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             bs->rd_bytes += (unsigned) len;
 526             bs->rd_ops ++;
 527             return 0;
 528         }
 529     } else {
 530         return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 531     }
 532 }
 533
 534 /* Return < 0 if error. Important errors are:
 535   -EIO         generic I/O error (may happen for all errors)
 536   -ENOMEDIUM   No media inserted.
 537   -EINVAL      Invalid sector number or nb_sectors
 538   -EACCES      Trying to write a read-only device
 539 */
 540 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 541                const uint8_t *buf, int nb_sectors)
 542 {
 543     BlockDriver *drv = bs->drv;
 544     if (!bs->drv)
 545         return -ENOMEDIUM;
 546     if (bs->read_only)
 547         return -EACCES;
 548     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
 549         memcpy(bs->boot_sector_data, buf, 512);
 550     }
 551     if (drv->bdrv_pwrite) {
 552         int ret, len;
 553         len = nb_sectors * 512;
 554         ret = drv->bdrv_pwrite(bs, sector_num * 512, buf, len);
 555         if (ret < 0)
 556             return ret;
 557         else if (ret != len)
 558             return -EIO;
 559         else {
 560             bs->wr_bytes += (unsigned) len;
 561             bs->wr_ops ++;
 562             return 0;
 563         }
 564     } else {
 565         return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 566     }
 567 }
 568
 569 static int bdrv_pread_em(BlockDriverState *bs, int64_t offset,
 570                          uint8_t *buf, int count1)
 571 {
 572     uint8_t tmp_buf[SECTOR_SIZE];
 573     int len, nb_sectors, count;
 574     int64_t sector_num;
 575
 576     count = count1;
 577     /* first read to align to sector start */
 578     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 579     if (len > count)
 580         len = count;
 581     sector_num = offset >> SECTOR_BITS;
 582     if (len > 0) {
 583         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 584             return -EIO;
 585         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
 586         count -= len;
 587         if (count == 0)
 588             return count1;
 589         sector_num++;
 590         buf += len;
 591     }
 592
 593     /* read the sectors "in place" */
 594     nb_sectors = count >> SECTOR_BITS;
 595     if (nb_sectors > 0) {
 596         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
 597             return -EIO;
 598         sector_num += nb_sectors;
 599         len = nb_sectors << SECTOR_BITS;
 600         buf += len;
 601         count -= len;
 602     }
 603
 604     /* add data from the last sector */
 605     if (count > 0) {
 606         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 607             return -EIO;
 608         memcpy(buf, tmp_buf, count);
 609     }
 610     return count1;
 611 }
 612
 613 static int bdrv_pwrite_em(BlockDriverState *bs, int64_t offset,
 614                           const uint8_t *buf, int count1)
 615 {
 616     uint8_t tmp_buf[SECTOR_SIZE];
 617     int len, nb_sectors, count;
 618     int64_t sector_num;
 619
 620     count = count1;
 621     /* first write to align to sector start */
 622     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 623     if (len > count)
 624         len = count;
 625     sector_num = offset >> SECTOR_BITS;
 626     if (len > 0) {
 627         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 628             return -EIO;
 629         memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
 630         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 631             return -EIO;
 632         count -= len;
 633         if (count == 0)
 634             return count1;
 635         sector_num++;
 636         buf += len;
 637     }
 638
 639     /* write the sectors "in place" */
 640     nb_sectors = count >> SECTOR_BITS;
 641     if (nb_sectors > 0) {
 642         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
 643             return -EIO;
 644         sector_num += nb_sectors;
 645         len = nb_sectors << SECTOR_BITS;
 646         buf += len;
 647         count -= len;
 648     }
 649
 650     /* add data from the last sector */
 651     if (count > 0) {
 652         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 653             return -EIO;
 654         memcpy(tmp_buf, buf, count);
 655         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 656             return -EIO;
 657     }
 658     return count1;
 659 }
 660
 661 /**
 662  * Read with byte offsets (needed only for file protocols)
 663  */
 664 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 665                void *buf1, int count1)
 666 {
 667     BlockDriver *drv = bs->drv;
 668
 669     if (!drv)
 670         return -ENOMEDIUM;
 671     if (!drv->bdrv_pread)
 672         return bdrv_pread_em(bs, offset, buf1, count1);
 673     return drv->bdrv_pread(bs, offset, buf1, count1);
 674 }
 675
 676 /**
 677  * Write with byte offsets (needed only for file protocols)
 678  */
 679 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 680                 const void *buf1, int count1)
 681 {
 682     BlockDriver *drv = bs->drv;
 683
 684     if (!drv)
 685         return -ENOMEDIUM;
 686     if (!drv->bdrv_pwrite)
 687         return bdrv_pwrite_em(bs, offset, buf1, count1);
 688     return drv->bdrv_pwrite(bs, offset, buf1, count1);
 689 }
 690
 691 /**
 692  * Truncate file to 'offset' bytes (needed only for file protocols)
 693  */
 694 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
 695 {
 696     BlockDriver *drv = bs->drv;
 697     if (!drv)
 698         return -ENOMEDIUM;
 699     if (!drv->bdrv_truncate)
 700         return -ENOTSUP;
 701     return drv->bdrv_truncate(bs, offset);
 702 }
 703
 704 /**
 705  * Length of a file in bytes. Return < 0 if error or unknown.
