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 #include "console.h"
  26 #include "block_int.h"
  27
  28 #ifdef _BSD
  29 #include <sys/types.h>
  30 #include <sys/stat.h>
  31 #include <sys/ioctl.h>
  32 #include <sys/queue.h>
  33 #include <sys/disk.h>
  34 #endif
  35
  36 #define SECTOR_BITS 9
  37 #define SECTOR_SIZE (1 << SECTOR_BITS)
  38
  39 typedef struct BlockDriverAIOCBSync {
  40     BlockDriverAIOCB common;
  41     QEMUBH *bh;
  42     int ret;
  43 } BlockDriverAIOCBSync;
  44
  45 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
  46         int64_t sector_num, uint8_t *buf, int nb_sectors,
  47         BlockDriverCompletionFunc *cb, void *opaque);
  48 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
  49         int64_t sector_num, const uint8_t *buf, int nb_sectors,
  50         BlockDriverCompletionFunc *cb, void *opaque);
  51 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
  52 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
  53                         uint8_t *buf, int nb_sectors);
  54 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
  55                          const uint8_t *buf, int nb_sectors);
  56
  57 BlockDriverState *bdrv_first;
  58
  59 static BlockDriver *first_drv;
  60
  61 int path_is_absolute(const char *path)
  62 {
  63     const char *p;
  64 #ifdef _WIN32
  65     /* specific case for names like: "\\.\d:" */
  66     if (*path == '/' || *path == '\\')
  67         return 1;
  68 #endif
  69     p = strchr(path, ':');
  70     if (p)
  71         p++;
  72     else
  73         p = path;
  74 #ifdef _WIN32
  75     return (*p == '/' || *p == '\\');
  76 #else
  77     return (*p == '/');
  78 #endif
  79 }
  80
  81 /* if filename is absolute, just copy it to dest. Otherwise, build a
  82    path to it by considering it is relative to base_path. URL are
  83    supported. */
  84 void path_combine(char *dest, int dest_size,
  85                   const char *base_path,
  86                   const char *filename)
  87 {
  88     const char *p, *p1;
  89     int len;
  90
  91     if (dest_size <= 0)
  92         return;
  93     if (path_is_absolute(filename)) {
  94         pstrcpy(dest, dest_size, filename);
  95     } else {
  96         p = strchr(base_path, ':');
  97         if (p)
  98             p++;
  99         else
 100             p = base_path;
 101         p1 = strrchr(base_path, '/');
 102 #ifdef _WIN32
 103         {
 104             const char *p2;
 105             p2 = strrchr(base_path, '\\');
 106             if (!p1 || p2 > p1)
 107                 p1 = p2;
 108         }
 109 #endif
 110         if (p1)
 111             p1++;
 112         else
 113             p1 = base_path;
 114         if (p1 > p)
 115             p = p1;
 116         len = p - base_path;
 117         if (len > dest_size - 1)
 118             len = dest_size - 1;
 119         memcpy(dest, base_path, len);
 120         dest[len] = '\0';
 121         pstrcat(dest, dest_size, filename);
 122     }
 123 }
 124
 125
 126 static void bdrv_register(BlockDriver *bdrv)
 127 {
 128     if (!bdrv->bdrv_aio_read) {
 129         /* add AIO emulation layer */
 130         bdrv->bdrv_aio_read = bdrv_aio_read_em;
 131         bdrv->bdrv_aio_write = bdrv_aio_write_em;
 132         bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
 133         bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
 134     } else if (!bdrv->bdrv_read && !bdrv->bdrv_pread) {
 135         /* add synchronous IO emulation layer */
 136         bdrv->bdrv_read = bdrv_read_em;
 137         bdrv->bdrv_write = bdrv_write_em;
 138     }
 139     bdrv->next = first_drv;
 140     first_drv = bdrv;
 141 }
 142
 143 /* create a new block device (by default it is empty) */
 144 BlockDriverState *bdrv_new(const char *device_name)
 145 {
 146     BlockDriverState **pbs, *bs;
 147
 148     bs = qemu_mallocz(sizeof(BlockDriverState));
 149     if(!bs)
 150         return NULL;
 151     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
 152     if (device_name[0] != '\0') {
 153         /* insert at the end */
 154         pbs = &bdrv_first;
 155         while (*pbs != NULL)
 156             pbs = &(*pbs)->next;
 157         *pbs = bs;
 158     }
 159     return bs;
 160 }
 161
 162 BlockDriver *bdrv_find_format(const char *format_name)
 163 {
 164     BlockDriver *drv1;
 165     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 166         if (!strcmp(drv1->format_name, format_name))
 167             return drv1;
 168     }
 169     return NULL;
 170 }
 171
 172 int bdrv_create(BlockDriver *drv,
 173                 const char *filename, int64_t size_in_sectors,
 174                 const char *backing_file, int flags)
 175 {
 176     if (!drv->bdrv_create)
 177         return -ENOTSUP;
 178     return drv->bdrv_create(filename, size_in_sectors, backing_file, flags);
 179 }
 180
 181 #ifdef _WIN32
 182 void get_tmp_filename(char *filename, int size)
 183 {
 184     char temp_dir[MAX_PATH];
 185
 186     GetTempPath(MAX_PATH, temp_dir);
 187     GetTempFileName(temp_dir, "qem", 0, filename);
 188 }
 189 #else
 190 void get_tmp_filename(char *filename, int size)
 191 {
 192     int fd;
 193     const char *tmpdir;
 194     /* XXX: race condition possible */
 195     tmpdir = getenv("TMPDIR");
 196     if (!tmpdir)
