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 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
  26 #include "qemu-timer.h"
  27 #include "exec-all.h"
  28 #endif
  29 #include "block_int.h"
  30 #include <assert.h>
  31 #ifdef CONFIG_AIO
  32 #include <aio.h>
  33 #endif
  34
  35 #ifdef CONFIG_COCOA
  36 #include <paths.h>
  37 #include <sys/param.h>
  38 #include <IOKit/IOKitLib.h>
  39 #include <IOKit/IOBSD.h>
  40 #include <IOKit/storage/IOMediaBSDClient.h>
  41 #include <IOKit/storage/IOMedia.h>
  42 #include <IOKit/storage/IOCDMedia.h>
  43 //#include <IOKit/storage/IOCDTypes.h>
  44 #include <CoreFoundation/CoreFoundation.h>
  45 #endif
  46
  47 #ifdef __sun__
  48 #define _POSIX_PTHREAD_SEMANTICS 1
  49 #include <signal.h>
  50 #include <sys/dkio.h>
  51 #endif
  52 #ifdef __linux__
  53 #include <sys/ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/fd.h>
  56 #endif
  57 #ifdef __FreeBSD__
  58 #include <sys/disk.h>
  59 #endif
  60
  61 //#define DEBUG_FLOPPY
  62
  63 //#define DEBUG_BLOCK
  64 #if defined(DEBUG_BLOCK) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
  65 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (loglevel != 0)  \
  66     { fprintf(logfile, formatCstr, ##args); fflush(logfile); } } while (0)
  67 #else
  68 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  69 #endif
  70
  71 #define FTYPE_FILE   0
  72 #define FTYPE_CD     1
  73 #define FTYPE_FD     2
  74
  75 #define ALIGNED_BUFFER_SIZE (32 * 512)
  76
  77 /* if the FD is not accessed during that time (in ms), we try to
  78    reopen it to see if the disk has been changed */
  79 #define FD_OPEN_TIMEOUT 1000
  80
  81 typedef struct BDRVRawState {
  82     int fd;
  83     int type;
  84     unsigned int lseek_err_cnt;
  85 #if defined(__linux__)
  86     /* linux floppy specific */
  87     int fd_open_flags;
  88     int64_t fd_open_time;
  89     int64_t fd_error_time;
  90     int fd_got_error;
  91     int fd_media_changed;
  92 #endif
  93 #if defined(O_DIRECT) && !defined(QEMU_IMG)
  94     uint8_t* aligned_buf;
  95 #endif
  96 } BDRVRawState;
  97
  98 static int fd_open(BlockDriverState *bs);
  99
 100 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 101 {
 102     BDRVRawState *s = bs->opaque;
 103     int fd, open_flags, ret;
 104
 105     s->lseek_err_cnt = 0;
 106
 107     open_flags = O_BINARY;
 108     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 109         open_flags |= O_RDWR;
 110     } else {
 111         open_flags |= O_RDONLY;
 112         bs->read_only = 1;
 113     }
 114     if (flags & BDRV_O_CREAT)
 115         open_flags |= O_CREAT | O_TRUNC;
 116 #ifdef O_DIRECT
 117     if (flags & BDRV_O_DIRECT)
 118         open_flags |= O_DIRECT;
 119 #endif
 120
 121     s->type = FTYPE_FILE;
 122
 123     fd = open(filename, open_flags, 0644);
 124     if (fd < 0) {
 125         ret = -errno;
 126         if (ret == -EROFS)
 127             ret = -EACCES;
 128         return ret;
 129     }
 130     s->fd = fd;
 131 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 132     s->aligned_buf = NULL;
 133     if (flags & BDRV_O_DIRECT) {
 134         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 135         if (s->aligned_buf == NULL) {
 136             ret = -errno;
 137             close(fd);
 138             return ret;
 139         }
 140     }
 141 #endif
 142     return 0;
 143 }
 144
 145 /* XXX: use host sector size if necessary with:
 146 #ifdef DIOCGSECTORSIZE
 147         {
 148             unsigned int sectorsize = 512;
 149             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 150                 sectorsize > bufsize)
 151                 bufsize = sectorsize;
 152         }
 153 #endif
 154 #ifdef CONFIG_COCOA
 155         u_int32_t   blockSize = 512;
 156         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 157             bufsize = blockSize;
 158         }
 159 #endif
 160 */
 161
 162 /*
 163  * offset and count are in bytes, but must be multiples of 512 for files
 164  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 165  *
 166  * This function may be called without alignment if the caller ensures
 167  * that O_DIRECT is not in effect.
