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 //#define DEBUG_FLOPPY
  67
  68 //#define DEBUG_BLOCK
  69 #if defined(DEBUG_BLOCK)
  70 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (qemu_log_enabled())     \
  71     { qemu_log(formatCstr, ##args); qemu_log_flush(); } } while (0)
  72 #else
  73 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  74 #endif
  75
  76 /* OS X does not have O_DSYNC */
  77 #ifndef O_DSYNC
  78 #define O_DSYNC O_SYNC
  79 #endif
  80
  81 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
  82 #ifndef O_DIRECT
  83 #define O_DIRECT O_DSYNC
  84 #endif
  85
  86 #define FTYPE_FILE   0
  87 #define FTYPE_CD     1
  88 #define FTYPE_FD     2
  89
  90 #define ALIGNED_BUFFER_SIZE (32 * 512)
  91
  92 /* if the FD is not accessed during that time (in ms), we try to
  93    reopen it to see if the disk has been changed */
  94 #define FD_OPEN_TIMEOUT 1000
  95
  96 typedef struct BDRVRawState {
  97     int fd;
  98     int type;
  99     unsigned int lseek_err_cnt;
 100 #if defined(__linux__)
 101     /* linux floppy specific */
 102     int fd_open_flags;
 103     int64_t fd_open_time;
 104     int64_t fd_error_time;
 105     int fd_got_error;
 106     int fd_media_changed;
 107 #endif
 108     uint8_t* aligned_buf;
 109 } BDRVRawState;
 110
 111 static int posix_aio_init(void);
 112
 113 static int fd_open(BlockDriverState *bs);
 114
 115 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 116 {
 117     BDRVRawState *s = bs->opaque;
 118     int fd, open_flags, ret;
 119
 120     posix_aio_init();
 121
 122     s->lseek_err_cnt = 0;
 123
 124     open_flags = O_BINARY;
 125     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 126         open_flags |= O_RDWR;
 127     } else {
 128         open_flags |= O_RDONLY;
 129         bs->read_only = 1;
 130     }
 131     if (flags & BDRV_O_CREAT)
 132         open_flags |= O_CREAT | O_TRUNC;
 133
 134     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 135      * and O_DIRECT for no caching. */
 136     if ((flags & BDRV_O_NOCACHE))
 137         open_flags |= O_DIRECT;
 138     else if (!(flags & BDRV_O_CACHE_WB))
 139         open_flags |= O_DSYNC;
 140
 141     s->type = FTYPE_FILE;
 142
 143     fd = open(filename, open_flags, 0644);
 144     if (fd < 0) {
 145         ret = -errno;
 146         if (ret == -EROFS)
 147             ret = -EACCES;
 148         return ret;
 149     }
 150     s->fd = fd;
 151     s->aligned_buf = NULL;
 152     if ((flags & BDRV_O_NOCACHE)) {
 153         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 154         if (s->aligned_buf == NULL) {
 155             ret = -errno;
 156             close(fd);
 157             return ret;
 158         }
 159     }
 160     return 0;
 161 }
 162
 163 /* XXX: use host sector size if necessary with:
 164 #ifdef DIOCGSECTORSIZE
 165         {
 166             unsigned int sectorsize = 512;
 167             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 168                 sectorsize > bufsize)
 169                 bufsize = sectorsize;
 170         }
 171 #endif
 172 #ifdef CONFIG_COCOA
 173         u_int32_t   blockSize = 512;
 174         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 175             bufsize = blockSize;
 176         }
 177 #endif
 178 */
 179
 180 /*
 181  * offset and count are in bytes, but must be multiples of 512 for files
 182  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 183  *
 184  * This function may be called without alignment if the caller ensures
 185  * that O_DIRECT is not in effect.
