block/raw-posix.c

   1 /*
   2  * Block driver for RAW files (posix)
   3  *
   4  * Copyright (c) 2006 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 "qemu-timer.h"
  26 #include "qemu-char.h"
  27 #include "qemu-log.h"
  28 #include "block_int.h"
  29 #include "module.h"
  30 #include "block/raw-posix-aio.h"
  31
  32 #ifdef CONFIG_COCOA
  33 #include <paths.h>
  34 #include <sys/param.h>
  35 #include <IOKit/IOKitLib.h>
  36 #include <IOKit/IOBSD.h>
  37 #include <IOKit/storage/IOMediaBSDClient.h>
  38 #include <IOKit/storage/IOMedia.h>
  39 #include <IOKit/storage/IOCDMedia.h>
  40 //#include <IOKit/storage/IOCDTypes.h>
  41 #include <CoreFoundation/CoreFoundation.h>
  42 #endif
  43
  44 #ifdef __sun__
  45 #define _POSIX_PTHREAD_SEMANTICS 1
  46 #include <sys/dkio.h>
  47 #endif
  48 #ifdef __linux__
  49 #include <sys/types.h>
  50 #include <sys/stat.h>
  51 #include <sys/ioctl.h>
  52 #include <sys/param.h>
  53 #include <linux/cdrom.h>
  54 #include <linux/fd.h>
  55 #endif
  56 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
  57 #include <sys/disk.h>
  58 #include <sys/cdio.h>
  59 #endif
  60
  61 #ifdef __OpenBSD__
  62 #include <sys/ioctl.h>
  63 #include <sys/disklabel.h>
  64 #include <sys/dkio.h>
  65 #endif
  66
  67 #ifdef __NetBSD__
  68 #include <sys/ioctl.h>
  69 #include <sys/disklabel.h>
  70 #include <sys/dkio.h>
  71 #include <sys/disk.h>
  72 #endif
  73
  74 #ifdef __DragonFly__
  75 #include <sys/ioctl.h>
  76 #include <sys/diskslice.h>
  77 #endif
  78
  79 #ifdef CONFIG_XFS
  80 #include <xfs/xfs.h>
  81 #endif
  82
  83 //#define DEBUG_FLOPPY
  84
  85 //#define DEBUG_BLOCK
  86 #if defined(DEBUG_BLOCK)
  87 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
  88     { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
  89 #else
  90 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
  91 #endif
  92
  93 /* OS X does not have O_DSYNC */
  94 #ifndef O_DSYNC
  95 #ifdef O_SYNC
  96 #define O_DSYNC O_SYNC
  97 #elif defined(O_FSYNC)
  98 #define O_DSYNC O_FSYNC
  99 #endif
 100 #endif
 101
 102 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
 103 #ifndef O_DIRECT
 104 #define O_DIRECT O_DSYNC
 105 #endif
 106
 107 #define FTYPE_FILE   0
 108 #define FTYPE_CD     1
 109 #define FTYPE_FD     2
 110
 111 /* if the FD is not accessed during that time (in ns), we try to
 112    reopen it to see if the disk has been changed */
 113 #define FD_OPEN_TIMEOUT (1000000000)
 114
 115 #define MAX_BLOCKSIZE   4096
 116
 117 typedef struct BDRVRawState {
 118     int fd;
 119     int type;
 120     int open_flags;
 121 #if defined(__linux__)
 122     /* linux floppy specific */
 123     int64_t fd_open_time;
 124     int64_t fd_error_time;
 125     int fd_got_error;
 126     int fd_media_changed;
 127 #endif
 128 #ifdef CONFIG_LINUX_AIO
 129     int use_aio;
 130     void *aio_ctx;
 131 #endif
 132     uint8_t *aligned_buf;
 133     unsigned aligned_buf_size;
 134 #ifdef CONFIG_XFS
 135     bool is_xfs : 1;
 136 #endif
 137 } BDRVRawState;
 138
 139 static int fd_open(BlockDriverState *bs);
 140 static int64_t raw_getlength(BlockDriverState *bs);
 141
 142 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
 143 static int cdrom_reopen(BlockDriverState *bs);
 144 #endif
 145
 146 #if defined(__NetBSD__)
 147 static int raw_normalize_devicepath(const char **filename)
 148 {
 149     static char namebuf[PATH_MAX];
 150     const char *dp, *fname;
 151     struct stat sb;
 152
 153     fname = *filename;
 154     dp = strrchr(fname, '/');
 155     if (lstat(fname, &sb) < 0) {
 156         fprintf(stderr, "%s: stat failed: %s\n",
 157             fname, strerror(errno));
 158         return -errno;
 159     }
 160
 161     if (!S_ISBLK(sb.st_mode)) {
 162         return 0;
 163     }
 164
 165     if (dp == NULL) {
 166         snprintf(namebuf, PATH_MAX, "r%s", fname);
 167     } else {
 168         snprintf(namebuf, PATH_MAX, "%.*s/r%s",
 169             (int)(dp - fname), fname, dp + 1);
 170     }
 171     fprintf(stderr, "%s is a block device", fname);
 172     *filename = namebuf;
 173     fprintf(stderr, ", using %s\n", *filename);
 174
 175     return 0;
 176 }
 177 #else
 178 static int raw_normalize_devicepath(const char **filename)
 179 {
 180     return 0;
 181 }
 182 #endif
 183
 184 static int raw_open_common(BlockDriverState *bs, const char *filename,
 185                            int bdrv_flags, int open_flags)
 186 {
 187     BDRVRawState *s = bs->opaque;
 188     int fd, ret;
 189
 190     ret = raw_normalize_devicepath(&filename);
 191     if (ret != 0) {
 192         return ret;
 193     }
 194
 195     s->open_flags = open_flags | O_BINARY;
 196     s->open_flags &= ~O_ACCMODE;
 197     if (bdrv_flags & BDRV_O_RDWR) {
 198         s->open_flags |= O_RDWR;
 199     } else {
 200         s->open_flags |= O_RDONLY;
 201     }
 202
 203     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 204      * and O_DIRECT for no caching. */
 205     if ((bdrv_flags & BDRV_O_NOCACHE))
 206         s->open_flags |= O_DIRECT;
 207     if (!(bdrv_flags & BDRV_O_CACHE_WB))
 208         s->open_flags |= O_DSYNC;
 209
 210     s->fd = -1;
 211     fd = qemu_open(filename, s->open_flags, 0644);
 212     if (fd < 0) {
 213         ret = -errno;
 214         if (ret == -EROFS)
 215             ret = -EACCES;
 216         return ret;
 217     }
 218     s->fd = fd;
 219     s->aligned_buf = NULL;
 220
 221     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 222         /*
 223          * Allocate a buffer for read/modify/write cycles.  Chose the size
 224          * pessimistically as we don't know the block size yet.