 168  */
 169 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 170                      uint8_t *buf, int count)
 171 {
 172     BDRVRawState *s = bs->opaque;
 173     int ret;
 174
 175     ret = fd_open(bs);
 176     if (ret < 0)
 177         return ret;
 178
 179     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 180         ++(s->lseek_err_cnt);
 181         if(s->lseek_err_cnt <= 10) {
 182             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 183                               "] lseek failed : %d = %s\n",
 184                               s->fd, bs->filename, offset, buf, count,
 185                               bs->total_sectors, errno, strerror(errno));
 186         }
 187         return -1;
 188     }
 189     s->lseek_err_cnt=0;
 190
 191     ret = read(s->fd, buf, count);
 192     if (ret == count)
 193         goto label__raw_read__success;
 194
 195     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 196                       "] read failed %d : %d = %s\n",
 197                       s->fd, bs->filename, offset, buf, count,
 198                       bs->total_sectors, ret, errno, strerror(errno));
 199
 200     /* Try harder for CDrom. */
 201     if (bs->type == BDRV_TYPE_CDROM) {
 202         lseek(s->fd, offset, SEEK_SET);
 203         ret = read(s->fd, buf, count);
 204         if (ret == count)
 205             goto label__raw_read__success;
 206         lseek(s->fd, offset, SEEK_SET);
 207         ret = read(s->fd, buf, count);
 208         if (ret == count)
 209             goto label__raw_read__success;
 210
 211         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 212                           "] retry read failed %d : %d = %s\n",
 213                           s->fd, bs->filename, offset, buf, count,
 214                           bs->total_sectors, ret, errno, strerror(errno));
 215     }
 216
 217 label__raw_read__success:
 218
 219     return ret;
 220 }
 221
 222 /*
 223  * offset and count are in bytes, but must be multiples of 512 for files
 224  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 225  *
 226  * This function may be called without alignment if the caller ensures
 227  * that O_DIRECT is not in effect.
 228  */
 229 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 230                       const uint8_t *buf, int count)
 231 {
 232     BDRVRawState *s = bs->opaque;
 233     int ret;
 234
 235     ret = fd_open(bs);
 236     if (ret < 0)
 237         return ret;
 238
 239     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 240         ++(s->lseek_err_cnt);
 241         if(s->lseek_err_cnt) {
 242             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 243                               PRId64 "] lseek failed : %d = %s\n",
 244                               s->fd, bs->filename, offset, buf, count,
 245                               bs->total_sectors, errno, strerror(errno));
 246         }
 247         return -1;
 248     }
 249     s->lseek_err_cnt = 0;
 250
 251     ret = write(s->fd, buf, count);
 252     if (ret == count)
 253         goto label__raw_write__success;
 254
 255     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 256                       "] write failed %d : %d = %s\n",
 257                       s->fd, bs->filename, offset, buf, count,
 258                       bs->total_sectors, ret, errno, strerror(errno));
 259
 260 label__raw_write__success:
 261
 262     return ret;
 263 }
 264
 265
 266 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 267 /*
 268  * offset and count are in bytes and possibly not aligned. For files opened
 269  * with O_DIRECT, necessary alignments are ensured before calling
 270  * raw_pread_aligned to do the actual read.