 186  */
 187 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 188                      uint8_t *buf, int count)
 189 {
 190     BDRVRawState *s = bs->opaque;
 191     int ret;
 192
 193     ret = fd_open(bs);
 194     if (ret < 0)
 195         return ret;
 196
 197     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 198         ++(s->lseek_err_cnt);
 199         if(s->lseek_err_cnt <= 10) {
 200             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 201                               "] lseek failed : %d = %s\n",
 202                               s->fd, bs->filename, offset, buf, count,
 203                               bs->total_sectors, errno, strerror(errno));
 204         }
 205         return -1;
 206     }
 207     s->lseek_err_cnt=0;
 208
 209     ret = read(s->fd, buf, count);
 210     if (ret == count)
 211         goto label__raw_read__success;
 212
 213     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 214                       "] read failed %d : %d = %s\n",
 215                       s->fd, bs->filename, offset, buf, count,
 216                       bs->total_sectors, ret, errno, strerror(errno));
 217
 218     /* Try harder for CDrom. */
 219     if (bs->type == BDRV_TYPE_CDROM) {
 220         lseek(s->fd, offset, SEEK_SET);
 221         ret = read(s->fd, buf, count);
 222         if (ret == count)
 223             goto label__raw_read__success;
 224         lseek(s->fd, offset, SEEK_SET);
 225         ret = read(s->fd, buf, count);
 226         if (ret == count)
 227             goto label__raw_read__success;
 228
 229         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 230                           "] retry read failed %d : %d = %s\n",
 231                           s->fd, bs->filename, offset, buf, count,
 232                           bs->total_sectors, ret, errno, strerror(errno));
 233     }
 234
 235 label__raw_read__success:
 236
 237     return ret;
 238 }
 239
 240 /*
 241  * offset and count are in bytes, but must be multiples of 512 for files
 242  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 243  *
 244  * This function may be called without alignment if the caller ensures
 245  * that O_DIRECT is not in effect.
 246  */
 247 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 248                       const uint8_t *buf, int count)
 249 {
 250     BDRVRawState *s = bs->opaque;
 251     int ret;
 252
 253     ret = fd_open(bs);
 254     if (ret < 0)
 255         return -errno;
 256
 257     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 258         ++(s->lseek_err_cnt);
 259         if(s->lseek_err_cnt) {
 260             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 261                               PRId64 "] lseek failed : %d = %s\n",
 262                               s->fd, bs->filename, offset, buf, count,
 263                               bs->total_sectors, errno, strerror(errno));
 264         }
 265         return -EIO;
 266     }
 267     s->lseek_err_cnt = 0;
 268
 269     ret = write(s->fd, buf, count);
 270     if (ret == count)
 271         goto label__raw_write__success;
 272
 273     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 274                       "] write failed %d : %d = %s\n",
 275                       s->fd, bs->filename, offset, buf, count,
 276                       bs->total_sectors, ret, errno, strerror(errno));
 277
 278 label__raw_write__success:
 279
 280     return  (ret < 0) ? -errno : ret;
 281 }
 282
 283
 284 /*
 285  * offset and count are in bytes and possibly not aligned. For files opened
 286  * with O_DIRECT, necessary alignments are ensured before calling
 287  * raw_pread_aligned to do the actual read.
 288  */
 289 static int raw_pread(BlockDriverState *bs, int64_t offset,
 290                      uint8_t *buf, int count)
 291 {
 292     BDRVRawState *s = bs->opaque;
 293     int size, ret, shift, sum;
 294
 295     sum = 0;
 296
 297     if (s->aligned_buf != NULL)  {
 298
 299         if (offset & 0x1ff) {
 300             /* align offset on a 512 bytes boundary */
 301
 302             shift = offset & 0x1ff;
 303             size = (shift + count + 0x1ff) & ~0x1ff;
 304             if (size > ALIGNED_BUFFER_SIZE)
 305                 size = ALIGNED_BUFFER_SIZE;
 306             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 307             if (ret < 0)
 308                 return ret;
 309
 310             size = 512 - shift;
 311             if (size > count)
 312                 size = count;
 313             memcpy(buf, s->aligned_buf + shift, size);
 314
 315             buf += size;
 316             offset += size;
 317             count -= size;
 318             sum += size;
 319
 320             if (count == 0)
 321                 return sum;
 322         }
 323         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 324
 325             /* read on aligned buffer */
 326
 327             while (count) {
 328
 329                 size = (count + 0x1ff) & ~0x1ff;
 330                 if (size > ALIGNED_BUFFER_SIZE)
 331                     size = ALIGNED_BUFFER_SIZE;
 332
 333                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 334                 if (ret < 0)
 335                     return ret;
 336
 337                 size = ret;
 338                 if (size > count)
 339                     size = count;
 340
 341                 memcpy(buf, s->aligned_buf, size);
 342
 343                 buf += size;
 344                 offset += size;
 345                 count -= size;
 346                 sum += size;
 347             }
 348
 349             return sum;
 350         }
 351     }
 352
 353     return raw_pread_aligned(bs, offset, buf, count) + sum;
 354 }
 355
 356 /*
 357  * offset and count are in bytes and possibly not aligned. For files opened
 358  * with O_DIRECT, necessary alignments are ensured before calling
 359  * raw_pwrite_aligned to do the actual write.
