block-raw-posix.c

   1 /*
   2  * Block driver for RAW files (posix)
   3  *
   4  * Copyright (c) 2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #include "qemu-timer.h"
  26 #include "qemu-char.h"
  27 #include "block_int.h"
  28 #include <assert.h>
  29 #ifdef CONFIG_AIO
  30 #include "posix-aio-compat.h"
  31 #endif
  32
  33 #ifdef CONFIG_COCOA
  34 #include <paths.h>
  35 #include <sys/param.h>
  36 #include <IOKit/IOKitLib.h>
  37 #include <IOKit/IOBSD.h>
  38 #include <IOKit/storage/IOMediaBSDClient.h>
  39 #include <IOKit/storage/IOMedia.h>
  40 #include <IOKit/storage/IOCDMedia.h>
  41 //#include <IOKit/storage/IOCDTypes.h>
  42 #include <CoreFoundation/CoreFoundation.h>
  43 #endif
  44
  45 #ifdef __sun__
  46 #define _POSIX_PTHREAD_SEMANTICS 1
  47 #include <signal.h>
  48 #include <sys/dkio.h>
  49 #endif
  50 #ifdef __linux__
  51 #include <sys/ioctl.h>
  52 #include <linux/cdrom.h>
  53 #include <linux/fd.h>
  54 #endif
  55 #ifdef __FreeBSD__
  56 #include <signal.h>
  57 #include <sys/disk.h>
  58 #endif
  59
  60 #ifdef __OpenBSD__
  61 #include <sys/ioctl.h>
  62 #include <sys/disklabel.h>
  63 #include <sys/dkio.h>
  64 #endif
  65
  66 #ifdef __DragonFly__
  67 #include <sys/ioctl.h>
  68 #include <sys/diskslice.h>
  69 #endif
  70
  71 //#define DEBUG_FLOPPY
  72
  73 //#define DEBUG_BLOCK
  74 #if defined(DEBUG_BLOCK)
  75 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (qemu_log_enabled())     \
  76     { qemu_log(formatCstr, ##args); qemu_log_flush(); } } while (0)
  77 #else
  78 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  79 #endif
  80
  81 /* OS X does not have O_DSYNC */
  82 #ifndef O_DSYNC
  83 #define O_DSYNC O_SYNC
  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;
 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 rfd, wfd;
 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     ssize_t len;
 462
 463     /* read all bytes from signal pipe */
 464     for (;;) {
 465         char bytes[16];
 466
 467         len = read(s->rfd, bytes, sizeof(bytes));
 468         if (len == -1 && errno == EINTR)
 469             continue; /* try again */
 470         if (len == sizeof(bytes))
 471             continue; /* more to read */
 472         break;
 473     }
 474
 475     for(;;) {
 476         pacb = &s->first_aio;
 477         for(;;) {
 478             acb = *pacb;
 479             if (!acb)
 480                 goto the_end;
 481             ret = qemu_paio_error(&acb->aiocb);
 482             if (ret == ECANCELED) {
 483                 /* remove the request */
 484                 *pacb = acb->next;
 485                 qemu_aio_release(acb);
 486             } else if (ret != EINPROGRESS) {
 487                 /* end of aio */
 488                 if (ret == 0) {
 489                     ret = qemu_paio_return(&acb->aiocb);
 490                     if (ret == acb->aiocb.aio_nbytes)
 491                         ret = 0;
 492                     else
 493                         ret = -EINVAL;
 494                 } else {
 495                     ret = -ret;
 496                 }
 497                 /* remove the request */
 498                 *pacb = acb->next;
 499                 /* call the callback */
 500                 acb->common.cb(acb->common.opaque, ret);
 501                 qemu_aio_release(acb);
 502                 break;
 503             } else {
 504                 pacb = &acb->next;
