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 <assert.h>
  29 #ifdef CONFIG_AIO
  30 #include <aio.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 #endif
  59
  60 #ifdef __OpenBSD__
  61 #include <sys/ioctl.h>
  62 #include <sys/disklabel.h>
  63 #include <sys/dkio.h>
  64 #endif
  65
  66 //#define DEBUG_FLOPPY
  67
  68 //#define DEBUG_BLOCK
  69 #if defined(DEBUG_BLOCK)
  70 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (loglevel != 0)  \
  71     { fprintf(logfile, formatCstr, ##args); fflush(logfile); } } while (0)
  72 #else
  73 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  74 #endif
  75
  76 /* OS X does not have O_DSYNC */
  77 #ifndef O_DSYNC
  78 #define O_DSYNC O_SYNC
  79 #endif
  80
  81 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
  82 #ifndef O_DIRECT
  83 #define O_DIRECT O_DSYNC
  84 #endif
  85
  86 #define FTYPE_FILE   0
  87 #define FTYPE_CD     1
  88 #define FTYPE_FD     2
  89
  90 #define ALIGNED_BUFFER_SIZE (32 * 512)
  91
  92 /* if the FD is not accessed during that time (in ms), we try to
  93    reopen it to see if the disk has been changed */
  94 #define FD_OPEN_TIMEOUT 1000
  95
  96 /* posix-aio doesn't allow multiple outstanding requests to a single file
  97  * descriptor.  we implement a pool of dup()'d file descriptors to work
  98  * around this */
  99 #define RAW_FD_POOL_SIZE        64
 100
 101 typedef struct BDRVRawState {
 102     int fd;
 103     int type;
 104     unsigned int lseek_err_cnt;
 105     int fd_pool[RAW_FD_POOL_SIZE];
 106 #if defined(__linux__)
 107     /* linux floppy specific */
 108     int fd_open_flags;
 109     int64_t fd_open_time;
 110     int64_t fd_error_time;
 111     int fd_got_error;
 112     int fd_media_changed;
 113 #endif
 114     uint8_t* aligned_buf;
 115 } BDRVRawState;
 116
 117 static int posix_aio_init(void);
 118
 119 static int fd_open(BlockDriverState *bs);
 120
 121 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 122 {
 123     BDRVRawState *s = bs->opaque;
 124     int fd, open_flags, ret;
 125     int i;
 126
 127     posix_aio_init();
 128
 129     s->lseek_err_cnt = 0;
 130
 131     open_flags = O_BINARY;
 132     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 133         open_flags |= O_RDWR;
 134     } else {
 135         open_flags |= O_RDONLY;
 136         bs->read_only = 1;
 137     }
 138     if (flags & BDRV_O_CREAT)
 139         open_flags |= O_CREAT | O_TRUNC;
 140
 141     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 142      * and O_DIRECT for no caching. */
 143     if ((flags & BDRV_O_NOCACHE))
 144         open_flags |= O_DIRECT;
 145     else if (!(flags & BDRV_O_CACHE_WB))
 146         open_flags |= O_DSYNC;
 147
 148     s->type = FTYPE_FILE;
 149
 150     fd = open(filename, open_flags, 0644);
 151     if (fd < 0) {
 152         ret = -errno;
 153         if (ret == -EROFS)
 154             ret = -EACCES;
 155         return ret;
 156     }
 157     s->fd = fd;
 158     for (i = 0; i < RAW_FD_POOL_SIZE; i++)
 159         s->fd_pool[i] = -1;
 160     s->aligned_buf = NULL;
 161     if ((flags & BDRV_O_NOCACHE)) {
 162         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 163         if (s->aligned_buf == NULL) {
 164             ret = -errno;
 165             close(fd);
 166             return ret;
 167         }
 168     }
 169     return 0;
 170 }
 171
 172 /* XXX: use host sector size if necessary with:
 173 #ifdef DIOCGSECTORSIZE
 174         {
 175             unsigned int sectorsize = 512;
 176             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 177                 sectorsize > bufsize)
 178                 bufsize = sectorsize;
 179         }
 180 #endif
 181 #ifdef CONFIG_COCOA
 182         u_int32_t   blockSize = 512;
 183         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 184             bufsize = blockSize;
 185         }
 186 #endif
 187 */
 188
 189 /*
 190  * offset and count are in bytes, but must be multiples of 512 for files
 191  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 192  *
 193  * This function may be called without alignment if the caller ensures
 194  * that O_DIRECT is not in effect.
