092
{
1093
    BlockDriver *drv = bs->drv;
1094
    int ret;
1095
    if (!drv)
1096
        return -ENOMEDIUM;
1097
    if (!drv->bdrv_truncate)
1098
        return -ENOTSUP;
1099
    if (bs->read_only)
1100
        return -EACCES;
1101
    ret = drv->bdrv_truncate(bs, offset);
1102
    if (ret == 0) {
1103
        ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1104
    }
1105
    return ret;
1106
}
1107

    
1108
/**
1109
 * Length of a file in bytes. Return < 0 if error or unknown.
1110
 */
1111
int64_t bdrv_getlength(BlockDriverState *bs)
1112
{
1113
    BlockDriver *drv = bs->drv;
1114
    if (!drv)
1115
        return -ENOMEDIUM;
1116

    
1117
    /* Fixed size devices use the total_sectors value for speed instead of
1118
       issuing a length query (like lseek) on each call.  Also, legacy block
1119
       drivers don't provide a bdrv_getlength function and must use
1120
       total_sectors. */
1121
    if (!bs->growable || !drv->bdrv_getlength) {
1122
        return bs->total_sectors * BDRV_SECTOR_SIZE;
1123
    }
1124
    return drv->bdrv_getlength(bs);
1125
}
1126

    
1127
/* return 0 as number of sectors if no device present or error */
1128
void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1129
{
1130
    int64_t length;
1131
    length = bdrv_getlength(bs);
1132
    if (length < 0)
1133
        length = 0;
1134
    else
1135
        length = length >> BDRV_SECTOR_BITS;
1136
    *nb_sectors_ptr = length;
1137
}
1138

    
1139
struct partition {
1140
        uint8_t boot_ind;           /* 0x80 - active */
1141
        uint8_t head;               /* starting head */
1142
        uint8_t sector;             /* starting sector */
1143
        uint8_t cyl;                /* starting cylinder */
1144
        uint8_t sys_ind;            /* What partition type */
1145
        uint8_t end_head;           /* end head */
1146
        uint8_t end_sector;         /* end sector */
1147
        uint8_t end_cyl;            /* end cylinder */
1148
        uint32_t start_sect;        /* starting sector counting from 0 */
1149
        uint32_t nr_sects;          /* nr of sectors in partition */
1150
} __attribute__((packed));
1151

    
1152
/* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1153
static int guess_disk_lchs(BlockDriverState *bs,
1154
                           int *pcylinders, int *pheads, int *psectors)
1155
{
1156
    uint8_t buf[BDRV_SECTOR_SIZE];
1157
    int ret, i, heads, sectors, cylinders;
1158
    struct partition *p;
1159
    uint32_t nr_sects;
1160
    uint64_t nb_sectors;
1161

    
1162
    bdrv_get_geometry(bs, &nb_sectors);
1163

    
1164
    ret = bdrv_read(bs, 0, buf, 1);
1165
    if (ret < 0)
1166
        return -1;
1167
    /* test msdos magic */
1168
    if (buf[510] != 0x55 || buf[511] != 0xaa)
1169
        return -1;
1170
    for(i = 0; i < 4; i++) {
1171
        p = ((struct partition *)(buf + 0x1be)) + i;
1172
        nr_sects = le32_to_cpu(p->nr_sects);
1173
        if (nr_sects && p->end_head) {
1174
            /* We make the assumption that the partition terminates on
1175
               a cylinder boundary */
1176
            heads = p->end_head + 1;
1177
            sectors = p->end_sector & 63;
1178
            if (sectors == 0)
1179
                continue;
1180
            cylinders = nb_sectors / (heads * sectors);
1181
            if (cylinders < 1 || cylinders > 16383)
1182
                continue;
1183
            *pheads = heads;
1184
            *psectors = sectors;
1185
            *pcylinders = cylinders;
1186
#if 0
1187
            printf("guessed geometry: LCHS=%d %d %d\n",
1188
                   cylinders, heads, sectors);
1189
#endif
1190
            return 0;
1191
        }
1192
    }
1193
    return -1;
1194
}
1195

    
1196
void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1197
{
1198
    int translation, lba_detected = 0;
1199
    int cylinders, heads, secs;
1200
    uint64_t nb_sectors;
1201

    
1202
    /* if a geometry hint is available, use it */
1203
    bdrv_get_geometry(bs, &nb_sectors);
1204
    bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1205
    translation = bdrv_get_translation_hint(bs);
1206
    if (cylinders != 0) {
1207
        *pcyls = cylinders;
1208
        *pheads = heads;
1209
        *psecs = secs;
1210
    } else {
1211
        if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1212
            if (heads > 16) {
1213
                /* if heads > 16, it means that a BIOS LBA
1214
                   translation was active, so the default
1215
                   hardware geometry is OK */
1216
                lba_detected = 1;
1217
                goto default_geometry;
1218
            } else {
1219
                *pcyls = cylinders;
1220
                *pheads = heads;
1221
                *psecs = secs;
1222
                /* disable any translation to be in sync with
1223
                   the logical geometry */
1224
                if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1225
                    bdrv_set_translation_hint(bs,
1226
                                              BIOS_ATA_TRANSLATION_NONE);
1227
                }
1228
            }
1229
        } else {
1230
        default_geometry:
1231
            /* if no geometry, use a standard physical disk geometry */
1232
            cylinders = nb_sectors / (16 * 63);
1233

    
1234
            if (cylinders > 16383)
1235
                cylinders = 16383;
1236
            else if (cylinders < 2)
1237
                cylinders = 2;
1238
            *pcyls = cylinders;
1239
            *pheads = 16;
1240
            *psecs = 63;
1241
            if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1242
                if ((*pcyls * *pheads) <= 131072) {
1243
                    bdrv_set_translation_hint(bs,
1244
                                              BIOS_ATA_TRANSLATION_LARGE);
1245
                } else {
1246
                    bdrv_set_translation_hint(bs,
1247
                                              BIOS_ATA_TRANSLATION_LBA);
1248
                }
1249
            }
1250
        }
1251
        bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1252
    }
1253
}
1254

    
1255
void bdrv_set_geometry_hint(BlockDriverState *bs,
1256
                            int cyls, int heads, int secs)
1257
{
1258
    bs->cyls = cyls;
1259
    bs->heads = heads;
1260
    bs->secs = secs;
1261
}
1262