 271  */
 272 static int raw_pread(BlockDriverState *bs, int64_t offset,
 273                      uint8_t *buf, int count)
 274 {
 275     BDRVRawState *s = bs->opaque;
 276     int size, ret, shift, sum;
 277
 278     sum = 0;
 279
 280     if (s->aligned_buf != NULL)  {
 281
 282         if (offset & 0x1ff) {
 283             /* align offset on a 512 bytes boundary */
 284
 285             shift = offset & 0x1ff;
 286             size = (shift + count + 0x1ff) & ~0x1ff;
 287             if (size > ALIGNED_BUFFER_SIZE)
 288                 size = ALIGNED_BUFFER_SIZE;
 289             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 290             if (ret < 0)
 291                 return ret;
 292
 293             size = 512 - shift;
 294             if (size > count)
 295                 size = count;
 296             memcpy(buf, s->aligned_buf + shift, size);
 297
 298             buf += size;
 299             offset += size;
 300             count -= size;
 301             sum += size;
 302
 303             if (count == 0)
 304                 return sum;
 305         }
 306         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 307
 308             /* read on aligned buffer */
 309
 310             while (count) {
 311
 312                 size = (count + 0x1ff) & ~0x1ff;
 313                 if (size > ALIGNED_BUFFER_SIZE)
 314                     size = ALIGNED_BUFFER_SIZE;
 315
 316                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 317                 if (ret < 0)
 318                     return ret;
 319
 320                 size = ret;
 321                 if (size > count)
 322                     size = count;
 323
 324                 memcpy(buf, s->aligned_buf, size);
 325
 326                 buf += size;
 327                 offset += size;
 328                 count -= size;
 329                 sum += size;
 330             }
 331
 332             return sum;
 333         }
 334     }
 335
 336     return raw_pread_aligned(bs, offset, buf, count) + sum;
 337 }
 338
 339 /*
 340  * offset and count are in bytes and possibly not aligned. For files opened
 341  * with O_DIRECT, necessary alignments are ensured before calling
 342  * raw_pwrite_aligned to do the actual write.
 343  */
 344 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 345                       const uint8_t *buf, int count)
 346 {
 347     BDRVRawState *s = bs->opaque;
 348     int size, ret, shift, sum;
 349
 350     sum = 0;
 351
 352     if (s->aligned_buf != NULL) {
 353
 354         if (offset & 0x1ff) {
 355             /* align offset on a 512 bytes boundary */
 356             shift = offset & 0x1ff;
 357             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 358             if (ret < 0)
 359                 return ret;
 360
 361             size = 512 - shift;
 362             if (size > count)
 363                 size = count;
 364             memcpy(s->aligned_buf + shift, buf, size);
 365
 366             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 367             if (ret < 0)
 368                 return ret;
 369
 370             buf += size;
 371             offset += size;
 372             count -= size;
 373             sum += size;
 374
 375             if (count == 0)
 376                 return sum;
 377         }
 378         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 379
 380             while ((size = (count & ~0x1ff)) != 0) {
 381
 382                 if (size > ALIGNED_BUFFER_SIZE)
 383                     size = ALIGNED_BUFFER_SIZE;
 384
 385                 memcpy(s->aligned_buf, buf, size);
 386
 387                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 388                 if (ret < 0)
 389                     return ret;
 390
 391                 buf += ret;
 392                 offset += ret;
 393                 count -= ret;
 394                 sum += ret;
 395             }
 396             /* here, count < 512 because (count & ~0x1ff) == 0 */
 397             if (count) {
 398                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 399                 if (ret < 0)
 400                     return ret;
 401                  memcpy(s->aligned_buf, buf, count);
 402
 403                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 404                  if (ret < 0)
 405                      return ret;
 406                  if (count < ret)
 407                      ret = count;
 408
 409                  sum += ret;
 410             }
 411             return sum;
 412         }
 413     }
 414     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 415 }
 416
 417 #else
 418 #define raw_pread raw_pread_aligned
 419 #define raw_pwrite raw_pwrite_aligned
 420 #endif
 421
 422
 423 #ifdef CONFIG_AIO
 424 /***********************************************************/
 425 /* Unix AIO using POSIX AIO */
 426
 427 typedef struct RawAIOCB {
 428     BlockDriverAIOCB common;
 429     struct aiocb aiocb;
 430     struct RawAIOCB *next;
