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.ev_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 void raw_aio_remove(RawAIOCB *acb)
 600 {
 601     RawAIOCB **pacb;
 602
 603     /* remove the callback from the queue */
 604     pacb = &posix_aio_state->first_aio;
 605     for(;;) {
 606         if (*pacb == NULL) {
 607             break;
 608         } else if (*pacb == acb) {
 609             *pacb = acb->next;
 610             qemu_aio_release(acb);
 611             break;
 612         }
 613         pacb = &acb->next;
 614     }
 615 }
 616
 617 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 618         int64_t sector_num, uint8_t *buf, int nb_sectors,
 619         BlockDriverCompletionFunc *cb, void *opaque)
 620 {
 621     RawAIOCB *acb;
 622
 623     /*
 624      * If O_DIRECT is used and the buffer is not aligned fall back
 625      * to synchronous IO.
 626      */
 627     BDRVRawState *s = bs->opaque;
 628
 629     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 630         QEMUBH *bh;
 631         acb = qemu_aio_get(bs, cb, opaque);
 632         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 633         bh = qemu_bh_new(raw_aio_em_cb, acb);
 634         qemu_bh_schedule(bh);
 635         return &acb->common;
 636     }
 637
 638     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 639     if (!acb)
 640         return NULL;
 641     if (qemu_paio_read(&acb->aiocb) < 0) {
 642         raw_aio_remove(acb);
 643         return NULL;
 644     }
 645     return &acb->common;
 646 }
 647
 648 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 649         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 650         BlockDriverCompletionFunc *cb, void *opaque)
 651 {
 652     RawAIOCB *acb;
 653
 654     /*
 655      * If O_DIRECT is used and the buffer is not aligned fall back
 656      * to synchronous IO.
 657      */
 658     BDRVRawState *s = bs->opaque;
 659
 660     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 661         QEMUBH *bh;
 662         acb = qemu_aio_get(bs, cb, opaque);
 663         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 664         bh = qemu_bh_new(raw_aio_em_cb, acb);
 665         qemu_bh_schedule(bh);
 666         return &acb->common;
 667     }
 668
 669     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 670     if (!acb)
 671         return NULL;
 672     if (qemu_paio_write(&acb->aiocb) < 0) {
 673         raw_aio_remove(acb);
 674         return NULL;
 675     }
 676     return &acb->common;
 677 }
 678
 679 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 680 {
 681     int ret;
 682     RawAIOCB *acb = (RawAIOCB *)blockacb;
 683
 684     ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 685     if (ret == QEMU_PAIO_NOTCANCELED) {
 686         /* fail safe: if the aio could not be canceled, we wait for
 687            it */
 688         while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
 689     }
 690
 691     raw_aio_remove(acb);
 692 }
 693 #else /* CONFIG_AIO */
 694 static int posix_aio_init(void)
 695 {
 696     return 0;
 697 }
 698 #endif /* CONFIG_AIO */
 699
 700
 701 static void raw_close(BlockDriverState *bs)
 702 {
 703     BDRVRawState *s = bs->opaque;
 704     if (s->fd >= 0) {
 705         close(s->fd);
 706         s->fd = -1;
 707         if (s->aligned_buf != NULL)
 708             qemu_free(s->aligned_buf);
 709     }
 710 }
 711
 712 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 713 {
 714     BDRVRawState *s = bs->opaque;
 715     if (s->type != FTYPE_FILE)
 716         return -ENOTSUP;
 717     if (ftruncate(s->fd, offset) < 0)
 718         return -errno;
 719     return 0;
 720 }
 721
 722 #ifdef __OpenBSD__
 723 static int64_t raw_getlength(BlockDriverState *bs)
 724 {
 725     BDRVRawState *s = bs->opaque;
 726     int fd = s->fd;
 727     struct stat st;
 728
 729     if (fstat(fd, &st))
 730         return -1;
 731     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 732         struct disklabel dl;
 733
 734         if (ioctl(fd, DIOCGDINFO, &dl))
 735             return -1;
 736         return (uint64_t)dl.d_secsize *
 737             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 738     } else
 739         return st.st_size;
 740 }
 741 #else /* !__OpenBSD__ */
 742 static int64_t  raw_getlength(BlockDriverState *bs)
 743 {
 744     BDRVRawState *s = bs->opaque;
 745     int fd = s->fd;
 746     int64_t size;
 747 #ifdef _BSD
 748     struct stat sb;
 749 #endif
 750 #ifdef __sun__
 751     struct dk_minfo minfo;
 752     int rv;
 753 #endif
 754     int ret;
 755
 756     ret = fd_open(bs);
 757     if (ret < 0)
 758         return ret;
 759
 760 #ifdef _BSD
 761     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 762 #ifdef DIOCGMEDIASIZE
 763         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 764 #endif
 765 #ifdef CONFIG_COCOA
 766         size = LONG_LONG_MAX;
 767 #else
 768         size = lseek(fd, 0LL, SEEK_END);
 769 #endif
 770     } else
 771 #endif
 772 #ifdef __sun__
 773     /*
 774      * use the DKIOCGMEDIAINFO ioctl to read the size.
