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 "block_int.h"
  28 #include "module.h"
  29 #ifdef CONFIG_AIO
  30 #include "posix-aio-compat.h"
  31 #endif
  32
  33 #ifdef CONFIG_COCOA
  34 #include <paths.h>
  35 #include <sys/param.h>
  36 #include <IOKit/IOKitLib.h>
  37 #include <IOKit/IOBSD.h>
  38 #include <IOKit/storage/IOMediaBSDClient.h>
  39 #include <IOKit/storage/IOMedia.h>
  40 #include <IOKit/storage/IOCDMedia.h>
  41 //#include <IOKit/storage/IOCDTypes.h>
  42 #include <CoreFoundation/CoreFoundation.h>
  43 #endif
  44
  45 #ifdef __sun__
  46 #define _POSIX_PTHREAD_SEMANTICS 1
  47 #include <signal.h>
  48 #include <sys/dkio.h>
  49 #endif
  50 #ifdef __linux__
  51 #include <sys/ioctl.h>
  52 #include <linux/cdrom.h>
  53 #include <linux/fd.h>
  54 #endif
  55 #ifdef __FreeBSD__
  56 #include <signal.h>
  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 __DragonFly__
  68 #include <sys/ioctl.h>
  69 #include <sys/diskslice.h>
  70 #endif
  71
  72 //#define DEBUG_FLOPPY
  73
  74 //#define DEBUG_BLOCK
  75 #if defined(DEBUG_BLOCK)
  76 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
  77     { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
  78 #else
  79 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
  80 #endif
  81
  82 /* OS X does not have O_DSYNC */
  83 #ifndef O_DSYNC
  84 #define O_DSYNC O_SYNC
  85 #endif
  86
  87 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
  88 #ifndef O_DIRECT
  89 #define O_DIRECT O_DSYNC
  90 #endif
  91
  92 #define FTYPE_FILE   0
  93 #define FTYPE_CD     1
  94 #define FTYPE_FD     2
  95
  96 #define ALIGNED_BUFFER_SIZE (32 * 512)
  97
  98 /* if the FD is not accessed during that time (in ms), we try to
  99    reopen it to see if the disk has been changed */
 100 #define FD_OPEN_TIMEOUT 1000
 101
 102 typedef struct BDRVRawState {
 103     int fd;
 104     int type;
 105     unsigned int lseek_err_cnt;
 106     int open_flags;
 107 #if defined(__linux__)
 108     /* linux floppy specific */
 109     int64_t fd_open_time;
 110     int64_t fd_error_time;
 111     int fd_got_error;
 112     int fd_media_changed;
 113 #endif
 114     uint8_t* aligned_buf;
 115 } BDRVRawState;
 116
 117 static int posix_aio_init(void);
 118
 119 static int fd_open(BlockDriverState *bs);
 120
 121 #if defined(__FreeBSD__)
 122 static int cdrom_reopen(BlockDriverState *bs);
 123 #endif
 124
 125 static int raw_open_common(BlockDriverState *bs, const char *filename,
 126         int flags)
 127 {
 128     BDRVRawState *s = bs->opaque;
 129     int fd, ret;
 130
 131     posix_aio_init();
 132
 133     s->lseek_err_cnt = 0;
 134
 135     s->open_flags |= O_BINARY;
 136     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 137         s->open_flags |= O_RDWR;
 138     } else {
 139         s->open_flags |= O_RDONLY;
 140         bs->read_only = 1;
 141     }
 142
 143     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 144      * and O_DIRECT for no caching. */
 145     if ((flags & BDRV_O_NOCACHE))
 146         s->open_flags |= O_DIRECT;
 147     else if (!(flags & BDRV_O_CACHE_WB))
 148         s->open_flags |= O_DSYNC;
 149
 150     s->fd = -1;
 151     fd = open(filename, s->open_flags, 0644);
 152     if (fd < 0) {
 153         ret = -errno;
 154         if (ret == -EROFS)
 155             ret = -EACCES;
 156         return ret;
 157     }
 158     s->fd = fd;
 159     s->aligned_buf = NULL;
 160     if ((flags & BDRV_O_NOCACHE)) {
 161         s->aligned_buf = qemu_blockalign(bs, ALIGNED_BUFFER_SIZE);
 162         if (s->aligned_buf == NULL) {
 163             ret = -errno;
 164             close(fd);
 165             return ret;
 166         }
 167     }
 168     return 0;
 169 }
 170
 171 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 172 {
 173     BDRVRawState *s = bs->opaque;
 174
 175     s->type = FTYPE_FILE;
 176     if (flags & BDRV_O_CREAT)
 177         s->open_flags |= O_CREAT | O_TRUNC;
 178
 179     return raw_open_common(bs, filename, flags);
 180 }
 181
 182 /* XXX: use host sector size if necessary with:
 183 #ifdef DIOCGSECTORSIZE
 184         {
 185             unsigned int sectorsize = 512;
 186             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 187                 sectorsize > bufsize)
 188                 bufsize = sectorsize;
 189         }
 190 #endif
 191 #ifdef CONFIG_COCOA
 192         u_int32_t   blockSize = 512;
 193         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 194             bufsize = blockSize;
 195         }
 196 #endif
 197 */
 198
 199 /*
 200  * offset and count are in bytes, but must be multiples of 512 for files
 201  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 202  *
 203  * This function may be called without alignment if the caller ensures
 204  * that O_DIRECT is not in effect.
