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