block.c

   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 "qemu-common.h"
  25 #include "console.h"
  26 #include "block_int.h"
  27 #include "osdep.h"
  28
  29 #ifdef _BSD
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <sys/ioctl.h>
  33 #include <sys/queue.h>
  34 #include <sys/disk.h>
  35 #endif
  36
  37 #define SECTOR_BITS 9
  38 #define SECTOR_SIZE (1 << SECTOR_BITS)
  39
  40 typedef struct BlockDriverAIOCBSync {
  41     BlockDriverAIOCB common;
  42     QEMUBH *bh;
  43     int ret;
  44 } BlockDriverAIOCBSync;
  45
  46 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
  47         int64_t sector_num, uint8_t *buf, int nb_sectors,
  48         BlockDriverCompletionFunc *cb, void *opaque);
  49 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
  50         int64_t sector_num, const uint8_t *buf, int nb_sectors,
  51         BlockDriverCompletionFunc *cb, void *opaque);
  52 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
  53 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
  54                         uint8_t *buf, int nb_sectors);
  55 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
  56                          const uint8_t *buf, int nb_sectors);
  57
  58 BlockDriverState *bdrv_first;
  59
  60 static BlockDriver *first_drv;
  61
  62 int path_is_absolute(const char *path)
  63 {
  64     const char *p;
  65 #ifdef _WIN32
  66     /* specific case for names like: "\\.\d:" */
  67     if (*path == '/' || *path == '\\')
  68         return 1;
  69 #endif
  70     p = strchr(path, ':');
  71     if (p)
  72         p++;
  73     else
  74         p = path;
  75 #ifdef _WIN32
  76     return (*p == '/' || *p == '\\');
  77 #else
  78     return (*p == '/');
  79 #endif
  80 }
  81
  82 /* if filename is absolute, just copy it to dest. Otherwise, build a
  83    path to it by considering it is relative to base_path. URL are
  84    supported. */
  85 void path_combine(char *dest, int dest_size,
  86                   const char *base_path,
  87                   const char *filename)
  88 {
  89     const char *p, *p1;
  90     int len;
  91
  92     if (dest_size <= 0)
  93         return;
  94     if (path_is_absolute(filename)) {
  95         pstrcpy(dest, dest_size, filename);
  96     } else {
  97         p = strchr(base_path, ':');
  98         if (p)
  99             p++;
 100         else
 101             p = base_path;
 102         p1 = strrchr(base_path, '/');
 103 #ifdef _WIN32
 104         {
 105             const char *p2;
 106             p2 = strrchr(base_path, '\\');
 107             if (!p1 || p2 > p1)
 108                 p1 = p2;
 109         }
 110 #endif
 111         if (p1)
 112             p1++;
 113         else
 114             p1 = base_path;
 115         if (p1 > p)
 116             p = p1;
 117         len = p - base_path;
 118         if (len > dest_size - 1)
 119             len = dest_size - 1;
 120         memcpy(dest, base_path, len);
 121         dest[len] = '\0';
 122         pstrcat(dest, dest_size, filename);
 123     }
 124 }
 125
 126
 127 static void bdrv_register(BlockDriver *bdrv)
 128 {
 129     if (!bdrv->bdrv_aio_read) {
 130         /* add AIO emulation layer */
 131         bdrv->bdrv_aio_read = bdrv_aio_read_em;
 132         bdrv->bdrv_aio_write = bdrv_aio_write_em;
 133         bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
 134         bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
 135     } else if (!bdrv->bdrv_read && !bdrv->bdrv_pread) {
 136         /* add synchronous IO emulation layer */
 137         bdrv->bdrv_read = bdrv_read_em;
 138         bdrv->bdrv_write = bdrv_write_em;
 139     }
 140     bdrv->next = first_drv;
 141     first_drv = bdrv;
 142 }
 143
 144 /* create a new block device (by default it is empty) */
 145 BlockDriverState *bdrv_new(const char *device_name)
 146 {
 147     BlockDriverState **pbs, *bs;
 148
 149     bs = qemu_mallocz(sizeof(BlockDriverState));
 150     if(!bs)
 151         return NULL;
 152     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
 153     if (device_name[0] != '\0') {
 154         /* insert at the end */
 155         pbs = &bdrv_first;
 156         while (*pbs != NULL)
 157             pbs = &(*pbs)->next;
 158         *pbs = bs;
 159     }
 160     return bs;
 161 }
 162
 163 BlockDriver *bdrv_find_format(const char *format_name)
 164 {
 165     BlockDriver *drv1;
 166     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 167         if (!strcmp(drv1->format_name, format_name))
 168             return drv1;
 169     }
 170     return NULL;
 171 }
 172
 173 int bdrv_create(BlockDriver *drv,
 174                 const char *filename, int64_t size_in_sectors,
 175                 const char *backing_file, int flags)
 176 {
 177     if (!drv->bdrv_create)
 178         return -ENOTSUP;
 179     return drv->bdrv_create(filename, size_in_sectors, backing_file, flags);
 180 }
 181
 182 #ifdef _WIN32
 183 void get_tmp_filename(char *filename, int size)
 184 {
 185     char temp_dir[MAX_PATH];
 186
 187     GetTempPath(MAX_PATH, temp_dir);
 188     GetTempFileName(temp_dir, "qem", 0, filename);
 189 }
 190 #else
 191 void get_tmp_filename(char *filename, int size)
 192 {
 193     int fd;
 194     const char *tmpdir;
 195     /* XXX: race condition possible */
 196     tmpdir = getenv("TMPDIR");
 197     if (!tmpdir)
 198         tmpdir = "/tmp";
 199     snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
