block.c

   1 /*
   2  * QEMU System Emulator block driver
   3  *
   4  * Copyright (c) 2003 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #ifndef QEMU_IMG
  26 #include "console.h"
  27 #endif
  28 #include "block_int.h"
  29
  30 #ifdef _BSD
  31 #include <sys/types.h>
  32 #include <sys/stat.h>
  33 #include <sys/ioctl.h>
  34 #include <sys/queue.h>
  35 #include <sys/disk.h>
  36 #endif
  37
  38 #define SECTOR_BITS 9
  39 #define SECTOR_SIZE (1 << SECTOR_BITS)
  40
  41 typedef struct BlockDriverAIOCBSync {
  42     BlockDriverAIOCB common;
  43     QEMUBH *bh;
  44     int ret;
  45 } BlockDriverAIOCBSync;
  46
  47 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
  48         int64_t sector_num, uint8_t *buf, int nb_sectors,
  49         BlockDriverCompletionFunc *cb, void *opaque);
  50 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
  51         int64_t sector_num, const uint8_t *buf, int nb_sectors,
  52         BlockDriverCompletionFunc *cb, void *opaque);
  53 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
  54 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
  55                         uint8_t *buf, int nb_sectors);
  56 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
  57                          const uint8_t *buf, int nb_sectors);
  58
  59 BlockDriverState *bdrv_first;
  60 static BlockDriver *first_drv;
  61
  62 int path_is_absolute(const char *path)
  63 {
  64     const char *p;
  65 #ifdef _WIN32
  66     /* specific case for names like: "\\.\d:" */
  67     if (*path == '/' || *path == '\\')
  68         return 1;
  69 #endif
  70     p = strchr(path, ':');
  71     if (p)
  72         p++;
  73     else
  74         p = path;
  75 #ifdef _WIN32
  76     return (*p == '/' || *p == '\\');
  77 #else
  78     return (*p == '/');
  79 #endif
  80 }
  81
  82 /* if filename is absolute, just copy it to dest. Otherwise, build a
  83    path to it by considering it is relative to base_path. URL are
  84    supported. */
  85 void path_combine(char *dest, int dest_size,
  86                   const char *base_path,
  87                   const char *filename)
  88 {
  89     const char *p, *p1;
  90     int len;
  91
  92     if (dest_size <= 0)
  93         return;
  94     if (path_is_absolute(filename)) {
  95         pstrcpy(dest, dest_size, filename);
  96     } else {
  97         p = strchr(base_path, ':');
  98         if (p)
  99             p++;
 100         else
 101             p = base_path;
 102         p1 = strrchr(base_path, '/');
 103 #ifdef _WIN32
 104         {
 105             const char *p2;
 106             p2 = strrchr(base_path, '\\');
 107             if (!p1 || p2 > p1)
 108                 p1 = p2;
 109         }
 110 #endif
 111         if (p1)
 112             p1++;
 113         else
 114             p1 = base_path;
 115         if (p1 > p)
 116             p = p1;
 117         len = p - base_path;
 118         if (len > dest_size - 1)
 119             len = dest_size - 1;
 120         memcpy(dest, base_path, len);
 121         dest[len] = '\0';
 122         pstrcat(dest, dest_size, filename);
 123     }
 124 }
 125
 126
 127 static void bdrv_register(BlockDriver *bdrv)
 128 {
 129     if (!bdrv->bdrv_aio_read) {
 130         /* add AIO emulation layer */
 131         bdrv->bdrv_aio_read = bdrv_aio_read_em;
 132         bdrv->bdrv_aio_write = bdrv_aio_write_em;
 133         bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
 134         bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
 135     } else if (!bdrv->bdrv_read && !bdrv->bdrv_pread) {
 136         /* add synchronous IO emulation layer */
 137         bdrv->bdrv_read = bdrv_read_em;
 138         bdrv->bdrv_write = bdrv_write_em;
 139     }
 140     bdrv->next = first_drv;
 141     first_drv = bdrv;
 142 }
 143
 144 /* create a new block device (by default it is empty) */
 145 BlockDriverState *bdrv_new(const char *device_name)
 146 {
 147     BlockDriverState **pbs, *bs;
 148
 149     bs = qemu_mallocz(sizeof(BlockDriverState));
 150     if(!bs)
 151         return NULL;
 152     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
 153     if (device_name[0] != '\0') {
 154         /* insert at the end */
 155         pbs = &bdrv_first;
 156         while (*pbs != NULL)
 157             pbs = &(*pbs)->next;
 158         *pbs = bs;
 159     }
 160     return bs;
 161 }
 162
 163 BlockDriver *bdrv_find_format(const char *format_name)
 164 {
 165     BlockDriver *drv1;
 166     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 167         if (!strcmp(drv1->format_name, format_name))
 168             return drv1;
 169     }
 170     return NULL;
 171 }
 172
 173 int bdrv_create(BlockDriver *drv,
 174                 const char *filename, int64_t size_in_sectors,
 175                 const char *backing_file, int flags)
 176 {
 177     if (!drv->bdrv_create)
 178         return -ENOTSUP;
 179     return drv->bdrv_create(filename, size_in_sectors, backing_file, flags);
 180 }
 181
 182 #ifdef _WIN32
 183 void get_tmp_filename(char *filename, int size)
 184 {
 185     char temp_dir[MAX_PATH];
 186
 187     GetTempPath(MAX_PATH, temp_dir);
 188     GetTempFileName(temp_dir, "qem", 0, filename);
