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     struct qemu_paioinit ai;
 531
 532     if (posix_aio_state)
 533         return 0;
 534
 535     s = qemu_malloc(sizeof(PosixAioState));
 536
 537     /* Make sure to block AIO signal */
 538     sigemptyset(&mask);
 539     sigaddset(&mask, SIGUSR2);
 540     sigprocmask(SIG_BLOCK, &mask, NULL);
 541
 542     s->first_aio = NULL;
 543     s->fd = qemu_signalfd(&mask);
 544     if (s->fd == -1) {
 545         fprintf(stderr, "failed to create signalfd\n");
 546         return -errno;
 547     }
 548
 549     fcntl(s->fd, F_SETFL, O_NONBLOCK);
 550
 551     qemu_aio_set_fd_handler(s->fd, posix_aio_read, NULL, posix_aio_flush, s);
 552
 553     memset(&ai, 0, sizeof(ai));
 554     ai.aio_threads = 64;
 555     ai.aio_num = 64;
 556     qemu_paio_init(&ai);
 557
 558     posix_aio_state = s;
 559
 560     return 0;
 561 }
 562
 563 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 564         int64_t sector_num, uint8_t *buf, int nb_sectors,
 565         BlockDriverCompletionFunc *cb, void *opaque)
 566 {
 567     BDRVRawState *s = bs->opaque;
 568     RawAIOCB *acb;
 569
 570     if (fd_open(bs) < 0)
 571         return NULL;
 572
 573     acb = qemu_aio_get(bs, cb, opaque);
 574     if (!acb)
 575         return NULL;
 576     acb->aiocb.aio_fildes = s->fd;
 577     acb->aiocb.ev_signo = SIGUSR2;
 578     acb->aiocb.aio_buf = buf;
 579     if (nb_sectors < 0)
 580         acb->aiocb.aio_nbytes = -nb_sectors;
 581     else
 582         acb->aiocb.aio_nbytes = nb_sectors * 512;
 583     acb->aiocb.aio_offset = sector_num * 512;
 584     acb->next = posix_aio_state->first_aio;
 585     posix_aio_state->first_aio = acb;
 586     return acb;
 587 }
 588
 589 static void raw_aio_em_cb(void* opaque)
 590 {
 591     RawAIOCB *acb = opaque;
 592     acb->common.cb(acb->common.opaque, acb->ret);
 593     qemu_aio_release(acb);
 594 }
 595
 596 static void raw_aio_remove(RawAIOCB *acb)
 597 {
 598     RawAIOCB **pacb;
 599
 600     /* remove the callback from the queue */
 601     pacb = &posix_aio_state->first_aio;
 602     for(;;) {
 603         if (*pacb == NULL) {
 604             fprintf(stderr, "raw_aio_remove: aio request not found!\n");
 605             break;
 606         } else if (*pacb == acb) {
 607             *pacb = acb->next;
 608             qemu_aio_release(acb);
 609             break;
 610         }
 611         pacb = &(*pacb)->next;
 612     }
 613 }
 614
 615 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 616         int64_t sector_num, uint8_t *buf, int nb_sectors,
 617         BlockDriverCompletionFunc *cb, void *opaque)
 618 {
 619     RawAIOCB *acb;
 620
 621     /*
 622      * If O_DIRECT is used and the buffer is not aligned fall back
 623      * to synchronous IO.
 624      */
 625     BDRVRawState *s = bs->opaque;
 626
 627     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 628         QEMUBH *bh;
 629         acb = qemu_aio_get(bs, cb, opaque);
 630         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 631         bh = qemu_bh_new(raw_aio_em_cb, acb);
 632         qemu_bh_schedule(bh);
 633         return &acb->common;
 634     }
 635
 636     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 637     if (!acb)
 638         return NULL;
 639     if (qemu_paio_read(&acb->aiocb) < 0) {
 640         raw_aio_remove(acb);
 641         return NULL;
 642     }
 643     return &acb->common;
 644 }
 645
 646 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 647         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 648         BlockDriverCompletionFunc *cb, void *opaque)
 649 {
 650     RawAIOCB *acb;
 651
 652     /*
 653      * If O_DIRECT is used and the buffer is not aligned fall back
 654      * to synchronous IO.