 225          */
 226         s->aligned_buf_size = 32 * MAX_BLOCKSIZE;
 227         s->aligned_buf = qemu_memalign(MAX_BLOCKSIZE, s->aligned_buf_size);
 228         if (s->aligned_buf == NULL) {
 229             goto out_close;
 230         }
 231     }
 232
 233 #ifdef CONFIG_LINUX_AIO
 234     if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
 235                       (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
 236
 237         /* We're falling back to POSIX AIO in some cases */
 238         paio_init();
 239
 240         s->aio_ctx = laio_init();
 241         if (!s->aio_ctx) {
 242             goto out_free_buf;
 243         }
 244         s->use_aio = 1;
 245     } else
 246 #endif
 247     {
 248         if (paio_init() < 0) {
 249             goto out_free_buf;
 250         }
 251 #ifdef CONFIG_LINUX_AIO
 252         s->use_aio = 0;
 253 #endif
 254     }
 255
 256 #ifdef CONFIG_XFS
 257     if (platform_test_xfs_fd(s->fd)) {
 258         s->is_xfs = 1;
 259     }
 260 #endif
 261
 262     return 0;
 263
 264 out_free_buf:
 265     qemu_vfree(s->aligned_buf);
 266 out_close:
 267     close(fd);
 268     return -errno;
 269 }
 270
 271 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 272 {
 273     BDRVRawState *s = bs->opaque;
 274
 275     s->type = FTYPE_FILE;
 276     return raw_open_common(bs, filename, flags, 0);
 277 }
 278
 279 /* XXX: use host sector size if necessary with:
 280 #ifdef DIOCGSECTORSIZE
 281         {
 282             unsigned int sectorsize = 512;
 283             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 284                 sectorsize > bufsize)
 285                 bufsize = sectorsize;
 286         }
 287 #endif
 288 #ifdef CONFIG_COCOA
 289         uint32_t blockSize = 512;
 290         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 291             bufsize = blockSize;
 292         }
 293 #endif
 294 */
 295
 296 /*
 297  * offset and count are in bytes, but must be multiples of 512 for files
 298  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 299  *
 300  * This function may be called without alignment if the caller ensures
 301  * that O_DIRECT is not in effect.
 302  */
 303 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 304                      uint8_t *buf, int count)
 305 {
 306     BDRVRawState *s = bs->opaque;
 307     int ret;
 308
 309     ret = fd_open(bs);
 310     if (ret < 0)
 311         return ret;
 312
 313     ret = pread(s->fd, buf, count, offset);
 314     if (ret == count)
 315         return ret;
 316
 317     /* Allow reads beyond the end (needed for pwrite) */
 318     if ((ret == 0) && bs->growable) {
 319         int64_t size = raw_getlength(bs);
 320         if (offset >= size) {
 321             memset(buf, 0, count);
 322             return count;
 323         }
 324     }
 325
 326     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 327                       "] read failed %d : %d = %s\n",
 328                       s->fd, bs->filename, offset, buf, count,
 329                       bs->total_sectors, ret, errno, strerror(errno));
 330
 331     /* Try harder for CDrom. */
 332     if (s->type != FTYPE_FILE) {
 333         ret = pread(s->fd, buf, count, offset);
 334         if (ret == count)
 335             return ret;
 336         ret = pread(s->fd, buf, count, offset);
 337         if (ret == count)
 338             return ret;
 339
 340         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 341                           "] retry read failed %d : %d = %s\n",
 342                           s->fd, bs->filename, offset, buf, count,
 343                           bs->total_sectors, ret, errno, strerror(errno));
 344     }
 345
 346     return  (ret < 0) ? -errno : ret;
 347 }
 348
 349 /*
 350  * offset and count are in bytes, but must be multiples of the sector size
 351  * for files opened with O_DIRECT. buf must be aligned to sector size bytes
 352  * then.
 353  *
 354  * This function may be called without alignment if the caller ensures
 355  * that O_DIRECT is not in effect.
 356  */
 357 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 358                       const uint8_t *buf, int count)
 359 {
 360     BDRVRawState *s = bs->opaque;
 361     int ret;
 362
 363     ret = fd_open(bs);
 364     if (ret < 0)
 365         return -errno;
 366
 367     ret = pwrite(s->fd, buf, count, offset);
 368     if (ret == count)
 369         return ret;
 370
 371     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 372                       "] write failed %d : %d = %s\n",
 373                       s->fd, bs->filename, offset, buf, count,
 374                       bs->total_sectors, ret, errno, strerror(errno));
 375
 376     return  (ret < 0) ? -errno : ret;
 377 }
 378
 379
 380 /*
 381  * offset and count are in bytes and possibly not aligned. For files opened
 382  * with O_DIRECT, necessary alignments are ensured before calling
 383  * raw_pread_aligned to do the actual read.
