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 "compatfd.h"
  29 #include <assert.h>
  30 #ifdef CONFIG_AIO
  31 #include "posix-aio-compat.h"
  32 #endif
  33
  34 #ifdef CONFIG_COCOA
  35 #include <paths.h>
  36 #include <sys/param.h>
  37 #include <IOKit/IOKitLib.h>
  38 #include <IOKit/IOBSD.h>
  39 #include <IOKit/storage/IOMediaBSDClient.h>
  40 #include <IOKit/storage/IOMedia.h>
  41 #include <IOKit/storage/IOCDMedia.h>
  42 //#include <IOKit/storage/IOCDTypes.h>
  43 #include <CoreFoundation/CoreFoundation.h>
  44 #endif
  45
  46 #ifdef __sun__
  47 #define _POSIX_PTHREAD_SEMANTICS 1
  48 #include <signal.h>
  49 #include <sys/dkio.h>
  50 #endif
  51 #ifdef __linux__
  52 #include <sys/ioctl.h>
  53 #include <linux/cdrom.h>
  54 #include <linux/fd.h>
  55 #endif
  56 #ifdef __FreeBSD__
  57 #include <signal.h>
  58 #include <sys/disk.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 //#define DEBUG_FLOPPY
  68
  69 //#define DEBUG_BLOCK
  70 #if defined(DEBUG_BLOCK)
  71 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (qemu_log_enabled())     \
  72     { qemu_log(formatCstr, ##args); qemu_log_flush(); } } while (0)
  73 #else
  74 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  75 #endif
  76
  77 /* OS X does not have O_DSYNC */
  78 #ifndef O_DSYNC
  79 #define O_DSYNC O_SYNC
  80 #endif
  81
  82 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
  83 #ifndef O_DIRECT
  84 #define O_DIRECT O_DSYNC
  85 #endif
  86
  87 #define FTYPE_FILE   0
  88 #define FTYPE_CD     1
  89 #define FTYPE_FD     2
  90
  91 #define ALIGNED_BUFFER_SIZE (32 * 512)
  92
  93 /* if the FD is not accessed during that time (in ms), we try to
  94    reopen it to see if the disk has been changed */
  95 #define FD_OPEN_TIMEOUT 1000
  96
  97 typedef struct BDRVRawState {
  98     int fd;
  99     int type;
 100     unsigned int lseek_err_cnt;
 101 #if defined(__linux__)
 102     /* linux floppy specific */
 103     int fd_open_flags;
 104     int64_t fd_open_time;
 105     int64_t fd_error_time;
 106     int fd_got_error;
 107     int fd_media_changed;
 108 #endif
 109     uint8_t* aligned_buf;
 110 } BDRVRawState;
 111
 112 static int posix_aio_init(void);
 113
 114 static int fd_open(BlockDriverState *bs);
 115
 116 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 117 {
 118     BDRVRawState *s = bs->opaque;
 119     int fd, open_flags, ret;
 120
 121     posix_aio_init();
 122
 123     s->lseek_err_cnt = 0;
 124
 125     open_flags = O_BINARY;
 126     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 127         open_flags |= O_RDWR;
 128     } else {
 129         open_flags |= O_RDONLY;
 130         bs->read_only = 1;
 131     }
 132     if (flags & BDRV_O_CREAT)
 133         open_flags |= O_CREAT | O_TRUNC;
 134
 135     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 136      * and O_DIRECT for no caching. */
 137     if ((flags & BDRV_O_NOCACHE))
 138         open_flags |= O_DIRECT;
 139     else if (!(flags & BDRV_O_CACHE_WB))
 140         open_flags |= O_DSYNC;
 141
 142     s->type = FTYPE_FILE;
 143
 144     fd = open(filename, open_flags, 0644);
 145     if (fd < 0) {
 146         ret = -errno;
 147         if (ret == -EROFS)
 148             ret = -EACCES;
 149         return ret;
 150     }
 151     s->fd = fd;
 152     s->aligned_buf = NULL;
 153     if ((flags & BDRV_O_NOCACHE)) {
 154         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 155         if (s->aligned_buf == NULL) {
 156             ret = -errno;
 157             close(fd);
 158             return ret;
 159         }
 160     }
 161     return 0;
 162 }
 163
 164 /* XXX: use host sector size if necessary with:
 165 #ifdef DIOCGSECTORSIZE
 166         {
 167             unsigned int sectorsize = 512;
 168             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 169                 sectorsize > bufsize)
 170                 bufsize = sectorsize;
 171         }
 172 #endif
 173 #ifdef CONFIG_COCOA
 174         u_int32_t   blockSize = 512;
 175         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 176             bufsize = blockSize;
 177         }
 178 #endif
 179 */
 180
 181 /*
 182  * offset and count are in bytes, but must be multiples of 512 for files
 183  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 184  *
 185  * This function may be called without alignment if the caller ensures
 186  * that O_DIRECT is not in effect.