 360  */
 361 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 362                       const uint8_t *buf, int count)
 363 {
 364     BDRVRawState *s = bs->opaque;
 365     int size, ret, shift, sum;
 366
 367     sum = 0;
 368
 369     if (s->aligned_buf != NULL) {
 370
 371         if (offset & 0x1ff) {
 372             /* align offset on a 512 bytes boundary */
 373             shift = offset & 0x1ff;
 374             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 375             if (ret < 0)
 376                 return ret;
 377
 378             size = 512 - shift;
 379             if (size > count)
 380                 size = count;
 381             memcpy(s->aligned_buf + shift, buf, size);
 382
 383             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 384             if (ret < 0)
 385                 return ret;
 386
 387             buf += size;
 388             offset += size;
 389             count -= size;
 390             sum += size;
 391
 392             if (count == 0)
 393                 return sum;
 394         }
 395         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 396
 397             while ((size = (count & ~0x1ff)) != 0) {
 398
 399                 if (size > ALIGNED_BUFFER_SIZE)
 400                     size = ALIGNED_BUFFER_SIZE;
 401
 402                 memcpy(s->aligned_buf, buf, size);
 403
 404                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 405                 if (ret < 0)
 406                     return ret;
 407
 408                 buf += ret;
 409                 offset += ret;
 410                 count -= ret;
 411                 sum += ret;
 412             }
 413             /* here, count < 512 because (count & ~0x1ff) == 0 */
 414             if (count) {
 415                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 416                 if (ret < 0)
 417                     return ret;
 418                  memcpy(s->aligned_buf, buf, count);
 419
 420                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 421                  if (ret < 0)
 422                      return ret;
 423                  if (count < ret)
 424                      ret = count;
 425
 426                  sum += ret;
 427             }
 428             return sum;
 429         }
 430     }
 431     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 432 }
 433
 434 #ifdef CONFIG_AIO
 435 /***********************************************************/
 436 /* Unix AIO using POSIX AIO */
 437
 438 typedef struct RawAIOCB {
 439     BlockDriverAIOCB common;
 440     struct qemu_paiocb aiocb;
 441     struct RawAIOCB *next;
 442     int ret;
 443 } RawAIOCB;
 444
 445 typedef struct PosixAioState
 446 {
 447     int rfd, wfd;
 448     RawAIOCB *first_aio;
 449 } PosixAioState;
 450
 451 static void posix_aio_read(void *opaque)
 452 {
 453     PosixAioState *s = opaque;
 454     RawAIOCB *acb, **pacb;
 455     int ret;
 456     ssize_t len;
 457
 458     /* read all bytes from signal pipe */
 459     for (;;) {
 460         char bytes[16];
 461
 462         len = read(s->rfd, bytes, sizeof(bytes));
 463         if (len == -1 && errno == EINTR)
 464             continue; /* try again */
 465         if (len == sizeof(bytes))
 466             continue; /* more to read */
 467         break;
 468     }
 469
 470     for(;;) {
 471         pacb = &s->first_aio;
 472         for(;;) {
 473             acb = *pacb;
 474             if (!acb)
 475                 goto the_end;
 476             ret = qemu_paio_error(&acb->aiocb);
 477             if (ret == ECANCELED) {
 478                 /* remove the request */
 479                 *pacb = acb->next;
 480                 qemu_aio_release(acb);
 481             } else if (ret != EINPROGRESS) {
 482                 /* end of aio */
 483                 if (ret == 0) {
 484                     ret = qemu_paio_return(&acb->aiocb);