 195  */
 196 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 197                      uint8_t *buf, int count)
 198 {
 199     BDRVRawState *s = bs->opaque;
 200     int ret;
 201
 202     ret = fd_open(bs);
 203     if (ret < 0)
 204         return ret;
 205
 206     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 207         ++(s->lseek_err_cnt);
 208         if(s->lseek_err_cnt <= 10) {
 209             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 210                               "] lseek failed : %d = %s\n",
 211                               s->fd, bs->filename, offset, buf, count,
 212                               bs->total_sectors, errno, strerror(errno));
 213         }
 214         return -1;
 215     }
 216     s->lseek_err_cnt=0;
 217
 218     ret = read(s->fd, buf, count);
 219     if (ret == count)
 220         goto label__raw_read__success;
 221
 222     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 223                       "] read failed %d : %d = %s\n",
 224                       s->fd, bs->filename, offset, buf, count,
 225                       bs->total_sectors, ret, errno, strerror(errno));
 226
 227     /* Try harder for CDrom. */
 228     if (bs->type == BDRV_TYPE_CDROM) {
 229         lseek(s->fd, offset, SEEK_SET);
 230         ret = read(s->fd, buf, count);
 231         if (ret == count)
 232             goto label__raw_read__success;
 233         lseek(s->fd, offset, SEEK_SET);
 234         ret = read(s->fd, buf, count);
 235         if (ret == count)
 236             goto label__raw_read__success;
 237
 238         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 239                           "] retry read failed %d : %d = %s\n",
 240                           s->fd, bs->filename, offset, buf, count,
 241                           bs->total_sectors, ret, errno, strerror(errno));
 242     }
 243
 244 label__raw_read__success:
 245
 246     return ret;
 247 }
 248
 249 /*
 250  * offset and count are in bytes, but must be multiples of 512 for files
 251  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 252  *
 253  * This function may be called without alignment if the caller ensures
 254  * that O_DIRECT is not in effect.
 255  */
 256 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 257                       const uint8_t *buf, int count)
 258 {
 259     BDRVRawState *s = bs->opaque;
 260     int ret;
 261
 262     ret = fd_open(bs);
 263     if (ret < 0)
 264         return ret;
 265
 266     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 267         ++(s->lseek_err_cnt);
 268         if(s->lseek_err_cnt) {
 269             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 270                               PRId64 "] lseek failed : %d = %s\n",
 271                               s->fd, bs->filename, offset, buf, count,
 272                               bs->total_sectors, errno, strerror(errno));
 273         }
 274         return -1;
 275     }
 276     s->lseek_err_cnt = 0;
 277
 278     ret = write(s->fd, buf, count);
 279     if (ret == count)
 280         goto label__raw_write__success;
 281
 282     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 283                       "] write failed %d : %d = %s\n",
 284                       s->fd, bs->filename, offset, buf, count,
 285                       bs->total_sectors, ret, errno, strerror(errno));
 286
 287 label__raw_write__success:
 288
 289     return ret;
 290 }
 291
 292
 293 /*
 294  * offset and count are in bytes and possibly not aligned. For files opened
 295  * with O_DIRECT, necessary alignments are ensured before calling
 296  * raw_pread_aligned to do the actual read.
 297  */
 298 static int raw_pread(BlockDriverState *bs, int64_t offset,
 299                      uint8_t *buf, int count)
 300 {
 301     BDRVRawState *s = bs->opaque;
 302     int size, ret, shift, sum;
 303
 304     sum = 0;
 305
 306     if (s->aligned_buf != NULL)  {
 307
 308         if (offset & 0x1ff) {
 309             /* align offset on a 512 bytes boundary */
 310
 311             shift = offset & 0x1ff;
 312             size = (shift + count + 0x1ff) & ~0x1ff;
 313             if (size > ALIGNED_BUFFER_SIZE)
 314                 size = ALIGNED_BUFFER_SIZE;
 315             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 316             if (ret < 0)
 317                 return ret;
 318
 319             size = 512 - shift;
 320             if (size > count)
 321                 size = count;
 322             memcpy(buf, s->aligned_buf + shift, size);
 323
 324             buf += size;
 325             offset += size;
 326             count -= size;
 327             sum += size;
 328
 329             if (count == 0)
 330                 return sum;
 331         }
 332         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 333
 334             /* read on aligned buffer */
 335
 336             while (count) {
 337
 338                 size = (count + 0x1ff) & ~0x1ff;
 339                 if (size > ALIGNED_BUFFER_SIZE)
 340                     size = ALIGNED_BUFFER_SIZE;
 341
 342                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 343                 if (ret < 0)
 344                     return ret;
 345
 346                 size = ret;
 347                 if (size > count)
 348                     size = count;
 349
 350                 memcpy(buf, s->aligned_buf, size);
 351
 352                 buf += size;
 353                 offset += size;
 354                 count -= size;
 355                 sum += size;
 356             }
 357
 358             return sum;
 359         }
 360     }
 361
 362     return raw_pread_aligned(bs, offset, buf, count) + sum;
 363 }
 364
 365 /*
 366  * offset and count are in bytes and possibly not aligned. For files opened
 367  * with O_DIRECT, necessary alignments are ensured before calling
 368  * raw_pwrite_aligned to do the actual write.
