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 "config-host.h"
  25 #ifdef _BSD
  26 /* include native header before sys-queue.h */
  27 #include <sys/queue.h>
  28 #endif
  29
  30 #include "qemu-common.h"
  31 #include "console.h"
  32 #include "block_int.h"
  33
  34 #ifdef _BSD
  35 #include <sys/types.h>
  36 #include <sys/stat.h>
  37 #include <sys/ioctl.h>
  38 #include <sys/disk.h>
  39 #endif
  40
  41 #define SECTOR_BITS 9
  42 #define SECTOR_SIZE (1 << SECTOR_BITS)
  43
  44 typedef struct BlockDriverAIOCBSync {
  45     BlockDriverAIOCB common;
  46     QEMUBH *bh;
  47     int ret;
  48 } BlockDriverAIOCBSync;
  49
  50 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
  51         int64_t sector_num, uint8_t *buf, int nb_sectors,
  52         BlockDriverCompletionFunc *cb, void *opaque);
  53 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
  54         int64_t sector_num, const uint8_t *buf, int nb_sectors,
  55         BlockDriverCompletionFunc *cb, void *opaque);
  56 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
  57 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
  58                         uint8_t *buf, int nb_sectors);
  59 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
  60                          const uint8_t *buf, int nb_sectors);
  61
  62 BlockDriverState *bdrv_first;
  63
  64 static BlockDriver *first_drv;
  65
  66 int path_is_absolute(const char *path)
  67 {
  68     const char *p;
  69 #ifdef _WIN32
  70     /* specific case for names like: "\\.\d:" */
  71     if (*path == '/' || *path == '\\')
  72         return 1;
  73 #endif
  74     p = strchr(path, ':');
  75     if (p)
  76         p++;
  77     else
  78         p = path;
  79 #ifdef _WIN32
  80     return (*p == '/' || *p == '\\');
  81 #else
  82     return (*p == '/');
  83 #endif
  84 }
  85
  86 /* if filename is absolute, just copy it to dest. Otherwise, build a
  87    path to it by considering it is relative to base_path. URL are
  88    supported. */
  89 void path_combine(char *dest, int dest_size,
  90                   const char *base_path,
  91                   const char *filename)
  92 {
  93     const char *p, *p1;
  94     int len;
  95
  96     if (dest_size <= 0)
  97         return;
  98     if (path_is_absolute(filename)) {
  99         pstrcpy(dest, dest_size, filename);
 100     } else {
 101         p = strchr(base_path, ':');
 102         if (p)
 103             p++;
 104         else
 105             p = base_path;
 106         p1 = strrchr(base_path, '/');
 107 #ifdef _WIN32
 108         {
 109             const char *p2;
 110             p2 = strrchr(base_path, '\\');
 111             if (!p1 || p2 > p1)
 112                 p1 = p2;
 113         }
 114 #endif
 115         if (p1)
 116             p1++;
 117         else
 118             p1 = base_path;
 119         if (p1 > p)
 120             p = p1;
 121         len = p - base_path;
 122         if (len > dest_size - 1)
 123             len = dest_size - 1;
 124         memcpy(dest, base_path, len);
 125         dest[len] = '\0';
 126         pstrcat(dest, dest_size, filename);
 127     }
 128 }
 129
 130
 131 static void bdrv_register(BlockDriver *bdrv)
 132 {
 133     if (!bdrv->bdrv_aio_read) {
 134         /* add AIO emulation layer */
 135         bdrv->bdrv_aio_read = bdrv_aio_read_em;
 136         bdrv->bdrv_aio_write = bdrv_aio_write_em;
 137         bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
 138         bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
 139     } else if (!bdrv->bdrv_read && !bdrv->bdrv_pread) {
 140         /* add synchronous IO emulation layer */
 141         bdrv->bdrv_read = bdrv_read_em;
 142         bdrv->bdrv_write = bdrv_write_em;
 143     }
 144     bdrv->next = first_drv;
 145     first_drv = bdrv;
 146 }
 147
 148 /* create a new block device (by default it is empty) */
 149 BlockDriverState *bdrv_new(const char *device_name)
 150 {
 151     BlockDriverState **pbs, *bs;
 152
 153     bs = qemu_mallocz(sizeof(BlockDriverState));
 154     if(!bs)
 155         return NULL;
 156     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
 157     if (device_name[0] != '\0') {
 158         /* insert at the end */
 159         pbs = &bdrv_first;
 160         while (*pbs != NULL)
 161             pbs = &(*pbs)->next;
 162         *pbs = bs;
 163     }
 164     return bs;
 165 }
 166
 167 BlockDriver *bdrv_find_format(const char *format_name)
 168 {
 169     BlockDriver *drv1;
 170     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 171         if (!strcmp(drv1->format_name, format_name))
 172             return drv1;
 173     }
 174     return NULL;
 175 }
 176
 177 int bdrv_create(BlockDriver *drv,
 178                 const char *filename, int64_t size_in_sectors,
 179                 const char *backing_file, int flags)
 180 {
 181     if (!drv->bdrv_create)
 182         return -ENOTSUP;
 183     return drv->bdrv_create(filename, size_in_sectors, backing_file, flags);
 184 }
 185
 186 #ifdef _WIN32
 187 void get_tmp_filename(char *filename, int size)
 188 {
 189     char temp_dir[MAX_PATH];
 190
 191     GetTempPath(MAX_PATH, temp_dir);
 192     GetTempFileName(temp_dir, "qem", 0, filename);
 193 }
 194 #else
 195 void get_tmp_filename(char *filename, int size)
 196 {
 197     int fd;
 198     const char *tmpdir;
 199     /* XXX: race condition possible */
 200     tmpdir = getenv("TMPDIR");
 201     if (!tmpdir)
 202         tmpdir = "/tmp";
 203     snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
 204     fd = mkstemp(filename);
 205     close(fd);
 206 }
 207 #endif
 208
 209 #ifdef _WIN32
 210 static int is_windows_drive_prefix(const char *filename)
 211 {
 212     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
 213              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
 214             filename[1] == ':');
 215 }
 216
 217 static int is_windows_drive(const char *filename)
 218 {
 219     if (is_windows_drive_prefix(filename) &&
 220         filename[2] == '\0')
 221         return 1;
 222     if (strstart(filename, "\\\\.\\", NULL) ||
 223         strstart(filename, "//./", NULL))
 224         return 1;