 655      */
 656     BDRVRawState *s = bs->opaque;
 657
 658     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 659         QEMUBH *bh;
 660         acb = qemu_aio_get(bs, cb, opaque);
 661         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 662         bh = qemu_bh_new(raw_aio_em_cb, acb);
 663         qemu_bh_schedule(bh);
 664         return &acb->common;
 665     }
 666
 667     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 668     if (!acb)
 669         return NULL;
 670     if (qemu_paio_write(&acb->aiocb) < 0) {
 671         raw_aio_remove(acb);
 672         return NULL;
 673     }
 674     return &acb->common;
 675 }
 676
 677 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 678 {
 679     int ret;
 680     RawAIOCB *acb = (RawAIOCB *)blockacb;
 681
 682     ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 683     if (ret == QEMU_PAIO_NOTCANCELED) {
 684         /* fail safe: if the aio could not be canceled, we wait for
 685            it */
 686         while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
 687     }
 688
 689     raw_aio_remove(acb);
 690 }
 691 #else /* CONFIG_AIO */
 692 static int posix_aio_init(void)
 693 {
 694     return 0;
 695 }
 696 #endif /* CONFIG_AIO */
 697
 698
 699 static void raw_close(BlockDriverState *bs)
 700 {
 701     BDRVRawState *s = bs->opaque;
 702     if (s->fd >= 0) {
 703         close(s->fd);
 704         s->fd = -1;
 705         if (s->aligned_buf != NULL)
 706             qemu_free(s->aligned_buf);
 707     }
 708 }
 709
 710 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 711 {
 712     BDRVRawState *s = bs->opaque;
 713     if (s->type != FTYPE_FILE)
 714         return -ENOTSUP;
 715     if (ftruncate(s->fd, offset) < 0)
 716         return -errno;
 717     return 0;
 718 }
 719
 720 #ifdef __OpenBSD__
 721 static int64_t raw_getlength(BlockDriverState *bs)
 722 {
 723     BDRVRawState *s = bs->opaque;
 724     int fd = s->fd;
 725     struct stat st;
 726
 727     if (fstat(fd, &st))
 728         return -1;
 729     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 730         struct disklabel dl;
 731
 732         if (ioctl(fd, DIOCGDINFO, &dl))
 733             return -1;
 734         return (uint64_t)dl.d_secsize *
 735             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 736     } else
 737         return st.st_size;
 738 }
 739 #else /* !__OpenBSD__ */
 740 static int64_t  raw_getlength(BlockDriverState *bs)
 741 {
 742     BDRVRawState *s = bs->opaque;
 743     int fd = s->fd;
 744     int64_t size;
 745 #ifdef _BSD
 746     struct stat sb;
 747 #endif
 748 #ifdef __sun__
 749     struct dk_minfo minfo;
 750     int rv;
 751 #endif
 752     int ret;
 753
 754     ret = fd_open(bs);
 755     if (ret < 0)
 756         return ret;
 757
 758 #ifdef _BSD
 759     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 760 #ifdef DIOCGMEDIASIZE
 761         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 762 #endif
 763 #ifdef CONFIG_COCOA
 764         size = LONG_LONG_MAX;
 765 #else
 766         size = lseek(fd, 0LL, SEEK_END);
 767 #endif
 768     } else
 769 #endif
 770 #ifdef __sun__
 771     /*
 772      * use the DKIOCGMEDIAINFO ioctl to read the size.