 200     fd = mkstemp(filename);
 201     close(fd);
 202 }
 203 #endif
 204
 205 #ifdef _WIN32
 206 static int is_windows_drive_prefix(const char *filename)
 207 {
 208     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
 209              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
 210             filename[1] == ':');
 211 }
 212
 213 static int is_windows_drive(const char *filename)
 214 {
 215     if (is_windows_drive_prefix(filename) &&
 216         filename[2] == '\0')
 217         return 1;
 218     if (strstart(filename, "\\\\.\\", NULL) ||
 219         strstart(filename, "//./", NULL))
 220         return 1;
 221     return 0;
 222 }
 223 #endif
 224
 225 static BlockDriver *find_protocol(const char *filename)
 226 {
 227     BlockDriver *drv1;
 228     char protocol[128];
 229     int len;
 230     const char *p;
 231
 232 #ifdef _WIN32
 233     if (is_windows_drive(filename) ||
 234         is_windows_drive_prefix(filename))
 235         return &bdrv_raw;
 236 #endif
 237     p = strchr(filename, ':');
 238     if (!p)
 239         return &bdrv_raw;
 240     len = p - filename;
 241     if (len > sizeof(protocol) - 1)
 242         len = sizeof(protocol) - 1;
 243     memcpy(protocol, filename, len);
 244     protocol[len] = '\0';
 245     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 246         if (drv1->protocol_name &&
 247             !strcmp(drv1->protocol_name, protocol))
 248             return drv1;
 249     }
 250     return NULL;
 251 }
 252
 253 /* XXX: force raw format if block or character device ? It would
 254    simplify the BSD case */
 255 static BlockDriver *find_image_format(const char *filename)
 256 {
 257     int ret, score, score_max;
 258     BlockDriver *drv1, *drv;
 259     uint8_t buf[2048];
 260     BlockDriverState *bs;
 261
 262     /* detect host devices. By convention, /dev/cdrom[N] is always
 263        recognized as a host CDROM */
 264     if (strstart(filename, "/dev/cdrom", NULL))
 265         return &bdrv_host_device;
 266 #ifdef _WIN32
 267     if (is_windows_drive(filename))
 268         return &bdrv_host_device;
 269 #else
 270     {
 271         struct stat st;
 272         if (stat(filename, &st) >= 0 &&
 273             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 274             return &bdrv_host_device;
 275         }
 276     }
 277 #endif
 278
 279     drv = find_protocol(filename);
 280     /* no need to test disk image formats for vvfat */
 281     if (drv == &bdrv_vvfat)
 282         return drv;
 283
 284     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
 285     if (ret < 0)
 286         return NULL;
 287     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
 288     bdrv_delete(bs);
 289     if (ret < 0) {
 290         return NULL;
 291     }
 292
 293     score_max = 0;
 294     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 295         if (drv1->bdrv_probe) {
 296             score = drv1->bdrv_probe(buf, ret, filename);
 297             if (score > score_max) {
 298                 score_max = score;
 299                 drv = drv1;
 300             }
 301         }
 302     }
 303     return drv;
 304 }
 305
 306 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
 307 {
 308     BlockDriverState *bs;
 309     int ret;
 310
 311     bs = bdrv_new("");
 312     if (!bs)
 313         return -ENOMEM;
 314     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
 315     if (ret < 0) {
 316         bdrv_delete(bs);
 317         return ret;
 318     }
 319     *pbs = bs;
 320     return 0;
 321 }
 322
 323 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
 324 {
 325     return bdrv_open2(bs, filename, flags, NULL);
 326 }
 327
 328 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
 329                BlockDriver *drv)
 330 {
 331     int ret, open_flags;
 332     char tmp_filename[PATH_MAX];
 333     char backing_filename[PATH_MAX];
 334
 335     bs->read_only = 0;
 336     bs->is_temporary = 0;
 337     bs->encrypted = 0;
 338
 339     if (flags & BDRV_O_SNAPSHOT) {
 340         BlockDriverState *bs1;
 341         int64_t total_size;
 342         int is_protocol = 0;
 343
 344         /* if snapshot, we create a temporary backing file and open it
 345            instead of opening 'filename' directly */
 346
 347         /* if there is a backing file, use it */
 348         bs1 = bdrv_new("");
 349         if (!bs1) {
 350             return -ENOMEM;
 351         }
 352         if (bdrv_open(bs1, filename, 0) < 0) {
 353             bdrv_delete(bs1);
 354             return -1;
 355         }
 356         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
 357
 358         if (bs1->drv && bs1->drv->protocol_name)
 359             is_protocol = 1;
 360
 361         bdrv_delete(bs1);
 362
 363         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
 364
 365         /* Real path is meaningless for protocols */
 366         if (is_protocol)
 367             snprintf(backing_filename, sizeof(backing_filename),
 368                      "%s", filename);
 369         else
 370             realpath(filename, backing_filename);
 371
 372         if (bdrv_create(&bdrv_qcow2, tmp_filename,
 373                         total_size, backing_filename, 0) < 0) {
 374             return -1;
 375         }
 376         filename = tmp_filename;
 377         bs->is_temporary = 1;
 378     }
 379