 775      */
 776     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 777     if ( rv != -1 ) {
 778         size = minfo.dki_lbsize * minfo.dki_capacity;
 779     } else /* there are reports that lseek on some devices
 780               fails, but irc discussion said that contingency
 781               on contingency was overkill */
 782 #endif
 783     {
 784         size = lseek(fd, 0, SEEK_END);
 785     }
 786     return size;
 787 }
 788 #endif
 789
 790 static int raw_create(const char *filename, int64_t total_size,
 791                       const char *backing_file, int flags)
 792 {
 793     int fd;
 794
 795     if (flags || backing_file)
 796         return -ENOTSUP;
 797
 798     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 799               0644);
 800     if (fd < 0)
 801         return -EIO;
 802     ftruncate(fd, total_size * 512);
 803     close(fd);
 804     return 0;
 805 }
 806
 807 static void raw_flush(BlockDriverState *bs)
 808 {
 809     BDRVRawState *s = bs->opaque;
 810     fsync(s->fd);
 811 }
 812
 813 BlockDriver bdrv_raw = {
 814     "raw",
 815     sizeof(BDRVRawState),
 816     NULL, /* no probe for protocols */
 817     raw_open,
 818     NULL,
 819     NULL,
 820     raw_close,
 821     raw_create,
 822     raw_flush,
 823
 824 #ifdef CONFIG_AIO
 825     .bdrv_aio_read = raw_aio_read,
 826     .bdrv_aio_write = raw_aio_write,
 827     .bdrv_aio_cancel = raw_aio_cancel,
 828     .aiocb_size = sizeof(RawAIOCB),
 829 #endif
 830
 831     .bdrv_pread = raw_pread,
 832     .bdrv_pwrite = raw_pwrite,
 833     .bdrv_truncate = raw_truncate,
 834     .bdrv_getlength = raw_getlength,
 835 };
 836
 837 /***********************************************/
 838 /* host device */
 839
 840 #ifdef CONFIG_COCOA
 841 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 842 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 843
 844 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 845 {
 846     kern_return_t       kernResult;
 847     mach_port_t     masterPort;
 848     CFMutableDictionaryRef  classesToMatch;
 849
 850     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 851     if ( KERN_SUCCESS != kernResult ) {
 852         printf( "IOMasterPort returned %d\n", kernResult );
 853     }
 854
 855     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 856     if ( classesToMatch == NULL ) {
 857         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 858     } else {
 859     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 860     }
 861     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 862     if ( KERN_SUCCESS != kernResult )
 863     {
 864         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 865     }
 866
 867     return kernResult;
 868 }
 869
 870 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 871 {
 872     io_object_t     nextMedia;
 873     kern_return_t   kernResult = KERN_FAILURE;
 874     *bsdPath = '\0';
 875     nextMedia = IOIteratorNext( mediaIterator );
 876     if ( nextMedia )
 877     {
 878         CFTypeRef   bsdPathAsCFString;
 879     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 880         if ( bsdPathAsCFString ) {
 881             size_t devPathLength;
 882             strcpy( bsdPath, _PATH_DEV );
 883             strcat( bsdPath, "r" );
 884             devPathLength = strlen( bsdPath );
 885             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 886                 kernResult = KERN_SUCCESS;
 887             }
 888             CFRelease( bsdPathAsCFString );
 889         }
 890         IOObjectRelease( nextMedia );
 891     }
 892
 893     return kernResult;
 894 }
 895
 896 #endif
 897
 898 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 899 {
 900     BDRVRawState *s = bs->opaque;
 901     int fd, open_flags, ret;
 902
 903     posix_aio_init();
 904
 905 #ifdef CONFIG_COCOA
 906     if (strstart(filename, "/dev/cdrom", NULL)) {
 907         kern_return_t kernResult;
 908         io_iterator_t mediaIterator;
 909         char bsdPath[ MAXPATHLEN ];
 910         int fd;
 911
 912         kernResult = FindEjectableCDMedia( &mediaIterator );
