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 #ifdef QEMU_IMG
  25 #include "qemu-common.h"
  26 #else
  27 #include "vl.h"
  28 #endif
  29 #include "block_int.h"
  30
  31 #ifdef _BSD
  32 #include <sys/types.h>
  33 #include <sys/stat.h>
  34 #include <sys/ioctl.h>
  35 #include <sys/queue.h>
  36 #include <sys/disk.h>
  37 #endif
  38
  39 #define SECTOR_BITS 9
  40 #define SECTOR_SIZE (1 << SECTOR_BITS)
  41
  42 typedef struct BlockDriverAIOCBSync {
  43     BlockDriverAIOCB common;
  44     QEMUBH *bh;
  45     int ret;
  46 } BlockDriverAIOCBSync;
  47
  48 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
  49         int64_t sector_num, uint8_t *buf, int nb_sectors,
  50         BlockDriverCompletionFunc *cb, void *opaque);
  51 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
  52         int64_t sector_num, const uint8_t *buf, int nb_sectors,
  53         BlockDriverCompletionFunc *cb, void *opaque);
  54 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
  55 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
  56                         uint8_t *buf, int nb_sectors);
  57 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
  58                          const uint8_t *buf, int nb_sectors);
  59
  60 BlockDriverState *bdrv_first;
  61 static BlockDriver *first_drv;
  62
  63 int path_is_absolute(const char *path)
  64 {
  65     const char *p;
  66 #ifdef _WIN32
  67     /* specific case for names like: "\\.\d:" */
  68     if (*path == '/' || *path == '\\')
  69         return 1;
  70 #endif
  71     p = strchr(path, ':');
  72     if (p)
  73         p++;
  74     else
  75         p = path;
  76 #ifdef _WIN32
  77     return (*p == '/' || *p == '\\');
  78 #else
  79     return (*p == '/');
  80 #endif
  81 }
  82
  83 /* if filename is absolute, just copy it to dest. Otherwise, build a
  84    path to it by considering it is relative to base_path. URL are
  85    supported. */
  86 void path_combine(char *dest, int dest_size,
  87                   const char *base_path,
  88                   const char *filename)
  89 {
  90     const char *p, *p1;
  91     int len;
  92
  93     if (dest_size <= 0)
  94         return;
  95     if (path_is_absolute(filename)) {
  96         pstrcpy(dest, dest_size, filename);
  97     } else {
  98         p = strchr(base_path, ':');
  99         if (p)
 100             p++;
 101         else
 102             p = base_path;
 103         p1 = strrchr(base_path, '/');
 104 #ifdef _WIN32
 105         {
 106             const char *p2;
 107             p2 = strrchr(base_path, '\\');
 108             if (!p1 || p2 > p1)
 109                 p1 = p2;
 110         }
 111 #endif
 112         if (p1)
 113             p1++;
 114         else
 115             p1 = base_path;
 116         if (p1 > p)
 117             p = p1;
 118         len = p - base_path;
 119         if (len > dest_size - 1)
 120             len = dest_size - 1;
 121         memcpy(dest, base_path, len);
 122         dest[len] = '\0';
 123         pstrcat(dest, dest_size, filename);
 124     }
 125 }
 126
 127
 128 void bdrv_register(BlockDriver *bdrv)
 129 {
 130     if (!bdrv->bdrv_aio_read) {
 131         /* add AIO emulation layer */
 132         bdrv->bdrv_aio_read = bdrv_aio_read_em;
 133         bdrv->bdrv_aio_write = bdrv_aio_write_em;
 134         bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
 135         bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
 136     } else if (!bdrv->bdrv_read && !bdrv->bdrv_pread) {
 137         /* add synchronous IO emulation layer */
 138         bdrv->bdrv_read = bdrv_read_em;
 139         bdrv->bdrv_write = bdrv_write_em;
 140     }
 141     bdrv->next = first_drv;
 142     first_drv = bdrv;
 143 }
 144
 145 /* create a new block device (by default it is empty) */
 146 BlockDriverState *bdrv_new(const char *device_name)
 147 {
 148     BlockDriverState **pbs, *bs;
 149
 150     bs = qemu_mallocz(sizeof(BlockDriverState));
 151     if(!bs)
 152         return NULL;
 153     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
 154     if (device_name[0] != '\0') {
 155         /* insert at the end */
 156         pbs = &bdrv_first;
 157         while (*pbs != NULL)
 158             pbs = &(*pbs)->next;
 159         *pbs = bs;
 160     }
 161     return bs;
 162 }
 163
 164 BlockDriver *bdrv_find_format(const char *format_name)
 165 {
 166     BlockDriver *drv1;
 167     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 168         if (!strcmp(drv1->format_name, format_name))
 169             return drv1;
 170     }
 171     return NULL;
 172 }
 173
 174 int bdrv_create(BlockDriver *drv,
 175                 const char *filename, int64_t size_in_sectors,
 176                 const char *backing_file, int flags)
 177 {
 178     if (!drv->bdrv_create)
 179         return -ENOTSUP;
 180     return drv->bdrv_create(filename, size_in_sectors, backing_file, flags);
 181 }
 182
 183 #ifdef _WIN32
 184 void get_tmp_filename(char *filename, int size)
 185 {
 186     char temp_dir[MAX_PATH];
 187
 188     GetTempPath(MAX_PATH, temp_dir);
 189     GetTempFileName(temp_dir, "qem", 0, filename);
 190 }
 191 #else
