block/raw-posix.c

   1 /*
   2  * Block driver for RAW files (posix)
   3  *
   4  * Copyright (c) 2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #include "qemu-timer.h"
  26 #include "qemu-char.h"
  27 #include "qemu-log.h"
  28 #include "block_int.h"
  29 #include "module.h"
  30 #include "posix-aio-compat.h"
  31
  32 #ifdef CONFIG_COCOA
  33 #include <paths.h>
  34 #include <sys/param.h>
  35 #include <IOKit/IOKitLib.h>
  36 #include <IOKit/IOBSD.h>
  37 #include <IOKit/storage/IOMediaBSDClient.h>
  38 #include <IOKit/storage/IOMedia.h>
  39 #include <IOKit/storage/IOCDMedia.h>
  40 //#include <IOKit/storage/IOCDTypes.h>
  41 #include <CoreFoundation/CoreFoundation.h>
  42 #endif
  43
  44 #ifdef __sun__
  45 #define _POSIX_PTHREAD_SEMANTICS 1
  46 #include <signal.h>
  47 #include <sys/dkio.h>
  48 #endif
  49 #ifdef __linux__
  50 #include <sys/ioctl.h>
  51 #include <linux/cdrom.h>
  52 #include <linux/fd.h>
  53 #endif
  54 #ifdef __FreeBSD__
  55 #include <signal.h>
  56 #include <sys/disk.h>
  57 #include <sys/cdio.h>
  58 #endif
  59
  60 #ifdef __OpenBSD__
  61 #include <sys/ioctl.h>
  62 #include <sys/disklabel.h>
  63 #include <sys/dkio.h>
  64 #endif
  65
  66 #ifdef __DragonFly__
  67 #include <sys/ioctl.h>
  68 #include <sys/diskslice.h>
  69 #endif
  70
  71 //#define DEBUG_FLOPPY
  72
  73 //#define DEBUG_BLOCK
  74 #if defined(DEBUG_BLOCK)
  75 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
  76     { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
  77 #else
  78 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
  79 #endif
  80
  81 /* OS X does not have O_DSYNC */
  82 #ifndef O_DSYNC
  83 #ifdef O_SYNC
  84 #define O_DSYNC O_SYNC
  85 #elif defined(O_FSYNC)
  86 #define O_DSYNC O_FSYNC
  87 #endif
  88 #endif
  89
  90 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
  91 #ifndef O_DIRECT
  92 #define O_DIRECT O_DSYNC
  93 #endif
  94
  95 #define FTYPE_FILE   0
  96 #define FTYPE_CD     1
  97 #define FTYPE_FD     2
  98
  99 #define ALIGNED_BUFFER_SIZE (32 * 512)
 100
 101 /* if the FD is not accessed during that time (in ms), we try to
 102    reopen it to see if the disk has been changed */
 103 #define FD_OPEN_TIMEOUT 1000
 104
 105 typedef struct BDRVRawState {
 106     int fd;
 107     int type;
 108     unsigned int lseek_err_cnt;
 109     int open_flags;
 110 #if defined(__linux__)
 111     /* linux floppy specific */
 112     int64_t fd_open_time;
 113     int64_t fd_error_time;
 114     int fd_got_error;
 115     int fd_media_changed;
 116 #endif
 117     uint8_t* aligned_buf;
 118 } BDRVRawState;
 119
 120 static int posix_aio_init(void);
 121
 122 static int fd_open(BlockDriverState *bs);
 123 static int64_t raw_getlength(BlockDriverState *bs);
 124
 125 #if defined(__FreeBSD__)
 126 static int cdrom_reopen(BlockDriverState *bs);
 127 #endif
 128
 129 static int raw_open_common(BlockDriverState *bs, const char *filename,
 130                            int bdrv_flags, int open_flags)
 131 {
 132     BDRVRawState *s = bs->opaque;
 133     int fd, ret;
 134
 135     posix_aio_init();
 136
 137     s->lseek_err_cnt = 0;
 138
 139     s->open_flags = open_flags | O_BINARY;
 140     s->open_flags &= ~O_ACCMODE;
 141     if ((bdrv_flags & BDRV_O_ACCESS) == BDRV_O_RDWR) {
 142         s->open_flags |= O_RDWR;
 143     } else {
 144         s->open_flags |= O_RDONLY;
 145         bs->read_only = 1;
 146     }
 147
 148     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 149      * and O_DIRECT for no caching. */
 150     if ((bdrv_flags & BDRV_O_NOCACHE))
 151         s->open_flags |= O_DIRECT;
 152     else if (!(bdrv_flags & BDRV_O_CACHE_WB))
 153         s->open_flags |= O_DSYNC;
 154
 155     s->fd = -1;
 156     fd = open(filename, s->open_flags, 0644);
 157     if (fd < 0) {
 158         ret = -errno;
 159         if (ret == -EROFS)
 160             ret = -EACCES;
 161         return ret;
 162     }
 163     s->fd = fd;
 164     s->aligned_buf = NULL;
 165     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 166         s->aligned_buf = qemu_blockalign(bs, ALIGNED_BUFFER_SIZE);
 167         if (s->aligned_buf == NULL) {
 168             ret = -errno;
 169             close(fd);
 170             return ret;
 171         }
 172     }
 173     return 0;
 174 }
 175
 176 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 177 {
 178     BDRVRawState *s = bs->opaque;
 179     int open_flags = 0;
 180
 181     s->type = FTYPE_FILE;
 182     if (flags & BDRV_O_CREAT)
 183         open_flags = O_CREAT | O_TRUNC;
 184
 185     return raw_open_common(bs, filename, flags, open_flags);
 186 }
 187
 188 /* XXX: use host sector size if necessary with:
 189 #ifdef DIOCGSECTORSIZE
 190         {
 191             unsigned int sectorsize = 512;
 192             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 193                 sectorsize > bufsize)
 194                 bufsize = sectorsize;
 195         }
 196 #endif
 197 #ifdef CONFIG_COCOA
 198         u_int32_t   blockSize = 512;
 199         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 200             bufsize = blockSize;
 201         }
 202 #endif
 203 */
 204
 205 /*
 206  * offset and count are in bytes, but must be multiples of 512 for files
 207  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 208  *
 209  * This function may be called without alignment if the caller ensures
 210  * that O_DIRECT is not in effect.
