Statistics
| Branch: | Revision:

root / block.c @ a55eb92c

History | View | Annotate | Download (53.4 kB)

1
/*
2
 * QEMU System Emulator block driver
3
 *
4
 * Copyright (c) 2003 Fabrice Bellard
5
 *
6
 * Permission is hereby granted, free of charge, to any person obtaining a copy
7
 * of this software and associated documentation files (the "Software"), to deal
8
 * in the Software without restriction, including without limitation the rights
9
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10
 * copies of the Software, and to permit persons to whom the Software is
11
 * furnished to do so, subject to the following conditions:
12
 *
13
 * The above copyright notice and this permission notice shall be included in
14
 * all copies or substantial portions of the Software.
15
 *
16
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22
 * THE SOFTWARE.
23
 */
24
#include "config-host.h"
25
#include "qemu-common.h"
26
#include "monitor.h"
27
#include "block_int.h"
28
#include "module.h"
29

    
30
#ifdef CONFIG_BSD
31
#include <sys/types.h>
32
#include <sys/stat.h>
33
#include <sys/ioctl.h>
34
#include <sys/queue.h>
35
#ifndef __DragonFly__
36
#include <sys/disk.h>
37
#endif
38
#endif
39

    
40
#ifdef _WIN32
41
#include <windows.h>
42
#endif
43

    
44
#define SECTOR_BITS 9
45
#define SECTOR_SIZE (1 << SECTOR_BITS)
46
#define SECTORS_PER_DIRTY_CHUNK 8
47

    
48
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
49
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
50
        BlockDriverCompletionFunc *cb, void *opaque);
51
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
52
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
53
        BlockDriverCompletionFunc *cb, void *opaque);
54
static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
55
        BlockDriverCompletionFunc *cb, void *opaque);
56
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
57
                        uint8_t *buf, int nb_sectors);
58
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
59
                         const uint8_t *buf, int nb_sectors);
60

    
61
BlockDriverState *bdrv_first;
62

    
63
static BlockDriver *first_drv;
64

    
65
/* If non-zero, use only whitelisted block drivers */
66
static int use_bdrv_whitelist;
67

    
68
int path_is_absolute(const char *path)
69
{
70
    const char *p;
71
#ifdef _WIN32
72
    /* specific case for names like: "\\.\d:" */
73
    if (*path == '/' || *path == '\\')
74
        return 1;
75
#endif
76
    p = strchr(path, ':');
77
    if (p)
78
        p++;
79
    else
80
        p = path;
81
#ifdef _WIN32
82
    return (*p == '/' || *p == '\\');
83
#else
84
    return (*p == '/');
85
#endif
86
}
87

    
88
/* if filename is absolute, just copy it to dest. Otherwise, build a
89
   path to it by considering it is relative to base_path. URL are
90
   supported. */
91
void path_combine(char *dest, int dest_size,
92
                  const char *base_path,
93
                  const char *filename)
94
{
95
    const char *p, *p1;
96
    int len;
97

    
98
    if (dest_size <= 0)
99
        return;
100
    if (path_is_absolute(filename)) {
101
        pstrcpy(dest, dest_size, filename);
102
    } else {
103
        p = strchr(base_path, ':');
104
        if (p)
105
            p++;
106
        else
107
            p = base_path;
108
        p1 = strrchr(base_path, '/');
109
#ifdef _WIN32
110
        {
111
            const char *p2;
112
            p2 = strrchr(base_path, '\\');
113
            if (!p1 || p2 > p1)
114
                p1 = p2;
115
        }
116
#endif
117
        if (p1)
118
            p1++;
119
        else
120
            p1 = base_path;
121
        if (p1 > p)
122
            p = p1;
123
        len = p - base_path;
124
        if (len > dest_size - 1)
125
            len = dest_size - 1;
126
        memcpy(dest, base_path, len);
127
        dest[len] = '\0';
128
        pstrcat(dest, dest_size, filename);
129
    }
130
}
131

    
132
void bdrv_register(BlockDriver *bdrv)
133
{
134
    if (!bdrv->bdrv_aio_readv) {
135
        /* add AIO emulation layer */
136
        bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
137
        bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
138
    } else if (!bdrv->bdrv_read) {
139
        /* add synchronous IO emulation layer */
140
        bdrv->bdrv_read = bdrv_read_em;
141
        bdrv->bdrv_write = bdrv_write_em;
142
    }
143

    
144
    if (!bdrv->bdrv_aio_flush)
145
        bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
146

    
147
    bdrv->next = first_drv;
148
    first_drv = bdrv;
149
}
150

    
151
/* create a new block device (by default it is empty) */
152
BlockDriverState *bdrv_new(const char *device_name)
153
{
154
    BlockDriverState **pbs, *bs;
155

    
156
    bs = qemu_mallocz(sizeof(BlockDriverState));
157
    pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
158
    if (device_name[0] != '\0') {
159
        /* insert at the end */
160
        pbs = &bdrv_first;
161
        while (*pbs != NULL)
162
            pbs = &(*pbs)->next;
163
        *pbs = bs;
164
    }
165
    return bs;
166
}
167

    
168
BlockDriver *bdrv_find_format(const char *format_name)
169
{
170
    BlockDriver *drv1;
171
    for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
172
        if (!strcmp(drv1->format_name, format_name))
173
            return drv1;
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
#ifdef _WIN32
212
void get_tmp_filename(char *filename, int size)
213
{
214
    char temp_dir[MAX_PATH];
215

    
216
    GetTempPath(MAX_PATH, temp_dir);
217
    GetTempFileName(temp_dir, "qem", 0, filename);
218
}
219
#else
220
void get_tmp_filename(char *filename, int size)
221
{
222
    int fd;
223
    const char *tmpdir;
224
    /* XXX: race condition possible */
225
    tmpdir = getenv("TMPDIR");
226
    if (!tmpdir)
227
        tmpdir = "/tmp";
228
    snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
229
    fd = mkstemp(filename);
230
    close(fd);
231
}
232
#endif
233

    
234
#ifdef _WIN32
235
static int is_windows_drive_prefix(const char *filename)
236
{
237
    return (((filename[0] >= 'a' && filename[0] <= 'z') ||
238
             (filename[0] >= 'A' && filename[0] <= 'Z')) &&
239
            filename[1] == ':');
240
}
241

    
242
int is_windows_drive(const char *filename)
243
{
244
    if (is_windows_drive_prefix(filename) &&
245
        filename[2] == '\0')
246
        return 1;
247
    if (strstart(filename, "\\\\.\\", NULL) ||
248
        strstart(filename, "//./", NULL))
249
        return 1;
250
    return 0;
251
}
252
#endif
253

    
254
static BlockDriver *find_protocol(const char *filename)
255
{
256
    BlockDriver *drv1;
257
    char protocol[128];
258
    int len;
259
    const char *p;
260

    
261
#ifdef _WIN32
262
    if (is_windows_drive(filename) ||
263
        is_windows_drive_prefix(filename))
264
        return bdrv_find_format("raw");
265
#endif
266
    p = strchr(filename, ':');
267
    if (!p)
268
        return bdrv_find_format("raw");
269
    len = p - filename;
270
    if (len > sizeof(protocol) - 1)
271
        len = sizeof(protocol) - 1;
272
    memcpy(protocol, filename, len);
273
    protocol[len] = '\0';
274
    for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
275
        if (drv1->protocol_name &&
276
            !strcmp(drv1->protocol_name, protocol))
277
            return drv1;
278
    }
279
    return NULL;
280
}
281