 192 void get_tmp_filename(char *filename, int size)
 193 {
 194     int fd;
 195     /* XXX: race condition possible */
 196     pstrcpy(filename, size, "/tmp/vl.XXXXXX");
 197     fd = mkstemp(filename);
 198     close(fd);
 199 }
 200 #endif
 201
 202 #ifdef _WIN32
 203 static int is_windows_drive_prefix(const char *filename)
 204 {
 205     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
 206              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
 207             filename[1] == ':');
 208 }
 209
 210 static int is_windows_drive(const char *filename)
 211 {
 212     if (is_windows_drive_prefix(filename) &&
 213         filename[2] == '\0')
 214         return 1;
 215     if (strstart(filename, "\\\\.\\", NULL) ||
 216         strstart(filename, "//./", NULL))
 217         return 1;
 218     return 0;
 219 }
 220 #endif
 221
 222 static BlockDriver *find_protocol(const char *filename)
 223 {
 224     BlockDriver *drv1;
 225     char protocol[128];
 226     int len;
 227     const char *p;
 228
 229 #ifdef _WIN32
 230     if (is_windows_drive(filename) ||
 231         is_windows_drive_prefix(filename))
 232         return &bdrv_raw;
 233 #endif
 234     p = strchr(filename, ':');
 235     if (!p)
 236         return &bdrv_raw;
 237     len = p - filename;
 238     if (len > sizeof(protocol) - 1)
 239         len = sizeof(protocol) - 1;
 240     memcpy(protocol, filename, len);
 241     protocol[len] = '\0';
 242     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 243         if (drv1->protocol_name &&
 244             !strcmp(drv1->protocol_name, protocol))
 245             return drv1;
 246     }
 247     return NULL;
 248 }
 249
 250 /* XXX: force raw format if block or character device ? It would
 251    simplify the BSD case */
 252 static BlockDriver *find_image_format(const char *filename)
 253 {
 254     int ret, score, score_max;
 255     BlockDriver *drv1, *drv;
 256     uint8_t buf[2048];
 257     BlockDriverState *bs;
 258
 259     /* detect host devices. By convention, /dev/cdrom[N] is always
 260        recognized as a host CDROM */
 261     if (strstart(filename, "/dev/cdrom", NULL))
 262         return &bdrv_host_device;
 263 #ifdef _WIN32
 264     if (is_windows_drive(filename))
 265         return &bdrv_host_device;
 266 #else
 267     {
 268         struct stat st;
 269         if (stat(filename, &st) >= 0 &&
 270             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 271             return &bdrv_host_device;
 272         }
 273     }
 274 #endif
 275
 276     drv = find_protocol(filename);
 277     /* no need to test disk image formats for vvfat */
 278     if (drv == &bdrv_vvfat)
 279         return drv;
 280
 281     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
 282     if (ret < 0)
 283         return NULL;
 284     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
 285     bdrv_delete(bs);
 286     if (ret < 0) {
 287         return NULL;
 288     }
 289
 290     score_max = 0;
 291     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 292         if (drv1->bdrv_probe) {
 293             score = drv1->bdrv_probe(buf, ret, filename);
 294             if (score > score_max) {
 295                 score_max = score;
 296                 drv = drv1;
 297             }
 298         }
 299     }
 300     return drv;
 301 }
 302
 303 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
 304 {
 305     BlockDriverState *bs;
 306     int ret;
 307
 308     bs = bdrv_new("");
 309     if (!bs)
 310         return -ENOMEM;
 311     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
 312     if (ret < 0) {
 313         bdrv_delete(bs);
 314         return ret;
 315     }
 316     *pbs = bs;
 317     return 0;
 318 }
 319
 320 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
 321 {
 322     return bdrv_open2(bs, filename, flags, NULL);
 323 }
 324
 325 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
 326                BlockDriver *drv)
 327 {
 328     int ret, open_flags;
 329     char tmp_filename[PATH_MAX];
 330     char backing_filename[PATH_MAX];
 331
 332     bs->read_only = 0;
 333     bs->is_temporary = 0;
 334     bs->encrypted = 0;
 335
 336     if (flags & BDRV_O_SNAPSHOT) {
 337         BlockDriverState *bs1;
 338         int64_t total_size;
 339
 340         /* if snapshot, we create a temporary backing file and open it
 341            instead of opening 'filename' directly */
 342
 343         /* if there is a backing file, use it */
 344         bs1 = bdrv_new("");
 345         if (!bs1) {
 346             return -ENOMEM;
 347         }
 348         if (bdrv_open(bs1, filename, 0) < 0) {
 349             bdrv_delete(bs1);
 350             return -1;
 351         }
 352         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
 353         bdrv_delete(bs1);
 354
 355         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
 356         realpath(filename, backing_filename);
 357         if (bdrv_create(&bdrv_qcow2, tmp_filename,
 358                         total_size, backing_filename, 0) < 0) {
 359             return -1;
 360         }
 361         filename = tmp_filename;
 362         bs->is_temporary = 1;