 369  */
 370 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 371                       const uint8_t *buf, int count)
 372 {
 373     BDRVRawState *s = bs->opaque;
 374     int size, ret, shift, sum;
 375
 376     sum = 0;
 377
 378     if (s->aligned_buf != NULL) {
 379
 380         if (offset & 0x1ff) {
 381             /* align offset on a 512 bytes boundary */
 382             shift = offset & 0x1ff;
 383             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 384             if (ret < 0)
 385                 return ret;
 386
 387             size = 512 - shift;
 388             if (size > count)
 389                 size = count;
 390             memcpy(s->aligned_buf + shift, buf, size);
 391
 392             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 393             if (ret < 0)
 394                 return ret;
 395
 396             buf += size;
 397             offset += size;
 398             count -= size;
 399             sum += size;
 400
 401             if (count == 0)
 402                 return sum;
 403         }
 404         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 405
 406             while ((size = (count & ~0x1ff)) != 0) {
 407
 408                 if (size > ALIGNED_BUFFER_SIZE)
 409                     size = ALIGNED_BUFFER_SIZE;
 410
 411                 memcpy(s->aligned_buf, buf, size);
 412
 413                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 414                 if (ret < 0)
 415                     return ret;
 416
 417                 buf += ret;
 418                 offset += ret;
 419                 count -= ret;
 420                 sum += ret;
 421             }
 422             /* here, count < 512 because (count & ~0x1ff) == 0 */
 423             if (count) {
 424                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 425                 if (ret < 0)
 426                     return ret;
 427                  memcpy(s->aligned_buf, buf, count);
 428
 429                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 430                  if (ret < 0)
 431                      return ret;
 432                  if (count < ret)
 433                      ret = count;
 434
 435                  sum += ret;
 436             }
 437             return sum;
 438         }
 439     }
 440     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 441 }
 442
 443 #ifdef CONFIG_AIO
 444 /***********************************************************/
 445 /* Unix AIO using POSIX AIO */
 446
 447 typedef struct RawAIOCB {
 448     BlockDriverAIOCB common;
 449     int fd;
 450     struct aiocb aiocb;
 451     struct RawAIOCB *next;
 452     int ret;
 453 } RawAIOCB;
 454
 455 typedef struct PosixAioState
 456 {
 457     int rfd, wfd;
 458     RawAIOCB *first_aio;
 459 } PosixAioState;
 460
 461 static int raw_fd_pool_get(BDRVRawState *s)
 462 {
 463     int i;
 464
 465     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 466         /* already in use */
 467         if (s->fd_pool[i] != -1)
 468             continue;
 469
 470         /* try to dup file descriptor */
 471         s->fd_pool[i] = dup(s->fd);
 472         if (s->fd_pool[i] != -1)
 473             return s->fd_pool[i];
 474     }
 475
 476     /* we couldn't dup the file descriptor so just use the main one */
 477     return s->fd;
 478 }
 479
 480 static void raw_fd_pool_put(RawAIOCB *acb)
 481 {
 482     BDRVRawState *s = acb->common.bs->opaque;
 483     int i;
 484
 485     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 486         if (s->fd_pool[i] == acb->fd) {
 487             close(s->fd_pool[i]);
 488             s->fd_pool[i] = -1;
 489         }
 490     }
 491 }
 492
 493 static void posix_aio_read(void *opaque)
 494 {
 495     PosixAioState *s = opaque;
 496     RawAIOCB *acb, **pacb;
 497     int ret;
 498     ssize_t len;
 499
 500     do {
 501         char byte;
 502
 503         len = read(s->rfd, &byte, 1);
 504         if (len == -1 && errno == EINTR)
 505             continue;
 506         if (len == -1 && errno == EAGAIN)
 507             break;
 508     } while (len == -1);
 509
 510     for(;;) {
 511         pacb = &s->first_aio;
 512         for(;;) {
 513             acb = *pacb;
 514             if (!acb)
 515                 goto the_end;
 516             ret = aio_error(&acb->aiocb);
 517             if (ret == ECANCELED) {