 197         tmpdir = "/tmp";
 198     snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
 199     fd = mkstemp(filename);
 200     close(fd);
 201 }
 202 #endif
 203
 204 #ifdef _WIN32
 205 static int is_windows_drive_prefix(const char *filename)
 206 {
 207     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
 208              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
 209             filename[1] == ':');
 210 }
 211
 212 static int is_windows_drive(const char *filename)
 213 {
 214     if (is_windows_drive_prefix(filename) &&
 215         filename[2] == '\0')
 216         return 1;
 217     if (strstart(filename, "\\\\.\\", NULL) ||
 218         strstart(filename, "//./", NULL))
 219         return 1;
 220     return 0;
 221 }
 222 #endif
 223
 224 static BlockDriver *find_protocol(const char *filename)
 225 {
 226     BlockDriver *drv1;
 227     char protocol[128];
 228     int len;
 229     const char *p;
 230
 231 #ifdef _WIN32
 232     if (is_windows_drive(filename) ||
 233         is_windows_drive_prefix(filename))
 234         return &bdrv_raw;
 235 #endif
 236     p = strchr(filename, ':');
 237     if (!p)
 238         return &bdrv_raw;
 239     len = p - filename;
 240     if (len > sizeof(protocol) - 1)
 241         len = sizeof(protocol) - 1;
 242     memcpy(protocol, filename, len);
 243     protocol[len] = '\0';
 244     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 245         if (drv1->protocol_name &&
 246             !strcmp(drv1->protocol_name, protocol))
 247             return drv1;
 248     }
 249     return NULL;
 250 }
 251
 252 /* XXX: force raw format if block or character device ? It would
 253    simplify the BSD case */
 254 static BlockDriver *find_image_format(const char *filename)
 255 {
 256     int ret, score, score_max;
 257     BlockDriver *drv1, *drv;
 258     uint8_t buf[2048];
 259     BlockDriverState *bs;
 260
 261     /* detect host devices. By convention, /dev/cdrom[N] is always
 262        recognized as a host CDROM */
 263     if (strstart(filename, "/dev/cdrom", NULL))
 264         return &bdrv_host_device;
 265 #ifdef _WIN32
 266     if (is_windows_drive(filename))
 267         return &bdrv_host_device;
 268 #else
 269     {
 270         struct stat st;
 271         if (stat(filename, &st) >= 0 &&
 272             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 273             return &bdrv_host_device;
 274         }
 275     }
 276 #endif
 277
 278     drv = find_protocol(filename);
 279     /* no need to test disk image formats for vvfat */
 280     if (drv == &bdrv_vvfat)
 281         return drv;
 282
 283     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
 284     if (ret < 0)
 285         return NULL;
 286     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
 287     bdrv_delete(bs);
 288     if (ret < 0) {
 289         return NULL;
 290     }
 291
 292     score_max = 0;
 293     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 294         if (drv1->bdrv_probe) {
 295             score = drv1->bdrv_probe(buf, ret, filename);
 296             if (score > score_max) {
 297                 score_max = score;
 298                 drv = drv1;
 299             }
 300         }
 301     }
 302     return drv;
 303 }
 304
 305 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
 306 {
 307     BlockDriverState *bs;
 308     int ret;
 309
 310     bs = bdrv_new("");
 311     if (!bs)
 312         return -ENOMEM;
 313     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
 314     if (ret < 0) {
 315         bdrv_delete(bs);
 316         return ret;
 317     }
 318     *pbs = bs;
 319     return 0;
 320 }
 321
 322 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
 323 {
 324     return bdrv_open2(bs, filename, flags, NULL);
 325 }
 326
 327 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
 328                BlockDriver *drv)
 329 {
 330     int ret, open_flags;
 331     char tmp_filename[PATH_MAX];
 332     char backing_filename[PATH_MAX];
 333
 334     bs->read_only = 0;
 335     bs->is_temporary = 0;
 336     bs->encrypted = 0;
 337
 338     if (flags & BDRV_O_SNAPSHOT) {
 339         BlockDriverState *bs1;
 340         int64_t total_size;
 341         int is_protocol = 0;
 342
 343         /* if snapshot, we create a temporary backing file and open it
 344            instead of opening 'filename' directly */
 345
 346         /* if there is a backing file, use it */
 347         bs1 = bdrv_new("");
 348         if (!bs1) {
 349             return -ENOMEM;
 350         }
 351         if (bdrv_open(bs1, filename, 0) < 0) {
 352             bdrv_delete(bs1);
 353             return -1;
 354         }
 355         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
 356
 357         if (bs1->drv && bs1->drv->protocol_name)
 358             is_protocol = 1;
 359
 360         bdrv_delete(bs1);
 361
 362         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
 363
 364         /* Real path is meaningless for protocols */
 365         if (is_protocol)
 366             snprintf(backing_filename, sizeof(backing_filename),
 367                      "%s", filename);
 368         else
 369             realpath(filename, backing_filename);
 370
 371         if (bdrv_create(&bdrv_qcow2, tmp_filename,
 372                         total_size, backing_filename, 0) < 0) {
