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 "config-host.h"
  25 #ifdef HOST_BSD
  26 /* include native header before sys-queue.h */
  27 #include <sys/queue.h>
  28 #endif
  29
  30 #include "qemu-common.h"
  31 #include "monitor.h"
  32 #include "block_int.h"
  33
  34 #ifdef HOST_BSD
  35 #include <sys/types.h>
  36 #include <sys/stat.h>
  37 #include <sys/ioctl.h>
  38 #ifndef __DragonFly__
  39 #include <sys/disk.h>
  40 #endif
  41 #endif
  42
  43 #ifdef _WIN32
  44 #include <windows.h>
  45 #endif
  46
  47 #define SECTOR_BITS 9
  48 #define SECTOR_SIZE (1 << SECTOR_BITS)
  49
  50 typedef struct BlockDriverAIOCBSync {
  51     BlockDriverAIOCB common;
  52     QEMUBH *bh;
  53     int ret;
  54 } BlockDriverAIOCBSync;
  55
  56 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
  57         int64_t sector_num, uint8_t *buf, int nb_sectors,
  58         BlockDriverCompletionFunc *cb, void *opaque);
  59 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
  60         int64_t sector_num, const uint8_t *buf, int nb_sectors,
  61         BlockDriverCompletionFunc *cb, void *opaque);
  62 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
  63 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
  64                         uint8_t *buf, int nb_sectors);
  65 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
  66                          const uint8_t *buf, int nb_sectors);
  67
  68 BlockDriverState *bdrv_first;
  69
  70 static BlockDriver *first_drv;
  71
  72 int path_is_absolute(const char *path)
  73 {
  74     const char *p;
  75 #ifdef _WIN32
  76     /* specific case for names like: "\\.\d:" */
  77     if (*path == '/' || *path == '\\')
  78         return 1;
  79 #endif
  80     p = strchr(path, ':');
  81     if (p)
  82         p++;
  83     else
  84         p = path;
  85 #ifdef _WIN32
  86     return (*p == '/' || *p == '\\');
  87 #else
  88     return (*p == '/');
  89 #endif
  90 }
  91
  92 /* if filename is absolute, just copy it to dest. Otherwise, build a
  93    path to it by considering it is relative to base_path. URL are
  94    supported. */
  95 void path_combine(char *dest, int dest_size,
  96                   const char *base_path,
  97                   const char *filename)
  98 {
  99     const char *p, *p1;
 100     int len;
 101
 102     if (dest_size <= 0)
 103         return;
 104     if (path_is_absolute(filename)) {
 105         pstrcpy(dest, dest_size, filename);
 106     } else {
 107         p = strchr(base_path, ':');
 108         if (p)
 109             p++;
 110         else
 111             p = base_path;
 112         p1 = strrchr(base_path, '/');
 113 #ifdef _WIN32
 114         {
 115             const char *p2;
 116             p2 = strrchr(base_path, '\\');
 117             if (!p1 || p2 > p1)
 118                 p1 = p2;
 119         }
 120 #endif
 121         if (p1)
 122             p1++;
 123         else
 124             p1 = base_path;
 125         if (p1 > p)
 126             p = p1;
 127         len = p - base_path;
 128         if (len > dest_size - 1)
 129             len = dest_size - 1;
 130         memcpy(dest, base_path, len);
 131         dest[len] = '\0';
 132         pstrcat(dest, dest_size, filename);
 133     }
 134 }
 135
 136
 137 static void bdrv_register(BlockDriver *bdrv)
 138 {
 139     if (!bdrv->bdrv_aio_read) {
 140         /* add AIO emulation layer */
 141         bdrv->bdrv_aio_read = bdrv_aio_read_em;
 142         bdrv->bdrv_aio_write = bdrv_aio_write_em;
 143         bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
 144         bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
 145     } else if (!bdrv->bdrv_read) {
 146         /* add synchronous IO emulation layer */
 147         bdrv->bdrv_read = bdrv_read_em;
 148         bdrv->bdrv_write = bdrv_write_em;
 149     }
 150     bdrv->next = first_drv;
 151     first_drv = bdrv;
 152 }
 153
 154 /* create a new block device (by default it is empty) */
 155 BlockDriverState *bdrv_new(const char *device_name)
 156 {
 157     BlockDriverState **pbs, *bs;
 158
 159     bs = qemu_mallocz(sizeof(BlockDriverState));
 160     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
 161     if (device_name[0] != '\0') {
 162         /* insert at the end */
 163         pbs = &bdrv_first;
 164         while (*pbs != NULL)
 165             pbs = &(*pbs)->next;
 166         *pbs = bs;
 167     }
 168     return bs;
 169 }
 170
 171 BlockDriver *bdrv_find_format(const char *format_name)
 172 {
 173     BlockDriver *drv1;
 174     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 175         if (!strcmp(drv1->format_name, format_name))
 176             return drv1;
 177     }
 178     return NULL;
 179 }
 180
 181 int bdrv_create(BlockDriver *drv,
 182                 const char *filename, int64_t size_in_sectors,
 183                 const char *backing_file, int flags)
 184 {
 185     if (!drv->bdrv_create)
 186         return -ENOTSUP;
 187     return drv->bdrv_create(filename, size_in_sectors, backing_file, flags);
 188 }
 189
 190 #ifdef _WIN32
 191 void get_tmp_filename(char *filename, int size)
 192 {
 193     char temp_dir[MAX_PATH];
 194
 195     GetTempPath(MAX_PATH, temp_dir);
 196     GetTempFileName(temp_dir, "qem", 0, filename);
 197 }
 198 #else
 199 void get_tmp_filename(char *filename, int size)
 200 {
 201     int fd;
 202     const char *tmpdir;
 203     /* XXX: race condition possible */
 204     tmpdir = getenv("TMPDIR");
 205     if (!tmpdir)
 206         tmpdir = "/tmp";
 207     snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
 208     fd = mkstemp(filename);
 209     close(fd);
 210 }
 211 #endif
 212
 213 #ifdef _WIN32
 214 static int is_windows_drive_prefix(const char *filename)
 215 {
 216     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
 217              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
 218             filename[1] == ':');
 219 }
 220
 221 static int is_windows_drive(const char *filename)
 222 {
 223     if (is_windows_drive_prefix(filename) &&
 224         filename[2] == '\0')
 225         return 1;
 226     if (strstart(filename, "\\\\.\\", NULL) ||
 227         strstart(filename, "//./", NULL))
 228         return 1;
 229     return 0;
 230 }
 231 #endif
 232
 233 static BlockDriver *find_protocol(const char *filename)
 234 {
 235     BlockDriver *drv1;
 236     char protocol[128];
 237     int len;
 238     const char *p;
 239
 240 #ifdef _WIN32
 241     if (is_windows_drive(filename) ||
 242         is_windows_drive_prefix(filename))
 243         return &bdrv_raw;
 244 #endif
 245     p = strchr(filename, ':');
 246     if (!p)
 247         return &bdrv_raw;
 248     len = p - filename;
 249     if (len > sizeof(protocol) - 1)
 250         len = sizeof(protocol) - 1;
 251     memcpy(protocol, filename, len);
 252     protocol[len] = '\0';