 363     }
 364
 365     pstrcpy(bs->filename, sizeof(bs->filename), filename);
 366     if (flags & BDRV_O_FILE) {
 367         drv = find_protocol(filename);
 368         if (!drv)
 369             return -ENOENT;
 370     } else {
 371         if (!drv) {
 372             drv = find_image_format(filename);
 373             if (!drv)
 374                 return -1;
 375         }
 376     }
 377     bs->drv = drv;
 378     bs->opaque = qemu_mallocz(drv->instance_size);
 379     if (bs->opaque == NULL && drv->instance_size > 0)
 380         return -1;
 381     /* Note: for compatibility, we open disk image files as RDWR, and
 382        RDONLY as fallback */
 383     if (!(flags & BDRV_O_FILE))
 384         open_flags = BDRV_O_RDWR;
 385     else
 386         open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
 387     ret = drv->bdrv_open(bs, filename, open_flags);
 388     if (ret == -EACCES && !(flags & BDRV_O_FILE)) {
 389         ret = drv->bdrv_open(bs, filename, BDRV_O_RDONLY);
 390         bs->read_only = 1;
 391     }
 392     if (ret < 0) {
 393         qemu_free(bs->opaque);
 394         bs->opaque = NULL;
 395         bs->drv = NULL;
 396         return ret;
 397     }
 398     if (drv->bdrv_getlength) {
 399         bs->total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 400     }
 401 #ifndef _WIN32
 402     if (bs->is_temporary) {
 403         unlink(filename);
 404     }
 405 #endif
 406     if (bs->backing_file[0] != '\0') {
 407         /* if there is a backing file, use it */
 408         bs->backing_hd = bdrv_new("");
 409         if (!bs->backing_hd) {
 410         fail:
 411             bdrv_close(bs);
 412             return -ENOMEM;
 413         }
 414         path_combine(backing_filename, sizeof(backing_filename),
 415                      filename, bs->backing_file);
 416         if (bdrv_open(bs->backing_hd, backing_filename, 0) < 0)
 417             goto fail;
 418     }
 419
 420     /* call the change callback */
 421     bs->media_changed = 1;
 422     if (bs->change_cb)
 423         bs->change_cb(bs->change_opaque);
 424
 425     return 0;
 426 }
 427
 428 void bdrv_close(BlockDriverState *bs)
 429 {
 430     if (bs->drv) {
 431         if (bs->backing_hd)
 432             bdrv_delete(bs->backing_hd);
 433         bs->drv->bdrv_close(bs);
 434         qemu_free(bs->opaque);
 435 #ifdef _WIN32
 436         if (bs->is_temporary) {
 437             unlink(bs->filename);
 438         }
 439 #endif
 440         bs->opaque = NULL;
 441         bs->drv = NULL;
 442
 443         /* call the change callback */
 444         bs->media_changed = 1;
 445         if (bs->change_cb)
 446             bs->change_cb(bs->change_opaque);
 447     }
 448 }
 449
 450 void bdrv_delete(BlockDriverState *bs)
 451 {
 452     /* XXX: remove the driver list */
 453     bdrv_close(bs);
 454     qemu_free(bs);
 455 }
 456
 457 /* commit COW file into the raw image */
 458 int bdrv_commit(BlockDriverState *bs)
 459 {
 460     BlockDriver *drv = bs->drv;
 461     int64_t i, total_sectors;
 462     int n, j;
 463     unsigned char sector[512];
 464
 465     if (!drv)
 466         return -ENOMEDIUM;
 467
 468     if (bs->read_only) {
 469         return -EACCES;
 470     }
 471
 472     if (!bs->backing_hd) {
 473         return -ENOTSUP;
 474     }
 475
 476     total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 477     for (i = 0; i < total_sectors;) {
 478         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
 479             for(j = 0; j < n; j++) {
 480                 if (bdrv_read(bs, i, sector, 1) != 0) {
 481                     return -EIO;
 482                 }
 483
 484                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
 485                     return -EIO;
 486                 }
 487                 i++;
 488             }
 489         } else {
 490             i += n;
 491         }
 492     }
 493
 494     if (drv->bdrv_make_empty)
 495         return drv->bdrv_make_empty(bs);
 496
 497     return 0;
 498 }
 499
 500 /* return < 0 if error. See bdrv_write() for the return codes */
 501 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 502               uint8_t *buf, int nb_sectors)
 503 {
 504     BlockDriver *drv = bs->drv;
 505
 506     if (!drv)
 507         return -ENOMEDIUM;
 508
 509     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
 510             memcpy(buf, bs->boot_sector_data, 512);
 511         sector_num++;
 512         nb_sectors--;
 513         buf += 512;
 514         if (nb_sectors == 0)
 515             return 0;
 516     }
 517     if (drv->bdrv_pread) {
 518         int ret, len;
 519         len = nb_sectors * 512;
 520         ret = drv->bdrv_pread(bs, sector_num * 512, buf, len);
 521         if (ret < 0)
 522             return ret;
 523         else if (ret != len)
 524             return -EINVAL;
 525         else
 526             return 0;
 527     } else {
 528         return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 529     }
 530 }
 531
 532 /* Return < 0 if error. Important errors are:
 533   -EIO         generic I/O error (may happen for all errors)
 534   -ENOMEDIUM   No media inserted.