 505             }
 506         }
 507     }
 508  the_end: ;
 509 }
 510
 511 static int posix_aio_flush(void *opaque)
 512 {
 513     PosixAioState *s = opaque;
 514     return !!s->first_aio;
 515 }
 516
 517 static PosixAioState *posix_aio_state;
 518
 519 static void aio_signal_handler(int signum)
 520 {
 521     if (posix_aio_state) {
 522         char byte = 0;
 523
 524         write(posix_aio_state->wfd, &byte, sizeof(byte));
 525     }
 526
 527     qemu_service_io();
 528 }
 529
 530 static int posix_aio_init(void)
 531 {
 532     struct sigaction act;
 533     PosixAioState *s;
 534     int fds[2];
 535     struct qemu_paioinit ai;
 536
 537     if (posix_aio_state)
 538         return 0;
 539
 540     s = qemu_malloc(sizeof(PosixAioState));
 541
 542     sigfillset(&act.sa_mask);
 543     act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
 544     act.sa_handler = aio_signal_handler;
 545     sigaction(SIGUSR2, &act, NULL);
 546
 547     s->first_aio = NULL;
 548     if (pipe(fds) == -1) {
 549         fprintf(stderr, "failed to create pipe\n");
 550         return -errno;
 551     }
 552
 553     s->rfd = fds[0];
 554     s->wfd = fds[1];
 555
 556     fcntl(s->rfd, F_SETFL, O_NONBLOCK);
 557     fcntl(s->wfd, F_SETFL, O_NONBLOCK);
 558
 559     qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
 560
 561     memset(&ai, 0, sizeof(ai));
 562     ai.aio_threads = 64;
 563     ai.aio_num = 64;
 564     qemu_paio_init(&ai);
 565
 566     posix_aio_state = s;
 567
 568     return 0;
 569 }
 570
 571 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 572         int64_t sector_num, uint8_t *buf, int nb_sectors,
 573         BlockDriverCompletionFunc *cb, void *opaque)
 574 {
 575     BDRVRawState *s = bs->opaque;
 576     RawAIOCB *acb;
 577
 578     if (fd_open(bs) < 0)
 579         return NULL;
 580
 581     acb = qemu_aio_get(bs, cb, opaque);
 582     if (!acb)
 583         return NULL;
 584     acb->aiocb.aio_fildes = s->fd;
 585     acb->aiocb.ev_signo = SIGUSR2;
 586     acb->aiocb.aio_buf = buf;
 587     if (nb_sectors < 0)
 588         acb->aiocb.aio_nbytes = -nb_sectors;
 589     else
 590         acb->aiocb.aio_nbytes = nb_sectors * 512;
 591     acb->aiocb.aio_offset = sector_num * 512;
 592     acb->next = posix_aio_state->first_aio;
 593     posix_aio_state->first_aio = acb;
 594     return acb;
 595 }
 596
 597 static void raw_aio_em_cb(void* opaque)
 598 {
 599     RawAIOCB *acb = opaque;
 600     acb->common.cb(acb->common.opaque, acb->ret);
 601     qemu_aio_release(acb);
 602 }
 603
 604 static void raw_aio_remove(RawAIOCB *acb)
 605 {
 606     RawAIOCB **pacb;
 607
 608     /* remove the callback from the queue */
 609     pacb = &posix_aio_state->first_aio;
 610     for(;;) {
 611         if (*pacb == NULL) {
 612             fprintf(stderr, "raw_aio_remove: aio request not found!\n");
 613             break;
 614         } else if (*pacb == acb) {
 615             *pacb = acb->next;
 616             qemu_aio_release(acb);
 617             break;
 618         }
 619         pacb = &(*pacb)->next;
 620     }
 621 }
 622
 623 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 624         int64_t sector_num, uint8_t *buf, int nb_sectors,
 625         BlockDriverCompletionFunc *cb, void *opaque)
 626 {
 627     RawAIOCB *acb;
 628
 629     /*
 630      * If O_DIRECT is used and the buffer is not aligned fall back
 631      * to synchronous IO.