 211  */
 212 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 213                      uint8_t *buf, int count)
 214 {
 215     BDRVRawState *s = bs->opaque;
 216     int ret;
 217
 218     ret = fd_open(bs);
 219     if (ret < 0)
 220         return ret;
 221
 222     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 223         ++(s->lseek_err_cnt);
 224         if(s->lseek_err_cnt <= 10) {
 225             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 226                               "] lseek failed : %d = %s\n",
 227                               s->fd, bs->filename, offset, buf, count,
 228                               bs->total_sectors, errno, strerror(errno));
 229         }
 230         return -1;
 231     }
 232     s->lseek_err_cnt=0;
 233
 234     ret = read(s->fd, buf, count);
 235     if (ret == count)
 236         goto label__raw_read__success;
 237
 238     /* Allow reads beyond the end (needed for pwrite) */
 239     if ((ret == 0) && bs->growable) {
 240         int64_t size = raw_getlength(bs);
 241         if (offset >= size) {
 242             memset(buf, 0, count);
 243             ret = count;
 244             goto label__raw_read__success;
 245         }
 246     }
 247
 248     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 249                       "] read failed %d : %d = %s\n",
 250                       s->fd, bs->filename, offset, buf, count,
 251                       bs->total_sectors, ret, errno, strerror(errno));
 252
 253     /* Try harder for CDrom. */
 254     if (bs->type == BDRV_TYPE_CDROM) {
 255         lseek(s->fd, offset, SEEK_SET);
 256         ret = read(s->fd, buf, count);
 257         if (ret == count)
 258             goto label__raw_read__success;
 259         lseek(s->fd, offset, SEEK_SET);
 260         ret = read(s->fd, buf, count);
 261         if (ret == count)
 262             goto label__raw_read__success;
 263
 264         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 265                           "] retry read failed %d : %d = %s\n",
 266                           s->fd, bs->filename, offset, buf, count,
 267                           bs->total_sectors, ret, errno, strerror(errno));
 268     }
 269
 270 label__raw_read__success:
 271
 272     return  (ret < 0) ? -errno : ret;
 273 }
 274
 275 /*
 276  * offset and count are in bytes, but must be multiples of 512 for files
 277  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 278  *
 279  * This function may be called without alignment if the caller ensures
 280  * that O_DIRECT is not in effect.
 281  */
 282 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 283                       const uint8_t *buf, int count)
 284 {
 285     BDRVRawState *s = bs->opaque;
 286     int ret;
 287
 288     ret = fd_open(bs);
 289     if (ret < 0)
 290         return -errno;
 291
 292     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 293         ++(s->lseek_err_cnt);
 294         if(s->lseek_err_cnt) {
 295             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 296                               PRId64 "] lseek failed : %d = %s\n",
 297                               s->fd, bs->filename, offset, buf, count,
 298                               bs->total_sectors, errno, strerror(errno));
 299         }
 300         return -EIO;
 301     }
 302     s->lseek_err_cnt = 0;
 303
 304     ret = write(s->fd, buf, count);
 305     if (ret == count)
 306         goto label__raw_write__success;
 307
 308     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 309                       "] write failed %d : %d = %s\n",
 310                       s->fd, bs->filename, offset, buf, count,
 311                       bs->total_sectors, ret, errno, strerror(errno));
 312
 313 label__raw_write__success:
 314
 315     return  (ret < 0) ? -errno : ret;
 316 }
 317
 318
 319 /*
 320  * offset and count are in bytes and possibly not aligned. For files opened
 321  * with O_DIRECT, necessary alignments are ensured before calling
 322  * raw_pread_aligned to do the actual read.
 323  */
 324 static int raw_pread(BlockDriverState *bs, int64_t offset,
 325                      uint8_t *buf, int count)
 326 {
 327     BDRVRawState *s = bs->opaque;
 328     int size, ret, shift, sum;
 329
 330     sum = 0;
 331
 332     if (s->aligned_buf != NULL)  {
 333
 334         if (offset & 0x1ff) {
 335             /* align offset on a 512 bytes boundary */
 336
 337             shift = offset & 0x1ff;
 338             size = (shift + count + 0x1ff) & ~0x1ff;
 339             if (size > ALIGNED_BUFFER_SIZE)
 340                 size = ALIGNED_BUFFER_SIZE;
 341             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 342             if (ret < 0)
 343                 return ret;
 344
 345             size = 512 - shift;
 346             if (size > count)
 347                 size = count;
 348             memcpy(buf, s->aligned_buf + shift, size);
 349
 350             buf += size;
 351             offset += size;
 352             count -= size;
 353             sum += size;
 354
 355             if (count == 0)
 356                 return sum;
 357         }
 358         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 359
 360             /* read on aligned buffer */
 361
 362             while (count) {
 363
 364                 size = (count + 0x1ff) & ~0x1ff;
 365                 if (size > ALIGNED_BUFFER_SIZE)
 366                     size = ALIGNED_BUFFER_SIZE;
 367
 368                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 369                 if (ret < 0)
 370                     return ret;
 371
 372                 size = ret;
 373                 if (size > count)
 374                     size = count;
 375
 376                 memcpy(buf, s->aligned_buf, size);
 377
 378                 buf += size;
 379                 offset += size;
 380                 count -= size;
 381                 sum += size;
 382             }
 383
 384             return sum;
 385         }
 386     }
 387
 388     return raw_pread_aligned(bs, offset, buf, count) + sum;
 389 }
 390
 391 static int raw_read(BlockDriverState *bs, int64_t sector_num,
 392                     uint8_t *buf, int nb_sectors)
 393 {
 394     int ret;
 395
 396     ret = raw_pread(bs, sector_num * 512, buf, nb_sectors * 512);
 397     if (ret == (nb_sectors * 512))
 398         ret = 0;
 399     return ret;
 400 }
 401
 402 /*
 403  * offset and count are in bytes and possibly not aligned. For files opened
 404  * with O_DIRECT, necessary alignments are ensured before calling
 405  * raw_pwrite_aligned to do the actual write.