 253     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 254         if (drv1->protocol_name &&
 255             !strcmp(drv1->protocol_name, protocol))
 256             return drv1;
 257     }
 258     return NULL;
 259 }
 260
 261 /* XXX: force raw format if block or character device ? It would
 262    simplify the BSD case */
 263 static BlockDriver *find_image_format(const char *filename)
 264 {
 265     int ret, score, score_max;
 266     BlockDriver *drv1, *drv;
 267     uint8_t buf[2048];
 268     BlockDriverState *bs;
 269
 270     /* detect host devices. By convention, /dev/cdrom[N] is always
 271        recognized as a host CDROM */
 272     if (strstart(filename, "/dev/cdrom", NULL))
 273         return &bdrv_host_device;
 274 #ifdef _WIN32
 275     if (is_windows_drive(filename))
 276         return &bdrv_host_device;
 277 #else
 278     {
 279         struct stat st;
 280         if (stat(filename, &st) >= 0 &&
 281             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 282             return &bdrv_host_device;
 283         }
 284     }
 285 #endif
 286
 287     drv = find_protocol(filename);
 288     /* no need to test disk image formats for vvfat */
 289     if (drv == &bdrv_vvfat)
 290         return drv;
 291
 292     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
 293     if (ret < 0)
 294         return NULL;
 295     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
 296     bdrv_delete(bs);
 297     if (ret < 0) {
 298         return NULL;
 299     }
 300
 301     score_max = 0;
 302     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 303         if (drv1->bdrv_probe) {
 304             score = drv1->bdrv_probe(buf, ret, filename);
 305             if (score > score_max) {
 306                 score_max = score;
 307                 drv = drv1;
 308             }
 309         }
 310     }
 311     return drv;
 312 }
 313
 314 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
 315 {
 316     BlockDriverState *bs;
 317     int ret;
 318
 319     bs = bdrv_new("");
 320     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
 321     if (ret < 0) {
 322         bdrv_delete(bs);
 323         return ret;
 324     }
 325     bs->growable = 1;
 326     *pbs = bs;
 327     return 0;
 328 }
 329
 330 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
 331 {
 332     return bdrv_open2(bs, filename, flags, NULL);
 333 }
 334
 335 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
 336                BlockDriver *drv)
 337 {
 338     int ret, open_flags;
 339     char tmp_filename[PATH_MAX];
 340     char backing_filename[PATH_MAX];
 341
 342     bs->read_only = 0;
 343     bs->is_temporary = 0;
 344     bs->encrypted = 0;
 345     bs->valid_key = 0;
 346
 347     if (flags & BDRV_O_SNAPSHOT) {
 348         BlockDriverState *bs1;
 349         int64_t total_size;
 350         int is_protocol = 0;
 351
 352         /* if snapshot, we create a temporary backing file and open it
 353            instead of opening 'filename' directly */
 354
 355         /* if there is a backing file, use it */
 356         bs1 = bdrv_new("");
 357         ret = bdrv_open(bs1, filename, 0);
 358         if (ret < 0) {
 359             bdrv_delete(bs1);
 360             return ret;
 361         }
 362         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
 363
 364         if (bs1->drv && bs1->drv->protocol_name)
 365             is_protocol = 1;
 366
 367         bdrv_delete(bs1);
 368
 369         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
 370
 371         /* Real path is meaningless for protocols */
 372         if (is_protocol)
 373             snprintf(backing_filename, sizeof(backing_filename),
 374                      "%s", filename);
 375         else
 376             realpath(filename, backing_filename);
 377
 378         ret = bdrv_create(&bdrv_qcow2, tmp_filename,
 379                           total_size, backing_filename, 0);
 380         if (ret < 0) {
 381             return ret;
 382         }
 383         filename = tmp_filename;
 384         bs->is_temporary = 1;
 385     }
 386
 387     pstrcpy(bs->filename, sizeof(bs->filename), filename);
 388     if (flags & BDRV_O_FILE) {
 389         drv = find_protocol(filename);
 390     } else if (!drv) {
 391         drv = find_image_format(filename);
 392     }
 393     if (!drv) {
 394         ret = -ENOENT;
 395         goto unlink_and_fail;
 396     }
 397     bs->drv = drv;
 398     bs->opaque = qemu_mallocz(drv->instance_size);
 399     /* Note: for compatibility, we open disk image files as RDWR, and
 400        RDONLY as fallback */
 401     if (!(flags & BDRV_O_FILE))
 402         open_flags = BDRV_O_RDWR | (flags & BDRV_O_CACHE_MASK);
 403     else
 404         open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
 405     ret = drv->bdrv_open(bs, filename, open_flags);
 406     if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
 407         ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
 408         bs->read_only = 1;
 409     }
 410     if (ret < 0) {
 411         qemu_free(bs->opaque);
 412         bs->opaque = NULL;
 413         bs->drv = NULL;
 414     unlink_and_fail:
 415         if (bs->is_temporary)
 416             unlink(filename);
 417         return ret;
 418     }
 419     if (drv->bdrv_getlength) {
 420         bs->total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 421     }
 422 #ifndef _WIN32
 423     if (bs->is_temporary) {
 424         unlink(filename);
 425     }
 426 #endif
 427     if (bs->backing_file[0] != '\0') {
 428         /* if there is a backing file, use it */
 429         bs->backing_hd = bdrv_new("");
 430         path_combine(backing_filename, sizeof(backing_filename),
 431                      filename, bs->backing_file);
 432         ret = bdrv_open(bs->backing_hd, backing_filename, open_flags);
 433         if (ret < 0) {
 434             bdrv_close(bs);
 435             return ret;
 436         }
 437     }
 438
 439     if (!bdrv_key_required(bs)) {
 440         /* call the change callback */
 441         bs->media_changed = 1;
 442         if (bs->change_cb)
 443             bs->change_cb(bs->change_opaque);
 444     }
 445     return 0;
 446 }
 447
 448 void bdrv_close(BlockDriverState *bs)
 449 {
 450     if (bs->drv) {
 451         if (bs->backing_hd)
 452             bdrv_delete(bs->backing_hd);
 453         bs->drv->bdrv_close(bs);
 454         qemu_free(bs->opaque);
 455 #ifdef _WIN32
 456         if (bs->is_temporary) {
 457             unlink(bs->filename);
 458         }
 459 #endif
 460         bs->opaque = NULL;
 461         bs->drv = NULL;
 462
 463         /* call the change callback */
 464         bs->media_changed = 1;
 465         if (bs->change_cb)
 466             bs->change_cb(bs->change_opaque);
 467     }
 468 }
 469
 470 void bdrv_delete(BlockDriverState *bs)