 187  */
 188 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 189                      uint8_t *buf, int count)
 190 {
 191     BDRVRawState *s = bs->opaque;
 192     int ret;
 193
 194     ret = fd_open(bs);
 195     if (ret < 0)
 196         return ret;
 197
 198     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 199         ++(s->lseek_err_cnt);
 200         if(s->lseek_err_cnt <= 10) {
 201             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 202                               "] lseek failed : %d = %s\n",
 203                               s->fd, bs->filename, offset, buf, count,
 204                               bs->total_sectors, errno, strerror(errno));
 205         }
 206         return -1;
 207     }
 208     s->lseek_err_cnt=0;
 209
 210     ret = read(s->fd, buf, count);
 211     if (ret == count)
 212         goto label__raw_read__success;
 213
 214     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 215                       "] read failed %d : %d = %s\n",
 216                       s->fd, bs->filename, offset, buf, count,
 217                       bs->total_sectors, ret, errno, strerror(errno));
 218
 219     /* Try harder for CDrom. */
 220     if (bs->type == BDRV_TYPE_CDROM) {
 221         lseek(s->fd, offset, SEEK_SET);
 222         ret = read(s->fd, buf, count);
 223         if (ret == count)
 224             goto label__raw_read__success;
 225         lseek(s->fd, offset, SEEK_SET);
 226         ret = read(s->fd, buf, count);
 227         if (ret == count)
 228             goto label__raw_read__success;
 229
 230         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 231                           "] retry read failed %d : %d = %s\n",
 232                           s->fd, bs->filename, offset, buf, count,
 233                           bs->total_sectors, ret, errno, strerror(errno));
 234     }
 235
 236 label__raw_read__success:
 237
 238     return ret;
 239 }
 240
 241 /*
 242  * offset and count are in bytes, but must be multiples of 512 for files
 243  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 244  *
 245  * This function may be called without alignment if the caller ensures
 246  * that O_DIRECT is not in effect.
 247  */
 248 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 249                       const uint8_t *buf, int count)
 250 {
 251     BDRVRawState *s = bs->opaque;
 252     int ret;
 253
 254     ret = fd_open(bs);
 255     if (ret < 0)
 256         return -errno;
 257
 258     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 259         ++(s->lseek_err_cnt);
 260         if(s->lseek_err_cnt) {
 261             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 262                               PRId64 "] lseek failed : %d = %s\n",
 263                               s->fd, bs->filename, offset, buf, count,
 264                               bs->total_sectors, errno, strerror(errno));
 265         }
 266         return -EIO;
 267     }
 268     s->lseek_err_cnt = 0;
 269
 270     ret = write(s->fd, buf, count);
 271     if (ret == count)
 272         goto label__raw_write__success;
 273
 274     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 275                       "] write failed %d : %d = %s\n",
 276                       s->fd, bs->filename, offset, buf, count,
 277                       bs->total_sectors, ret, errno, strerror(errno));
 278
 279 label__raw_write__success:
 280
 281     return  (ret < 0) ? -errno : ret;
 282 }
 283
 284
 285 /*
 286  * offset and count are in bytes and possibly not aligned. For files opened
 287  * with O_DIRECT, necessary alignments are ensured before calling
 288  * raw_pread_aligned to do the actual read.