 535   -EINVAL      Invalid sector number or nb_sectors
 536   -EACCES      Trying to write a read-only device
 537 */
 538 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 539                const uint8_t *buf, int nb_sectors)
 540 {
 541     BlockDriver *drv = bs->drv;
 542     if (!bs->drv)
 543         return -ENOMEDIUM;
 544     if (bs->read_only)
 545         return -EACCES;
 546     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
 547         memcpy(bs->boot_sector_data, buf, 512);
 548     }
 549     if (drv->bdrv_pwrite) {
 550         int ret, len;
 551         len = nb_sectors * 512;
 552         ret = drv->bdrv_pwrite(bs, sector_num * 512, buf, len);
 553         if (ret < 0)
 554             return ret;
 555         else if (ret != len)
 556             return -EIO;
 557         else
 558             return 0;
 559     } else {
 560         return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 561     }
 562 }
 563
 564 static int bdrv_pread_em(BlockDriverState *bs, int64_t offset,
 565                          uint8_t *buf, int count1)
 566 {
 567     uint8_t tmp_buf[SECTOR_SIZE];
 568     int len, nb_sectors, count;
 569     int64_t sector_num;
 570
 571     count = count1;
 572     /* first read to align to sector start */
 573     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 574     if (len > count)
 575         len = count;
 576     sector_num = offset >> SECTOR_BITS;
 577     if (len > 0) {
 578         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 579             return -EIO;
 580         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
 581         count -= len;
 582         if (count == 0)
 583             return count1;
 584         sector_num++;
 585         buf += len;
 586     }
 587
 588     /* read the sectors "in place" */
 589     nb_sectors = count >> SECTOR_BITS;
 590     if (nb_sectors > 0) {
 591         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
 592             return -EIO;
 593         sector_num += nb_sectors;
 594         len = nb_sectors << SECTOR_BITS;
 595         buf += len;
 596         count -= len;
 597     }
 598
 599     /* add data from the last sector */
 600     if (count > 0) {
 601         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 602             return -EIO;
 603         memcpy(buf, tmp_buf, count);
 604     }
 605     return count1;
 606 }
 607
 608 static int bdrv_pwrite_em(BlockDriverState *bs, int64_t offset,
 609                           const uint8_t *buf, int count1)
 610 {
 611     uint8_t tmp_buf[SECTOR_SIZE];
 612     int len, nb_sectors, count;
 613     int64_t sector_num;
 614
 615     count = count1;
 616     /* first write to align to sector start */
 617     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 618     if (len > count)
 619         len = count;
 620     sector_num = offset >> SECTOR_BITS;
 621     if (len > 0) {
 622         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 623             return -EIO;
 624         memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
 625         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 626             return -EIO;
 627         count -= len;
 628         if (count == 0)
 629             return count1;
 630         sector_num++;
 631         buf += len;
 632     }
 633
 634     /* write the sectors "in place" */
 635     nb_sectors = count >> SECTOR_BITS;
 636     if (nb_sectors > 0) {
 637         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
 638             return -EIO;
 639         sector_num += nb_sectors;
 640         len = nb_sectors << SECTOR_BITS;
 641         buf += len;
 642         count -= len;
 643     }
 644
 645     /* add data from the last sector */
 646     if (count > 0) {
 647         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 648             return -EIO;
 649         memcpy(tmp_buf, buf, count);
 650         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 651             return -EIO;
 652     }
 653     return count1;
 654 }
 655
 656 /**
 657  * Read with byte offsets (needed only for file protocols)
 658  */
 659 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 660                void *buf1, int count1)
 661 {
 662     BlockDriver *drv = bs->drv;
 663
 664     if (!drv)
 665         return -ENOMEDIUM;
 666     if (!drv->bdrv_pread)
 667         return bdrv_pread_em(bs, offset, buf1, count1);
 668     return drv->bdrv_pread(bs, offset, buf1, count1);
 669 }
 670
 671 /**
 672  * Write with byte offsets (needed only for file protocols)
 673  */
 674 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 675                 const void *buf1, int count1)
 676 {
 677     BlockDriver *drv = bs->drv;
 678
 679     if (!drv)
 680         return -ENOMEDIUM;
 681     if (!drv->bdrv_pwrite)
 682         return bdrv_pwrite_em(bs, offset, buf1, count1);
 683     return drv->bdrv_pwrite(bs, offset, buf1, count1);
 684 }
 685
 686 /**
 687  * Truncate file to 'offset' bytes (needed only for file protocols)
 688  */
 689 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
 690 {
 691     BlockDriver *drv = bs->drv;
 692     if (!drv)
 693         return -ENOMEDIUM;
 694     if (!drv->bdrv_truncate)
 695         return -ENOTSUP;
 696     return drv->bdrv_truncate(bs, offset);
 697 }
 698
 699 /**
 700  * Length of a file in bytes. Return < 0 if error or unknown.