 471 {
 472     BlockDriverState **pbs;
 473
 474     pbs = &bdrv_first;
 475     while (*pbs != bs && *pbs != NULL)
 476         pbs = &(*pbs)->next;
 477     if (*pbs == bs)
 478         *pbs = bs->next;
 479
 480     bdrv_close(bs);
 481     qemu_free(bs);
 482 }
 483
 484 /* commit COW file into the raw image */
 485 int bdrv_commit(BlockDriverState *bs)
 486 {
 487     BlockDriver *drv = bs->drv;
 488     int64_t i, total_sectors;
 489     int n, j;
 490     unsigned char sector[512];
 491
 492     if (!drv)
 493         return -ENOMEDIUM;
 494
 495     if (bs->read_only) {
 496         return -EACCES;
 497     }
 498
 499     if (!bs->backing_hd) {
 500         return -ENOTSUP;
 501     }
 502
 503     total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 504     for (i = 0; i < total_sectors;) {
 505         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
 506             for(j = 0; j < n; j++) {
 507                 if (bdrv_read(bs, i, sector, 1) != 0) {
 508                     return -EIO;
 509                 }
 510
 511                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
 512                     return -EIO;
 513                 }
 514                 i++;
 515             }
 516         } else {
 517             i += n;
 518         }
 519     }
 520
 521     if (drv->bdrv_make_empty)
 522         return drv->bdrv_make_empty(bs);
 523
 524     return 0;
 525 }
 526
 527 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
 528                                    size_t size)
 529 {
 530     int64_t len;
 531
 532     if (!bdrv_is_inserted(bs))
 533         return -ENOMEDIUM;
 534
 535     if (bs->growable)
 536         return 0;
 537
 538     len = bdrv_getlength(bs);
 539
 540     if ((offset + size) > len)
 541         return -EIO;
 542
 543     return 0;
 544 }
 545
 546 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
 547                               int nb_sectors)
 548 {
 549     int64_t offset;
 550
 551     /* Deal with byte accesses */
 552     if (sector_num < 0)
 553         offset = -sector_num;
 554     else
 555         offset = sector_num * 512;
 556
 557     return bdrv_check_byte_request(bs, offset, nb_sectors * 512);
 558 }
 559
 560 /* return < 0 if error. See bdrv_write() for the return codes */
 561 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 562               uint8_t *buf, int nb_sectors)
 563 {
 564     BlockDriver *drv = bs->drv;
 565
 566     if (!drv)
 567         return -ENOMEDIUM;
 568     if (bdrv_check_request(bs, sector_num, nb_sectors))
 569         return -EIO;
 570
 571     return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 572 }
 573
 574 /* Return < 0 if error. Important errors are:
 575   -EIO         generic I/O error (may happen for all errors)
 576   -ENOMEDIUM   No media inserted.
 577   -EINVAL      Invalid sector number or nb_sectors
 578   -EACCES      Trying to write a read-only device
 579 */
 580 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 581                const uint8_t *buf, int nb_sectors)
 582 {
 583     BlockDriver *drv = bs->drv;
 584     if (!bs->drv)
 585         return -ENOMEDIUM;
 586     if (bs->read_only)
 587         return -EACCES;
 588     if (bdrv_check_request(bs, sector_num, nb_sectors))
 589         return -EIO;
 590
 591     return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 592 }
 593
 594 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 595                void *buf, int count1)
 596 {
 597     uint8_t tmp_buf[SECTOR_SIZE];
 598     int len, nb_sectors, count;
 599     int64_t sector_num;
 600
 601     count = count1;
 602     /* first read to align to sector start */
 603     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 604     if (len > count)
 605         len = count;
 606     sector_num = offset >> SECTOR_BITS;
 607     if (len > 0) {
 608         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 609             return -EIO;
 610         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
 611         count -= len;
 612         if (count == 0)
 613             return count1;
 614         sector_num++;
 615         buf += len;
 616     }
 617
 618     /* read the sectors "in place" */
 619     nb_sectors = count >> SECTOR_BITS;
 620     if (nb_sectors > 0) {
 621         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
 622             return -EIO;
 623         sector_num += nb_sectors;
 624         len = nb_sectors << SECTOR_BITS;
 625         buf += len;
 626         count -= len;
 627     }
 628
 629     /* add data from the last sector */
 630     if (count > 0) {
 631         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 632             return -EIO;
 633         memcpy(buf, tmp_buf, count);
 634     }
 635     return count1;
 636 }
 637
 638 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 639                 const void *buf, int count1)
 640 {
 641     uint8_t tmp_buf[SECTOR_SIZE];
 642     int len, nb_sectors, count;
 643     int64_t sector_num;
 644
 645     count = count1;
 646     /* first write to align to sector start */
 647     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 648     if (len > count)
 649         len = count;
 650     sector_num = offset >> SECTOR_BITS;
 651     if (len > 0) {
 652         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 653             return -EIO;
 654         memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
 655         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 656             return -EIO;
 657         count -= len;
 658         if (count == 0)
 659             return count1;
 660         sector_num++;
 661         buf += len;
 662     }
 663
 664     /* write the sectors "in place" */
 665     nb_sectors = count >> SECTOR_BITS;
 666     if (nb_sectors > 0) {
 667         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
 668             return -EIO;
 669         sector_num += nb_sectors;
 670         len = nb_sectors << SECTOR_BITS;
 671         buf += len;
 672         count -= len;
 673     }
 674
 675     /* add data from the last sector */
 676     if (count > 0) {
 677         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 678             return -EIO;
 679         memcpy(tmp_buf, buf, count);
 680         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 681             return -EIO;
 682     }
 683     return count1;
 684 }
 685
 686 /**
 687  * Truncate file to 'offset' bytes (needed only for file protocols)
 688  */
 689 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
 690 {
 691     BlockDriver *drv = bs->drv;
 692     if (!drv)
 693         return -ENOMEDIUM;
 694     if (!drv->bdrv_truncate)
 695         return -ENOTSUP;
 696     return drv->bdrv_truncate(bs, offset);
 697 }
 698
 699 /**
 700  * Length of a file in bytes. Return < 0 if error or unknown.