 384  */
 385 static int raw_pread(BlockDriverState *bs, int64_t offset,
 386                      uint8_t *buf, int count)
 387 {
 388     BDRVRawState *s = bs->opaque;
 389     unsigned sector_mask = bs->buffer_alignment - 1;
 390     int size, ret, shift, sum;
 391
 392     sum = 0;
 393
 394     if (s->aligned_buf != NULL)  {
 395
 396         if (offset & sector_mask) {
 397             /* align offset on a sector size bytes boundary */
 398
 399             shift = offset & sector_mask;
 400             size = (shift + count + sector_mask) & ~sector_mask;
 401             if (size > s->aligned_buf_size)
 402                 size = s->aligned_buf_size;
 403             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 404             if (ret < 0)
 405                 return ret;
 406
 407             size = bs->buffer_alignment - shift;
 408             if (size > count)
 409                 size = count;
 410             memcpy(buf, s->aligned_buf + shift, size);
 411
 412             buf += size;
 413             offset += size;
 414             count -= size;
 415             sum += size;
 416
 417             if (count == 0)
 418                 return sum;
 419         }
 420         if (count & sector_mask || (uintptr_t) buf & sector_mask) {
 421
 422             /* read on aligned buffer */
 423
 424             while (count) {
 425
 426                 size = (count + sector_mask) & ~sector_mask;
 427                 if (size > s->aligned_buf_size)
 428                     size = s->aligned_buf_size;
 429
 430                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 431                 if (ret < 0) {
 432                     return ret;
 433                 } else if (ret == 0) {
 434                     fprintf(stderr, "raw_pread: read beyond end of file\n");
 435                     abort();
 436                 }
 437
 438                 size = ret;
 439                 if (size > count)
 440                     size = count;
 441
 442                 memcpy(buf, s->aligned_buf, size);
 443
 444                 buf += size;
 445                 offset += size;
 446                 count -= size;
 447                 sum += size;
 448             }
 449
 450             return sum;
 451         }
 452     }
 453
 454     return raw_pread_aligned(bs, offset, buf, count) + sum;
 455 }
 456
 457 static int raw_read(BlockDriverState *bs, int64_t sector_num,
 458                     uint8_t *buf, int nb_sectors)
 459 {
 460     int ret;
 461
 462     ret = raw_pread(bs, sector_num * BDRV_SECTOR_SIZE, buf,
 463                     nb_sectors * BDRV_SECTOR_SIZE);
 464     if (ret == (nb_sectors * BDRV_SECTOR_SIZE))
 465         ret = 0;
 466     return ret;
 467 }
 468
 469 /*
 470  * offset and count are in bytes and possibly not aligned. For files opened
 471  * with O_DIRECT, necessary alignments are ensured before calling
 472  * raw_pwrite_aligned to do the actual write.
 473  */
 474 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 475                       const uint8_t *buf, int count)
 476 {
 477     BDRVRawState *s = bs->opaque;
 478     unsigned sector_mask = bs->buffer_alignment - 1;
 479     int size, ret, shift, sum;
 480
 481     sum = 0;
 482
 483     if (s->aligned_buf != NULL) {
 484
 485         if (offset & sector_mask) {
 486             /* align offset on a sector size bytes boundary */
 487             shift = offset & sector_mask;
 488             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf,
 489                                     bs->buffer_alignment);
 490             if (ret < 0)
 491                 return ret;
 492
 493             size = bs->buffer_alignment - shift;
 494             if (size > count)
 495                 size = count;
 496             memcpy(s->aligned_buf + shift, buf, size);
 497
 498             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf,
 499                                      bs->buffer_alignment);
 500             if (ret < 0)
 501                 return ret;
 502
 503             buf += size;
 504             offset += size;
 505             count -= size;
 506             sum += size;
 507
 508             if (count == 0)
 509                 return sum;
 510         }
 511         if (count & sector_mask || (uintptr_t) buf & sector_mask) {
 512
 513             while ((size = (count & ~sector_mask)) != 0) {
 514
 515                 if (size > s->aligned_buf_size)
 516                     size = s->aligned_buf_size;
 517
 518                 memcpy(s->aligned_buf, buf, size);
 519
 520                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 521                 if (ret < 0)
 522                     return ret;
 523
 524                 buf += ret;
 525                 offset += ret;
 526                 count -= ret;
 527                 sum += ret;
 528             }
 529             /* here, count < sector_size because (count & ~sector_mask) == 0 */
 530             if (count) {
 531                 ret = raw_pread_aligned(bs, offset, s->aligned_buf,
 532                                      bs->buffer_alignment);
 533                 if (ret < 0)
 534                     return ret;
 535                  memcpy(s->aligned_buf, buf, count);
 536
 537                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf,
 538                                      bs->buffer_alignment);
 539                  if (ret < 0)
 540                      return ret;
 541                  if (count < ret)
 542                      ret = count;
 543
 544                  sum += ret;
 545             }
 546             return sum;
 547         }
 548     }
 549     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 550 }
 551
 552 static int raw_write(BlockDriverState *bs, int64_t sector_num,
 553                      const uint8_t *buf, int nb_sectors)
 554 {
 555     int ret;
 556     ret = raw_pwrite(bs, sector_num * BDRV_SECTOR_SIZE, buf,
 557                      nb_sectors * BDRV_SECTOR_SIZE);
 558     if (ret == (nb_sectors * BDRV_SECTOR_SIZE))
 559         ret = 0;
 560     return ret;
 561 }
 562
 563 /*
 564  * Check if all memory in this vector is sector aligned.
 565  */
 566 static int qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
 567 {
 568     int i;
 569
 570     for (i = 0; i < qiov->niov; i++) {
 571         if ((uintptr_t) qiov->iov[i].iov_base % bs->buffer_alignment) {
 572             return 0;
 573         }
 574     }
 575
 576     return 1;
 577 }
 578
 579 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
 580         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 581         BlockDriverCompletionFunc *cb, void *opaque, int type)
 582 {
 583     BDRVRawState *s = bs->opaque;
 584
 585     if (fd_open(bs) < 0)
 586         return NULL;
 587
 588     /*
 589      * If O_DIRECT is used the buffer needs to be aligned on a sector
 590      * boundary.  Check if this is the case or telll the low-level
 591      * driver that it needs to copy the buffer.