 406  */
 407 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 408                       const uint8_t *buf, int count)
 409 {
 410     BDRVRawState *s = bs->opaque;
 411     int size, ret, shift, sum;
 412
 413     sum = 0;
 414
 415     if (s->aligned_buf != NULL) {
 416
 417         if (offset & 0x1ff) {
 418             /* align offset on a 512 bytes boundary */
 419             shift = offset & 0x1ff;
 420             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 421             if (ret < 0)
 422                 return ret;
 423
 424             size = 512 - shift;
 425             if (size > count)
 426                 size = count;
 427             memcpy(s->aligned_buf + shift, buf, size);
 428
 429             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 430             if (ret < 0)
 431                 return ret;
 432
 433             buf += size;
 434             offset += size;
 435             count -= size;
 436             sum += size;
 437
 438             if (count == 0)
 439                 return sum;
 440         }
 441         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 442
 443             while ((size = (count & ~0x1ff)) != 0) {
 444
 445                 if (size > ALIGNED_BUFFER_SIZE)
 446                     size = ALIGNED_BUFFER_SIZE;
 447
 448                 memcpy(s->aligned_buf, buf, size);
 449
 450                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 451                 if (ret < 0)
 452                     return ret;
 453
 454                 buf += ret;
 455                 offset += ret;
 456                 count -= ret;
 457                 sum += ret;
 458             }
 459             /* here, count < 512 because (count & ~0x1ff) == 0 */
 460             if (count) {
 461                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 462                 if (ret < 0)
 463                     return ret;
 464                  memcpy(s->aligned_buf, buf, count);
 465
 466                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 467                  if (ret < 0)
 468                      return ret;
 469                  if (count < ret)
 470                      ret = count;
 471
 472                  sum += ret;
 473             }
 474             return sum;
 475         }
 476     }
 477     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 478 }
 479
 480 static int raw_write(BlockDriverState *bs, int64_t sector_num,
 481                      const uint8_t *buf, int nb_sectors)
 482 {
 483     int ret;
 484     ret = raw_pwrite(bs, sector_num * 512, buf, nb_sectors * 512);
 485     if (ret == (nb_sectors * 512))
 486         ret = 0;
 487     return ret;
 488 }
 489
 490 /***********************************************************/
 491 /* Unix AIO using POSIX AIO */
 492
 493 typedef struct RawAIOCB {
 494     BlockDriverAIOCB common;
 495     struct qemu_paiocb aiocb;
 496     struct RawAIOCB *next;
 497     int ret;
 498 } RawAIOCB;
 499
 500 typedef struct PosixAioState
 501 {
 502     int rfd, wfd;
 503     RawAIOCB *first_aio;
 504 } PosixAioState;
 505
 506 static void posix_aio_read(void *opaque)
 507 {
 508     PosixAioState *s = opaque;
 509     RawAIOCB *acb, **pacb;
 510     int ret;
 511     ssize_t len;
 512
 513     /* read all bytes from signal pipe */
 514     for (;;) {
 515         char bytes[16];
 516
 517         len = read(s->rfd, bytes, sizeof(bytes));
 518         if (len == -1 && errno == EINTR)
 519             continue; /* try again */
 520         if (len == sizeof(bytes))
 521             continue; /* more to read */
 522         break;
 523     }
 524
 525     for(;;) {
 526         pacb = &s->first_aio;
 527         for(;;) {
 528             acb = *pacb;
 529             if (!acb)
 530                 goto the_end;
 531             ret = qemu_paio_error(&acb->aiocb);
 532             if (ret == ECANCELED) {
 533                 /* remove the request */
 534                 *pacb = acb->next;
 535                 qemu_aio_release(acb);
 536             } else if (ret != EINPROGRESS) {
 537                 /* end of aio */
 538                 if (ret == 0) {
 539                     ret = qemu_paio_return(&acb->aiocb);
 540                     if (ret == acb->aiocb.aio_nbytes)
 541                         ret = 0;
 542                     else
 543                         ret = -EINVAL;
 544                 } else {
 545                     ret = -ret;
 546                 }
 547                 /* remove the request */
 548                 *pacb = acb->next;
 549                 /* call the callback */
 550                 acb->common.cb(acb->common.opaque, ret);
 551                 qemu_aio_release(acb);
 552                 break;
 553             } else {
 554                 pacb = &acb->next;
 555             }
 556         }
 557     }
 558  the_end: ;
 559 }
 560
 561 static int posix_aio_flush(void *opaque)
 562 {
 563     PosixAioState *s = opaque;
 564     return !!s->first_aio;
 565 }
 566
 567 static PosixAioState *posix_aio_state;
 568
 569 static void aio_signal_handler(int signum)
 570 {
 571     if (posix_aio_state) {
 572         char byte = 0;
 573
 574         write(posix_aio_state->wfd, &byte, sizeof(byte));
 575     }
 576
 577     qemu_service_io();
 578 }