    
1263
void bdrv_set_type_hint(BlockDriverState *bs, int type)
1264
{
1265
    bs->type = type;
1266
    bs->removable = ((type == BDRV_TYPE_CDROM ||
1267
                      type == BDRV_TYPE_FLOPPY));
1268
}
1269

    
1270
void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1271
{
1272
    bs->translation = translation;
1273
}
1274

    
1275
void bdrv_get_geometry_hint(BlockDriverState *bs,
1276
                            int *pcyls, int *pheads, int *psecs)
1277
{
1278
    *pcyls = bs->cyls;
1279
    *pheads = bs->heads;
1280
    *psecs = bs->secs;
1281
}
1282

    
1283
int bdrv_get_type_hint(BlockDriverState *bs)
1284
{
1285
    return bs->type;
1286
}
1287

    
1288
int bdrv_get_translation_hint(BlockDriverState *bs)
1289
{
1290
    return bs->translation;
1291
}
1292

    
1293
void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1294
                       BlockErrorAction on_write_error)
1295
{
1296
    bs->on_read_error = on_read_error;
1297
    bs->on_write_error = on_write_error;
1298
}
1299

    
1300
BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1301
{
1302
    return is_read ? bs->on_read_error : bs->on_write_error;
1303
}
1304

    
1305
void bdrv_set_removable(BlockDriverState *bs, int removable)
1306
{
1307
    bs->removable = removable;
1308
    if (removable && bs == bs_snapshots) {
1309
        bs_snapshots = NULL;
1310
    }
1311
}
1312

    
1313
int bdrv_is_removable(BlockDriverState *bs)
1314
{
1315
    return bs->removable;
1316
}
1317

    
1318
int bdrv_is_read_only(BlockDriverState *bs)
1319
{
1320
    return bs->read_only;
1321
}
1322

    
1323
int bdrv_is_sg(BlockDriverState *bs)
1324
{
1325
    return bs->sg;
1326
}
1327

    
1328
int bdrv_enable_write_cache(BlockDriverState *bs)
1329
{
1330
    return bs->enable_write_cache;
1331
}
1332

    
1333
/* XXX: no longer used */
1334
void bdrv_set_change_cb(BlockDriverState *bs,
1335
                        void (*change_cb)(void *opaque), void *opaque)
1336
{
1337
    bs->change_cb = change_cb;
1338
    bs->change_opaque = opaque;
1339
}
1340

    
1341
int bdrv_is_encrypted(BlockDriverState *bs)
1342
{
1343
    if (bs->backing_hd && bs->backing_hd->encrypted)
1344
        return 1;
1345
    return bs->encrypted;
1346
}
1347

    
1348
int bdrv_key_required(BlockDriverState *bs)
1349
{
1350
    BlockDriverState *backing_hd = bs->backing_hd;
1351

    
1352
    if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1353
        return 1;
1354
    return (bs->encrypted && !bs->valid_key);
1355
}
1356

    
1357
int bdrv_set_key(BlockDriverState *bs, const char *key)
1358
{
1359
    int ret;
1360
    if (bs->backing_hd && bs->backing_hd->encrypted) {
1361
        ret = bdrv_set_key(bs->backing_hd, key);
1362
        if (ret < 0)
1363
            return ret;
1364
        if (!bs->encrypted)
1365
            return 0;
1366
    }
1367
    if (!bs->encrypted) {
1368
        return -EINVAL;
1369
    } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1370
        return -ENOMEDIUM;
1371
    }
1372
    ret = bs->drv->bdrv_set_key(bs, key);
1373
    if (ret < 0) {
1374
        bs->valid_key = 0;
1375
    } else if (!bs->valid_key) {
1376
        bs->valid_key = 1;
1377
        /* call the change callback now, we skipped it on open */
1378
        bs->media_changed = 1;
1379
        if (bs->change_cb)
1380
            bs->change_cb(bs->change_opaque);
1381
    }
1382
    return ret;
1383
}
1384

    
1385
void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1386
{
1387
    if (!bs->drv) {
1388
        buf[0] = '\0';
1389
    } else {
1390
        pstrcpy(buf, buf_size, bs->drv->format_name);
1391
    }
1392
}
1393

    
1394
void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1395
                         void *opaque)
1396
{
1397
    BlockDriver *drv;
1398

    
1399
    QLIST_FOREACH(drv, &bdrv_drivers, list) {
1400
        it(opaque, drv->format_name);
1401
    }
1402
}
1403

    
1404
BlockDriverState *bdrv_find(const char *name)
1405
{
1406
    BlockDriverState *bs;
1407

    
1408
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1409
        if (!strcmp(name, bs->device_name)) {
1410
            return bs;
1411
        }
1412
    }
1413
    return NULL;
1414
}
1415

    
1416
BlockDriverState *bdrv_next(BlockDriverState *bs)
1417
{
1418
    if (!bs) {
1419
        return QTAILQ_FIRST(&bdrv_states);
1420
    }
1421
    return QTAILQ_NEXT(bs, list);
1422
}
1423

    
1424
void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1425
{
1426
    BlockDriverState *bs;
1427

    
1428
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1429
        it(opaque, bs);
1430
    }
1431
}
1432

    
1433
const char *bdrv_get_device_name(BlockDriverState *bs)
1434
{
1435
    return bs->device_name;
1436
}
1437

    
1438
void bdrv_flush(BlockDriverState *bs)
1439
{
1440
    if (bs->open_flags & BDRV_O_NO_FLUSH) {
1441
        return;
1442
    }
1443

    
1444
    if (bs->drv && bs->drv->bdrv_flush)
1445
        bs->drv->bdrv_flush(bs);
1446
}
1447

    
1448
void bdrv_flush_all(void)
1449
{
1450
    BlockDriverState *bs;
1451

    
1452
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1453
        if (bs->drv && !bdrv_is_read_only(bs) &&
1454
            (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1455
            bdrv_flush(bs);
1456
        }
1457
    }
1458
}
1459

    
1460
int bdrv_has_zero_init(BlockDriverState *bs)
1461
{
1462
    assert(bs->drv);
1463

    
1464
    if (bs->drv->no_zero_init) {
1465
        return 0;
1466
    } else if (bs->file) {
1467
        return bdrv_has_zero_init(bs->file);
1468
    }
1469