 431     int ret;
 432 } RawAIOCB;
 433
 434 static int aio_sig_num = SIGUSR2;
 435 static RawAIOCB *first_aio; /* AIO issued */
 436 static int aio_initialized = 0;
 437
 438 static void aio_signal_handler(int signum)
 439 {
 440 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 441     CPUState *env = cpu_single_env;
 442     if (env) {
 443         /* stop the currently executing cpu because a timer occured */
 444         cpu_interrupt(env, CPU_INTERRUPT_EXIT);
 445 #ifdef USE_KQEMU
 446         if (env->kqemu_enabled) {
 447             kqemu_cpu_interrupt(env);
 448         }
 449 #endif
 450     }
 451 #endif
 452 }
 453
 454 void qemu_aio_init(void)
 455 {
 456     struct sigaction act;
 457
 458     aio_initialized = 1;
 459
 460     sigfillset(&act.sa_mask);
 461     act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
 462     act.sa_handler = aio_signal_handler;
 463     sigaction(aio_sig_num, &act, NULL);
 464
 465 #if defined(__GLIBC__) && defined(__linux__)
 466     {
 467         /* XXX: aio thread exit seems to hang on RedHat 9 and this init
 468            seems to fix the problem. */
 469         struct aioinit ai;
 470         memset(&ai, 0, sizeof(ai));
 471         ai.aio_threads = 1;
 472         ai.aio_num = 1;
 473         ai.aio_idle_time = 365 * 100000;
 474         aio_init(&ai);
 475     }
 476 #endif
 477 }
 478
 479 void qemu_aio_poll(void)
 480 {
 481     RawAIOCB *acb, **pacb;
 482     int ret;
 483
 484     for(;;) {
 485         pacb = &first_aio;
 486         for(;;) {
 487             acb = *pacb;
 488             if (!acb)
 489                 goto the_end;
 490             ret = aio_error(&acb->aiocb);
 491             if (ret == ECANCELED) {
 492                 /* remove the request */
 493                 *pacb = acb->next;
 494                 qemu_aio_release(acb);
 495             } else if (ret != EINPROGRESS) {
 496                 /* end of aio */
 497                 if (ret == 0) {
 498                     ret = aio_return(&acb->aiocb);
 499                     if (ret == acb->aiocb.aio_nbytes)
 500                         ret = 0;
 501                     else
 502                         ret = -EINVAL;
 503                 } else {
 504                     ret = -ret;
 505                 }
 506                 /* remove the request */
 507                 *pacb = acb->next;
 508                 /* call the callback */
 509                 acb->common.cb(acb->common.opaque, ret);
 510                 qemu_aio_release(acb);
 511                 break;
 512             } else {
 513                 pacb = &acb->next;
 514             }
 515         }
 516     }
 517  the_end: ;
 518 }
 519
 520 /* Wait for all IO requests to complete.  */
 521 void qemu_aio_flush(void)
 522 {
 523     qemu_aio_wait_start();
 524     qemu_aio_poll();
 525     while (first_aio) {
 526         qemu_aio_wait();
 527     }
 528     qemu_aio_wait_end();
 529 }
 530
 531 /* wait until at least one AIO was handled */
 532 static sigset_t wait_oset;
 533
 534 void qemu_aio_wait_start(void)
 535 {
 536     sigset_t set;
 537
 538     if (!aio_initialized)
 539         qemu_aio_init();
 540     sigemptyset(&set);
 541     sigaddset(&set, aio_sig_num);
 542     sigprocmask(SIG_BLOCK, &set, &wait_oset);
 543 }
 544
 545 void qemu_aio_wait(void)
 546 {
 547     sigset_t set;
 548     int nb_sigs;
 549
 550 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 551     if (qemu_bh_poll())
 552         return;
 553 #endif
 554     sigemptyset(&set);
 555     sigaddset(&set, aio_sig_num);
 556     sigwait(&set, &nb_sigs);
 557     qemu_aio_poll();
 558 }
 559
 560 void qemu_aio_wait_end(void)
 561 {
 562     sigprocmask(SIG_SETMASK, &wait_oset, NULL);
 563 }
 564
 565 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 566         int64_t sector_num, uint8_t *buf, int nb_sectors,
 567         BlockDriverCompletionFunc *cb, void *opaque)
 568 {
 569     BDRVRawState *s = bs->opaque;
 570     RawAIOCB *acb;
 571
 572     if (fd_open(bs) < 0)
 573         return NULL;
 574
 575     acb = qemu_aio_get(bs, cb, opaque);
 576     if (!acb)
 577         return NULL;
 578     acb->aiocb.aio_fildes = s->fd;
 579     acb->aiocb.aio_sigevent.sigev_signo = aio_sig_num;
 580     acb->aiocb.aio_sigevent.sigev_notify = SIGEV_SIGNAL;
 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 = first_aio;
 588     first_aio = acb;
 589     return acb;
 590 }
 591
 592 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 593 static void raw_aio_em_cb(void* opaque)
 594 {
 595     RawAIOCB *acb = opaque;
 596     acb->common.cb(acb->common.opaque, acb->ret);
 597     qemu_aio_release(acb);
 598 }
 599 #endif
 600
 601 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 602         int64_t sector_num, uint8_t *buf, int nb_sectors,
 603         BlockDriverCompletionFunc *cb, void *opaque)
 604 {
 605     RawAIOCB *acb;
 606
 607     /*
 608      * If O_DIRECT is used and the buffer is not aligned fall back
 609      * to synchronous IO.