 592      */
 593     if (s->aligned_buf) {
 594         if (!qiov_is_aligned(bs, qiov)) {
 595             type |= QEMU_AIO_MISALIGNED;
 596 #ifdef CONFIG_LINUX_AIO
 597         } else if (s->use_aio) {
 598             return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
 599                                nb_sectors, cb, opaque, type);
 600 #endif
 601         }
 602     }
 603
 604     return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
 605                        cb, opaque, type);
 606 }
 607
 608 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 609         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 610         BlockDriverCompletionFunc *cb, void *opaque)
 611 {
 612     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 613                           cb, opaque, QEMU_AIO_READ);
 614 }
 615
 616 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 617         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 618         BlockDriverCompletionFunc *cb, void *opaque)
 619 {
 620     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 621                           cb, opaque, QEMU_AIO_WRITE);
 622 }
 623
 624 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
 625         BlockDriverCompletionFunc *cb, void *opaque)
 626 {
 627     BDRVRawState *s = bs->opaque;
 628
 629     if (fd_open(bs) < 0)
 630         return NULL;
 631
 632     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
 633 }
 634
 635 static void raw_close(BlockDriverState *bs)
 636 {
 637     BDRVRawState *s = bs->opaque;
 638     if (s->fd >= 0) {
 639         close(s->fd);
 640         s->fd = -1;
 641         if (s->aligned_buf != NULL)
 642             qemu_vfree(s->aligned_buf);
 643     }
 644 }
 645
 646 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 647 {
 648     BDRVRawState *s = bs->opaque;
 649     if (s->type != FTYPE_FILE)
 650         return -ENOTSUP;
 651     if (ftruncate(s->fd, offset) < 0)
 652         return -errno;
 653     return 0;
 654 }
 655
 656 #ifdef __OpenBSD__
 657 static int64_t raw_getlength(BlockDriverState *bs)
 658 {
 659     BDRVRawState *s = bs->opaque;
 660     int fd = s->fd;
 661     struct stat st;
 662
 663     if (fstat(fd, &st))
 664         return -1;
 665     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 666         struct disklabel dl;
 667
 668         if (ioctl(fd, DIOCGDINFO, &dl))
 669             return -1;
 670         return (uint64_t)dl.d_secsize *
 671             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 672     } else
 673         return st.st_size;
 674 }
 675 #elif defined(__NetBSD__)
 676 static int64_t raw_getlength(BlockDriverState *bs)
 677 {
 678     BDRVRawState *s = bs->opaque;
 679     int fd = s->fd;
 680     struct stat st;
 681
 682     if (fstat(fd, &st))
 683         return -1;
 684     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 685         struct dkwedge_info dkw;
 686
 687         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
 688             return dkw.dkw_size * 512;
 689         } else {
 690             struct disklabel dl;
 691
 692             if (ioctl(fd, DIOCGDINFO, &dl))
 693                 return -1;
 694             return (uint64_t)dl.d_secsize *
 695                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 696         }
 697     } else
 698         return st.st_size;
 699 }
 700 #elif defined(__sun__)
 701 static int64_t raw_getlength(BlockDriverState *bs)
 702 {
 703     BDRVRawState *s = bs->opaque;
 704     struct dk_minfo minfo;
 705     int ret;
 706
 707     ret = fd_open(bs);
 708     if (ret < 0) {
 709         return ret;
 710     }
 711
 712     /*
 713      * Use the DKIOCGMEDIAINFO ioctl to read the size.
 714      */
 715     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
 716     if (ret != -1) {
 717         return minfo.dki_lbsize * minfo.dki_capacity;
 718     }
 719
 720     /*
 721      * There are reports that lseek on some devices fails, but
 722      * irc discussion said that contingency on contingency was overkill.
 723      */
 724     return lseek(s->fd, 0, SEEK_END);
 725 }
 726 #elif defined(CONFIG_BSD)
 727 static int64_t raw_getlength(BlockDriverState *bs)
 728 {
 729     BDRVRawState *s = bs->opaque;
 730     int fd = s->fd;
 731     int64_t size;
 732     struct stat sb;
 733 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 734     int reopened = 0;
 735 #endif
 736     int ret;
 737
 738     ret = fd_open(bs);
 739     if (ret < 0)
 740         return ret;
 741
 742 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 743 again:
 744 #endif
 745     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 746 #ifdef DIOCGMEDIASIZE
 747         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 748 #elif defined(DIOCGPART)
 749         {
 750                 struct partinfo pi;
 751                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 752                         size = pi.media_size;
 753                 else
 754                         size = 0;
 755         }
 756         if (size == 0)
 757 #endif
 758 #ifdef CONFIG_COCOA
 759         size = LONG_LONG_MAX;
 760 #else
 761         size = lseek(fd, 0LL, SEEK_END);
 762 #endif
 763 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
 764         switch(s->type) {
 765         case FTYPE_CD:
 766             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
 767             if (size == 2048LL * (unsigned)-1)
 768                 size = 0;
 769             /* XXX no disc?  maybe we need to reopen... */
 770             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
 771                 reopened = 1;
 772                 goto again;
 773             }
 774         }
 775 #endif
 776     } else {
 777         size = lseek(fd, 0, SEEK_END);
 778     }
 779     return size;
 780 }
 781 #else
 782 static int64_t raw_getlength(BlockDriverState *bs)
 783 {
 784     BDRVRawState *s = bs->opaque;
 785     int ret;
 786
 787     ret = fd_open(bs);
 788     if (ret < 0) {
 789         return ret;
 790     }
 791
 792     return lseek(s->fd, 0, SEEK_END);
 793 }
 794 #endif
 795
 796 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
 797 {
 798     struct stat st;
 799     BDRVRawState *s = bs->opaque;
 800
 801     if (fstat(s->fd, &st) < 0) {
 802         return -errno;
 803     }
 804     return (int64_t)st.st_blocks * 512;
 805 }
 806
 807 static int raw_create(const char *filename, QEMUOptionParameter *options)