 701  */
 702 int64_t bdrv_getlength(BlockDriverState *bs)
 703 {
 704     BlockDriver *drv = bs->drv;
 705     if (!drv)
 706         return -ENOMEDIUM;
 707     if (!drv->bdrv_getlength) {
 708         /* legacy mode */
 709         return bs->total_sectors * SECTOR_SIZE;
 710     }
 711     return drv->bdrv_getlength(bs);
 712 }
 713
 714 /* return 0 as number of sectors if no device present or error */
 715 void bdrv_get_geometry(BlockDriverState *bs, int64_t *nb_sectors_ptr)
 716 {
 717     int64_t length;
 718     length = bdrv_getlength(bs);
 719     if (length < 0)
 720         length = 0;
 721     else
 722         length = length >> SECTOR_BITS;
 723     *nb_sectors_ptr = length;
 724 }
 725
 726 /* force a given boot sector. */
 727 void bdrv_set_boot_sector(BlockDriverState *bs, const uint8_t *data, int size)
 728 {
 729     bs->boot_sector_enabled = 1;
 730     if (size > 512)
 731         size = 512;
 732     memcpy(bs->boot_sector_data, data, size);
 733     memset(bs->boot_sector_data + size, 0, 512 - size);
 734 }
 735
 736 void bdrv_set_geometry_hint(BlockDriverState *bs,
 737                             int cyls, int heads, int secs)
 738 {
 739     bs->cyls = cyls;
 740     bs->heads = heads;
 741     bs->secs = secs;
 742 }
 743
 744 void bdrv_set_type_hint(BlockDriverState *bs, int type)
 745 {
 746     bs->type = type;
 747     bs->removable = ((type == BDRV_TYPE_CDROM ||
 748                       type == BDRV_TYPE_FLOPPY));
 749 }
 750
 751 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
 752 {
 753     bs->translation = translation;
 754 }
 755
 756 void bdrv_get_geometry_hint(BlockDriverState *bs,
 757                             int *pcyls, int *pheads, int *psecs)
 758 {
 759     *pcyls = bs->cyls;
 760     *pheads = bs->heads;
 761     *psecs = bs->secs;
 762 }
 763
 764 int bdrv_get_type_hint(BlockDriverState *bs)
 765 {
 766     return bs->type;
 767 }
 768
 769 int bdrv_get_translation_hint(BlockDriverState *bs)
 770 {
 771     return bs->translation;
 772 }
 773
 774 int bdrv_is_removable(BlockDriverState *bs)
 775 {
 776     return bs->removable;
 777 }
 778
 779 int bdrv_is_read_only(BlockDriverState *bs)
 780 {
 781     return bs->read_only;
 782 }
 783
 784 /* XXX: no longer used */
 785 void bdrv_set_change_cb(BlockDriverState *bs,
 786                         void (*change_cb)(void *opaque), void *opaque)
 787 {
 788     bs->change_cb = change_cb;
 789     bs->change_opaque = opaque;
 790 }
 791
 792 int bdrv_is_encrypted(BlockDriverState *bs)
 793 {
 794     if (bs->backing_hd && bs->backing_hd->encrypted)
 795         return 1;
 796     return bs->encrypted;
 797 }
 798
 799 int bdrv_set_key(BlockDriverState *bs, const char *key)
 800 {
 801     int ret;
 802     if (bs->backing_hd && bs->backing_hd->encrypted) {
 803         ret = bdrv_set_key(bs->backing_hd, key);
 804         if (ret < 0)
 805             return ret;
 806         if (!bs->encrypted)
 807             return 0;
 808     }
 809     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
 810         return -1;
 811     return bs->drv->bdrv_set_key(bs, key);
 812 }
 813
 814 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
 815 {
 816     if (!bs->drv) {
 817         buf[0] = '\0';
 818     } else {
 819         pstrcpy(buf, buf_size, bs->drv->format_name);
 820     }
 821 }
 822
 823 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
 824                          void *opaque)
 825 {
 826     BlockDriver *drv;
 827
 828     for (drv = first_drv; drv != NULL; drv = drv->next) {
 829         it(opaque, drv->format_name);
 830     }
 831 }
 832
 833 BlockDriverState *bdrv_find(const char *name)
 834 {
 835     BlockDriverState *bs;
 836
 837     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 838         if (!strcmp(name, bs->device_name))
 839             return bs;
 840     }
 841     return NULL;
 842 }
 843
 844 void bdrv_iterate(void (*it)(void *opaque, const char *name), void *opaque)
 845 {
 846     BlockDriverState *bs;
 847
 848     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 849         it(opaque, bs->device_name);
 850     }
 851 }
 852
 853 const char *bdrv_get_device_name(BlockDriverState *bs)
 854 {
 855     return bs->device_name;
 856 }
 857
 858 void bdrv_flush(BlockDriverState *bs)
 859 {
 860     if (bs->drv->bdrv_flush)
 861         bs->drv->bdrv_flush(bs);
 862     if (bs->backing_hd)
 863         bdrv_flush(bs->backing_hd);
 864 }
 865
 866 #ifndef QEMU_IMG
 867 void bdrv_info(void)
 868 {
 869     BlockDriverState *bs;
 870
 871     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 872         term_printf("%s:", bs->device_name);
 873         term_printf(" type=");
 874         switch(bs->type) {
 875         case BDRV_TYPE_HD:
 876             term_printf("hd");
 877             break;