    
1470
    return 1;
1471
}
1472

    
1473
/*
1474
 * Returns true iff the specified sector is present in the disk image. Drivers
1475
 * not implementing the functionality are assumed to not support backing files,
1476
 * hence all their sectors are reported as allocated.
1477
 *
1478
 * 'pnum' is set to the number of sectors (including and immediately following
1479
 * the specified sector) that are known to be in the same
1480
 * allocated/unallocated state.
1481
 *
1482
 * 'nb_sectors' is the max value 'pnum' should be set to.
1483
 */
1484
int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1485
        int *pnum)
1486
{
1487
    int64_t n;
1488
    if (!bs->drv->bdrv_is_allocated) {
1489
        if (sector_num >= bs->total_sectors) {
1490
            *pnum = 0;
1491
            return 0;
1492
        }
1493
        n = bs->total_sectors - sector_num;
1494
        *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1495
        return 1;
1496
    }
1497
    return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1498
}
1499

    
1500
void bdrv_mon_event(const BlockDriverState *bdrv,
1501
                    BlockMonEventAction action, int is_read)
1502
{
1503
    QObject *data;
1504
    const char *action_str;
1505

    
1506
    switch (action) {
1507
    case BDRV_ACTION_REPORT:
1508
        action_str = "report";
1509
        break;
1510
    case BDRV_ACTION_IGNORE:
1511
        action_str = "ignore";
1512
        break;
1513
    case BDRV_ACTION_STOP:
1514
        action_str = "stop";
1515
        break;
1516
    default:
1517
        abort();
1518
    }
1519

    
1520
    data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1521
                              bdrv->device_name,
1522
                              action_str,
1523
                              is_read ? "read" : "write");
1524
    monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1525

    
1526
    qobject_decref(data);
1527
}
1528

    
1529
static void bdrv_print_dict(QObject *obj, void *opaque)
1530
{
1531
    QDict *bs_dict;
1532
    Monitor *mon = opaque;
1533

    
1534
    bs_dict = qobject_to_qdict(obj);
1535

    
1536
    monitor_printf(mon, "%s: type=%s removable=%d",
1537
                        qdict_get_str(bs_dict, "device"),
1538
                        qdict_get_str(bs_dict, "type"),
1539
                        qdict_get_bool(bs_dict, "removable"));
1540

    
1541
    if (qdict_get_bool(bs_dict, "removable")) {
1542
        monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1543
    }
1544

    
1545
    if (qdict_haskey(bs_dict, "inserted")) {
1546
        QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1547

    
1548
        monitor_printf(mon, " file=");
1549
        monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1550
        if (qdict_haskey(qdict, "backing_file")) {
1551
            monitor_printf(mon, " backing_file=");
1552
            monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1553
        }
1554
        monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1555
                            qdict_get_bool(qdict, "ro"),
1556
                            qdict_get_str(qdict, "drv"),
1557
                            qdict_get_bool(qdict, "encrypted"));
1558
    } else {
1559
        monitor_printf(mon, " [not inserted]");
1560
    }
1561

    
1562
    monitor_printf(mon, "\n");
1563
}
1564

    
1565
void bdrv_info_print(Monitor *mon, const QObject *data)
1566
{
1567
    qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1568
}
1569

    
1570
void bdrv_info(Monitor *mon, QObject **ret_data)
1571
{
1572
    QList *bs_list;
1573
    BlockDriverState *bs;
1574

    
1575
    bs_list = qlist_new();
1576

    
1577
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1578
        QObject *bs_obj;
1579
        const char *type = "unknown";
1580

    
1581
        switch(bs->type) {
1582
        case BDRV_TYPE_HD:
1583
            type = "hd";
1584
            break;
1585
        case BDRV_TYPE_CDROM:
1586
            type = "cdrom";
1587
            break;
1588
        case BDRV_TYPE_FLOPPY:
1589
            type = "floppy";
1590
            break;
1591
        }
1592

    
1593
        bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1594
                                    "'removable': %i, 'locked': %i }",
1595
                                    bs->device_name, type, bs->removable,
1596
                                    bs->locked);
1597

    
1598
        if (bs->drv) {
1599
            QObject *obj;
1600
            QDict *bs_dict = qobject_to_qdict(bs_obj);
1601

    
1602
            obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1603
                                     "'encrypted': %i }",
1604
                                     bs->filename, bs->read_only,
1605
                                     bs->drv->format_name,
1606
                                     bdrv_is_encrypted(bs));
1607
            if (bs->backing_file[0] != '\0') {
1608
                QDict *qdict = qobject_to_qdict(obj);
1609
                qdict_put(qdict, "backing_file",
1610
                          qstring_from_str(bs->backing_file));
1611
            }
1612

    
1613
            qdict_put_obj(bs_dict, "inserted", obj);
1614
        }
1615
        qlist_append_obj(bs_list, bs_obj);
1616
    }
1617

    
1618
    *ret_data = QOBJECT(bs_list);
1619
}
1620

    
1621
static void bdrv_stats_iter(QObject *data, void *opaque)
1622
{
1623
    QDict *qdict;
1624
    Monitor *mon = opaque;
1625

    
1626
    qdict = qobject_to_qdict(data);
1627
    monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1628

    
1629
    qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1630
    monitor_printf(mon, " rd_bytes=%" PRId64
1631
                        " wr_bytes=%" PRId64
1632
                        " rd_operations=%" PRId64
1633
                        " wr_operations=%" PRId64
1634
                        "\n",
1635
                        qdict_get_int(qdict, "rd_bytes"),
1636
                        qdict_get_int(qdict, "wr_bytes"),
1637
                        qdict_get_int(qdict, "rd_operations"),
1638
                        qdict_get_int(qdict, "wr_operations"));
1639
}
1640

    
1641
void bdrv_stats_print(Monitor *mon, const QObject *data)
1642
{
1643
    qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1644
}
1645

    
1646
static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1647
{
1648
    QObject *res;
1649
    QDict *dict;
1650