 518                 /* remove the request */
 519                 *pacb = acb->next;
 520                 raw_fd_pool_put(acb);
 521                 qemu_aio_release(acb);
 522             } else if (ret != EINPROGRESS) {
 523                 /* end of aio */
 524                 if (ret == 0) {
 525                     ret = aio_return(&acb->aiocb);
 526                     if (ret == acb->aiocb.aio_nbytes)
 527                         ret = 0;
 528                     else
 529                         ret = -EINVAL;
 530                 } else {
 531                     ret = -ret;
 532                 }
 533                 /* remove the request */
 534                 *pacb = acb->next;
 535                 /* call the callback */
 536                 acb->common.cb(acb->common.opaque, ret);
 537                 raw_fd_pool_put(acb);
 538                 qemu_aio_release(acb);
 539                 break;
 540             } else {
 541                 pacb = &acb->next;
 542             }
 543         }
 544     }
 545  the_end: ;
 546 }
 547
 548 static int posix_aio_flush(void *opaque)
 549 {
 550     PosixAioState *s = opaque;
 551     return !!s->first_aio;
 552 }
 553
 554 static PosixAioState *posix_aio_state;
 555
 556 static void aio_signal_handler(int signum)
 557 {
 558     if (posix_aio_state) {
 559         char byte = 0;
 560
 561         write(posix_aio_state->wfd, &byte, sizeof(byte));
 562     }
 563
 564     qemu_service_io();
 565 }
 566
 567 static int posix_aio_init(void)
 568 {
 569     struct sigaction act;
 570     PosixAioState *s;
 571     int fds[2];
 572
 573     if (posix_aio_state)
 574         return 0;
 575
 576     s = qemu_malloc(sizeof(PosixAioState));
 577     if (s == NULL)
 578         return -ENOMEM;
 579
 580     sigfillset(&act.sa_mask);
 581     act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
 582     act.sa_handler = aio_signal_handler;
 583     sigaction(SIGUSR2, &act, NULL);
 584
 585     s->first_aio = NULL;
 586     if (pipe(fds) == -1) {
 587         fprintf(stderr, "failed to create pipe\n");
 588         return -errno;
 589     }
 590
 591     s->rfd = fds[0];
 592     s->wfd = fds[1];
 593
 594     fcntl(s->wfd, F_SETFL, O_NONBLOCK);
 595
 596     qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
 597
 598 #if defined(__linux__)
 599     {
 600         struct aioinit ai;
 601
 602         memset(&ai, 0, sizeof(ai));
 603 #if defined(__GLIBC_PREREQ) && __GLIBC_PREREQ(2, 4)
 604         ai.aio_threads = 64;
 605         ai.aio_num = 64;
 606 #else
 607         /* XXX: aio thread exit seems to hang on RedHat 9 and this init
 608            seems to fix the problem. */
 609         ai.aio_threads = 1;
 610         ai.aio_num = 1;
 611         ai.aio_idle_time = 365 * 100000;
 612 #endif
 613         aio_init(&ai);
 614     }
 615 #endif
 616     posix_aio_state = s;
 617
 618     return 0;
 619 }
 620
 621 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 622         int64_t sector_num, uint8_t *buf, int nb_sectors,
 623         BlockDriverCompletionFunc *cb, void *opaque)
 624 {
 625     BDRVRawState *s = bs->opaque;
 626     RawAIOCB *acb;
 627
 628     if (fd_open(bs) < 0)
 629         return NULL;
 630
 631     acb = qemu_aio_get(bs, cb, opaque);
 632     if (!acb)
 633         return NULL;
 634     acb->fd = raw_fd_pool_get(s);
 635     acb->aiocb.aio_fildes = acb->fd;
 636     acb->aiocb.aio_sigevent.sigev_signo = SIGUSR2;
 637     acb->aiocb.aio_sigevent.sigev_notify = SIGEV_SIGNAL;
 638     acb->aiocb.aio_buf = buf;
 639     if (nb_sectors < 0)
 640         acb->aiocb.aio_nbytes = -nb_sectors;
 641     else
 642         acb->aiocb.aio_nbytes = nb_sectors * 512;
 643     acb->aiocb.aio_offset = sector_num * 512;
 644     acb->next = posix_aio_state->first_aio;
 645     posix_aio_state->first_aio = acb;
 646     return acb;
 647 }
 648
 649 static void raw_aio_em_cb(void* opaque)
 650 {
 651     RawAIOCB *acb = opaque;
 652     acb->common.cb(acb->common.opaque, acb->ret);
 653     qemu_aio_release(acb);
 654 }
 655
 656 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 657         int64_t sector_num, uint8_t *buf, int nb_sectors,
 658         BlockDriverCompletionFunc *cb, void *opaque)
 659 {
 660     RawAIOCB *acb;
 661
 662     /*
 663      * If O_DIRECT is used and the buffer is not aligned fall back
 664      * to synchronous IO.