 225     return 0;
 226 }
 227 #endif
 228
 229 static BlockDriver *find_protocol(const char *filename)
 230 {
 231     BlockDriver *drv1;
 232     char protocol[128];
 233     int len;
 234     const char *p;
 235
 236 #ifdef _WIN32
 237     if (is_windows_drive(filename) ||
 238         is_windows_drive_prefix(filename))
 239         return &bdrv_raw;
 240 #endif
 241     p = strchr(filename, ':');
 242     if (!p)
 243         return &bdrv_raw;
 244     len = p - filename;
 245     if (len > sizeof(protocol) - 1)
 246         len = sizeof(protocol) - 1;
 247     memcpy(protocol, filename, len);
 248     protocol[len] = '\0';
 249     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 250         if (drv1->protocol_name &&
 251             !strcmp(drv1->protocol_name, protocol))
 252             return drv1;
 253     }
 254     return NULL;
 255 }
 256
 257 /* XXX: force raw format if block or character device ? It would
 258    simplify the BSD case */
 259 static BlockDriver *find_image_format(const char *filename)
 260 {
 261     int ret, score, score_max;
 262     BlockDriver *drv1, *drv;
 263     uint8_t buf[2048];
 264     BlockDriverState *bs;
 265
 266     /* detect host devices. By convention, /dev/cdrom[N] is always
 267        recognized as a host CDROM */
 268     if (strstart(filename, "/dev/cdrom", NULL))
 269         return &bdrv_host_device;
 270 #ifdef _WIN32
 271     if (is_windows_drive(filename))
 272         return &bdrv_host_device;
 273 #else
 274     {
 275         struct stat st;
 276         if (stat(filename, &st) >= 0 &&
 277             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 278             return &bdrv_host_device;
 279         }
 280     }
 281 #endif
 282
 283     drv = find_protocol(filename);
 284     /* no need to test disk image formats for vvfat */
 285     if (drv == &bdrv_vvfat)
 286         return drv;
 287
 288     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
 289     if (ret < 0)
 290         return NULL;
 291     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
 292     bdrv_delete(bs);
 293     if (ret < 0) {
 294         return NULL;
 295     }
 296
 297     score_max = 0;
 298     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 299         if (drv1->bdrv_probe) {
 300             score = drv1->bdrv_probe(buf, ret, filename);
 301             if (score > score_max) {
 302                 score_max = score;
 303                 drv = drv1;
 304             }
 305         }
 306     }
 307     return drv;
 308 }
 309
 310 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
 311 {
 312     BlockDriverState *bs;
 313     int ret;
 314
 315     bs = bdrv_new("");
 316     if (!bs)
 317         return -ENOMEM;
 318     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
 319     if (ret < 0) {
 320         bdrv_delete(bs);
 321         return ret;
 322     }
 323     *pbs = bs;
 324     return 0;
 325 }
 326
 327 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
 328 {
 329     return bdrv_open2(bs, filename, flags, NULL);
 330 }
 331
 332 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
 333                BlockDriver *drv)
 334 {
 335     int ret, open_flags;
 336     char tmp_filename[PATH_MAX];
 337     char backing_filename[PATH_MAX];
 338
 339     bs->read_only = 0;
 340     bs->is_temporary = 0;
 341     bs->encrypted = 0;
 342
 343     if (flags & BDRV_O_SNAPSHOT) {
 344         BlockDriverState *bs1;
 345         int64_t total_size;
 346         int is_protocol = 0;
 347
 348         /* if snapshot, we create a temporary backing file and open it
 349            instead of opening 'filename' directly */
 350
 351         /* if there is a backing file, use it */
 352         bs1 = bdrv_new("");
 353         if (!bs1) {
 354             return -ENOMEM;
 355         }
 356         if (bdrv_open(bs1, filename, 0) < 0) {
 357             bdrv_delete(bs1);
 358             return -1;
 359         }
 360         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
 361
 362         if (bs1->drv && bs1->drv->protocol_name)
 363             is_protocol = 1;
 364
 365         bdrv_delete(bs1);
 366
 367         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
 368
 369         /* Real path is meaningless for protocols */
 370         if (is_protocol)
 371             snprintf(backing_filename, sizeof(backing_filename),
 372                      "%s", filename);
 373         else
 374             realpath(filename, backing_filename);
 375
 376         if (bdrv_create(&bdrv_qcow2, tmp_filename,
 377                         total_size, backing_filename, 0) < 0) {
 378             return -1;
 379         }
 380         filename = tmp_filename;
 381         bs->is_temporary = 1;
 382     }
 383
 384     pstrcpy(bs->filename, sizeof(bs->filename), filename);
 385     if (flags & BDRV_O_FILE) {
 386         drv = find_protocol(filename);
 387         if (!drv)
 388             return -ENOENT;
 389     } else {
 390         if (!drv) {
 391             drv = find_image_format(filename);
 392             if (!drv)
 393                 return -1;
 394         }
 395     }
 396     bs->drv = drv;
 397     bs->opaque = qemu_mallocz(drv->instance_size);
 398     if (bs->opaque == NULL && drv->instance_size > 0)
 399         return -1;
 400     /* Note: for compatibility, we open disk image files as RDWR, and
 401        RDONLY as fallback */
 402     if (!(flags & BDRV_O_FILE))
 403         open_flags = BDRV_O_RDWR | (flags & BDRV_O_CACHE_MASK);
 404     else
 405         open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
 406     ret = drv->bdrv_open(bs, filename, open_flags);
 407     if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
 408         ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
 409         bs->read_only = 1;
 410     }
 411     if (ret < 0) {
 412         qemu_free(bs->opaque);
 413         bs->opaque = NULL;
 414         bs->drv = NULL;
 415         return ret;
 416     }
 417     if (drv->bdrv_getlength) {
 418         bs->total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 419     }
 420 #ifndef _WIN32
 421     if (bs->is_temporary) {
 422         unlink(filename);
 423     }