 808 {
 809     int fd;
 810     int result = 0;
 811     int64_t total_size = 0;
 812
 813     /* Read out options */
 814     while (options && options->name) {
 815         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 816             total_size = options->value.n / BDRV_SECTOR_SIZE;
 817         }
 818         options++;
 819     }
 820
 821     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 822               0644);
 823     if (fd < 0) {
 824         result = -errno;
 825     } else {
 826         if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
 827             result = -errno;
 828         }
 829         if (close(fd) != 0) {
 830             result = -errno;
 831         }
 832     }
 833     return result;
 834 }
 835
 836 static int raw_flush(BlockDriverState *bs)
 837 {
 838     BDRVRawState *s = bs->opaque;
 839     return qemu_fdatasync(s->fd);
 840 }
 841
 842 #ifdef CONFIG_XFS
 843 static int xfs_discard(BDRVRawState *s, int64_t sector_num, int nb_sectors)
 844 {
 845     struct xfs_flock64 fl;
 846
 847     memset(&fl, 0, sizeof(fl));
 848     fl.l_whence = SEEK_SET;
 849     fl.l_start = sector_num << 9;
 850     fl.l_len = (int64_t)nb_sectors << 9;
 851
 852     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
 853         DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
 854         return -errno;
 855     }
 856
 857     return 0;
 858 }
 859 #endif
 860
 861 static int raw_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
 862 {
 863 #ifdef CONFIG_XFS
 864     BDRVRawState *s = bs->opaque;
 865
 866     if (s->is_xfs) {
 867         return xfs_discard(s, sector_num, nb_sectors);
 868     }
 869 #endif
 870
 871     return 0;
 872 }
 873
 874 static QEMUOptionParameter raw_create_options[] = {
 875     {
 876         .name = BLOCK_OPT_SIZE,
 877         .type = OPT_SIZE,
 878         .help = "Virtual disk size"
 879     },
 880     { NULL }
 881 };
 882
 883 static BlockDriver bdrv_file = {
 884     .format_name = "file",
 885     .protocol_name = "file",
 886     .instance_size = sizeof(BDRVRawState),
 887     .bdrv_probe = NULL, /* no probe for protocols */
 888     .bdrv_file_open = raw_open,
 889     .bdrv_read = raw_read,
 890     .bdrv_write = raw_write,
 891     .bdrv_close = raw_close,
 892     .bdrv_create = raw_create,
 893     .bdrv_flush = raw_flush,
 894     .bdrv_discard = raw_discard,
 895
 896     .bdrv_aio_readv = raw_aio_readv,
 897     .bdrv_aio_writev = raw_aio_writev,
 898     .bdrv_aio_flush = raw_aio_flush,
 899
 900     .bdrv_truncate = raw_truncate,
 901     .bdrv_getlength = raw_getlength,
 902     .bdrv_get_allocated_file_size
 903                         = raw_get_allocated_file_size,
 904
 905     .create_options = raw_create_options,
 906 };
 907
 908 /***********************************************/
 909 /* host device */
 910
 911 #ifdef CONFIG_COCOA
 912 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 913 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 914
 915 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 916 {
 917     kern_return_t       kernResult;
 918     mach_port_t     masterPort;
 919     CFMutableDictionaryRef  classesToMatch;
 920
 921     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 922     if ( KERN_SUCCESS != kernResult ) {
 923         printf( "IOMasterPort returned %d\n", kernResult );
 924     }
 925
 926     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 927     if ( classesToMatch == NULL ) {
 928         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 929     } else {
 930     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 931     }
 932     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 933     if ( KERN_SUCCESS != kernResult )
 934     {
 935         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 936     }
 937
 938     return kernResult;
 939 }
 940
 941 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 942 {
 943     io_object_t     nextMedia;
 944     kern_return_t   kernResult = KERN_FAILURE;
 945     *bsdPath = '\0';
 946     nextMedia = IOIteratorNext( mediaIterator );
 947     if ( nextMedia )
 948     {
 949         CFTypeRef   bsdPathAsCFString;
 950     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 951         if ( bsdPathAsCFString ) {
 952             size_t devPathLength;
 953             strcpy( bsdPath, _PATH_DEV );
 954             strcat( bsdPath, "r" );
 955             devPathLength = strlen( bsdPath );
 956             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 957                 kernResult = KERN_SUCCESS;
 958             }
 959             CFRelease( bsdPathAsCFString );
 960         }
 961         IOObjectRelease( nextMedia );
 962     }
 963
 964     return kernResult;
 965 }
 966
 967 #endif
 968
 969 static int hdev_probe_device(const char *filename)
 970 {
 971     struct stat st;
 972
 973     /* allow a dedicated CD-ROM driver to match with a higher priority */
 974     if (strstart(filename, "/dev/cdrom", NULL))
 975         return 50;
 976
 977     if (stat(filename, &st) >= 0 &&
 978             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 979         return 100;
 980     }
 981
 982     return 0;
 983 }
 984
 985 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 986 {
 987     BDRVRawState *s = bs->opaque;
 988
 989 #ifdef CONFIG_COCOA
 990     if (strstart(filename, "/dev/cdrom", NULL)) {
 991         kern_return_t kernResult;
 992         io_iterator_t mediaIterator;
 993         char bsdPath[ MAXPATHLEN ];
 994         int fd;
 995
 996         kernResult = FindEjectableCDMedia( &mediaIterator );
 997         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 998
 999         if ( bsdPath[ 0 ] != '\0' ) {
1000             strcat(bsdPath,"s0");
1001             /* some CDs don't have a partition 0 */
1002             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1003             if (fd < 0) {
1004                 bsdPath[strlen(bsdPath)-1] = '1';
1005             } else {
1006                 close(fd);
1007             }
1008             filename = bsdPath;
1009         }
1010
1011         if ( mediaIterator )
1012             IOObjectRelease( mediaIterator );
1013     }
1014 #endif
1015
1016     s->type = FTYPE_FILE;
1017 #if defined(__linux__)
1018     {
1019         char resolved_path[ MAXPATHLEN ], *temp;
1020
1021         temp = realpath(filename, resolved_path);