 701  */
 702 int64_t bdrv_getlength(BlockDriverState *bs)
 703 {
 704     BlockDriver *drv = bs->drv;
 705     if (!drv)
 706         return -ENOMEDIUM;
 707     if (!drv->bdrv_getlength) {
 708         /* legacy mode */
 709         return bs->total_sectors * SECTOR_SIZE;
 710     }
 711     return drv->bdrv_getlength(bs);
 712 }
 713
 714 /* return 0 as number of sectors if no device present or error */
 715 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
 716 {
 717     int64_t length;
 718     length = bdrv_getlength(bs);
 719     if (length < 0)
 720         length = 0;
 721     else
 722         length = length >> SECTOR_BITS;
 723     *nb_sectors_ptr = length;
 724 }
 725
 726 struct partition {
 727         uint8_t boot_ind;           /* 0x80 - active */
 728         uint8_t head;               /* starting head */
 729         uint8_t sector;             /* starting sector */
 730         uint8_t cyl;                /* starting cylinder */
 731         uint8_t sys_ind;            /* What partition type */
 732         uint8_t end_head;           /* end head */
 733         uint8_t end_sector;         /* end sector */
 734         uint8_t end_cyl;            /* end cylinder */
 735         uint32_t start_sect;        /* starting sector counting from 0 */
 736         uint32_t nr_sects;          /* nr of sectors in partition */
 737 } __attribute__((packed));
 738
 739 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
 740 static int guess_disk_lchs(BlockDriverState *bs,
 741                            int *pcylinders, int *pheads, int *psectors)
 742 {
 743     uint8_t buf[512];
 744     int ret, i, heads, sectors, cylinders;
 745     struct partition *p;
 746     uint32_t nr_sects;
 747     uint64_t nb_sectors;
 748
 749     bdrv_get_geometry(bs, &nb_sectors);
 750
 751     ret = bdrv_read(bs, 0, buf, 1);
 752     if (ret < 0)
 753         return -1;
 754     /* test msdos magic */
 755     if (buf[510] != 0x55 || buf[511] != 0xaa)
 756         return -1;
 757     for(i = 0; i < 4; i++) {
 758         p = ((struct partition *)(buf + 0x1be)) + i;
 759         nr_sects = le32_to_cpu(p->nr_sects);
 760         if (nr_sects && p->end_head) {
 761             /* We make the assumption that the partition terminates on
 762                a cylinder boundary */
 763             heads = p->end_head + 1;
 764             sectors = p->end_sector & 63;
 765             if (sectors == 0)
 766                 continue;
 767             cylinders = nb_sectors / (heads * sectors);
 768             if (cylinders < 1 || cylinders > 16383)
 769                 continue;
 770             *pheads = heads;
 771             *psectors = sectors;
 772             *pcylinders = cylinders;
 773 #if 0
 774             printf("guessed geometry: LCHS=%d %d %d\n",
 775                    cylinders, heads, sectors);
 776 #endif
 777             return 0;
 778         }
 779     }
 780     return -1;
 781 }
 782
 783 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
 784 {
 785     int translation, lba_detected = 0;
 786     int cylinders, heads, secs;
 787     uint64_t nb_sectors;
 788
 789     /* if a geometry hint is available, use it */
 790     bdrv_get_geometry(bs, &nb_sectors);
 791     bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
 792     translation = bdrv_get_translation_hint(bs);
 793     if (cylinders != 0) {
 794         *pcyls = cylinders;
 795         *pheads = heads;
 796         *psecs = secs;
 797     } else {
 798         if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
 799             if (heads > 16) {
 800                 /* if heads > 16, it means that a BIOS LBA
 801                    translation was active, so the default
 802                    hardware geometry is OK */
 803                 lba_detected = 1;
 804                 goto default_geometry;
 805             } else {
 806                 *pcyls = cylinders;
 807                 *pheads = heads;
 808                 *psecs = secs;
 809                 /* disable any translation to be in sync with
 810                    the logical geometry */
 811                 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
 812                     bdrv_set_translation_hint(bs,
 813                                               BIOS_ATA_TRANSLATION_NONE);
 814                 }
 815             }
 816         } else {
 817         default_geometry:
 818             /* if no geometry, use a standard physical disk geometry */
 819             cylinders = nb_sectors / (16 * 63);
 820
 821             if (cylinders > 16383)
 822                 cylinders = 16383;
 823             else if (cylinders < 2)
 824                 cylinders = 2;
 825             *pcyls = cylinders;
 826             *pheads = 16;
 827             *psecs = 63;
 828             if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
 829                 if ((*pcyls * *pheads) <= 131072) {
 830                     bdrv_set_translation_hint(bs,
 831                                               BIOS_ATA_TRANSLATION_LARGE);
 832                 } else {
 833                     bdrv_set_translation_hint(bs,