 610      */
 611 #if defined(O_DIRECT) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
 612     BDRVRawState *s = bs->opaque;
 613
 614     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 615         QEMUBH *bh;
 616         acb = qemu_aio_get(bs, cb, opaque);
 617         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 618         bh = qemu_bh_new(raw_aio_em_cb, acb);
 619         qemu_bh_schedule(bh);
 620         return &acb->common;
 621     }
 622 #endif
 623
 624     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 625     if (!acb)
 626         return NULL;
 627     if (aio_read(&acb->aiocb) < 0) {
 628         qemu_aio_release(acb);
 629         return NULL;
 630     }
 631     return &acb->common;
 632 }
 633
 634 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 635         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 636         BlockDriverCompletionFunc *cb, void *opaque)
 637 {
 638     RawAIOCB *acb;
 639
 640     /*
 641      * If O_DIRECT is used and the buffer is not aligned fall back
 642      * to synchronous IO.
 643      */
 644 #if defined(O_DIRECT) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
 645     BDRVRawState *s = bs->opaque;
 646
 647     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 648         QEMUBH *bh;
 649         acb = qemu_aio_get(bs, cb, opaque);
 650         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 651         bh = qemu_bh_new(raw_aio_em_cb, acb);
 652         qemu_bh_schedule(bh);
 653         return &acb->common;
 654     }
 655 #endif
 656
 657     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 658     if (!acb)
 659         return NULL;
 660     if (aio_write(&acb->aiocb) < 0) {
 661         qemu_aio_release(acb);
 662         return NULL;
 663     }
 664     return &acb->common;
 665 }
 666
 667 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 668 {
 669     int ret;
 670     RawAIOCB *acb = (RawAIOCB *)blockacb;
 671     RawAIOCB **pacb;
 672
 673     ret = aio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 674     if (ret == AIO_NOTCANCELED) {
 675         /* fail safe: if the aio could not be canceled, we wait for
 676            it */
 677         while (aio_error(&acb->aiocb) == EINPROGRESS);
 678     }
 679
 680     /* remove the callback from the queue */
 681     pacb = &first_aio;
 682     for(;;) {
 683         if (*pacb == NULL) {
 684             break;
 685         } else if (*pacb == acb) {
 686             *pacb = acb->next;
 687             qemu_aio_release(acb);
 688             break;
 689         }
 690         pacb = &acb->next;
 691     }
 692 }
 693
 694 # else /* CONFIG_AIO */
 695
 696 void qemu_aio_init(void)
 697 {
 698 }
 699
 700 void qemu_aio_poll(void)
 701 {
 702 }
 703
 704 void qemu_aio_flush(void)
 705 {
 706 }
 707
 708 void qemu_aio_wait_start(void)
 709 {
 710 }
 711
 712 void qemu_aio_wait(void)
 713 {
 714 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 715     qemu_bh_poll();
 716 #endif
 717 }
 718
 719 void qemu_aio_wait_end(void)
 720 {
 721 }
 722
 723 #endif /* CONFIG_AIO */
 724
 725 static void raw_close(BlockDriverState *bs)
 726 {
 727     BDRVRawState *s = bs->opaque;
 728     if (s->fd >= 0) {
 729         close(s->fd);
 730         s->fd = -1;
 731 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 732         if (s->aligned_buf != NULL)
 733             qemu_free(s->aligned_buf);
 734 #endif
 735     }
 736 }
 737
 738 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 739 {
 740     BDRVRawState *s = bs->opaque;
 741     if (s->type != FTYPE_FILE)
 742         return -ENOTSUP;
 743     if (ftruncate(s->fd, offset) < 0)
 744         return -errno;
 745     return 0;
 746 }
 747
 748 static int64_t  raw_getlength(BlockDriverState *bs)
 749 {
 750     BDRVRawState *s = bs->opaque;
 751     int fd = s->fd;
 752     int64_t size;
 753 #ifdef _BSD
 754     struct stat sb;
 755 #endif
 756 #ifdef __sun__
 757     struct dk_minfo minfo;
 758     int rv;
 759 #endif
 760     int ret;
 761
 762     ret = fd_open(bs);
 763     if (ret < 0)
 764         return ret;
 765
 766 #ifdef _BSD
 767     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 768 #ifdef DIOCGMEDIASIZE
 769         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 770 #endif
 771 #ifdef CONFIG_COCOA
 772         size = LONG_LONG_MAX;
 773 #else
 774         size = lseek(fd, 0LL, SEEK_END);
 775 #endif
 776     } else
 777 #endif
 778 #ifdef __sun__
 779     /*
 780      * use the DKIOCGMEDIAINFO ioctl to read the size.