 706  */
 707 int64_t bdrv_getlength(BlockDriverState *bs)
 708 {
 709     BlockDriver *drv = bs->drv;
 710     if (!drv)
 711         return -ENOMEDIUM;
 712     if (!drv->bdrv_getlength) {
 713         /* legacy mode */
 714         return bs->total_sectors * SECTOR_SIZE;
 715     }
 716     return drv->bdrv_getlength(bs);
 717 }
 718
 719 /* return 0 as number of sectors if no device present or error */
 720 void bdrv_get_geometry(BlockDriverState *bs, int64_t *nb_sectors_ptr)
 721 {
 722     int64_t length;
 723     length = bdrv_getlength(bs);
 724     if (length < 0)
 725         length = 0;
 726     else
 727         length = length >> SECTOR_BITS;
 728     *nb_sectors_ptr = length;
 729 }
 730
 731 /* force a given boot sector. */
 732 void bdrv_set_boot_sector(BlockDriverState *bs, const uint8_t *data, int size)
 733 {
 734     bs->boot_sector_enabled = 1;
 735     if (size > 512)
 736         size = 512;
 737     memcpy(bs->boot_sector_data, data, size);
 738     memset(bs->boot_sector_data + size, 0, 512 - size);
 739 }
 740
 741 void bdrv_set_geometry_hint(BlockDriverState *bs,
 742                             int cyls, int heads, int secs)
 743 {
 744     bs->cyls = cyls;
 745     bs->heads = heads;
 746     bs->secs = secs;
 747 }
 748
 749 void bdrv_set_type_hint(BlockDriverState *bs, int type)
 750 {
 751     bs->type = type;
 752     bs->removable = ((type == BDRV_TYPE_CDROM ||
 753                       type == BDRV_TYPE_FLOPPY));
 754 }
 755
 756 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
 757 {
 758     bs->translation = translation;
 759 }
 760
 761 void bdrv_get_geometry_hint(BlockDriverState *bs,
 762                             int *pcyls, int *pheads, int *psecs)
 763 {
 764     *pcyls = bs->cyls;
 765     *pheads = bs->heads;
 766     *psecs = bs->secs;
 767 }
 768
 769 int bdrv_get_type_hint(BlockDriverState *bs)
 770 {
 771     return bs->type;
 772 }
 773
 774 int bdrv_get_translation_hint(BlockDriverState *bs)
 775 {
 776     return bs->translation;
 777 }
 778
 779 int bdrv_is_removable(BlockDriverState *bs)
 780 {
 781     return bs->removable;
 782 }
 783
 784 int bdrv_is_read_only(BlockDriverState *bs)
 785 {
 786     return bs->read_only;
 787 }
 788
 789 /* XXX: no longer used */
 790 void bdrv_set_change_cb(BlockDriverState *bs,
 791                         void (*change_cb)(void *opaque), void *opaque)
 792 {
 793     bs->change_cb = change_cb;
 794     bs->change_opaque = opaque;
 795 }
 796
 797 int bdrv_is_encrypted(BlockDriverState *bs)
 798 {
 799     if (bs->backing_hd && bs->backing_hd->encrypted)
 800         return 1;
 801     return bs->encrypted;
 802 }
 803
 804 int bdrv_set_key(BlockDriverState *bs, const char *key)
 805 {
 806     int ret;
 807     if (bs->backing_hd && bs->backing_hd->encrypted) {
 808         ret = bdrv_set_key(bs->backing_hd, key);
 809         if (ret < 0)
 810             return ret;
 811         if (!bs->encrypted)
 812             return 0;
 813     }
 814     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
 815         return -1;
 816     return bs->drv->bdrv_set_key(bs, key);
 817 }
 818
 819 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
 820 {
 821     if (!bs->drv) {
 822         buf[0] = '\0';
 823     } else {
 824         pstrcpy(buf, buf_size, bs->drv->format_name);
 825     }
 826 }
 827
 828 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
 829                          void *opaque)
 830 {
 831     BlockDriver *drv;
 832
 833     for (drv = first_drv; drv != NULL; drv = drv->next) {
 834         it(opaque, drv->format_name);
 835     }
 836 }
 837
 838 BlockDriverState *bdrv_find(const char *name)
 839 {
 840     BlockDriverState *bs;
 841
 842     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 843         if (!strcmp(name, bs->device_name))
 844             return bs;
 845     }
 846     return NULL;
 847 }
 848
 849 void bdrv_iterate(void (*it)(void *opaque, const char *name), void *opaque)
 850 {
 851     BlockDriverState *bs;
 852
 853     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 854         it(opaque, bs->device_name);