 289  */
 290 static int raw_pread(BlockDriverState *bs, int64_t offset,
 291                      uint8_t *buf, int count)
 292 {
 293     BDRVRawState *s = bs->opaque;
 294     int size, ret, shift, sum;
 295
 296     sum = 0;
 297
 298     if (s->aligned_buf != NULL)  {
 299
 300         if (offset & 0x1ff) {
 301             /* align offset on a 512 bytes boundary */
 302
 303             shift = offset & 0x1ff;
 304             size = (shift + count + 0x1ff) & ~0x1ff;
 305             if (size > ALIGNED_BUFFER_SIZE)
 306                 size = ALIGNED_BUFFER_SIZE;
 307             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 308             if (ret < 0)
 309                 return ret;
 310
 311             size = 512 - shift;
 312             if (size > count)
 313                 size = count;
 314             memcpy(buf, s->aligned_buf + shift, size);
 315
 316             buf += size;
 317             offset += size;
 318             count -= size;
 319             sum += size;
 320
 321             if (count == 0)
 322                 return sum;
 323         }
 324         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 325
 326             /* read on aligned buffer */
 327
 328             while (count) {
 329
 330                 size = (count + 0x1ff) & ~0x1ff;
 331                 if (size > ALIGNED_BUFFER_SIZE)
 332                     size = ALIGNED_BUFFER_SIZE;
 333
 334                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 335                 if (ret < 0)
 336                     return ret;
 337
 338                 size = ret;
 339                 if (size > count)
 340                     size = count;
 341
 342                 memcpy(buf, s->aligned_buf, size);
 343
 344                 buf += size;
 345                 offset += size;
 346                 count -= size;
 347                 sum += size;
 348             }
 349
 350             return sum;
 351         }
 352     }
 353
 354     return raw_pread_aligned(bs, offset, buf, count) + sum;
 355 }
 356
 357 /*
 358  * offset and count are in bytes and possibly not aligned. For files opened
 359  * with O_DIRECT, necessary alignments are ensured before calling
 360  * raw_pwrite_aligned to do the actual write.
 361  */
 362 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 363                       const uint8_t *buf, int count)
 364 {
 365     BDRVRawState *s = bs->opaque;
 366     int size, ret, shift, sum;
 367
 368     sum = 0;
 369
 370     if (s->aligned_buf != NULL) {
 371
 372         if (offset & 0x1ff) {
 373             /* align offset on a 512 bytes boundary */
 374             shift = offset & 0x1ff;
 375             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 376             if (ret < 0)
 377                 return ret;
 378
 379             size = 512 - shift;
 380             if (size > count)
 381                 size = count;
 382             memcpy(s->aligned_buf + shift, buf, size);
 383
 384             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 385             if (ret < 0)
 386                 return ret;
 387
 388             buf += size;
 389             offset += size;
 390             count -= size;
 391             sum += size;
 392
 393             if (count == 0)
 394                 return sum;
 395         }
 396         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 397
 398             while ((size = (count & ~0x1ff)) != 0) {
 399
 400                 if (size > ALIGNED_BUFFER_SIZE)
 401                     size = ALIGNED_BUFFER_SIZE;
 402
 403                 memcpy(s->aligned_buf, buf, size);
 404
 405                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 406                 if (ret < 0)
 407                     return ret;
 408
 409                 buf += ret;
 410                 offset += ret;
 411                 count -= ret;
 412                 sum += ret;
 413             }
 414             /* here, count < 512 because (count & ~0x1ff) == 0 */
 415             if (count) {
 416                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 417                 if (ret < 0)
 418                     return ret;
 419                  memcpy(s->aligned_buf, buf, count);
 420
 421                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 422                  if (ret < 0)
 423                      return ret;
 424                  if (count < ret)
 425                      ret = count;
 426
 427                  sum += ret;
 428             }
 429             return sum;
 430         }
 431     }
 432     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 433 }
 434
 435 #ifdef CONFIG_AIO
 436 /***********************************************************/
 437 /* Unix AIO using POSIX AIO */
 438
 439 typedef struct RawAIOCB {
 440     BlockDriverAIOCB common;
 441     struct qemu_paiocb aiocb;
 442     struct RawAIOCB *next;