 424 #endif
 425     if (bs->backing_file[0] != '\0') {
 426         /* if there is a backing file, use it */
 427         bs->backing_hd = bdrv_new("");
 428         if (!bs->backing_hd) {
 429         fail:
 430             bdrv_close(bs);
 431             return -ENOMEM;
 432         }
 433         path_combine(backing_filename, sizeof(backing_filename),
 434                      filename, bs->backing_file);
 435         if (bdrv_open(bs->backing_hd, backing_filename, open_flags) < 0)
 436             goto fail;
 437     }
 438
 439     /* call the change callback */
 440     bs->media_changed = 1;
 441     if (bs->change_cb)
 442         bs->change_cb(bs->change_opaque);
 443
 444     return 0;
 445 }
 446
 447 void bdrv_close(BlockDriverState *bs)
 448 {
 449     if (bs->drv) {
 450         if (bs->backing_hd)
 451             bdrv_delete(bs->backing_hd);
 452         bs->drv->bdrv_close(bs);
 453         qemu_free(bs->opaque);
 454 #ifdef _WIN32
 455         if (bs->is_temporary) {
 456             unlink(bs->filename);
 457         }
 458 #endif
 459         bs->opaque = NULL;
 460         bs->drv = NULL;
 461
 462         /* call the change callback */
 463         bs->media_changed = 1;
 464         if (bs->change_cb)
 465             bs->change_cb(bs->change_opaque);
 466     }
 467 }
 468
 469 void bdrv_delete(BlockDriverState *bs)
 470 {
 471     BlockDriverState **pbs;
 472
 473     pbs = &bdrv_first;
 474     while (*pbs != bs && *pbs != NULL)
 475         pbs = &(*pbs)->next;
 476     if (*pbs == bs)
 477         *pbs = bs->next;
 478
 479     bdrv_close(bs);
 480     qemu_free(bs);
 481 }
 482
 483 /* commit COW file into the raw image */
 484 int bdrv_commit(BlockDriverState *bs)
 485 {
 486     BlockDriver *drv = bs->drv;
 487     int64_t i, total_sectors;
 488     int n, j;
 489     unsigned char sector[512];
 490
 491     if (!drv)
 492         return -ENOMEDIUM;
 493
 494     if (bs->read_only) {
 495         return -EACCES;
 496     }
 497
 498     if (!bs->backing_hd) {
 499         return -ENOTSUP;
 500     }
 501
 502     total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 503     for (i = 0; i < total_sectors;) {
 504         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
 505             for(j = 0; j < n; j++) {
 506                 if (bdrv_read(bs, i, sector, 1) != 0) {
 507                     return -EIO;
 508                 }
 509
 510                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
 511                     return -EIO;
 512                 }
 513                 i++;
 514             }
 515         } else {
 516             i += n;
 517         }
 518     }
 519
 520     if (drv->bdrv_make_empty)
 521         return drv->bdrv_make_empty(bs);
 522
 523     return 0;
 524 }
 525
 526 /* return < 0 if error. See bdrv_write() for the return codes */
 527 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 528               uint8_t *buf, int nb_sectors)
 529 {
 530     BlockDriver *drv = bs->drv;
 531
 532     if (!drv)
 533         return -ENOMEDIUM;
 534
 535     if (drv->bdrv_pread) {
 536         int ret, len;
 537         len = nb_sectors * 512;
 538         ret = drv->bdrv_pread(bs, sector_num * 512, buf, len);
 539         if (ret < 0)
 540             return ret;
 541         else if (ret != len)
 542             return -EINVAL;
 543         else {
 544             bs->rd_bytes += (unsigned) len;
 545             bs->rd_ops ++;
 546             return 0;
 547         }
 548     } else {
 549         return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 550     }
 551 }
 552
 553 /* Return < 0 if error. Important errors are:
 554   -EIO         generic I/O error (may happen for all errors)
 555   -ENOMEDIUM   No media inserted.
 556   -EINVAL      Invalid sector number or nb_sectors
 557   -EACCES      Trying to write a read-only device
 558 */
 559 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 560                const uint8_t *buf, int nb_sectors)
 561 {
 562     BlockDriver *drv = bs->drv;
 563     if (!bs->drv)
 564         return -ENOMEDIUM;
 565     if (bs->read_only)
 566         return -EACCES;
 567     if (drv->bdrv_pwrite) {
 568         int ret, len, count = 0;
 569         len = nb_sectors * 512;
 570         do {
 571             ret = drv->bdrv_pwrite(bs, sector_num * 512, buf, len - count);
 572             if (ret < 0) {
 573                 printf("bdrv_write ret=%d\n", ret);
 574                 return ret;
 575             }
 576             count += ret;
 577             buf += ret;
 578         } while (count != len);
 579         bs->wr_bytes += (unsigned) len;
 580         bs->wr_ops ++;
 581         return 0;
 582     }
 583     return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 584 }
 585
 586 static int bdrv_pread_em(BlockDriverState *bs, int64_t offset,
 587                          uint8_t *buf, int count1)
 588 {
 589     uint8_t tmp_buf[SECTOR_SIZE];
 590     int len, nb_sectors, count;
 591     int64_t sector_num;
 592
 593     count = count1;
 594     /* first read to align to sector start */
 595     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 596     if (len > count)
 597         len = count;
 598     sector_num = offset >> SECTOR_BITS;
 599     if (len > 0) {
 600         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 601             return -EIO;
 602         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
 603         count -= len;
 604         if (count == 0)
 605             return count1;
 606         sector_num++;
 607         buf += len;
 608     }
 609
 610     /* read the sectors "in place" */
 611     nb_sectors = count >> SECTOR_BITS;
 612     if (nb_sectors > 0) {
 613         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
 614             return -EIO;
 615         sector_num += nb_sectors;
 616         len = nb_sectors << SECTOR_BITS;
 617         buf += len;
 618         count -= len;
 619     }
 620
 621     /* add data from the last sector */
 622     if (count > 0) {
 623         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 624             return -EIO;
 625         memcpy(buf, tmp_buf, count);
 626     }
 627     return count1;