    
1651
    res = qobject_from_jsonf("{ 'stats': {"
1652
                             "'rd_bytes': %" PRId64 ","
1653
                             "'wr_bytes': %" PRId64 ","
1654
                             "'rd_operations': %" PRId64 ","
1655
                             "'wr_operations': %" PRId64 ","
1656
                             "'wr_highest_offset': %" PRId64
1657
                             "} }",
1658
                             bs->rd_bytes, bs->wr_bytes,
1659
                             bs->rd_ops, bs->wr_ops,
1660
                             bs->wr_highest_sector *
1661
                             (uint64_t)BDRV_SECTOR_SIZE);
1662
    dict  = qobject_to_qdict(res);
1663

    
1664
    if (*bs->device_name) {
1665
        qdict_put(dict, "device", qstring_from_str(bs->device_name));
1666
    }
1667

    
1668
    if (bs->file) {
1669
        QObject *parent = bdrv_info_stats_bs(bs->file);
1670
        qdict_put_obj(dict, "parent", parent);
1671
    }
1672

    
1673
    return res;
1674
}
1675

    
1676
void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1677
{
1678
    QObject *obj;
1679
    QList *devices;
1680
    BlockDriverState *bs;
1681

    
1682
    devices = qlist_new();
1683

    
1684
    QTAILQ_FOREACH(bs, &bdrv_states, list) {
1685
        obj = bdrv_info_stats_bs(bs);
1686
        qlist_append_obj(devices, obj);
1687
    }
1688

    
1689
    *ret_data = QOBJECT(devices);
1690
}
1691

    
1692
const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1693
{
1694
    if (bs->backing_hd && bs->backing_hd->encrypted)
1695
        return bs->backing_file;
1696
    else if (bs->encrypted)
1697
        return bs->filename;
1698
    else
1699
        return NULL;
1700
}
1701

    
1702
void bdrv_get_backing_filename(BlockDriverState *bs,
1703
                               char *filename, int filename_size)
1704
{
1705
    if (!bs->backing_file) {
1706
        pstrcpy(filename, filename_size, "");
1707
    } else {
1708
        pstrcpy(filename, filename_size, bs->backing_file);
1709
    }
1710
}
1711

    
1712
int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1713
                          const uint8_t *buf, int nb_sectors)
1714
{
1715
    BlockDriver *drv = bs->drv;
1716
    if (!drv)
1717
        return -ENOMEDIUM;
1718
    if (!drv->bdrv_write_compressed)
1719
        return -ENOTSUP;
1720
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1721
        return -EIO;
1722

    
1723
    if (bs->dirty_bitmap) {
1724
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1725
    }
1726

    
1727
    return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1728
}
1729

    
1730
int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1731
{
1732
    BlockDriver *drv = bs->drv;
1733
    if (!drv)
1734
        return -ENOMEDIUM;
1735
    if (!drv->bdrv_get_info)
1736
        return -ENOTSUP;
1737
    memset(bdi, 0, sizeof(*bdi));
1738
    return drv->bdrv_get_info(bs, bdi);
1739
}
1740

    
1741
int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1742
                      int64_t pos, int size)
1743
{
1744
    BlockDriver *drv = bs->drv;
1745
    if (!drv)
1746
        return -ENOMEDIUM;
1747
    if (drv->bdrv_save_vmstate)
1748
        return drv->bdrv_save_vmstate(bs, buf, pos, size);
1749
    if (bs->file)
1750
        return bdrv_save_vmstate(bs->file, buf, pos, size);
1751
    return -ENOTSUP;
1752
}
1753

    
1754
int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1755
                      int64_t pos, int size)
1756
{
1757
    BlockDriver *drv = bs->drv;
1758
    if (!drv)
1759
        return -ENOMEDIUM;
1760
    if (drv->bdrv_load_vmstate)
1761
        return drv->bdrv_load_vmstate(bs, buf, pos, size);
1762
    if (bs->file)
1763
        return bdrv_load_vmstate(bs->file, buf, pos, size);
1764
    return -ENOTSUP;
1765
}
1766

    
1767
void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1768
{
1769
    BlockDriver *drv = bs->drv;
1770

    
1771
    if (!drv || !drv->bdrv_debug_event) {
1772
        return;
1773
    }
1774

    
1775
    return drv->bdrv_debug_event(bs, event);
1776

    
1777
}
1778

    
1779
/**************************************************************/
1780
/* handling of snapshots */
1781

    
1782
int bdrv_can_snapshot(BlockDriverState *bs)
1783
{
1784
    BlockDriver *drv = bs->drv;
1785
    if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1786
        return 0;
1787
    }
1788

    
1789
    if (!drv->bdrv_snapshot_create) {
1790
        if (bs->file != NULL) {
1791
            return bdrv_can_snapshot(bs->file);
1792
        }
1793
        return 0;
1794
    }
1795

    
1796
    return 1;
1797
}
1798

    
1799
BlockDriverState *bdrv_snapshots(void)
1800
{
1801
    BlockDriverState *bs;
1802

    
1803
    if (bs_snapshots) {
1804
        return bs_snapshots;
1805
    }
1806

    
1807
    bs = NULL;
1808
    while ((bs = bdrv_next(bs))) {
1809
        if (bdrv_can_snapshot(bs)) {
1810
            bs_snapshots = bs;
1811
            return bs;
1812
        }
1813
    }
1814
    return NULL;
1815
}
1816

    
1817
int bdrv_snapshot_create(BlockDriverState *bs,
1818
                         QEMUSnapshotInfo *sn_info)
1819
{
1820
    BlockDriver *drv = bs->drv;
1821
    if (!drv)
1822
        return -ENOMEDIUM;
1823
    if (drv->bdrv_snapshot_create)
1824
        return drv->bdrv_snapshot_create(bs, sn_info);
1825
    if (bs->file)
1826
        return bdrv_snapshot_create(bs->file, sn_info);
1827
    return -ENOTSUP;
1828
}
1829

    
1830
int bdrv_snapshot_goto(BlockDriverState *bs,
1831
                       const char *snapshot_id)
1832
{
1833
    BlockDriver *drv = bs->drv;
1834
    int ret, open_ret;
1835

    
1836
    if (!drv)
1837
        return -ENOMEDIUM;
1838
    if (drv->bdrv_snapshot_goto)
1839
        return drv->bdrv_snapshot_goto(bs, snapshot_id);
1840