 855     }
 856 }
 857
 858 const char *bdrv_get_device_name(BlockDriverState *bs)
 859 {
 860     return bs->device_name;
 861 }
 862
 863 void bdrv_flush(BlockDriverState *bs)
 864 {
 865     if (bs->drv->bdrv_flush)
 866         bs->drv->bdrv_flush(bs);
 867     if (bs->backing_hd)
 868         bdrv_flush(bs->backing_hd);
 869 }
 870
 871 #ifndef QEMU_IMG
 872 void bdrv_info(void)
 873 {
 874     BlockDriverState *bs;
 875
 876     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 877         term_printf("%s:", bs->device_name);
 878         term_printf(" type=");
 879         switch(bs->type) {
 880         case BDRV_TYPE_HD:
 881             term_printf("hd");
 882             break;
 883         case BDRV_TYPE_CDROM:
 884             term_printf("cdrom");
 885             break;
 886         case BDRV_TYPE_FLOPPY:
 887             term_printf("floppy");
 888             break;
 889         }
 890         term_printf(" removable=%d", bs->removable);
 891         if (bs->removable) {
 892             term_printf(" locked=%d", bs->locked);
 893         }
 894         if (bs->drv) {
 895             term_printf(" file=");
 896             term_print_filename(bs->filename);
 897             if (bs->backing_file[0] != '\0') {
 898                 term_printf(" backing_file=");
 899                 term_print_filename(bs->backing_file);
 900             }
 901             term_printf(" ro=%d", bs->read_only);
 902             term_printf(" drv=%s", bs->drv->format_name);
 903             if (bs->encrypted)
 904                 term_printf(" encrypted");
 905         } else {
 906             term_printf(" [not inserted]");
 907         }
 908         term_printf("\n");
 909     }
 910 }
 911
 912 /* The "info blockstats" command. */
 913 void bdrv_info_stats (void)
 914 {
 915     BlockDriverState *bs;
 916
 917     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 918         term_printf ("%s:"
 919                      " rd_bytes=%" PRIu64
 920                      " wr_bytes=%" PRIu64
 921                      " rd_operations=%" PRIu64
 922                      " wr_operations=%" PRIu64
 923                      "\n",
 924                      bs->device_name,
 925                      bs->rd_bytes, bs->wr_bytes,
 926                      bs->rd_ops, bs->wr_ops);
 927     }
 928 }
 929 #endif
 930
 931 void bdrv_get_backing_filename(BlockDriverState *bs,
 932                                char *filename, int filename_size)
 933 {
 934     if (!bs->backing_hd) {
 935         pstrcpy(filename, filename_size, "");
 936     } else {
 937         pstrcpy(filename, filename_size, bs->backing_file);
 938     }
 939 }
 940
 941 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
 942                           const uint8_t *buf, int nb_sectors)
 943 {
 944     BlockDriver *drv = bs->drv;
 945     if (!drv)
 946         return -ENOMEDIUM;
 947     if (!drv->bdrv_write_compressed)
 948         return -ENOTSUP;
 949     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
 950 }
 951
 952 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 953 {
 954     BlockDriver *drv = bs->drv;
 955     if (!drv)
 956         return -ENOMEDIUM;
 957     if (!drv->bdrv_get_info)
 958         return -ENOTSUP;
 959     memset(bdi, 0, sizeof(*bdi));
 960     return drv->bdrv_get_info(bs, bdi);
 961 }
 962
 963 /**************************************************************/
 964 /* handling of snapshots */
 965
 966 int bdrv_snapshot_create(BlockDriverState *bs,
 967                          QEMUSnapshotInfo *sn_info)
 968 {
 969     BlockDriver *drv = bs->drv;
 970     if (!drv)
 971         return -ENOMEDIUM;
 972     if (!drv->bdrv_snapshot_create)
 973         return -ENOTSUP;
 974     return drv->bdrv_snapshot_create(bs, sn_info);
 975 }
 976
 977 int bdrv_snapshot_goto(BlockDriverState *bs,
 978                        const char *snapshot_id)
 979 {
 980     BlockDriver *drv = bs->drv;
 981     if (!drv)
 982         return -ENOMEDIUM;
 983     if (!drv->bdrv_snapshot_goto)
 984         return -ENOTSUP;