 834                                               BIOS_ATA_TRANSLATION_LBA);
 835                 }
 836             }
 837         }
 838         bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
 839     }
 840 }
 841
 842 void bdrv_set_geometry_hint(BlockDriverState *bs,
 843                             int cyls, int heads, int secs)
 844 {
 845     bs->cyls = cyls;
 846     bs->heads = heads;
 847     bs->secs = secs;
 848 }
 849
 850 void bdrv_set_type_hint(BlockDriverState *bs, int type)
 851 {
 852     bs->type = type;
 853     bs->removable = ((type == BDRV_TYPE_CDROM ||
 854                       type == BDRV_TYPE_FLOPPY));
 855 }
 856
 857 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
 858 {
 859     bs->translation = translation;
 860 }
 861
 862 void bdrv_get_geometry_hint(BlockDriverState *bs,
 863                             int *pcyls, int *pheads, int *psecs)
 864 {
 865     *pcyls = bs->cyls;
 866     *pheads = bs->heads;
 867     *psecs = bs->secs;
 868 }
 869
 870 int bdrv_get_type_hint(BlockDriverState *bs)
 871 {
 872     return bs->type;
 873 }
 874
 875 int bdrv_get_translation_hint(BlockDriverState *bs)
 876 {
 877     return bs->translation;
 878 }
 879
 880 int bdrv_is_removable(BlockDriverState *bs)
 881 {
 882     return bs->removable;
 883 }
 884
 885 int bdrv_is_read_only(BlockDriverState *bs)
 886 {
 887     return bs->read_only;
 888 }
 889
 890 int bdrv_is_sg(BlockDriverState *bs)
 891 {
 892     return bs->sg;
 893 }
 894
 895 /* XXX: no longer used */
 896 void bdrv_set_change_cb(BlockDriverState *bs,
 897                         void (*change_cb)(void *opaque), void *opaque)
 898 {
 899     bs->change_cb = change_cb;
 900     bs->change_opaque = opaque;
 901 }
 902
 903 int bdrv_is_encrypted(BlockDriverState *bs)
 904 {
 905     if (bs->backing_hd && bs->backing_hd->encrypted)
 906         return 1;
 907     return bs->encrypted;
 908 }
 909
 910 int bdrv_key_required(BlockDriverState *bs)
 911 {
 912     BlockDriverState *backing_hd = bs->backing_hd;
 913
 914     if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
 915         return 1;
 916     return (bs->encrypted && !bs->valid_key);
 917 }
 918
 919 int bdrv_set_key(BlockDriverState *bs, const char *key)
 920 {
 921     int ret;
 922     if (bs->backing_hd && bs->backing_hd->encrypted) {
 923         ret = bdrv_set_key(bs->backing_hd, key);
 924         if (ret < 0)
 925             return ret;
 926         if (!bs->encrypted)
 927             return 0;
 928     }
 929     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
 930         return -1;
 931     ret = bs->drv->bdrv_set_key(bs, key);
 932     if (ret < 0) {
 933         bs->valid_key = 0;
 934     } else if (!bs->valid_key) {
 935         bs->valid_key = 1;
 936         /* call the change callback now, we skipped it on open */
 937         bs->media_changed = 1;
 938         if (bs->change_cb)
 939             bs->change_cb(bs->change_opaque);
 940     }
 941     return ret;
 942 }
 943
 944 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
 945 {
 946     if (!bs->drv) {
 947         buf[0] = '\0';
 948     } else {
 949         pstrcpy(buf, buf_size, bs->drv->format_name);
 950     }
 951 }
 952
 953 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
 954                          void *opaque)
 955 {
 956     BlockDriver *drv;
 957
 958     for (drv = first_drv; drv != NULL; drv = drv->next) {
 959         it(opaque, drv->format_name);
 960     }
 961 }
 962
 963 BlockDriverState *bdrv_find(const char *name)
 964 {
 965     BlockDriverState *bs;
 966
 967     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 968         if (!strcmp(name, bs->device_name))
 969             return bs;
 970     }
 971     return NULL;
 972 }
 973
 974 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
 975 {
 976     BlockDriverState *bs;
 977
 978     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 979         it(opaque, bs);
 980     }
 981 }
 982
 983 const char *bdrv_get_device_name(BlockDriverState *bs)
 984 {
 985     return bs->device_name;
 986 }
 987
 988 void bdrv_flush(BlockDriverState *bs)
 989 {
 990     if (bs->drv->bdrv_flush)
 991         bs->drv->bdrv_flush(bs);
 992     if (bs->backing_hd)
 993         bdrv_flush(bs->backing_hd);
 994 }
 995
 996 void bdrv_flush_all(void)
 997 {
 998     BlockDriverState *bs;
 999
1000     for (bs = bdrv_first; bs != NULL; bs = bs->next)
1001         if (bs->drv && !bdrv_is_read_only(bs) &&
1002             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1003             bdrv_flush(bs);
1004 }
1005
1006 /*
1007  * Returns true iff the specified sector is present in the disk image. Drivers
1008  * not implementing the functionality are assumed to not support backing files,
1009  * hence all their sectors are reported as allocated.
1010  *
1011  * 'pnum' is set to the number of sectors (including and immediately following
1012  * the specified sector) that are known to be in the same
1013  * allocated/unallocated state.
1014  *
1015  * 'nb_sectors' is the max value 'pnum' should be set to.