    
282
/*
283
 * Detect host devices. By convention, /dev/cdrom[N] is always
284
 * recognized as a host CDROM.
285
 */
286
static BlockDriver *find_hdev_driver(const char *filename)
287
{
288
    int score_max = 0, score;
289
    BlockDriver *drv = NULL, *d;
290

    
291
    for (d = first_drv; d; d = d->next) {
292
        if (d->bdrv_probe_device) {
293
            score = d->bdrv_probe_device(filename);
294
            if (score > score_max) {
295
                score_max = score;
296
                drv = d;
297
            }
298
        }
299
    }
300

    
301
    return drv;
302
}
303

    
304
static BlockDriver *find_image_format(const char *filename)
305
{
306
    int ret, score, score_max;
307
    BlockDriver *drv1, *drv;
308
    uint8_t buf[2048];
309
    BlockDriverState *bs;
310

    
311
    drv = find_protocol(filename);
312
    /* no need to test disk image formats for vvfat */
313
    if (drv && strcmp(drv->format_name, "vvfat") == 0)
314
        return drv;
315

    
316
    ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
317
    if (ret < 0)
318
        return NULL;
319
    ret = bdrv_pread(bs, 0, buf, sizeof(buf));
320
    bdrv_delete(bs);
321
    if (ret < 0) {
322
        return NULL;
323
    }
324

    
325
    score_max = 0;
326
    for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
327
        if (drv1->bdrv_probe) {
328
            score = drv1->bdrv_probe(buf, ret, filename);
329
            if (score > score_max) {
330
                score_max = score;
331
                drv = drv1;
332
            }
333
        }
334
    }
335
    return drv;
336
}
337

    
338
int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
339
{
340
    BlockDriverState *bs;
341
    int ret;
342

    
343
    bs = bdrv_new("");
344
    ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
345
    if (ret < 0) {
346
        bdrv_delete(bs);
347
        return ret;
348
    }
349
    bs->growable = 1;
350
    *pbs = bs;
351
    return 0;
352
}
353

    
354
int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
355
{
356
    return bdrv_open2(bs, filename, flags, NULL);
357
}
358

    
359
int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
360
               BlockDriver *drv)
361
{
362
    int ret, open_flags, try_rw;
363
    char tmp_filename[PATH_MAX];
364
    char backing_filename[PATH_MAX];
365

    
366
    bs->is_temporary = 0;
367
    bs->encrypted = 0;
368
    bs->valid_key = 0;
369
    /* buffer_alignment defaulted to 512, drivers can change this value */
370
    bs->buffer_alignment = 512;
371

    
372
    if (flags & BDRV_O_SNAPSHOT) {
373
        BlockDriverState *bs1;
374
        int64_t total_size;
375
        int is_protocol = 0;
376
        BlockDriver *bdrv_qcow2;
377
        QEMUOptionParameter *options;
378

    
379
        /* if snapshot, we create a temporary backing file and open it
380
           instead of opening 'filename' directly */
381

    
382
        /* if there is a backing file, use it */
383
        bs1 = bdrv_new("");
384
        ret = bdrv_open2(bs1, filename, 0, drv);
385
        if (ret < 0) {
386
            bdrv_delete(bs1);
387
            return ret;
388
        }
389
        total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
390

    
391
        if (bs1->drv && bs1->drv->protocol_name)
392
            is_protocol = 1;
393

    
394
        bdrv_delete(bs1);
395

    
396
        get_tmp_filename(tmp_filename, sizeof(tmp_filename));
397

    
398
        /* Real path is meaningless for protocols */
399
        if (is_protocol)
400
            snprintf(backing_filename, sizeof(backing_filename),
401
                     "%s", filename);
402
        else
403
            realpath(filename, backing_filename);
404

    
405
        bdrv_qcow2 = bdrv_find_format("qcow2");
406
        options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
407

    
408
        set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size * 512);
409
        set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
410
        if (drv) {
411
            set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
412
                drv->format_name);
413
        }
414

    
415
        ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
416
        if (ret < 0) {
417
            return ret;
418
        }
419

    
420
        filename = tmp_filename;
421
        drv = bdrv_qcow2;
422
        bs->is_temporary = 1;
423
    }
424

    
425
    pstrcpy(bs->filename, sizeof(bs->filename), filename);
426
    if (flags & BDRV_O_FILE) {
427
        drv = find_protocol(filename);
428
    } else if (!drv) {
429
        drv = find_hdev_driver(filename);
430
        if (!drv) {
431
            drv = find_image_format(filename);
432
        }
433
    }
434
    if (!drv) {
435
        ret = -ENOENT;
436
        goto unlink_and_fail;
437
    }
438
    bs->drv = drv;
439
    bs->opaque = qemu_mallocz(drv->instance_size);
440

    
441
    /*
442
     * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
443
     * write cache to the guest.  We do need the fdatasync to flush
444
     * out transactions for block allocations, and we maybe have a
445
     * volatile write cache in our backing device to deal with.
446
     */
447
    if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
448
        bs->enable_write_cache = 1;
449

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
600
    return 0;
601
}
602

    
603
static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
604
                                   size_t size)
605
{
606
    int64_t len;
607

    
608
    if (!bdrv_is_inserted(bs))
609
        return -ENOMEDIUM;
610

    
611
    if (bs->growable)
612
        return 0;
613

    
614
    len = bdrv_getlength(bs);
615

    
616
    if (offset < 0)
617
        return -EIO;
618

    
619
    if ((offset > len) || (len - offset < size))
620
        return -EIO;
621

    
622
    return 0;
623
}
624

    
625
static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
626
                              int nb_sectors)
627
{
628
    return bdrv_check_byte_request(bs, sector_num * 512, nb_sectors * 512);
629
}
630

    
631
/* return < 0 if error. See bdrv_write() for the return codes */
632
int bdrv_read(BlockDriverState *bs, int64_t sector_num,
633
              uint8_t *buf, int nb_sectors)
634
{
635
    BlockDriver *drv = bs->drv;
636

    
637
    if (!drv)
638
        return -ENOMEDIUM;
639
    if (bdrv_check_request(bs, sector_num, nb_sectors))
640
        return -EIO;
641

    
642
    return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
643
}
644

    
645
static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
646
                             int nb_sectors, int dirty)
647
{
648
    int64_t start, end;
649

    
650
    start = sector_num / SECTORS_PER_DIRTY_CHUNK;
651
    end = (sector_num + nb_sectors) / SECTORS_PER_DIRTY_CHUNK;
652

    
653
    for (; start <= end; start++) {
654
        bs->dirty_bitmap[start] = dirty;
655
    }
656
}
657

    
658
/* Return < 0 if error. Important errors are:
659
  -EIO         generic I/O error (may happen for all errors)
660
  -ENOMEDIUM   No media inserted.
661
  -EINVAL      Invalid sector number or nb_sectors
662
  -EACCES      Trying to write a read-only device
663
*/
664
int bdrv_write(BlockDriverState *bs, int64_t sector_num,
665
               const uint8_t *buf, int nb_sectors)
666
{
667
    BlockDriver *drv = bs->drv;
668
    if (!bs->drv)
669
        return -ENOMEDIUM;
670
    if (bs->read_only)
671
        return -EACCES;
672
    if (bdrv_check_request(bs, sector_num, nb_sectors))
673
        return -EIO;
674