 913         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 914
 915         if ( bsdPath[ 0 ] != '\0' ) {
 916             strcat(bsdPath,"s0");
 917             /* some CDs don't have a partition 0 */
 918             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 919             if (fd < 0) {
 920                 bsdPath[strlen(bsdPath)-1] = '1';
 921             } else {
 922                 close(fd);
 923             }
 924             filename = bsdPath;
 925         }
 926
 927         if ( mediaIterator )
 928             IOObjectRelease( mediaIterator );
 929     }
 930 #endif
 931     open_flags = O_BINARY;
 932     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 933         open_flags |= O_RDWR;
 934     } else {
 935         open_flags |= O_RDONLY;
 936         bs->read_only = 1;
 937     }
 938     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 939      * and O_DIRECT for no caching. */
 940     if ((flags & BDRV_O_NOCACHE))
 941         open_flags |= O_DIRECT;
 942     else if (!(flags & BDRV_O_CACHE_WB))
 943         open_flags |= O_DSYNC;
 944
 945     s->type = FTYPE_FILE;
 946 #if defined(__linux__)
 947     if (strstart(filename, "/dev/cd", NULL)) {
 948         /* open will not fail even if no CD is inserted */
 949         open_flags |= O_NONBLOCK;
 950         s->type = FTYPE_CD;
 951     } else if (strstart(filename, "/dev/fd", NULL)) {
 952         s->type = FTYPE_FD;
 953         s->fd_open_flags = open_flags;
 954         /* open will not fail even if no floppy is inserted */
 955         open_flags |= O_NONBLOCK;
 956     } else if (strstart(filename, "/dev/sg", NULL)) {
 957         bs->sg = 1;
 958     }
 959 #endif
 960     fd = open(filename, open_flags, 0644);
 961     if (fd < 0) {
 962         ret = -errno;
 963         if (ret == -EROFS)
 964             ret = -EACCES;
 965         return ret;
 966     }
 967     s->fd = fd;
 968 #if defined(__linux__)
 969     /* close fd so that we can reopen it as needed */
 970     if (s->type == FTYPE_FD) {
 971         close(s->fd);
 972         s->fd = -1;
 973         s->fd_media_changed = 1;
 974     }
 975 #endif
 976     return 0;
 977 }
 978
 979 #if defined(__linux__)
 980 /* Note: we do not have a reliable method to detect if the floppy is
 981    present. The current method is to try to open the floppy at every
 982    I/O and to keep it opened during a few hundreds of ms. */
 983 static int fd_open(BlockDriverState *bs)
 984 {
 985     BDRVRawState *s = bs->opaque;
 986     int last_media_present;
 987
 988     if (s->type != FTYPE_FD)
 989         return 0;
 990     last_media_present = (s->fd >= 0);
 991     if (s->fd >= 0 &&
 992         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
 993         close(s->fd);
 994         s->fd = -1;
 995 #ifdef DEBUG_FLOPPY
 996         printf("Floppy closed\n");
 997 #endif
 998     }
 999     if (s->fd < 0) {
1000         if (s->fd_got_error &&
1001             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1002 #ifdef DEBUG_FLOPPY
1003             printf("No floppy (open delayed)\n");
1004 #endif
1005             return -EIO;
1006         }
1007         s->fd = open(bs->filename, s->fd_open_flags);
1008         if (s->fd < 0) {
1009             s->fd_error_time = qemu_get_clock(rt_clock);
1010             s->fd_got_error = 1;
1011             if (last_media_present)
1012                 s->fd_media_changed = 1;
1013 #ifdef DEBUG_FLOPPY
1014             printf("No floppy\n");
1015 #endif
1016             return -EIO;
1017         }
1018 #ifdef DEBUG_FLOPPY
1019         printf("Floppy opened\n");
1020 #endif
1021     }
1022     if (!last_media_present)
1023         s->fd_media_changed = 1;
1024     s->fd_open_time = qemu_get_clock(rt_clock);
1025     s->fd_got_error = 0;
1026     return 0;
1027 }
1028
1029 static int raw_is_inserted(BlockDriverState *bs)
1030 {
1031     BDRVRawState *s = bs->opaque;
1032     int ret;
1033
1034     switch(s->type) {
1035     case FTYPE_CD:
1036         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1037         if (ret == CDS_DISC_OK)
1038             return 1;
1039         else
1040             return 0;