 781      */
 782     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 783     if ( rv != -1 ) {
 784         size = minfo.dki_lbsize * minfo.dki_capacity;
 785     } else /* there are reports that lseek on some devices
 786               fails, but irc discussion said that contingency
 787               on contingency was overkill */
 788 #endif
 789     {
 790         size = lseek(fd, 0, SEEK_END);
 791     }
 792     return size;
 793 }
 794
 795 static int raw_create(const char *filename, int64_t total_size,
 796                       const char *backing_file, int flags)
 797 {
 798     int fd;
 799
 800     if (flags || backing_file)
 801         return -ENOTSUP;
 802
 803     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 804               0644);
 805     if (fd < 0)
 806         return -EIO;
 807     ftruncate(fd, total_size * 512);
 808     close(fd);
 809     return 0;
 810 }
 811
 812 static void raw_flush(BlockDriverState *bs)
 813 {
 814     BDRVRawState *s = bs->opaque;
 815     fsync(s->fd);
 816 }
 817
 818 BlockDriver bdrv_raw = {
 819     "raw",
 820     sizeof(BDRVRawState),
 821     NULL, /* no probe for protocols */
 822     raw_open,
 823     NULL,
 824     NULL,
 825     raw_close,
 826     raw_create,
 827     raw_flush,
 828
 829 #ifdef CONFIG_AIO
 830     .bdrv_aio_read = raw_aio_read,
 831     .bdrv_aio_write = raw_aio_write,
 832     .bdrv_aio_cancel = raw_aio_cancel,
 833     .aiocb_size = sizeof(RawAIOCB),
 834 #endif
 835     .protocol_name = "file",
 836     .bdrv_pread = raw_pread,
 837     .bdrv_pwrite = raw_pwrite,
 838     .bdrv_truncate = raw_truncate,
 839     .bdrv_getlength = raw_getlength,
 840 };
 841
 842 /***********************************************/
 843 /* host device */
 844
 845 #ifdef CONFIG_COCOA
 846 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 847 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 848
 849 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 850 {
 851     kern_return_t       kernResult;
 852     mach_port_t     masterPort;
 853     CFMutableDictionaryRef  classesToMatch;
 854
 855     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 856     if ( KERN_SUCCESS != kernResult ) {
 857         printf( "IOMasterPort returned %d\n", kernResult );
 858     }
 859
 860     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 861     if ( classesToMatch == NULL ) {
 862         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 863     } else {
 864     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 865     }
 866     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 867     if ( KERN_SUCCESS != kernResult )
 868     {
 869         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 870     }
 871
 872     return kernResult;
 873 }
 874
 875 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 876 {
 877     io_object_t     nextMedia;
 878     kern_return_t   kernResult = KERN_FAILURE;
 879     *bsdPath = '\0';
 880     nextMedia = IOIteratorNext( mediaIterator );
 881     if ( nextMedia )
 882     {
 883         CFTypeRef   bsdPathAsCFString;
 884     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 885         if ( bsdPathAsCFString ) {
 886             size_t devPathLength;
 887             strcpy( bsdPath, _PATH_DEV );
 888             strcat( bsdPath, "r" );
 889             devPathLength = strlen( bsdPath );
 890             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 891                 kernResult = KERN_SUCCESS;
 892             }
 893             CFRelease( bsdPathAsCFString );
 894         }
 895         IOObjectRelease( nextMedia );
 896     }
 897
 898     return kernResult;
 899 }
 900
 901 #endif
 902
 903 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 904 {
 905     BDRVRawState *s = bs->opaque;
 906     int fd, open_flags, ret;
 907
 908 #ifdef CONFIG_COCOA
 909     if (strstart(filename, "/dev/cdrom", NULL)) {
 910         kern_return_t kernResult;
 911         io_iterator_t mediaIterator;
 912         char bsdPath[ MAXPATHLEN ];
 913         int fd;
 914
 915         kernResult = FindEjectableCDMedia( &mediaIterator );
