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