 579
 580 static int posix_aio_init(void)
 581 {
 582     struct sigaction act;
 583     PosixAioState *s;
 584     int fds[2];
 585     struct qemu_paioinit ai;
 586
 587     if (posix_aio_state)
 588         return 0;
 589
 590     s = qemu_malloc(sizeof(PosixAioState));
 591
 592     sigfillset(&act.sa_mask);
 593     act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
 594     act.sa_handler = aio_signal_handler;
 595     sigaction(SIGUSR2, &act, NULL);
 596
 597     s->first_aio = NULL;
 598     if (pipe(fds) == -1) {
 599         fprintf(stderr, "failed to create pipe\n");
 600         return -errno;
 601     }
 602
 603     s->rfd = fds[0];
 604     s->wfd = fds[1];
 605
 606     fcntl(s->rfd, F_SETFL, O_NONBLOCK);
 607     fcntl(s->wfd, F_SETFL, O_NONBLOCK);
 608
 609     qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
 610
 611     memset(&ai, 0, sizeof(ai));
 612     ai.aio_threads = 64;
 613     ai.aio_num = 64;
 614     qemu_paio_init(&ai);
 615
 616     posix_aio_state = s;
 617
 618     return 0;
 619 }
 620
 621 static void raw_aio_remove(RawAIOCB *acb)
 622 {
 623     RawAIOCB **pacb;
 624
 625     /* remove the callback from the queue */
 626     pacb = &posix_aio_state->first_aio;
 627     for(;;) {
 628         if (*pacb == NULL) {
 629             fprintf(stderr, "raw_aio_remove: aio request not found!\n");
 630             break;
 631         } else if (*pacb == acb) {
 632             *pacb = acb->next;
 633             qemu_aio_release(acb);
 634             break;
 635         }
 636         pacb = &(*pacb)->next;
 637     }
 638 }
 639
 640 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 641 {
 642     int ret;
 643     RawAIOCB *acb = (RawAIOCB *)blockacb;
 644
 645     ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 646     if (ret == QEMU_PAIO_NOTCANCELED) {
 647         /* fail safe: if the aio could not be canceled, we wait for
 648            it */
 649         while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
 650     }
 651
 652     raw_aio_remove(acb);
 653 }
 654
 655 static AIOPool raw_aio_pool = {
 656     .aiocb_size         = sizeof(RawAIOCB),
 657     .cancel             = raw_aio_cancel,
 658 };
 659
 660 static RawAIOCB *raw_aio_setup(BlockDriverState *bs, int64_t sector_num,
 661         QEMUIOVector *qiov, int nb_sectors,
 662         BlockDriverCompletionFunc *cb, void *opaque)
 663 {
 664     BDRVRawState *s = bs->opaque;
 665     RawAIOCB *acb;
 666
 667     if (fd_open(bs) < 0)
 668         return NULL;
 669
 670     acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque);
 671     if (!acb)
 672         return NULL;
 673     acb->aiocb.aio_fildes = s->fd;
 674     acb->aiocb.ev_signo = SIGUSR2;
 675     acb->aiocb.aio_iov = qiov->iov;
 676     acb->aiocb.aio_niov = qiov->niov;
 677     acb->aiocb.aio_nbytes = nb_sectors * 512;
 678     acb->aiocb.aio_offset = sector_num * 512;
 679     acb->aiocb.aio_flags = 0;
 680
 681     /*
 682      * If O_DIRECT is used the buffer needs to be aligned on a sector
 683      * boundary. Tell the low level code to ensure that in case it's
 684      * not done yet.
 685      */
 686     if (s->aligned_buf)
 687         acb->aiocb.aio_flags |= QEMU_AIO_SECTOR_ALIGNED;
 688
 689     acb->next = posix_aio_state->first_aio;
 690     posix_aio_state->first_aio = acb;
 691     return acb;
 692 }
 693
 694 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 695         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 696         BlockDriverCompletionFunc *cb, void *opaque)
 697 {
 698     RawAIOCB *acb;
 699
 700     acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
 701     if (!acb)
 702         return NULL;
 703     if (qemu_paio_read(&acb->aiocb) < 0) {
 704         raw_aio_remove(acb);
 705         return NULL;
 706     }
 707     return &acb->common;
 708 }
 709
 710 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 711         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 712         BlockDriverCompletionFunc *cb, void *opaque)
 713 {
 714     RawAIOCB *acb;
 715
 716     acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
 717     if (!acb)
 718         return NULL;
 719     if (qemu_paio_write(&acb->aiocb) < 0) {
 720         raw_aio_remove(acb);
 721         return NULL;
 722     }
 723     return &acb->common;
 724 }
 725
 726 static void raw_close(BlockDriverState *bs)
 727 {
 728     BDRVRawState *s = bs->opaque;
 729     if (s->fd >= 0) {
 730         close(s->fd);
 731         s->fd = -1;
 732         if (s->aligned_buf != NULL)
 733             qemu_free(s->aligned_buf);
 734     }
 735 }
 736
 737 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 738 {
 739     BDRVRawState *s = bs->opaque;
 740     if (s->type != FTYPE_FILE)
 741         return -ENOTSUP;
 742     if (ftruncate(s->fd, offset) < 0)
 743         return -errno;
 744     return 0;
 745 }
 746
 747 #ifdef __OpenBSD__
 748 static int64_t raw_getlength(BlockDriverState *bs)
 749 {
 750     BDRVRawState *s = bs->opaque;
 751     int fd = s->fd;
 752     struct stat st;
 753
 754     if (fstat(fd, &st))
 755         return -1;
 756     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 757         struct disklabel dl;
 758
 759         if (ioctl(fd, DIOCGDINFO, &dl))
 760             return -1;