 773      */
 774     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 775     if ( rv != -1 ) {
 776         size = minfo.dki_lbsize * minfo.dki_capacity;
 777     } else /* there are reports that lseek on some devices
 778               fails, but irc discussion said that contingency
 779               on contingency was overkill */
 780 #endif
 781     {
 782         size = lseek(fd, 0, SEEK_END);
 783     }
 784     return size;
 785 }
 786 #endif
 787
 788 static int raw_create(const char *filename, int64_t total_size,
 789                       const char *backing_file, int flags)
 790 {
 791     int fd;
 792
 793     if (flags || backing_file)
 794         return -ENOTSUP;
 795
 796     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 797               0644);
 798     if (fd < 0)
 799         return -EIO;
 800     ftruncate(fd, total_size * 512);
 801     close(fd);
 802     return 0;
 803 }
 804
 805 static void raw_flush(BlockDriverState *bs)
 806 {
 807     BDRVRawState *s = bs->opaque;
 808     fsync(s->fd);
 809 }
 810
 811 BlockDriver bdrv_raw = {
 812     "raw",
 813     sizeof(BDRVRawState),
 814     NULL, /* no probe for protocols */
 815     raw_open,
 816     NULL,
 817     NULL,
 818     raw_close,
 819     raw_create,
 820     raw_flush,
 821
 822 #ifdef CONFIG_AIO
 823     .bdrv_aio_read = raw_aio_read,
 824     .bdrv_aio_write = raw_aio_write,
 825     .bdrv_aio_cancel = raw_aio_cancel,
 826     .aiocb_size = sizeof(RawAIOCB),
 827 #endif
 828
 829     .bdrv_pread = raw_pread,
 830     .bdrv_pwrite = raw_pwrite,
 831     .bdrv_truncate = raw_truncate,
 832     .bdrv_getlength = raw_getlength,
 833 };
 834
 835 /***********************************************/
 836 /* host device */
 837
 838 #ifdef CONFIG_COCOA
 839 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 840 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 841
 842 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 843 {
 844     kern_return_t       kernResult;
 845     mach_port_t     masterPort;
 846     CFMutableDictionaryRef  classesToMatch;
 847
 848     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 849     if ( KERN_SUCCESS != kernResult ) {
 850         printf( "IOMasterPort returned %d\n", kernResult );
 851     }
 852
 853     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 854     if ( classesToMatch == NULL ) {
 855         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 856     } else {
 857     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 858     }
 859     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 860     if ( KERN_SUCCESS != kernResult )
 861     {
 862         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 863     }
 864
 865     return kernResult;
 866 }
 867
 868 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 869 {
 870     io_object_t     nextMedia;
 871     kern_return_t   kernResult = KERN_FAILURE;
 872     *bsdPath = '\0';
 873     nextMedia = IOIteratorNext( mediaIterator );
 874     if ( nextMedia )
 875     {
 876         CFTypeRef   bsdPathAsCFString;
 877     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 878         if ( bsdPathAsCFString ) {
 879             size_t devPathLength;
 880             strcpy( bsdPath, _PATH_DEV );
 881             strcat( bsdPath, "r" );
 882             devPathLength = strlen( bsdPath );
 883             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 884                 kernResult = KERN_SUCCESS;
 885             }
 886             CFRelease( bsdPathAsCFString );
 887         }
 888         IOObjectRelease( nextMedia );
 889     }
 890
 891     return kernResult;
 892 }
 893
 894 #endif
 895
 896 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 897 {
 898     BDRVRawState *s = bs->opaque;
 899     int fd, open_flags, ret;
 900
 901     posix_aio_init();
 902
 903 #ifdef CONFIG_COCOA
 904     if (strstart(filename, "/dev/cdrom", NULL)) {
 905         kern_return_t kernResult;
 906         io_iterator_t mediaIterator;
 907         char bsdPath[ MAXPATHLEN ];
 908         int fd;
 909
 910         kernResult = FindEjectableCDMedia( &mediaIterator );