 485                     if (ret == acb->aiocb.aio_nbytes)
 486                         ret = 0;
 487                     else
 488                         ret = -EINVAL;
 489                 } else {
 490                     ret = -ret;
 491                 }
 492                 /* remove the request */
 493                 *pacb = acb->next;
 494                 /* call the callback */
 495                 acb->common.cb(acb->common.opaque, ret);
 496                 qemu_aio_release(acb);
 497                 break;
 498             } else {
 499                 pacb = &acb->next;
 500             }
 501         }
 502     }
 503  the_end: ;
 504 }
 505
 506 static int posix_aio_flush(void *opaque)
 507 {
 508     PosixAioState *s = opaque;
 509     return !!s->first_aio;
 510 }
 511
 512 static PosixAioState *posix_aio_state;
 513
 514 static void aio_signal_handler(int signum)
 515 {
 516     if (posix_aio_state) {
 517         char byte = 0;
 518
 519         write(posix_aio_state->wfd, &byte, sizeof(byte));
 520     }
 521
 522     qemu_service_io();
 523 }
 524
 525 static int posix_aio_init(void)
 526 {
 527     struct sigaction act;
 528     PosixAioState *s;
 529     int fds[2];
 530     struct qemu_paioinit ai;
 531
 532     if (posix_aio_state)
 533         return 0;
 534
 535     s = qemu_malloc(sizeof(PosixAioState));
 536
 537     sigfillset(&act.sa_mask);
 538     act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
 539     act.sa_handler = aio_signal_handler;
 540     sigaction(SIGUSR2, &act, NULL);
 541
 542     s->first_aio = NULL;
 543     if (pipe(fds) == -1) {
 544         fprintf(stderr, "failed to create pipe\n");
 545         return -errno;
 546     }
 547
 548     s->rfd = fds[0];
 549     s->wfd = fds[1];
 550
 551     fcntl(s->rfd, F_SETFL, O_NONBLOCK);
 552     fcntl(s->wfd, F_SETFL, O_NONBLOCK);
 553
 554     qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
 555
 556     memset(&ai, 0, sizeof(ai));
 557     ai.aio_threads = 64;
 558     ai.aio_num = 64;
 559     qemu_paio_init(&ai);
 560
 561     posix_aio_state = s;
 562
 563     return 0;
 564 }
 565
 566 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 567         int64_t sector_num, uint8_t *buf, int nb_sectors,
 568         BlockDriverCompletionFunc *cb, void *opaque)
 569 {
 570     BDRVRawState *s = bs->opaque;
 571     RawAIOCB *acb;
 572
 573     if (fd_open(bs) < 0)
 574         return NULL;
 575
 576     acb = qemu_aio_get(bs, cb, opaque);
 577     if (!acb)
 578         return NULL;
 579     acb->aiocb.aio_fildes = s->fd;
 580     acb->aiocb.ev_signo = SIGUSR2;
 581     acb->aiocb.aio_buf = buf;
 582     if (nb_sectors < 0)
 583         acb->aiocb.aio_nbytes = -nb_sectors;
 584     else
 585         acb->aiocb.aio_nbytes = nb_sectors * 512;
 586     acb->aiocb.aio_offset = sector_num * 512;
 587     acb->next = posix_aio_state->first_aio;
 588     posix_aio_state->first_aio = acb;
 589     return acb;
 590 }
 591
 592 static void raw_aio_em_cb(void* opaque)
 593 {
 594     RawAIOCB *acb = opaque;
 595     acb->common.cb(acb->common.opaque, acb->ret);
 596     qemu_aio_release(acb);
 597 }
 598
 599 static void raw_aio_remove(RawAIOCB *acb)
 600 {
 601     RawAIOCB **pacb;
 602
 603     /* remove the callback from the queue */
 604     pacb = &posix_aio_state->first_aio;
 605     for(;;) {
 606         if (*pacb == NULL) {
 607             break;
 608         } else if (*pacb == acb) {
 609             *pacb = acb->next;
 610             qemu_aio_release(acb);
 611             break;
 612         }
 613         pacb = &acb->next;
 614     }
 615 }
 616
 617 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 618         int64_t sector_num, uint8_t *buf, int nb_sectors,
 619         BlockDriverCompletionFunc *cb, void *opaque)
 620 {
 621     RawAIOCB *acb;
 622
 623     /*
 624      * If O_DIRECT is used and the buffer is not aligned fall back
 625      * to synchronous IO.