 985     return drv->bdrv_snapshot_goto(bs, snapshot_id);
 986 }
 987
 988 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
 989 {
 990     BlockDriver *drv = bs->drv;
 991     if (!drv)
 992         return -ENOMEDIUM;
 993     if (!drv->bdrv_snapshot_delete)
 994         return -ENOTSUP;
 995     return drv->bdrv_snapshot_delete(bs, snapshot_id);
 996 }
 997
 998 int bdrv_snapshot_list(BlockDriverState *bs,
 999                        QEMUSnapshotInfo **psn_info)
1000 {
1001     BlockDriver *drv = bs->drv;
1002     if (!drv)
1003         return -ENOMEDIUM;
1004     if (!drv->bdrv_snapshot_list)
1005         return -ENOTSUP;
1006     return drv->bdrv_snapshot_list(bs, psn_info);
1007 }
1008
1009 #define NB_SUFFIXES 4
1010
1011 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1012 {
1013     static const char suffixes[NB_SUFFIXES] = "KMGT";
1014     int64_t base;
1015     int i;
1016
1017     if (size <= 999) {
1018         snprintf(buf, buf_size, "%" PRId64, size);
1019     } else {
1020         base = 1024;
1021         for(i = 0; i < NB_SUFFIXES; i++) {
1022             if (size < (10 * base)) {
1023                 snprintf(buf, buf_size, "%0.1f%c",
1024                          (double)size / base,
1025                          suffixes[i]);
1026                 break;
1027             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1028                 snprintf(buf, buf_size, "%" PRId64 "%c",
1029                          ((size + (base >> 1)) / base),
1030                          suffixes[i]);
1031                 break;
1032             }
1033             base = base * 1024;
1034         }
1035     }
1036     return buf;
1037 }
1038
1039 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1040 {
1041     char buf1[128], date_buf[128], clock_buf[128];
1042 #ifdef _WIN32
1043     struct tm *ptm;
1044 #else
1045     struct tm tm;
1046 #endif
1047     time_t ti;
1048     int64_t secs;
1049
1050     if (!sn) {
1051         snprintf(buf, buf_size,
1052                  "%-10s%-20s%7s%20s%15s",
1053                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1054     } else {
1055         ti = sn->date_sec;
1056 #ifdef _WIN32
1057         ptm = localtime(&ti);
1058         strftime(date_buf, sizeof(date_buf),
1059                  "%Y-%m-%d %H:%M:%S", ptm);
1060 #else
1061         localtime_r(&ti, &tm);
1062         strftime(date_buf, sizeof(date_buf),
1063                  "%Y-%m-%d %H:%M:%S", &tm);
1064 #endif
1065         secs = sn->vm_clock_nsec / 1000000000;
1066         snprintf(clock_buf, sizeof(clock_buf),
1067                  "%02d:%02d:%02d.%03d",
1068                  (int)(secs / 3600),
1069                  (int)((secs / 60) % 60),
1070                  (int)(secs % 60),
1071                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1072         snprintf(buf, buf_size,
1073                  "%-10s%-20s%7s%20s%15s",
1074                  sn->id_str, sn->name,
1075                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1076                  date_buf,
1077                  clock_buf);
1078     }
1079     return buf;
1080 }
1081
1082
1083 /**************************************************************/
1084 /* async I/Os */
1085
1086 BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
1087                                 uint8_t *buf, int nb_sectors,
1088                                 BlockDriverCompletionFunc *cb, void *opaque)
1089 {
1090     BlockDriver *drv = bs->drv;
1091     BlockDriverAIOCB *ret;
1092
1093     if (!drv)
1094         return NULL;
1095
1096     /* XXX: we assume that nb_sectors == 0 is suppored by the async read */
1097     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
1098         memcpy(buf, bs->boot_sector_data, 512);
1099         sector_num++;
1100         nb_sectors--;
1101         buf += 512;
1102     }
1103
1104     ret = drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
1105
1106     if (ret) {
1107         /* Update stats even though technically transfer has not happened. */
1108         bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1109         bs->rd_ops ++;