 911         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 912
 913         if ( bsdPath[ 0 ] != '\0' ) {
 914             strcat(bsdPath,"s0");
 915             /* some CDs don't have a partition 0 */
 916             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 917             if (fd < 0) {
 918                 bsdPath[strlen(bsdPath)-1] = '1';
 919             } else {
 920                 close(fd);
 921             }
 922             filename = bsdPath;
 923         }
 924
 925         if ( mediaIterator )
 926             IOObjectRelease( mediaIterator );
 927     }
 928 #endif
 929     open_flags = O_BINARY;
 930     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 931         open_flags |= O_RDWR;
 932     } else {
 933         open_flags |= O_RDONLY;
 934         bs->read_only = 1;
 935     }
 936     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 937      * and O_DIRECT for no caching. */
 938     if ((flags & BDRV_O_NOCACHE))
 939         open_flags |= O_DIRECT;
 940     else if (!(flags & BDRV_O_CACHE_WB))
 941         open_flags |= O_DSYNC;
 942
 943     s->type = FTYPE_FILE;
 944 #if defined(__linux__)
 945     if (strstart(filename, "/dev/cd", NULL)) {
 946         /* open will not fail even if no CD is inserted */
 947         open_flags |= O_NONBLOCK;
 948         s->type = FTYPE_CD;
 949     } else if (strstart(filename, "/dev/fd", NULL)) {
 950         s->type = FTYPE_FD;
 951         s->fd_open_flags = open_flags;
 952         /* open will not fail even if no floppy is inserted */
 953         open_flags |= O_NONBLOCK;
 954     } else if (strstart(filename, "/dev/sg", NULL)) {
 955         bs->sg = 1;
 956     }
 957 #endif
 958     fd = open(filename, open_flags, 0644);
 959     if (fd < 0) {
 960         ret = -errno;
 961         if (ret == -EROFS)
 962             ret = -EACCES;
 963         return ret;
 964     }
 965     s->fd = fd;
 966 #if defined(__linux__)
 967     /* close fd so that we can reopen it as needed */
 968     if (s->type == FTYPE_FD) {
 969         close(s->fd);
 970         s->fd = -1;
 971         s->fd_media_changed = 1;
 972     }
 973 #endif
 974     return 0;
 975 }
 976
 977 #if defined(__linux__)
 978 /* Note: we do not have a reliable method to detect if the floppy is
 979    present. The current method is to try to open the floppy at every
 980    I/O and to keep it opened during a few hundreds of ms. */
 981 static int fd_open(BlockDriverState *bs)
 982 {
 983     BDRVRawState *s = bs->opaque;
 984     int last_media_present;
 985
 986     if (s->type != FTYPE_FD)
 987         return 0;
 988     last_media_present = (s->fd >= 0);
 989     if (s->fd >= 0 &&
 990         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
 991         close(s->fd);
 992         s->fd = -1;
 993 #ifdef DEBUG_FLOPPY
 994         printf("Floppy closed\n");
 995 #endif
 996     }
 997     if (s->fd < 0) {
 998         if (s->fd_got_error &&
 999             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1000 #ifdef DEBUG_FLOPPY
1001             printf("No floppy (open delayed)\n");
1002 #endif
1003             return -EIO;
1004         }
1005         s->fd = open(bs->filename, s->fd_open_flags);
1006         if (s->fd < 0) {
1007             s->fd_error_time = qemu_get_clock(rt_clock);
1008             s->fd_got_error = 1;
1009             if (last_media_present)
1010                 s->fd_media_changed = 1;
1011 #ifdef DEBUG_FLOPPY
1012             printf("No floppy\n");
1013 #endif
1014             return -EIO;
1015         }
1016 #ifdef DEBUG_FLOPPY
1017         printf("Floppy opened\n");
1018 #endif
1019     }
1020     if (!last_media_present)
1021         s->fd_media_changed = 1;
1022     s->fd_open_time = qemu_get_clock(rt_clock);
1023     s->fd_got_error = 0;
1024     return 0;
1025 }
1026
1027 static int raw_is_inserted(BlockDriverState *bs)
1028 {
1029     BDRVRawState *s = bs->opaque;
1030     int ret;
1031
1032     switch(s->type) {
1033     case FTYPE_CD:
1034         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1035         if (ret == CDS_DISC_OK)
1036             return 1;
1037         else
1038             return 0;