    
675
    if (bs->dirty_tracking) {
676
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
677
    }
678

    
679
    return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
680
}
681

    
682
int bdrv_pread(BlockDriverState *bs, int64_t offset,
683
               void *buf, int count1)
684
{
685
    uint8_t tmp_buf[SECTOR_SIZE];
686
    int len, nb_sectors, count;
687
    int64_t sector_num;
688

    
689
    count = count1;
690
    /* first read to align to sector start */
691
    len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
692
    if (len > count)
693
        len = count;
694
    sector_num = offset >> SECTOR_BITS;
695
    if (len > 0) {
696
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
697
            return -EIO;
698
        memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
699
        count -= len;
700
        if (count == 0)
701
            return count1;
702
        sector_num++;
703
        buf += len;
704
    }
705

    
706
    /* read the sectors "in place" */
707
    nb_sectors = count >> SECTOR_BITS;
708
    if (nb_sectors > 0) {
709
        if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
710
            return -EIO;
711
        sector_num += nb_sectors;
712
        len = nb_sectors << SECTOR_BITS;
713
        buf += len;
714
        count -= len;
715
    }
716

    
717
    /* add data from the last sector */
718
    if (count > 0) {
719
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
720
            return -EIO;
721
        memcpy(buf, tmp_buf, count);
722
    }
723
    return count1;
724
}
725

    
726
int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
727
                const void *buf, int count1)
728
{
729
    uint8_t tmp_buf[SECTOR_SIZE];
730
    int len, nb_sectors, count;
731
    int64_t sector_num;
732

    
733
    count = count1;
734
    /* first write to align to sector start */
735
    len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
736
    if (len > count)
737
        len = count;
738
    sector_num = offset >> SECTOR_BITS;
739
    if (len > 0) {
740
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
741
            return -EIO;
742
        memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
743
        if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
744
            return -EIO;
745
        count -= len;
746
        if (count == 0)
747
            return count1;
748
        sector_num++;
749
        buf += len;
750
    }
751

    
752
    /* write the sectors "in place" */
753
    nb_sectors = count >> SECTOR_BITS;
754
    if (nb_sectors > 0) {
755
        if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
756
            return -EIO;
757
        sector_num += nb_sectors;
758
        len = nb_sectors << SECTOR_BITS;
759
        buf += len;
760
        count -= len;
761
    }
762

    
763
    /* add data from the last sector */
764
    if (count > 0) {
765
        if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
766
            return -EIO;
767
        memcpy(tmp_buf, buf, count);
768
        if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
769
            return -EIO;
770
    }
771
    return count1;
772
}
773

    
774
/**
775
 * Truncate file to 'offset' bytes (needed only for file protocols)
776
 */
777
int bdrv_truncate(BlockDriverState *bs, int64_t offset)
778
{
779
    BlockDriver *drv = bs->drv;
780
    if (!drv)
781
        return -ENOMEDIUM;
782
    if (!drv->bdrv_truncate)
783
        return -ENOTSUP;
784
    if (bs->read_only)
785
        return -EACCES;
786
    return drv->bdrv_truncate(bs, offset);
787
}
788

    
789
/**
790
 * Length of a file in bytes. Return < 0 if error or unknown.
791
 */
792
int64_t bdrv_getlength(BlockDriverState *bs)
793
{
794
    BlockDriver *drv = bs->drv;
795
    if (!drv)
796
        return -ENOMEDIUM;
797
    if (!drv->bdrv_getlength) {
798
        /* legacy mode */
799
        return bs->total_sectors * SECTOR_SIZE;
800
    }
801
    return drv->bdrv_getlength(bs);
802
}
803

    
804
/* return 0 as number of sectors if no device present or error */
805
void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
806
{
807
    int64_t length;
808
    length = bdrv_getlength(bs);
809
    if (length < 0)
810
        length = 0;
811
    else
812
        length = length >> SECTOR_BITS;
813
    *nb_sectors_ptr = length;
814
}
815

    
816
struct partition {
817
        uint8_t boot_ind;           /* 0x80 - active */
818
        uint8_t head;               /* starting head */
819
        uint8_t sector;             /* starting sector */
820
        uint8_t cyl;                /* starting cylinder */
821
        uint8_t sys_ind;            /* What partition type */
822
        uint8_t end_head;           /* end head */
823
        uint8_t end_sector;         /* end sector */
824
        uint8_t end_cyl;            /* end cylinder */
825
        uint32_t start_sect;        /* starting sector counting from 0 */
826
        uint32_t nr_sects;          /* nr of sectors in partition */
827
} __attribute__((packed));
828

    
829
/* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
830
static int guess_disk_lchs(BlockDriverState *bs,
831
                           int *pcylinders, int *pheads, int *psectors)
832
{
833
    uint8_t buf[512];
834
    int ret, i, heads, sectors, cylinders;
835
    struct partition *p;
836
    uint32_t nr_sects;
837
    uint64_t nb_sectors;
838

    
839
    bdrv_get_geometry(bs, &nb_sectors);
840

    
841
    ret = bdrv_read(bs, 0, buf, 1);
842
    if (ret < 0)
843
        return -1;
844
    /* test msdos magic */
845
    if (buf[510] != 0x55 || buf[511] != 0xaa)
846
        return -1;
847
    for(i = 0; i < 4; i++) {
848
        p = ((struct partition *)(buf + 0x1be)) + i;
849
        nr_sects = le32_to_cpu(p->nr_sects);
850
        if (nr_sects && p->end_head) {
851
            /* We make the assumption that the partition terminates on
852
               a cylinder boundary */
853
            heads = p->end_head + 1;
854
            sectors = p->end_sector & 63;
855
            if (sectors == 0)
856
                continue;
857
            cylinders = nb_sectors / (heads * sectors);
858
            if (cylinders < 1 || cylinders > 16383)
859
                continue;
860
            *pheads = heads;
861
            *psectors = sectors;
862
            *pcylinders = cylinders;
863
#if 0
864
            printf("guessed geometry: LCHS=%d %d %d\n",
865
                   cylinders, heads, sectors);
866
#endif
867
            return 0;
868
        }
869
    }
870
    return -1;
871
}
872

    
873
void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
874
{
875
    int translation, lba_detected = 0;
876
    int cylinders, heads, secs;
877
    uint64_t nb_sectors;
878

    
879
    /* if a geometry hint is available, use it */
880
    bdrv_get_geometry(bs, &nb_sectors);
881
    bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
882
    translation = bdrv_get_translation_hint(bs);
883
    if (cylinders != 0) {
884
        *pcyls = cylinders;
885
        *pheads = heads;
886
        *psecs = secs;
887
    } else {
888
        if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
889
            if (heads > 16) {
890
                /* if heads > 16, it means that a BIOS LBA
891
                   translation was active, so the default
892
                   hardware geometry is OK */
893
                lba_detected = 1;
894
                goto default_geometry;
895
            } else {
896
                *pcyls = cylinders;
897
                *pheads = heads;
898
                *psecs = secs;
899
                /* disable any translation to be in sync with
900
                   the logical geometry */
901
                if (translation == BIOS_ATA_TRANSLATION_AUTO) {
902
                    bdrv_set_translation_hint(bs,
903
                                              BIOS_ATA_TRANSLATION_NONE);
904
                }
905
            }
906
        } else {
907
        default_geometry:
908
            /* if no geometry, use a standard physical disk geometry */
909
            cylinders = nb_sectors / (16 * 63);
910