1110     }
1111
1112     return ret;
1113 }
1114
1115 BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
1116                                  const uint8_t *buf, int nb_sectors,
1117                                  BlockDriverCompletionFunc *cb, void *opaque)
1118 {
1119     BlockDriver *drv = bs->drv;
1120     BlockDriverAIOCB *ret;
1121
1122     if (!drv)
1123         return NULL;
1124     if (bs->read_only)
1125         return NULL;
1126     if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
1127         memcpy(bs->boot_sector_data, buf, 512);
1128     }
1129
1130     ret = drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
1131
1132     if (ret) {
1133         /* Update stats even though technically transfer has not happened. */
1134         bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1135         bs->wr_ops ++;
1136     }
1137
1138     return ret;
1139 }
1140
1141 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1142 {
1143     BlockDriver *drv = acb->bs->drv;
1144
1145     drv->bdrv_aio_cancel(acb);
1146 }
1147
1148
1149 /**************************************************************/
1150 /* async block device emulation */
1151
1152 #ifdef QEMU_IMG
1153 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1154         int64_t sector_num, uint8_t *buf, int nb_sectors,
1155         BlockDriverCompletionFunc *cb, void *opaque)
1156 {
1157     int ret;
1158     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1159     cb(opaque, ret);
1160     return NULL;
1161 }
1162
1163 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1164         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1165         BlockDriverCompletionFunc *cb, void *opaque)
1166 {
1167     int ret;
1168     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1169     cb(opaque, ret);
1170     return NULL;
1171 }
1172
1173 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb)
1174 {
1175 }
1176 #else
1177 static void bdrv_aio_bh_cb(void *opaque)
1178 {
1179     BlockDriverAIOCBSync *acb = opaque;
1180     acb->common.cb(acb->common.opaque, acb->ret);
1181     qemu_aio_release(acb);
1182 }
1183
1184 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1185         int64_t sector_num, uint8_t *buf, int nb_sectors,
1186         BlockDriverCompletionFunc *cb, void *opaque)
1187 {
1188     BlockDriverAIOCBSync *acb;
1189     int ret;
1190
1191     acb = qemu_aio_get(bs, cb, opaque);
1192     if (!acb->bh)
1193         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1194     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1195     acb->ret = ret;
1196     qemu_bh_schedule(acb->bh);
1197     return &acb->common;
1198 }
1199
1200 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1201         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1202         BlockDriverCompletionFunc *cb, void *opaque)
1203 {
1204     BlockDriverAIOCBSync *acb;
1205     int ret;
1206
1207     acb = qemu_aio_get(bs, cb, opaque);
1208     if (!acb->bh)
1209         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1210     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1211     acb->ret = ret;
1212     qemu_bh_schedule(acb->bh);
1213     return &acb->common;
1214 }
1215
1216 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1217 {
1218     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1219     qemu_bh_cancel(acb->bh);
1220     qemu_aio_release(acb);
1221 }
1222 #endif /* !QEMU_IMG */
1223
1224 /**************************************************************/
1225 /* sync block device emulation */
1226
1227 static void bdrv_rw_em_cb(void *opaque, int ret)
1228 {
1229     *(int *)opaque = ret;
1230 }
1231
1232 #define NOT_DONE 0x7fffffff
1233
1234 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1235                         uint8_t *buf, int nb_sectors)
1236 {
1237     int async_ret;
1238     BlockDriverAIOCB *acb;
1239
1240     async_ret = NOT_DONE;
1241     qemu_aio_wait_start();
1242     acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