 632      */
 633     BDRVRawState *s = bs->opaque;
 634
 635     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 636         QEMUBH *bh;
 637         acb = qemu_aio_get(bs, cb, opaque);
 638         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 639         bh = qemu_bh_new(raw_aio_em_cb, acb);
 640         qemu_bh_schedule(bh);
 641         return &acb->common;
 642     }
 643
 644     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 645     if (!acb)
 646         return NULL;
 647     if (qemu_paio_read(&acb->aiocb) < 0) {
 648         raw_aio_remove(acb);
 649         return NULL;
 650     }
 651     return &acb->common;
 652 }
 653
 654 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 655         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 656         BlockDriverCompletionFunc *cb, void *opaque)
 657 {
 658     RawAIOCB *acb;
 659
 660     /*
 661      * If O_DIRECT is used and the buffer is not aligned fall back
 662      * to synchronous IO.
 663      */
 664     BDRVRawState *s = bs->opaque;
 665
 666     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 667         QEMUBH *bh;
 668         acb = qemu_aio_get(bs, cb, opaque);
 669         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 670         bh = qemu_bh_new(raw_aio_em_cb, acb);
 671         qemu_bh_schedule(bh);
 672         return &acb->common;
 673     }
 674
 675     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 676     if (!acb)
 677         return NULL;
 678     if (qemu_paio_write(&acb->aiocb) < 0) {
 679         raw_aio_remove(acb);
 680         return NULL;
 681     }
 682     return &acb->common;
 683 }
 684
 685 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 686 {
 687     int ret;
 688     RawAIOCB *acb = (RawAIOCB *)blockacb;
 689
 690     ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 691     if (ret == QEMU_PAIO_NOTCANCELED) {
 692         /* fail safe: if the aio could not be canceled, we wait for
 693            it */
 694         while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
 695     }
 696
 697     raw_aio_remove(acb);
 698 }
 699 #else /* CONFIG_AIO */
 700 static int posix_aio_init(void)
 701 {
 702     return 0;
 703 }
 704 #endif /* CONFIG_AIO */
 705
 706
 707 static void raw_close(BlockDriverState *bs)
 708 {
 709     BDRVRawState *s = bs->opaque;
 710     if (s->fd >= 0) {
 711         close(s->fd);
 712         s->fd = -1;
 713         if (s->aligned_buf != NULL)
 714             qemu_free(s->aligned_buf);
 715     }
 716 }
 717
 718 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 719 {
 720     BDRVRawState *s = bs->opaque;
 721     if (s->type != FTYPE_FILE)
 722         return -ENOTSUP;
 723     if (ftruncate(s->fd, offset) < 0)
 724         return -errno;
 725     return 0;
 726 }
 727
 728 #ifdef __OpenBSD__
 729 static int64_t raw_getlength(BlockDriverState *bs)
 730 {
 731     BDRVRawState *s = bs->opaque;
 732     int fd = s->fd;
 733     struct stat st;
 734
 735     if (fstat(fd, &st))
 736         return -1;
 737     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 738         struct disklabel dl;
 739
 740         if (ioctl(fd, DIOCGDINFO, &dl))
 741             return -1;
 742         return (uint64_t)dl.d_secsize *
 743             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 744     } else
 745         return st.st_size;
 746 }
 747 #else /* !__OpenBSD__ */
 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 HOST_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 HOST_BSD
 767     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 768 #ifdef DIOCGMEDIASIZE
 769         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 770 #elif defined(DIOCGPART)
 771         {
 772                 struct partinfo pi;
 773                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 774                         size = pi.media_size;
 775                 else
 776                         size = 0;
 777         }
 778         if (size == 0)
 779 #endif
 780 #ifdef CONFIG_COCOA
 781         size = LONG_LONG_MAX;
 782 #else
 783         size = lseek(fd, 0LL, SEEK_END);
 784 #endif
 785     } else
 786 #endif
 787 #ifdef __sun__
 788     /*
 789      * use the DKIOCGMEDIAINFO ioctl to read the size.