 205  */
 206 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 207                      uint8_t *buf, int count)
 208 {
 209     BDRVRawState *s = bs->opaque;
 210     int ret;
 211
 212     ret = fd_open(bs);
 213     if (ret < 0)
 214         return ret;
 215
 216     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 217         ++(s->lseek_err_cnt);
 218         if(s->lseek_err_cnt <= 10) {
 219             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 220                               "] lseek failed : %d = %s\n",
 221                               s->fd, bs->filename, offset, buf, count,
 222                               bs->total_sectors, errno, strerror(errno));
 223         }
 224         return -1;
 225     }
 226     s->lseek_err_cnt=0;
 227
 228     ret = read(s->fd, buf, count);
 229     if (ret == count)
 230         goto label__raw_read__success;
 231
 232     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 233                       "] read failed %d : %d = %s\n",
 234                       s->fd, bs->filename, offset, buf, count,
 235                       bs->total_sectors, ret, errno, strerror(errno));
 236
 237     /* Try harder for CDrom. */
 238     if (bs->type == BDRV_TYPE_CDROM) {
 239         lseek(s->fd, offset, SEEK_SET);
 240         ret = read(s->fd, buf, count);
 241         if (ret == count)
 242             goto label__raw_read__success;
 243         lseek(s->fd, offset, SEEK_SET);
 244         ret = read(s->fd, buf, count);
 245         if (ret == count)
 246             goto label__raw_read__success;
 247
 248         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 249                           "] retry read failed %d : %d = %s\n",
 250                           s->fd, bs->filename, offset, buf, count,
 251                           bs->total_sectors, ret, errno, strerror(errno));
 252     }
 253
 254 label__raw_read__success:
 255
 256     return  (ret < 0) ? -errno : ret;
 257 }
 258
 259 /*
 260  * offset and count are in bytes, but must be multiples of 512 for files
 261  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 262  *
 263  * This function may be called without alignment if the caller ensures
 264  * that O_DIRECT is not in effect.
 265  */
 266 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 267                       const uint8_t *buf, int count)
 268 {
 269     BDRVRawState *s = bs->opaque;
 270     int ret;
 271
 272     ret = fd_open(bs);
 273     if (ret < 0)
 274         return -errno;
 275
 276     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 277         ++(s->lseek_err_cnt);
 278         if(s->lseek_err_cnt) {
 279             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 280                               PRId64 "] lseek failed : %d = %s\n",
 281                               s->fd, bs->filename, offset, buf, count,
 282                               bs->total_sectors, errno, strerror(errno));
 283         }
 284         return -EIO;
 285     }
 286     s->lseek_err_cnt = 0;
 287
 288     ret = write(s->fd, buf, count);
 289     if (ret == count)
 290         goto label__raw_write__success;
 291
 292     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 293                       "] write failed %d : %d = %s\n",
 294                       s->fd, bs->filename, offset, buf, count,
 295                       bs->total_sectors, ret, errno, strerror(errno));
 296
 297 label__raw_write__success:
 298
 299     return  (ret < 0) ? -errno : ret;
 300 }
 301
 302
 303 /*
 304  * offset and count are in bytes and possibly not aligned. For files opened
 305  * with O_DIRECT, necessary alignments are ensured before calling
 306  * raw_pread_aligned to do the actual read.
 307  */
 308 static int raw_pread(BlockDriverState *bs, int64_t offset,
 309                      uint8_t *buf, int count)
 310 {
 311     BDRVRawState *s = bs->opaque;
 312     int size, ret, shift, sum;
 313
 314     sum = 0;
 315
 316     if (s->aligned_buf != NULL)  {
 317
 318         if (offset & 0x1ff) {
 319             /* align offset on a 512 bytes boundary */
 320
 321             shift = offset & 0x1ff;
 322             size = (shift + count + 0x1ff) & ~0x1ff;
 323             if (size > ALIGNED_BUFFER_SIZE)
 324                 size = ALIGNED_BUFFER_SIZE;
 325             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 326             if (ret < 0)
 327                 return ret;
 328
 329             size = 512 - shift;
 330             if (size > count)
 331                 size = count;
 332             memcpy(buf, s->aligned_buf + shift, size);
 333
 334             buf += size;
 335             offset += size;
 336             count -= size;
 337             sum += size;
 338
 339             if (count == 0)
 340                 return sum;
 341         }
 342         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 343
 344             /* read on aligned buffer */
 345
 346             while (count) {
 347
 348                 size = (count + 0x1ff) & ~0x1ff;
 349                 if (size > ALIGNED_BUFFER_SIZE)
 350                     size = ALIGNED_BUFFER_SIZE;
 351
 352                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 353                 if (ret < 0)
 354                     return ret;
 355
 356                 size = ret;
 357                 if (size > count)
 358                     size = count;
 359
 360                 memcpy(buf, s->aligned_buf, size);
 361
 362                 buf += size;
 363                 offset += size;
 364                 count -= size;
 365                 sum += size;
 366             }
 367
 368             return sum;
 369         }
 370     }
 371
 372     return raw_pread_aligned(bs, offset, buf, count) + sum;
 373 }
 374
 375 static int raw_read(BlockDriverState *bs, int64_t sector_num,
 376                     uint8_t *buf, int nb_sectors)
 377 {
 378     int ret;
 379
 380     ret = raw_pread(bs, sector_num * 512, buf, nb_sectors * 512);
 381     if (ret == (nb_sectors * 512))
 382         ret = 0;
 383     return ret;
 384 }
 385
 386 /*
 387  * offset and count are in bytes and possibly not aligned. For files opened
 388  * with O_DIRECT, necessary alignments are ensured before calling
 389  * raw_pwrite_aligned to do the actual write.