 878         case BDRV_TYPE_CDROM:
 879             term_printf("cdrom");
 880             break;
 881         case BDRV_TYPE_FLOPPY:
 882             term_printf("floppy");
 883             break;
 884         }
 885         term_printf(" removable=%d", bs->removable);
 886         if (bs->removable) {
 887             term_printf(" locked=%d", bs->locked);
 888         }
 889         if (bs->drv) {
 890             term_printf(" file=");
 891             term_print_filename(bs->filename);
 892             if (bs->backing_file[0] != '\0') {
 893                 term_printf(" backing_file=");
 894                 term_print_filename(bs->backing_file);
 895             }
 896             term_printf(" ro=%d", bs->read_only);
 897             term_printf(" drv=%s", bs->drv->format_name);
 898             if (bs->encrypted)
 899                 term_printf(" encrypted");
 900         } else {
 901             term_printf(" [not inserted]");
 902         }
 903         term_printf("\n");
 904     }
 905 }
 906 #endif
 907
 908 void bdrv_get_backing_filename(BlockDriverState *bs,
 909                                char *filename, int filename_size)
 910 {
 911     if (!bs->backing_hd) {
 912         pstrcpy(filename, filename_size, "");
 913     } else {
 914         pstrcpy(filename, filename_size, bs->backing_file);
 915     }
 916 }
 917
 918 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
 919                           const uint8_t *buf, int nb_sectors)
 920 {
 921     BlockDriver *drv = bs->drv;
 922     if (!drv)
 923         return -ENOMEDIUM;
 924     if (!drv->bdrv_write_compressed)
 925         return -ENOTSUP;
 926     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
 927 }
 928
 929 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 930 {
 931     BlockDriver *drv = bs->drv;
 932     if (!drv)
 933         return -ENOMEDIUM;
 934     if (!drv->bdrv_get_info)
 935         return -ENOTSUP;
 936     memset(bdi, 0, sizeof(*bdi));
 937     return drv->bdrv_get_info(bs, bdi);
 938 }
 939
 940 /**************************************************************/
 941 /* handling of snapshots */
 942
 943 int bdrv_snapshot_create(BlockDriverState *bs,
 944                          QEMUSnapshotInfo *sn_info)
 945 {
 946     BlockDriver *drv = bs->drv;
 947     if (!drv)
 948         return -ENOMEDIUM;
 949     if (!drv->bdrv_snapshot_create)
 950         return -ENOTSUP;
 951     return drv->bdrv_snapshot_create(bs, sn_info);
 952 }
 953
 954 int bdrv_snapshot_goto(BlockDriverState *bs,
 955                        const char *snapshot_id)
 956 {
 957     BlockDriver *drv = bs->drv;
 958     if (!drv)
 959         return -ENOMEDIUM;
 960     if (!drv->bdrv_snapshot_goto)
 961         return -ENOTSUP;
 962     return drv->bdrv_snapshot_goto(bs, snapshot_id);
 963 }
 964
 965 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
 966 {
 967     BlockDriver *drv = bs->drv;
 968     if (!drv)
 969         return -ENOMEDIUM;
 970     if (!drv->bdrv_snapshot_delete)
 971         return -ENOTSUP;
 972     return drv->bdrv_snapshot_delete(bs, snapshot_id);
 973 }
 974
 975 int bdrv_snapshot_list(BlockDriverState *bs,
 976                        QEMUSnapshotInfo **psn_info)
 977 {
 978     BlockDriver *drv = bs->drv;
 979     if (!drv)
 980         return -ENOMEDIUM;
 981     if (!drv->bdrv_snapshot_list)
 982         return -ENOTSUP;
 983     return drv->bdrv_snapshot_list(bs, psn_info);
 984 }
 985
 986 #define NB_SUFFIXES 4
 987
 988 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
 989 {
 990     static const char suffixes[NB_SUFFIXES] = "KMGT";
 991     int64_t base;
 992     int i;
 993
 994     if (size <= 999) {
 995         snprintf(buf, buf_size, "%" PRId64, size);
 996     } else {
 997         base = 1024;
 998         for(i = 0; i < NB_SUFFIXES; i++) {
 999             if (size < (10 * base)) {
1000                 snprintf(buf, buf_size, "%0.1f%c",
1001                          (double)size / base,
1002                          suffixes[i]);
1003                 break;
1004             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1005                 snprintf(buf, buf_size, "%" PRId64 "%c",
1006                          ((size + (base >> 1)) / base),
1007                          suffixes[i]);
1008                 break;
1009             }
1010             base = base * 1024;
1011         }
1012     }
1013     return buf;
1014 }
1015
1016 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1017 {
1018     char buf1[128], date_buf[128], clock_buf[128];
1019 #ifdef _WIN32
1020     struct tm *ptm;
1021 #else
1022     struct tm tm;
1023 #endif
1024     time_t ti;
1025     int64_t secs;
1026
1027     if (!sn) {
1028         snprintf(buf, buf_size,
1029                  "%-10s%-20s%7s%20s%15s",