 790      */
 791     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 792     if ( rv != -1 ) {
 793         size = minfo.dki_lbsize * minfo.dki_capacity;
 794     } else /* there are reports that lseek on some devices
 795               fails, but irc discussion said that contingency
 796               on contingency was overkill */
 797 #endif
 798     {
 799         size = lseek(fd, 0, SEEK_END);
 800     }
 801     return size;
 802 }
 803 #endif
 804
 805 static int raw_create(const char *filename, int64_t total_size,
 806                       const char *backing_file, int flags)
 807 {
 808     int fd;
 809
 810     if (flags || backing_file)
 811         return -ENOTSUP;
 812
 813     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 814               0644);
 815     if (fd < 0)
 816         return -EIO;
 817     ftruncate(fd, total_size * 512);
 818     close(fd);
 819     return 0;
 820 }
 821
 822 static void raw_flush(BlockDriverState *bs)
 823 {
 824     BDRVRawState *s = bs->opaque;
 825     fsync(s->fd);
 826 }
 827
 828 BlockDriver bdrv_raw = {
 829     "raw",
 830     sizeof(BDRVRawState),
 831     NULL, /* no probe for protocols */
 832     raw_open,
 833     NULL,
 834     NULL,
 835     raw_close,
 836     raw_create,
 837     raw_flush,
 838
 839 #ifdef CONFIG_AIO
 840     .bdrv_aio_read = raw_aio_read,
 841     .bdrv_aio_write = raw_aio_write,
 842     .bdrv_aio_cancel = raw_aio_cancel,
 843     .aiocb_size = sizeof(RawAIOCB),
 844 #endif
 845
 846     .bdrv_pread = raw_pread,
 847     .bdrv_pwrite = raw_pwrite,
 848     .bdrv_truncate = raw_truncate,
 849     .bdrv_getlength = raw_getlength,
 850 };
 851
 852 /***********************************************/
 853 /* host device */
 854
 855 #ifdef CONFIG_COCOA
 856 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 857 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 858
 859 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 860 {
 861     kern_return_t       kernResult;
 862     mach_port_t     masterPort;
 863     CFMutableDictionaryRef  classesToMatch;
 864
 865     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 866     if ( KERN_SUCCESS != kernResult ) {
 867         printf( "IOMasterPort returned %d\n", kernResult );
 868     }
 869
 870     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 871     if ( classesToMatch == NULL ) {
 872         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 873     } else {
 874     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 875     }
 876     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 877     if ( KERN_SUCCESS != kernResult )
 878     {
 879         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 880     }
 881
 882     return kernResult;
 883 }
 884
 885 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 886 {
 887     io_object_t     nextMedia;
 888     kern_return_t   kernResult = KERN_FAILURE;
 889     *bsdPath = '\0';
 890     nextMedia = IOIteratorNext( mediaIterator );
 891     if ( nextMedia )
 892     {
 893         CFTypeRef   bsdPathAsCFString;
 894     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 895         if ( bsdPathAsCFString ) {
 896             size_t devPathLength;
 897             strcpy( bsdPath, _PATH_DEV );
 898             strcat( bsdPath, "r" );
 899             devPathLength = strlen( bsdPath );
 900             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 901                 kernResult = KERN_SUCCESS;
 902             }
 903             CFRelease( bsdPathAsCFString );
 904         }
 905         IOObjectRelease( nextMedia );
 906     }
 907
 908     return kernResult;
 909 }
 910
 911 #endif
 912
 913 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 914 {
 915     BDRVRawState *s = bs->opaque;
 916     int fd, open_flags, ret;
 917
 918     posix_aio_init();
 919
 920 #ifdef CONFIG_COCOA
 921     if (strstart(filename, "/dev/cdrom", NULL)) {
 922         kern_return_t kernResult;
 923         io_iterator_t mediaIterator;
 924         char bsdPath[ MAXPATHLEN ];
 925         int fd;
 926
 927         kernResult = FindEjectableCDMedia( &mediaIterator );
 928         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 929