 443     int ret;
 444 } RawAIOCB;
 445
 446 typedef struct PosixAioState
 447 {
 448     int fd;
 449     RawAIOCB *first_aio;
 450 } PosixAioState;
 451
 452 static void posix_aio_read(void *opaque)
 453 {
 454     PosixAioState *s = opaque;
 455     RawAIOCB *acb, **pacb;
 456     int ret;
 457     size_t offset;
 458     union {
 459         struct qemu_signalfd_siginfo siginfo;
 460         char buf[128];
 461     } sig;
 462
 463     /* try to read from signalfd, don't freak out if we can't read anything */
 464     offset = 0;
 465     while (offset < 128) {
 466         ssize_t len;
 467
 468         len = read(s->fd, sig.buf + offset, 128 - offset);
 469         if (len == -1 && errno == EINTR)
 470             continue;
 471         if (len == -1 && errno == EAGAIN) {
 472             /* there is no natural reason for this to happen,
 473              * so we'll spin hard until we get everything just
 474              * to be on the safe side. */
 475             if (offset > 0)
 476                 continue;
 477         }
 478
 479         offset += len;
 480     }
 481
 482     for(;;) {
 483         pacb = &s->first_aio;
 484         for(;;) {
 485             acb = *pacb;
 486             if (!acb)
 487                 goto the_end;
 488             ret = qemu_paio_error(&acb->aiocb);
 489             if (ret == ECANCELED) {
 490                 /* remove the request */
 491                 *pacb = acb->next;
 492                 qemu_aio_release(acb);
 493             } else if (ret != EINPROGRESS) {
 494                 /* end of aio */
 495                 if (ret == 0) {
 496                     ret = qemu_paio_return(&acb->aiocb);
 497                     if (ret == acb->aiocb.aio_nbytes)
 498                         ret = 0;
 499                     else
 500                         ret = -EINVAL;
 501                 } else {
 502                     ret = -ret;
 503                 }
 504                 /* remove the request */
 505                 *pacb = acb->next;
 506                 /* call the callback */
 507                 acb->common.cb(acb->common.opaque, ret);
 508                 qemu_aio_release(acb);
 509                 break;
 510             } else {
 511                 pacb = &acb->next;
 512             }
 513         }
 514     }
 515  the_end: ;
 516 }
 517
 518 static int posix_aio_flush(void *opaque)
 519 {
 520     PosixAioState *s = opaque;
 521     return !!s->first_aio;
 522 }
 523
 524 static PosixAioState *posix_aio_state;
 525
 526 static int posix_aio_init(void)
 527 {
 528     sigset_t mask;
 529     PosixAioState *s;
 530     int fds[2];
 531     struct qemu_paioinit ai;
 532
 533     if (posix_aio_state)
 534         return 0;
 535
 536     s = qemu_malloc(sizeof(PosixAioState));
 537     if (s == NULL)
 538         return -ENOMEM;
 539
 540     /* Make sure to block AIO signal */
 541     sigemptyset(&mask);
 542     sigaddset(&mask, SIGUSR2);
 543     sigprocmask(SIG_BLOCK, &mask, NULL);
 544
 545     s->first_aio = NULL;
 546     s->fd = qemu_signalfd(&mask);
 547     if (s->fd == -1) {
 548         fprintf(stderr, "failed to create signalfd\n");
 549         return -errno;
 550     }
 551
 552     fcntl(s->fd, F_SETFL, O_NONBLOCK);
 553
 554     qemu_aio_set_fd_handler(s->fd, posix_aio_read, NULL, posix_aio_flush, s);
 555
 556     memset(&ai, 0, sizeof(ai));
 557     ai.aio_threads = 64;
 558     ai.aio_num = 64;
 559     qemu_paio_init(&ai);
 560
 561     posix_aio_state = s;
 562
 563     return 0;
 564 }
 565
 566 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 567         int64_t sector_num, uint8_t *buf, int nb_sectors,
 568         BlockDriverCompletionFunc *cb, void *opaque)
 569 {
 570     BDRVRawState *s = bs->opaque;
 571     RawAIOCB *acb;
 572
 573     if (fd_open(bs) < 0)
 574         return NULL;
 575
 576     acb = qemu_aio_get(bs, cb, opaque);
 577     if (!acb)
 578         return NULL;
 579     acb->aiocb.aio_fildes = s->fd;
 580     acb->aiocb.sigev_signo = SIGUSR2;
 581     acb->aiocb.aio_buf = buf;
 582     if (nb_sectors < 0)
 583         acb->aiocb.aio_nbytes = -nb_sectors;
 584     else
 585         acb->aiocb.aio_nbytes = nb_sectors * 512;
 586     acb->aiocb.aio_offset = sector_num * 512;
 587     acb->next = posix_aio_state->first_aio;
 588     posix_aio_state->first_aio = acb;
 589     return acb;
 590 }
 591
 592 static void raw_aio_em_cb(void* opaque)
 593 {
 594     RawAIOCB *acb = opaque;
 595     acb->common.cb(acb->common.opaque, acb->ret);
 596     qemu_aio_release(acb);
 597 }
 598
 599 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 600         int64_t sector_num, uint8_t *buf, int nb_sectors,
 601         BlockDriverCompletionFunc *cb, void *opaque)
 602 {
 603     RawAIOCB *acb;
 604
 605     /*
 606      * If O_DIRECT is used and the buffer is not aligned fall back
 607      * to synchronous IO.