 373             return -1;
 374         }
 375         filename = tmp_filename;
 376         bs->is_temporary = 1;
 377     }
 378
 379     pstrcpy(bs->filename, sizeof(bs->filename), filename);
 380     if (flags & BDRV_O_FILE) {
 381         drv = find_protocol(filename);
 382         if (!drv)
 383             return -ENOENT;
 384     } else {
 385         if (!drv) {
 386             drv = find_image_format(filename);
 387             if (!drv)
 388                 return -1;
 389         }
 390     }
 391     bs->drv = drv;
 392     bs->opaque = qemu_mallocz(drv->instance_size);
 393     if (bs->opaque == NULL && drv->instance_size > 0)
 394         return -1;
 395     /* Note: for compatibility, we open disk image files as RDWR, and
 396        RDONLY as fallback */
 397     if (!(flags & BDRV_O_FILE))
 398         open_flags = BDRV_O_RDWR | (flags & BDRV_O_CACHE_MASK);
 399     else
 400         open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
 401     ret = drv->bdrv_open(bs, filename, open_flags);
 402     if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
 403         ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
 404         bs->read_only = 1;
 405     }
 406     if (ret < 0) {
 407         qemu_free(bs->opaque);
 408         bs->opaque = NULL;
 409         bs->drv = NULL;
 410         return ret;
 411     }
 412     if (drv->bdrv_getlength) {
 413         bs->total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 414     }
 415 #ifndef _WIN32
 416     if (bs->is_temporary) {
 417         unlink(filename);
 418     }
 419 #endif
 420     if (bs->backing_file[0] != '\0') {
 421         /* if there is a backing file, use it */
 422         bs->backing_hd = bdrv_new("");
 423         if (!bs->backing_hd) {
 424         fail:
 425             bdrv_close(bs);
 426             return -ENOMEM;
 427         }
 428         path_combine(backing_filename, sizeof(backing_filename),
 429                      filename, bs->backing_file);
 430         if (bdrv_open(bs->backing_hd, backing_filename, open_flags) < 0)
 431             goto fail;
 432     }
 433
 434     /* call the change callback */
 435     bs->media_changed = 1;
 436     if (bs->change_cb)
 437         bs->change_cb(bs->change_opaque);
 438
 439     return 0;
 440 }
 441
 442 void bdrv_close(BlockDriverState *bs)
 443 {
 444     if (bs->drv) {
 445         if (bs->backing_hd)
 446             bdrv_delete(bs->backing_hd);
 447         bs->drv->bdrv_close(bs);
 448         qemu_free(bs->opaque);
 449 #ifdef _WIN32
 450         if (bs->is_temporary) {
 451             unlink(bs->filename);
 452         }
 453 #endif
 454         bs->opaque = NULL;
 455         bs->drv = NULL;
 456
 457         /* call the change callback */
 458         bs->media_changed = 1;
 459         if (bs->change_cb)
 460             bs->change_cb(bs->change_opaque);
 461     }
 462 }
 463
 464 void bdrv_delete(BlockDriverState *bs)
 465 {
 466     BlockDriverState **pbs;
 467
 468     pbs = &bdrv_first;
 469     while (*pbs != bs && *pbs != NULL)
 470         pbs = &(*pbs)->next;
 471     if (*pbs == bs)
 472         *pbs = bs->next;
 473
 474     bdrv_close(bs);
 475     qemu_free(bs);
 476 }
 477
 478 /* commit COW file into the raw image */
 479 int bdrv_commit(BlockDriverState *bs)
 480 {
 481     BlockDriver *drv = bs->drv;
 482     int64_t i, total_sectors;
 483     int n, j;
 484     unsigned char sector[512];
 485
 486     if (!drv)
 487         return -ENOMEDIUM;
 488
 489     if (bs->read_only) {
 490         return -EACCES;
 491     }
 492
 493     if (!bs->backing_hd) {
 494         return -ENOTSUP;
 495     }
 496
 497     total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 498     for (i = 0; i < total_sectors;) {
 499         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
 500             for(j = 0; j < n; j++) {
 501                 if (bdrv_read(bs, i, sector, 1) != 0) {
 502                     return -EIO;
 503                 }
 504
 505                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
 506                     return -EIO;
 507                 }
 508                 i++;
 509             }
 510         } else {
 511             i += n;
 512         }
 513     }
 514
 515     if (drv->bdrv_make_empty)
 516         return drv->bdrv_make_empty(bs);
 517
 518     return 0;
 519 }
 520
 521 /* return < 0 if error. See bdrv_write() for the return codes */
 522 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 523               uint8_t *buf, int nb_sectors)
 524 {
 525     BlockDriver *drv = bs->drv;
 526
 527     if (!drv)
 528         return -ENOMEDIUM;
 529
 530     if (drv->bdrv_pread) {
 531         int ret, len;
 532         len = nb_sectors * 512;
 533         ret = drv->bdrv_pread(bs, sector_num * 512, buf, len);
 534         if (ret < 0)
 535             return ret;
 536         else if (ret != len)
 537             return -EINVAL;
 538         else {
 539             bs->rd_bytes += (unsigned) len;
 540             bs->rd_ops ++;
 541             return 0;
 542         }
 543     } else {
 544         return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 545     }
 546 }
 547
 548 /* Return < 0 if error. Important errors are:
 549   -EIO         generic I/O error (may happen for all errors)
 550   -ENOMEDIUM   No media inserted.