1022         if (temp && strstart(temp, "/dev/sg", NULL)) {
1023             bs->sg = 1;
1024         }
1025     }
1026 #endif
1027
1028     return raw_open_common(bs, filename, flags, 0);
1029 }
1030
1031 #if defined(__linux__)
1032 /* Note: we do not have a reliable method to detect if the floppy is
1033    present. The current method is to try to open the floppy at every
1034    I/O and to keep it opened during a few hundreds of ms. */
1035 static int fd_open(BlockDriverState *bs)
1036 {
1037     BDRVRawState *s = bs->opaque;
1038     int last_media_present;
1039
1040     if (s->type != FTYPE_FD)
1041         return 0;
1042     last_media_present = (s->fd >= 0);
1043     if (s->fd >= 0 &&
1044         (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1045         close(s->fd);
1046         s->fd = -1;
1047 #ifdef DEBUG_FLOPPY
1048         printf("Floppy closed\n");
1049 #endif
1050     }
1051     if (s->fd < 0) {
1052         if (s->fd_got_error &&
1053             (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1054 #ifdef DEBUG_FLOPPY
1055             printf("No floppy (open delayed)\n");
1056 #endif
1057             return -EIO;
1058         }
1059         s->fd = open(bs->filename, s->open_flags & ~O_NONBLOCK);
1060         if (s->fd < 0) {
1061             s->fd_error_time = get_clock();
1062             s->fd_got_error = 1;
1063             if (last_media_present)
1064                 s->fd_media_changed = 1;
1065 #ifdef DEBUG_FLOPPY
1066             printf("No floppy\n");
1067 #endif
1068             return -EIO;
1069         }
1070 #ifdef DEBUG_FLOPPY
1071         printf("Floppy opened\n");
1072 #endif
1073     }
1074     if (!last_media_present)
1075         s->fd_media_changed = 1;
1076     s->fd_open_time = get_clock();
1077     s->fd_got_error = 0;
1078     return 0;
1079 }
1080
1081 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1082 {
1083     BDRVRawState *s = bs->opaque;
1084
1085     return ioctl(s->fd, req, buf);
1086 }
1087
1088 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1089         unsigned long int req, void *buf,
1090         BlockDriverCompletionFunc *cb, void *opaque)
1091 {
1092     BDRVRawState *s = bs->opaque;
1093
1094     if (fd_open(bs) < 0)
1095         return NULL;
1096     return paio_ioctl(bs, s->fd, req, buf, cb, opaque);
1097 }
1098
1099 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1100 static int fd_open(BlockDriverState *bs)
1101 {
1102     BDRVRawState *s = bs->opaque;
1103
1104     /* this is just to ensure s->fd is sane (its called by io ops) */
1105     if (s->fd >= 0)
1106         return 0;
1107     return -EIO;
1108 }
1109 #else /* !linux && !FreeBSD */
1110
1111 static int fd_open(BlockDriverState *bs)
1112 {
1113     return 0;
1114 }
1115
1116 #endif /* !linux && !FreeBSD */
1117
1118 static int hdev_create(const char *filename, QEMUOptionParameter *options)
1119 {
1120     int fd;
1121     int ret = 0;
1122     struct stat stat_buf;
1123     int64_t total_size = 0;
1124
1125     /* Read out options */
1126     while (options && options->name) {
1127         if (!strcmp(options->name, "size")) {
1128             total_size = options->value.n / BDRV_SECTOR_SIZE;
1129         }
1130         options++;
1131     }
1132
1133     fd = open(filename, O_WRONLY | O_BINARY);
1134     if (fd < 0)
1135         return -errno;
1136
1137     if (fstat(fd, &stat_buf) < 0)
1138         ret = -errno;
1139     else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
1140         ret = -ENODEV;
1141     else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE)
1142         ret = -ENOSPC;
1143
1144     close(fd);
1145     return ret;
1146 }
1147
1148 static int hdev_has_zero_init(BlockDriverState *bs)
1149 {
1150     return 0;
1151 }
1152
1153 static BlockDriver bdrv_host_device = {
1154     .format_name        = "host_device",
1155     .protocol_name        = "host_device",
1156     .instance_size      = sizeof(BDRVRawState),
1157     .bdrv_probe_device  = hdev_probe_device,
1158     .bdrv_file_open     = hdev_open,
1159     .bdrv_close         = raw_close,
1160     .bdrv_create        = hdev_create,
1161     .create_options     = raw_create_options,
1162     .bdrv_has_zero_init = hdev_has_zero_init,
1163     .bdrv_flush         = raw_flush,
1164
1165     .bdrv_aio_readv     = raw_aio_readv,
1166     .bdrv_aio_writev    = raw_aio_writev,
1167     .bdrv_aio_flush     = raw_aio_flush,
1168
1169     .bdrv_read          = raw_read,
1170     .bdrv_write         = raw_write,
1171     .bdrv_getlength     = raw_getlength,
1172     .bdrv_get_allocated_file_size
1173                         = raw_get_allocated_file_size,
1174
1175     /* generic scsi device */
1176 #ifdef __linux__
1177     .bdrv_ioctl         = hdev_ioctl,
1178     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1179 #endif
1180 };
1181
1182 #ifdef __linux__
1183 static int floppy_open(BlockDriverState *bs, const char *filename, int flags)
1184 {
1185     BDRVRawState *s = bs->opaque;
1186     int ret;
1187
1188     s->type = FTYPE_FD;
1189
1190     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1191     ret = raw_open_common(bs, filename, flags, O_NONBLOCK);
1192     if (ret)
1193         return ret;
1194
1195     /* close fd so that we can reopen it as needed */
1196     close(s->fd);
1197     s->fd = -1;
1198     s->fd_media_changed = 1;
1199
1200     return 0;
1201 }
1202
1203 static int floppy_probe_device(const char *filename)
1204 {
1205     int fd, ret;
1206     int prio = 0;
1207     struct floppy_struct fdparam;
1208     struct stat st;
1209
1210     if (strstart(filename, "/dev/fd", NULL))
1211         prio = 50;
1212
1213     fd = open(filename, O_RDONLY | O_NONBLOCK);
1214     if (fd < 0) {
1215         goto out;
1216     }
1217     ret = fstat(fd, &st);
1218     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1219         goto outc;
1220     }
1221
1222     /* Attempt to detect via a floppy specific ioctl */
1223     ret = ioctl(fd, FDGETPRM, &fdparam);
1224     if (ret >= 0)
1225         prio = 100;
1226
1227 outc:
1228     close(fd);
1229 out:
1230     return prio;
1231 }
1232
1233
1234 static int floppy_is_inserted(BlockDriverState *bs)
1235 {
1236     return fd_open(bs) >= 0;
1237 }
1238
1239 static int floppy_media_changed(BlockDriverState *bs)
1240 {
1241     BDRVRawState *s = bs->opaque;
1242     int ret;
1243
1244     /*
1245      * XXX: we do not have a true media changed indication.