1016  */
1017 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1018         int *pnum)
1019 {
1020     int64_t n;
1021     if (!bs->drv->bdrv_is_allocated) {
1022         if (sector_num >= bs->total_sectors) {
1023             *pnum = 0;
1024             return 0;
1025         }
1026         n = bs->total_sectors - sector_num;
1027         *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1028         return 1;
1029     }
1030     return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1031 }
1032
1033 void bdrv_info(Monitor *mon)
1034 {
1035     BlockDriverState *bs;
1036
1037     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1038         monitor_printf(mon, "%s:", bs->device_name);
1039         monitor_printf(mon, " type=");
1040         switch(bs->type) {
1041         case BDRV_TYPE_HD:
1042             monitor_printf(mon, "hd");
1043             break;
1044         case BDRV_TYPE_CDROM:
1045             monitor_printf(mon, "cdrom");
1046             break;
1047         case BDRV_TYPE_FLOPPY:
1048             monitor_printf(mon, "floppy");
1049             break;
1050         }
1051         monitor_printf(mon, " removable=%d", bs->removable);
1052         if (bs->removable) {
1053             monitor_printf(mon, " locked=%d", bs->locked);
1054         }
1055         if (bs->drv) {
1056             monitor_printf(mon, " file=");
1057             monitor_print_filename(mon, bs->filename);
1058             if (bs->backing_file[0] != '\0') {
1059                 monitor_printf(mon, " backing_file=");
1060                 monitor_print_filename(mon, bs->backing_file);
1061             }
1062             monitor_printf(mon, " ro=%d", bs->read_only);
1063             monitor_printf(mon, " drv=%s", bs->drv->format_name);
1064             monitor_printf(mon, " encrypted=%d", bdrv_is_encrypted(bs));
1065         } else {
1066             monitor_printf(mon, " [not inserted]");
1067         }
1068         monitor_printf(mon, "\n");
1069     }
1070 }
1071
1072 /* The "info blockstats" command. */
1073 void bdrv_info_stats(Monitor *mon)
1074 {
1075     BlockDriverState *bs;
1076
1077     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1078         monitor_printf(mon, "%s:"
1079                        " rd_bytes=%" PRIu64
1080                        " wr_bytes=%" PRIu64
1081                        " rd_operations=%" PRIu64
1082                        " wr_operations=%" PRIu64
1083                        "\n",
1084                        bs->device_name,
1085                        bs->rd_bytes, bs->wr_bytes,
1086                        bs->rd_ops, bs->wr_ops);
1087     }
1088 }
1089
1090 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1091 {
1092     if (bs->backing_hd && bs->backing_hd->encrypted)
1093         return bs->backing_file;
1094     else if (bs->encrypted)
1095         return bs->filename;
1096     else
1097         return NULL;
1098 }
1099
1100 void bdrv_get_backing_filename(BlockDriverState *bs,
1101                                char *filename, int filename_size)
1102 {
1103     if (!bs->backing_hd) {
1104         pstrcpy(filename, filename_size, "");
1105     } else {
1106         pstrcpy(filename, filename_size, bs->backing_file);
1107     }
1108 }
1109
1110 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1111                           const uint8_t *buf, int nb_sectors)
1112 {
1113     BlockDriver *drv = bs->drv;
1114     if (!drv)
1115         return -ENOMEDIUM;
1116     if (!drv->bdrv_write_compressed)
1117         return -ENOTSUP;
1118     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1119 }
1120
1121 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1122 {
1123     BlockDriver *drv = bs->drv;
1124     if (!drv)
1125         return -ENOMEDIUM;
1126     if (!drv->bdrv_get_info)
1127         return -ENOTSUP;
1128     memset(bdi, 0, sizeof(*bdi));
1129     return drv->bdrv_get_info(bs, bdi);
1130 }
1131
1132 /**************************************************************/
1133 /* handling of snapshots */
1134
1135 int bdrv_snapshot_create(BlockDriverState *bs,
1136                          QEMUSnapshotInfo *sn_info)
1137 {
1138     BlockDriver *drv = bs->drv;
1139     if (!drv)
1140         return -ENOMEDIUM;
1141     if (!drv->bdrv_snapshot_create)
1142         return -ENOTSUP;
1143     return drv->bdrv_snapshot_create(bs, sn_info);
1144 }
1145
1146 int bdrv_snapshot_goto(BlockDriverState *bs,
1147                        const char *snapshot_id)
1148 {
1149     BlockDriver *drv = bs->drv;
1150     if (!drv)
1151         return -ENOMEDIUM;
1152     if (!drv->bdrv_snapshot_goto)
1153         return -ENOTSUP;
1154     return drv->bdrv_snapshot_goto(bs, snapshot_id);
1155 }
1156
1157 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1158 {
1159     BlockDriver *drv = bs->drv;
1160     if (!drv)
1161         return -ENOMEDIUM;
1162     if (!drv->bdrv_snapshot_delete)
1163         return -ENOTSUP;
1164     return drv->bdrv_snapshot_delete(bs, snapshot_id);
1165 }
1166
1167 int bdrv_snapshot_list(BlockDriverState *bs,
1168                        QEMUSnapshotInfo **psn_info)
1169 {
1170     BlockDriver *drv = bs->drv;
1171     if (!drv)
1172         return -ENOMEDIUM;
1173     if (!drv->bdrv_snapshot_list)
1174         return -ENOTSUP;
1175     return drv->bdrv_snapshot_list(bs, psn_info);
1176 }
1177
1178 #define NB_SUFFIXES 4
1179
1180 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1181 {
1182     static const char suffixes[NB_SUFFIXES] = "KMGT";
1183     int64_t base;
1184     int i;
1185
1186     if (size <= 999) {
1187         snprintf(buf, buf_size, "%" PRId64, size);
1188     } else {
1189         base = 1024;
1190         for(i = 0; i < NB_SUFFIXES; i++) {
1191             if (size < (10 * base)) {
1192                 snprintf(buf, buf_size, "%0.1f%c",
1193                          (double)size / base,
1194                          suffixes[i]);
1195                 break;
1196             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1197                 snprintf(buf, buf_size, "%" PRId64 "%c",
1198                          ((size + (base >> 1)) / base),
1199                          suffixes[i]);
1200                 break;
1201             }
1202             base = base * 1024;
1203         }
1204     }
1205     return buf;
1206 }
1207
1208 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1209 {
1210     char buf1[128], date_buf[128], clock_buf[128];
1211 #ifdef _WIN32
1212     struct tm *ptm;
1213 #else
1214     struct tm tm;
1215 #endif
1216     time_t ti;