 189 }
 190 #else
 191 void get_tmp_filename(char *filename, int size)
 192 {
 193     int fd;
 194     /* XXX: race condition possible */
 195     pstrcpy(filename, size, "/tmp/vl.XXXXXX");
 196     fd = mkstemp(filename);
 197     close(fd);
 198 }
 199 #endif
 200
 201 #ifdef _WIN32
 202 static int is_windows_drive_prefix(const char *filename)
 203 {
 204     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
 205              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
 206             filename[1] == ':');
 207 }
 208
 209 static int is_windows_drive(const char *filename)
 210 {
 211     if (is_windows_drive_prefix(filename) &&
 212         filename[2] == '\0')
 213         return 1;
 214     if (strstart(filename, "\\\\.\\", NULL) ||
 215         strstart(filename, "//./", NULL))
 216         return 1;
 217     return 0;
 218 }
 219 #endif
 220
 221 static BlockDriver *find_protocol(const char *filename)
 222 {
 223     BlockDriver *drv1;
 224     char protocol[128];
 225     int len;
 226     const char *p;
 227
 228 #ifdef _WIN32
 229     if (is_windows_drive(filename) ||
 230         is_windows_drive_prefix(filename))
 231         return &bdrv_raw;
 232 #endif
 233     p = strchr(filename, ':');
 234     if (!p)
 235         return &bdrv_raw;
 236     len = p - filename;
 237     if (len > sizeof(protocol) - 1)
 238         len = sizeof(protocol) - 1;
 239     memcpy(protocol, filename, len);
 240     protocol[len] = '\0';
 241     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 242         if (drv1->protocol_name &&
 243             !strcmp(drv1->protocol_name, protocol))
 244             return drv1;
 245     }
 246     return NULL;
 247 }
 248
 249 /* XXX: force raw format if block or character device ? It would
 250    simplify the BSD case */
 251 static BlockDriver *find_image_format(const char *filename)
 252 {
 253     int ret, score, score_max;
 254     BlockDriver *drv1, *drv;
 255     uint8_t buf[2048];
 256     BlockDriverState *bs;
 257
 258     /* detect host devices. By convention, /dev/cdrom[N] is always
 259        recognized as a host CDROM */
 260     if (strstart(filename, "/dev/cdrom", NULL))
 261         return &bdrv_host_device;
 262 #ifdef _WIN32
 263     if (is_windows_drive(filename))
 264         return &bdrv_host_device;
 265 #else
 266     {
 267         struct stat st;
 268         if (stat(filename, &st) >= 0 &&
 269             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 270             return &bdrv_host_device;
 271         }
 272     }
 273 #endif
 274
 275     drv = find_protocol(filename);
 276     /* no need to test disk image formats for vvfat */
 277     if (drv == &bdrv_vvfat)
 278         return drv;
 279
 280     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
 281     if (ret < 0)
 282         return NULL;
 283     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
 284     bdrv_delete(bs);
 285     if (ret < 0) {
 286         return NULL;
 287     }
 288
 289     score_max = 0;
 290     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 291         if (drv1->bdrv_probe) {
 292             score = drv1->bdrv_probe(buf, ret, filename);
 293             if (score > score_max) {
 294                 score_max = score;
 295                 drv = drv1;
 296             }
 297         }
 298     }
 299     return drv;
 300 }
 301
 302 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
 303 {
 304     BlockDriverState *bs;
 305     int ret;
 306
 307     bs = bdrv_new("");
 308     if (!bs)
 309         return -ENOMEM;
 310     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
 311     if (ret < 0) {
 312         bdrv_delete(bs);
 313         return ret;
 314     }
 315     *pbs = bs;
 316     return 0;
 317 }
 318
 319 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
 320 {
 321     return bdrv_open2(bs, filename, flags, NULL);
 322 }
 323
 324 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
 325                BlockDriver *drv)
 326 {
 327     int ret, open_flags;
 328     char tmp_filename[PATH_MAX];
 329     char backing_filename[PATH_MAX];
 330
 331     bs->read_only = 0;
 332     bs->is_temporary = 0;
 333     bs->encrypted = 0;
 334
 335     if (flags & BDRV_O_SNAPSHOT) {
 336         BlockDriverState *bs1;
 337         int64_t total_size;
 338
 339         /* if snapshot, we create a temporary backing file and open it
 340            instead of opening 'filename' directly */
 341
 342         /* if there is a backing file, use it */
 343         bs1 = bdrv_new("");
 344         if (!bs1) {
 345             return -ENOMEM;
 346         }
 347         if (bdrv_open(bs1, filename, 0) < 0) {
 348             bdrv_delete(bs1);
 349             return -1;
 350         }
 351         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
 352         bdrv_delete(bs1);
 353
 354         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
 355         realpath(filename, backing_filename);
 356         if (bdrv_create(&bdrv_qcow2, tmp_filename,
 357                         total_size, backing_filename, 0) < 0) {
 358             return -1;
 359         }
 360         filename = tmp_filename;