 628 }
 629
 630 static int bdrv_pwrite_em(BlockDriverState *bs, int64_t offset,
 631                           const uint8_t *buf, int count1)
 632 {
 633     uint8_t tmp_buf[SECTOR_SIZE];
 634     int len, nb_sectors, count;
 635     int64_t sector_num;
 636
 637     count = count1;
 638     /* first write to align to sector start */
 639     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 640     if (len > count)
 641         len = count;
 642     sector_num = offset >> SECTOR_BITS;
 643     if (len > 0) {
 644         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 645             return -EIO;
 646         memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
 647         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 648             return -EIO;
 649         count -= len;
 650         if (count == 0)
 651             return count1;
 652         sector_num++;
 653         buf += len;
 654     }
 655
 656     /* write the sectors "in place" */
 657     nb_sectors = count >> SECTOR_BITS;
 658     if (nb_sectors > 0) {
 659         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
 660             return -EIO;
 661         sector_num += nb_sectors;
 662         len = nb_sectors << SECTOR_BITS;
 663         buf += len;
 664         count -= len;
 665     }
 666
 667     /* add data from the last sector */
 668     if (count > 0) {
 669         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 670             return -EIO;
 671         memcpy(tmp_buf, buf, count);
 672         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 673             return -EIO;
 674     }
 675     return count1;
 676 }
 677
 678 /**
 679  * Read with byte offsets (needed only for file protocols)
 680  */
 681 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 682                void *buf1, int count1)
 683 {
 684     BlockDriver *drv = bs->drv;
 685
 686     if (!drv)
 687         return -ENOMEDIUM;
 688     if (!drv->bdrv_pread)
 689         return bdrv_pread_em(bs, offset, buf1, count1);
 690     return drv->bdrv_pread(bs, offset, buf1, count1);
 691 }
 692
 693 /**
 694  * Write with byte offsets (needed only for file protocols)
 695  */
 696 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 697                 const void *buf1, int count1)
 698 {
 699     BlockDriver *drv = bs->drv;
 700
 701     if (!drv)
 702         return -ENOMEDIUM;
 703     if (!drv->bdrv_pwrite)
 704         return bdrv_pwrite_em(bs, offset, buf1, count1);
 705     return drv->bdrv_pwrite(bs, offset, buf1, count1);
 706 }
 707
 708 /**
 709  * Truncate file to 'offset' bytes (needed only for file protocols)
 710  */
 711 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
 712 {
 713     BlockDriver *drv = bs->drv;
 714     if (!drv)
 715         return -ENOMEDIUM;
 716     if (!drv->bdrv_truncate)
 717         return -ENOTSUP;
 718     return drv->bdrv_truncate(bs, offset);
 719 }
 720
 721 /**
 722  * Length of a file in bytes. Return < 0 if error or unknown.
 723  */
 724 int64_t bdrv_getlength(BlockDriverState *bs)
 725 {
 726     BlockDriver *drv = bs->drv;
 727     if (!drv)
 728         return -ENOMEDIUM;
 729     if (!drv->bdrv_getlength) {
 730         /* legacy mode */
 731         return bs->total_sectors * SECTOR_SIZE;
 732     }
 733     return drv->bdrv_getlength(bs);
 734 }
 735
 736 /* return 0 as number of sectors if no device present or error */
 737 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
 738 {
 739     int64_t length;
 740     length = bdrv_getlength(bs);
 741     if (length < 0)
 742         length = 0;
 743     else
 744         length = length >> SECTOR_BITS;
 745     *nb_sectors_ptr = length;
 746 }
 747
 748 struct partition {
 749         uint8_t boot_ind;           /* 0x80 - active */
 750         uint8_t head;               /* starting head */
 751         uint8_t sector;             /* starting sector */
 752         uint8_t cyl;                /* starting cylinder */
 753         uint8_t sys_ind;            /* What partition type */
 754         uint8_t end_head;           /* end head */
 755         uint8_t end_sector;         /* end sector */
 756         uint8_t end_cyl;            /* end cylinder */
 757         uint32_t start_sect;        /* starting sector counting from 0 */
 758         uint32_t nr_sects;          /* nr of sectors in partition */
 759 } __attribute__((packed));
 760
 761 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
 762 static int guess_disk_lchs(BlockDriverState *bs,
 763                            int *pcylinders, int *pheads, int *psectors)
 764 {
 765     uint8_t buf[512];
 766     int ret, i, heads, sectors, cylinders;
 767     struct partition *p;
 768     uint32_t nr_sects;
 769     uint64_t nb_sectors;
 770
 771     bdrv_get_geometry(bs, &nb_sectors);
 772
 773     ret = bdrv_read(bs, 0, buf, 1);
 774     if (ret < 0)
 775         return -1;
 776     /* test msdos magic */
 777     if (buf[510] != 0x55 || buf[511] != 0xaa)
 778         return -1;
 779     for(i = 0; i < 4; i++) {
 780         p = ((struct partition *)(buf + 0x1be)) + i;
 781         nr_sects = le32_to_cpu(p->nr_sects);
 782         if (nr_sects && p->end_head) {
 783             /* We make the assumption that the partition terminates on
 784                a cylinder boundary */
 785             heads = p->end_head + 1;
 786             sectors = p->end_sector & 63;
 787             if (sectors == 0)
 788                 continue;
 789             cylinders = nb_sectors / (heads * sectors);
 790             if (cylinders < 1 || cylinders > 16383)
 791                 continue;
 792             *pheads = heads;
 793             *psectors = sectors;
 794             *pcylinders = cylinders;
 795 #if 0
 796             printf("guessed geometry: LCHS=%d %d %d\n",
 797                    cylinders, heads, sectors);
 798 #endif
 799             return 0;
 800         }
 801     }
 802     return -1;
 803 }
 804
 805 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
 806 {
 807     int translation, lba_detected = 0;
 808     int cylinders, heads, secs;
 809     uint64_t nb_sectors;
 810
 811     /* if a geometry hint is available, use it */