 626      */
 627     BDRVRawState *s = bs->opaque;
 628
 629     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 630         QEMUBH *bh;
 631         acb = qemu_aio_get(bs, cb, opaque);
 632         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 633         bh = qemu_bh_new(raw_aio_em_cb, acb);
 634         qemu_bh_schedule(bh);
 635         return &acb->common;
 636     }
 637
 638     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 639     if (!acb)
 640         return NULL;
 641     if (qemu_paio_read(&acb->aiocb) < 0) {
 642         raw_aio_remove(acb);
 643         return NULL;
 644     }
 645     return &acb->common;
 646 }
 647
 648 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 649         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 650         BlockDriverCompletionFunc *cb, void *opaque)
 651 {
 652     RawAIOCB *acb;
 653
 654     /*
 655      * If O_DIRECT is used and the buffer is not aligned fall back
 656      * to synchronous IO.
 657      */
 658     BDRVRawState *s = bs->opaque;
 659
 660     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 661         QEMUBH *bh;
 662         acb = qemu_aio_get(bs, cb, opaque);
 663         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 664         bh = qemu_bh_new(raw_aio_em_cb, acb);
 665         qemu_bh_schedule(bh);
 666         return &acb->common;
 667     }
 668
 669     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 670     if (!acb)
 671         return NULL;
 672     if (qemu_paio_write(&acb->aiocb) < 0) {
 673         raw_aio_remove(acb);
 674         return NULL;
 675     }
 676     return &acb->common;
 677 }
 678
 679 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 680 {
 681     int ret;
 682     RawAIOCB *acb = (RawAIOCB *)blockacb;
 683
 684     ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 685     if (ret == QEMU_PAIO_NOTCANCELED) {
 686         /* fail safe: if the aio could not be canceled, we wait for
 687            it */
 688         while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
 689     }
 690
 691     raw_aio_remove(acb);
 692 }
 693 #else /* CONFIG_AIO */
 694 static int posix_aio_init(void)
 695 {
 696     return 0;
 697 }
 698 #endif /* CONFIG_AIO */
 699
 700
 701 static void raw_close(BlockDriverState *bs)
 702 {
 703     BDRVRawState *s = bs->opaque;
 704     if (s->fd >= 0) {
 705         close(s->fd);
 706         s->fd = -1;
 707         if (s->aligned_buf != NULL)
 708             qemu_free(s->aligned_buf);
 709     }
 710 }
 711
 712 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 713 {
 714     BDRVRawState *s = bs->opaque;
 715     if (s->type != FTYPE_FILE)
 716         return -ENOTSUP;
 717     if (ftruncate(s->fd, offset) < 0)
 718         return -errno;
 719     return 0;
 720 }
 721
 722 #ifdef __OpenBSD__
 723 static int64_t raw_getlength(BlockDriverState *bs)
 724 {
 725     BDRVRawState *s = bs->opaque;
 726     int fd = s->fd;
 727     struct stat st;
 728
 729     if (fstat(fd, &st))
 730         return -1;
 731     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 732         struct disklabel dl;
 733
 734         if (ioctl(fd, DIOCGDINFO, &dl))
 735             return -1;
 736         return (uint64_t)dl.d_secsize *
 737             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 738     } else
 739         return st.st_size;
 740 }
 741 #else /* !__OpenBSD__ */
 742 static int64_t  raw_getlength(BlockDriverState *bs)
 743 {
 744     BDRVRawState *s = bs->opaque;
 745     int fd = s->fd;
 746     int64_t size;
 747 #ifdef _BSD
 748     struct stat sb;
 749 #endif
 750 #ifdef __sun__
 751     struct dk_minfo minfo;
 752     int rv;
 753 #endif
 754     int ret;
 755
 756     ret = fd_open(bs);
 757     if (ret < 0)
 758         return ret;
 759
 760 #ifdef _BSD
 761     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 762 #ifdef DIOCGMEDIASIZE
 763         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 764 #endif
 765 #ifdef CONFIG_COCOA
 766         size = LONG_LONG_MAX;
 767 #else
 768         size = lseek(fd, 0LL, SEEK_END);
 769 #endif
 770     } else
 771 #endif
 772 #ifdef __sun__
 773     /*
 774      * use the DKIOCGMEDIAINFO ioctl to read the size.