 361         bs->is_temporary = 1;
 362     }
 363
 364     pstrcpy(bs->filename, sizeof(bs->filename), filename);
 365     if (flags & BDRV_O_FILE) {
 366         drv = find_protocol(filename);
 367         if (!drv)
 368             return -ENOENT;
 369     } else {
 370         if (!drv) {
 371             drv = find_image_format(filename);
 372             if (!drv)
 373                 return -1;
 374         }
 375     }
 376     bs->drv = drv;
 377     bs->opaque = qemu_mallocz(drv->instance_size);
 378     if (bs->opaque == NULL && drv->instance_size > 0)
 379         return -1;
 380     /* Note: for compatibility, we open disk image files as RDWR, and
 381        RDONLY as fallback */
 382     if (!(flags & BDRV_O_FILE))
 383         open_flags = BDRV_O_RDWR;
 384     else
 385         open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
 386     ret = drv->bdrv_open(bs, filename, open_flags);
 387     if (ret == -EACCES && !(flags & BDRV_O_FILE)) {
 388         ret = drv->bdrv_open(bs, filename, BDRV_O_RDONLY);
 389         bs->read_only = 1;
 390     }
 391     if (ret < 0) {
 392         qemu_free(bs->opaque);
 393         bs->opaque = NULL;
 394         bs->drv = NULL;
 395         return ret;
 396     }
 397     if (drv->bdrv_getlength) {
 398         bs->total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 399     }
 400 #ifndef _WIN32
 401     if (bs->is_temporary) {
 402         unlink(filename);
 403     }
 404 #endif
 405     if (bs->backing_file[0] != '\0') {
 406         /* if there is a backing file, use it */
 407         bs->backing_hd = bdrv_new("");
 408         if (!bs->backing_hd) {
 409         fail:
 410             bdrv_close(bs);
 411             return -ENOMEM;
 412         }
 413         path_combine(backing_filename, sizeof(backing_filename),
 414                      filename, bs->backing_file);
 415         if (bdrv_open(bs->backing_hd, backing_filename, 0) < 0)
 416             goto fail;
 417     }
 418
 419     /* call the change callback */
 420     bs->media_changed = 1;
 421     if (bs->change_cb)
 422         bs->change_cb(bs->change_opaque);
 423
 424     return 0;
 425 }
 426
 427 void bdrv_close(BlockDriverState *bs)
 428 {
 429     if (bs->drv) {
 430         if (bs->backing_hd)
 431             bdrv_delete(bs->backing_hd);
 432         bs->drv->bdrv_close(bs);
 433         qemu_free(bs->opaque);
 434 #ifdef _WIN32
 435         if (bs->is_temporary) {
 436             unlink(bs->filename);
 437         }
 438 #endif
 439         bs->opaque = NULL;
 440         bs->drv = NULL;
 441
 442         /* call the change callback */
 443         bs->media_changed = 1;
 444         if (bs->change_cb)
 445             bs->change_cb(bs->change_opaque);
 446     }
 447 }
 448
 449 void bdrv_delete(BlockDriverState *bs)
 450 {
 451     /* XXX: remove the driver list */
 452     bdrv_close(bs);
 453     qemu_free(bs);
 454 }
 455
 456 /* commit COW file into the raw image */
 457 int bdrv_commit(BlockDriverState *bs)
 458 {
 459     BlockDriver *drv = bs->drv;
 460     int64_t i, total_sectors;
 461     int n, j;
 462     unsigned char sector[512];
 463
 464     if (!drv)
 465         return -ENOMEDIUM;
 466
 467     if (bs->read_only) {
 468         return -EACCES;
 469     }
 470
 471     if (!bs->backing_hd) {
 472         return -ENOTSUP;
 473     }
 474
 475     total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 476     for (i = 0; i < total_sectors;) {
 477         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
 478             for(j = 0; j < n; j++) {
 479                 if (bdrv_read(bs, i, sector, 1) != 0) {
 480                     return -EIO;
 481                 }
 482
 483                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
 484                     return -EIO;
 485                 }
 486                 i++;
 487             }
 488         } else {
 489             i += n;
 490         }
 491     }
 492
 493     if (drv->bdrv_make_empty)
 494         return drv->bdrv_make_empty(bs);
 495
 496     return 0;
 497 }
 498
 499 /* return < 0 if error. See bdrv_write() for the return codes */
 500 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 501               uint8_t *buf, int nb_sectors)
 502 {
 503     BlockDriver *drv = bs->drv;
 504
 505     if (!drv)
 506         return -ENOMEDIUM;
 507
 508     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
 509             memcpy(buf, bs->boot_sector_data, 512);
 510         sector_num++;
 511         nb_sectors--;
 512         buf += 512;
 513         if (nb_sectors == 0)
 514             return 0;
 515     }
 516     if (drv->bdrv_pread) {
 517         int ret, len;
 518         len = nb_sectors * 512;
 519         ret = drv->bdrv_pread(bs, sector_num * 512, buf, len);
 520         if (ret < 0)
 521             return ret;
 522         else if (ret != len)
 523             return -EINVAL;
 524         else
 525             return 0;
 526     } else {
 527         return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 528     }
 529 }
 530
 531 /* Return < 0 if error. Important errors are:
 532   -EIO         generic I/O error (may happen for all errors)
 533   -ENOMEDIUM   No media inserted.
