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