 812     bdrv_get_geometry(bs, &nb_sectors);
 813     bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
 814     translation = bdrv_get_translation_hint(bs);
 815     if (cylinders != 0) {
 816         *pcyls = cylinders;
 817         *pheads = heads;
 818         *psecs = secs;
 819     } else {
 820         if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
 821             if (heads > 16) {
 822                 /* if heads > 16, it means that a BIOS LBA
 823                    translation was active, so the default
 824                    hardware geometry is OK */
 825                 lba_detected = 1;
 826                 goto default_geometry;
 827             } else {
 828                 *pcyls = cylinders;
 829                 *pheads = heads;
 830                 *psecs = secs;
 831                 /* disable any translation to be in sync with
 832                    the logical geometry */
 833                 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
 834                     bdrv_set_translation_hint(bs,
 835                                               BIOS_ATA_TRANSLATION_NONE);
 836                 }
 837             }
 838         } else {
 839         default_geometry:
 840             /* if no geometry, use a standard physical disk geometry */
 841             cylinders = nb_sectors / (16 * 63);
 842
 843             if (cylinders > 16383)
 844                 cylinders = 16383;
 845             else if (cylinders < 2)
 846                 cylinders = 2;
 847             *pcyls = cylinders;
 848             *pheads = 16;
 849             *psecs = 63;
 850             if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
 851                 if ((*pcyls * *pheads) <= 131072) {
 852                     bdrv_set_translation_hint(bs,
 853                                               BIOS_ATA_TRANSLATION_LARGE);
 854                 } else {
 855                     bdrv_set_translation_hint(bs,
 856                                               BIOS_ATA_TRANSLATION_LBA);
 857                 }
 858             }
 859         }
 860         bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
 861     }
 862 }
 863
 864 void bdrv_set_geometry_hint(BlockDriverState *bs,
 865                             int cyls, int heads, int secs)
 866 {
 867     bs->cyls = cyls;
 868     bs->heads = heads;
 869     bs->secs = secs;
 870 }
 871
 872 void bdrv_set_type_hint(BlockDriverState *bs, int type)
 873 {
 874     bs->type = type;
 875     bs->removable = ((type == BDRV_TYPE_CDROM ||
 876                       type == BDRV_TYPE_FLOPPY));
 877 }
 878
 879 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
 880 {
 881     bs->translation = translation;
 882 }
 883
 884 void bdrv_get_geometry_hint(BlockDriverState *bs,
 885                             int *pcyls, int *pheads, int *psecs)
 886 {
 887     *pcyls = bs->cyls;
 888     *pheads = bs->heads;
 889     *psecs = bs->secs;
 890 }
 891
 892 int bdrv_get_type_hint(BlockDriverState *bs)
 893 {
 894     return bs->type;
 895 }
 896
 897 int bdrv_get_translation_hint(BlockDriverState *bs)
 898 {
 899     return bs->translation;
 900 }
 901
 902 int bdrv_is_removable(BlockDriverState *bs)
 903 {
 904     return bs->removable;
 905 }
 906
 907 int bdrv_is_read_only(BlockDriverState *bs)
 908 {
 909     return bs->read_only;
 910 }
 911
 912 int bdrv_is_sg(BlockDriverState *bs)
 913 {
 914     return bs->sg;
 915 }
 916
 917 /* XXX: no longer used */
 918 void bdrv_set_change_cb(BlockDriverState *bs,
 919                         void (*change_cb)(void *opaque), void *opaque)
 920 {
 921     bs->change_cb = change_cb;
 922     bs->change_opaque = opaque;
 923 }
 924
 925 int bdrv_is_encrypted(BlockDriverState *bs)
 926 {
 927     if (bs->backing_hd && bs->backing_hd->encrypted)
 928         return 1;
 929     return bs->encrypted;
 930 }
 931
 932 int bdrv_set_key(BlockDriverState *bs, const char *key)
 933 {
 934     int ret;
 935     if (bs->backing_hd && bs->backing_hd->encrypted) {
 936         ret = bdrv_set_key(bs->backing_hd, key);
 937         if (ret < 0)
 938             return ret;
 939         if (!bs->encrypted)
 940             return 0;
 941     }
 942     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
 943         return -1;
 944     return bs->drv->bdrv_set_key(bs, key);
 945 }
 946
 947 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
 948 {
 949     if (!bs->drv) {
 950         buf[0] = '\0';
 951     } else {
 952         pstrcpy(buf, buf_size, bs->drv->format_name);
 953     }
 954 }
 955
 956 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
 957                          void *opaque)
 958 {
 959     BlockDriver *drv;
 960
 961     for (drv = first_drv; drv != NULL; drv = drv->next) {
 962         it(opaque, drv->format_name);
 963     }
 964 }
 965
 966 BlockDriverState *bdrv_find(const char *name)
 967 {
 968     BlockDriverState *bs;
 969
 970     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 971         if (!strcmp(name, bs->device_name))
 972             return bs;
 973     }
 974     return NULL;
 975 }
 976
 977 void bdrv_iterate(void (*it)(void *opaque, const char *name), void *opaque)
 978 {
 979     BlockDriverState *bs;
 980
 981     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 982         it(opaque, bs->device_name);
 983     }
 984 }
 985
 986 const char *bdrv_get_device_name(BlockDriverState *bs)
 987 {
 988     return bs->device_name;
 989 }
 990
 991 void bdrv_flush(BlockDriverState *bs)
 992 {
 993     if (bs->drv->bdrv_flush)
 994         bs->drv->bdrv_flush(bs);
 995     if (bs->backing_hd)
 996         bdrv_flush(bs->backing_hd);
 997 }
 998
 999 void bdrv_flush_all(void)
1000 {
1001     BlockDriverState *bs;
1002
1003     for (bs = bdrv_first; bs != NULL; bs = bs->next)
1004         if (bs->drv && !bdrv_is_read_only(bs) &&
1005             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1006             bdrv_flush(bs);
1007 }
1008
1009 /*
1010  * Returns true iff the specified sector is present in the disk image. Drivers
1011  * not implementing the functionality are assumed to not support backing files,
1012  * hence all their sectors are reported as allocated.