 380     pstrcpy(bs->filename, sizeof(bs->filename), filename);
 381     if (flags & BDRV_O_FILE) {
 382         drv = find_protocol(filename);
 383         if (!drv)
 384             return -ENOENT;
 385     } else {
 386         if (!drv) {
 387             drv = find_image_format(filename);
 388             if (!drv)
 389                 return -1;
 390         }
 391     }
 392     bs->drv = drv;
 393     bs->opaque = qemu_mallocz(drv->instance_size);
 394     if (bs->opaque == NULL && drv->instance_size > 0)
 395         return -1;
 396     /* Note: for compatibility, we open disk image files as RDWR, and
 397        RDONLY as fallback */
 398     if (!(flags & BDRV_O_FILE))
 399         open_flags = BDRV_O_RDWR | (flags & BDRV_O_CACHE_MASK);
 400     else
 401         open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
 402     ret = drv->bdrv_open(bs, filename, open_flags);
 403     if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
 404         ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
 405         bs->read_only = 1;
 406     }
 407     if (ret < 0) {
 408         qemu_free(bs->opaque);
 409         bs->opaque = NULL;
 410         bs->drv = NULL;
 411         return ret;
 412     }
 413     if (drv->bdrv_getlength) {
 414         bs->total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 415     }
 416 #ifndef _WIN32
 417     if (bs->is_temporary) {
 418         unlink(filename);
 419     }
 420 #endif
 421     if (bs->backing_file[0] != '\0') {
 422         /* if there is a backing file, use it */
 423         bs->backing_hd = bdrv_new("");
 424         if (!bs->backing_hd) {
 425         fail:
 426             bdrv_close(bs);
 427             return -ENOMEM;
 428         }
 429         path_combine(backing_filename, sizeof(backing_filename),
 430                      filename, bs->backing_file);
 431         if (bdrv_open(bs->backing_hd, backing_filename, open_flags) < 0)
 432             goto fail;
 433     }
 434
 435     /* call the change callback */
 436     bs->media_changed = 1;
 437     if (bs->change_cb)
 438         bs->change_cb(bs->change_opaque);
 439
 440     return 0;
 441 }
 442
 443 void bdrv_close(BlockDriverState *bs)
 444 {
 445     if (bs->drv) {
 446         if (bs->backing_hd)
 447             bdrv_delete(bs->backing_hd);
 448         bs->drv->bdrv_close(bs);
 449         qemu_free(bs->opaque);
 450 #ifdef _WIN32
 451         if (bs->is_temporary) {
 452             unlink(bs->filename);
 453         }
 454 #endif
 455         bs->opaque = NULL;
 456         bs->drv = NULL;
 457
 458         /* call the change callback */
 459         bs->media_changed = 1;
 460         if (bs->change_cb)
 461             bs->change_cb(bs->change_opaque);
 462     }
 463 }
 464
 465 void bdrv_delete(BlockDriverState *bs)
 466 {
 467     BlockDriverState **pbs;
 468
 469     pbs = &bdrv_first;
 470     while (*pbs != bs && *pbs != NULL)
 471         pbs = &(*pbs)->next;
 472     if (*pbs == bs)
 473         *pbs = bs->next;
 474
 475     bdrv_close(bs);
 476     qemu_free(bs);
 477 }
 478
 479 /* commit COW file into the raw image */
 480 int bdrv_commit(BlockDriverState *bs)
 481 {
 482     BlockDriver *drv = bs->drv;
 483     int64_t i, total_sectors;
 484     int n, j;
 485     unsigned char sector[512];
 486
 487     if (!drv)
 488         return -ENOMEDIUM;
 489
 490     if (bs->read_only) {
 491         return -EACCES;
 492     }
 493
 494     if (!bs->backing_hd) {
 495         return -ENOTSUP;
 496     }
 497
 498     total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 499     for (i = 0; i < total_sectors;) {
 500         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
 501             for(j = 0; j < n; j++) {
 502                 if (bdrv_read(bs, i, sector, 1) != 0) {
 503                     return -EIO;
 504                 }
 505
 506                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
 507                     return -EIO;
 508                 }
 509                 i++;
 510             }
 511         } else {
 512             i += n;
 513         }
 514     }
 515
 516     if (drv->bdrv_make_empty)
 517         return drv->bdrv_make_empty(bs);
 518
 519     return 0;
 520 }
 521
 522 /* return < 0 if error. See bdrv_write() for the return codes */
 523 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 524               uint8_t *buf, int nb_sectors)
 525 {
 526     BlockDriver *drv = bs->drv;
 527
 528     if (!drv)
 529         return -ENOMEDIUM;
 530
 531     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
 532             memcpy(buf, bs->boot_sector_data, 512);
 533         sector_num++;
 534         nb_sectors--;
 535         buf += 512;
 536         if (nb_sectors == 0)
 537             return 0;
 538     }
 539     if (drv->bdrv_pread) {
 540         int ret, len;
 541         len = nb_sectors * 512;
 542         ret = drv->bdrv_pread(bs, sector_num * 512, buf, len);
 543         if (ret < 0)
 544             return ret;
 545         else if (ret != len)
 546             return -EINVAL;
 547         else {
 548             bs->rd_bytes += (unsigned) len;
 549             bs->rd_ops ++;
 550             return 0;
 551         }
 552     } else {
 553         return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 554     }
 555 }
 556
 557 /* Return < 0 if error. Important errors are:
 558   -EIO         generic I/O error (may happen for all errors)
 559   -ENOMEDIUM   No media inserted.