 551   -EINVAL      Invalid sector number or nb_sectors
 552   -EACCES      Trying to write a read-only device
 553 */
 554 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 555                const uint8_t *buf, int nb_sectors)
 556 {
 557     BlockDriver *drv = bs->drv;
 558     if (!bs->drv)
 559         return -ENOMEDIUM;
 560     if (bs->read_only)
 561         return -EACCES;
 562     if (drv->bdrv_pwrite) {
 563         int ret, len;
 564         len = nb_sectors * 512;
 565         ret = drv->bdrv_pwrite(bs, sector_num * 512, buf, len);
 566         if (ret < 0)
 567             return ret;
 568         else if (ret != len)
 569             return -EIO;
 570         else {
 571             bs->wr_bytes += (unsigned) len;
 572             bs->wr_ops ++;
 573             return 0;
 574         }
 575     } else {
 576         return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 577     }
 578 }
 579
 580 static int bdrv_pread_em(BlockDriverState *bs, int64_t offset,
 581                          uint8_t *buf, int count1)
 582 {
 583     uint8_t tmp_buf[SECTOR_SIZE];
 584     int len, nb_sectors, count;
 585     int64_t sector_num;
 586
 587     count = count1;
 588     /* first read to align to sector start */
 589     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 590     if (len > count)
 591         len = count;
 592     sector_num = offset >> SECTOR_BITS;
 593     if (len > 0) {
 594         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 595             return -EIO;
 596         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
 597         count -= len;
 598         if (count == 0)
 599             return count1;
 600         sector_num++;
 601         buf += len;
 602     }
 603
 604     /* read the sectors "in place" */
 605     nb_sectors = count >> SECTOR_BITS;
 606     if (nb_sectors > 0) {
 607         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
 608             return -EIO;
 609         sector_num += nb_sectors;
 610         len = nb_sectors << SECTOR_BITS;
 611         buf += len;
 612         count -= len;
 613     }
 614
 615     /* add data from the last sector */
 616     if (count > 0) {
 617         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 618             return -EIO;
 619         memcpy(buf, tmp_buf, count);
 620     }
 621     return count1;
 622 }
 623
 624 static int bdrv_pwrite_em(BlockDriverState *bs, int64_t offset,
 625                           const uint8_t *buf, int count1)
 626 {
 627     uint8_t tmp_buf[SECTOR_SIZE];
 628     int len, nb_sectors, count;
 629     int64_t sector_num;
 630
 631     count = count1;
 632     /* first write to align to sector start */
 633     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 634     if (len > count)
 635         len = count;
 636     sector_num = offset >> SECTOR_BITS;
 637     if (len > 0) {
 638         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 639             return -EIO;
 640         memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
 641         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 642             return -EIO;
 643         count -= len;
 644         if (count == 0)
 645             return count1;
 646         sector_num++;
 647         buf += len;
 648     }
 649
 650     /* write the sectors "in place" */
 651     nb_sectors = count >> SECTOR_BITS;
 652     if (nb_sectors > 0) {
 653         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
 654             return -EIO;
 655         sector_num += nb_sectors;
 656         len = nb_sectors << SECTOR_BITS;
 657         buf += len;
 658         count -= len;
 659     }
 660
 661     /* add data from the last sector */
 662     if (count > 0) {
 663         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 664             return -EIO;
 665         memcpy(tmp_buf, buf, count);
 666         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 667             return -EIO;
 668     }
 669     return count1;
 670 }
 671
 672 /**
 673  * Read with byte offsets (needed only for file protocols)
 674  */
 675 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 676                void *buf1, int count1)
 677 {
 678     BlockDriver *drv = bs->drv;
 679
 680     if (!drv)
 681         return -ENOMEDIUM;
 682     if (!drv->bdrv_pread)
 683         return bdrv_pread_em(bs, offset, buf1, count1);
 684     return drv->bdrv_pread(bs, offset, buf1, count1);
 685 }
 686
 687 /**
 688  * Write with byte offsets (needed only for file protocols)
 689  */
 690 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 691                 const void *buf1, int count1)
 692 {
 693     BlockDriver *drv = bs->drv;
 694
 695     if (!drv)
 696         return -ENOMEDIUM;
 697     if (!drv->bdrv_pwrite)
 698         return bdrv_pwrite_em(bs, offset, buf1, count1);
 699     return drv->bdrv_pwrite(bs, offset, buf1, count1);
 700 }
 701
 702 /**
 703  * Truncate file to 'offset' bytes (needed only for file protocols)
 704  */
 705 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
 706 {
 707     BlockDriver *drv = bs->drv;
 708     if (!drv)
 709         return -ENOMEDIUM;
 710     if (!drv->bdrv_truncate)
 711         return -ENOTSUP;
 712     return drv->bdrv_truncate(bs, offset);
 713 }
 714
 715 /**
 716  * Length of a file in bytes. Return < 0 if error or unknown.