1013  *
1014  * 'pnum' is set to the number of sectors (including and immediately following
1015  * the specified sector) that are known to be in the same
1016  * allocated/unallocated state.
1017  *
1018  * 'nb_sectors' is the max value 'pnum' should be set to.
1019  */
1020 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1021         int *pnum)
1022 {
1023     int64_t n;
1024     if (!bs->drv->bdrv_is_allocated) {
1025         if (sector_num >= bs->total_sectors) {
1026             *pnum = 0;
1027             return 0;
1028         }
1029         n = bs->total_sectors - sector_num;
1030         *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1031         return 1;
1032     }
1033     return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1034 }
1035
1036 void bdrv_info(void)
1037 {
1038     BlockDriverState *bs;
1039
1040     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1041         term_printf("%s:", bs->device_name);
1042         term_printf(" type=");
1043         switch(bs->type) {
1044         case BDRV_TYPE_HD:
1045             term_printf("hd");
1046             break;
1047         case BDRV_TYPE_CDROM:
1048             term_printf("cdrom");
1049             break;
1050         case BDRV_TYPE_FLOPPY:
1051             term_printf("floppy");
1052             break;
1053         }
1054         term_printf(" removable=%d", bs->removable);
1055         if (bs->removable) {
1056             term_printf(" locked=%d", bs->locked);
1057         }
1058         if (bs->drv) {
1059             term_printf(" file=");
1060             term_print_filename(bs->filename);
1061             if (bs->backing_file[0] != '\0') {
1062                 term_printf(" backing_file=");
1063                 term_print_filename(bs->backing_file);
1064             }
1065             term_printf(" ro=%d", bs->read_only);
1066             term_printf(" drv=%s", bs->drv->format_name);
1067             if (bs->encrypted)
1068                 term_printf(" encrypted");
1069         } else {
1070             term_printf(" [not inserted]");
1071         }
1072         term_printf("\n");
1073     }
1074 }
1075
1076 /* The "info blockstats" command. */
1077 void bdrv_info_stats (void)
1078 {
1079     BlockDriverState *bs;
1080
1081     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1082         term_printf ("%s:"
1083                      " rd_bytes=%" PRIu64
1084                      " wr_bytes=%" PRIu64
1085                      " rd_operations=%" PRIu64
1086                      " wr_operations=%" PRIu64
1087                      "\n",
1088                      bs->device_name,
1089                      bs->rd_bytes, bs->wr_bytes,
1090                      bs->rd_ops, bs->wr_ops);
1091     }
1092 }
1093
1094 void bdrv_get_backing_filename(BlockDriverState *bs,
1095                                char *filename, int filename_size)
1096 {
1097     if (!bs->backing_hd) {
1098         pstrcpy(filename, filename_size, "");
1099     } else {
1100         pstrcpy(filename, filename_size, bs->backing_file);
1101     }
1102 }
1103
1104 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1105                           const uint8_t *buf, int nb_sectors)
1106 {
1107     BlockDriver *drv = bs->drv;
1108     if (!drv)
1109         return -ENOMEDIUM;
1110     if (!drv->bdrv_write_compressed)
1111         return -ENOTSUP;
1112     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1113 }
1114
1115 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1116 {
1117     BlockDriver *drv = bs->drv;
1118     if (!drv)
1119         return -ENOMEDIUM;
1120     if (!drv->bdrv_get_info)
1121         return -ENOTSUP;
1122     memset(bdi, 0, sizeof(*bdi));
1123     return drv->bdrv_get_info(bs, bdi);
1124 }
1125
1126 /**************************************************************/
1127 /* handling of snapshots */
1128
1129 int bdrv_snapshot_create(BlockDriverState *bs,
1130                          QEMUSnapshotInfo *sn_info)
1131 {
1132     BlockDriver *drv = bs->drv;
1133     if (!drv)
1134         return -ENOMEDIUM;
1135     if (!drv->bdrv_snapshot_create)
1136         return -ENOTSUP;
1137     return drv->bdrv_snapshot_create(bs, sn_info);
1138 }
1139
1140 int bdrv_snapshot_goto(BlockDriverState *bs,
1141                        const char *snapshot_id)
1142 {
1143     BlockDriver *drv = bs->drv;
1144     if (!drv)
1145         return -ENOMEDIUM;
1146     if (!drv->bdrv_snapshot_goto)
1147         return -ENOTSUP;
1148     return drv->bdrv_snapshot_goto(bs, snapshot_id);
1149 }
1150
1151 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1152 {
1153     BlockDriver *drv = bs->drv;
1154     if (!drv)
1155         return -ENOMEDIUM;
1156     if (!drv->bdrv_snapshot_delete)
1157         return -ENOTSUP;
1158     return drv->bdrv_snapshot_delete(bs, snapshot_id);
1159 }
1160
1161 int bdrv_snapshot_list(BlockDriverState *bs,
1162                        QEMUSnapshotInfo **psn_info)
1163 {
1164     BlockDriver *drv = bs->drv;
1165     if (!drv)
1166         return -ENOMEDIUM;
1167     if (!drv->bdrv_snapshot_list)
1168         return -ENOTSUP;
1169     return drv->bdrv_snapshot_list(bs, psn_info);
1170 }
1171
1172 #define NB_SUFFIXES 4
1173
1174 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1175 {
1176     static const char suffixes[NB_SUFFIXES] = "KMGT";
1177     int64_t base;
1178     int i;
1179
1180     if (size <= 999) {
1181         snprintf(buf, buf_size, "%" PRId64, size);
1182     } else {
1183         base = 1024;
1184         for(i = 0; i < NB_SUFFIXES; i++) {
1185             if (size < (10 * base)) {
1186                 snprintf(buf, buf_size, "%0.1f%c",
1187                          (double)size / base,
1188                          suffixes[i]);
1189                 break;
1190             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1191                 snprintf(buf, buf_size, "%" PRId64 "%c",
1192                          ((size + (base >> 1)) / base),
1193                          suffixes[i]);
1194                 break;
1195             }
1196             base = base * 1024;
1197         }