 560   -EINVAL      Invalid sector number or nb_sectors
 561   -EACCES      Trying to write a read-only device
 562 */
 563 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 564                const uint8_t *buf, int nb_sectors)
 565 {
 566     BlockDriver *drv = bs->drv;
 567     if (!bs->drv)
 568         return -ENOMEDIUM;
 569     if (bs->read_only)
 570         return -EACCES;
 571     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
 572         memcpy(bs->boot_sector_data, buf, 512);
 573     }
 574     if (drv->bdrv_pwrite) {
 575         int ret, len;
 576         len = nb_sectors * 512;
 577         ret = drv->bdrv_pwrite(bs, sector_num * 512, buf, len);
 578         if (ret < 0)
 579             return ret;
 580         else if (ret != len)
 581             return -EIO;
 582         else {
 583             bs->wr_bytes += (unsigned) len;
 584             bs->wr_ops ++;
 585             return 0;
 586         }
 587     } else {
 588         return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 589     }
 590 }
 591
 592 static int bdrv_pread_em(BlockDriverState *bs, int64_t offset,
 593                          uint8_t *buf, int count1)
 594 {
 595     uint8_t tmp_buf[SECTOR_SIZE];
 596     int len, nb_sectors, count;
 597     int64_t sector_num;
 598
 599     count = count1;
 600     /* first read to align to sector start */
 601     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 602     if (len > count)
 603         len = count;
 604     sector_num = offset >> SECTOR_BITS;
 605     if (len > 0) {
 606         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 607             return -EIO;
 608         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
 609         count -= len;
 610         if (count == 0)
 611             return count1;
 612         sector_num++;
 613         buf += len;
 614     }
 615
 616     /* read the sectors "in place" */
 617     nb_sectors = count >> SECTOR_BITS;
 618     if (nb_sectors > 0) {
 619         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
 620             return -EIO;
 621         sector_num += nb_sectors;
 622         len = nb_sectors << SECTOR_BITS;
 623         buf += len;
 624         count -= len;
 625     }
 626
 627     /* add data from the last sector */
 628     if (count > 0) {
 629         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 630             return -EIO;
 631         memcpy(buf, tmp_buf, count);
 632     }
 633     return count1;
 634 }
 635
 636 static int bdrv_pwrite_em(BlockDriverState *bs, int64_t offset,
 637                           const uint8_t *buf, int count1)
 638 {
 639     uint8_t tmp_buf[SECTOR_SIZE];
 640     int len, nb_sectors, count;
 641     int64_t sector_num;
 642
 643     count = count1;
 644     /* first write to align to sector start */
 645     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 646     if (len > count)
 647         len = count;
 648     sector_num = offset >> SECTOR_BITS;
 649     if (len > 0) {
 650         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 651             return -EIO;
 652         memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
 653         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 654             return -EIO;
 655         count -= len;
 656         if (count == 0)
 657             return count1;
 658         sector_num++;
 659         buf += len;
 660     }
 661
 662     /* write the sectors "in place" */
 663     nb_sectors = count >> SECTOR_BITS;
 664     if (nb_sectors > 0) {
 665         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
 666             return -EIO;
 667         sector_num += nb_sectors;
 668         len = nb_sectors << SECTOR_BITS;
 669         buf += len;
 670         count -= len;
 671     }
 672
 673     /* add data from the last sector */
 674     if (count > 0) {
 675         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 676             return -EIO;
 677         memcpy(tmp_buf, buf, count);
 678         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 679             return -EIO;
 680     }
 681     return count1;
 682 }
 683
 684 /**
 685  * Read with byte offsets (needed only for file protocols)
 686  */
 687 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 688                void *buf1, int count1)
 689 {
 690     BlockDriver *drv = bs->drv;
 691
 692     if (!drv)
 693         return -ENOMEDIUM;
 694     if (!drv->bdrv_pread)
 695         return bdrv_pread_em(bs, offset, buf1, count1);
 696     return drv->bdrv_pread(bs, offset, buf1, count1);
 697 }
 698
 699 /**
 700  * Write with byte offsets (needed only for file protocols)
 701  */
 702 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 703                 const void *buf1, int count1)
 704 {
 705     BlockDriver *drv = bs->drv;
 706
 707     if (!drv)
 708         return -ENOMEDIUM;
 709     if (!drv->bdrv_pwrite)
 710         return bdrv_pwrite_em(bs, offset, buf1, count1);
 711     return drv->bdrv_pwrite(bs, offset, buf1, count1);
 712 }
 713
 714 /**
 715  * Truncate file to 'offset' bytes (needed only for file protocols)
 716  */
 717 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
 718 {
 719     BlockDriver *drv = bs->drv;
 720     if (!drv)
 721         return -ENOMEDIUM;
 722     if (!drv->bdrv_truncate)
 723         return -ENOTSUP;
 724     return drv->bdrv_truncate(bs, offset);
 725 }
 726
 727 /**
 728  * Length of a file in bytes. Return < 0 if error or unknown.
 729  */
 730 int64_t bdrv_getlength(BlockDriverState *bs)
 731 {
 732     BlockDriver *drv = bs->drv;
 733     if (!drv)
 734         return -ENOMEDIUM;
 735     if (!drv->bdrv_getlength) {
 736         /* legacy mode */
 737         return bs->total_sectors * SECTOR_SIZE;
 738     }
 739     return drv->bdrv_getlength(bs);
 740 }
 741
 742 /* return 0 as number of sectors if no device present or error */
 743 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
 744 {
 745     int64_t length;
 746     length = bdrv_getlength(bs);
 747     if (length < 0)
 748         length = 0;
 749     else
 750         length = length >> SECTOR_BITS;
 751     *nb_sectors_ptr = length;
 752 }
 753
 754 /* force a given boot sector. */
 755 void bdrv_set_boot_sector(BlockDriverState *bs, const uint8_t *data, int size)
 756 {
 757     bs->boot_sector_enabled = 1;
 758     if (size > 512)
 759         size = 512;
 760     memcpy(bs->boot_sector_data, data, size);
 761     memset(bs->boot_sector_data + size, 0, 512 - size);
 762 }
 763
 764 struct partition {
 765         uint8_t boot_ind;               /* 0x80 - active */
 766         uint8_t head;           /* starting head */
 767         uint8_t sector;         /* starting sector */
 768         uint8_t cyl;            /* starting cylinder */
 769         uint8_t sys_ind;                /* What partition type */
 770         uint8_t end_head;               /* end head */
 771         uint8_t end_sector;     /* end sector */
 772         uint8_t end_cyl;                /* end cylinder */
 773         uint32_t start_sect;    /* starting sector counting from 0 */
 774         uint32_t nr_sects;              /* nr of sectors in partition */
 775 } __attribute__((packed));
 776
 777 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
 778 static int guess_disk_lchs(BlockDriverState *bs,
 779                            int *pcylinders, int *pheads, int *psectors)
 780 {
 781     uint8_t *buf;
 782     int ret, i, heads, sectors, cylinders;
 783     struct partition *p;
 784     uint32_t nr_sects;
 785     int64_t nb_sectors;
 786
 787     buf = qemu_memalign(512, 512);
 788     if (buf == NULL)
 789         return -1;
 790
 791     bdrv_get_geometry(bs, &nb_sectors);
 792
 793     ret = bdrv_read(bs, 0, buf, 1);
 794     if (ret < 0)
 795         return -1;
 796     /* test msdos magic */
 797     if (buf[510] != 0x55 || buf[511] != 0xaa) {