 930         if ( bsdPath[ 0 ] != '\0' ) {
 931             strcat(bsdPath,"s0");
 932             /* some CDs don't have a partition 0 */
 933             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 934             if (fd < 0) {
 935                 bsdPath[strlen(bsdPath)-1] = '1';
 936             } else {
 937                 close(fd);
 938             }
 939             filename = bsdPath;
 940         }
 941
 942         if ( mediaIterator )
 943             IOObjectRelease( mediaIterator );
 944     }
 945 #endif
 946     open_flags = O_BINARY;
 947     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 948         open_flags |= O_RDWR;
 949     } else {
 950         open_flags |= O_RDONLY;
 951         bs->read_only = 1;
 952     }
 953     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 954      * and O_DIRECT for no caching. */
 955     if ((flags & BDRV_O_NOCACHE))
 956         open_flags |= O_DIRECT;
 957     else if (!(flags & BDRV_O_CACHE_WB))
 958         open_flags |= O_DSYNC;
 959
 960     s->type = FTYPE_FILE;
 961 #if defined(__linux__)
 962     if (strstart(filename, "/dev/cd", NULL)) {
 963         /* open will not fail even if no CD is inserted */
 964         open_flags |= O_NONBLOCK;
 965         s->type = FTYPE_CD;
 966     } else if (strstart(filename, "/dev/fd", NULL)) {
 967         s->type = FTYPE_FD;
 968         s->fd_open_flags = open_flags;
 969         /* open will not fail even if no floppy is inserted */
 970         open_flags |= O_NONBLOCK;
 971     } else if (strstart(filename, "/dev/sg", NULL)) {
 972         bs->sg = 1;
 973     }
 974 #endif
 975     fd = open(filename, open_flags, 0644);
 976     if (fd < 0) {
 977         ret = -errno;
 978         if (ret == -EROFS)
 979             ret = -EACCES;
 980         return ret;
 981     }
 982     s->fd = fd;
 983 #if defined(__linux__)
 984     /* close fd so that we can reopen it as needed */
 985     if (s->type == FTYPE_FD) {
 986         close(s->fd);
 987         s->fd = -1;
 988         s->fd_media_changed = 1;
 989     }
 990 #endif
 991     return 0;
 992 }
 993
 994 #if defined(__linux__)
 995 /* Note: we do not have a reliable method to detect if the floppy is
 996    present. The current method is to try to open the floppy at every
 997    I/O and to keep it opened during a few hundreds of ms. */
 998 static int fd_open(BlockDriverState *bs)
 999 {
1000     BDRVRawState *s = bs->opaque;
1001     int last_media_present;
1002
1003     if (s->type != FTYPE_FD)
1004         return 0;
1005     last_media_present = (s->fd >= 0);
1006     if (s->fd >= 0 &&
1007         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1008         close(s->fd);
1009         s->fd = -1;
1010 #ifdef DEBUG_FLOPPY
1011         printf("Floppy closed\n");
1012 #endif
1013     }
1014     if (s->fd < 0) {
1015         if (s->fd_got_error &&
1016             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1017 #ifdef DEBUG_FLOPPY
1018             printf("No floppy (open delayed)\n");
1019 #endif
1020             return -EIO;
1021         }
1022         s->fd = open(bs->filename, s->fd_open_flags);
1023         if (s->fd < 0) {
1024             s->fd_error_time = qemu_get_clock(rt_clock);
1025             s->fd_got_error = 1;
1026             if (last_media_present)
1027                 s->fd_media_changed = 1;
1028 #ifdef DEBUG_FLOPPY
1029             printf("No floppy\n");
1030 #endif
1031             return -EIO;
1032         }
1033 #ifdef DEBUG_FLOPPY
1034         printf("Floppy opened\n");
1035 #endif
1036     }
1037     if (!last_media_present)
1038         s->fd_media_changed = 1;
1039     s->fd_open_time = qemu_get_clock(rt_clock);
1040     s->fd_got_error = 0;
1041     return 0;
1042 }
1043
1044 static int raw_is_inserted(BlockDriverState *bs)
1045 {
1046     BDRVRawState *s = bs->opaque;
1047     int ret;
1048
1049     switch(s->type) {
1050     case FTYPE_CD:
1051         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1052         if (ret == CDS_DISC_OK)
1053             return 1;
1054         else
1055             return 0;
1056         break;
1057     case FTYPE_FD:
1058         ret = fd_open(bs);
