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