 798         qemu_free(buf);
 799         return -1;
 800     }
 801     for(i = 0; i < 4; i++) {
 802         p = ((struct partition *)(buf + 0x1be)) + i;
 803         nr_sects = le32_to_cpu(p->nr_sects);
 804         if (nr_sects && p->end_head) {
 805             /* We make the assumption that the partition terminates on
 806                a cylinder boundary */
 807             heads = p->end_head + 1;
 808             sectors = p->end_sector & 63;
 809             if (sectors == 0)
 810                 continue;
 811             cylinders = nb_sectors / (heads * sectors);
 812             if (cylinders < 1 || cylinders > 16383)
 813                 continue;
 814             *pheads = heads;
 815             *psectors = sectors;
 816             *pcylinders = cylinders;
 817 #if 0
 818             printf("guessed geometry: LCHS=%d %d %d\n",
 819                    cylinders, heads, sectors);
 820 #endif
 821             qemu_free(buf);
 822             return 0;
 823         }
 824     }
 825     qemu_free(buf);
 826     return -1;
 827 }
 828
 829 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
 830 {
 831     int translation, lba_detected = 0;
 832     int cylinders, heads, secs;
 833     int64_t nb_sectors;
 834
 835     /* if a geometry hint is available, use it */
 836     bdrv_get_geometry(bs, &nb_sectors);
 837     bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
 838     translation = bdrv_get_translation_hint(bs);
 839     if (cylinders != 0) {
 840         *pcyls = cylinders;
 841         *pheads = heads;
 842         *psecs = secs;
 843     } else {
 844         if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
 845             if (heads > 16) {
 846                 /* if heads > 16, it means that a BIOS LBA
 847                    translation was active, so the default
 848                    hardware geometry is OK */
 849                 lba_detected = 1;
 850                 goto default_geometry;
 851             } else {
 852                 *pcyls = cylinders;
 853                 *pheads = heads;
 854                 *psecs = secs;
 855                 /* disable any translation to be in sync with
 856                    the logical geometry */
 857                 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
 858                     bdrv_set_translation_hint(bs,
 859                                               BIOS_ATA_TRANSLATION_NONE);
 860                 }
 861             }
 862         } else {
 863         default_geometry:
 864             /* if no geometry, use a standard physical disk geometry */
 865             cylinders = nb_sectors / (16 * 63);
 866
 867             if (cylinders > 16383)
 868                 cylinders = 16383;
 869             else if (cylinders < 2)
 870                 cylinders = 2;
 871             *pcyls = cylinders;
 872             *pheads = 16;
 873             *psecs = 63;
 874             if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
 875                 if ((*pcyls * *pheads) <= 131072) {
 876                     bdrv_set_translation_hint(bs,
 877                                               BIOS_ATA_TRANSLATION_LARGE);
 878                 } else {
 879                     bdrv_set_translation_hint(bs,
 880                                               BIOS_ATA_TRANSLATION_LBA);
 881                 }
 882             }
 883         }
 884         bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
 885     }
 886 }
 887
 888 void bdrv_set_geometry_hint(BlockDriverState *bs,
 889                             int cyls, int heads, int secs)
 890 {
 891     bs->cyls = cyls;
 892     bs->heads = heads;
 893     bs->secs = secs;
 894 }
 895
 896 void bdrv_set_type_hint(BlockDriverState *bs, int type)
 897 {
 898     bs->type = type;
 899     bs->removable = ((type == BDRV_TYPE_CDROM ||
 900                       type == BDRV_TYPE_FLOPPY));
 901 }
 902
 903 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
 904 {
 905     bs->translation = translation;
 906 }
 907
 908 void bdrv_get_geometry_hint(BlockDriverState *bs,
 909                             int *pcyls, int *pheads, int *psecs)
 910 {
 911     *pcyls = bs->cyls;
 912     *pheads = bs->heads;
 913     *psecs = bs->secs;
 914 }
 915
 916 int bdrv_get_type_hint(BlockDriverState *bs)
 917 {
 918     return bs->type;
 919 }
 920
 921 int bdrv_get_translation_hint(BlockDriverState *bs)
 922 {
 923     return bs->translation;
 924 }
 925
 926 int bdrv_is_removable(BlockDriverState *bs)
 927 {
 928     return bs->removable;
 929 }
 930
 931 int bdrv_is_read_only(BlockDriverState *bs)
 932 {
 933     return bs->read_only;
 934 }
 935
 936 int bdrv_is_sg(BlockDriverState *bs)
 937 {
 938     return bs->sg;
 939 }
 940
 941 /* XXX: no longer used */
 942 void bdrv_set_change_cb(BlockDriverState *bs,
 943                         void (*change_cb)(void *opaque), void *opaque)
 944 {
 945     bs->change_cb = change_cb;
 946     bs->change_opaque = opaque;
 947 }
 948
 949 int bdrv_is_encrypted(BlockDriverState *bs)
 950 {
 951     if (bs->backing_hd && bs->backing_hd->encrypted)
 952         return 1;
 953     return bs->encrypted;
 954 }
 955
 956 int bdrv_set_key(BlockDriverState *bs, const char *key)
 957 {
 958     int ret;
 959     if (bs->backing_hd && bs->backing_hd->encrypted) {
 960         ret = bdrv_set_key(bs->backing_hd, key);
 961         if (ret < 0)
 962             return ret;
 963         if (!bs->encrypted)
 964             return 0;
 965     }
 966     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
 967         return -1;
 968     return bs->drv->bdrv_set_key(bs, key);
 969 }
 970
 971 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
 972 {
 973     if (!bs->drv) {
 974         buf[0] = '\0';
 975     } else {
 976         pstrcpy(buf, buf_size, bs->drv->format_name);
 977     }
 978 }
 979
 980 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
 981                          void *opaque)
 982 {
 983     BlockDriver *drv;
 984
 985     for (drv = first_drv; drv != NULL; drv = drv->next) {
 986         it(opaque, drv->format_name);
 987     }
 988 }
 989
 990 BlockDriverState *bdrv_find(const char *name)
 991 {
 992     BlockDriverState *bs;
 993
 994     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 995         if (!strcmp(name, bs->device_name))
 996             return bs;
 997     }
 998     return NULL;
 999 }
1000
1001 void bdrv_iterate(void (*it)(void *opaque, const char *name), void *opaque)
1002 {
1003     BlockDriverState *bs;
1004
1005     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1006         it(opaque, bs->device_name);
1007     }
1008 }
1009
1010 const char *bdrv_get_device_name(BlockDriverState *bs)
1011 {
1012     return bs->device_name;
1013 }
1014
1015 void bdrv_flush(BlockDriverState *bs)
1016 {
1017     if (bs->drv->bdrv_flush)
1018         bs->drv->bdrv_flush(bs);
1019     if (bs->backing_hd)
1020         bdrv_flush(bs->backing_hd);
1021 }
1022
1023 void bdrv_iterate_writeable(void (*it)(BlockDriverState *bs))
1024 {
1025     BlockDriverState *bs;
1026
1027     for (bs = bdrv_first; bs != NULL; bs = bs->next)
1028         if (bs->drv && !bdrv_is_read_only(bs) &&
1029             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1030             it(bs);
1031 }
1032
1033 void bdrv_flush_all(void)
1034 {
1035     BlockDriverState *bs;
1036
1037     for (bs = bdrv_first; bs != NULL; bs = bs->next)
1038         if (bs->drv && !bdrv_is_read_only(bs) &&
1039             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1040             bdrv_flush(bs);
1041 }
1042
1043 /*
1044  * Returns true iff the specified sector is present in the disk image. Drivers
1045  * not implementing the functionality are assumed to not support backing files,
1046  * hence all their sectors are reported as allocated.