 390  */
 391 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 392                       const uint8_t *buf, int count)
 393 {
 394     BDRVRawState *s = bs->opaque;
 395     int size, ret, shift, sum;
 396
 397     sum = 0;
 398
 399     if (s->aligned_buf != NULL) {
 400
 401         if (offset & 0x1ff) {
 402             /* align offset on a 512 bytes boundary */
 403             shift = offset & 0x1ff;
 404             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 405             if (ret < 0)
 406                 return ret;
 407
 408             size = 512 - shift;
 409             if (size > count)
 410                 size = count;
 411             memcpy(s->aligned_buf + shift, buf, size);
 412
 413             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 414             if (ret < 0)
 415                 return ret;
 416
 417             buf += size;
 418             offset += size;
 419             count -= size;
 420             sum += size;
 421
 422             if (count == 0)
 423                 return sum;
 424         }
 425         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 426
 427             while ((size = (count & ~0x1ff)) != 0) {
 428
 429                 if (size > ALIGNED_BUFFER_SIZE)
 430                     size = ALIGNED_BUFFER_SIZE;
 431
 432                 memcpy(s->aligned_buf, buf, size);
 433
 434                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 435                 if (ret < 0)
 436                     return ret;
 437
 438                 buf += ret;
 439                 offset += ret;
 440                 count -= ret;
 441                 sum += ret;
 442             }
 443             /* here, count < 512 because (count & ~0x1ff) == 0 */
 444             if (count) {
 445                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 446                 if (ret < 0)
 447                     return ret;
 448                  memcpy(s->aligned_buf, buf, count);
 449
 450                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 451                  if (ret < 0)
 452                      return ret;
 453                  if (count < ret)
 454                      ret = count;
 455
 456                  sum += ret;
 457             }
 458             return sum;
 459         }
 460     }
 461     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 462 }
 463
 464 static int raw_write(BlockDriverState *bs, int64_t sector_num,
 465                      const uint8_t *buf, int nb_sectors)
 466 {
 467     int ret;
 468     ret = raw_pwrite(bs, sector_num * 512, buf, nb_sectors * 512);
 469     if (ret == (nb_sectors * 512))
 470         ret = 0;
 471     return ret;
 472 }
 473
 474 #ifdef CONFIG_AIO
 475 /***********************************************************/
 476 /* Unix AIO using POSIX AIO */
 477
 478 typedef struct RawAIOCB {
 479     BlockDriverAIOCB common;
 480     struct qemu_paiocb aiocb;
 481     struct RawAIOCB *next;
 482     int ret;
 483 } RawAIOCB;
 484
 485 typedef struct PosixAioState
 486 {
 487     int rfd, wfd;
 488     RawAIOCB *first_aio;
 489 } PosixAioState;
 490
 491 static void posix_aio_read(void *opaque)
 492 {
 493     PosixAioState *s = opaque;
 494     RawAIOCB *acb, **pacb;
 495     int ret;
 496     ssize_t len;
 497
 498     /* read all bytes from signal pipe */
 499     for (;;) {
 500         char bytes[16];
 501
 502         len = read(s->rfd, bytes, sizeof(bytes));
 503         if (len == -1 && errno == EINTR)
 504             continue; /* try again */
 505         if (len == sizeof(bytes))
 506             continue; /* more to read */
 507         break;
 508     }
 509
 510     for(;;) {
 511         pacb = &s->first_aio;
 512         for(;;) {
 513             acb = *pacb;
 514             if (!acb)
 515                 goto the_end;
 516             ret = qemu_paio_error(&acb->aiocb);
 517             if (ret == ECANCELED) {
 518                 /* remove the request */
 519                 *pacb = acb->next;
 520                 qemu_aio_release(acb);
 521             } else if (ret != EINPROGRESS) {
 522                 /* end of aio */
 523                 if (ret == 0) {
 524                     ret = qemu_paio_return(&acb->aiocb);
 525                     if (ret == acb->aiocb.aio_nbytes)
 526                         ret = 0;
 527                     else
 528                         ret = -EINVAL;
 529                 } else {
 530                     ret = -ret;
 531                 }
 532                 /* remove the request */
 533                 *pacb = acb->next;
 534                 /* call the callback */
 535                 acb->common.cb(acb->common.opaque, ret);
 536                 qemu_aio_release(acb);
 537                 break;
 538             } else {
 539                 pacb = &acb->next;
 540             }
 541         }
 542     }
 543  the_end: ;
 544 }
 545
 546 static int posix_aio_flush(void *opaque)
 547 {
 548     PosixAioState *s = opaque;
 549     return !!s->first_aio;
 550 }
 551
 552 static PosixAioState *posix_aio_state;
 553
 554 static void aio_signal_handler(int signum)
 555 {
 556     if (posix_aio_state) {
 557         char byte = 0;
 558
 559         write(posix_aio_state->wfd, &byte, sizeof(byte));
 560     }
 561
 562     qemu_service_io();
 563 }
 564
 565 static int posix_aio_init(void)
 566 {
 567     struct sigaction act;
 568     PosixAioState *s;
 569     int fds[2];
 570     struct qemu_paioinit ai;
 571
 572     if (posix_aio_state)
 573         return 0;
 574
 575     s = qemu_malloc(sizeof(PosixAioState));
 576
 577     sigfillset(&act.sa_mask);
 578     act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
 579     act.sa_handler = aio_signal_handler;
 580     sigaction(SIGUSR2, &act, NULL);
 581
 582     s->first_aio = NULL;
 583     if (pipe(fds) == -1) {
 584         fprintf(stderr, "failed to create pipe\n");
 585         return -errno;
 586     }
 587
 588     s->rfd = fds[0];
 589     s->wfd = fds[1];
 590
 591     fcntl(s->rfd, F_SETFL, O_NONBLOCK);
 592     fcntl(s->wfd, F_SETFL, O_NONBLOCK);
 593
 594     qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
 595
 596     memset(&ai, 0, sizeof(ai));
 597     ai.aio_threads = 64;
 598     ai.aio_num = 64;
 599     qemu_paio_init(&ai);
 600
 601     posix_aio_state = s;
 602