1041         break;
1042     case FTYPE_FD:
1043         ret = fd_open(bs);
1044         return (ret >= 0);
1045     default:
1046         return 1;
1047     }
1048 }
1049
1050 /* currently only used by fdc.c, but a CD version would be good too */
1051 static int raw_media_changed(BlockDriverState *bs)
1052 {
1053     BDRVRawState *s = bs->opaque;
1054
1055     switch(s->type) {
1056     case FTYPE_FD:
1057         {
1058             int ret;
1059             /* XXX: we do not have a true media changed indication. It
1060                does not work if the floppy is changed without trying
1061                to read it */
1062             fd_open(bs);
1063             ret = s->fd_media_changed;
1064             s->fd_media_changed = 0;
1065 #ifdef DEBUG_FLOPPY
1066             printf("Floppy changed=%d\n", ret);
1067 #endif
1068             return ret;
1069         }
1070     default:
1071         return -ENOTSUP;
1072     }
1073 }
1074
1075 static int raw_eject(BlockDriverState *bs, int eject_flag)
1076 {
1077     BDRVRawState *s = bs->opaque;
1078
1079     switch(s->type) {
1080     case FTYPE_CD:
1081         if (eject_flag) {
1082             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1083                 perror("CDROMEJECT");
1084         } else {
1085             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1086                 perror("CDROMEJECT");
1087         }
1088         break;
1089     case FTYPE_FD:
1090         {
1091             int fd;
1092             if (s->fd >= 0) {
1093                 close(s->fd);
1094                 s->fd = -1;
1095             }
1096             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1097             if (fd >= 0) {
1098                 if (ioctl(fd, FDEJECT, 0) < 0)
1099                     perror("FDEJECT");
1100                 close(fd);
1101             }
1102         }
1103         break;
1104     default:
1105         return -ENOTSUP;
1106     }
1107     return 0;
1108 }
1109
1110 static int raw_set_locked(BlockDriverState *bs, int locked)
1111 {
1112     BDRVRawState *s = bs->opaque;
1113
1114     switch(s->type) {
1115     case FTYPE_CD:
1116         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1117             /* Note: an error can happen if the distribution automatically
1118                mounts the CD-ROM */
1119             //        perror("CDROM_LOCKDOOR");
1120         }
1121         break;
1122     default:
1123         return -ENOTSUP;
1124     }
1125     return 0;
1126 }
1127
1128 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1129 {
1130     BDRVRawState *s = bs->opaque;
1131
1132     return ioctl(s->fd, req, buf);
1133 }
1134 #else
1135
1136 static int fd_open(BlockDriverState *bs)
1137 {
1138     return 0;
1139 }
1140
1141 static int raw_is_inserted(BlockDriverState *bs)
1142 {
1143     return 1;
1144 }
1145
1146 static int raw_media_changed(BlockDriverState *bs)
1147 {
1148     return -ENOTSUP;
1149 }
1150
1151 static int raw_eject(BlockDriverState *bs, int eject_flag)
1152 {
1153     return -ENOTSUP;
1154 }
1155
1156 static int raw_set_locked(BlockDriverState *bs, int locked)
1157 {
1158     return -ENOTSUP;
1159 }
1160
1161 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1162 {
1163     return -ENOTSUP;
1164 }
1165 #endif /* !linux */
1166
1167 BlockDriver bdrv_host_device = {
1168     "host_device",
1169     sizeof(BDRVRawState),
1170     NULL, /* no probe for protocols */
1171     hdev_open,
1172     NULL,
1173     NULL,
1174     raw_close,
1175     NULL,
1176     raw_flush,
1177
1178 #ifdef CONFIG_AIO
1179     .bdrv_aio_read = raw_aio_read,
1180     .bdrv_aio_write = raw_aio_write,
1181     .bdrv_aio_cancel = raw_aio_cancel,
1182     .aiocb_size = sizeof(RawAIOCB),
1183 #endif
1184
1185     .bdrv_pread = raw_pread,
1186     .bdrv_pwrite = raw_pwrite,
1187     .bdrv_getlength = raw_getlength,
1188
1189     /* removable device support */
1190     .bdrv_is_inserted = raw_is_inserted,
1191     .bdrv_media_changed = raw_media_changed,
1192     .bdrv_eject = raw_eject,
1193     .bdrv_set_locked = raw_set_locked,
1194     /* generic scsi device */
1195     .bdrv_ioctl = raw_ioctl,
1196 };