 717  */
 718 int64_t bdrv_getlength(BlockDriverState *bs)
 719 {
 720     BlockDriver *drv = bs->drv;
 721     if (!drv)
 722         return -ENOMEDIUM;
 723     if (!drv->bdrv_getlength) {
 724         /* legacy mode */
 725         return bs->total_sectors * SECTOR_SIZE;
 726     }
 727     return drv->bdrv_getlength(bs);
 728 }
 729
 730 /* return 0 as number of sectors if no device present or error */
 731 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
 732 {
 733     int64_t length;
 734     length = bdrv_getlength(bs);
 735     if (length < 0)
 736         length = 0;
 737     else
 738         length = length >> SECTOR_BITS;
 739     *nb_sectors_ptr = length;
 740 }
 741
 742 void bdrv_set_geometry_hint(BlockDriverState *bs,
 743                             int cyls, int heads, int secs)
 744 {
 745     bs->cyls = cyls;
 746     bs->heads = heads;
 747     bs->secs = secs;
 748 }
 749
 750 void bdrv_set_type_hint(BlockDriverState *bs, int type)
 751 {
 752     bs->type = type;
 753     bs->removable = ((type == BDRV_TYPE_CDROM ||
 754                       type == BDRV_TYPE_FLOPPY));
 755 }
 756
 757 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
 758 {
 759     bs->translation = translation;
 760 }
 761
 762 void bdrv_get_geometry_hint(BlockDriverState *bs,
 763                             int *pcyls, int *pheads, int *psecs)
 764 {
 765     *pcyls = bs->cyls;
 766     *pheads = bs->heads;
 767     *psecs = bs->secs;
 768 }
 769
 770 int bdrv_get_type_hint(BlockDriverState *bs)
 771 {
 772     return bs->type;
 773 }
 774
 775 int bdrv_get_translation_hint(BlockDriverState *bs)
 776 {
 777     return bs->translation;
 778 }
 779
 780 int bdrv_is_removable(BlockDriverState *bs)
 781 {
 782     return bs->removable;
 783 }
 784
 785 int bdrv_is_read_only(BlockDriverState *bs)
 786 {
 787     return bs->read_only;
 788 }
 789
 790 int bdrv_is_sg(BlockDriverState *bs)
 791 {
 792     return bs->sg;
 793 }
 794
 795 /* XXX: no longer used */
 796 void bdrv_set_change_cb(BlockDriverState *bs,
 797                         void (*change_cb)(void *opaque), void *opaque)
 798 {
 799     bs->change_cb = change_cb;
 800     bs->change_opaque = opaque;
 801 }
 802
 803 int bdrv_is_encrypted(BlockDriverState *bs)
 804 {
 805     if (bs->backing_hd && bs->backing_hd->encrypted)
 806         return 1;
 807     return bs->encrypted;
 808 }
 809
 810 int bdrv_set_key(BlockDriverState *bs, const char *key)
 811 {
 812     int ret;
 813     if (bs->backing_hd && bs->backing_hd->encrypted) {
 814         ret = bdrv_set_key(bs->backing_hd, key);
 815         if (ret < 0)
 816             return ret;
 817         if (!bs->encrypted)
 818             return 0;
 819     }
 820     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
 821         return -1;
 822     return bs->drv->bdrv_set_key(bs, key);
 823 }
 824
 825 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
 826 {
 827     if (!bs->drv) {
 828         buf[0] = '\0';
 829     } else {
 830         pstrcpy(buf, buf_size, bs->drv->format_name);
 831     }
 832 }
 833
 834 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
 835                          void *opaque)
 836 {
 837     BlockDriver *drv;
 838
 839     for (drv = first_drv; drv != NULL; drv = drv->next) {
 840         it(opaque, drv->format_name);
 841     }
 842 }
 843
 844 BlockDriverState *bdrv_find(const char *name)
 845 {
 846     BlockDriverState *bs;
 847
 848     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 849         if (!strcmp(name, bs->device_name))
 850             return bs;
 851     }
 852     return NULL;
 853 }
 854
 855 void bdrv_iterate(void (*it)(void *opaque, const char *name), void *opaque)
 856 {
 857     BlockDriverState *bs;
 858
 859     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 860         it(opaque, bs->device_name);
 861     }
 862 }
 863
 864 const char *bdrv_get_device_name(BlockDriverState *bs)
 865 {
 866     return bs->device_name;
 867 }
 868
 869 void bdrv_flush(BlockDriverState *bs)
 870 {
 871     if (bs->drv->bdrv_flush)
 872         bs->drv->bdrv_flush(bs);
 873     if (bs->backing_hd)
 874         bdrv_flush(bs->backing_hd);
 875 }
 876
 877 void bdrv_flush_all(void)
 878 {
 879     BlockDriverState *bs;
 880
 881     for (bs = bdrv_first; bs != NULL; bs = bs->next)
 882         if (bs->drv && !bdrv_is_read_only(bs) &&
 883             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
 884             bdrv_flush(bs);
 885 }
 886
 887 /*
 888  * Returns true iff the specified sector is present in the disk image. Drivers
 889  * not implementing the functionality are assumed to not support backing files,
 890  * hence all their sectors are reported as allocated.
 891  *
 892  * 'pnum' is set to the number of sectors (including and immediately following
 893  * the specified sector) that are known to be in the same
 894  * allocated/unallocated state.