 775      */
 776     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 777     if ( rv != -1 ) {
 778         size = minfo.dki_lbsize * minfo.dki_capacity;
 779     } else /* there are reports that lseek on some devices
 780               fails, but irc discussion said that contingency
 781               on contingency was overkill */
 782 #endif
 783     {
 784         size = lseek(fd, 0, SEEK_END);
 785     }
 786     return size;
 787 }
 788 #endif
 789
 790 static int raw_create(const char *filename, int64_t total_size,
 791                       const char *backing_file, int flags)
 792 {
 793     int fd;
 794
 795     if (flags || backing_file)
 796         return -ENOTSUP;
 797
 798     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 799               0644);
 800     if (fd < 0)
 801         return -EIO;
 802     ftruncate(fd, total_size * 512);
 803     close(fd);
 804     return 0;
 805 }
 806
 807 static void raw_flush(BlockDriverState *bs)
 808 {
 809     BDRVRawState *s = bs->opaque;
 810     fsync(s->fd);
 811 }
 812
 813 BlockDriver bdrv_raw = {
 814     "raw",
 815     sizeof(BDRVRawState),
 816     NULL, /* no probe for protocols */
 817     raw_open,
 818     NULL,
 819     NULL,
 820     raw_close,
 821     raw_create,
 822     raw_flush,
 823
 824 #ifdef CONFIG_AIO
 825     .bdrv_aio_read = raw_aio_read,
 826     .bdrv_aio_write = raw_aio_write,
 827     .bdrv_aio_cancel = raw_aio_cancel,
 828     .aiocb_size = sizeof(RawAIOCB),
 829 #endif
 830
 831     .bdrv_pread = raw_pread,
 832     .bdrv_pwrite = raw_pwrite,
 833     .bdrv_truncate = raw_truncate,
 834     .bdrv_getlength = raw_getlength,
 835 };
 836
 837 /***********************************************/
 838 /* host device */
 839
 840 #ifdef CONFIG_COCOA
 841 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 842 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 843
 844 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 845 {
 846     kern_return_t       kernResult;
 847     mach_port_t     masterPort;
 848     CFMutableDictionaryRef  classesToMatch;
 849
 850     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 851     if ( KERN_SUCCESS != kernResult ) {
 852         printf( "IOMasterPort returned %d\n", kernResult );
 853     }
 854
 855     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 856     if ( classesToMatch == NULL ) {
 857         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 858     } else {
 859     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 860     }
 861     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 862     if ( KERN_SUCCESS != kernResult )
 863     {
 864         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 865     }
 866
 867     return kernResult;
 868 }
 869
 870 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 871 {
 872     io_object_t     nextMedia;
 873     kern_return_t   kernResult = KERN_FAILURE;
 874     *bsdPath = '\0';
 875     nextMedia = IOIteratorNext( mediaIterator );
 876     if ( nextMedia )
 877     {
 878         CFTypeRef   bsdPathAsCFString;
 879     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 880         if ( bsdPathAsCFString ) {
 881             size_t devPathLength;
 882             strcpy( bsdPath, _PATH_DEV );
 883             strcat( bsdPath, "r" );
 884             devPathLength = strlen( bsdPath );
 885             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 886                 kernResult = KERN_SUCCESS;
 887             }
 888             CFRelease( bsdPathAsCFString );
 889         }
 890         IOObjectRelease( nextMedia );
 891     }
 892
 893     return kernResult;
 894 }
 895
 896 #endif
 897
 898 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 899 {
 900     BDRVRawState *s = bs->opaque;
 901     int fd, open_flags, ret;
 902
 903     posix_aio_init();
 904
 905 #ifdef CONFIG_COCOA
 906     if (strstart(filename, "/dev/cdrom", NULL)) {
 907         kern_return_t kernResult;
 908         io_iterator_t mediaIterator;