 603     return 0;
 604 }
 605
 606 static void raw_aio_remove(RawAIOCB *acb)
 607 {
 608     RawAIOCB **pacb;
 609
 610     /* remove the callback from the queue */
 611     pacb = &posix_aio_state->first_aio;
 612     for(;;) {
 613         if (*pacb == NULL) {
 614             fprintf(stderr, "raw_aio_remove: aio request not found!\n");
 615             break;
 616         } else if (*pacb == acb) {
 617             *pacb = acb->next;
 618             qemu_aio_release(acb);
 619             break;
 620         }
 621         pacb = &(*pacb)->next;
 622     }
 623 }
 624
 625 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 626 {
 627     int ret;
 628     RawAIOCB *acb = (RawAIOCB *)blockacb;
 629
 630     ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 631     if (ret == QEMU_PAIO_NOTCANCELED) {
 632         /* fail safe: if the aio could not be canceled, we wait for
 633            it */
 634         while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
 635     }
 636
 637     raw_aio_remove(acb);
 638 }
 639
 640 static AIOPool raw_aio_pool = {
 641     .aiocb_size         = sizeof(RawAIOCB),
 642     .cancel             = raw_aio_cancel,
 643 };
 644
 645 static RawAIOCB *raw_aio_setup(BlockDriverState *bs, int64_t sector_num,
 646         QEMUIOVector *qiov, int nb_sectors,
 647         BlockDriverCompletionFunc *cb, void *opaque)
 648 {
 649     BDRVRawState *s = bs->opaque;
 650     RawAIOCB *acb;
 651
 652     if (fd_open(bs) < 0)
 653         return NULL;
 654
 655     acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque);
 656     if (!acb)
 657         return NULL;
 658     acb->aiocb.aio_fildes = s->fd;
 659     acb->aiocb.ev_signo = SIGUSR2;
 660     acb->aiocb.aio_iov = qiov->iov;
 661     acb->aiocb.aio_niov = qiov->niov;
 662     acb->aiocb.aio_nbytes = nb_sectors * 512;
 663     acb->aiocb.aio_offset = sector_num * 512;
 664     acb->aiocb.aio_flags = 0;
 665
 666     /*
 667      * If O_DIRECT is used the buffer needs to be aligned on a sector
 668      * boundary. Tell the low level code to ensure that in case it's
 669      * not done yet.
 670      */
 671     if (s->aligned_buf)
 672         acb->aiocb.aio_flags |= QEMU_AIO_SECTOR_ALIGNED;
 673
 674     acb->next = posix_aio_state->first_aio;
 675     posix_aio_state->first_aio = acb;
 676     return acb;
 677 }
 678
 679 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 680         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 681         BlockDriverCompletionFunc *cb, void *opaque)
 682 {
 683     RawAIOCB *acb;
 684
 685     acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
 686     if (!acb)
 687         return NULL;
 688     if (qemu_paio_read(&acb->aiocb) < 0) {
 689         raw_aio_remove(acb);
 690         return NULL;
 691     }
 692     return &acb->common;
 693 }
 694
 695 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 696         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 697         BlockDriverCompletionFunc *cb, void *opaque)
 698 {
 699     RawAIOCB *acb;
 700
 701     acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
 702     if (!acb)
 703         return NULL;
 704     if (qemu_paio_write(&acb->aiocb) < 0) {
 705         raw_aio_remove(acb);
 706         return NULL;
 707     }
 708     return &acb->common;
 709 }
 710 #else /* CONFIG_AIO */
 711 static int posix_aio_init(void)
 712 {
 713     return 0;
 714 }
 715 #endif /* CONFIG_AIO */
 716
 717
 718 static void raw_close(BlockDriverState *bs)
 719 {
 720     BDRVRawState *s = bs->opaque;
 721     if (s->fd >= 0) {
 722         close(s->fd);
 723         s->fd = -1;
 724         if (s->aligned_buf != NULL)
 725             qemu_free(s->aligned_buf);
 726     }
 727 }
 728
 729 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 730 {
 731     BDRVRawState *s = bs->opaque;
 732     if (s->type != FTYPE_FILE)
 733         return -ENOTSUP;
 734     if (ftruncate(s->fd, offset) < 0)
 735         return -errno;
 736     return 0;
 737 }
 738
 739 #ifdef __OpenBSD__
 740 static int64_t raw_getlength(BlockDriverState *bs)
 741 {
 742     BDRVRawState *s = bs->opaque;
 743     int fd = s->fd;
 744     struct stat st;
 745
 746     if (fstat(fd, &st))
 747         return -1;
 748     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 749         struct disklabel dl;
 750
 751         if (ioctl(fd, DIOCGDINFO, &dl))
 752             return -1;
 753         return (uint64_t)dl.d_secsize *
 754             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 755     } else
 756         return st.st_size;
 757 }
 758 #else /* !__OpenBSD__ */
 759 static int64_t  raw_getlength(BlockDriverState *bs)
 760 {
 761     BDRVRawState *s = bs->opaque;
 762     int fd = s->fd;
 763     int64_t size;
 764 #ifdef HOST_BSD
 765     struct stat sb;
 766 #ifdef __FreeBSD__
 767     int reopened = 0;
 768 #endif
 769 #endif
 770 #ifdef __sun__
 771     struct dk_minfo minfo;
 772     int rv;
 773 #endif
 774     int ret;
 775
 776     ret = fd_open(bs);
 777     if (ret < 0)
 778         return ret;
 779
 780 #ifdef HOST_BSD
 781 #ifdef __FreeBSD__
 782 again:
 783 #endif
 784     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 785 #ifdef DIOCGMEDIASIZE
 786         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 787 #elif defined(DIOCGPART)
 788         {
 789                 struct partinfo pi;
 790                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 791                         size = pi.media_size;
 792                 else
 793                         size = 0;
 794         }
 795         if (size == 0)
 796 #endif
 797 #ifdef CONFIG_COCOA
 798         size = LONG_LONG_MAX;
 799 #else
 800         size = lseek(fd, 0LL, SEEK_END);
 801 #endif
 802 #ifdef __FreeBSD__
 803         switch(s->type) {
 804         case FTYPE_CD:
 805             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
 806             if (size == 2048LL * (unsigned)-1)
 807                 size = 0;
 808             /* XXX no disc?  maybe we need to reopen... */
 809             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
 810                 reopened = 1;
 811                 goto again;
 812             }
 813         }
 814 #endif
 815     } else
 816 #endif
 817 #ifdef __sun__
 818     /*
 819      * use the DKIOCGMEDIAINFO ioctl to read the size.