 895  *
 896  * 'nb_sectors' is the max value 'pnum' should be set to.
 897  */
 898 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
 899         int *pnum)
 900 {
 901     int64_t n;
 902     if (!bs->drv->bdrv_is_allocated) {
 903         if (sector_num >= bs->total_sectors) {
 904             *pnum = 0;
 905             return 0;
 906         }
 907         n = bs->total_sectors - sector_num;
 908         *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
 909         return 1;
 910     }
 911     return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
 912 }
 913
 914 void bdrv_info(void)
 915 {
 916     BlockDriverState *bs;
 917
 918     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 919         term_printf("%s:", bs->device_name);
 920         term_printf(" type=");
 921         switch(bs->type) {
 922         case BDRV_TYPE_HD:
 923             term_printf("hd");
 924             break;
 925         case BDRV_TYPE_CDROM:
 926             term_printf("cdrom");
 927             break;
 928         case BDRV_TYPE_FLOPPY:
 929             term_printf("floppy");
 930             break;
 931         }
 932         term_printf(" removable=%d", bs->removable);
 933         if (bs->removable) {
 934             term_printf(" locked=%d", bs->locked);
 935         }
 936         if (bs->drv) {
 937             term_printf(" file=");
 938             term_print_filename(bs->filename);
 939             if (bs->backing_file[0] != '\0') {
 940                 term_printf(" backing_file=");
 941                 term_print_filename(bs->backing_file);
 942             }
 943             term_printf(" ro=%d", bs->read_only);
 944             term_printf(" drv=%s", bs->drv->format_name);
 945             if (bs->encrypted)
 946                 term_printf(" encrypted");
 947         } else {
 948             term_printf(" [not inserted]");
 949         }
 950         term_printf("\n");
 951     }
 952 }
 953
 954 /* The "info blockstats" command. */
 955 void bdrv_info_stats (void)
 956 {
 957     BlockDriverState *bs;
 958
 959     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 960         term_printf ("%s:"
 961                      " rd_bytes=%" PRIu64
 962                      " wr_bytes=%" PRIu64
 963                      " rd_operations=%" PRIu64
 964                      " wr_operations=%" PRIu64
 965                      "\n",
 966                      bs->device_name,
 967                      bs->rd_bytes, bs->wr_bytes,
 968                      bs->rd_ops, bs->wr_ops);
 969     }
 970 }
 971
 972 void bdrv_get_backing_filename(BlockDriverState *bs,
 973                                char *filename, int filename_size)
 974 {
 975     if (!bs->backing_hd) {
 976         pstrcpy(filename, filename_size, "");
 977     } else {
 978         pstrcpy(filename, filename_size, bs->backing_file);
 979     }
 980 }
 981
 982 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
 983                           const uint8_t *buf, int nb_sectors)
 984 {
 985     BlockDriver *drv = bs->drv;
 986     if (!drv)
 987         return -ENOMEDIUM;
 988     if (!drv->bdrv_write_compressed)
 989         return -ENOTSUP;
 990     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
 991 }
 992
 993 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 994 {
 995     BlockDriver *drv = bs->drv;
 996     if (!drv)
 997         return -ENOMEDIUM;
 998     if (!drv->bdrv_get_info)
 999         return -ENOTSUP;
1000     memset(bdi, 0, sizeof(*bdi));
1001     return drv->bdrv_get_info(bs, bdi);
1002 }
1003
1004 /**************************************************************/
1005 /* handling of snapshots */
1006
1007 int bdrv_snapshot_create(BlockDriverState *bs,
1008                          QEMUSnapshotInfo *sn_info)
1009 {
1010     BlockDriver *drv = bs->drv;
1011     if (!drv)
1012         return -ENOMEDIUM;
1013     if (!drv->bdrv_snapshot_create)
1014         return -ENOTSUP;
1015     return drv->bdrv_snapshot_create(bs, sn_info);
1016 }
1017
1018 int bdrv_snapshot_goto(BlockDriverState *bs,
1019                        const char *snapshot_id)
1020 {
1021     BlockDriver *drv = bs->drv;
1022     if (!drv)
1023         return -ENOMEDIUM;
1024     if (!drv->bdrv_snapshot_goto)
1025         return -ENOTSUP;
1026     return drv->bdrv_snapshot_goto(bs, snapshot_id);
1027 }
1028
1029 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1030 {
1031     BlockDriver *drv = bs->drv;
1032     if (!drv)
1033         return -ENOMEDIUM;
1034     if (!drv->bdrv_snapshot_delete)
1035         return -ENOTSUP;
1036     return drv->bdrv_snapshot_delete(bs, snapshot_id);
1037 }
1038
1039 int bdrv_snapshot_list(BlockDriverState *bs,
1040                        QEMUSnapshotInfo **psn_info)
1041 {
1042     BlockDriver *drv = bs->drv;
1043     if (!drv)
1044         return -ENOMEDIUM;
1045     if (!drv->bdrv_snapshot_list)
1046         return -ENOTSUP;
1047     return drv->bdrv_snapshot_list(bs, psn_info);
1048 }
1049
1050 #define NB_SUFFIXES 4
1051
1052 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1053 {
1054     static const char suffixes[NB_SUFFIXES] = "KMGT";
1055     int64_t base;