    
911
            if (cylinders > 16383)
912
                cylinders = 16383;
913
            else if (cylinders < 2)
914
                cylinders = 2;
915
            *pcyls = cylinders;
916
            *pheads = 16;
917
            *psecs = 63;
918
            if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
919
                if ((*pcyls * *pheads) <= 131072) {
920
                    bdrv_set_translation_hint(bs,
921
                                              BIOS_ATA_TRANSLATION_LARGE);
922
                } else {
923
                    bdrv_set_translation_hint(bs,
924
                                              BIOS_ATA_TRANSLATION_LBA);
925
                }
926
            }
927
        }
928
        bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
929
    }
930
}
931

    
932
void bdrv_set_geometry_hint(BlockDriverState *bs,
933
                            int cyls, int heads, int secs)
934
{
935
    bs->cyls = cyls;
936
    bs->heads = heads;
937
    bs->secs = secs;
938
}
939

    
940
void bdrv_set_type_hint(BlockDriverState *bs, int type)
941
{
942
    bs->type = type;
943
    bs->removable = ((type == BDRV_TYPE_CDROM ||
944
                      type == BDRV_TYPE_FLOPPY));
945
}
946

    
947
void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
948
{
949
    bs->translation = translation;
950
}
951

    
952
void bdrv_get_geometry_hint(BlockDriverState *bs,
953
                            int *pcyls, int *pheads, int *psecs)
954
{
955
    *pcyls = bs->cyls;
956
    *pheads = bs->heads;
957
    *psecs = bs->secs;
958
}
959

    
960
int bdrv_get_type_hint(BlockDriverState *bs)
961
{
962
    return bs->type;
963
}
964

    
965
int bdrv_get_translation_hint(BlockDriverState *bs)
966
{
967
    return bs->translation;
968
}
969

    
970
int bdrv_is_removable(BlockDriverState *bs)
971
{
972
    return bs->removable;
973
}
974

    
975
int bdrv_is_read_only(BlockDriverState *bs)
976
{
977
    return bs->read_only;
978
}
979

    
980
int bdrv_set_read_only(BlockDriverState *bs, int read_only)
981
{
982
    int ret = bs->read_only;
983
    bs->read_only = read_only;
984
    return ret;
985
}
986

    
987
int bdrv_is_sg(BlockDriverState *bs)
988
{
989
    return bs->sg;
990
}
991

    
992
int bdrv_enable_write_cache(BlockDriverState *bs)
993
{
994
    return bs->enable_write_cache;
995
}
996

    
997
/* XXX: no longer used */
998
void bdrv_set_change_cb(BlockDriverState *bs,
999
                        void (*change_cb)(void *opaque), void *opaque)
1000
{
1001
    bs->change_cb = change_cb;
1002
    bs->change_opaque = opaque;
1003
}
1004

    
1005
int bdrv_is_encrypted(BlockDriverState *bs)
1006
{
1007
    if (bs->backing_hd && bs->backing_hd->encrypted)
1008
        return 1;
1009
    return bs->encrypted;
1010
}
1011

    
1012
int bdrv_key_required(BlockDriverState *bs)
1013
{
1014
    BlockDriverState *backing_hd = bs->backing_hd;
1015

    
1016
    if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1017
        return 1;
1018
    return (bs->encrypted && !bs->valid_key);
1019
}
1020

    
1021
int bdrv_set_key(BlockDriverState *bs, const char *key)
1022
{
1023
    int ret;
1024
    if (bs->backing_hd && bs->backing_hd->encrypted) {
1025
        ret = bdrv_set_key(bs->backing_hd, key);
1026
        if (ret < 0)
1027
            return ret;
1028
        if (!bs->encrypted)
1029
            return 0;
1030
    }
1031
    if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
1032
        return -1;
1033
    ret = bs->drv->bdrv_set_key(bs, key);
1034
    if (ret < 0) {
1035
        bs->valid_key = 0;
1036
    } else if (!bs->valid_key) {
1037
        bs->valid_key = 1;
1038
        /* call the change callback now, we skipped it on open */
1039
        bs->media_changed = 1;
1040
        if (bs->change_cb)
1041
            bs->change_cb(bs->change_opaque);
1042
    }
1043
    return ret;
1044
}
1045

    
1046
void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1047
{
1048
    if (!bs->drv) {
1049
        buf[0] = '\0';
1050
    } else {
1051
        pstrcpy(buf, buf_size, bs->drv->format_name);
1052
    }
1053
}
1054

    
1055
void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1056
                         void *opaque)
1057
{
1058
    BlockDriver *drv;
1059

    
1060
    for (drv = first_drv; drv != NULL; drv = drv->next) {
1061
        it(opaque, drv->format_name);
1062
    }
1063
}
1064

    
1065
BlockDriverState *bdrv_find(const char *name)
1066
{
1067
    BlockDriverState *bs;
1068

    
1069
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1070
        if (!strcmp(name, bs->device_name))
1071
            return bs;
1072
    }
1073
    return NULL;
1074
}
1075

    
1076
void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1077
{
1078
    BlockDriverState *bs;
1079

    
1080
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1081
        it(opaque, bs);
1082
    }
1083
}
1084

    
1085
const char *bdrv_get_device_name(BlockDriverState *bs)
1086
{
1087
    return bs->device_name;
1088
}
1089

    
1090
void bdrv_flush(BlockDriverState *bs)
1091
{
1092
    if (!bs->drv)
1093
        return;
1094
    if (bs->drv->bdrv_flush)
1095
        bs->drv->bdrv_flush(bs);
1096
    if (bs->backing_hd)
1097
        bdrv_flush(bs->backing_hd);
1098
}
1099

    
1100
void bdrv_flush_all(void)
1101
{
1102
    BlockDriverState *bs;
1103

    
1104
    for (bs = bdrv_first; bs != NULL; bs = bs->next)
1105
        if (bs->drv && !bdrv_is_read_only(bs) && 
1106
            (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1107
            bdrv_flush(bs);
1108
}
1109

    
1110
/*
1111
 * Returns true iff the specified sector is present in the disk image. Drivers
1112
 * not implementing the functionality are assumed to not support backing files,
1113
 * hence all their sectors are reported as allocated.
1114
 *
1115
 * 'pnum' is set to the number of sectors (including and immediately following
1116
 * the specified sector) that are known to be in the same
1117
 * allocated/unallocated state.
1118
 *
1119
 * 'nb_sectors' is the max value 'pnum' should be set to.
1120
 */
1121
int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1122
        int *pnum)
1123
{
1124
    int64_t n;
1125
    if (!bs->drv->bdrv_is_allocated) {
1126
        if (sector_num >= bs->total_sectors) {
1127
            *pnum = 0;
1128
            return 0;
1129
        }
1130
        n = bs->total_sectors - sector_num;
1131
        *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1132
        return 1;
1133
    }
1134
    return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1135
}
1136