    
1841
    if (bs->file) {
1842
        drv->bdrv_close(bs);
1843
        ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1844
        open_ret = drv->bdrv_open(bs, bs->open_flags);
1845
        if (open_ret < 0) {
1846
            bdrv_delete(bs->file);
1847
            bs->drv = NULL;
1848
            return open_ret;
1849
        }
1850
        return ret;
1851
    }
1852

    
1853
    return -ENOTSUP;
1854
}
1855

    
1856
int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1857
{
1858
    BlockDriver *drv = bs->drv;
1859
    if (!drv)
1860
        return -ENOMEDIUM;
1861
    if (drv->bdrv_snapshot_delete)
1862
        return drv->bdrv_snapshot_delete(bs, snapshot_id);
1863
    if (bs->file)
1864
        return bdrv_snapshot_delete(bs->file, snapshot_id);
1865
    return -ENOTSUP;
1866
}
1867

    
1868
int bdrv_snapshot_list(BlockDriverState *bs,
1869
                       QEMUSnapshotInfo **psn_info)
1870
{
1871
    BlockDriver *drv = bs->drv;
1872
    if (!drv)
1873
        return -ENOMEDIUM;
1874
    if (drv->bdrv_snapshot_list)
1875
        return drv->bdrv_snapshot_list(bs, psn_info);
1876
    if (bs->file)
1877
        return bdrv_snapshot_list(bs->file, psn_info);
1878
    return -ENOTSUP;
1879
}
1880

    
1881
#define NB_SUFFIXES 4
1882

    
1883
char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1884
{
1885
    static const char suffixes[NB_SUFFIXES] = "KMGT";
1886
    int64_t base;
1887
    int i;
1888

    
1889
    if (size <= 999) {
1890
        snprintf(buf, buf_size, "%" PRId64, size);
1891
    } else {
1892
        base = 1024;
1893
        for(i = 0; i < NB_SUFFIXES; i++) {
1894
            if (size < (10 * base)) {
1895
                snprintf(buf, buf_size, "%0.1f%c",
1896
                         (double)size / base,
1897
                         suffixes[i]);
1898
                break;
1899
            } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1900
                snprintf(buf, buf_size, "%" PRId64 "%c",
1901
                         ((size + (base >> 1)) / base),
1902
                         suffixes[i]);
1903
                break;
1904
            }
1905
            base = base * 1024;
1906
        }
1907
    }
1908
    return buf;
1909
}
1910

    
1911
char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1912
{
1913
    char buf1[128], date_buf[128], clock_buf[128];
1914
#ifdef _WIN32
1915
    struct tm *ptm;
1916
#else
1917
    struct tm tm;
1918
#endif
1919
    time_t ti;
1920
    int64_t secs;
1921

    
1922
    if (!sn) {
1923
        snprintf(buf, buf_size,
1924
                 "%-10s%-20s%7s%20s%15s",
1925
                 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1926
    } else {
1927
        ti = sn->date_sec;
1928
#ifdef _WIN32
1929
        ptm = localtime(&ti);
1930
        strftime(date_buf, sizeof(date_buf),
1931
                 "%Y-%m-%d %H:%M:%S", ptm);
1932
#else
1933
        localtime_r(&ti, &tm);
1934
        strftime(date_buf, sizeof(date_buf),
1935
                 "%Y-%m-%d %H:%M:%S", &tm);
1936
#endif
1937
        secs = sn->vm_clock_nsec / 1000000000;
1938
        snprintf(clock_buf, sizeof(clock_buf),
1939
                 "%02d:%02d:%02d.%03d",
1940
                 (int)(secs / 3600),
1941
                 (int)((secs / 60) % 60),
1942
                 (int)(secs % 60),
1943
                 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1944
        snprintf(buf, buf_size,
1945
                 "%-10s%-20s%7s%20s%15s",
1946
                 sn->id_str, sn->name,
1947
                 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1948
                 date_buf,
1949
                 clock_buf);
1950
    }
1951
    return buf;
1952
}
1953

    
1954

    
1955
/**************************************************************/
1956
/* async I/Os */
1957

    
1958
BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1959
                                 QEMUIOVector *qiov, int nb_sectors,
1960
                                 BlockDriverCompletionFunc *cb, void *opaque)
1961
{
1962
    BlockDriver *drv = bs->drv;
1963
    BlockDriverAIOCB *ret;
1964

    
1965
    if (!drv)
1966
        return NULL;
1967
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1968
        return NULL;
1969

    
1970
    ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1971
                              cb, opaque);
1972

    
1973
    if (ret) {
1974
        /* Update stats even though technically transfer has not happened. */
1975
        bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1976
        bs->rd_ops ++;
1977
    }
1978

    
1979
    return ret;
1980
}
1981

    
1982
BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1983
                                  QEMUIOVector *qiov, int nb_sectors,
1984
                                  BlockDriverCompletionFunc *cb, void *opaque)
1985
{
1986
    BlockDriver *drv = bs->drv;
1987
    BlockDriverAIOCB *ret;
1988

    
1989
    if (!drv)
1990
        return NULL;
1991
    if (bs->read_only)
1992
        return NULL;
1993
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1994
        return NULL;
1995

    
1996
    if (bs->dirty_bitmap) {
1997
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1998
    }
1999

    
2000
    ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
2001
                               cb, opaque);
2002

    
2003
    if (ret) {
2004
        /* Update stats even though technically transfer has not happened. */
2005
        bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
2006
        bs->wr_ops ++;
2007
        if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2008
            bs->wr_highest_sector = sector_num + nb_sectors - 1;
2009
        }
2010
    }
2011

    
2012
    return ret;
2013
}
2014

    
2015

    
2016
typedef struct MultiwriteCB {
2017
    int error;
2018
    int num_requests;
2019
    int num_callbacks;
2020
    struct {
2021
        BlockDriverCompletionFunc *cb;
2022
        void *opaque;
2023
        QEMUIOVector *free_qiov;
2024
        void *free_buf;
2025
    } callbacks[];
2026
} MultiwriteCB;
2027

    
2028
static void multiwrite_user_cb(MultiwriteCB *mcb)
2029
{
2030
    int i;
2031

    
2032
    for (i = 0; i < mcb->num_callbacks; i++) {
2033
        mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2034
        if (mcb->callbacks[i].free_qiov) {
2035
            qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2036
        }
2037
        qemu_free(mcb->callbacks[i].free_qiov);
2038
        qemu_vfree(mcb->callbacks[i].free_buf);
2039
    }
2040
}
2041