1056     int i;
1057
1058     if (size <= 999) {
1059         snprintf(buf, buf_size, "%" PRId64, size);
1060     } else {
1061         base = 1024;
1062         for(i = 0; i < NB_SUFFIXES; i++) {
1063             if (size < (10 * base)) {
1064                 snprintf(buf, buf_size, "%0.1f%c",
1065                          (double)size / base,
1066                          suffixes[i]);
1067                 break;
1068             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1069                 snprintf(buf, buf_size, "%" PRId64 "%c",
1070                          ((size + (base >> 1)) / base),
1071                          suffixes[i]);
1072                 break;
1073             }
1074             base = base * 1024;
1075         }
1076     }
1077     return buf;
1078 }
1079
1080 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1081 {
1082     char buf1[128], date_buf[128], clock_buf[128];
1083 #ifdef _WIN32
1084     struct tm *ptm;
1085 #else
1086     struct tm tm;
1087 #endif
1088     time_t ti;
1089     int64_t secs;
1090
1091     if (!sn) {
1092         snprintf(buf, buf_size,
1093                  "%-10s%-20s%7s%20s%15s",
1094                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1095     } else {
1096         ti = sn->date_sec;
1097 #ifdef _WIN32
1098         ptm = localtime(&ti);
1099         strftime(date_buf, sizeof(date_buf),
1100                  "%Y-%m-%d %H:%M:%S", ptm);
1101 #else
1102         localtime_r(&ti, &tm);
1103         strftime(date_buf, sizeof(date_buf),
1104                  "%Y-%m-%d %H:%M:%S", &tm);
1105 #endif
1106         secs = sn->vm_clock_nsec / 1000000000;
1107         snprintf(clock_buf, sizeof(clock_buf),
1108                  "%02d:%02d:%02d.%03d",
1109                  (int)(secs / 3600),
1110                  (int)((secs / 60) % 60),
1111                  (int)(secs % 60),
1112                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1113         snprintf(buf, buf_size,
1114                  "%-10s%-20s%7s%20s%15s",
1115                  sn->id_str, sn->name,
1116                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1117                  date_buf,
1118                  clock_buf);
1119     }
1120     return buf;
1121 }
1122
1123
1124 /**************************************************************/
1125 /* async I/Os */
1126
1127 BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
1128                                 uint8_t *buf, int nb_sectors,
1129                                 BlockDriverCompletionFunc *cb, void *opaque)
1130 {
1131     BlockDriver *drv = bs->drv;
1132     BlockDriverAIOCB *ret;
1133
1134     if (!drv)
1135         return NULL;
1136
1137     ret = drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
1138
1139     if (ret) {
1140         /* Update stats even though technically transfer has not happened. */
1141         bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1142         bs->rd_ops ++;
1143     }
1144
1145     return ret;
1146 }
1147
1148 BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
1149                                  const uint8_t *buf, int nb_sectors,
1150                                  BlockDriverCompletionFunc *cb, void *opaque)
1151 {
1152     BlockDriver *drv = bs->drv;
1153     BlockDriverAIOCB *ret;
1154
1155     if (!drv)
1156         return NULL;
1157     if (bs->read_only)
1158         return NULL;
1159
1160     ret = drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
1161
1162     if (ret) {
1163         /* Update stats even though technically transfer has not happened. */
1164         bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1165         bs->wr_ops ++;
1166     }
1167
1168     return ret;
1169 }
1170
1171 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1172 {
1173     BlockDriver *drv = acb->bs->drv;
1174
1175     drv->bdrv_aio_cancel(acb);
1176 }
1177
1178
1179 /**************************************************************/
1180 /* async block device emulation */
1181
1182 static void bdrv_aio_bh_cb(void *opaque)
1183 {
1184     BlockDriverAIOCBSync *acb = opaque;
1185     acb->common.cb(acb->common.opaque, acb->ret);
1186     qemu_aio_release(acb);
1187 }
1188
1189 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1190         int64_t sector_num, uint8_t *buf, int nb_sectors,
1191         BlockDriverCompletionFunc *cb, void *opaque)
1192 {
1193     BlockDriverAIOCBSync *acb;
1194     int ret;
1195
1196     acb = qemu_aio_get(bs, cb, opaque);
1197     if (!acb->bh)
1198         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1199     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1200     acb->ret = ret;
1201     qemu_bh_schedule(acb->bh);
1202     return &acb->common;
1203 }
1204
1205 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1206         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1207         BlockDriverCompletionFunc *cb, void *opaque)
1208 {
1209     BlockDriverAIOCBSync *acb;
1210     int ret;
1211
1212     acb = qemu_aio_get(bs, cb, opaque);
1213     if (!acb->bh)
1214         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1215     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1216     acb->ret = ret;