1246      * It does not work if the floppy is changed without trying to read it.
1247      */
1248     fd_open(bs);
1249     ret = s->fd_media_changed;
1250     s->fd_media_changed = 0;
1251 #ifdef DEBUG_FLOPPY
1252     printf("Floppy changed=%d\n", ret);
1253 #endif
1254     return ret;
1255 }
1256
1257 static int floppy_eject(BlockDriverState *bs, int eject_flag)
1258 {
1259     BDRVRawState *s = bs->opaque;
1260     int fd;
1261
1262     if (s->fd >= 0) {
1263         close(s->fd);
1264         s->fd = -1;
1265     }
1266     fd = open(bs->filename, s->open_flags | O_NONBLOCK);
1267     if (fd >= 0) {
1268         if (ioctl(fd, FDEJECT, 0) < 0)
1269             perror("FDEJECT");
1270         close(fd);
1271     }
1272
1273     return 0;
1274 }
1275
1276 static BlockDriver bdrv_host_floppy = {
1277     .format_name        = "host_floppy",
1278     .protocol_name      = "host_floppy",
1279     .instance_size      = sizeof(BDRVRawState),
1280     .bdrv_probe_device  = floppy_probe_device,
1281     .bdrv_file_open     = floppy_open,
1282     .bdrv_close         = raw_close,
1283     .bdrv_create        = hdev_create,
1284     .create_options     = raw_create_options,
1285     .bdrv_has_zero_init = hdev_has_zero_init,
1286     .bdrv_flush         = raw_flush,
1287
1288     .bdrv_aio_readv     = raw_aio_readv,
1289     .bdrv_aio_writev    = raw_aio_writev,
1290     .bdrv_aio_flush     = raw_aio_flush,
1291
1292     .bdrv_read          = raw_read,
1293     .bdrv_write         = raw_write,
1294     .bdrv_getlength     = raw_getlength,
1295     .bdrv_get_allocated_file_size
1296                         = raw_get_allocated_file_size,
1297
1298     /* removable device support */
1299     .bdrv_is_inserted   = floppy_is_inserted,
1300     .bdrv_media_changed = floppy_media_changed,
1301     .bdrv_eject         = floppy_eject,
1302 };
1303
1304 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1305 {
1306     BDRVRawState *s = bs->opaque;
1307
1308     s->type = FTYPE_CD;
1309
1310     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1311     return raw_open_common(bs, filename, flags, O_NONBLOCK);
1312 }
1313
1314 static int cdrom_probe_device(const char *filename)
1315 {
1316     int fd, ret;
1317     int prio = 0;
1318     struct stat st;
1319
1320     fd = open(filename, O_RDONLY | O_NONBLOCK);
1321     if (fd < 0) {
1322         goto out;
1323     }
1324     ret = fstat(fd, &st);
1325     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1326         goto outc;
1327     }
1328
1329     /* Attempt to detect via a CDROM specific ioctl */
1330     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1331     if (ret >= 0)
1332         prio = 100;
1333
1334 outc:
1335     close(fd);
1336 out:
1337     return prio;
1338 }
1339
1340 static int cdrom_is_inserted(BlockDriverState *bs)
1341 {
1342     BDRVRawState *s = bs->opaque;
1343     int ret;
1344
1345     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1346     if (ret == CDS_DISC_OK)
1347         return 1;
1348     return 0;
1349 }
1350
1351 static int cdrom_eject(BlockDriverState *bs, int eject_flag)
1352 {
1353     BDRVRawState *s = bs->opaque;
1354
1355     if (eject_flag) {
1356         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1357             perror("CDROMEJECT");
1358     } else {
1359         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1360             perror("CDROMEJECT");
1361     }
1362
1363     return 0;
1364 }
1365
1366 static int cdrom_set_locked(BlockDriverState *bs, int locked)
1367 {
1368     BDRVRawState *s = bs->opaque;
1369
1370     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1371         /*
1372          * Note: an error can happen if the distribution automatically
1373          * mounts the CD-ROM
1374          */
1375         /* perror("CDROM_LOCKDOOR"); */
1376     }
1377
1378     return 0;
1379 }
1380
1381 static BlockDriver bdrv_host_cdrom = {
1382     .format_name        = "host_cdrom",
1383     .protocol_name      = "host_cdrom",
1384     .instance_size      = sizeof(BDRVRawState),
1385     .bdrv_probe_device  = cdrom_probe_device,
1386     .bdrv_file_open     = cdrom_open,
1387     .bdrv_close         = raw_close,
1388     .bdrv_create        = hdev_create,
1389     .create_options     = raw_create_options,
1390     .bdrv_has_zero_init = hdev_has_zero_init,
1391     .bdrv_flush         = raw_flush,
1392
1393     .bdrv_aio_readv     = raw_aio_readv,
1394     .bdrv_aio_writev    = raw_aio_writev,
1395     .bdrv_aio_flush     = raw_aio_flush,
1396
1397     .bdrv_read          = raw_read,
1398     .bdrv_write         = raw_write,