 761         return (uint64_t)dl.d_secsize *
 762             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 763     } else
 764         return st.st_size;
 765 }
 766 #else /* !__OpenBSD__ */
 767 static int64_t  raw_getlength(BlockDriverState *bs)
 768 {
 769     BDRVRawState *s = bs->opaque;
 770     int fd = s->fd;
 771     int64_t size;
 772 #ifdef CONFIG_BSD
 773     struct stat sb;
 774 #ifdef __FreeBSD__
 775     int reopened = 0;
 776 #endif
 777 #endif
 778 #ifdef __sun__
 779     struct dk_minfo minfo;
 780     int rv;
 781 #endif
 782     int ret;
 783
 784     ret = fd_open(bs);
 785     if (ret < 0)
 786         return ret;
 787
 788 #ifdef CONFIG_BSD
 789 #ifdef __FreeBSD__
 790 again:
 791 #endif
 792     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 793 #ifdef DIOCGMEDIASIZE
 794         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 795 #elif defined(DIOCGPART)
 796         {
 797                 struct partinfo pi;
 798                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 799                         size = pi.media_size;
 800                 else
 801                         size = 0;
 802         }
 803         if (size == 0)
 804 #endif
 805 #ifdef CONFIG_COCOA
 806         size = LONG_LONG_MAX;
 807 #else
 808         size = lseek(fd, 0LL, SEEK_END);
 809 #endif
 810 #ifdef __FreeBSD__
 811         switch(s->type) {
 812         case FTYPE_CD:
 813             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
 814             if (size == 2048LL * (unsigned)-1)
 815                 size = 0;
 816             /* XXX no disc?  maybe we need to reopen... */
 817             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
 818                 reopened = 1;
 819                 goto again;
 820             }
 821         }
 822 #endif
 823     } else
 824 #endif
 825 #ifdef __sun__
 826     /*
 827      * use the DKIOCGMEDIAINFO ioctl to read the size.
 828      */
 829     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 830     if ( rv != -1 ) {
 831         size = minfo.dki_lbsize * minfo.dki_capacity;
 832     } else /* there are reports that lseek on some devices
 833               fails, but irc discussion said that contingency
 834               on contingency was overkill */
 835 #endif
 836     {
 837         size = lseek(fd, 0, SEEK_END);
 838     }
 839     return size;
 840 }
 841 #endif
 842
 843 static int raw_create(const char *filename, QEMUOptionParameter *options)
 844 {
 845     int fd;
 846     int result = 0;
 847     int64_t total_size = 0;
 848
 849     /* Read out options */
 850     while (options && options->name) {
 851         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 852             total_size = options->value.n / 512;
 853         }
 854         options++;
 855     }
 856
 857     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 858               0644);
 859     if (fd < 0) {
 860         result = -errno;
 861     } else {
 862         if (ftruncate(fd, total_size * 512) != 0) {
 863             result = -errno;
 864         }
 865         if (close(fd) != 0) {
 866             result = -errno;
 867         }
 868     }
 869     return result;
 870 }
 871
 872 static void raw_flush(BlockDriverState *bs)
 873 {
 874     BDRVRawState *s = bs->opaque;
 875     fsync(s->fd);
 876 }
 877
 878
 879 static QEMUOptionParameter raw_create_options[] = {
 880     {
 881         .name = BLOCK_OPT_SIZE,
 882         .type = OPT_SIZE,
 883         .help = "Virtual disk size"
 884     },
 885     { NULL }
 886 };
 887
 888 static BlockDriver bdrv_raw = {
 889     .format_name = "raw",
 890     .instance_size = sizeof(BDRVRawState),
 891     .bdrv_probe = NULL, /* no probe for protocols */
 892     .bdrv_open = raw_open,
 893     .bdrv_read = raw_read,
 894     .bdrv_write = raw_write,
 895     .bdrv_close = raw_close,
 896     .bdrv_create = raw_create,
 897     .bdrv_flush = raw_flush,
 898
 899     .bdrv_aio_readv = raw_aio_readv,
 900     .bdrv_aio_writev = raw_aio_writev,
 901
 902     .bdrv_truncate = raw_truncate,
 903     .bdrv_getlength = raw_getlength,
 904
 905     .create_options = raw_create_options,
 906 };
 907
 908 /***********************************************/
 909 /* host device */
 910
 911 #ifdef CONFIG_COCOA
 912 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 913 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 914
 915 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 916 {
 917     kern_return_t       kernResult;
 918     mach_port_t     masterPort;
 919     CFMutableDictionaryRef  classesToMatch;
 920
 921     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 922     if ( KERN_SUCCESS != kernResult ) {
 923         printf( "IOMasterPort returned %d\n", kernResult );
 924     }
 925
 926     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 927     if ( classesToMatch == NULL ) {
 928         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 929     } else {
 930     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 931     }
 932     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 933     if ( KERN_SUCCESS != kernResult )
 934     {
 935         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 936     }
 937
 938     return kernResult;
 939 }
 940