 534   -EINVAL      Invalid sector number or nb_sectors
 535   -EACCES      Trying to write a read-only device
 536 */
 537 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 538                const uint8_t *buf, int nb_sectors)
 539 {
 540     BlockDriver *drv = bs->drv;
 541     if (!bs->drv)
 542         return -ENOMEDIUM;
 543     if (bs->read_only)
 544         return -EACCES;
 545     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
 546         memcpy(bs->boot_sector_data, buf, 512);
 547     }
 548     if (drv->bdrv_pwrite) {
 549         int ret, len;
 550         len = nb_sectors * 512;
 551         ret = drv->bdrv_pwrite(bs, sector_num * 512, buf, len);
 552         if (ret < 0)
 553             return ret;
 554         else if (ret != len)
 555             return -EIO;
 556         else
 557             return 0;
 558     } else {
 559         return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 560     }
 561 }
 562
 563 static int bdrv_pread_em(BlockDriverState *bs, int64_t offset,
 564                          uint8_t *buf, int count1)
 565 {
 566     uint8_t tmp_buf[SECTOR_SIZE];
 567     int len, nb_sectors, count;
 568     int64_t sector_num;
 569
 570     count = count1;
 571     /* first read to align to sector start */
 572     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 573     if (len > count)
 574         len = count;
 575     sector_num = offset >> SECTOR_BITS;
 576     if (len > 0) {
 577         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 578             return -EIO;
 579         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
 580         count -= len;
 581         if (count == 0)
 582             return count1;
 583         sector_num++;
 584         buf += len;
 585     }
 586
 587     /* read the sectors "in place" */
 588     nb_sectors = count >> SECTOR_BITS;
 589     if (nb_sectors > 0) {
 590         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
 591             return -EIO;
 592         sector_num += nb_sectors;
 593         len = nb_sectors << SECTOR_BITS;
 594         buf += len;
 595         count -= len;
 596     }
 597
 598     /* add data from the last sector */
 599     if (count > 0) {
 600         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 601             return -EIO;
 602         memcpy(buf, tmp_buf, count);
 603     }
 604     return count1;
 605 }
 606
 607 static int bdrv_pwrite_em(BlockDriverState *bs, int64_t offset,
 608                           const uint8_t *buf, int count1)
 609 {
 610     uint8_t tmp_buf[SECTOR_SIZE];
 611     int len, nb_sectors, count;
 612     int64_t sector_num;
 613
 614     count = count1;
 615     /* first write to align to sector start */
 616     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 617     if (len > count)
 618         len = count;
 619     sector_num = offset >> SECTOR_BITS;
 620     if (len > 0) {
 621         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 622             return -EIO;
 623         memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
 624         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 625             return -EIO;
 626         count -= len;
 627         if (count == 0)
 628             return count1;
 629         sector_num++;
 630         buf += len;
 631     }
 632
 633     /* write the sectors "in place" */
 634     nb_sectors = count >> SECTOR_BITS;
 635     if (nb_sectors > 0) {
 636         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
 637             return -EIO;
 638         sector_num += nb_sectors;
 639         len = nb_sectors << SECTOR_BITS;
 640         buf += len;
 641         count -= len;
 642     }
 643
 644     /* add data from the last sector */
 645     if (count > 0) {
 646         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 647             return -EIO;
 648         memcpy(tmp_buf, buf, count);
 649         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 650             return -EIO;
 651     }
 652     return count1;
 653 }
 654
 655 /**
 656  * Read with byte offsets (needed only for file protocols)
 657  */
 658 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 659                void *buf1, int count1)
 660 {
 661     BlockDriver *drv = bs->drv;
 662
 663     if (!drv)
 664         return -ENOMEDIUM;
 665     if (!drv->bdrv_pread)
 666         return bdrv_pread_em(bs, offset, buf1, count1);
 667     return drv->bdrv_pread(bs, offset, buf1, count1);
 668 }
 669
 670 /**
 671  * Write with byte offsets (needed only for file protocols)
 672  */
 673 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 674                 const void *buf1, int count1)
 675 {
 676     BlockDriver *drv = bs->drv;
 677
 678     if (!drv)
 679         return -ENOMEDIUM;
 680     if (!drv->bdrv_pwrite)
 681         return bdrv_pwrite_em(bs, offset, buf1, count1);
 682     return drv->bdrv_pwrite(bs, offset, buf1, count1);
 683 }
 684
 685 /**
 686  * Truncate file to 'offset' bytes (needed only for file protocols)
 687  */
 688 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
 689 {
 690     BlockDriver *drv = bs->drv;
 691     if (!drv)
 692         return -ENOMEDIUM;
 693     if (!drv->bdrv_truncate)
 694         return -ENOTSUP;
 695     return drv->bdrv_truncate(bs, offset);
 696 }
 697
 698 /**
 699  * Length of a file in bytes. Return < 0 if error or unknown.