1059         return (ret >= 0);
1060     default:
1061         return 1;
1062     }
1063 }
1064
1065 /* currently only used by fdc.c, but a CD version would be good too */
1066 static int raw_media_changed(BlockDriverState *bs)
1067 {
1068     BDRVRawState *s = bs->opaque;
1069
1070     switch(s->type) {
1071     case FTYPE_FD:
1072         {
1073             int ret;
1074             /* XXX: we do not have a true media changed indication. It
1075                does not work if the floppy is changed without trying
1076                to read it */
1077             fd_open(bs);
1078             ret = s->fd_media_changed;
1079             s->fd_media_changed = 0;
1080 #ifdef DEBUG_FLOPPY
1081             printf("Floppy changed=%d\n", ret);
1082 #endif
1083             return ret;
1084         }
1085     default:
1086         return -ENOTSUP;
1087     }
1088 }
1089
1090 static int raw_eject(BlockDriverState *bs, int eject_flag)
1091 {
1092     BDRVRawState *s = bs->opaque;
1093
1094     switch(s->type) {
1095     case FTYPE_CD:
1096         if (eject_flag) {
1097             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1098                 perror("CDROMEJECT");
1099         } else {
1100             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1101                 perror("CDROMEJECT");
1102         }
1103         break;
1104     case FTYPE_FD:
1105         {
1106             int fd;
1107             if (s->fd >= 0) {
1108                 close(s->fd);
1109                 s->fd = -1;
1110             }
1111             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1112             if (fd >= 0) {
1113                 if (ioctl(fd, FDEJECT, 0) < 0)
1114                     perror("FDEJECT");
1115                 close(fd);
1116             }
1117         }
1118         break;
1119     default:
1120         return -ENOTSUP;
1121     }
1122     return 0;
1123 }
1124
1125 static int raw_set_locked(BlockDriverState *bs, int locked)
1126 {
1127     BDRVRawState *s = bs->opaque;
1128
1129     switch(s->type) {
1130     case FTYPE_CD:
1131         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1132             /* Note: an error can happen if the distribution automatically
1133                mounts the CD-ROM */
1134             //        perror("CDROM_LOCKDOOR");
1135         }
1136         break;
1137     default:
1138         return -ENOTSUP;
1139     }
1140     return 0;
1141 }
1142
1143 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1144 {
1145     BDRVRawState *s = bs->opaque;
1146
1147     return ioctl(s->fd, req, buf);
1148 }
1149 #else
1150
1151 static int fd_open(BlockDriverState *bs)
1152 {
1153     return 0;
1154 }
1155
1156 static int raw_is_inserted(BlockDriverState *bs)
1157 {
1158     return 1;
1159 }
1160
1161 static int raw_media_changed(BlockDriverState *bs)
1162 {
1163     return -ENOTSUP;
1164 }
1165
1166 static int raw_eject(BlockDriverState *bs, int eject_flag)
1167 {
1168     return -ENOTSUP;
1169 }
1170
1171 static int raw_set_locked(BlockDriverState *bs, int locked)
1172 {
1173     return -ENOTSUP;
1174 }
1175
1176 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1177 {
1178     return -ENOTSUP;
1179 }
1180 #endif /* !linux */
1181
1182 BlockDriver bdrv_host_device = {
1183     .format_name        = "host_device",
1184     .instance_size      = sizeof(BDRVRawState),
1185     .bdrv_open          = hdev_open,
1186     .bdrv_close         = raw_close,
1187     .bdrv_flush         = raw_flush,
1188
1189 #ifdef CONFIG_AIO
1190     .bdrv_aio_read      = raw_aio_read,
1191     .bdrv_aio_write     = raw_aio_write,
1192     .bdrv_aio_cancel    = raw_aio_cancel,
1193     .aiocb_size         = sizeof(RawAIOCB),
1194 #endif
1195
1196     .bdrv_pread         = raw_pread,
1197     .bdrv_pwrite        = raw_pwrite,
1198     .bdrv_getlength     = raw_getlength,
1199
1200     /* removable device support */
1201     .bdrv_is_inserted   = raw_is_inserted,
1202     .bdrv_media_changed = raw_media_changed,
1203     .bdrv_eject         = raw_eject,
1204     .bdrv_set_locked    = raw_set_locked,
1205     /* generic scsi device */
1206     .bdrv_ioctl         = raw_ioctl,
1207 };