1047  *
1048  * 'pnum' is set to the number of sectors (including and immediately following
1049  * the specified sector) that are known to be in the same
1050  * allocated/unallocated state.
1051  *
1052  * 'nb_sectors' is the max value 'pnum' should be set to.
1053  */
1054 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1055         int *pnum)
1056 {
1057     int64_t n;
1058     if (!bs->drv->bdrv_is_allocated) {
1059         if (sector_num >= bs->total_sectors) {
1060             *pnum = 0;
1061             return 0;
1062         }
1063         n = bs->total_sectors - sector_num;
1064         *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1065         return 1;
1066     }
1067     return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1068 }
1069
1070 void bdrv_info(void)
1071 {
1072     BlockDriverState *bs;
1073
1074     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1075         term_printf("%s:", bs->device_name);
1076         term_printf(" type=");
1077         switch(bs->type) {
1078         case BDRV_TYPE_HD:
1079             term_printf("hd");
1080             break;
1081         case BDRV_TYPE_CDROM:
1082             term_printf("cdrom");
1083             break;
1084         case BDRV_TYPE_FLOPPY:
1085             term_printf("floppy");
1086             break;
1087         }
1088         term_printf(" removable=%d", bs->removable);
1089         if (bs->removable) {
1090             term_printf(" locked=%d", bs->locked);
1091         }
1092         if (bs->drv) {
1093             term_printf(" file=");
1094             term_print_filename(bs->filename);
1095             if (bs->backing_file[0] != '\0') {
1096                 term_printf(" backing_file=");
1097                 term_print_filename(bs->backing_file);
1098             }
1099             term_printf(" ro=%d", bs->read_only);
1100             term_printf(" drv=%s", bs->drv->format_name);
1101             if (bs->encrypted)
1102                 term_printf(" encrypted");
1103         } else {
1104             term_printf(" [not inserted]");
1105         }
1106         term_printf("\n");
1107     }
1108 }
1109
1110 /* The "info blockstats" command. */
1111 void bdrv_info_stats (void)
1112 {
1113     BlockDriverState *bs;
1114
1115     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1116         term_printf ("%s:"
1117                      " rd_bytes=%" PRIu64
1118                      " wr_bytes=%" PRIu64
1119                      " rd_operations=%" PRIu64
1120                      " wr_operations=%" PRIu64
1121                      "\n",
1122                      bs->device_name,
1123                      bs->rd_bytes, bs->wr_bytes,
1124                      bs->rd_ops, bs->wr_ops);
1125     }
1126 }
1127
1128 void bdrv_get_backing_filename(BlockDriverState *bs,
1129                                char *filename, int filename_size)
1130 {
1131     if (!bs->backing_hd) {
1132         pstrcpy(filename, filename_size, "");
1133     } else {
1134         pstrcpy(filename, filename_size, bs->backing_file);
1135     }
1136 }
1137
1138 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1139                           const uint8_t *buf, int nb_sectors)
1140 {
1141     BlockDriver *drv = bs->drv;
1142     if (!drv)
1143         return -ENOMEDIUM;
1144     if (!drv->bdrv_write_compressed)
1145         return -ENOTSUP;
1146     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1147 }
1148
1149 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1150 {
1151     BlockDriver *drv = bs->drv;
1152     if (!drv)
1153         return -ENOMEDIUM;
1154     if (!drv->bdrv_get_info)
1155         return -ENOTSUP;
1156     memset(bdi, 0, sizeof(*bdi));
1157     return drv->bdrv_get_info(bs, bdi);
1158 }
1159
1160 /**************************************************************/
1161 /* handling of snapshots */
1162
1163 int bdrv_snapshot_create(BlockDriverState *bs,
1164                          QEMUSnapshotInfo *sn_info)
1165 {
1166     BlockDriver *drv = bs->drv;
1167     if (!drv)
1168         return -ENOMEDIUM;
1169     if (!drv->bdrv_snapshot_create)
1170         return -ENOTSUP;
1171     return drv->bdrv_snapshot_create(bs, sn_info);
1172 }
1173
1174 int bdrv_snapshot_goto(BlockDriverState *bs,
1175                        const char *snapshot_id)
1176 {
1177     BlockDriver *drv = bs->drv;
1178     if (!drv)
1179         return -ENOMEDIUM;
1180     if (!drv->bdrv_snapshot_goto)
1181         return -ENOTSUP;
1182     return drv->bdrv_snapshot_goto(bs, snapshot_id);
1183 }
1184
1185 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1186 {
1187     BlockDriver *drv = bs->drv;
1188     if (!drv)
1189         return -ENOMEDIUM;
1190     if (!drv->bdrv_snapshot_delete)
1191         return -ENOTSUP;
1192     return drv->bdrv_snapshot_delete(bs, snapshot_id);
1193 }
1194
1195 int bdrv_snapshot_list(BlockDriverState *bs,
1196                        QEMUSnapshotInfo **psn_info)
1197 {
1198     BlockDriver *drv = bs->drv;
1199     if (!drv)
1200         return -ENOMEDIUM;
1201     if (!drv->bdrv_snapshot_list)
1202         return -ENOTSUP;
1203     return drv->bdrv_snapshot_list(bs, psn_info);
1204 }
1205
1206 #define NB_SUFFIXES 4
1207
1208 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1209 {
1210     static const char suffixes[NB_SUFFIXES] = "KMGT";
1211     int64_t base;
1212     int i;