1030                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1031     } else {
1032         ti = sn->date_sec;
1033 #ifdef _WIN32
1034         ptm = localtime(&ti);
1035         strftime(date_buf, sizeof(date_buf),
1036                  "%Y-%m-%d %H:%M:%S", ptm);
1037 #else
1038         localtime_r(&ti, &tm);
1039         strftime(date_buf, sizeof(date_buf),
1040                  "%Y-%m-%d %H:%M:%S", &tm);
1041 #endif
1042         secs = sn->vm_clock_nsec / 1000000000;
1043         snprintf(clock_buf, sizeof(clock_buf),
1044                  "%02d:%02d:%02d.%03d",
1045                  (int)(secs / 3600),
1046                  (int)((secs / 60) % 60),
1047                  (int)(secs % 60),
1048                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1049         snprintf(buf, buf_size,
1050                  "%-10s%-20s%7s%20s%15s",
1051                  sn->id_str, sn->name,
1052                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1053                  date_buf,
1054                  clock_buf);
1055     }
1056     return buf;
1057 }
1058
1059
1060 /**************************************************************/
1061 /* async I/Os */
1062
1063 BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
1064                                 uint8_t *buf, int nb_sectors,
1065                                 BlockDriverCompletionFunc *cb, void *opaque)
1066 {
1067     BlockDriver *drv = bs->drv;
1068
1069     if (!drv)
1070         return NULL;
1071
1072     /* XXX: we assume that nb_sectors == 0 is suppored by the async read */
1073     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
1074         memcpy(buf, bs->boot_sector_data, 512);
1075         sector_num++;
1076         nb_sectors--;
1077         buf += 512;
1078     }
1079
1080     return drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
1081 }
1082
1083 BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
1084                                  const uint8_t *buf, int nb_sectors,
1085                                  BlockDriverCompletionFunc *cb, void *opaque)
1086 {
1087     BlockDriver *drv = bs->drv;
1088
1089     if (!drv)
1090         return NULL;
1091     if (bs->read_only)
1092         return NULL;
1093     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
1094         memcpy(bs->boot_sector_data, buf, 512);
1095     }
1096
1097     return drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
1098 }
1099
1100 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1101 {
1102     BlockDriver *drv = acb->bs->drv;
1103
1104     drv->bdrv_aio_cancel(acb);
1105 }
1106
1107
1108 /**************************************************************/
1109 /* async block device emulation */
1110
1111 #ifdef QEMU_IMG
1112 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1113         int64_t sector_num, uint8_t *buf, int nb_sectors,
1114         BlockDriverCompletionFunc *cb, void *opaque)
1115 {
1116     int ret;
1117     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1118     cb(opaque, ret);
1119     return NULL;
1120 }
1121
1122 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1123         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1124         BlockDriverCompletionFunc *cb, void *opaque)
1125 {
1126     int ret;
1127     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1128     cb(opaque, ret);
1129     return NULL;
1130 }
1131
1132 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb)
1133 {
1134 }
1135 #else
1136 static void bdrv_aio_bh_cb(void *opaque)
1137 {
1138     BlockDriverAIOCBSync *acb = opaque;
1139     acb->common.cb(acb->common.opaque, acb->ret);
1140     qemu_aio_release(acb);
1141 }
1142
1143 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1144         int64_t sector_num, uint8_t *buf, int nb_sectors,
1145         BlockDriverCompletionFunc *cb, void *opaque)
1146 {
1147     BlockDriverAIOCBSync *acb;
1148     int ret;
1149
1150     acb = qemu_aio_get(bs, cb, opaque);
1151     if (!acb->bh)
1152         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1153     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1154     acb->ret = ret;
1155     qemu_bh_schedule(acb->bh);
1156     return &acb->common;
1157 }
1158
1159 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1160         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1161         BlockDriverCompletionFunc *cb, void *opaque)
1162 {
1163     BlockDriverAIOCBSync *acb;
1164     int ret;
1165
1166     acb = qemu_aio_get(bs, cb, opaque);
1167     if (!acb->bh)
1168         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1169     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1170     acb->ret = ret;
1171     qemu_bh_schedule(acb->bh);
1172     return &acb->common;
1173 }
1174
1175 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1176 {