 941 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 942 {
 943     io_object_t     nextMedia;
 944     kern_return_t   kernResult = KERN_FAILURE;
 945     *bsdPath = '\0';
 946     nextMedia = IOIteratorNext( mediaIterator );
 947     if ( nextMedia )
 948     {
 949         CFTypeRef   bsdPathAsCFString;
 950     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 951         if ( bsdPathAsCFString ) {
 952             size_t devPathLength;
 953             strcpy( bsdPath, _PATH_DEV );
 954             strcat( bsdPath, "r" );
 955             devPathLength = strlen( bsdPath );
 956             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 957                 kernResult = KERN_SUCCESS;
 958             }
 959             CFRelease( bsdPathAsCFString );
 960         }
 961         IOObjectRelease( nextMedia );
 962     }
 963
 964     return kernResult;
 965 }
 966
 967 #endif
 968
 969 static int hdev_probe_device(const char *filename)
 970 {
 971     struct stat st;
 972
 973     /* allow a dedicated CD-ROM driver to match with a higher priority */
 974     if (strstart(filename, "/dev/cdrom", NULL))
 975         return 50;
 976
 977     if (stat(filename, &st) >= 0 &&
 978             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 979         return 100;
 980     }
 981
 982     return 0;
 983 }
 984
 985 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 986 {
 987     BDRVRawState *s = bs->opaque;
 988
 989 #ifdef CONFIG_COCOA
 990     if (strstart(filename, "/dev/cdrom", NULL)) {
 991         kern_return_t kernResult;
 992         io_iterator_t mediaIterator;
 993         char bsdPath[ MAXPATHLEN ];
 994         int fd;
 995
 996         kernResult = FindEjectableCDMedia( &mediaIterator );
 997         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 998
 999         if ( bsdPath[ 0 ] != '\0' ) {
1000             strcat(bsdPath,"s0");
1001             /* some CDs don't have a partition 0 */
1002             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1003             if (fd < 0) {
1004                 bsdPath[strlen(bsdPath)-1] = '1';
1005             } else {
1006                 close(fd);
1007             }
1008             filename = bsdPath;
1009         }
1010
1011         if ( mediaIterator )
1012             IOObjectRelease( mediaIterator );
1013     }
1014 #endif
1015
1016     s->type = FTYPE_FILE;
1017 #if defined(__linux__)
1018     if (strstart(filename, "/dev/sg", NULL)) {
1019         bs->sg = 1;
1020     }
1021 #endif
1022
1023     return raw_open_common(bs, filename, flags, 0);
1024 }
1025
1026 #if defined(__linux__)
1027 /* Note: we do not have a reliable method to detect if the floppy is
1028    present. The current method is to try to open the floppy at every
1029    I/O and to keep it opened during a few hundreds of ms. */
1030 static int fd_open(BlockDriverState *bs)
1031 {
1032     BDRVRawState *s = bs->opaque;
1033     int last_media_present;
1034
1035     if (s->type != FTYPE_FD)
1036         return 0;
1037     last_media_present = (s->fd >= 0);
1038     if (s->fd >= 0 &&
1039         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1040         close(s->fd);
1041         s->fd = -1;
1042 #ifdef DEBUG_FLOPPY
1043         printf("Floppy closed\n");
1044 #endif
1045     }
1046     if (s->fd < 0) {
1047         if (s->fd_got_error &&
1048             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1049 #ifdef DEBUG_FLOPPY
1050             printf("No floppy (open delayed)\n");
1051 #endif
1052             return -EIO;
1053         }
1054         s->fd = open(bs->filename, s->open_flags & ~O_NONBLOCK);
1055         if (s->fd < 0) {
1056             s->fd_error_time = qemu_get_clock(rt_clock);
1057             s->fd_got_error = 1;
1058             if (last_media_present)
1059                 s->fd_media_changed = 1;
1060 #ifdef DEBUG_FLOPPY
1061             printf("No floppy\n");
1062 #endif
1063             return -EIO;
1064         }
1065 #ifdef DEBUG_FLOPPY
1066         printf("Floppy opened\n");
1067 #endif
1068     }
1069     if (!last_media_present)
1070         s->fd_media_changed = 1;
1071     s->fd_open_time = qemu_get_clock(rt_clock);
1072     s->fd_got_error = 0;
1073     return 0;
1074 }
1075
1076 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1077 {
1078     BDRVRawState *s = bs->opaque;
1079
1080     return ioctl(s->fd, req, buf);
1081 }
1082
1083 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1084         unsigned long int req, void *buf,
1085         BlockDriverCompletionFunc *cb, void *opaque)
1086 {
1087     BDRVRawState *s = bs->opaque;
1088     RawAIOCB *acb;
1089
1090     if (fd_open(bs) < 0)
1091         return NULL;
1092
1093     acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque);
1094     if (!acb)
1095         return NULL;
1096     acb->aiocb.aio_fildes = s->fd;
1097     acb->aiocb.ev_signo = SIGUSR2;
1098     acb->aiocb.aio_offset = 0;
1099     acb->aiocb.aio_flags = 0;
1100
1101     acb->next = posix_aio_state->first_aio;
1102     posix_aio_state->first_aio = acb;
1103
1104     acb->aiocb.aio_ioctl_buf = buf;
1105     acb->aiocb.aio_ioctl_cmd = req;
1106     if (qemu_paio_ioctl(&acb->aiocb) < 0) {
1107         raw_aio_remove(acb);
1108         return NULL;
1109     }
1110
1111     return &acb->common;
1112 }
1113