1243                         bdrv_rw_em_cb, &async_ret);
1244     if (acb == NULL) {
1245         qemu_aio_wait_end();
1246         return -1;
1247     }
1248     while (async_ret == NOT_DONE) {
1249         qemu_aio_wait();
1250     }
1251     qemu_aio_wait_end();
1252     return async_ret;
1253 }
1254
1255 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1256                          const uint8_t *buf, int nb_sectors)
1257 {
1258     int async_ret;
1259     BlockDriverAIOCB *acb;
1260
1261     async_ret = NOT_DONE;
1262     qemu_aio_wait_start();
1263     acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
1264                          bdrv_rw_em_cb, &async_ret);
1265     if (acb == NULL) {
1266         qemu_aio_wait_end();
1267         return -1;
1268     }
1269     while (async_ret == NOT_DONE) {
1270         qemu_aio_wait();
1271     }
1272     qemu_aio_wait_end();
1273     return async_ret;
1274 }
1275
1276 void bdrv_init(void)
1277 {
1278     bdrv_register(&bdrv_raw);
1279     bdrv_register(&bdrv_host_device);
1280 #ifndef _WIN32
1281     bdrv_register(&bdrv_cow);
1282 #endif
1283     bdrv_register(&bdrv_qcow);
1284     bdrv_register(&bdrv_vmdk);
1285     bdrv_register(&bdrv_cloop);
1286     bdrv_register(&bdrv_dmg);
1287     bdrv_register(&bdrv_bochs);
1288     bdrv_register(&bdrv_vpc);
1289     bdrv_register(&bdrv_vvfat);
1290     bdrv_register(&bdrv_qcow2);
1291     bdrv_register(&bdrv_parallels);
1292 }
1293
1294 void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
1295                    void *opaque)
1296 {
1297     BlockDriver *drv;
1298     BlockDriverAIOCB *acb;
1299
1300     drv = bs->drv;
1301     if (drv->free_aiocb) {
1302         acb = drv->free_aiocb;
1303         drv->free_aiocb = acb->next;
1304     } else {
1305         acb = qemu_mallocz(drv->aiocb_size);
1306         if (!acb)
1307             return NULL;
1308     }
1309     acb->bs = bs;
1310     acb->cb = cb;
1311     acb->opaque = opaque;
1312     return acb;
1313 }
1314
1315 void qemu_aio_release(void *p)
1316 {
1317     BlockDriverAIOCB *acb = p;
1318     BlockDriver *drv = acb->bs->drv;
1319     acb->next = drv->free_aiocb;
1320     drv->free_aiocb = acb;
1321 }
1322
1323 /**************************************************************/
1324 /* removable device support */
1325
1326 /**
1327  * Return TRUE if the media is present
1328  */
1329 int bdrv_is_inserted(BlockDriverState *bs)
1330 {
1331     BlockDriver *drv = bs->drv;
1332     int ret;
1333     if (!drv)
1334         return 0;
1335     if (!drv->bdrv_is_inserted)
1336         return 1;
1337     ret = drv->bdrv_is_inserted(bs);
1338     return ret;
1339 }
1340
1341 /**
1342  * Return TRUE if the media changed since the last call to this
1343  * function. It is currently only used for floppy disks
1344  */
1345 int bdrv_media_changed(BlockDriverState *bs)
1346 {
1347     BlockDriver *drv = bs->drv;
1348     int ret;
1349
1350     if (!drv || !drv->bdrv_media_changed)
1351         ret = -ENOTSUP;
1352     else
1353         ret = drv->bdrv_media_changed(bs);
1354     if (ret == -ENOTSUP)
1355         ret = bs->media_changed;
1356     bs->media_changed = 0;
1357     return ret;
1358 }
1359
1360 /**
1361  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1362  */
1363 void bdrv_eject(BlockDriverState *bs, int eject_flag)
1364 {
1365     BlockDriver *drv = bs->drv;
1366     int ret;
1367
1368     if (!drv || !drv->bdrv_eject) {
1369         ret = -ENOTSUP;
1370     } else {
1371         ret = drv->bdrv_eject(bs, eject_flag);
1372     }
1373     if (ret == -ENOTSUP) {
1374         if (eject_flag)
1375             bdrv_close(bs);
1376     }
1377 }
1378
1379 int bdrv_is_locked(BlockDriverState *bs)
1380 {
1381     return bs->locked;
1382 }
1383
1384 /**
1385  * Lock or unlock the media (if it is locked, the user won't be able
1386  * to eject it manually).
1387  */
1388 void bdrv_set_locked(BlockDriverState *bs, int locked)
1389 {
1390     BlockDriver *drv = bs->drv;
1391
1392     bs->locked = locked;
1393     if (drv && drv->bdrv_set_locked) {
1394         drv->bdrv_set_locked(bs, locked);
1395     }
1396 }