 665      */
 666     BDRVRawState *s = bs->opaque;
 667
 668     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 669         QEMUBH *bh;
 670         acb = qemu_aio_get(bs, cb, opaque);
 671         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 672         bh = qemu_bh_new(raw_aio_em_cb, acb);
 673         qemu_bh_schedule(bh);
 674         return &acb->common;
 675     }
 676
 677     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 678     if (!acb)
 679         return NULL;
 680     if (aio_read(&acb->aiocb) < 0) {
 681         qemu_aio_release(acb);
 682         return NULL;
 683     }
 684     return &acb->common;
 685 }
 686
 687 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 688         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 689         BlockDriverCompletionFunc *cb, void *opaque)
 690 {
 691     RawAIOCB *acb;
 692
 693     /*
 694      * If O_DIRECT is used and the buffer is not aligned fall back
 695      * to synchronous IO.
 696      */
 697     BDRVRawState *s = bs->opaque;
 698
 699     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 700         QEMUBH *bh;
 701         acb = qemu_aio_get(bs, cb, opaque);
 702         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 703         bh = qemu_bh_new(raw_aio_em_cb, acb);
 704         qemu_bh_schedule(bh);
 705         return &acb->common;
 706     }
 707
 708     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 709     if (!acb)
 710         return NULL;
 711     if (aio_write(&acb->aiocb) < 0) {
 712         qemu_aio_release(acb);
 713         return NULL;
 714     }
 715     return &acb->common;
 716 }
 717
 718 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 719 {
 720     int ret;
 721     RawAIOCB *acb = (RawAIOCB *)blockacb;
 722     RawAIOCB **pacb;
 723
 724     ret = aio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 725     if (ret == AIO_NOTCANCELED) {
 726         /* fail safe: if the aio could not be canceled, we wait for
 727            it */
 728         while (aio_error(&acb->aiocb) == EINPROGRESS);
 729     }
 730
 731     /* remove the callback from the queue */
 732     pacb = &posix_aio_state->first_aio;
 733     for(;;) {
 734         if (*pacb == NULL) {
 735             break;
 736         } else if (*pacb == acb) {
 737             *pacb = acb->next;
 738             raw_fd_pool_put(acb);
 739             qemu_aio_release(acb);
 740             break;
 741         }
 742         pacb = &acb->next;
 743     }
 744 }
 745
 746 #else /* CONFIG_AIO */
 747 static int posix_aio_init(void)
 748 {
 749     return 0;
 750 }
 751 #endif /* CONFIG_AIO */
 752
 753 static void raw_close_fd_pool(BDRVRawState *s)
 754 {
 755     int i;
 756
 757     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 758         if (s->fd_pool[i] != -1) {
 759             close(s->fd_pool[i]);
 760             s->fd_pool[i] = -1;
 761         }
 762     }
 763 }
 764
 765 static void raw_close(BlockDriverState *bs)
 766 {
 767     BDRVRawState *s = bs->opaque;
 768     if (s->fd >= 0) {
 769         close(s->fd);
 770         s->fd = -1;
 771         if (s->aligned_buf != NULL)
 772             qemu_free(s->aligned_buf);
 773     }
 774     raw_close_fd_pool(s);
 775 }
 776
 777 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 778 {
 779     BDRVRawState *s = bs->opaque;
 780     if (s->type != FTYPE_FILE)
 781         return -ENOTSUP;
 782     if (ftruncate(s->fd, offset) < 0)
 783         return -errno;
 784     return 0;
 785 }
 786
 787 #ifdef __OpenBSD__
 788 static int64_t raw_getlength(BlockDriverState *bs)
 789 {
 790     BDRVRawState *s = bs->opaque;
 791     int fd = s->fd;
 792     struct stat st;
 793
 794     if (fstat(fd, &st))
 795         return -1;
 796     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 797         struct disklabel dl;
 798
 799         if (ioctl(fd, DIOCGDINFO, &dl))
 800             return -1;
 801         return (uint64_t)dl.d_secsize *
 802             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 803     } else
 804         return st.st_size;
 805 }
 806 #else /* !__OpenBSD__ */
 807 static int64_t  raw_getlength(BlockDriverState *bs)
 808 {
 809     BDRVRawState *s = bs->opaque;
 810     int fd = s->fd;
 811     int64_t size;
 812 #ifdef _BSD
 813     struct stat sb;
 814 #endif
 815 #ifdef __sun__
 816     struct dk_minfo minfo;
 817     int rv;
 818 #endif
 819     int ret;
 820
 821     ret = fd_open(bs);
 822     if (ret < 0)
 823         return ret;
 824
 825 #ifdef _BSD
 826     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 827 #ifdef DIOCGMEDIASIZE
 828         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 829 #endif
 830 #ifdef CONFIG_COCOA
 831         size = LONG_LONG_MAX;
 832 #else
 833         size = lseek(fd, 0LL, SEEK_END);
 834 #endif
 835     } else
 836 #endif
 837 #ifdef __sun__
 838     /*
 839      * use the DKIOCGMEDIAINFO ioctl to read the size.