1114 #elif defined(__FreeBSD__)
1115 static int fd_open(BlockDriverState *bs)
1116 {
1117     BDRVRawState *s = bs->opaque;
1118
1119     /* this is just to ensure s->fd is sane (its called by io ops) */
1120     if (s->fd >= 0)
1121         return 0;
1122     return -EIO;
1123 }
1124 #else /* !linux && !FreeBSD */
1125
1126 static int fd_open(BlockDriverState *bs)
1127 {
1128     return 0;
1129 }
1130
1131 #endif /* !linux && !FreeBSD */
1132
1133 static int hdev_create(const char *filename, QEMUOptionParameter *options)
1134 {
1135     int fd;
1136     int ret = 0;
1137     struct stat stat_buf;
1138     int64_t total_size = 0;
1139
1140     /* Read out options */
1141     while (options && options->name) {
1142         if (!strcmp(options->name, "size")) {
1143             total_size = options->value.n / 512;
1144         }
1145         options++;
1146     }
1147
1148     fd = open(filename, O_WRONLY | O_BINARY);
1149     if (fd < 0)
1150         return -EIO;
1151
1152     if (fstat(fd, &stat_buf) < 0)
1153         ret = -EIO;
1154     else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
1155         ret = -EIO;
1156     else if (lseek(fd, 0, SEEK_END) < total_size * 512)
1157         ret = -ENOSPC;
1158
1159     close(fd);
1160     return ret;
1161 }
1162
1163 static BlockDriver bdrv_host_device = {
1164     .format_name        = "host_device",
1165     .instance_size      = sizeof(BDRVRawState),
1166     .bdrv_probe_device  = hdev_probe_device,
1167     .bdrv_open          = hdev_open,
1168     .bdrv_close         = raw_close,
1169     .bdrv_create        = hdev_create,
1170     .bdrv_flush         = raw_flush,
1171
1172     .bdrv_aio_readv     = raw_aio_readv,
1173     .bdrv_aio_writev    = raw_aio_writev,
1174
1175     .bdrv_read          = raw_read,
1176     .bdrv_write         = raw_write,
1177     .bdrv_getlength     = raw_getlength,
1178
1179     /* generic scsi device */
1180 #ifdef __linux__
1181     .bdrv_ioctl         = hdev_ioctl,
1182     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1183 #endif
1184 };
1185
1186 #ifdef __linux__
1187 static int floppy_open(BlockDriverState *bs, const char *filename, int flags)
1188 {
1189     BDRVRawState *s = bs->opaque;
1190     int ret;
1191
1192     posix_aio_init();
1193
1194     s->type = FTYPE_FD;
1195
1196     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1197     ret = raw_open_common(bs, filename, flags, O_NONBLOCK);
1198     if (ret)
1199         return ret;
1200
1201     /* close fd so that we can reopen it as needed */
1202     close(s->fd);
1203     s->fd = -1;
1204     s->fd_media_changed = 1;
1205
1206     return 0;
1207 }
1208
1209 static int floppy_probe_device(const char *filename)
1210 {
1211     if (strstart(filename, "/dev/fd", NULL))
1212         return 100;
1213     return 0;
1214 }
1215
1216
1217 static int floppy_is_inserted(BlockDriverState *bs)
1218 {
1219     return fd_open(bs) >= 0;
1220 }
1221
1222 static int floppy_media_changed(BlockDriverState *bs)
1223 {
1224     BDRVRawState *s = bs->opaque;
1225     int ret;
1226
1227     /*
1228      * XXX: we do not have a true media changed indication.
1229      * It does not work if the floppy is changed without trying to read it.
1230      */
1231     fd_open(bs);
1232     ret = s->fd_media_changed;
1233     s->fd_media_changed = 0;
1234 #ifdef DEBUG_FLOPPY
1235     printf("Floppy changed=%d\n", ret);
1236 #endif
1237     return ret;
1238 }
1239
1240 static int floppy_eject(BlockDriverState *bs, int eject_flag)
1241 {
1242     BDRVRawState *s = bs->opaque;
1243     int fd;
1244
1245     if (s->fd >= 0) {
1246         close(s->fd);
1247         s->fd = -1;
1248     }
1249     fd = open(bs->filename, s->open_flags | O_NONBLOCK);
1250     if (fd >= 0) {
1251         if (ioctl(fd, FDEJECT, 0) < 0)
1252             perror("FDEJECT");
1253         close(fd);
1254     }
1255
1256     return 0;
1257 }
1258
1259 static BlockDriver bdrv_host_floppy = {
1260     .format_name        = "host_floppy",
1261     .instance_size      = sizeof(BDRVRawState),
1262     .bdrv_probe_device  = floppy_probe_device,
1263     .bdrv_open          = floppy_open,
1264     .bdrv_close         = raw_close,
1265     .bdrv_create        = hdev_create,
1266     .bdrv_flush         = raw_flush,
1267
1268     .bdrv_aio_readv     = raw_aio_readv,
1269     .bdrv_aio_writev    = raw_aio_writev,
1270
1271     .bdrv_read          = raw_read,
1272     .bdrv_write         = raw_write,
1273     .bdrv_getlength     = raw_getlength,
1274
1275     /* removable device support */
1276     .bdrv_is_inserted   = floppy_is_inserted,
1277     .bdrv_media_changed = floppy_media_changed,
1278     .bdrv_eject         = floppy_eject,
1279 };
1280
1281 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1282 {
1283     BDRVRawState *s = bs->opaque;
1284
1285     s->type = FTYPE_CD;
1286
1287     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1288     return raw_open_common(bs, filename, flags, O_NONBLOCK);
1289 }
1290
1291 static int cdrom_probe_device(const char *filename)
1292 {
1293     if (strstart(filename, "/dev/cd", NULL))
1294         return 100;
1295     return 0;
1296 }
1297
1298 static int cdrom_is_inserted(BlockDriverState *bs)
1299 {
1300     BDRVRawState *s = bs->opaque;
1301     int ret;
1302