    
1137
void bdrv_info(Monitor *mon)
1138
{
1139
    BlockDriverState *bs;
1140

    
1141
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1142
        monitor_printf(mon, "%s:", bs->device_name);
1143
        monitor_printf(mon, " type=");
1144
        switch(bs->type) {
1145
        case BDRV_TYPE_HD:
1146
            monitor_printf(mon, "hd");
1147
            break;
1148
        case BDRV_TYPE_CDROM:
1149
            monitor_printf(mon, "cdrom");
1150
            break;
1151
        case BDRV_TYPE_FLOPPY:
1152
            monitor_printf(mon, "floppy");
1153
            break;
1154
        }
1155
        monitor_printf(mon, " removable=%d", bs->removable);
1156
        if (bs->removable) {
1157
            monitor_printf(mon, " locked=%d", bs->locked);
1158
        }
1159
        if (bs->drv) {
1160
            monitor_printf(mon, " file=");
1161
            monitor_print_filename(mon, bs->filename);
1162
            if (bs->backing_file[0] != '\0') {
1163
                monitor_printf(mon, " backing_file=");
1164
                monitor_print_filename(mon, bs->backing_file);
1165
            }
1166
            monitor_printf(mon, " ro=%d", bs->read_only);
1167
            monitor_printf(mon, " drv=%s", bs->drv->format_name);
1168
            monitor_printf(mon, " encrypted=%d", bdrv_is_encrypted(bs));
1169
        } else {
1170
            monitor_printf(mon, " [not inserted]");
1171
        }
1172
        monitor_printf(mon, "\n");
1173
    }
1174
}
1175

    
1176
/* The "info blockstats" command. */
1177
void bdrv_info_stats(Monitor *mon)
1178
{
1179
    BlockDriverState *bs;
1180

    
1181
    for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1182
        monitor_printf(mon, "%s:"
1183
                       " rd_bytes=%" PRIu64
1184
                       " wr_bytes=%" PRIu64
1185
                       " rd_operations=%" PRIu64
1186
                       " wr_operations=%" PRIu64
1187
                       "\n",
1188
                       bs->device_name,
1189
                       bs->rd_bytes, bs->wr_bytes,
1190
                       bs->rd_ops, bs->wr_ops);
1191
    }
1192
}
1193

    
1194
const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1195
{
1196
    if (bs->backing_hd && bs->backing_hd->encrypted)
1197
        return bs->backing_file;
1198
    else if (bs->encrypted)
1199
        return bs->filename;
1200
    else
1201
        return NULL;
1202
}
1203

    
1204
void bdrv_get_backing_filename(BlockDriverState *bs,
1205
                               char *filename, int filename_size)
1206
{
1207
    if (!bs->backing_hd) {
1208
        pstrcpy(filename, filename_size, "");
1209
    } else {
1210
        pstrcpy(filename, filename_size, bs->backing_file);
1211
    }
1212
}
1213

    
1214
int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1215
                          const uint8_t *buf, int nb_sectors)
1216
{
1217
    BlockDriver *drv = bs->drv;
1218
    if (!drv)
1219
        return -ENOMEDIUM;
1220
    if (!drv->bdrv_write_compressed)
1221
        return -ENOTSUP;
1222
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1223
        return -EIO;
1224

    
1225
    if (bs->dirty_tracking) {
1226
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1227
    }
1228

    
1229
    return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1230
}
1231

    
1232
int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1233
{
1234
    BlockDriver *drv = bs->drv;
1235
    if (!drv)
1236
        return -ENOMEDIUM;
1237
    if (!drv->bdrv_get_info)
1238
        return -ENOTSUP;
1239
    memset(bdi, 0, sizeof(*bdi));
1240
    return drv->bdrv_get_info(bs, bdi);
1241
}
1242

    
1243
int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1244
                      int64_t pos, int size)
1245
{
1246
    BlockDriver *drv = bs->drv;
1247
    if (!drv)
1248
        return -ENOMEDIUM;
1249
    if (!drv->bdrv_save_vmstate)
1250
        return -ENOTSUP;
1251
    return drv->bdrv_save_vmstate(bs, buf, pos, size);
1252
}
1253

    
1254
int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1255
                      int64_t pos, int size)
1256
{
1257
    BlockDriver *drv = bs->drv;
1258
    if (!drv)
1259
        return -ENOMEDIUM;
1260
    if (!drv->bdrv_load_vmstate)
1261
        return -ENOTSUP;
1262
    return drv->bdrv_load_vmstate(bs, buf, pos, size);
1263
}
1264

    
1265
/**************************************************************/
1266
/* handling of snapshots */
1267

    
1268
int bdrv_snapshot_create(BlockDriverState *bs,
1269
                         QEMUSnapshotInfo *sn_info)
1270
{
1271
    BlockDriver *drv = bs->drv;
1272
    if (!drv)
1273
        return -ENOMEDIUM;
1274
    if (!drv->bdrv_snapshot_create)
1275
        return -ENOTSUP;
1276
    return drv->bdrv_snapshot_create(bs, sn_info);
1277
}
1278

    
1279
int bdrv_snapshot_goto(BlockDriverState *bs,
1280
                       const char *snapshot_id)
1281
{
1282
    BlockDriver *drv = bs->drv;
1283
    if (!drv)
1284
        return -ENOMEDIUM;
1285
    if (!drv->bdrv_snapshot_goto)
1286
        return -ENOTSUP;
1287
    return drv->bdrv_snapshot_goto(bs, snapshot_id);
1288
}
1289

    
1290
int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1291
{
1292
    BlockDriver *drv = bs->drv;
1293
    if (!drv)
1294
        return -ENOMEDIUM;
1295
    if (!drv->bdrv_snapshot_delete)
1296
        return -ENOTSUP;
1297
    return drv->bdrv_snapshot_delete(bs, snapshot_id);
1298
}
1299

    
1300
int bdrv_snapshot_list(BlockDriverState *bs,
1301
                       QEMUSnapshotInfo **psn_info)
1302
{
1303
    BlockDriver *drv = bs->drv;
1304
    if (!drv)
1305
        return -ENOMEDIUM;
1306
    if (!drv->bdrv_snapshot_list)
1307
        return -ENOTSUP;
1308
    return drv->bdrv_snapshot_list(bs, psn_info);
1309
}
1310

    
1311
#define NB_SUFFIXES 4
1312

    
1313
char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1314
{
1315
    static const char suffixes[NB_SUFFIXES] = "KMGT";
1316
    int64_t base;
1317
    int i;
1318

    
1319
    if (size <= 999) {
1320
        snprintf(buf, buf_size, "%" PRId64, size);
1321
    } else {
1322
        base = 1024;
1323
        for(i = 0; i < NB_SUFFIXES; i++) {
1324
            if (size < (10 * base)) {
1325
                snprintf(buf, buf_size, "%0.1f%c",
1326
                         (double)size / base,
1327
                         suffixes[i]);
1328
                break;
1329
            } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1330
                snprintf(buf, buf_size, "%" PRId64 "%c",
1331
                         ((size + (base >> 1)) / base),
1332
                         suffixes[i]);
1333
                break;
1334
            }
1335
            base = base * 1024;
1336
        }
1337
    }
1338
    return buf;
1339
}
1340