1217     int64_t secs;
1218
1219     if (!sn) {
1220         snprintf(buf, buf_size,
1221                  "%-10s%-20s%7s%20s%15s",
1222                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1223     } else {
1224         ti = sn->date_sec;
1225 #ifdef _WIN32
1226         ptm = localtime(&ti);
1227         strftime(date_buf, sizeof(date_buf),
1228                  "%Y-%m-%d %H:%M:%S", ptm);
1229 #else
1230         localtime_r(&ti, &tm);
1231         strftime(date_buf, sizeof(date_buf),
1232                  "%Y-%m-%d %H:%M:%S", &tm);
1233 #endif
1234         secs = sn->vm_clock_nsec / 1000000000;
1235         snprintf(clock_buf, sizeof(clock_buf),
1236                  "%02d:%02d:%02d.%03d",
1237                  (int)(secs / 3600),
1238                  (int)((secs / 60) % 60),
1239                  (int)(secs % 60),
1240                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1241         snprintf(buf, buf_size,
1242                  "%-10s%-20s%7s%20s%15s",
1243                  sn->id_str, sn->name,
1244                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1245                  date_buf,
1246                  clock_buf);
1247     }
1248     return buf;
1249 }
1250
1251
1252 /**************************************************************/
1253 /* async I/Os */
1254
1255 typedef struct VectorTranslationState {
1256     QEMUIOVector *iov;
1257     uint8_t *bounce;
1258     int is_write;
1259     BlockDriverAIOCB *aiocb;
1260     BlockDriverAIOCB *this_aiocb;
1261 } VectorTranslationState;
1262
1263 static void bdrv_aio_rw_vector_cb(void *opaque, int ret)
1264 {
1265     VectorTranslationState *s = opaque;
1266
1267     if (!s->is_write) {
1268         qemu_iovec_from_buffer(s->iov, s->bounce, s->iov->size);
1269     }
1270     qemu_vfree(s->bounce);
1271     s->this_aiocb->cb(s->this_aiocb->opaque, ret);
1272     qemu_aio_release(s->this_aiocb);
1273 }
1274
1275 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
1276                                             int64_t sector_num,
1277                                             QEMUIOVector *iov,
1278                                             int nb_sectors,
1279                                             BlockDriverCompletionFunc *cb,
1280                                             void *opaque,
1281                                             int is_write)
1282
1283 {
1284     VectorTranslationState *s = qemu_mallocz(sizeof(*s));
1285     BlockDriverAIOCB *aiocb = qemu_aio_get(bs, cb, opaque);
1286
1287     s->this_aiocb = aiocb;
1288     s->iov = iov;
1289     s->bounce = qemu_memalign(512, nb_sectors * 512);
1290     s->is_write = is_write;
1291     if (is_write) {
1292         qemu_iovec_to_buffer(s->iov, s->bounce);
1293         s->aiocb = bdrv_aio_write(bs, sector_num, s->bounce, nb_sectors,
1294                                   bdrv_aio_rw_vector_cb, s);
1295     } else {
1296         s->aiocb = bdrv_aio_read(bs, sector_num, s->bounce, nb_sectors,
1297                                  bdrv_aio_rw_vector_cb, s);
1298     }
1299     return aiocb;
1300 }
1301
1302 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1303                                  QEMUIOVector *iov, int nb_sectors,
1304                                  BlockDriverCompletionFunc *cb, void *opaque)
1305 {
1306     if (bdrv_check_request(bs, sector_num, nb_sectors))
1307         return NULL;
1308
1309     return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,
1310                               cb, opaque, 0);
1311 }
1312
1313 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1314                                   QEMUIOVector *iov, int nb_sectors,
1315                                   BlockDriverCompletionFunc *cb, void *opaque)
1316 {
1317     if (bdrv_check_request(bs, sector_num, nb_sectors))
1318         return NULL;
1319
1320     return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,
1321                               cb, opaque, 1);
1322 }
1323
1324 BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
1325                                 uint8_t *buf, int nb_sectors,
1326                                 BlockDriverCompletionFunc *cb, void *opaque)
1327 {
1328     BlockDriver *drv = bs->drv;
1329     BlockDriverAIOCB *ret;
1330
1331     if (!drv)
1332         return NULL;
1333     if (bdrv_check_request(bs, sector_num, nb_sectors))
1334         return NULL;
1335
1336     ret = drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
1337
1338     if (ret) {
1339         /* Update stats even though technically transfer has not happened. */
1340         bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1341         bs->rd_ops ++;
1342     }
1343
1344     return ret;
1345 }
1346
1347 BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
1348                                  const uint8_t *buf, int nb_sectors,
1349                                  BlockDriverCompletionFunc *cb, void *opaque)
1350 {
1351     BlockDriver *drv = bs->drv;
1352     BlockDriverAIOCB *ret;
1353
1354     if (!drv)
1355         return NULL;
1356     if (bs->read_only)
1357         return NULL;
1358     if (bdrv_check_request(bs, sector_num, nb_sectors))
1359         return NULL;
1360
1361     ret = drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
1362
1363     if (ret) {
1364         /* Update stats even though technically transfer has not happened. */
1365         bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1366         bs->wr_ops ++;
1367     }
1368
1369     return ret;
1370 }
1371
1372 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1373 {
1374     BlockDriver *drv = acb->bs->drv;
1375
1376     if (acb->cb == bdrv_aio_rw_vector_cb) {
1377         VectorTranslationState *s = acb->opaque;
1378         acb = s->aiocb;
1379     }
1380
1381     drv->bdrv_aio_cancel(acb);
1382 }
1383
1384
1385 /**************************************************************/
1386 /* async block device emulation */
1387
1388 static void bdrv_aio_bh_cb(void *opaque)
1389 {
1390     BlockDriverAIOCBSync *acb = opaque;
1391     acb->common.cb(acb->common.opaque, acb->ret);
1392     qemu_aio_release(acb);
1393 }
1394
1395 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1396         int64_t sector_num, uint8_t *buf, int nb_sectors,
1397         BlockDriverCompletionFunc *cb, void *opaque)
1398 {
1399     BlockDriverAIOCBSync *acb;
1400     int ret;
1401
1402     acb = qemu_aio_get(bs, cb, opaque);
1403     if (!acb->bh)