 700  */
 701 int64_t bdrv_getlength(BlockDriverState *bs)
 702 {
 703     BlockDriver *drv = bs->drv;
 704     if (!drv)
 705         return -ENOMEDIUM;
 706     if (!drv->bdrv_getlength) {
 707         /* legacy mode */
 708         return bs->total_sectors * SECTOR_SIZE;
 709     }
 710     return drv->bdrv_getlength(bs);
 711 }
 712
 713 /* return 0 as number of sectors if no device present or error */
 714 void bdrv_get_geometry(BlockDriverState *bs, int64_t *nb_sectors_ptr)
 715 {
 716     int64_t length;
 717     length = bdrv_getlength(bs);
 718     if (length < 0)
 719         length = 0;
 720     else
 721         length = length >> SECTOR_BITS;
 722     *nb_sectors_ptr = length;
 723 }
 724
 725 /* force a given boot sector. */
 726 void bdrv_set_boot_sector(BlockDriverState *bs, const uint8_t *data, int size)
 727 {
 728     bs->boot_sector_enabled = 1;
 729     if (size > 512)
 730         size = 512;
 731     memcpy(bs->boot_sector_data, data, size);
 732     memset(bs->boot_sector_data + size, 0, 512 - size);
 733 }
 734
 735 void bdrv_set_geometry_hint(BlockDriverState *bs,
 736                             int cyls, int heads, int secs)
 737 {
 738     bs->cyls = cyls;
 739     bs->heads = heads;
 740     bs->secs = secs;
 741 }
 742
 743 void bdrv_set_type_hint(BlockDriverState *bs, int type)
 744 {
 745     bs->type = type;
 746     bs->removable = ((type == BDRV_TYPE_CDROM ||
 747                       type == BDRV_TYPE_FLOPPY));
 748 }
 749
 750 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
 751 {
 752     bs->translation = translation;
 753 }
 754
 755 void bdrv_get_geometry_hint(BlockDriverState *bs,
 756                             int *pcyls, int *pheads, int *psecs)
 757 {
 758     *pcyls = bs->cyls;
 759     *pheads = bs->heads;
 760     *psecs = bs->secs;
 761 }
 762
 763 int bdrv_get_type_hint(BlockDriverState *bs)
 764 {
 765     return bs->type;
 766 }
 767
 768 int bdrv_get_translation_hint(BlockDriverState *bs)
 769 {
 770     return bs->translation;
 771 }
 772
 773 int bdrv_is_removable(BlockDriverState *bs)
 774 {
 775     return bs->removable;
 776 }
 777
 778 int bdrv_is_read_only(BlockDriverState *bs)
 779 {
 780     return bs->read_only;
 781 }
 782
 783 /* XXX: no longer used */
 784 void bdrv_set_change_cb(BlockDriverState *bs,
 785                         void (*change_cb)(void *opaque), void *opaque)
 786 {
 787     bs->change_cb = change_cb;
 788     bs->change_opaque = opaque;
 789 }
 790
 791 int bdrv_is_encrypted(BlockDriverState *bs)
 792 {
 793     if (bs->backing_hd && bs->backing_hd->encrypted)
 794         return 1;
 795     return bs->encrypted;
 796 }
 797
 798 int bdrv_set_key(BlockDriverState *bs, const char *key)
 799 {
 800     int ret;
 801     if (bs->backing_hd && bs->backing_hd->encrypted) {
 802         ret = bdrv_set_key(bs->backing_hd, key);
 803         if (ret < 0)
 804             return ret;
 805         if (!bs->encrypted)
 806             return 0;
 807     }
 808     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
 809         return -1;
 810     return bs->drv->bdrv_set_key(bs, key);
 811 }
 812
 813 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
 814 {
 815     if (!bs->drv) {
 816         buf[0] = '\0';
 817     } else {
 818         pstrcpy(buf, buf_size, bs->drv->format_name);
 819     }
 820 }
 821
 822 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
 823                          void *opaque)
 824 {
 825     BlockDriver *drv;
 826
 827     for (drv = first_drv; drv != NULL; drv = drv->next) {
 828         it(opaque, drv->format_name);
 829     }
 830 }
 831
 832 BlockDriverState *bdrv_find(const char *name)
 833 {
 834     BlockDriverState *bs;
 835
 836     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 837         if (!strcmp(name, bs->device_name))
 838             return bs;
 839     }
 840     return NULL;
 841 }
 842
 843 void bdrv_iterate(void (*it)(void *opaque, const char *name), void *opaque)
 844 {
 845     BlockDriverState *bs;
 846
 847     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 848         it(opaque, bs->device_name);
 849     }
 850 }
 851
 852 const char *bdrv_get_device_name(BlockDriverState *bs)
 853 {
 854     return bs->device_name;
 855 }
 856
 857 void bdrv_flush(BlockDriverState *bs)
 858 {
 859     if (bs->drv->bdrv_flush)
 860         bs->drv->bdrv_flush(bs);
 861     if (bs->backing_hd)
 862         bdrv_flush(bs->backing_hd);
 863 }
 864
 865 #ifndef QEMU_IMG
 866 void bdrv_info(void)
 867 {
 868     BlockDriverState *bs;
 869
 870     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 871         term_printf("%s:", bs->device_name);
 872         term_printf(" type=");
 873         switch(bs->type) {
 874         case BDRV_TYPE_HD:
 875             term_printf("hd");
 876             break;