1303     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1304     if (ret == CDS_DISC_OK)
1305         return 1;
1306     return 0;
1307 }
1308
1309 static int cdrom_eject(BlockDriverState *bs, int eject_flag)
1310 {
1311     BDRVRawState *s = bs->opaque;
1312
1313     if (eject_flag) {
1314         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1315             perror("CDROMEJECT");
1316     } else {
1317         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1318             perror("CDROMEJECT");
1319     }
1320
1321     return 0;
1322 }
1323
1324 static int cdrom_set_locked(BlockDriverState *bs, int locked)
1325 {
1326     BDRVRawState *s = bs->opaque;
1327
1328     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1329         /*
1330          * Note: an error can happen if the distribution automatically
1331          * mounts the CD-ROM
1332          */
1333         /* perror("CDROM_LOCKDOOR"); */
1334     }
1335
1336     return 0;
1337 }
1338
1339 static BlockDriver bdrv_host_cdrom = {
1340     .format_name        = "host_cdrom",
1341     .instance_size      = sizeof(BDRVRawState),
1342     .bdrv_probe_device  = cdrom_probe_device,
1343     .bdrv_open          = cdrom_open,
1344     .bdrv_close         = raw_close,
1345     .bdrv_create        = hdev_create,
1346     .bdrv_flush         = raw_flush,
1347
1348     .bdrv_aio_readv     = raw_aio_readv,
1349     .bdrv_aio_writev    = raw_aio_writev,
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     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1363 };
1364 #endif /* __linux__ */
1365
1366 #ifdef __FreeBSD__
1367 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1368 {
1369     BDRVRawState *s = bs->opaque;
1370     int ret;
1371
1372     s->type = FTYPE_CD;
1373
1374     ret = raw_open_common(bs, filename, flags, 0);
1375     if (ret)
1376         return ret;
1377
1378     /* make sure the door isnt locked at this time */
1379     ioctl(s->fd, CDIOCALLOW);
1380     return 0;
1381 }
1382
1383 static int cdrom_probe_device(const char *filename)
1384 {
1385     if (strstart(filename, "/dev/cd", NULL) ||
1386             strstart(filename, "/dev/acd", NULL))
1387         return 100;
1388     return 0;
1389 }
1390
1391 static int cdrom_reopen(BlockDriverState *bs)
1392 {
1393     BDRVRawState *s = bs->opaque;
1394     int fd;
1395
1396     /*
1397      * Force reread of possibly changed/newly loaded disc,
1398      * FreeBSD seems to not notice sometimes...
1399      */
1400     if (s->fd >= 0)
1401         close(s->fd);
1402     fd = open(bs->filename, s->open_flags, 0644);
1403     if (fd < 0) {
1404         s->fd = -1;
1405         return -EIO;
1406     }
1407     s->fd = fd;
1408
1409     /* make sure the door isnt locked at this time */
1410     ioctl(s->fd, CDIOCALLOW);
1411     return 0;
1412 }
1413
1414 static int cdrom_is_inserted(BlockDriverState *bs)
1415 {
1416     return raw_getlength(bs) > 0;
1417 }
1418
1419 static int cdrom_eject(BlockDriverState *bs, int eject_flag)
1420 {
1421     BDRVRawState *s = bs->opaque;
1422
1423     if (s->fd < 0)
1424         return -ENOTSUP;
1425
1426     (void) ioctl(s->fd, CDIOCALLOW);
1427
1428     if (eject_flag) {
1429         if (ioctl(s->fd, CDIOCEJECT) < 0)
1430             perror("CDIOCEJECT");
1431     } else {
1432         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1433             perror("CDIOCCLOSE");
1434     }
1435
1436     if (cdrom_reopen(bs) < 0)
1437         return -ENOTSUP;
1438     return 0;
1439 }
1440
1441 static int cdrom_set_locked(BlockDriverState *bs, int locked)
1442 {
1443     BDRVRawState *s = bs->opaque;
1444
1445     if (s->fd < 0)
1446         return -ENOTSUP;
1447     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1448         /*
1449          * Note: an error can happen if the distribution automatically
1450          * mounts the CD-ROM
1451          */
1452         /* perror("CDROM_LOCKDOOR"); */
1453     }
1454
1455     return 0;
1456 }
1457
1458 static BlockDriver bdrv_host_cdrom = {
1459     .format_name        = "host_cdrom",
1460     .instance_size      = sizeof(BDRVRawState),
1461     .bdrv_probe_device  = cdrom_probe_device,
1462     .bdrv_open          = cdrom_open,
1463     .bdrv_close         = raw_close,
1464     .bdrv_create        = hdev_create,
1465     .bdrv_flush         = raw_flush,
1466
1467     .bdrv_aio_readv     = raw_aio_readv,
1468     .bdrv_aio_writev    = raw_aio_writev,
1469
1470     .bdrv_read          = raw_read,
1471     .bdrv_write         = raw_write,
1472     .bdrv_getlength     = raw_getlength,
1473
1474     /* removable device support */
1475     .bdrv_is_inserted   = cdrom_is_inserted,
1476     .bdrv_eject         = cdrom_eject,
1477     .bdrv_set_locked    = cdrom_set_locked,
1478 };
1479 #endif /* __FreeBSD__ */
1480
1481 static void bdrv_raw_init(void)
1482 {
1483     /*
1484      * Register all the drivers.  Note that order is important, the driver
1485      * registered last will get probed first.
1486      */
1487     bdrv_register(&bdrv_raw);
1488     bdrv_register(&bdrv_host_device);
1489 #ifdef __linux__
1490     bdrv_register(&bdrv_host_floppy);
1491     bdrv_register(&bdrv_host_cdrom);
1492 #endif
1493 #ifdef __FreeBSD__
1494     bdrv_register(&bdrv_host_cdrom);
1495 #endif
1496 }
1497
1498 block_init(bdrv_raw_init);