 840      */
 841     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 842     if ( rv != -1 ) {
 843         size = minfo.dki_lbsize * minfo.dki_capacity;
 844     } else /* there are reports that lseek on some devices
 845               fails, but irc discussion said that contingency
 846               on contingency was overkill */
 847 #endif
 848     {
 849         size = lseek(fd, 0, SEEK_END);
 850     }
 851     return size;
 852 }
 853 #endif
 854
 855 static int raw_create(const char *filename, int64_t total_size,
 856                       const char *backing_file, int flags)
 857 {
 858     int fd;
 859
 860     if (flags || backing_file)
 861         return -ENOTSUP;
 862
 863     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 864               0644);
 865     if (fd < 0)
 866         return -EIO;
 867     ftruncate(fd, total_size * 512);
 868     close(fd);
 869     return 0;
 870 }
 871
 872 static void raw_flush(BlockDriverState *bs)
 873 {
 874     BDRVRawState *s = bs->opaque;
 875     fsync(s->fd);
 876 }
 877
 878 BlockDriver bdrv_raw = {
 879     "raw",
 880     sizeof(BDRVRawState),
 881     NULL, /* no probe for protocols */
 882     raw_open,
 883     NULL,
 884     NULL,
 885     raw_close,
 886     raw_create,
 887     raw_flush,
 888
 889 #ifdef CONFIG_AIO
 890     .bdrv_aio_read = raw_aio_read,
 891     .bdrv_aio_write = raw_aio_write,
 892     .bdrv_aio_cancel = raw_aio_cancel,
 893     .aiocb_size = sizeof(RawAIOCB),
 894 #endif
 895     .bdrv_pread = raw_pread,
 896     .bdrv_pwrite = raw_pwrite,
 897     .bdrv_truncate = raw_truncate,
 898     .bdrv_getlength = raw_getlength,
 899 };
 900
 901 /***********************************************/
 902 /* host device */
 903
 904 #ifdef CONFIG_COCOA
 905 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 906 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 907
 908 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 909 {
 910     kern_return_t       kernResult;
 911     mach_port_t     masterPort;
 912     CFMutableDictionaryRef  classesToMatch;
 913
 914     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 915     if ( KERN_SUCCESS != kernResult ) {
 916         printf( "IOMasterPort returned %d\n", kernResult );
 917     }
 918
 919     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 920     if ( classesToMatch == NULL ) {
 921         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 922     } else {
 923     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 924     }
 925     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 926     if ( KERN_SUCCESS != kernResult )
 927     {
 928         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 929     }
 930
 931     return kernResult;
 932 }
 933
 934 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 935 {
 936     io_object_t     nextMedia;
 937     kern_return_t   kernResult = KERN_FAILURE;
 938     *bsdPath = '\0';
 939     nextMedia = IOIteratorNext( mediaIterator );
 940     if ( nextMedia )
 941     {
 942         CFTypeRef   bsdPathAsCFString;
 943     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 944         if ( bsdPathAsCFString ) {
 945             size_t devPathLength;
 946             strcpy( bsdPath, _PATH_DEV );
 947             strcat( bsdPath, "r" );
 948             devPathLength = strlen( bsdPath );
 949             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 950                 kernResult = KERN_SUCCESS;
 951             }
 952             CFRelease( bsdPathAsCFString );
 953         }
 954         IOObjectRelease( nextMedia );
 955     }
 956
 957     return kernResult;
 958 }
 959
 960 #endif
 961
 962 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 963 {
 964     BDRVRawState *s = bs->opaque;
 965     int fd, open_flags, ret, i;
 966
 967     posix_aio_init();
 968
 969 #ifdef CONFIG_COCOA
 970     if (strstart(filename, "/dev/cdrom", NULL)) {
 971         kern_return_t kernResult;
 972         io_iterator_t mediaIterator;
 973         char bsdPath[ MAXPATHLEN ];
 974         int fd;
 975
 976         kernResult = FindEjectableCDMedia( &mediaIterator );
 977         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 978