    
2042
static void multiwrite_cb(void *opaque, int ret)
2043
{
2044
    MultiwriteCB *mcb = opaque;
2045

    
2046
    if (ret < 0 && !mcb->error) {
2047
        mcb->error = ret;
2048
    }
2049

    
2050
    mcb->num_requests--;
2051
    if (mcb->num_requests == 0) {
2052
        multiwrite_user_cb(mcb);
2053
        qemu_free(mcb);
2054
    }
2055
}
2056

    
2057
static int multiwrite_req_compare(const void *a, const void *b)
2058
{
2059
    const BlockRequest *req1 = a, *req2 = b;
2060

    
2061
    /*
2062
     * Note that we can't simply subtract req2->sector from req1->sector
2063
     * here as that could overflow the return value.
2064
     */
2065
    if (req1->sector > req2->sector) {
2066
        return 1;
2067
    } else if (req1->sector < req2->sector) {
2068
        return -1;
2069
    } else {
2070
        return 0;
2071
    }
2072
}
2073

    
2074
/*
2075
 * Takes a bunch of requests and tries to merge them. Returns the number of
2076
 * requests that remain after merging.
2077
 */
2078
static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2079
    int num_reqs, MultiwriteCB *mcb)
2080
{
2081
    int i, outidx;
2082

    
2083
    // Sort requests by start sector
2084
    qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2085

    
2086
    // Check if adjacent requests touch the same clusters. If so, combine them,
2087
    // filling up gaps with zero sectors.
2088
    outidx = 0;
2089
    for (i = 1; i < num_reqs; i++) {
2090
        int merge = 0;
2091
        int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2092

    
2093
        // This handles the cases that are valid for all block drivers, namely
2094
        // exactly sequential writes and overlapping writes.
2095
        if (reqs[i].sector <= oldreq_last) {
2096
            merge = 1;
2097
        }
2098

    
2099
        // The block driver may decide that it makes sense to combine requests
2100
        // even if there is a gap of some sectors between them. In this case,
2101
        // the gap is filled with zeros (therefore only applicable for yet
2102
        // unused space in format like qcow2).
2103
        if (!merge && bs->drv->bdrv_merge_requests) {
2104
            merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2105
        }
2106

    
2107
        if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2108
            merge = 0;
2109
        }
2110

    
2111
        if (merge) {
2112
            size_t size;
2113
            QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2114
            qemu_iovec_init(qiov,
2115
                reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2116

    
2117
            // Add the first request to the merged one. If the requests are
2118
            // overlapping, drop the last sectors of the first request.
2119
            size = (reqs[i].sector - reqs[outidx].sector) << 9;
2120
            qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2121

    
2122
            // We might need to add some zeros between the two requests
2123
            if (reqs[i].sector > oldreq_last) {
2124
                size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2125
                uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2126
                memset(buf, 0, zero_bytes);
2127
                qemu_iovec_add(qiov, buf, zero_bytes);
2128
                mcb->callbacks[i].free_buf = buf;
2129
            }
2130

    
2131
            // Add the second request
2132
            qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2133

    
2134
            reqs[outidx].nb_sectors = qiov->size >> 9;
2135
            reqs[outidx].qiov = qiov;
2136

    
2137
            mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2138
        } else {
2139
            outidx++;
2140
            reqs[outidx].sector     = reqs[i].sector;
2141
            reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2142
            reqs[outidx].qiov       = reqs[i].qiov;
2143
        }
2144
    }
2145

    
2146
    return outidx + 1;
2147
}
2148

    
2149
/*
2150
 * Submit multiple AIO write requests at once.
2151
 *
2152
 * On success, the function returns 0 and all requests in the reqs array have
2153
 * been submitted. In error case this function returns -1, and any of the
2154
 * requests may or may not be submitted yet. In particular, this means that the
2155
 * callback will be called for some of the requests, for others it won't. The
2156
 * caller must check the error field of the BlockRequest to wait for the right
2157
 * callbacks (if error != 0, no callback will be called).
2158
 *
2159
 * The implementation may modify the contents of the reqs array, e.g. to merge
2160
 * requests. However, the fields opaque and error are left unmodified as they
2161
 * are used to signal failure for a single request to the caller.
2162
 */
2163
int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2164
{
2165
    BlockDriverAIOCB *acb;
2166
    MultiwriteCB *mcb;
2167
    int i;
2168

    
2169
    if (num_reqs == 0) {
2170
        return 0;
2171
    }
2172

    
2173
    // Create MultiwriteCB structure
2174
    mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2175
    mcb->num_requests = 0;
2176
    mcb->num_callbacks = num_reqs;
2177

    
2178
    for (i = 0; i < num_reqs; i++) {
2179
        mcb->callbacks[i].cb = reqs[i].cb;
2180
        mcb->callbacks[i].opaque = reqs[i].opaque;
2181
    }
2182

    
2183
    // Check for mergable requests
2184
    num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2185

    
2186
    /*
2187
     * Run the aio requests. As soon as one request can't be submitted
2188
     * successfully, fail all requests that are not yet submitted (we must
2189
     * return failure for all requests anyway)
2190
     *
2191
     * num_requests cannot be set to the right value immediately: If
2192
     * bdrv_aio_writev fails for some request, num_requests would be too high
2193
     * and therefore multiwrite_cb() would never recognize the multiwrite
2194
     * request as completed. We also cannot use the loop variable i to set it
2195
     * when the first request fails because the callback may already have been
2196
     * called for previously submitted requests. Thus, num_requests must be
2197
     * incremented for each request that is submitted.
2198
     *
2199
     * The problem that callbacks may be called early also means that we need
2200
     * to take care that num_requests doesn't become 0 before all requests are
2201
     * submitted - multiwrite_cb() would consider the multiwrite request
2202
     * completed. A dummy request that is "completed" by a manual call to
2203
     * multiwrite_cb() takes care of this.
2204
     */
2205
    mcb->num_requests = 1;
2206

    
2207
    for (i = 0; i < num_reqs; i++) {
2208
        mcb->num_requests++;
2209
        acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2210
            reqs[i].nb_sectors, multiwrite_cb, mcb);
2211