    
1341
char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1342
{
1343
    char buf1[128], date_buf[128], clock_buf[128];
1344
#ifdef _WIN32
1345
    struct tm *ptm;
1346
#else
1347
    struct tm tm;
1348
#endif
1349
    time_t ti;
1350
    int64_t secs;
1351

    
1352
    if (!sn) {
1353
        snprintf(buf, buf_size,
1354
                 "%-10s%-20s%7s%20s%15s",
1355
                 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1356
    } else {
1357
        ti = sn->date_sec;
1358
#ifdef _WIN32
1359
        ptm = localtime(&ti);
1360
        strftime(date_buf, sizeof(date_buf),
1361
                 "%Y-%m-%d %H:%M:%S", ptm);
1362
#else
1363
        localtime_r(&ti, &tm);
1364
        strftime(date_buf, sizeof(date_buf),
1365
                 "%Y-%m-%d %H:%M:%S", &tm);
1366
#endif
1367
        secs = sn->vm_clock_nsec / 1000000000;
1368
        snprintf(clock_buf, sizeof(clock_buf),
1369
                 "%02d:%02d:%02d.%03d",
1370
                 (int)(secs / 3600),
1371
                 (int)((secs / 60) % 60),
1372
                 (int)(secs % 60),
1373
                 (int)((sn->vm_clock_nsec / 1000000) % 1000));
1374
        snprintf(buf, buf_size,
1375
                 "%-10s%-20s%7s%20s%15s",
1376
                 sn->id_str, sn->name,
1377
                 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1378
                 date_buf,
1379
                 clock_buf);
1380
    }
1381
    return buf;
1382
}
1383

    
1384

    
1385
/**************************************************************/
1386
/* async I/Os */
1387

    
1388
BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1389
                                 QEMUIOVector *qiov, int nb_sectors,
1390
                                 BlockDriverCompletionFunc *cb, void *opaque)
1391
{
1392
    BlockDriver *drv = bs->drv;
1393
    BlockDriverAIOCB *ret;
1394

    
1395
    if (!drv)
1396
        return NULL;
1397
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1398
        return NULL;
1399

    
1400
    ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1401
                              cb, opaque);
1402

    
1403
    if (ret) {
1404
        /* Update stats even though technically transfer has not happened. */
1405
        bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1406
        bs->rd_ops ++;
1407
    }
1408

    
1409
    return ret;
1410
}
1411

    
1412
BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1413
                                  QEMUIOVector *qiov, int nb_sectors,
1414
                                  BlockDriverCompletionFunc *cb, void *opaque)
1415
{
1416
    BlockDriver *drv = bs->drv;
1417
    BlockDriverAIOCB *ret;
1418

    
1419
    if (!drv)
1420
        return NULL;
1421
    if (bs->read_only)
1422
        return NULL;
1423
    if (bdrv_check_request(bs, sector_num, nb_sectors))
1424
        return NULL;
1425

    
1426
    if (bs->dirty_tracking) {
1427
        set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1428
    }
1429

    
1430
    ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
1431
                               cb, opaque);
1432

    
1433
    if (ret) {
1434
        /* Update stats even though technically transfer has not happened. */
1435
        bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1436
        bs->wr_ops ++;
1437
    }
1438

    
1439
    return ret;
1440
}
1441

    
1442

    
1443
typedef struct MultiwriteCB {
1444
    int error;
1445
    int num_requests;
1446
    int num_callbacks;
1447
    struct {
1448
        BlockDriverCompletionFunc *cb;
1449
        void *opaque;
1450
        QEMUIOVector *free_qiov;
1451
        void *free_buf;
1452
    } callbacks[];
1453
} MultiwriteCB;
1454

    
1455
static void multiwrite_user_cb(MultiwriteCB *mcb)
1456
{
1457
    int i;
1458

    
1459
    for (i = 0; i < mcb->num_callbacks; i++) {
1460
        mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
1461
        qemu_free(mcb->callbacks[i].free_qiov);
1462
        qemu_free(mcb->callbacks[i].free_buf);
1463
    }
1464
}
1465

    
1466
static void multiwrite_cb(void *opaque, int ret)
1467
{
1468
    MultiwriteCB *mcb = opaque;
1469

    
1470
    if (ret < 0) {
1471
        mcb->error = ret;
1472
        multiwrite_user_cb(mcb);
1473
    }
1474

    
1475
    mcb->num_requests--;
1476
    if (mcb->num_requests == 0) {
1477
        if (mcb->error == 0) {
1478
            multiwrite_user_cb(mcb);
1479
        }
1480
        qemu_free(mcb);
1481
    }
1482
}
1483

    
1484
static int multiwrite_req_compare(const void *a, const void *b)
1485
{
1486
    return (((BlockRequest*) a)->sector - ((BlockRequest*) b)->sector);
1487
}
1488

    
1489
/*
1490
 * Takes a bunch of requests and tries to merge them. Returns the number of
1491
 * requests that remain after merging.
1492
 */
1493
static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
1494
    int num_reqs, MultiwriteCB *mcb)
1495
{
1496
    int i, outidx;
1497

    
1498
    // Sort requests by start sector
1499
    qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
1500

    
1501
    // Check if adjacent requests touch the same clusters. If so, combine them,
1502
    // filling up gaps with zero sectors.
1503
    outidx = 0;
1504
    for (i = 1; i < num_reqs; i++) {
1505
        int merge = 0;
1506
        int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
1507

    
1508
        // This handles the cases that are valid for all block drivers, namely
1509
        // exactly sequential writes and overlapping writes.
1510
        if (reqs[i].sector <= oldreq_last) {
1511
            merge = 1;
1512
        }
1513

    
1514
        // The block driver may decide that it makes sense to combine requests
1515
        // even if there is a gap of some sectors between them. In this case,
1516
        // the gap is filled with zeros (therefore only applicable for yet
1517
        // unused space in format like qcow2).
1518
        if (!merge && bs->drv->bdrv_merge_requests) {
1519
            merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
1520
        }
1521

    
1522
        if (merge) {
1523
            size_t size;
1524
            QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
1525
            qemu_iovec_init(qiov,
1526
                reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
1527

    
1528
            // Add the first request to the merged one. If the requests are
1529
            // overlapping, drop the last sectors of the first request.
1530
            size = (reqs[i].sector - reqs[outidx].sector) << 9;
1531
            qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
1532

    
1533
            // We might need to add some zeros between the two requests
1534
            if (reqs[i].sector > oldreq_last) {
1535
                size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
1536
                uint8_t *buf = qemu_blockalign(bs, zero_bytes);
1537
                memset(buf, 0, zero_bytes);
1538
                qemu_iovec_add(qiov, buf, zero_bytes);
1539
                mcb->callbacks[i].free_buf = buf;
1540
            }
1541

    
1542
            // Add the second request
1543
            qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
1544

    
1545
            reqs[outidx].nb_sectors += reqs[i].nb_sectors;
1546
            reqs[outidx].qiov = qiov;
1547

    
1548
            mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
1549
        } else {
1550
            outidx++;
1551
            reqs[outidx].sector     = reqs[i].sector;
1552
            reqs[outidx].nb_sectors = reqs[i].nb_sectors;
1553
            reqs[outidx].qiov       = reqs[i].qiov;
1554
        }
1555
    }
1556