1039         break;
1040     case FTYPE_FD:
1041         ret = fd_open(bs);
1042         return (ret >= 0);
1043     default:
1044         return 1;
1045     }
1046 }
1047
1048 /* currently only used by fdc.c, but a CD version would be good too */
1049 static int raw_media_changed(BlockDriverState *bs)
1050 {
1051     BDRVRawState *s = bs->opaque;
1052
1053     switch(s->type) {
1054     case FTYPE_FD:
1055         {
1056             int ret;
1057             /* XXX: we do not have a true media changed indication. It
1058                does not work if the floppy is changed without trying
1059                to read it */
1060             fd_open(bs);
1061             ret = s->fd_media_changed;
1062             s->fd_media_changed = 0;
1063 #ifdef DEBUG_FLOPPY
1064             printf("Floppy changed=%d\n", ret);
1065 #endif
1066             return ret;
1067         }
1068     default:
1069         return -ENOTSUP;
1070     }
1071 }
1072
1073 static int raw_eject(BlockDriverState *bs, int eject_flag)
1074 {
1075     BDRVRawState *s = bs->opaque;
1076
1077     switch(s->type) {
1078     case FTYPE_CD:
1079         if (eject_flag) {
1080             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1081                 perror("CDROMEJECT");
1082         } else {
1083             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1084                 perror("CDROMEJECT");
1085         }
1086         break;
1087     case FTYPE_FD:
1088         {
1089             int fd;
1090             if (s->fd >= 0) {
1091                 close(s->fd);
1092                 s->fd = -1;
1093             }
1094             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1095             if (fd >= 0) {
1096                 if (ioctl(fd, FDEJECT, 0) < 0)
1097                     perror("FDEJECT");
1098                 close(fd);
1099             }
1100         }
1101         break;
1102     default:
1103         return -ENOTSUP;
1104     }
1105     return 0;
1106 }
1107
1108 static int raw_set_locked(BlockDriverState *bs, int locked)
1109 {
1110     BDRVRawState *s = bs->opaque;
1111
1112     switch(s->type) {
1113     case FTYPE_CD:
1114         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1115             /* Note: an error can happen if the distribution automatically
1116                mounts the CD-ROM */
1117             //        perror("CDROM_LOCKDOOR");
1118         }
1119         break;
1120     default:
1121         return -ENOTSUP;
1122     }
1123     return 0;
1124 }
1125
1126 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1127 {
1128     BDRVRawState *s = bs->opaque;
1129
1130     return ioctl(s->fd, req, buf);
1131 }
1132 #else
1133
1134 static int fd_open(BlockDriverState *bs)
1135 {
1136     return 0;
1137 }
1138
1139 static int raw_is_inserted(BlockDriverState *bs)
1140 {
1141     return 1;
1142 }
1143
1144 static int raw_media_changed(BlockDriverState *bs)
1145 {
1146     return -ENOTSUP;
1147 }
1148
1149 static int raw_eject(BlockDriverState *bs, int eject_flag)
1150 {
1151     return -ENOTSUP;
1152 }
1153
1154 static int raw_set_locked(BlockDriverState *bs, int locked)
1155 {
1156     return -ENOTSUP;
1157 }
1158
1159 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1160 {
1161     return -ENOTSUP;
1162 }
1163 #endif /* !linux */
1164
1165 BlockDriver bdrv_host_device = {
1166     "host_device",
1167     sizeof(BDRVRawState),
1168     NULL, /* no probe for protocols */
1169     hdev_open,
1170     NULL,
1171     NULL,
1172     raw_close,
1173     NULL,
1174     raw_flush,
1175
1176 #ifdef CONFIG_AIO
1177     .bdrv_aio_read = raw_aio_read,
1178     .bdrv_aio_write = raw_aio_write,
1179     .bdrv_aio_cancel = raw_aio_cancel,
1180     .aiocb_size = sizeof(RawAIOCB),
1181 #endif
1182
1183     .bdrv_pread = raw_pread,
1184     .bdrv_pwrite = raw_pwrite,
1185     .bdrv_getlength = raw_getlength,
1186
1187     /* removable device support */
1188     .bdrv_is_inserted = raw_is_inserted,
1189     .bdrv_media_changed = raw_media_changed,
1190     .bdrv_eject = raw_eject,
1191     .bdrv_set_locked = raw_set_locked,
1192     /* generic scsi device */
1193     .bdrv_ioctl = raw_ioctl,
1194 };