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