1404         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1405     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1406     acb->ret = ret;
1407     qemu_bh_schedule(acb->bh);
1408     return &acb->common;
1409 }
1410
1411 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1412         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1413         BlockDriverCompletionFunc *cb, void *opaque)
1414 {
1415     BlockDriverAIOCBSync *acb;
1416     int ret;
1417
1418     acb = qemu_aio_get(bs, cb, opaque);
1419     if (!acb->bh)
1420         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1421     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1422     acb->ret = ret;
1423     qemu_bh_schedule(acb->bh);
1424     return &acb->common;
1425 }
1426
1427 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1428 {
1429     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1430     qemu_bh_cancel(acb->bh);
1431     qemu_aio_release(acb);
1432 }
1433
1434 /**************************************************************/
1435 /* sync block device emulation */
1436
1437 static void bdrv_rw_em_cb(void *opaque, int ret)
1438 {
1439     *(int *)opaque = ret;
1440 }
1441
1442 #define NOT_DONE 0x7fffffff
1443
1444 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1445                         uint8_t *buf, int nb_sectors)
1446 {
1447     int async_ret;
1448     BlockDriverAIOCB *acb;
1449
1450     async_ret = NOT_DONE;
1451     acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
1452                         bdrv_rw_em_cb, &async_ret);
1453     if (acb == NULL)
1454         return -1;
1455
1456     while (async_ret == NOT_DONE) {
1457         qemu_aio_wait();
1458     }
1459
1460     return async_ret;
1461 }
1462
1463 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1464                          const uint8_t *buf, int nb_sectors)
1465 {
1466     int async_ret;
1467     BlockDriverAIOCB *acb;
1468
1469     async_ret = NOT_DONE;
1470     acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
1471                          bdrv_rw_em_cb, &async_ret);
1472     if (acb == NULL)
1473         return -1;
1474     while (async_ret == NOT_DONE) {
1475         qemu_aio_wait();
1476     }
1477     return async_ret;
1478 }
1479
1480 void bdrv_init(void)
1481 {
1482     bdrv_register(&bdrv_raw);
1483     bdrv_register(&bdrv_host_device);
1484 #ifndef _WIN32
1485     bdrv_register(&bdrv_cow);
1486 #endif
1487     bdrv_register(&bdrv_qcow);
1488     bdrv_register(&bdrv_vmdk);
1489     bdrv_register(&bdrv_cloop);
1490     bdrv_register(&bdrv_dmg);
1491     bdrv_register(&bdrv_bochs);
1492     bdrv_register(&bdrv_vpc);
1493     bdrv_register(&bdrv_vvfat);
1494     bdrv_register(&bdrv_qcow2);
1495     bdrv_register(&bdrv_parallels);
1496     bdrv_register(&bdrv_nbd);
1497 }
1498
1499 void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
1500                    void *opaque)
1501 {
1502     BlockDriver *drv;
1503     BlockDriverAIOCB *acb;
1504
1505     drv = bs->drv;
1506     if (drv->free_aiocb) {
1507         acb = drv->free_aiocb;
1508         drv->free_aiocb = acb->next;
1509     } else {
1510         acb = qemu_mallocz(drv->aiocb_size);
1511     }
1512     acb->bs = bs;
1513     acb->cb = cb;
1514     acb->opaque = opaque;
1515     return acb;
1516 }
1517
1518 void qemu_aio_release(void *p)
1519 {
1520     BlockDriverAIOCB *acb = p;
1521     BlockDriver *drv = acb->bs->drv;
1522     acb->next = drv->free_aiocb;
1523     drv->free_aiocb = acb;
1524 }
1525
1526 /**************************************************************/
1527 /* removable device support */
1528
1529 /**
1530  * Return TRUE if the media is present
1531  */
1532 int bdrv_is_inserted(BlockDriverState *bs)
1533 {
1534     BlockDriver *drv = bs->drv;
1535     int ret;
1536     if (!drv)
1537         return 0;
1538     if (!drv->bdrv_is_inserted)
1539         return 1;
1540     ret = drv->bdrv_is_inserted(bs);
1541     return ret;
1542 }
1543
1544 /**
1545  * Return TRUE if the media changed since the last call to this
1546  * function. It is currently only used for floppy disks
1547  */
1548 int bdrv_media_changed(BlockDriverState *bs)
1549 {
1550     BlockDriver *drv = bs->drv;
1551     int ret;
1552
1553     if (!drv || !drv->bdrv_media_changed)
1554         ret = -ENOTSUP;
1555     else
1556         ret = drv->bdrv_media_changed(bs);
1557     if (ret == -ENOTSUP)
1558         ret = bs->media_changed;
1559     bs->media_changed = 0;
1560     return ret;
1561 }
1562
1563 /**
1564  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1565  */
1566 void bdrv_eject(BlockDriverState *bs, int eject_flag)
1567 {
1568     BlockDriver *drv = bs->drv;
1569     int ret;
1570
1571     if (!drv || !drv->bdrv_eject) {
1572         ret = -ENOTSUP;
1573     } else {
1574         ret = drv->bdrv_eject(bs, eject_flag);
1575     }
1576     if (ret == -ENOTSUP) {
1577         if (eject_flag)
1578             bdrv_close(bs);
1579     }
1580 }
1581
1582 int bdrv_is_locked(BlockDriverState *bs)
1583 {
1584     return bs->locked;
1585 }
1586
1587 /**
1588  * Lock or unlock the media (if it is locked, the user won't be able
1589  * to eject it manually).
1590  */
1591 void bdrv_set_locked(BlockDriverState *bs, int locked)
1592 {
1593     BlockDriver *drv = bs->drv;
1594
1595     bs->locked = locked;
1596     if (drv && drv->bdrv_set_locked) {
1597         drv->bdrv_set_locked(bs, locked);
1598     }
1599 }
1600
1601 /* needed for generic scsi interface */
1602
1603 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1604 {
1605     BlockDriver *drv = bs->drv;
1606
1607     if (drv && drv->bdrv_ioctl)
1608         return drv->bdrv_ioctl(bs, req, buf);
1609     return -ENOTSUP;
1610 }
1611
1612 int bdrv_sg_send_command(BlockDriverState *bs, void *buf, int count)
1613 {
1614     return bs->drv->bdrv_sg_send_command(bs, buf, count);
1615 }
1616
1617 int bdrv_sg_recv_response(BlockDriverState *bs, void *buf, int count)
1618 {
1619     return bs->drv->bdrv_sg_recv_response(bs, buf, count);
1620 }
1621
1622 BlockDriverAIOCB *bdrv_sg_aio_read(BlockDriverState *bs, void *buf, int count,
1623                                    BlockDriverCompletionFunc *cb, void *opaque)
1624 {
1625     return bs->drv->bdrv_sg_aio_read(bs, buf, count, cb, opaque);
1626 }
1627
1628 BlockDriverAIOCB *bdrv_sg_aio_write(BlockDriverState *bs, void *buf, int count,
1629                                     BlockDriverCompletionFunc *cb, void *opaque)
1630 {
1631     return bs->drv->bdrv_sg_aio_write(bs, buf, count, cb, opaque);
1632 }