    
2212
        if (acb == NULL) {
2213
            // We can only fail the whole thing if no request has been
2214
            // submitted yet. Otherwise we'll wait for the submitted AIOs to
2215
            // complete and report the error in the callback.
2216
            if (i == 0) {
2217
                goto fail;
2218
            } else {
2219
                multiwrite_cb(mcb, -EIO);
2220
                break;
2221
            }
2222
        }
2223
    }
2224

    
2225
    /* Complete the dummy request */
2226
    multiwrite_cb(mcb, 0);
2227

    
2228
    return 0;
2229

    
2230
fail:
2231
    for (i = 0; i < mcb->num_callbacks; i++) {
2232
        reqs[i].error = -EIO;
2233
    }
2234
    qemu_free(mcb);
2235
    return -1;
2236
}
2237

    
2238
BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2239
        BlockDriverCompletionFunc *cb, void *opaque)
2240
{
2241
    BlockDriver *drv = bs->drv;
2242

    
2243
    if (bs->open_flags & BDRV_O_NO_FLUSH) {
2244
        return bdrv_aio_noop_em(bs, cb, opaque);
2245
    }
2246

    
2247
    if (!drv)
2248
        return NULL;
2249
    return drv->bdrv_aio_flush(bs, cb, opaque);
2250
}
2251

    
2252
void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2253
{
2254
    acb->pool->cancel(acb);
2255
}
2256

    
2257

    
2258
/**************************************************************/
2259
/* async block device emulation */
2260

    
2261
typedef struct BlockDriverAIOCBSync {
2262
    BlockDriverAIOCB common;
2263
    QEMUBH *bh;
2264
    int ret;
2265
    /* vector translation state */
2266
    QEMUIOVector *qiov;
2267
    uint8_t *bounce;
2268
    int is_write;
2269
} BlockDriverAIOCBSync;
2270

    
2271
static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2272
{
2273
    BlockDriverAIOCBSync *acb =
2274
        container_of(blockacb, BlockDriverAIOCBSync, common);
2275
    qemu_bh_delete(acb->bh);
2276
    acb->bh = NULL;
2277
    qemu_aio_release(acb);
2278
}
2279

    
2280
static AIOPool bdrv_em_aio_pool = {
2281
    .aiocb_size         = sizeof(BlockDriverAIOCBSync),
2282
    .cancel             = bdrv_aio_cancel_em,
2283
};
2284

    
2285
static void bdrv_aio_bh_cb(void *opaque)
2286
{
2287
    BlockDriverAIOCBSync *acb = opaque;
2288

    
2289
    if (!acb->is_write)
2290
        qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2291
    qemu_vfree(acb->bounce);
2292
    acb->common.cb(acb->common.opaque, acb->ret);
2293
    qemu_bh_delete(acb->bh);
2294
    acb->bh = NULL;
2295
    qemu_aio_release(acb);
2296
}
2297

    
2298
static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2299
                                            int64_t sector_num,
2300
                                            QEMUIOVector *qiov,
2301
                                            int nb_sectors,
2302
                                            BlockDriverCompletionFunc *cb,
2303
                                            void *opaque,
2304
                                            int is_write)
2305

    
2306
{
2307
    BlockDriverAIOCBSync *acb;
2308

    
2309
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2310
    acb->is_write = is_write;
2311
    acb->qiov = qiov;
2312
    acb->bounce = qemu_blockalign(bs, qiov->size);
2313

    
2314
    if (!acb->bh)
2315
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2316

    
2317
    if (is_write) {
2318
        qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2319
        acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2320
    } else {
2321
        acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2322
    }
2323

    
2324
    qemu_bh_schedule(acb->bh);
2325

    
2326
    return &acb->common;
2327
}
2328

    
2329
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2330
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2331
        BlockDriverCompletionFunc *cb, void *opaque)
2332
{
2333
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2334
}
2335

    
2336
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2337
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2338
        BlockDriverCompletionFunc *cb, void *opaque)
2339
{
2340
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2341
}
2342

    
2343
static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2344
        BlockDriverCompletionFunc *cb, void *opaque)
2345
{
2346
    BlockDriverAIOCBSync *acb;
2347

    
2348
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2349
    acb->is_write = 1; /* don't bounce in the completion hadler */
2350
    acb->qiov = NULL;
2351
    acb->bounce = NULL;
2352
    acb->ret = 0;
2353

    
2354
    if (!acb->bh)
2355
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2356

    
2357
    bdrv_flush(bs);
2358
    qemu_bh_schedule(acb->bh);
2359
    return &acb->common;
2360
}
2361

    
2362
static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2363
        BlockDriverCompletionFunc *cb, void *opaque)
2364
{
2365
    BlockDriverAIOCBSync *acb;
2366

    
2367
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2368
    acb->is_write = 1; /* don't bounce in the completion handler */
2369
    acb->qiov = NULL;
2370
    acb->bounce = NULL;
2371
    acb->ret = 0;
2372

    
2373
    if (!acb->bh) {
2374
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2375
    }
2376

    
2377
    qemu_bh_schedule(acb->bh);
2378
    return &acb->common;
2379
}
2380

    
2381
/**************************************************************/
2382
/* sync block device emulation */
2383

    
2384
static void bdrv_rw_em_cb(void *opaque, int ret)
2385
{
2386
    *(int *)opaque = ret;
2387
}
2388

    
2389
#define NOT_DONE 0x7fffffff
2390

    
2391
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2392
                        uint8_t *buf, int nb_sectors)
2393
{
2394
    int async_ret;
2395
    BlockDriverAIOCB *acb;
2396
    struct iovec iov;
2397
    QEMUIOVector qiov;
2398

    
2399
    async_context_push();
2400

    
2401
    async_ret = NOT_DONE;
2402
    iov.iov_base = (void *)buf;
2403
    iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2404
    qemu_iovec_init_external(&qiov, &iov, 1);
2405
    acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2406
        bdrv_rw_em_cb, &async_ret);
2407
    if (acb == NULL) {
2408
        async_ret = -1;
2409
        goto fail;
2410
    }
2411

    
2412
    while (async_ret == NOT_DONE) {
2413
        qemu_aio_wait();
2414
    }
2415

    
2416

    
2417
fail:
2418
    async_context_pop();
2419
    return async_ret;
2420
}
2421

    
2422
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2423
                         const uint8_t *buf, int nb_sectors)
2424
{
2425
    int async_ret;
2426
    BlockDriverAIOCB *acb;
2427
    struct iovec iov;
2428
    QEMUIOVector qiov;
2429

