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