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