1213
1214     if (size <= 999) {
1215         snprintf(buf, buf_size, "%" PRId64, size);
1216     } else {
1217         base = 1024;
1218         for(i = 0; i < NB_SUFFIXES; i++) {
1219             if (size < (10 * base)) {
1220                 snprintf(buf, buf_size, "%0.1f%c",
1221                          (double)size / base,
1222                          suffixes[i]);
1223                 break;
1224             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1225                 snprintf(buf, buf_size, "%" PRId64 "%c",
1226                          ((size + (base >> 1)) / base),
1227                          suffixes[i]);
1228                 break;
1229             }
1230             base = base * 1024;
1231         }
1232     }
1233     return buf;
1234 }
1235
1236 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1237 {
1238     char buf1[128], date_buf[128], clock_buf[128];
1239 #ifdef _WIN32
1240     struct tm *ptm;
1241 #else
1242     struct tm tm;
1243 #endif
1244     time_t ti;
1245     int64_t secs;
1246
1247     if (!sn) {
1248         snprintf(buf, buf_size,
1249                  "%-10s%-20s%7s%20s%15s",
1250                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1251     } else {
1252         ti = sn->date_sec;
1253 #ifdef _WIN32
1254         ptm = localtime(&ti);
1255         strftime(date_buf, sizeof(date_buf),
1256                  "%Y-%m-%d %H:%M:%S", ptm);
1257 #else
1258         localtime_r(&ti, &tm);
1259         strftime(date_buf, sizeof(date_buf),
1260                  "%Y-%m-%d %H:%M:%S", &tm);
1261 #endif
1262         secs = sn->vm_clock_nsec / 1000000000;
1263         snprintf(clock_buf, sizeof(clock_buf),
1264                  "%02d:%02d:%02d.%03d",
1265                  (int)(secs / 3600),
1266                  (int)((secs / 60) % 60),
1267                  (int)(secs % 60),
1268                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1269         snprintf(buf, buf_size,
1270                  "%-10s%-20s%7s%20s%15s",
1271                  sn->id_str, sn->name,
1272                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1273                  date_buf,
1274                  clock_buf);
1275     }
1276     return buf;
1277 }
1278
1279
1280 /**************************************************************/
1281 /* async I/Os */
1282
1283 BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
1284                                 uint8_t *buf, int nb_sectors,
1285                                 BlockDriverCompletionFunc *cb, void *opaque)
1286 {
1287     BlockDriver *drv = bs->drv;
1288     BlockDriverAIOCB *ret;
1289
1290     if (!drv)
1291         return NULL;
1292
1293     /* XXX: we assume that nb_sectors == 0 is suppored by the async read */
1294     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
1295         memcpy(buf, bs->boot_sector_data, 512);
1296         sector_num++;
1297         nb_sectors--;
1298         buf += 512;
1299     }
1300
1301     ret = drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
1302
1303     if (ret) {
1304         /* Update stats even though technically transfer has not happened. */
1305         bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1306         bs->rd_ops ++;
1307     }
1308
1309     return ret;
1310 }
1311
1312 BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
1313                                  const uint8_t *buf, int nb_sectors,
1314                                  BlockDriverCompletionFunc *cb, void *opaque)
1315 {
1316     BlockDriver *drv = bs->drv;
1317     BlockDriverAIOCB *ret;
1318
1319     if (!drv)
1320         return NULL;
1321     if (bs->read_only)
1322         return NULL;
1323     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
1324         memcpy(bs->boot_sector_data, buf, 512);
1325     }
1326
1327     ret = drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
1328
1329     if (ret) {
1330         /* Update stats even though technically transfer has not happened. */
1331         bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1332         bs->wr_ops ++;
1333     }
1334
1335     return ret;
1336 }
1337
1338 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1339 {
1340     BlockDriver *drv = acb->bs->drv;
1341
1342     drv->bdrv_aio_cancel(acb);
1343 }
1344
1345
1346 /**************************************************************/
1347 /* async block device emulation */
1348
1349 static void bdrv_aio_bh_cb(void *opaque)
1350 {
1351     BlockDriverAIOCBSync *acb = opaque;
1352     acb->common.cb(acb->common.opaque, acb->ret);
1353     qemu_aio_release(acb);
1354 }
1355
1356 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1357         int64_t sector_num, uint8_t *buf, int nb_sectors,
1358         BlockDriverCompletionFunc *cb, void *opaque)
1359 {
1360     BlockDriverAIOCBSync *acb;
1361     int ret;
1362
1363     acb = qemu_aio_get(bs, cb, opaque);
1364     if (!acb->bh)
1365         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1366     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1367     acb->ret = ret;
1368     qemu_bh_schedule(acb->bh);
1369     return &acb->common;
1370 }
1371
1372 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1373         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1374         BlockDriverCompletionFunc *cb, void *opaque)
1375 {