1198     }
1199     return buf;
1200 }
1201
1202 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1203 {
1204     char buf1[128], date_buf[128], clock_buf[128];
1205 #ifdef _WIN32
1206     struct tm *ptm;
1207 #else
1208     struct tm tm;
1209 #endif
1210     time_t ti;
1211     int64_t secs;
1212
1213     if (!sn) {
1214         snprintf(buf, buf_size,
1215                  "%-10s%-20s%7s%20s%15s",
1216                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1217     } else {
1218         ti = sn->date_sec;
1219 #ifdef _WIN32
1220         ptm = localtime(&ti);
1221         strftime(date_buf, sizeof(date_buf),
1222                  "%Y-%m-%d %H:%M:%S", ptm);
1223 #else
1224         localtime_r(&ti, &tm);
1225         strftime(date_buf, sizeof(date_buf),
1226                  "%Y-%m-%d %H:%M:%S", &tm);
1227 #endif
1228         secs = sn->vm_clock_nsec / 1000000000;
1229         snprintf(clock_buf, sizeof(clock_buf),
1230                  "%02d:%02d:%02d.%03d",
1231                  (int)(secs / 3600),
1232                  (int)((secs / 60) % 60),
1233                  (int)(secs % 60),
1234                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1235         snprintf(buf, buf_size,
1236                  "%-10s%-20s%7s%20s%15s",
1237                  sn->id_str, sn->name,
1238                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1239                  date_buf,
1240                  clock_buf);
1241     }
1242     return buf;
1243 }
1244
1245
1246 /**************************************************************/
1247 /* async I/Os */
1248
1249 typedef struct VectorTranslationState {
1250     QEMUIOVector *iov;
1251     uint8_t *bounce;
1252     int is_write;
1253     BlockDriverAIOCB *aiocb;
1254     BlockDriverAIOCB *this_aiocb;
1255 } VectorTranslationState;
1256
1257 static void bdrv_aio_rw_vector_cb(void *opaque, int ret)
1258 {
1259     VectorTranslationState *s = opaque;
1260
1261     if (!s->is_write) {
1262         qemu_iovec_from_buffer(s->iov, s->bounce);
1263     }
1264     qemu_free(s->bounce);
1265     s->this_aiocb->cb(s->this_aiocb->opaque, ret);
1266     qemu_aio_release(s->this_aiocb);
1267 }
1268
1269 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
1270                                             int64_t sector_num,
1271                                             QEMUIOVector *iov,
1272                                             int nb_sectors,
1273                                             BlockDriverCompletionFunc *cb,
1274                                             void *opaque,
1275                                             int is_write)
1276
1277 {
1278     VectorTranslationState *s = qemu_mallocz(sizeof(*s));
1279     BlockDriverAIOCB *aiocb = qemu_aio_get(bs, cb, opaque);
1280
1281     s->this_aiocb = aiocb;
1282     s->iov = iov;
1283     s->bounce = qemu_memalign(512, nb_sectors * 512);
1284     s->is_write = is_write;
1285     if (is_write) {
1286         qemu_iovec_to_buffer(s->iov, s->bounce);
1287         s->aiocb = bdrv_aio_write(bs, sector_num, s->bounce, nb_sectors,
1288                                   bdrv_aio_rw_vector_cb, s);
1289     } else {
1290         s->aiocb = bdrv_aio_read(bs, sector_num, s->bounce, nb_sectors,
1291                                  bdrv_aio_rw_vector_cb, s);
1292     }
1293     return aiocb;
1294 }
1295
1296 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1297                                  QEMUIOVector *iov, int nb_sectors,
1298                                  BlockDriverCompletionFunc *cb, void *opaque)
1299 {
1300     return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,
1301                               cb, opaque, 0);
1302 }
1303
1304 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1305                                   QEMUIOVector *iov, int nb_sectors,
1306                                   BlockDriverCompletionFunc *cb, void *opaque)
1307 {
1308     return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,
1309                               cb, opaque, 1);
1310 }
1311
1312 BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
1313                                 uint8_t *buf, int nb_sectors,
1314                                 BlockDriverCompletionFunc *cb, void *opaque)
1315 {
1316     BlockDriver *drv = bs->drv;
1317     BlockDriverAIOCB *ret;
1318
1319     if (!drv)
1320         return NULL;
1321
1322     ret = drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
1323
1324     if (ret) {
1325         /* Update stats even though technically transfer has not happened. */
1326         bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1327         bs->rd_ops ++;
1328     }
1329
1330     return ret;
1331 }
1332
1333 BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
1334                                  const uint8_t *buf, int nb_sectors,
1335                                  BlockDriverCompletionFunc *cb, void *opaque)
1336 {
1337     BlockDriver *drv = bs->drv;
1338     BlockDriverAIOCB *ret;
1339
1340     if (!drv)
1341         return NULL;
1342     if (bs->read_only)
1343         return NULL;
1344
1345     ret = drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
1346
1347     if (ret) {
1348         /* Update stats even though technically transfer has not happened. */
1349         bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1350         bs->wr_ops ++;
1351     }
1352
1353     return ret;
1354 }
1355
1356 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1357 {
1358     BlockDriver *drv = acb->bs->drv;
1359
1360     if (acb->cb == bdrv_aio_rw_vector_cb) {
1361         VectorTranslationState *s = acb->opaque;
1362         acb = s->aiocb;
1363     }
1364
1365     drv->bdrv_aio_cancel(acb);
1366 }
1367
1368
1369 /**************************************************************/
1370 /* async block device emulation */
1371
1372 static void bdrv_aio_bh_cb(void *opaque)
1373 {
1374     BlockDriverAIOCBSync *acb = opaque;
1375     acb->common.cb(acb->common.opaque, acb->ret);
1376     qemu_aio_release(acb);
1377 }
1378