1217     qemu_bh_schedule(acb->bh);
1218     return &acb->common;
1219 }
1220
1221 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1222 {
1223     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1224     qemu_bh_cancel(acb->bh);
1225     qemu_aio_release(acb);
1226 }
1227
1228 /**************************************************************/
1229 /* sync block device emulation */
1230
1231 static void bdrv_rw_em_cb(void *opaque, int ret)
1232 {
1233     *(int *)opaque = ret;
1234 }
1235
1236 #define NOT_DONE 0x7fffffff
1237
1238 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1239                         uint8_t *buf, int nb_sectors)
1240 {
1241     int async_ret;
1242     BlockDriverAIOCB *acb;
1243
1244     async_ret = NOT_DONE;
1245     acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
1246                         bdrv_rw_em_cb, &async_ret);
1247     if (acb == NULL)
1248         return -1;
1249
1250     while (async_ret == NOT_DONE) {
1251         qemu_aio_wait();
1252     }
1253
1254     return async_ret;
1255 }
1256
1257 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1258                          const uint8_t *buf, int nb_sectors)
1259 {
1260     int async_ret;
1261     BlockDriverAIOCB *acb;
1262
1263     async_ret = NOT_DONE;
1264     acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
1265                          bdrv_rw_em_cb, &async_ret);
1266     if (acb == NULL)
1267         return -1;
1268     while (async_ret == NOT_DONE) {
1269         qemu_aio_wait();
1270     }
1271     return async_ret;
1272 }
1273
1274 void bdrv_init(void)
1275 {
1276     bdrv_register(&bdrv_raw);
1277     bdrv_register(&bdrv_host_device);
1278 #ifndef _WIN32
1279     bdrv_register(&bdrv_cow);
1280 #endif
1281     bdrv_register(&bdrv_qcow);
1282     bdrv_register(&bdrv_vmdk);
1283     bdrv_register(&bdrv_cloop);
1284     bdrv_register(&bdrv_dmg);
1285     bdrv_register(&bdrv_bochs);
1286     bdrv_register(&bdrv_vpc);
1287     bdrv_register(&bdrv_vvfat);
1288     bdrv_register(&bdrv_qcow2);
1289     bdrv_register(&bdrv_parallels);
1290     bdrv_register(&bdrv_nbd);
1291 }
1292
1293 void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
1294                    void *opaque)
1295 {
1296     BlockDriver *drv;
1297     BlockDriverAIOCB *acb;
1298
1299     drv = bs->drv;
1300     if (drv->free_aiocb) {
1301         acb = drv->free_aiocb;
1302         drv->free_aiocb = acb->next;
1303     } else {
1304         acb = qemu_mallocz(drv->aiocb_size);
1305         if (!acb)
1306             return NULL;
1307     }
1308     acb->bs = bs;
1309     acb->cb = cb;
1310     acb->opaque = opaque;
1311     return acb;
1312 }
1313
1314 void qemu_aio_release(void *p)
1315 {
1316     BlockDriverAIOCB *acb = p;
1317     BlockDriver *drv = acb->bs->drv;
1318     acb->next = drv->free_aiocb;
1319     drv->free_aiocb = acb;
1320 }
1321
1322 /**************************************************************/
1323 /* removable device support */
1324
1325 /**
1326  * Return TRUE if the media is present
1327  */
1328 int bdrv_is_inserted(BlockDriverState *bs)
1329 {
1330     BlockDriver *drv = bs->drv;
1331     int ret;
1332     if (!drv)
1333         return 0;
1334     if (!drv->bdrv_is_inserted)
1335         return 1;
1336     ret = drv->bdrv_is_inserted(bs);
1337     return ret;
1338 }
1339
1340 /**
1341  * Return TRUE if the media changed since the last call to this
1342  * function. It is currently only used for floppy disks
1343  */
1344 int bdrv_media_changed(BlockDriverState *bs)
1345 {
1346     BlockDriver *drv = bs->drv;
1347     int ret;
1348
1349     if (!drv || !drv->bdrv_media_changed)
1350         ret = -ENOTSUP;
1351     else
1352         ret = drv->bdrv_media_changed(bs);
1353     if (ret == -ENOTSUP)
1354         ret = bs->media_changed;
1355     bs->media_changed = 0;
1356     return ret;
1357 }
1358
1359 /**
1360  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1361  */
1362 void bdrv_eject(BlockDriverState *bs, int eject_flag)
1363 {
1364     BlockDriver *drv = bs->drv;
1365     int ret;
1366
1367     if (!drv || !drv->bdrv_eject) {
1368         ret = -ENOTSUP;
1369     } else {
1370         ret = drv->bdrv_eject(bs, eject_flag);
1371     }
1372     if (ret == -ENOTSUP) {
1373         if (eject_flag)
1374             bdrv_close(bs);
1375     }
1376 }
1377
1378 int bdrv_is_locked(BlockDriverState *bs)
1379 {
1380     return bs->locked;
1381 }
1382
1383 /**
1384  * Lock or unlock the media (if it is locked, the user won't be able
1385  * to eject it manually).
1386  */
1387 void bdrv_set_locked(BlockDriverState *bs, int locked)
1388 {
1389     BlockDriver *drv = bs->drv;
1390
1391     bs->locked = locked;
1392     if (drv && drv->bdrv_set_locked) {
1393         drv->bdrv_set_locked(bs, locked);
1394     }
1395 }
1396
1397 /* needed for generic scsi interface */
1398
1399 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1400 {
1401     BlockDriver *drv = bs->drv;
1402
1403     if (drv && drv->bdrv_ioctl)
1404         return drv->bdrv_ioctl(bs, req, buf);
1405     return -ENOTSUP;
1406 }