1376     BlockDriverAIOCBSync *acb;
1377     int ret;
1378
1379     acb = qemu_aio_get(bs, cb, opaque);
1380     if (!acb->bh)
1381         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1382     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1383     acb->ret = ret;
1384     qemu_bh_schedule(acb->bh);
1385     return &acb->common;
1386 }
1387
1388 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1389 {
1390     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1391     qemu_bh_cancel(acb->bh);
1392     qemu_aio_release(acb);
1393 }
1394
1395 /**************************************************************/
1396 /* sync block device emulation */
1397
1398 static void bdrv_rw_em_cb(void *opaque, int ret)
1399 {
1400     *(int *)opaque = ret;
1401 }
1402
1403 #define NOT_DONE 0x7fffffff
1404
1405 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1406                         uint8_t *buf, int nb_sectors)
1407 {
1408     int async_ret;
1409     BlockDriverAIOCB *acb;
1410
1411     async_ret = NOT_DONE;
1412     acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
1413                         bdrv_rw_em_cb, &async_ret);
1414     if (acb == NULL)
1415         return -1;
1416
1417     while (async_ret == NOT_DONE) {
1418         qemu_aio_wait();
1419     }
1420
1421     return async_ret;
1422 }
1423
1424 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1425                          const uint8_t *buf, int nb_sectors)
1426 {
1427     int async_ret;
1428     BlockDriverAIOCB *acb;
1429
1430     async_ret = NOT_DONE;
1431     acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
1432                          bdrv_rw_em_cb, &async_ret);
1433     if (acb == NULL)
1434         return -1;
1435     while (async_ret == NOT_DONE) {
1436         qemu_aio_wait();
1437     }
1438     return async_ret;
1439 }
1440
1441 void bdrv_init(void)
1442 {
1443     bdrv_register(&bdrv_raw);
1444     bdrv_register(&bdrv_host_device);
1445 #ifndef _WIN32
1446     bdrv_register(&bdrv_cow);
1447 #endif
1448     bdrv_register(&bdrv_qcow);
1449     bdrv_register(&bdrv_vmdk);
1450     bdrv_register(&bdrv_cloop);
1451     bdrv_register(&bdrv_dmg);
1452     bdrv_register(&bdrv_bochs);
1453     bdrv_register(&bdrv_vpc);
1454     bdrv_register(&bdrv_vvfat);
1455     bdrv_register(&bdrv_qcow2);
1456     bdrv_register(&bdrv_parallels);
1457     bdrv_register(&bdrv_nbd);
1458 }
1459
1460 void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
1461                    void *opaque)
1462 {
1463     BlockDriver *drv;
1464     BlockDriverAIOCB *acb;
1465
1466     drv = bs->drv;
1467     if (drv->free_aiocb) {
1468         acb = drv->free_aiocb;
1469         drv->free_aiocb = acb->next;
1470     } else {
1471         acb = qemu_mallocz(drv->aiocb_size);
1472         if (!acb)
1473             return NULL;
1474     }
1475     acb->bs = bs;
1476     acb->cb = cb;
1477     acb->opaque = opaque;
1478     return acb;
1479 }
1480
1481 void qemu_aio_release(void *p)
1482 {
1483     BlockDriverAIOCB *acb = p;
1484     BlockDriver *drv = acb->bs->drv;
1485     acb->next = drv->free_aiocb;
1486     drv->free_aiocb = acb;
1487 }
1488
1489 /**************************************************************/
1490 /* removable device support */
1491
1492 /**
1493  * Return TRUE if the media is present
1494  */
1495 int bdrv_is_inserted(BlockDriverState *bs)
1496 {
1497     BlockDriver *drv = bs->drv;
1498     int ret;
1499     if (!drv)
1500         return 0;
1501     if (!drv->bdrv_is_inserted)
1502         return 1;
1503     ret = drv->bdrv_is_inserted(bs);
1504     return ret;
1505 }
1506
1507 /**
1508  * Return TRUE if the media changed since the last call to this
1509  * function. It is currently only used for floppy disks
1510  */
1511 int bdrv_media_changed(BlockDriverState *bs)
1512 {
1513     BlockDriver *drv = bs->drv;
1514     int ret;
1515
1516     if (!drv || !drv->bdrv_media_changed)
1517         ret = -ENOTSUP;
1518     else
1519         ret = drv->bdrv_media_changed(bs);
1520     if (ret == -ENOTSUP)
1521         ret = bs->media_changed;
1522     bs->media_changed = 0;
1523     return ret;
1524 }
1525
1526 /**
1527  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1528  */
1529 void bdrv_eject(BlockDriverState *bs, int eject_flag)
1530 {
1531     BlockDriver *drv = bs->drv;
1532     int ret;
1533
1534     if (!drv || !drv->bdrv_eject) {
1535         ret = -ENOTSUP;
1536     } else {
1537         ret = drv->bdrv_eject(bs, eject_flag);
1538     }
1539     if (ret == -ENOTSUP) {
1540         if (eject_flag)
1541             bdrv_close(bs);
1542     }
1543 }
1544
1545 int bdrv_is_locked(BlockDriverState *bs)
1546 {
1547     return bs->locked;
1548 }
1549
1550 /**
1551  * Lock or unlock the media (if it is locked, the user won't be able
1552  * to eject it manually).
1553  */
1554 void bdrv_set_locked(BlockDriverState *bs, int locked)
1555 {
1556     BlockDriver *drv = bs->drv;
1557
1558     bs->locked = locked;
1559     if (drv && drv->bdrv_set_locked) {
1560         drv->bdrv_set_locked(bs, locked);
1561     }
1562 }
1563
1564 /* needed for generic scsi interface */
1565
1566 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1567 {
1568     BlockDriver *drv = bs->drv;
1569
1570     if (drv && drv->bdrv_ioctl)
1571         return drv->bdrv_ioctl(bs, req, buf);
1572     return -ENOTSUP;
1573 }