    
1557
    return outidx + 1;
1558
}
1559

    
1560
/*
1561
 * Submit multiple AIO write requests at once.
1562
 *
1563
 * On success, the function returns 0 and all requests in the reqs array have
1564
 * been submitted. In error case this function returns -1, and any of the
1565
 * requests may or may not be submitted yet. In particular, this means that the
1566
 * callback will be called for some of the requests, for others it won't. The
1567
 * caller must check the error field of the BlockRequest to wait for the right
1568
 * callbacks (if error != 0, no callback will be called).
1569
 *
1570
 * The implementation may modify the contents of the reqs array, e.g. to merge
1571
 * requests. However, the fields opaque and error are left unmodified as they
1572
 * are used to signal failure for a single request to the caller.
1573
 */
1574
int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
1575
{
1576
    BlockDriverAIOCB *acb;
1577
    MultiwriteCB *mcb;
1578
    int i;
1579

    
1580
    if (num_reqs == 0) {
1581
        return 0;
1582
    }
1583

    
1584
    // Create MultiwriteCB structure
1585
    mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
1586
    mcb->num_requests = 0;
1587
    mcb->num_callbacks = num_reqs;
1588

    
1589
    for (i = 0; i < num_reqs; i++) {
1590
        mcb->callbacks[i].cb = reqs[i].cb;
1591
        mcb->callbacks[i].opaque = reqs[i].opaque;
1592
    }
1593

    
1594
    // Check for mergable requests
1595
    num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
1596

    
1597
    // Run the aio requests
1598
    for (i = 0; i < num_reqs; i++) {
1599
        acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
1600
            reqs[i].nb_sectors, multiwrite_cb, mcb);
1601

    
1602
        if (acb == NULL) {
1603
            // We can only fail the whole thing if no request has been
1604
            // submitted yet. Otherwise we'll wait for the submitted AIOs to
1605
            // complete and report the error in the callback.
1606
            if (mcb->num_requests == 0) {
1607
                reqs[i].error = EIO;
1608
                goto fail;
1609
            } else {
1610
                mcb->error = EIO;
1611
                break;
1612
            }
1613
        } else {
1614
            mcb->num_requests++;
1615
        }
1616
    }
1617

    
1618
    return 0;
1619

    
1620
fail:
1621
    free(mcb);
1622
    return -1;
1623
}
1624

    
1625
BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
1626
        BlockDriverCompletionFunc *cb, void *opaque)
1627
{
1628
    BlockDriver *drv = bs->drv;
1629

    
1630
    if (!drv)
1631
        return NULL;
1632

    
1633
    /*
1634
     * Note that unlike bdrv_flush the driver is reponsible for flushing a
1635
     * backing image if it exists.
1636
     */
1637
    return drv->bdrv_aio_flush(bs, cb, opaque);
1638
}
1639

    
1640
void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1641
{
1642
    acb->pool->cancel(acb);
1643
}
1644

    
1645

    
1646
/**************************************************************/
1647
/* async block device emulation */
1648

    
1649
typedef struct BlockDriverAIOCBSync {
1650
    BlockDriverAIOCB common;
1651
    QEMUBH *bh;
1652
    int ret;
1653
    /* vector translation state */
1654
    QEMUIOVector *qiov;
1655
    uint8_t *bounce;
1656
    int is_write;
1657
} BlockDriverAIOCBSync;
1658

    
1659
static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1660
{
1661
    BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1662
    qemu_bh_delete(acb->bh);
1663
    acb->bh = NULL;
1664
    qemu_aio_release(acb);
1665
}
1666

    
1667
static AIOPool bdrv_em_aio_pool = {
1668
    .aiocb_size         = sizeof(BlockDriverAIOCBSync),
1669
    .cancel             = bdrv_aio_cancel_em,
1670
};
1671

    
1672
static void bdrv_aio_bh_cb(void *opaque)
1673
{
1674
    BlockDriverAIOCBSync *acb = opaque;
1675

    
1676
    if (!acb->is_write)
1677
        qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
1678
    qemu_vfree(acb->bounce);
1679
    acb->common.cb(acb->common.opaque, acb->ret);
1680
    qemu_bh_delete(acb->bh);
1681
    acb->bh = NULL;
1682
    qemu_aio_release(acb);
1683
}
1684

    
1685
static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
1686
                                            int64_t sector_num,
1687
                                            QEMUIOVector *qiov,
1688
                                            int nb_sectors,
1689
                                            BlockDriverCompletionFunc *cb,
1690
                                            void *opaque,
1691
                                            int is_write)
1692

    
1693
{
1694
    BlockDriverAIOCBSync *acb;
1695

    
1696
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
1697
    acb->is_write = is_write;
1698
    acb->qiov = qiov;
1699
    acb->bounce = qemu_blockalign(bs, qiov->size);
1700

    
1701
    if (!acb->bh)
1702
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1703

    
1704
    if (is_write) {
1705
        qemu_iovec_to_buffer(acb->qiov, acb->bounce);
1706
        acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
1707
    } else {
1708
        acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
1709
    }
1710

    
1711
    qemu_bh_schedule(acb->bh);
1712

    
1713
    return &acb->common;
1714
}
1715

    
1716
static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
1717
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1718
        BlockDriverCompletionFunc *cb, void *opaque)
1719
{
1720
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
1721
}
1722

    
1723
static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
1724
        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1725
        BlockDriverCompletionFunc *cb, void *opaque)
1726
{
1727
    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
1728
}
1729

    
1730
static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
1731
        BlockDriverCompletionFunc *cb, void *opaque)
1732
{
1733
    BlockDriverAIOCBSync *acb;
1734

    
1735
    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
1736
    acb->is_write = 1; /* don't bounce in the completion hadler */
1737
    acb->qiov = NULL;
1738
    acb->bounce = NULL;
1739
    acb->ret = 0;
1740

    
1741
    if (!acb->bh)
1742
        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1743

    
1744
    bdrv_flush(bs);
1745
    qemu_bh_schedule(acb->bh);
1746
    return &acb->common;
1747
}
1748

    
1749
/**************************************************************/
1750
/* sync block device emulation */
1751

    
1752
static void bdrv_rw_em_cb(void *opaque, int ret)
1753
{
1754
    *(int *)opaque = ret;
1755
}
1756

    
1757
#define NOT_DONE 0x7fffffff
1758

    
1759
static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1760
                        uint8_t *buf, int nb_sectors)
1761
{
1762
    int async_ret;
1763
    BlockDriverAIOCB *acb;
1764
    struct iovec iov;
1765
    QEMUIOVector qiov;
1766

    
1767
    async_context_push();
1768

    
1769
    async_ret = NOT_DONE;
1770
    iov.iov_base = (void *)buf;
1771
    iov.iov_len = nb_sectors * 512;
1772
    qemu_iovec_init_external(&qiov, &iov, 1);
1773
    acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
1774
        bdrv_rw_em_cb, &async_ret);
1775
    if (acb == NULL) {
1776
        async_ret = -1;
1777
        goto fail;
1778
    }
1779

    
1780
    while (async_ret == NOT_DONE) {
1781
        qemu_aio_wait();
1782
    }
1783