1399     .bdrv_getlength     = raw_getlength,
1400     .bdrv_get_allocated_file_size
1401                         = raw_get_allocated_file_size,
1402
1403     /* removable device support */
1404     .bdrv_is_inserted   = cdrom_is_inserted,
1405     .bdrv_eject         = cdrom_eject,
1406     .bdrv_set_locked    = cdrom_set_locked,
1407
1408     /* generic scsi device */
1409     .bdrv_ioctl         = hdev_ioctl,
1410     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1411 };
1412 #endif /* __linux__ */
1413
1414 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1415 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1416 {
1417     BDRVRawState *s = bs->opaque;
1418     int ret;
1419
1420     s->type = FTYPE_CD;
1421
1422     ret = raw_open_common(bs, filename, flags, 0);
1423     if (ret)
1424         return ret;
1425
1426     /* make sure the door isnt locked at this time */
1427     ioctl(s->fd, CDIOCALLOW);
1428     return 0;
1429 }
1430
1431 static int cdrom_probe_device(const char *filename)
1432 {
1433     if (strstart(filename, "/dev/cd", NULL) ||
1434             strstart(filename, "/dev/acd", NULL))
1435         return 100;
1436     return 0;
1437 }
1438
1439 static int cdrom_reopen(BlockDriverState *bs)
1440 {
1441     BDRVRawState *s = bs->opaque;
1442     int fd;
1443
1444     /*
1445      * Force reread of possibly changed/newly loaded disc,
1446      * FreeBSD seems to not notice sometimes...
1447      */
1448     if (s->fd >= 0)
1449         close(s->fd);
1450     fd = open(bs->filename, s->open_flags, 0644);
1451     if (fd < 0) {
1452         s->fd = -1;
1453         return -EIO;
1454     }
1455     s->fd = fd;
1456
1457     /* make sure the door isnt locked at this time */
1458     ioctl(s->fd, CDIOCALLOW);
1459     return 0;
1460 }
1461
1462 static int cdrom_is_inserted(BlockDriverState *bs)
1463 {
1464     return raw_getlength(bs) > 0;
1465 }
1466
1467 static int cdrom_eject(BlockDriverState *bs, int eject_flag)
1468 {
1469     BDRVRawState *s = bs->opaque;
1470
1471     if (s->fd < 0)
1472         return -ENOTSUP;
1473
1474     (void) ioctl(s->fd, CDIOCALLOW);
1475
1476     if (eject_flag) {
1477         if (ioctl(s->fd, CDIOCEJECT) < 0)
1478             perror("CDIOCEJECT");
1479     } else {
1480         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1481             perror("CDIOCCLOSE");
1482     }
1483
1484     if (cdrom_reopen(bs) < 0)
1485         return -ENOTSUP;
1486     return 0;
1487 }
1488
1489 static int cdrom_set_locked(BlockDriverState *bs, int locked)
1490 {
1491     BDRVRawState *s = bs->opaque;
1492
1493     if (s->fd < 0)
1494         return -ENOTSUP;
1495     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1496         /*
1497          * Note: an error can happen if the distribution automatically
1498          * mounts the CD-ROM
1499          */
1500         /* perror("CDROM_LOCKDOOR"); */
1501     }
1502
1503     return 0;
1504 }
1505
1506 static BlockDriver bdrv_host_cdrom = {
1507     .format_name        = "host_cdrom",
1508     .protocol_name      = "host_cdrom",
1509     .instance_size      = sizeof(BDRVRawState),
1510     .bdrv_probe_device  = cdrom_probe_device,
1511     .bdrv_file_open     = cdrom_open,
1512     .bdrv_close         = raw_close,
1513     .bdrv_create        = hdev_create,
1514     .create_options     = raw_create_options,
1515     .bdrv_has_zero_init = hdev_has_zero_init,
1516     .bdrv_flush         = raw_flush,
1517
1518     .bdrv_aio_readv     = raw_aio_readv,
1519     .bdrv_aio_writev    = raw_aio_writev,
1520     .bdrv_aio_flush     = raw_aio_flush,
1521
1522     .bdrv_read          = raw_read,
1523     .bdrv_write         = raw_write,
1524     .bdrv_getlength     = raw_getlength,
1525     .bdrv_get_allocated_file_size
1526                         = raw_get_allocated_file_size,
1527
1528     /* removable device support */
1529     .bdrv_is_inserted   = cdrom_is_inserted,
1530     .bdrv_eject         = cdrom_eject,
1531     .bdrv_set_locked    = cdrom_set_locked,
1532 };
1533 #endif /* __FreeBSD__ */
1534
1535 static void bdrv_file_init(void)
1536 {
1537     /*
1538      * Register all the drivers.  Note that order is important, the driver
1539      * registered last will get probed first.
1540      */
1541     bdrv_register(&bdrv_file);
1542     bdrv_register(&bdrv_host_device);
1543 #ifdef __linux__
1544     bdrv_register(&bdrv_host_floppy);
1545     bdrv_register(&bdrv_host_cdrom);
1546 #endif
1547 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1548     bdrv_register(&bdrv_host_cdrom);
1549 #endif
1550 }
1551
1552 block_init(bdrv_file_init);