 877         case BDRV_TYPE_CDROM:
 878             term_printf("cdrom");
 879             break;
 880         case BDRV_TYPE_FLOPPY:
 881             term_printf("floppy");
 882             break;
 883         }
 884         term_printf(" removable=%d", bs->removable);
 885         if (bs->removable) {
 886             term_printf(" locked=%d", bs->locked);
 887         }
 888         if (bs->drv) {
 889             term_printf(" file=");
 890             term_print_filename(bs->filename);
 891             if (bs->backing_file[0] != '\0') {
 892                 term_printf(" backing_file=");
 893                 term_print_filename(bs->backing_file);
 894             }
 895             term_printf(" ro=%d", bs->read_only);
 896             term_printf(" drv=%s", bs->drv->format_name);
 897             if (bs->encrypted)
 898                 term_printf(" encrypted");
 899         } else {
 900             term_printf(" [not inserted]");
 901         }
 902         term_printf("\n");
 903     }
 904 }
 905 #endif
 906
 907 void bdrv_get_backing_filename(BlockDriverState *bs,
 908                                char *filename, int filename_size)
 909 {
 910     if (!bs->backing_hd) {
 911         pstrcpy(filename, filename_size, "");
 912     } else {
 913         pstrcpy(filename, filename_size, bs->backing_file);
 914     }
 915 }
 916
 917 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
 918                           const uint8_t *buf, int nb_sectors)
 919 {
 920     BlockDriver *drv = bs->drv;
 921     if (!drv)
 922         return -ENOMEDIUM;
 923     if (!drv->bdrv_write_compressed)
 924         return -ENOTSUP;
 925     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
 926 }
 927
 928 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 929 {
 930     BlockDriver *drv = bs->drv;
 931     if (!drv)
 932         return -ENOMEDIUM;
 933     if (!drv->bdrv_get_info)
 934         return -ENOTSUP;
 935     memset(bdi, 0, sizeof(*bdi));
 936     return drv->bdrv_get_info(bs, bdi);
 937 }
 938
 939 /**************************************************************/
 940 /* handling of snapshots */
 941
 942 int bdrv_snapshot_create(BlockDriverState *bs,
 943                          QEMUSnapshotInfo *sn_info)
 944 {
 945     BlockDriver *drv = bs->drv;
 946     if (!drv)
 947         return -ENOMEDIUM;
 948     if (!drv->bdrv_snapshot_create)
 949         return -ENOTSUP;
 950     return drv->bdrv_snapshot_create(bs, sn_info);
 951 }
 952
 953 int bdrv_snapshot_goto(BlockDriverState *bs,
 954                        const char *snapshot_id)
 955 {
 956     BlockDriver *drv = bs->drv;
 957     if (!drv)
 958         return -ENOMEDIUM;
 959     if (!drv->bdrv_snapshot_goto)
 960         return -ENOTSUP;
 961     return drv->bdrv_snapshot_goto(bs, snapshot_id);
 962 }
 963
 964 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
 965 {
 966     BlockDriver *drv = bs->drv;
 967     if (!drv)
 968         return -ENOMEDIUM;
 969     if (!drv->bdrv_snapshot_delete)
 970         return -ENOTSUP;
 971     return drv->bdrv_snapshot_delete(bs, snapshot_id);
 972 }
 973
 974 int bdrv_snapshot_list(BlockDriverState *bs,
 975                        QEMUSnapshotInfo **psn_info)
 976 {
 977     BlockDriver *drv = bs->drv;
 978     if (!drv)
 979         return -ENOMEDIUM;
 980     if (!drv->bdrv_snapshot_list)
 981         return -ENOTSUP;
 982     return drv->bdrv_snapshot_list(bs, psn_info);
 983 }
 984
 985 #define NB_SUFFIXES 4
 986
 987 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
 988 {
 989     static const char suffixes[NB_SUFFIXES] = "KMGT";
 990     int64_t base;
 991     int i;
 992
 993     if (size <= 999) {
 994         snprintf(buf, buf_size, "%" PRId64, size);
 995     } else {
 996         base = 1024;
 997         for(i = 0; i < NB_SUFFIXES; i++) {
 998             if (size < (10 * base)) {
 999                 snprintf(buf, buf_size, "%0.1f%c",
1000                          (double)size / base,
1001                          suffixes[i]);
1002                 break;
1003             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1004                 snprintf(buf, buf_size, "%" PRId64 "%c",
1005                          ((size + (base >> 1)) / base),
1006                          suffixes[i]);
1007                 break;
1008             }
1009             base = base * 1024;
1010         }
1011     }
1012     return buf;
1013 }
1014
1015 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1016 {
1017     char buf1[128], date_buf[128], clock_buf[128];
1018 #ifdef _WIN32
1019     struct tm *ptm;
1020 #else
1021     struct tm tm;
1022 #endif
1023     time_t ti;
1024     int64_t secs;
1025
1026     if (!sn) {
1027         snprintf(buf, buf_size,
1028                  "%-10s%-20s%7s%20s%15s",