 820      */
 821     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 822     if ( rv != -1 ) {
 823         size = minfo.dki_lbsize * minfo.dki_capacity;
 824     } else /* there are reports that lseek on some devices
 825               fails, but irc discussion said that contingency
 826               on contingency was overkill */
 827 #endif
 828     {
 829         size = lseek(fd, 0, SEEK_END);
 830     }
 831     return size;
 832 }
 833 #endif
 834
 835 static int raw_create(const char *filename, QEMUOptionParameter *options)
 836 {
 837     int fd;
 838     int64_t total_size = 0;
 839
 840     /* Read out options */
 841     while (options && options->name) {
 842         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 843             total_size = options->value.n / 512;
 844         }
 845         options++;
 846     }
 847
 848     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 849               0644);
 850     if (fd < 0)
 851         return -EIO;
 852     ftruncate(fd, total_size * 512);
 853     close(fd);
 854     return 0;
 855 }
 856
 857 static void raw_flush(BlockDriverState *bs)
 858 {
 859     BDRVRawState *s = bs->opaque;
 860     fsync(s->fd);
 861 }
 862
 863
 864 static QEMUOptionParameter raw_create_options[] = {
 865     {
 866         .name = BLOCK_OPT_SIZE,
 867         .type = OPT_SIZE,
 868         .help = "Virtual disk size"
 869     },
 870     { NULL }
 871 };
 872
 873 static BlockDriver bdrv_raw = {
 874     .format_name = "raw",
 875     .instance_size = sizeof(BDRVRawState),
 876     .bdrv_probe = NULL, /* no probe for protocols */
 877     .bdrv_open = raw_open,
 878     .bdrv_read = raw_read,
 879     .bdrv_write = raw_write,
 880     .bdrv_close = raw_close,
 881     .bdrv_create = raw_create,
 882     .bdrv_flush = raw_flush,
 883
 884 #ifdef CONFIG_AIO
 885     .bdrv_aio_readv = raw_aio_readv,
 886     .bdrv_aio_writev = raw_aio_writev,
 887 #endif
 888
 889     .bdrv_truncate = raw_truncate,
 890     .bdrv_getlength = raw_getlength,
 891
 892     .create_options = raw_create_options,
 893 };
 894
 895 /***********************************************/
 896 /* host device */
 897
 898 #ifdef CONFIG_COCOA
 899 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 900 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 901
 902 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 903 {
 904     kern_return_t       kernResult;
 905     mach_port_t     masterPort;
 906     CFMutableDictionaryRef  classesToMatch;
 907
 908     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 909     if ( KERN_SUCCESS != kernResult ) {
 910         printf( "IOMasterPort returned %d\n", kernResult );
 911     }
 912
 913     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 914     if ( classesToMatch == NULL ) {
 915         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 916     } else {
 917     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 918     }
 919     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 920     if ( KERN_SUCCESS != kernResult )
 921     {
 922         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 923     }
 924
 925     return kernResult;
 926 }
 927
 928 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 929 {
 930     io_object_t     nextMedia;
 931     kern_return_t   kernResult = KERN_FAILURE;
 932     *bsdPath = '\0';
 933     nextMedia = IOIteratorNext( mediaIterator );
 934     if ( nextMedia )
 935     {
 936         CFTypeRef   bsdPathAsCFString;
 937     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 938         if ( bsdPathAsCFString ) {
 939             size_t devPathLength;
 940             strcpy( bsdPath, _PATH_DEV );
 941             strcat( bsdPath, "r" );
 942             devPathLength = strlen( bsdPath );
 943             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 944                 kernResult = KERN_SUCCESS;
 945             }
 946             CFRelease( bsdPathAsCFString );
 947         }
 948         IOObjectRelease( nextMedia );
 949     }
 950
 951     return kernResult;
 952 }
 953
 954 #endif
 955
 956 static int hdev_probe_device(const char *filename)
 957 {
 958     struct stat st;
 959
 960     /* allow a dedicated CD-ROM driver to match with a higher priority */
 961     if (strstart(filename, "/dev/cdrom", NULL))
 962         return 50;
 963
 964     if (stat(filename, &st) >= 0 &&
 965             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 966         return 100;
 967     }
 968
 969     return 0;
 970 }
 971
 972 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 973 {
 974     BDRVRawState *s = bs->opaque;
 975
 976 #ifdef CONFIG_COCOA
 977     if (strstart(filename, "/dev/cdrom", NULL)) {
 978         kern_return_t kernResult;
 979         io_iterator_t mediaIterator;
 980         char bsdPath[ MAXPATHLEN ];
 981         int fd;
 982
 983         kernResult = FindEjectableCDMedia( &mediaIterator );
 984         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 985
 986         if ( bsdPath[ 0 ] != '\0' ) {
 987             strcat(bsdPath,"s0");