 608      */
 609     BDRVRawState *s = bs->opaque;
 610
 611     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 612         QEMUBH *bh;
 613         acb = qemu_aio_get(bs, cb, opaque);
 614         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 615         bh = qemu_bh_new(raw_aio_em_cb, acb);
 616         qemu_bh_schedule(bh);
 617         return &acb->common;
 618     }
 619
 620     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 621     if (!acb)
 622         return NULL;
 623     if (qemu_paio_read(&acb->aiocb) < 0) {
 624         qemu_aio_release(acb);
 625         return NULL;
 626     }
 627     return &acb->common;
 628 }
 629
 630 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 631         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 632         BlockDriverCompletionFunc *cb, void *opaque)
 633 {
 634     RawAIOCB *acb;
 635
 636     /*
 637      * If O_DIRECT is used and the buffer is not aligned fall back
 638      * to synchronous IO.
 639      */
 640     BDRVRawState *s = bs->opaque;
 641
 642     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 643         QEMUBH *bh;
 644         acb = qemu_aio_get(bs, cb, opaque);
 645         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 646         bh = qemu_bh_new(raw_aio_em_cb, acb);
 647         qemu_bh_schedule(bh);
 648         return &acb->common;
 649     }
 650
 651     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 652     if (!acb)
 653         return NULL;
 654     if (qemu_paio_write(&acb->aiocb) < 0) {
 655         qemu_aio_release(acb);
 656         return NULL;
 657     }
 658     return &acb->common;
 659 }
 660
 661 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 662 {
 663     int ret;
 664     RawAIOCB *acb = (RawAIOCB *)blockacb;
 665     RawAIOCB **pacb;
 666
 667     ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 668     if (ret == QEMU_PAIO_NOTCANCELED) {
 669         /* fail safe: if the aio could not be canceled, we wait for
 670            it */
 671         while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
 672     }
 673
 674     /* remove the callback from the queue */
 675     pacb = &posix_aio_state->first_aio;
 676     for(;;) {
 677         if (*pacb == NULL) {
 678             break;
 679         } else if (*pacb == acb) {
 680             *pacb = acb->next;
 681             qemu_aio_release(acb);
 682             break;
 683         }
 684         pacb = &acb->next;
 685     }
 686 }
 687 #else /* CONFIG_AIO */
 688 static int posix_aio_init(void)
 689 {
 690     return 0;
 691 }
 692 #endif /* CONFIG_AIO */
 693
 694
 695 static void raw_close(BlockDriverState *bs)
 696 {
 697     BDRVRawState *s = bs->opaque;
 698     if (s->fd >= 0) {
 699         close(s->fd);
 700         s->fd = -1;
 701         if (s->aligned_buf != NULL)
 702             qemu_free(s->aligned_buf);
 703     }
 704 }
 705
 706 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 707 {
 708     BDRVRawState *s = bs->opaque;
 709     if (s->type != FTYPE_FILE)
 710         return -ENOTSUP;
 711     if (ftruncate(s->fd, offset) < 0)
 712         return -errno;
 713     return 0;
 714 }
 715
 716 #ifdef __OpenBSD__
 717 static int64_t raw_getlength(BlockDriverState *bs)
 718 {
 719     BDRVRawState *s = bs->opaque;
 720     int fd = s->fd;
 721     struct stat st;
 722
 723     if (fstat(fd, &st))
 724         return -1;
 725     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 726         struct disklabel dl;
 727
 728         if (ioctl(fd, DIOCGDINFO, &dl))
 729             return -1;
 730         return (uint64_t)dl.d_secsize *
 731             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 732     } else
 733         return st.st_size;
 734 }
 735 #else /* !__OpenBSD__ */
 736 static int64_t  raw_getlength(BlockDriverState *bs)
 737 {
 738     BDRVRawState *s = bs->opaque;
 739     int fd = s->fd;
 740     int64_t size;
 741 #ifdef _BSD
 742     struct stat sb;
 743 #endif
 744 #ifdef __sun__
 745     struct dk_minfo minfo;
 746     int rv;
 747 #endif
 748     int ret;
 749
 750     ret = fd_open(bs);
 751     if (ret < 0)
 752         return ret;
 753
 754 #ifdef _BSD
 755     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 756 #ifdef DIOCGMEDIASIZE
 757         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 758 #endif
 759 #ifdef CONFIG_COCOA
 760         size = LONG_LONG_MAX;
 761 #else
 762         size = lseek(fd, 0LL, SEEK_END);
 763 #endif
 764     } else
 765 #endif
 766 #ifdef __sun__
 767     /*
 768      * use the DKIOCGMEDIAINFO ioctl to read the size.