1029                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1030     } else {
1031         ti = sn->date_sec;
1032 #ifdef _WIN32
1033         ptm = localtime(&ti);
1034         strftime(date_buf, sizeof(date_buf),
1035                  "%Y-%m-%d %H:%M:%S", ptm);
1036 #else
1037         localtime_r(&ti, &tm);
1038         strftime(date_buf, sizeof(date_buf),
1039                  "%Y-%m-%d %H:%M:%S", &tm);
1040 #endif
1041         secs = sn->vm_clock_nsec / 1000000000;
1042         snprintf(clock_buf, sizeof(clock_buf),
1043                  "%02d:%02d:%02d.%03d",
1044                  (int)(secs / 3600),
1045                  (int)((secs / 60) % 60),
1046                  (int)(secs % 60),
1047                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1048         snprintf(buf, buf_size,
1049                  "%-10s%-20s%7s%20s%15s",
1050                  sn->id_str, sn->name,
1051                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1052                  date_buf,
1053                  clock_buf);
1054     }
1055     return buf;
1056 }
1057
1058
1059 /**************************************************************/
1060 /* async I/Os */
1061
1062 BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
1063                                 uint8_t *buf, int nb_sectors,
1064                                 BlockDriverCompletionFunc *cb, void *opaque)
1065 {
1066     BlockDriver *drv = bs->drv;
1067
1068     if (!drv)
1069         return NULL;
1070
1071     /* XXX: we assume that nb_sectors == 0 is suppored by the async read */
1072     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
1073         memcpy(buf, bs->boot_sector_data, 512);
1074         sector_num++;
1075         nb_sectors--;
1076         buf += 512;
1077     }
1078
1079     return drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
1080 }
1081
1082 BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
1083                                  const uint8_t *buf, int nb_sectors,
1084                                  BlockDriverCompletionFunc *cb, void *opaque)
1085 {
1086     BlockDriver *drv = bs->drv;
1087
1088     if (!drv)
1089         return NULL;
1090     if (bs->read_only)
1091         return NULL;
1092     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
1093         memcpy(bs->boot_sector_data, buf, 512);
1094     }
1095
1096     return drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
1097 }
1098
1099 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1100 {
1101     BlockDriver *drv = acb->bs->drv;
1102
1103     drv->bdrv_aio_cancel(acb);
1104 }
1105
1106
1107 /**************************************************************/
1108 /* async block device emulation */
1109
1110 #ifdef QEMU_IMG
1111 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1112         int64_t sector_num, uint8_t *buf, int nb_sectors,
1113         BlockDriverCompletionFunc *cb, void *opaque)
1114 {
1115     int ret;
1116     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1117     cb(opaque, ret);
1118     return NULL;
1119 }
1120
1121 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1122         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1123         BlockDriverCompletionFunc *cb, void *opaque)
1124 {
1125     int ret;
1126     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1127     cb(opaque, ret);
1128     return NULL;
1129 }
1130
1131 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb)
1132 {
1133 }
1134 #else
1135 static void bdrv_aio_bh_cb(void *opaque)
1136 {
1137     BlockDriverAIOCBSync *acb = opaque;
1138     acb->common.cb(acb->common.opaque, acb->ret);
1139     qemu_aio_release(acb);
1140 }
1141
1142 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1143         int64_t sector_num, uint8_t *buf, int nb_sectors,
1144         BlockDriverCompletionFunc *cb, void *opaque)
1145 {
1146     BlockDriverAIOCBSync *acb;
1147     int ret;
1148
1149     acb = qemu_aio_get(bs, cb, opaque);
1150     if (!acb->bh)
1151         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1152     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1153     acb->ret = ret;
1154     qemu_bh_schedule(acb->bh);
1155     return &acb->common;
1156 }
1157
1158 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1159         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1160         BlockDriverCompletionFunc *cb, void *opaque)
1161 {
1162     BlockDriverAIOCBSync *acb;
1163     int ret;
1164
1165     acb = qemu_aio_get(bs, cb, opaque);
1166     if (!acb->bh)
1167         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1168     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1169     acb->ret = ret;
1170     qemu_bh_schedule(acb->bh);
1171     return &acb->common;
1172 }
1173
1174 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1175 {