1379 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1380         int64_t sector_num, uint8_t *buf, int nb_sectors,
1381         BlockDriverCompletionFunc *cb, void *opaque)
1382 {
1383     BlockDriverAIOCBSync *acb;
1384     int ret;
1385
1386     acb = qemu_aio_get(bs, cb, opaque);
1387     if (!acb->bh)
1388         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1389     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1390     acb->ret = ret;
1391     qemu_bh_schedule(acb->bh);
1392     return &acb->common;
1393 }
1394
1395 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1396         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1397         BlockDriverCompletionFunc *cb, void *opaque)
1398 {
1399     BlockDriverAIOCBSync *acb;
1400     int ret;
1401
1402     acb = qemu_aio_get(bs, cb, opaque);
1403     if (!acb->bh)
1404         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1405     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1406     acb->ret = ret;
1407     qemu_bh_schedule(acb->bh);
1408     return &acb->common;
1409 }
1410
1411 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1412 {
1413     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1414     qemu_bh_cancel(acb->bh);
1415     qemu_aio_release(acb);
1416 }
1417
1418 /**************************************************************/
1419 /* sync block device emulation */
1420
1421 static void bdrv_rw_em_cb(void *opaque, int ret)
1422 {
1423     *(int *)opaque = ret;
1424 }
1425
1426 #define NOT_DONE 0x7fffffff
1427
1428 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1429                         uint8_t *buf, int nb_sectors)
1430 {
1431     int async_ret;
1432     BlockDriverAIOCB *acb;
1433
1434     async_ret = NOT_DONE;
1435     acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
1436                         bdrv_rw_em_cb, &async_ret);
1437     if (acb == NULL)
1438         return -1;
1439
1440     while (async_ret == NOT_DONE) {
1441         qemu_aio_wait();
1442     }
1443
1444     return async_ret;
1445 }
1446
1447 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1448                          const uint8_t *buf, int nb_sectors)
1449 {
1450     int async_ret;
1451     BlockDriverAIOCB *acb;
1452
1453     async_ret = NOT_DONE;
1454     acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
1455                          bdrv_rw_em_cb, &async_ret);
1456     if (acb == NULL)
1457         return -1;
1458     while (async_ret == NOT_DONE) {
1459         qemu_aio_wait();
1460     }
1461     return async_ret;
1462 }
1463
1464 void bdrv_init(void)
1465 {
1466     bdrv_register(&bdrv_raw);
1467     bdrv_register(&bdrv_host_device);
1468 #ifndef _WIN32
1469     bdrv_register(&bdrv_cow);
1470 #endif
1471     bdrv_register(&bdrv_qcow);
1472     bdrv_register(&bdrv_vmdk);
1473     bdrv_register(&bdrv_cloop);
1474     bdrv_register(&bdrv_dmg);
1475     bdrv_register(&bdrv_bochs);
1476     bdrv_register(&bdrv_vpc);
1477     bdrv_register(&bdrv_vvfat);
1478     bdrv_register(&bdrv_qcow2);
1479     bdrv_register(&bdrv_parallels);
1480     bdrv_register(&bdrv_nbd);
1481 }
1482
1483 void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
1484                    void *opaque)
1485 {
1486     BlockDriver *drv;
1487     BlockDriverAIOCB *acb;
1488
1489     drv = bs->drv;
1490     if (drv->free_aiocb) {
1491         acb = drv->free_aiocb;
1492         drv->free_aiocb = acb->next;
1493     } else {
1494         acb = qemu_mallocz(drv->aiocb_size);
1495         if (!acb)
1496             return NULL;
1497     }
1498     acb->bs = bs;
1499     acb->cb = cb;
1500     acb->opaque = opaque;
1501     return acb;
1502 }
1503
1504 void qemu_aio_release(void *p)
1505 {
1506     BlockDriverAIOCB *acb = p;
1507     BlockDriver *drv = acb->bs->drv;
1508     acb->next = drv->free_aiocb;
1509     drv->free_aiocb = acb;
1510 }
1511
1512 /**************************************************************/
1513 /* removable device support */
1514
1515 /**
1516  * Return TRUE if the media is present
1517  */
1518 int bdrv_is_inserted(BlockDriverState *bs)
1519 {
1520     BlockDriver *drv = bs->drv;
1521     int ret;
1522     if (!drv)
1523         return 0;
1524     if (!drv->bdrv_is_inserted)
1525         return 1;
1526     ret = drv->bdrv_is_inserted(bs);
1527     return ret;
1528 }
1529
1530 /**
1531  * Return TRUE if the media changed since the last call to this
1532  * function. It is currently only used for floppy disks
1533  */
1534 int bdrv_media_changed(BlockDriverState *bs)
1535 {
1536     BlockDriver *drv = bs->drv;
1537     int ret;
1538
1539     if (!drv || !drv->bdrv_media_changed)
1540         ret = -ENOTSUP;
1541     else
1542         ret = drv->bdrv_media_changed(bs);
1543     if (ret == -ENOTSUP)
1544         ret = bs->media_changed;
1545     bs->media_changed = 0;
1546     return ret;
1547 }
1548
1549 /**
1550  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1551  */
1552 void bdrv_eject(BlockDriverState *bs, int eject_flag)
1553 {
1554     BlockDriver *drv = bs->drv;
1555     int ret;
1556
1557     if (!drv || !drv->bdrv_eject) {
1558         ret = -ENOTSUP;
1559     } else {
1560         ret = drv->bdrv_eject(bs, eject_flag);
1561     }
1562     if (ret == -ENOTSUP) {
1563         if (eject_flag)
1564             bdrv_close(bs);
1565     }
1566 }
1567
1568 int bdrv_is_locked(BlockDriverState *bs)
1569 {
1570     return bs->locked;
1571 }
1572
1573 /**
1574  * Lock or unlock the media (if it is locked, the user won't be able
1575  * to eject it manually).
1576  */
1577 void bdrv_set_locked(BlockDriverState *bs, int locked)
1578 {
1579     BlockDriver *drv = bs->drv;
1580
1581     bs->locked = locked;
1582     if (drv && drv->bdrv_set_locked) {
1583         drv->bdrv_set_locked(bs, locked);
1584     }
1585 }
1586
1587 /* needed for generic scsi interface */
1588
1589 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1590 {
1591     BlockDriver *drv = bs->drv;
1592
1593     if (drv && drv->bdrv_ioctl)
1594         return drv->bdrv_ioctl(bs, req, buf);
1595     return -ENOTSUP;
1596 }