 909         char bsdPath[ MAXPATHLEN ];
 910         int fd;
 911
 912         kernResult = FindEjectableCDMedia( &mediaIterator );
 913         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 914
 915         if ( bsdPath[ 0 ] != '\0' ) {
 916             strcat(bsdPath,"s0");
 917             /* some CDs don't have a partition 0 */
 918             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 919             if (fd < 0) {
 920                 bsdPath[strlen(bsdPath)-1] = '1';
 921             } else {
 922                 close(fd);
 923             }
 924             filename = bsdPath;
 925         }
 926
 927         if ( mediaIterator )
 928             IOObjectRelease( mediaIterator );
 929     }
 930 #endif
 931     open_flags = O_BINARY;
 932     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 933         open_flags |= O_RDWR;
 934     } else {
 935         open_flags |= O_RDONLY;
 936         bs->read_only = 1;
 937     }
 938     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 939      * and O_DIRECT for no caching. */
 940     if ((flags & BDRV_O_NOCACHE))
 941         open_flags |= O_DIRECT;
 942     else if (!(flags & BDRV_O_CACHE_WB))
 943         open_flags |= O_DSYNC;
 944
 945     s->type = FTYPE_FILE;
 946 #if defined(__linux__)
 947     if (strstart(filename, "/dev/cd", NULL)) {
 948         /* open will not fail even if no CD is inserted */
 949         open_flags |= O_NONBLOCK;
 950         s->type = FTYPE_CD;
 951     } else if (strstart(filename, "/dev/fd", NULL)) {
 952         s->type = FTYPE_FD;
 953         s->fd_open_flags = open_flags;
 954         /* open will not fail even if no floppy is inserted */
 955         open_flags |= O_NONBLOCK;
 956     } else if (strstart(filename, "/dev/sg", NULL)) {
 957         bs->sg = 1;
 958     }
 959 #endif
 960     fd = open(filename, open_flags, 0644);
 961     if (fd < 0) {
 962         ret = -errno;
 963         if (ret == -EROFS)
 964             ret = -EACCES;
 965         return ret;
 966     }
 967     s->fd = fd;
 968 #if defined(__linux__)
 969     /* close fd so that we can reopen it as needed */
 970     if (s->type == FTYPE_FD) {
 971         close(s->fd);
 972         s->fd = -1;
 973         s->fd_media_changed = 1;
 974     }
 975 #endif
 976     return 0;
 977 }
 978
 979 #if defined(__linux__)
 980 /* Note: we do not have a reliable method to detect if the floppy is
 981    present. The current method is to try to open the floppy at every
 982    I/O and to keep it opened during a few hundreds of ms. */
 983 static int fd_open(BlockDriverState *bs)
 984 {
 985     BDRVRawState *s = bs->opaque;
 986     int last_media_present;
 987
 988     if (s->type != FTYPE_FD)
 989         return 0;
 990     last_media_present = (s->fd >= 0);
 991     if (s->fd >= 0 &&
 992         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
 993         close(s->fd);
 994         s->fd = -1;
 995 #ifdef DEBUG_FLOPPY
 996         printf("Floppy closed\n");
 997 #endif
 998     }
 999     if (s->fd < 0) {
1000         if (s->fd_got_error &&
1001             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1002 #ifdef DEBUG_FLOPPY
1003             printf("No floppy (open delayed)\n");
1004 #endif
1005             return -EIO;
1006         }
1007         s->fd = open(bs->filename, s->fd_open_flags);
1008         if (s->fd < 0) {
1009             s->fd_error_time = qemu_get_clock(rt_clock);
1010             s->fd_got_error = 1;
1011             if (last_media_present)
1012                 s->fd_media_changed = 1;
1013 #ifdef DEBUG_FLOPPY
1014             printf("No floppy\n");
1015 #endif
1016             return -EIO;
1017         }
1018 #ifdef DEBUG_FLOPPY
1019         printf("Floppy opened\n");
1020 #endif
1021     }
1022     if (!last_media_present)
1023         s->fd_media_changed = 1;
1024     s->fd_open_time = qemu_get_clock(rt_clock);
1025     s->fd_got_error = 0;
1026     return 0;
1027 }
1028