 979         if ( bsdPath[ 0 ] != '\0' ) {
 980             strcat(bsdPath,"s0");
 981             /* some CDs don't have a partition 0 */
 982             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 983             if (fd < 0) {
 984                 bsdPath[strlen(bsdPath)-1] = '1';
 985             } else {
 986                 close(fd);
 987             }
 988             filename = bsdPath;
 989         }
 990
 991         if ( mediaIterator )
 992             IOObjectRelease( mediaIterator );
 993     }
 994 #endif
 995     open_flags = O_BINARY;
 996     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 997         open_flags |= O_RDWR;
 998     } else {
 999         open_flags |= O_RDONLY;
1000         bs->read_only = 1;
1001     }
1002     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
1003      * and O_DIRECT for no caching. */
1004     if ((flags & BDRV_O_NOCACHE))
1005         open_flags |= O_DIRECT;
1006     else if (!(flags & BDRV_O_CACHE_WB))
1007         open_flags |= O_DSYNC;
1008
1009     s->type = FTYPE_FILE;
1010 #if defined(__linux__)
1011     if (strstart(filename, "/dev/cd", NULL)) {
1012         /* open will not fail even if no CD is inserted */
1013         open_flags |= O_NONBLOCK;
1014         s->type = FTYPE_CD;
1015     } else if (strstart(filename, "/dev/fd", NULL)) {
1016         s->type = FTYPE_FD;
1017         s->fd_open_flags = open_flags;
1018         /* open will not fail even if no floppy is inserted */
1019         open_flags |= O_NONBLOCK;
1020     } else if (strstart(filename, "/dev/sg", NULL)) {
1021         bs->sg = 1;
1022     }
1023 #endif
1024     fd = open(filename, open_flags, 0644);
1025     if (fd < 0) {
1026         ret = -errno;
1027         if (ret == -EROFS)
1028             ret = -EACCES;
1029         return ret;
1030     }
1031     s->fd = fd;
1032     for (i = 0; i < RAW_FD_POOL_SIZE; i++)
1033         s->fd_pool[i] = -1;
1034 #if defined(__linux__)
1035     /* close fd so that we can reopen it as needed */
1036     if (s->type == FTYPE_FD) {
1037         close(s->fd);
1038         s->fd = -1;
1039         s->fd_media_changed = 1;
1040     }
1041 #endif
1042     return 0;
1043 }
1044
1045 #if defined(__linux__)
1046 /* Note: we do not have a reliable method to detect if the floppy is
1047    present. The current method is to try to open the floppy at every
1048    I/O and to keep it opened during a few hundreds of ms. */
1049 static int fd_open(BlockDriverState *bs)
1050 {
1051     BDRVRawState *s = bs->opaque;
1052     int last_media_present;
1053
1054     if (s->type != FTYPE_FD)
1055         return 0;
1056     last_media_present = (s->fd >= 0);
1057     if (s->fd >= 0 &&
1058         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1059         close(s->fd);
1060         s->fd = -1;
1061         raw_close_fd_pool(s);
1062 #ifdef DEBUG_FLOPPY
1063         printf("Floppy closed\n");
1064 #endif
1065     }
1066     if (s->fd < 0) {
1067         if (s->fd_got_error &&
1068             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1069 #ifdef DEBUG_FLOPPY
1070             printf("No floppy (open delayed)\n");
1071 #endif
1072             return -EIO;
1073         }
1074         s->fd = open(bs->filename, s->fd_open_flags);
1075         if (s->fd < 0) {
1076             s->fd_error_time = qemu_get_clock(rt_clock);
1077             s->fd_got_error = 1;
1078             if (last_media_present)
1079                 s->fd_media_changed = 1;
1080 #ifdef DEBUG_FLOPPY
1081             printf("No floppy\n");
1082 #endif
1083             return -EIO;
1084         }
1085 #ifdef DEBUG_FLOPPY
1086         printf("Floppy opened\n");
1087 #endif
1088     }
1089     if (!last_media_present)
1090         s->fd_media_changed = 1;
1091     s->fd_open_time = qemu_get_clock(rt_clock);
1092     s->fd_got_error = 0;
1093     return 0;
1094 }
1095
1096 static int raw_is_inserted(BlockDriverState *bs)
1097 {
1098     BDRVRawState *s = bs->opaque;
1099     int ret;
1100
1101     switch(s->type) {
1102     case FTYPE_CD:
1103         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1104         if (ret == CDS_DISC_OK)
1105             return 1;
1106         else
1107             return 0;