 916         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 917
 918         if ( bsdPath[ 0 ] != '\0' ) {
 919             strcat(bsdPath,"s0");
 920             /* some CDs don't have a partition 0 */
 921             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 922             if (fd < 0) {
 923                 bsdPath[strlen(bsdPath)-1] = '1';
 924             } else {
 925                 close(fd);
 926             }
 927             filename = bsdPath;
 928         }
 929
 930         if ( mediaIterator )
 931             IOObjectRelease( mediaIterator );
 932     }
 933 #endif
 934     open_flags = O_BINARY;
 935     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 936         open_flags |= O_RDWR;
 937     } else {
 938         open_flags |= O_RDONLY;
 939         bs->read_only = 1;
 940     }
 941 #ifdef O_DIRECT
 942     if (flags & BDRV_O_DIRECT)
 943         open_flags |= O_DIRECT;
 944 #endif
 945
 946     s->type = FTYPE_FILE;
 947 #if defined(__linux__)
 948     if (strstart(filename, "/dev/cd", NULL)) {
 949         /* open will not fail even if no CD is inserted */
 950         open_flags |= O_NONBLOCK;
 951         s->type = FTYPE_CD;
 952     } else if (strstart(filename, "/dev/fd", NULL)) {
 953         s->type = FTYPE_FD;
 954         s->fd_open_flags = open_flags;
 955         /* open will not fail even if no floppy is inserted */
 956         open_flags |= O_NONBLOCK;
 957     } else if (strstart(filename, "/dev/sg", NULL)) {
 958         bs->sg = 1;
 959     }
 960 #endif
 961     fd = open(filename, open_flags, 0644);
 962     if (fd < 0) {
 963         ret = -errno;
 964         if (ret == -EROFS)
 965             ret = -EACCES;
 966         return ret;
 967     }
 968     s->fd = fd;
 969 #if defined(__linux__)
 970     /* close fd so that we can reopen it as needed */
 971     if (s->type == FTYPE_FD) {
 972         close(s->fd);
 973         s->fd = -1;
 974         s->fd_media_changed = 1;
 975     }
 976 #endif
 977     return 0;
 978 }
 979
 980 #if defined(__linux__) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
 981
 982 /* Note: we do not have a reliable method to detect if the floppy is
 983    present. The current method is to try to open the floppy at every
 984    I/O and to keep it opened during a few hundreds of ms. */
 985 static int fd_open(BlockDriverState *bs)
 986 {
 987     BDRVRawState *s = bs->opaque;
 988     int last_media_present;
 989
 990     if (s->type != FTYPE_FD)
 991         return 0;
 992     last_media_present = (s->fd >= 0);
 993     if (s->fd >= 0 &&
 994         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
 995         close(s->fd);
 996         s->fd = -1;
 997 #ifdef DEBUG_FLOPPY
 998         printf("Floppy closed\n");
 999 #endif
1000     }
1001     if (s->fd < 0) {
1002         if (s->fd_got_error &&
1003             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1004 #ifdef DEBUG_FLOPPY
1005             printf("No floppy (open delayed)\n");
1006 #endif
1007             return -EIO;
1008         }
1009         s->fd = open(bs->filename, s->fd_open_flags);
1010         if (s->fd < 0) {
1011             s->fd_error_time = qemu_get_clock(rt_clock);
1012             s->fd_got_error = 1;
1013             if (last_media_present)
1014                 s->fd_media_changed = 1;
1015 #ifdef DEBUG_FLOPPY
1016             printf("No floppy\n");
1017 #endif
1018             return -EIO;
1019         }
1020 #ifdef DEBUG_FLOPPY
1021         printf("Floppy opened\n");
1022 #endif
1023     }
1024     if (!last_media_present)
1025         s->fd_media_changed = 1;
1026     s->fd_open_time = qemu_get_clock(rt_clock);
1027     s->fd_got_error = 0;
1028     return 0;
1029 }
1030 #else
1031 static int fd_open(BlockDriverState *bs)
1032 {
1033     return 0;
1034 }
1035 #endif
1036
1037 #if defined(__linux__)
1038
1039 static int raw_is_inserted(BlockDriverState *bs)
1040 {
1041     BDRVRawState *s = bs->opaque;
1042     int ret;
1043
1044     switch(s->type) {
1045     case FTYPE_CD:
1046         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1047         if (ret == CDS_DISC_OK)
