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