 988             /* some CDs don't have a partition 0 */
 989             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 990             if (fd < 0) {
 991                 bsdPath[strlen(bsdPath)-1] = '1';
 992             } else {
 993                 close(fd);
 994             }
 995             filename = bsdPath;
 996         }
 997
 998         if ( mediaIterator )
 999             IOObjectRelease( mediaIterator );
1000     }
1001 #endif
1002
1003     s->type = FTYPE_FILE;
1004 #if defined(__linux__) && defined(CONFIG_AIO)
1005     if (strstart(filename, "/dev/sg", NULL)) {
1006         bs->sg = 1;
1007     }
1008 #endif
1009
1010     return raw_open_common(bs, filename, flags);
1011 }
1012
1013 #if defined(__linux__)
1014 /* Note: we do not have a reliable method to detect if the floppy is
1015    present. The current method is to try to open the floppy at every
1016    I/O and to keep it opened during a few hundreds of ms. */
1017 static int fd_open(BlockDriverState *bs)
1018 {
1019     BDRVRawState *s = bs->opaque;
1020     int last_media_present;
1021
1022     if (s->type != FTYPE_FD)
1023         return 0;
1024     last_media_present = (s->fd >= 0);
1025     if (s->fd >= 0 &&
1026         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1027         close(s->fd);
1028         s->fd = -1;
1029 #ifdef DEBUG_FLOPPY
1030         printf("Floppy closed\n");
1031 #endif
1032     }
1033     if (s->fd < 0) {
1034         if (s->fd_got_error &&
1035             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1036 #ifdef DEBUG_FLOPPY
1037             printf("No floppy (open delayed)\n");
1038 #endif
1039             return -EIO;
1040         }
1041         s->fd = open(bs->filename, s->open_flags & ~O_NONBLOCK);
1042         if (s->fd < 0) {
1043             s->fd_error_time = qemu_get_clock(rt_clock);
1044             s->fd_got_error = 1;
1045             if (last_media_present)
1046                 s->fd_media_changed = 1;
1047 #ifdef DEBUG_FLOPPY
1048             printf("No floppy\n");
1049 #endif
1050             return -EIO;
1051         }
1052 #ifdef DEBUG_FLOPPY
1053         printf("Floppy opened\n");
1054 #endif
1055     }
1056     if (!last_media_present)
1057         s->fd_media_changed = 1;
1058     s->fd_open_time = qemu_get_clock(rt_clock);
1059     s->fd_got_error = 0;
1060     return 0;
1061 }
1062
1063 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1064 {
1065     BDRVRawState *s = bs->opaque;
1066
1067     return ioctl(s->fd, req, buf);
1068 }
1069
1070 #ifdef CONFIG_AIO
1071 static BlockDriverAIOCB *raw_aio_ioctl(BlockDriverState *bs,
1072         unsigned long int req, void *buf,
1073         BlockDriverCompletionFunc *cb, void *opaque)
1074 {
1075     BDRVRawState *s = bs->opaque;
1076     RawAIOCB *acb;
1077
1078     if (fd_open(bs) < 0)
1079         return NULL;
1080
1081     acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque);
1082     if (!acb)
1083         return NULL;
1084     acb->aiocb.aio_fildes = s->fd;
1085     acb->aiocb.ev_signo = SIGUSR2;
1086     acb->aiocb.aio_offset = 0;
1087     acb->aiocb.aio_flags = 0;
1088
1089     acb->next = posix_aio_state->first_aio;
1090     posix_aio_state->first_aio = acb;
1091
1092     acb->aiocb.aio_ioctl_buf = buf;
1093     acb->aiocb.aio_ioctl_cmd = req;
1094     if (qemu_paio_ioctl(&acb->aiocb) < 0) {
1095         raw_aio_remove(acb);
1096         return NULL;
1097     }
1098
1099     return &acb->common;
1100 }
1101 #endif
1102
1103 #elif defined(__FreeBSD__)
1104 static int fd_open(BlockDriverState *bs)
1105 {
1106     BDRVRawState *s = bs->opaque;
1107
1108     /* this is just to ensure s->fd is sane (its called by io ops) */
1109     if (s->fd >= 0)
1110         return 0;
1111     return -EIO;
1112 }
1113
1114 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1115 {
1116     return -ENOTSUP;
1117 }
1118 #else /* !linux && !FreeBSD */
1119
1120 static int fd_open(BlockDriverState *bs)
1121 {
1122     return 0;
1123 }
1124
1125 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1126 {
1127     return -ENOTSUP;
1128 }
1129
1130 static BlockDriverAIOCB *raw_aio_ioctl(BlockDriverState *bs,
1131         unsigned long int req, void *buf,
1132         BlockDriverCompletionFunc *cb, void *opaque)
1133 {
1134     return NULL;
1135 }
1136 #endif /* !linux && !FreeBSD */
1137
1138 static int hdev_create(const char *filename, QEMUOptionParameter *options)
1139 {
1140     int fd;
1141     int ret = 0;
1142     struct stat stat_buf;
1143     int64_t total_size = 0;
1144
1145     /* Read out options */
1146     while (options && options->name) {
1147         if (!strcmp(options->name, "size")) {
1148             total_size = options->value.n / 512;
1149         }
1150         options++;
1151     }
1152
1153     fd = open(filename, O_WRONLY | O_BINARY);
1154     if (fd < 0)
1155         return -EIO;
1156
1157     if (fstat(fd, &stat_buf) < 0)
1158         ret = -EIO;
1159     else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
1160         ret = -EIO;
1161     else if (lseek(fd, 0, SEEK_END) < total_size * 512)
1162         ret = -ENOSPC;
1163
1164     close(fd);
