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