 769      */
 770     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 771     if ( rv != -1 ) {
 772         size = minfo.dki_lbsize * minfo.dki_capacity;
 773     } else /* there are reports that lseek on some devices
 774               fails, but irc discussion said that contingency
 775               on contingency was overkill */
 776 #endif
 777     {
 778         size = lseek(fd, 0, SEEK_END);
 779     }
 780     return size;
 781 }
 782 #endif
 783
 784 static int raw_create(const char *filename, int64_t total_size,
 785                       const char *backing_file, int flags)
 786 {
 787     int fd;
 788
 789     if (flags || backing_file)
 790         return -ENOTSUP;
 791
 792     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 793               0644);
 794     if (fd < 0)
 795         return -EIO;
 796     ftruncate(fd, total_size * 512);
 797     close(fd);
 798     return 0;
 799 }
 800
 801 static void raw_flush(BlockDriverState *bs)
 802 {
 803     BDRVRawState *s = bs->opaque;
 804     fsync(s->fd);
 805 }
 806
 807 BlockDriver bdrv_raw = {
 808     "raw",
 809     sizeof(BDRVRawState),
 810     NULL, /* no probe for protocols */
 811     raw_open,
 812     NULL,
 813     NULL,
 814     raw_close,
 815     raw_create,
 816     raw_flush,
 817
 818 #ifdef CONFIG_AIO
 819     .bdrv_aio_read = raw_aio_read,
 820     .bdrv_aio_write = raw_aio_write,
 821     .bdrv_aio_cancel = raw_aio_cancel,
 822     .aiocb_size = sizeof(RawAIOCB),
 823 #endif
 824
 825     .bdrv_pread = raw_pread,
 826     .bdrv_pwrite = raw_pwrite,
 827     .bdrv_truncate = raw_truncate,
 828     .bdrv_getlength = raw_getlength,
 829 };
 830
 831 /***********************************************/
 832 /* host device */
 833
 834 #ifdef CONFIG_COCOA
 835 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 836 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 837
 838 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 839 {
 840     kern_return_t       kernResult;
 841     mach_port_t     masterPort;
 842     CFMutableDictionaryRef  classesToMatch;
 843
 844     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 845     if ( KERN_SUCCESS != kernResult ) {
 846         printf( "IOMasterPort returned %d\n", kernResult );
 847     }
 848
 849     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 850     if ( classesToMatch == NULL ) {
 851         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 852     } else {
 853     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 854     }
 855     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 856     if ( KERN_SUCCESS != kernResult )
 857     {
 858         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 859     }
 860
 861     return kernResult;
 862 }
 863
 864 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 865 {
 866     io_object_t     nextMedia;
 867     kern_return_t   kernResult = KERN_FAILURE;
 868     *bsdPath = '\0';
 869     nextMedia = IOIteratorNext( mediaIterator );
 870     if ( nextMedia )
 871     {
 872         CFTypeRef   bsdPathAsCFString;
 873     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 874         if ( bsdPathAsCFString ) {
 875             size_t devPathLength;
 876             strcpy( bsdPath, _PATH_DEV );
 877             strcat( bsdPath, "r" );
 878             devPathLength = strlen( bsdPath );
 879             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 880                 kernResult = KERN_SUCCESS;
 881             }
 882             CFRelease( bsdPathAsCFString );
 883         }
 884         IOObjectRelease( nextMedia );
 885     }
 886
 887     return kernResult;
 888 }
 889
 890 #endif
 891
 892 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 893 {
 894     BDRVRawState *s = bs->opaque;
 895     int fd, open_flags, ret;
 896
 897     posix_aio_init();
 898
 899 #ifdef CONFIG_COCOA
 900     if (strstart(filename, "/dev/cdrom", NULL)) {
 901         kern_return_t kernResult;
 902         io_iterator_t mediaIterator;
 903         char bsdPath[ MAXPATHLEN ];
 904         int fd;
 905
 906         kernResult = FindEjectableCDMedia( &mediaIterator );