1108         break;
1109     case FTYPE_FD:
1110         ret = fd_open(bs);
1111         return (ret >= 0);
1112     default:
1113         return 1;
1114     }
1115 }
1116
1117 /* currently only used by fdc.c, but a CD version would be good too */
1118 static int raw_media_changed(BlockDriverState *bs)
1119 {
1120     BDRVRawState *s = bs->opaque;
1121
1122     switch(s->type) {
1123     case FTYPE_FD:
1124         {
1125             int ret;
1126             /* XXX: we do not have a true media changed indication. It
1127                does not work if the floppy is changed without trying
1128                to read it */
1129             fd_open(bs);
1130             ret = s->fd_media_changed;
1131             s->fd_media_changed = 0;
1132 #ifdef DEBUG_FLOPPY
1133             printf("Floppy changed=%d\n", ret);
1134 #endif
1135             return ret;
1136         }
1137     default:
1138         return -ENOTSUP;
1139     }
1140 }
1141
1142 static int raw_eject(BlockDriverState *bs, int eject_flag)
1143 {
1144     BDRVRawState *s = bs->opaque;
1145
1146     switch(s->type) {
1147     case FTYPE_CD:
1148         if (eject_flag) {
1149             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1150                 perror("CDROMEJECT");
1151         } else {
1152             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1153                 perror("CDROMEJECT");
1154         }
1155         break;
1156     case FTYPE_FD:
1157         {
1158             int fd;
1159             if (s->fd >= 0) {
1160                 close(s->fd);
1161                 s->fd = -1;
1162                 raw_close_fd_pool(s);
1163             }
1164             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1165             if (fd >= 0) {
1166                 if (ioctl(fd, FDEJECT, 0) < 0)
1167                     perror("FDEJECT");
1168                 close(fd);
1169             }
1170         }
1171         break;
1172     default:
1173         return -ENOTSUP;
1174     }
1175     return 0;
1176 }
1177
1178 static int raw_set_locked(BlockDriverState *bs, int locked)
1179 {
1180     BDRVRawState *s = bs->opaque;
1181
1182     switch(s->type) {
1183     case FTYPE_CD:
1184         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1185             /* Note: an error can happen if the distribution automatically
1186                mounts the CD-ROM */
1187             //        perror("CDROM_LOCKDOOR");
1188         }
1189         break;
1190     default:
1191         return -ENOTSUP;
1192     }
1193     return 0;
1194 }
1195
1196 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1197 {
1198     BDRVRawState *s = bs->opaque;
1199
1200     return ioctl(s->fd, req, buf);
1201 }
1202 #else
1203
1204 static int fd_open(BlockDriverState *bs)
1205 {
1206     return 0;
1207 }
1208
1209 static int raw_is_inserted(BlockDriverState *bs)
1210 {
1211     return 1;
1212 }
1213
1214 static int raw_media_changed(BlockDriverState *bs)
1215 {
1216     return -ENOTSUP;
1217 }
1218
1219 static int raw_eject(BlockDriverState *bs, int eject_flag)
1220 {
1221     return -ENOTSUP;
1222 }
1223
1224 static int raw_set_locked(BlockDriverState *bs, int locked)
1225 {
1226     return -ENOTSUP;
1227 }
1228
1229 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1230 {
1231     return -ENOTSUP;
1232 }
1233 #endif /* !linux */
1234
1235 BlockDriver bdrv_host_device = {
1236     "host_device",
1237     sizeof(BDRVRawState),
1238     NULL, /* no probe for protocols */
1239     hdev_open,
1240     NULL,
1241     NULL,
1242     raw_close,
1243     NULL,
1244     raw_flush,
1245
1246 #ifdef CONFIG_AIO
1247     .bdrv_aio_read = raw_aio_read,
1248     .bdrv_aio_write = raw_aio_write,
1249     .bdrv_aio_cancel = raw_aio_cancel,
1250     .aiocb_size = sizeof(RawAIOCB),
1251 #endif
1252     .bdrv_pread = raw_pread,
1253     .bdrv_pwrite = raw_pwrite,
1254     .bdrv_getlength = raw_getlength,
1255
1256     /* removable device support */
1257     .bdrv_is_inserted = raw_is_inserted,
1258     .bdrv_media_changed = raw_media_changed,
1259     .bdrv_eject = raw_eject,
1260     .bdrv_set_locked = raw_set_locked,
1261     /* generic scsi device */
1262     .bdrv_ioctl = raw_ioctl,
1263 };