1176     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1177     qemu_bh_cancel(acb->bh);
1178     qemu_aio_release(acb);
1179 }
1180 #endif /* !QEMU_IMG */
1181
1182 /**************************************************************/
1183 /* sync block device emulation */
1184
1185 static void bdrv_rw_em_cb(void *opaque, int ret)
1186 {
1187     *(int *)opaque = ret;
1188 }
1189
1190 #define NOT_DONE 0x7fffffff
1191
1192 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1193                         uint8_t *buf, int nb_sectors)
1194 {
1195     int async_ret;
1196     BlockDriverAIOCB *acb;
1197
1198     async_ret = NOT_DONE;
1199     qemu_aio_wait_start();
1200     acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
1201                         bdrv_rw_em_cb, &async_ret);
1202     if (acb == NULL) {
1203         qemu_aio_wait_end();
1204         return -1;
1205     }
1206     while (async_ret == NOT_DONE) {
1207         qemu_aio_wait();
1208     }
1209     qemu_aio_wait_end();
1210     return async_ret;
1211 }
1212
1213 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1214                          const uint8_t *buf, int nb_sectors)
1215 {
1216     int async_ret;
1217     BlockDriverAIOCB *acb;
1218
1219     async_ret = NOT_DONE;
1220     qemu_aio_wait_start();
1221     acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
1222                          bdrv_rw_em_cb, &async_ret);
1223     if (acb == NULL) {
1224         qemu_aio_wait_end();
1225         return -1;
1226     }
1227     while (async_ret == NOT_DONE) {
1228         qemu_aio_wait();
1229     }
1230     qemu_aio_wait_end();
1231     return async_ret;
1232 }
1233
1234 void bdrv_init(void)
1235 {
1236     bdrv_register(&bdrv_raw);
1237     bdrv_register(&bdrv_host_device);
1238 #ifndef _WIN32
1239     bdrv_register(&bdrv_cow);
1240 #endif
1241     bdrv_register(&bdrv_qcow);
1242     bdrv_register(&bdrv_vmdk);
1243     bdrv_register(&bdrv_cloop);
1244     bdrv_register(&bdrv_dmg);
1245     bdrv_register(&bdrv_bochs);
1246     bdrv_register(&bdrv_vpc);
1247     bdrv_register(&bdrv_vvfat);
1248     bdrv_register(&bdrv_qcow2);
1249     bdrv_register(&bdrv_parallels);
1250 }
1251
1252 void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
1253                    void *opaque)
1254 {
1255     BlockDriver *drv;
1256     BlockDriverAIOCB *acb;
1257
1258     drv = bs->drv;
1259     if (drv->free_aiocb) {
1260         acb = drv->free_aiocb;
1261         drv->free_aiocb = acb->next;
1262     } else {
1263         acb = qemu_mallocz(drv->aiocb_size);
1264         if (!acb)
1265             return NULL;
1266     }
1267     acb->bs = bs;
1268     acb->cb = cb;
1269     acb->opaque = opaque;
1270     return acb;
1271 }
1272
1273 void qemu_aio_release(void *p)
1274 {
1275     BlockDriverAIOCB *acb = p;
1276     BlockDriver *drv = acb->bs->drv;
1277     acb->next = drv->free_aiocb;
1278     drv->free_aiocb = acb;
1279 }
1280
1281 /**************************************************************/
1282 /* removable device support */
1283
1284 /**
1285  * Return TRUE if the media is present
1286  */
1287 int bdrv_is_inserted(BlockDriverState *bs)
1288 {
1289     BlockDriver *drv = bs->drv;
1290     int ret;
1291     if (!drv)
1292         return 0;
1293     if (!drv->bdrv_is_inserted)
1294         return 1;
1295     ret = drv->bdrv_is_inserted(bs);
1296     return ret;
1297 }
1298
1299 /**
1300  * Return TRUE if the media changed since the last call to this
1301  * function. It is currently only used for floppy disks
1302  */
1303 int bdrv_media_changed(BlockDriverState *bs)
1304 {
1305     BlockDriver *drv = bs->drv;
1306     int ret;
1307
1308     if (!drv || !drv->bdrv_media_changed)
1309         ret = -ENOTSUP;
1310     else
1311         ret = drv->bdrv_media_changed(bs);
1312     if (ret == -ENOTSUP)
1313         ret = bs->media_changed;
1314     bs->media_changed = 0;
1315     return ret;
1316 }
1317
1318 /**
1319  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1320  */
1321 void bdrv_eject(BlockDriverState *bs, int eject_flag)
1322 {
1323     BlockDriver *drv = bs->drv;
1324     int ret;
1325
1326     if (!drv || !drv->bdrv_eject) {
1327         ret = -ENOTSUP;
1328     } else {
1329         ret = drv->bdrv_eject(bs, eject_flag);
1330     }
1331     if (ret == -ENOTSUP) {
1332         if (eject_flag)
1333             bdrv_close(bs);
1334     }
1335 }
1336
1337 int bdrv_is_locked(BlockDriverState *bs)
1338 {
1339     return bs->locked;
1340 }
1341
1342 /**
1343  * Lock or unlock the media (if it is locked, the user won't be able
1344  * to eject it manually).
1345  */
1346 void bdrv_set_locked(BlockDriverState *bs, int locked)
1347 {
1348     BlockDriver *drv = bs->drv;
1349
1350     bs->locked = locked;
1351     if (drv && drv->bdrv_set_locked) {
1352         drv->bdrv_set_locked(bs, locked);
1353     }
1354 }