1165     return ret;
1166 }
1167
1168 static BlockDriver bdrv_host_device = {
1169     .format_name        = "host_device",
1170     .instance_size      = sizeof(BDRVRawState),
1171     .bdrv_probe_device  = hdev_probe_device,
1172     .bdrv_open          = hdev_open,
1173     .bdrv_close         = raw_close,
1174     .bdrv_create        = hdev_create,
1175     .bdrv_flush         = raw_flush,
1176
1177 #ifdef CONFIG_AIO
1178     .bdrv_aio_readv     = raw_aio_readv,
1179     .bdrv_aio_writev    = raw_aio_writev,
1180 #endif
1181
1182     .bdrv_read          = raw_read,
1183     .bdrv_write         = raw_write,
1184     .bdrv_getlength     = raw_getlength,
1185
1186     /* generic scsi device */
1187     .bdrv_ioctl         = raw_ioctl,
1188 #ifdef CONFIG_AIO
1189     .bdrv_aio_ioctl     = raw_aio_ioctl,
1190 #endif
1191 };
1192
1193 #ifdef __linux__
1194 static int floppy_open(BlockDriverState *bs, const char *filename, int flags)
1195 {
1196     BDRVRawState *s = bs->opaque;
1197     int ret;
1198
1199     posix_aio_init();
1200
1201     s->type = FTYPE_FD;
1202     /* open will not fail even if no floppy is inserted */
1203     s->open_flags |= O_NONBLOCK;
1204
1205     ret = raw_open_common(bs, filename, flags);
1206     if (ret)
1207         return ret;
1208
1209     /* close fd so that we can reopen it as needed */
1210     close(s->fd);
1211     s->fd = -1;
1212     s->fd_media_changed = 1;
1213
1214     return 0;
1215 }
1216
1217 static int floppy_probe_device(const char *filename)
1218 {
1219     if (strstart(filename, "/dev/fd", NULL))
1220         return 100;
1221     return 0;
1222 }
1223
1224
1225 static int floppy_is_inserted(BlockDriverState *bs)
1226 {
1227     return fd_open(bs) >= 0;
1228 }
1229
1230 static int floppy_media_changed(BlockDriverState *bs)
1231 {
1232     BDRVRawState *s = bs->opaque;
1233     int ret;
1234
1235     /*
1236      * XXX: we do not have a true media changed indication.
1237      * It does not work if the floppy is changed without trying to read it.
1238      */
1239     fd_open(bs);
1240     ret = s->fd_media_changed;
1241     s->fd_media_changed = 0;
1242 #ifdef DEBUG_FLOPPY
1243     printf("Floppy changed=%d\n", ret);
1244 #endif
1245     return ret;
1246 }
1247
1248 static int floppy_eject(BlockDriverState *bs, int eject_flag)
1249 {
1250     BDRVRawState *s = bs->opaque;
1251     int fd;
1252
1253     if (s->fd >= 0) {
1254         close(s->fd);
1255         s->fd = -1;
1256     }
1257     fd = open(bs->filename, s->open_flags | O_NONBLOCK);
1258     if (fd >= 0) {
1259         if (ioctl(fd, FDEJECT, 0) < 0)
1260             perror("FDEJECT");
1261         close(fd);
1262     }
1263
1264     return 0;
1265 }
1266
1267 static BlockDriver bdrv_host_floppy = {
1268     .format_name        = "host_floppy",
1269     .instance_size      = sizeof(BDRVRawState),
1270     .bdrv_probe_device  = floppy_probe_device,
1271     .bdrv_open          = floppy_open,
1272     .bdrv_close         = raw_close,
1273     .bdrv_create        = hdev_create,
1274     .bdrv_flush         = raw_flush,
1275
1276 #ifdef CONFIG_AIO
1277     .bdrv_aio_readv     = raw_aio_readv,
1278     .bdrv_aio_writev    = raw_aio_writev,
1279 #endif
1280
1281     .bdrv_read          = raw_read,
1282     .bdrv_write         = raw_write,
1283     .bdrv_getlength     = raw_getlength,
1284
1285     /* removable device support */
1286     .bdrv_is_inserted   = floppy_is_inserted,
1287     .bdrv_media_changed = floppy_media_changed,
1288     .bdrv_eject         = floppy_eject,
1289
1290     /* generic scsi device */
1291     .bdrv_ioctl         = raw_ioctl,
1292 #ifdef CONFIG_AIO
1293     .bdrv_aio_ioctl     = raw_aio_ioctl,
1294 #endif
1295 };
1296
1297 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1298 {
1299     BDRVRawState *s = bs->opaque;
1300
1301     /* open will not fail even if no CD is inserted */
1302     s->open_flags |= O_NONBLOCK;
1303     s->type = FTYPE_CD;
1304
1305     return raw_open_common(bs, filename, flags);
1306 }
1307
1308 static int cdrom_probe_device(const char *filename)
1309 {
1310     if (strstart(filename, "/dev/cd", NULL))
1311         return 100;
1312     return 0;
1313 }
1314
1315 static int cdrom_is_inserted(BlockDriverState *bs)
1316 {
1317     BDRVRawState *s = bs->opaque;
1318     int ret;
1319
1320     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1321     if (ret == CDS_DISC_OK)
1322         return 1;
1323     return 0;
1324 }
1325
1326 static int cdrom_eject(BlockDriverState *bs, int eject_flag)
1327 {
1328     BDRVRawState *s = bs->opaque;
1329
1330     if (eject_flag) {
1331         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1332             perror("CDROMEJECT");
1333     } else {
1334         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1335             perror("CDROMEJECT");
1336     }
1337
1338     return 0;
1339 }
1340
1341 static int cdrom_set_locked(BlockDriverState *bs, int locked)
1342 {
1343     BDRVRawState *s = bs->opaque;
1344
1345     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1346         /*
1347          * Note: an error can happen if the distribution automatically