 907         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 908
 909         if ( bsdPath[ 0 ] != '\0' ) {
 910             strcat(bsdPath,"s0");
 911             /* some CDs don't have a partition 0 */
 912             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 913             if (fd < 0) {
 914                 bsdPath[strlen(bsdPath)-1] = '1';
 915             } else {
 916                 close(fd);
 917             }
 918             filename = bsdPath;
 919         }
 920
 921         if ( mediaIterator )
 922             IOObjectRelease( mediaIterator );
 923     }
 924 #endif
 925     open_flags = O_BINARY;
 926     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 927         open_flags |= O_RDWR;
 928     } else {
 929         open_flags |= O_RDONLY;
 930         bs->read_only = 1;
 931     }
 932     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 933      * and O_DIRECT for no caching. */
 934     if ((flags & BDRV_O_NOCACHE))
 935         open_flags |= O_DIRECT;
 936     else if (!(flags & BDRV_O_CACHE_WB))
 937         open_flags |= O_DSYNC;
 938
 939     s->type = FTYPE_FILE;
 940 #if defined(__linux__)
 941     if (strstart(filename, "/dev/cd", NULL)) {
 942         /* open will not fail even if no CD is inserted */
 943         open_flags |= O_NONBLOCK;
 944         s->type = FTYPE_CD;
 945     } else if (strstart(filename, "/dev/fd", NULL)) {
 946         s->type = FTYPE_FD;
 947         s->fd_open_flags = open_flags;
 948         /* open will not fail even if no floppy is inserted */
 949         open_flags |= O_NONBLOCK;
 950     } else if (strstart(filename, "/dev/sg", NULL)) {
 951         bs->sg = 1;
 952     }
 953 #endif
 954     fd = open(filename, open_flags, 0644);
 955     if (fd < 0) {
 956         ret = -errno;
 957         if (ret == -EROFS)
 958             ret = -EACCES;
 959         return ret;
 960     }
 961     s->fd = fd;
 962 #if defined(__linux__)
 963     /* close fd so that we can reopen it as needed */
 964     if (s->type == FTYPE_FD) {
 965         close(s->fd);
 966         s->fd = -1;
 967         s->fd_media_changed = 1;
 968     }
 969 #endif
 970     return 0;
 971 }
 972
 973 #if defined(__linux__)
 974 /* Note: we do not have a reliable method to detect if the floppy is
 975    present. The current method is to try to open the floppy at every
 976    I/O and to keep it opened during a few hundreds of ms. */
 977 static int fd_open(BlockDriverState *bs)
 978 {
 979     BDRVRawState *s = bs->opaque;
 980     int last_media_present;
 981
 982     if (s->type != FTYPE_FD)
 983         return 0;
 984     last_media_present = (s->fd >= 0);
 985     if (s->fd >= 0 &&
 986         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
 987         close(s->fd);
 988         s->fd = -1;
 989 #ifdef DEBUG_FLOPPY
 990         printf("Floppy closed\n");
 991 #endif
 992     }
 993     if (s->fd < 0) {
 994         if (s->fd_got_error &&
 995             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
 996 #ifdef DEBUG_FLOPPY
 997             printf("No floppy (open delayed)\n");
 998 #endif
 999             return -EIO;
1000         }
1001         s->fd = open(bs->filename, s->fd_open_flags);
1002         if (s->fd < 0) {
1003             s->fd_error_time = qemu_get_clock(rt_clock);
1004             s->fd_got_error = 1;
1005             if (last_media_present)
1006                 s->fd_media_changed = 1;
1007 #ifdef DEBUG_FLOPPY
1008             printf("No floppy\n");
1009 #endif
1010             return -EIO;
1011         }
1012 #ifdef DEBUG_FLOPPY
1013         printf("Floppy opened\n");
1014 #endif
1015     }
1016     if (!last_media_present)
1017         s->fd_media_changed = 1;
1018     s->fd_open_time = qemu_get_clock(rt_clock);
1019     s->fd_got_error = 0;
1020     return 0;
1021 }
1022
1023 static int raw_is_inserted(BlockDriverState *bs)
1024 {
1025     BDRVRawState *s = bs->opaque;
1026     int ret;
1027
1028     switch(s->type) {
1029     case FTYPE_CD:
1030         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1031         if (ret == CDS_DISC_OK)
1032             return 1;
1033         else
1034             return 0;