1048             return 1;
1049         else
1050             return 0;
1051         break;
1052     case FTYPE_FD:
1053         ret = fd_open(bs);
1054         return (ret >= 0);
1055     default:
1056         return 1;
1057     }
1058 }
1059
1060 /* currently only used by fdc.c, but a CD version would be good too */
1061 static int raw_media_changed(BlockDriverState *bs)
1062 {
1063     BDRVRawState *s = bs->opaque;
1064
1065     switch(s->type) {
1066     case FTYPE_FD:
1067         {
1068             int ret;
1069             /* XXX: we do not have a true media changed indication. It
1070                does not work if the floppy is changed without trying
1071                to read it */
1072             fd_open(bs);
1073             ret = s->fd_media_changed;
1074             s->fd_media_changed = 0;
1075 #ifdef DEBUG_FLOPPY
1076             printf("Floppy changed=%d\n", ret);
1077 #endif
1078             return ret;
1079         }
1080     default:
1081         return -ENOTSUP;
1082     }
1083 }
1084
1085 static int raw_eject(BlockDriverState *bs, int eject_flag)
1086 {
1087     BDRVRawState *s = bs->opaque;
1088
1089     switch(s->type) {
1090     case FTYPE_CD:
1091         if (eject_flag) {
1092             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1093                 perror("CDROMEJECT");
1094         } else {
1095             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1096                 perror("CDROMEJECT");
1097         }
1098         break;
1099     case FTYPE_FD:
1100         {
1101             int fd;
1102             if (s->fd >= 0) {
1103                 close(s->fd);
1104                 s->fd = -1;
1105             }
1106             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1107             if (fd >= 0) {
1108                 if (ioctl(fd, FDEJECT, 0) < 0)
1109                     perror("FDEJECT");
1110                 close(fd);
1111             }
1112         }
1113         break;
1114     default:
1115         return -ENOTSUP;
1116     }
1117     return 0;
1118 }
1119
1120 static int raw_set_locked(BlockDriverState *bs, int locked)
1121 {
1122     BDRVRawState *s = bs->opaque;
1123
1124     switch(s->type) {
1125     case FTYPE_CD:
1126         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1127             /* Note: an error can happen if the distribution automatically
1128                mounts the CD-ROM */
1129             //        perror("CDROM_LOCKDOOR");
1130         }
1131         break;
1132     default:
1133         return -ENOTSUP;
1134     }
1135     return 0;
1136 }
1137
1138 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1139 {
1140     BDRVRawState *s = bs->opaque;
1141
1142     return ioctl(s->fd, req, buf);
1143 }
1144 #else
1145
1146 static int raw_is_inserted(BlockDriverState *bs)
1147 {
1148     return 1;
1149 }
1150
1151 static int raw_media_changed(BlockDriverState *bs)
1152 {
1153     return -ENOTSUP;
1154 }
1155
1156 static int raw_eject(BlockDriverState *bs, int eject_flag)
1157 {
1158     return -ENOTSUP;
1159 }
1160
1161 static int raw_set_locked(BlockDriverState *bs, int locked)
1162 {
1163     return -ENOTSUP;
1164 }
1165
1166 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1167 {
1168     return -ENOTSUP;
1169 }
1170 #endif /* !linux */
1171
1172 BlockDriver bdrv_host_device = {
1173     "host_device",
1174     sizeof(BDRVRawState),
1175     NULL, /* no probe for protocols */
1176     hdev_open,
1177     NULL,
1178     NULL,
1179     raw_close,
1180     NULL,
1181     raw_flush,
1182
1183 #ifdef CONFIG_AIO
1184     .bdrv_aio_read = raw_aio_read,
1185     .bdrv_aio_write = raw_aio_write,
1186     .bdrv_aio_cancel = raw_aio_cancel,
1187     .aiocb_size = sizeof(RawAIOCB),
1188 #endif
1189     .bdrv_pread = raw_pread,
1190     .bdrv_pwrite = raw_pwrite,
1191     .bdrv_getlength = raw_getlength,
1192
1193     /* removable device support */
1194     .bdrv_is_inserted = raw_is_inserted,
1195     .bdrv_media_changed = raw_media_changed,
1196     .bdrv_eject = raw_eject,
1197     .bdrv_set_locked = raw_set_locked,
1198     /* generic scsi device */
1199     .bdrv_ioctl = raw_ioctl,
1200 };