1348          * mounts the CD-ROM
1349          */
1350         /* perror("CDROM_LOCKDOOR"); */
1351     }
1352
1353     return 0;
1354 }
1355
1356 static BlockDriver bdrv_host_cdrom = {
1357     .format_name        = "host_cdrom",
1358     .instance_size      = sizeof(BDRVRawState),
1359     .bdrv_probe_device  = cdrom_probe_device,
1360     .bdrv_open          = cdrom_open,
1361     .bdrv_close         = raw_close,
1362     .bdrv_create        = hdev_create,
1363     .bdrv_flush         = raw_flush,
1364
1365 #ifdef CONFIG_AIO
1366     .bdrv_aio_readv     = raw_aio_readv,
1367     .bdrv_aio_writev    = raw_aio_writev,
1368 #endif
1369
1370     .bdrv_read          = raw_read,
1371     .bdrv_write         = raw_write,
1372     .bdrv_getlength     = raw_getlength,
1373
1374     /* removable device support */
1375     .bdrv_is_inserted   = cdrom_is_inserted,
1376     .bdrv_eject         = cdrom_eject,
1377     .bdrv_set_locked    = cdrom_set_locked,
1378
1379     /* generic scsi device */
1380     .bdrv_ioctl         = raw_ioctl,
1381 #ifdef CONFIG_AIO
1382     .bdrv_aio_ioctl     = raw_aio_ioctl,
1383 #endif
1384 };
1385 #endif /* __linux__ */
1386
1387 #ifdef __FreeBSD__
1388 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1389 {
1390     BDRVRawState *s = bs->opaque;
1391     int ret;
1392
1393     s->type = FTYPE_CD;
1394
1395     ret = raw_open_common(bs, filename, flags);
1396     if (ret)
1397         return ret;
1398
1399     /* make sure the door isnt locked at this time */
1400     ioctl(s->fd, CDIOCALLOW);
1401     return 0;
1402 }
1403
1404 static int cdrom_probe_device(const char *filename)
1405 {
1406     if (strstart(filename, "/dev/cd", NULL) ||
1407             strstart(filename, "/dev/acd", NULL))
1408         return 100;
1409     return 0;
1410 }
1411
1412 static int cdrom_reopen(BlockDriverState *bs)
1413 {
1414     BDRVRawState *s = bs->opaque;
1415     int fd;
1416
1417     /*
1418      * Force reread of possibly changed/newly loaded disc,
1419      * FreeBSD seems to not notice sometimes...
1420      */
1421     if (s->fd >= 0)
1422         close(s->fd);
1423     fd = open(bs->filename, s->open_flags, 0644);
1424     if (fd < 0) {
1425         s->fd = -1;
1426         return -EIO;
1427     }
1428     s->fd = fd;
1429
1430     /* make sure the door isnt locked at this time */
1431     ioctl(s->fd, CDIOCALLOW);
1432     return 0;
1433 }
1434
1435 static int cdrom_is_inserted(BlockDriverState *bs)
1436 {
1437     return raw_getlength(bs) > 0;
1438 }
1439
1440 static int cdrom_eject(BlockDriverState *bs, int eject_flag)
1441 {
1442     BDRVRawState *s = bs->opaque;
1443
1444     if (s->fd < 0)
1445         return -ENOTSUP;
1446
1447     (void) ioctl(s->fd, CDIOCALLOW);
1448
1449     if (eject_flag) {
1450         if (ioctl(s->fd, CDIOCEJECT) < 0)
1451             perror("CDIOCEJECT");
1452     } else {
1453         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1454             perror("CDIOCCLOSE");
1455     }
1456
1457     if (cdrom_reopen(bs) < 0)
1458         return -ENOTSUP;
1459     return 0;
1460 }
1461
1462 static int cdrom_set_locked(BlockDriverState *bs, int locked)
1463 {
1464     BDRVRawState *s = bs->opaque;
1465
1466     if (s->fd < 0)
1467         return -ENOTSUP;
1468     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1469         /*
1470          * Note: an error can happen if the distribution automatically
1471          * mounts the CD-ROM
1472          */
1473         /* perror("CDROM_LOCKDOOR"); */
1474     }
1475
1476     return 0;
1477 }
1478
1479 static BlockDriver bdrv_host_cdrom = {
1480     .format_name        = "host_cdrom",
1481     .instance_size      = sizeof(BDRVRawState),
1482     .bdrv_probe_device  = cdrom_probe_device,
1483     .bdrv_open          = cdrom_open,
1484     .bdrv_close         = raw_close,
1485     .bdrv_create        = hdev_create,
1486     .bdrv_flush         = raw_flush,
1487
1488 #ifdef CONFIG_AIO
1489     .bdrv_aio_readv     = raw_aio_readv,
1490     .bdrv_aio_writev    = raw_aio_writev,
1491 #endif
1492
1493     .bdrv_read          = raw_read,
1494     .bdrv_write         = raw_write,
1495     .bdrv_getlength     = raw_getlength,
1496
1497     /* removable device support */
1498     .bdrv_is_inserted   = cdrom_is_inserted,
1499     .bdrv_eject         = cdrom_eject,
1500     .bdrv_set_locked    = cdrom_set_locked,
1501
1502     /* generic scsi device */
1503     .bdrv_ioctl         = raw_ioctl,
1504 #ifdef CONFIG_AIO
1505     .bdrv_aio_ioctl     = raw_aio_ioctl,
1506 #endif
1507 };
1508 #endif /* __FreeBSD__ */
1509
1510 static void bdrv_raw_init(void)
1511 {
1512     /*
1513      * Register all the drivers.  Note that order is important, the driver
1514      * registered last will get probed first.
1515      */
1516     bdrv_register(&bdrv_raw);
1517     bdrv_register(&bdrv_host_device);
1518 #ifdef __linux__
1519     bdrv_register(&bdrv_host_floppy);
1520     bdrv_register(&bdrv_host_cdrom);
1521 #endif
1522 #ifdef __FreeBSD__
1523     bdrv_register(&bdrv_host_cdrom);
1524 #endif
1525 }
1526
1527 block_init(bdrv_raw_init);