1177     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1178     qemu_bh_cancel(acb->bh);
1179     qemu_aio_release(acb);
1180 }
1181 #endif /* !QEMU_IMG */
1182
1183 /**************************************************************/
1184 /* sync block device emulation */
1185
1186 static void bdrv_rw_em_cb(void *opaque, int ret)
1187 {
1188     *(int *)opaque = ret;
1189 }
1190
1191 #define NOT_DONE 0x7fffffff
1192
1193 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1194                         uint8_t *buf, int nb_sectors)
1195 {
1196     int async_ret;
1197     BlockDriverAIOCB *acb;
1198
1199     async_ret = NOT_DONE;
1200     qemu_aio_wait_start();
1201     acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
1202                         bdrv_rw_em_cb, &async_ret);
1203     if (acb == NULL) {
1204         qemu_aio_wait_end();
1205         return -1;
1206     }
1207     while (async_ret == NOT_DONE) {
1208         qemu_aio_wait();
1209     }
1210     qemu_aio_wait_end();
1211     return async_ret;
1212 }
1213
1214 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1215                          const uint8_t *buf, int nb_sectors)
1216 {
1217     int async_ret;
1218     BlockDriverAIOCB *acb;
1219
1220     async_ret = NOT_DONE;
1221     qemu_aio_wait_start();
1222     acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
1223                          bdrv_rw_em_cb, &async_ret);
1224     if (acb == NULL) {
1225         qemu_aio_wait_end();
1226         return -1;
1227     }
1228     while (async_ret == NOT_DONE) {
1229         qemu_aio_wait();
1230     }
1231     qemu_aio_wait_end();
1232     return async_ret;
1233 }
1234
1235 void bdrv_init(void)
1236 {
1237     bdrv_register(&bdrv_raw);
1238     bdrv_register(&bdrv_host_device);
1239 #ifndef _WIN32
1240     bdrv_register(&bdrv_cow);
1241 #endif
1242     bdrv_register(&bdrv_qcow);
1243     bdrv_register(&bdrv_vmdk);
1244     bdrv_register(&bdrv_cloop);
1245     bdrv_register(&bdrv_dmg);
1246     bdrv_register(&bdrv_bochs);
1247     bdrv_register(&bdrv_vpc);
1248     bdrv_register(&bdrv_vvfat);
1249     bdrv_register(&bdrv_qcow2);
1250     bdrv_register(&bdrv_parallels);
1251 }
1252
1253 void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
1254                    void *opaque)
1255 {
1256     BlockDriver *drv;
1257     BlockDriverAIOCB *acb;
1258
1259     drv = bs->drv;
1260     if (drv->free_aiocb) {
1261         acb = drv->free_aiocb;
1262         drv->free_aiocb = acb->next;
1263     } else {
1264         acb = qemu_mallocz(drv->aiocb_size);
1265         if (!acb)
1266             return NULL;
1267     }
1268     acb->bs = bs;
1269     acb->cb = cb;
1270     acb->opaque = opaque;
1271     return acb;
1272 }
1273
1274 void qemu_aio_release(void *p)
1275 {
1276     BlockDriverAIOCB *acb = p;
1277     BlockDriver *drv = acb->bs->drv;
1278     acb->next = drv->free_aiocb;
1279     drv->free_aiocb = acb;
1280 }
1281
1282 /**************************************************************/
1283 /* removable device support */
1284
1285 /**
1286  * Return TRUE if the media is present
1287  */
1288 int bdrv_is_inserted(BlockDriverState *bs)
1289 {
1290     BlockDriver *drv = bs->drv;
1291     int ret;
1292     if (!drv)
1293         return 0;
1294     if (!drv->bdrv_is_inserted)
1295         return 1;
1296     ret = drv->bdrv_is_inserted(bs);
1297     return ret;
1298 }
1299
1300 /**
1301  * Return TRUE if the media changed since the last call to this
1302  * function. It is currently only used for floppy disks
1303  */
1304 int bdrv_media_changed(BlockDriverState *bs)
1305 {
1306     BlockDriver *drv = bs->drv;
1307     int ret;
1308
1309     if (!drv || !drv->bdrv_media_changed)
1310         ret = -ENOTSUP;
1311     else
1312         ret = drv->bdrv_media_changed(bs);
1313     if (ret == -ENOTSUP)
1314         ret = bs->media_changed;
1315     bs->media_changed = 0;
1316     return ret;
1317 }
1318
1319 /**
1320  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1321  */
1322 void bdrv_eject(BlockDriverState *bs, int eject_flag)
1323 {
1324     BlockDriver *drv = bs->drv;
1325     int ret;
1326
1327     if (!drv || !drv->bdrv_eject) {
1328         ret = -ENOTSUP;
1329     } else {
1330         ret = drv->bdrv_eject(bs, eject_flag);
1331     }
1332     if (ret == -ENOTSUP) {
1333         if (eject_flag)
1334             bdrv_close(bs);
1335     }
1336 }
1337
1338 int bdrv_is_locked(BlockDriverState *bs)
1339 {
1340     return bs->locked;
1341 }
1342
1343 /**
1344  * Lock or unlock the media (if it is locked, the user won't be able
1345  * to eject it manually).
1346  */
1347 void bdrv_set_locked(BlockDriverState *bs, int locked)
1348 {
1349     BlockDriver *drv = bs->drv;
1350
1351     bs->locked = locked;
1352     if (drv && drv->bdrv_set_locked) {
1353         drv->bdrv_set_locked(bs, locked);
1354     }
1355 }