1035         break;
1036     case FTYPE_FD:
1037         ret = fd_open(bs);
1038         return (ret >= 0);
1039     default:
1040         return 1;
1041     }
1042 }
1043
1044 /* currently only used by fdc.c, but a CD version would be good too */
1045 static int raw_media_changed(BlockDriverState *bs)
1046 {
1047     BDRVRawState *s = bs->opaque;
1048
1049     switch(s->type) {
1050     case FTYPE_FD:
1051         {
1052             int ret;
1053             /* XXX: we do not have a true media changed indication. It
1054                does not work if the floppy is changed without trying
1055                to read it */
1056             fd_open(bs);
1057             ret = s->fd_media_changed;
1058             s->fd_media_changed = 0;
1059 #ifdef DEBUG_FLOPPY
1060             printf("Floppy changed=%d\n", ret);
1061 #endif
1062             return ret;
1063         }
1064     default:
1065         return -ENOTSUP;
1066     }
1067 }
1068
1069 static int raw_eject(BlockDriverState *bs, int eject_flag)
1070 {
1071     BDRVRawState *s = bs->opaque;
1072
1073     switch(s->type) {
1074     case FTYPE_CD:
1075         if (eject_flag) {
1076             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1077                 perror("CDROMEJECT");
1078         } else {
1079             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1080                 perror("CDROMEJECT");
1081         }
1082         break;
1083     case FTYPE_FD:
1084         {
1085             int fd;
1086             if (s->fd >= 0) {
1087                 close(s->fd);
1088                 s->fd = -1;
1089             }
1090             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1091             if (fd >= 0) {
1092                 if (ioctl(fd, FDEJECT, 0) < 0)
1093                     perror("FDEJECT");
1094                 close(fd);
1095             }
1096         }
1097         break;
1098     default:
1099         return -ENOTSUP;
1100     }
1101     return 0;
1102 }
1103
1104 static int raw_set_locked(BlockDriverState *bs, int locked)
1105 {
1106     BDRVRawState *s = bs->opaque;
1107
1108     switch(s->type) {
1109     case FTYPE_CD:
1110         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1111             /* Note: an error can happen if the distribution automatically
1112                mounts the CD-ROM */
1113             //        perror("CDROM_LOCKDOOR");
1114         }
1115         break;
1116     default:
1117         return -ENOTSUP;
1118     }
1119     return 0;
1120 }
1121
1122 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1123 {
1124     BDRVRawState *s = bs->opaque;
1125
1126     return ioctl(s->fd, req, buf);
1127 }
1128 #else
1129
1130 static int fd_open(BlockDriverState *bs)
1131 {
1132     return 0;
1133 }
1134
1135 static int raw_is_inserted(BlockDriverState *bs)
1136 {
1137     return 1;
1138 }
1139
1140 static int raw_media_changed(BlockDriverState *bs)
1141 {
1142     return -ENOTSUP;
1143 }
1144
1145 static int raw_eject(BlockDriverState *bs, int eject_flag)
1146 {
1147     return -ENOTSUP;
1148 }
1149
1150 static int raw_set_locked(BlockDriverState *bs, int locked)
1151 {
1152     return -ENOTSUP;
1153 }
1154
1155 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1156 {
1157     return -ENOTSUP;
1158 }
1159 #endif /* !linux */
1160
1161 BlockDriver bdrv_host_device = {
1162     "host_device",
1163     sizeof(BDRVRawState),
1164     NULL, /* no probe for protocols */
1165     hdev_open,
1166     NULL,
1167     NULL,
1168     raw_close,
1169     NULL,
1170     raw_flush,
1171
1172 #ifdef CONFIG_AIO
1173     .bdrv_aio_read = raw_aio_read,
1174     .bdrv_aio_write = raw_aio_write,
1175     .bdrv_aio_cancel = raw_aio_cancel,
1176     .aiocb_size = sizeof(RawAIOCB),
1177 #endif
1178
1179     .bdrv_pread = raw_pread,
1180     .bdrv_pwrite = raw_pwrite,
1181     .bdrv_getlength = raw_getlength,
1182
1183     /* removable device support */
1184     .bdrv_is_inserted = raw_is_inserted,
1185     .bdrv_media_changed = raw_media_changed,
1186     .bdrv_eject = raw_eject,
1187     .bdrv_set_locked = raw_set_locked,
1188     /* generic scsi device */
1189     .bdrv_ioctl = raw_ioctl,
1190 };