    
1784

    
1785
fail:
1786
    async_context_pop();
1787
    return async_ret;
1788
}
1789

    
1790
static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1791
                         const uint8_t *buf, int nb_sectors)
1792
{
1793
    int async_ret;
1794
    BlockDriverAIOCB *acb;
1795
    struct iovec iov;
1796
    QEMUIOVector qiov;
1797

    
1798
    async_context_push();
1799

    
1800
    async_ret = NOT_DONE;
1801
    iov.iov_base = (void *)buf;
1802
    iov.iov_len = nb_sectors * 512;
1803
    qemu_iovec_init_external(&qiov, &iov, 1);
1804
    acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
1805
        bdrv_rw_em_cb, &async_ret);
1806
    if (acb == NULL) {
1807
        async_ret = -1;
1808
        goto fail;
1809
    }
1810
    while (async_ret == NOT_DONE) {
1811
        qemu_aio_wait();
1812
    }
1813

    
1814
fail:
1815
    async_context_pop();
1816
    return async_ret;
1817
}
1818

    
1819
void bdrv_init(void)
1820
{
1821
    module_call_init(MODULE_INIT_BLOCK);
1822
}
1823

    
1824
void bdrv_init_with_whitelist(void)
1825
{
1826
    use_bdrv_whitelist = 1;
1827
    bdrv_init();
1828
}
1829

    
1830
void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
1831
                   BlockDriverCompletionFunc *cb, void *opaque)
1832
{
1833
    BlockDriverAIOCB *acb;
1834

    
1835
    if (pool->free_aiocb) {
1836
        acb = pool->free_aiocb;
1837
        pool->free_aiocb = acb->next;
1838
    } else {
1839
        acb = qemu_mallocz(pool->aiocb_size);
1840
        acb->pool = pool;
1841
    }
1842
    acb->bs = bs;
1843
    acb->cb = cb;
1844
    acb->opaque = opaque;
1845
    return acb;
1846
}
1847

    
1848
void qemu_aio_release(void *p)
1849
{
1850
    BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
1851
    AIOPool *pool = acb->pool;
1852
    acb->next = pool->free_aiocb;
1853
    pool->free_aiocb = acb;
1854
}
1855

    
1856
/**************************************************************/
1857
/* removable device support */
1858

    
1859
/**
1860
 * Return TRUE if the media is present
1861
 */
1862
int bdrv_is_inserted(BlockDriverState *bs)
1863
{
1864
    BlockDriver *drv = bs->drv;
1865
    int ret;
1866
    if (!drv)
1867
        return 0;
1868
    if (!drv->bdrv_is_inserted)
1869
        return 1;
1870
    ret = drv->bdrv_is_inserted(bs);
1871
    return ret;
1872
}
1873

    
1874
/**
1875
 * Return TRUE if the media changed since the last call to this
1876
 * function. It is currently only used for floppy disks
1877
 */
1878
int bdrv_media_changed(BlockDriverState *bs)
1879
{
1880
    BlockDriver *drv = bs->drv;
1881
    int ret;
1882

    
1883
    if (!drv || !drv->bdrv_media_changed)
1884
        ret = -ENOTSUP;
1885
    else
1886
        ret = drv->bdrv_media_changed(bs);
1887
    if (ret == -ENOTSUP)
1888
        ret = bs->media_changed;
1889
    bs->media_changed = 0;
1890
    return ret;
1891
}
1892

    
1893
/**
1894
 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1895
 */
1896
int bdrv_eject(BlockDriverState *bs, int eject_flag)
1897
{
1898
    BlockDriver *drv = bs->drv;
1899
    int ret;
1900

    
1901
    if (bs->locked) {
1902
        return -EBUSY;
1903
    }
1904

    
1905
    if (!drv || !drv->bdrv_eject) {
1906
        ret = -ENOTSUP;
1907
    } else {
1908
        ret = drv->bdrv_eject(bs, eject_flag);
1909
    }
1910
    if (ret == -ENOTSUP) {
1911
        if (eject_flag)
1912
            bdrv_close(bs);
1913
        ret = 0;
1914
    }
1915

    
1916
    return ret;
1917
}
1918

    
1919
int bdrv_is_locked(BlockDriverState *bs)
1920
{
1921
    return bs->locked;
1922
}
1923

    
1924
/**
1925
 * Lock or unlock the media (if it is locked, the user won't be able
1926
 * to eject it manually).
1927
 */
1928
void bdrv_set_locked(BlockDriverState *bs, int locked)
1929
{
1930
    BlockDriver *drv = bs->drv;
1931

    
1932
    bs->locked = locked;
1933
    if (drv && drv->bdrv_set_locked) {
1934
        drv->bdrv_set_locked(bs, locked);
1935
    }
1936
}
1937

    
1938
/* needed for generic scsi interface */
1939

    
1940
int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1941
{
1942
    BlockDriver *drv = bs->drv;
1943

    
1944
    if (drv && drv->bdrv_ioctl)
1945
        return drv->bdrv_ioctl(bs, req, buf);
1946
    return -ENOTSUP;
1947
}
1948

    
1949
BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
1950
        unsigned long int req, void *buf,
1951
        BlockDriverCompletionFunc *cb, void *opaque)
1952
{
1953
    BlockDriver *drv = bs->drv;
1954

    
1955
    if (drv && drv->bdrv_aio_ioctl)
1956
        return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
1957
    return NULL;
1958
}
1959

    
1960

    
1961

    
1962
void *qemu_blockalign(BlockDriverState *bs, size_t size)
1963
{
1964
    return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
1965
}
1966

    
1967
void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
1968
{
1969
    int64_t bitmap_size;
1970

    
1971
    if (enable) {
1972
        if (bs->dirty_tracking == 0) {
1973
            int64_t i;
1974
            uint8_t test;
1975

    
1976
            bitmap_size = (bdrv_getlength(bs) >> SECTOR_BITS);
1977
            bitmap_size /= SECTORS_PER_DIRTY_CHUNK;
1978
            bitmap_size++;
1979

    
1980
            bs->dirty_bitmap = qemu_mallocz(bitmap_size);
1981

    
1982
            bs->dirty_tracking = enable;
1983
            for(i = 0; i < bitmap_size; i++) test = bs->dirty_bitmap[i]; 
1984
        }
1985
    } else {
1986
        if (bs->dirty_tracking != 0) {
1987
            qemu_free(bs->dirty_bitmap);
1988
            bs->dirty_tracking = enable;
1989
        }
1990
    }
1991
}
1992

    
1993
int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
1994
{
1995
    int64_t chunk = sector / (int64_t)SECTORS_PER_DIRTY_CHUNK;
1996

    
1997
    if (bs->dirty_bitmap != NULL &&
1998
        (sector << SECTOR_BITS) <= bdrv_getlength(bs)) {
1999
        return bs->dirty_bitmap[chunk];
2000
    } else {
2001
        return 0;
2002
    }
2003
}
2004

    
2005
void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2006
                      int nr_sectors)
2007
{
2008
    set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2009
}
2010

    
2011
int bdrv_get_sectors_per_chunk(void)
2012
{
2013
    /* size must be 2^x */
2014
    return SECTORS_PER_DIRTY_CHUNK;
2015
}