    
2430
    async_context_push();
2431

    
2432
    async_ret = NOT_DONE;
2433
    iov.iov_base = (void *)buf;
2434
    iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2435
    qemu_iovec_init_external(&qiov, &iov, 1);
2436
    acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2437
        bdrv_rw_em_cb, &async_ret);
2438
    if (acb == NULL) {
2439
        async_ret = -1;
2440
        goto fail;
2441
    }
2442
    while (async_ret == NOT_DONE) {
2443
        qemu_aio_wait();
2444
    }
2445

    
2446
fail:
2447
    async_context_pop();
2448
    return async_ret;
2449
}
2450

    
2451
void bdrv_init(void)
2452
{
2453
    module_call_init(MODULE_INIT_BLOCK);
2454
}
2455

    
2456
void bdrv_init_with_whitelist(void)
2457
{
2458
    use_bdrv_whitelist = 1;
2459
    bdrv_init();
2460
}
2461

    
2462
void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2463
                   BlockDriverCompletionFunc *cb, void *opaque)
2464
{
2465
    BlockDriverAIOCB *acb;
2466

    
2467
    if (pool->free_aiocb) {
2468
        acb = pool->free_aiocb;
2469
        pool->free_aiocb = acb->next;
2470
    } else {
2471
        acb = qemu_mallocz(pool->aiocb_size);
2472
        acb->pool = pool;
2473
    }
2474
    acb->bs = bs;
2475
    acb->cb = cb;
2476
    acb->opaque = opaque;
2477
    return acb;
2478
}
2479

    
2480
void qemu_aio_release(void *p)
2481
{
2482
    BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2483
    AIOPool *pool = acb->pool;
2484
    acb->next = pool->free_aiocb;
2485
    pool->free_aiocb = acb;
2486
}
2487

    
2488
/**************************************************************/
2489
/* removable device support */
2490

    
2491
/**
2492
 * Return TRUE if the media is present
2493
 */
2494
int bdrv_is_inserted(BlockDriverState *bs)
2495
{
2496
    BlockDriver *drv = bs->drv;
2497
    int ret;
2498
    if (!drv)
2499
        return 0;
2500
    if (!drv->bdrv_is_inserted)
2501
        return 1;
2502
    ret = drv->bdrv_is_inserted(bs);
2503
    return ret;
2504
}
2505

    
2506
/**
2507
 * Return TRUE if the media changed since the last call to this
2508
 * function. It is currently only used for floppy disks
2509
 */
2510
int bdrv_media_changed(BlockDriverState *bs)
2511
{
2512
    BlockDriver *drv = bs->drv;
2513
    int ret;
2514

    
2515
    if (!drv || !drv->bdrv_media_changed)
2516
        ret = -ENOTSUP;
2517
    else
2518
        ret = drv->bdrv_media_changed(bs);
2519
    if (ret == -ENOTSUP)
2520
        ret = bs->media_changed;
2521
    bs->media_changed = 0;
2522
    return ret;
2523
}
2524

    
2525
/**
2526
 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2527
 */
2528
int bdrv_eject(BlockDriverState *bs, int eject_flag)
2529
{
2530
    BlockDriver *drv = bs->drv;
2531
    int ret;
2532

    
2533
    if (bs->locked) {
2534
        return -EBUSY;
2535
    }
2536

    
2537
    if (!drv || !drv->bdrv_eject) {
2538
        ret = -ENOTSUP;
2539
    } else {
2540
        ret = drv->bdrv_eject(bs, eject_flag);
2541
    }
2542
    if (ret == -ENOTSUP) {
2543
        if (eject_flag)
2544
            bdrv_close(bs);
2545
        ret = 0;
2546
    }
2547

    
2548
    return ret;
2549
}
2550

    
2551
int bdrv_is_locked(BlockDriverState *bs)
2552
{
2553
    return bs->locked;
2554
}
2555

    
2556
/**
2557
 * Lock or unlock the media (if it is locked, the user won't be able
2558
 * to eject it manually).
2559
 */
2560
void bdrv_set_locked(BlockDriverState *bs, int locked)
2561
{
2562
    BlockDriver *drv = bs->drv;
2563

    
2564
    bs->locked = locked;
2565
    if (drv && drv->bdrv_set_locked) {
2566
        drv->bdrv_set_locked(bs, locked);
2567
    }
2568
}
2569

    
2570
/* needed for generic scsi interface */
2571

    
2572
int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2573
{
2574
    BlockDriver *drv = bs->drv;
2575

    
2576
    if (drv && drv->bdrv_ioctl)
2577
        return drv->bdrv_ioctl(bs, req, buf);
2578
    return -ENOTSUP;
2579
}
2580

    
2581
BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2582
        unsigned long int req, void *buf,
2583
        BlockDriverCompletionFunc *cb, void *opaque)
2584
{
2585
    BlockDriver *drv = bs->drv;
2586

    
2587
    if (drv && drv->bdrv_aio_ioctl)
2588
        return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2589
    return NULL;
2590
}
2591

    
2592

    
2593

    
2594
void *qemu_blockalign(BlockDriverState *bs, size_t size)
2595
{
2596
    return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2597
}
2598

    
2599
void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2600
{
2601
    int64_t bitmap_size;
2602

    
2603
    bs->dirty_count = 0;
2604
    if (enable) {
2605
        if (!bs->dirty_bitmap) {
2606
            bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2607
                    BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2608
            bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2609

    
2610
            bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2611
        }
2612
    } else {
2613
        if (bs->dirty_bitmap) {
2614
            qemu_free(bs->dirty_bitmap);
2615
            bs->dirty_bitmap = NULL;
2616
        }
2617
    }
2618
}
2619

    
2620
int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2621
{
2622
    int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2623

    
2624
    if (bs->dirty_bitmap &&
2625
        (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2626
        return bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2627
            (1 << (chunk % (sizeof(unsigned long) * 8)));
2628
    } else {
2629
        return 0;
2630
    }
2631
}
2632

    
2633
void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2634
                      int nr_sectors)
2635
{
2636
    set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2637
}
2638

    
2639
int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2640
{
2641
    return bs->dirty_count;
2642
}