1029 static int raw_is_inserted(BlockDriverState *bs)
1030 {
1031     BDRVRawState *s = bs->opaque;
1032     int ret;
1033
1034     switch(s->type) {
1035     case FTYPE_CD:
1036         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1037         if (ret == CDS_DISC_OK)
1038             return 1;
1039         else
1040             return 0;
1041         break;
1042     case FTYPE_FD:
1043         ret = fd_open(bs);
1044         return (ret >= 0);
1045     default:
1046         return 1;
1047     }
1048 }
1049
1050 /* currently only used by fdc.c, but a CD version would be good too */
1051 static int raw_media_changed(BlockDriverState *bs)
1052 {
1053     BDRVRawState *s = bs->opaque;
1054
1055     switch(s->type) {
1056     case FTYPE_FD:
1057         {
1058             int ret;
1059             /* XXX: we do not have a true media changed indication. It
1060                does not work if the floppy is changed without trying
1061                to read it */
1062             fd_open(bs);
1063             ret = s->fd_media_changed;
1064             s->fd_media_changed = 0;
1065 #ifdef DEBUG_FLOPPY
1066             printf("Floppy changed=%d\n", ret);
1067 #endif
1068             return ret;
1069         }
1070     default:
1071         return -ENOTSUP;
1072     }
1073 }
1074
1075 static int raw_eject(BlockDriverState *bs, int eject_flag)
1076 {
1077     BDRVRawState *s = bs->opaque;
1078
1079     switch(s->type) {
1080     case FTYPE_CD:
1081         if (eject_flag) {
1082             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1083                 perror("CDROMEJECT");
1084         } else {
1085             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1086                 perror("CDROMEJECT");
1087         }
1088         break;
1089     case FTYPE_FD:
1090         {
1091             int fd;
1092             if (s->fd >= 0) {
1093                 close(s->fd);
1094                 s->fd = -1;
1095             }
1096             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1097             if (fd >= 0) {
1098                 if (ioctl(fd, FDEJECT, 0) < 0)
1099                     perror("FDEJECT");
1100                 close(fd);
1101             }
1102         }
1103         break;
1104     default:
1105         return -ENOTSUP;
1106     }
1107     return 0;
1108 }
1109
1110 static int raw_set_locked(BlockDriverState *bs, int locked)
1111 {
1112     BDRVRawState *s = bs->opaque;
1113
1114     switch(s->type) {
1115     case FTYPE_CD:
1116         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1117             /* Note: an error can happen if the distribution automatically
1118                mounts the CD-ROM */
1119             //        perror("CDROM_LOCKDOOR");
1120         }
1121         break;
1122     default:
1123         return -ENOTSUP;
1124     }
1125     return 0;
1126 }
1127
1128 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1129 {
1130     BDRVRawState *s = bs->opaque;
1131
1132     return ioctl(s->fd, req, buf);
1133 }
1134 #else
1135
1136 static int fd_open(BlockDriverState *bs)
1137 {
1138     return 0;
1139 }
1140
1141 static int raw_is_inserted(BlockDriverState *bs)
1142 {
1143     return 1;
1144 }
1145
1146 static int raw_media_changed(BlockDriverState *bs)
1147 {
1148     return -ENOTSUP;
1149 }
1150
1151 static int raw_eject(BlockDriverState *bs, int eject_flag)
1152 {
1153     return -ENOTSUP;
1154 }
1155
1156 static int raw_set_locked(BlockDriverState *bs, int locked)
1157 {
1158     return -ENOTSUP;
1159 }
1160
1161 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1162 {
1163     return -ENOTSUP;
1164 }
1165 #endif /* !linux */
1166
1167 BlockDriver bdrv_host_device = {
1168     "host_device",
1169     sizeof(BDRVRawState),
1170     NULL, /* no probe for protocols */
1171     hdev_open,
1172     NULL,
1173     NULL,
1174     raw_close,
1175     NULL,
1176     raw_flush,
1177
1178 #ifdef CONFIG_AIO
1179     .bdrv_aio_read = raw_aio_read,
1180     .bdrv_aio_write = raw_aio_write,
1181     .bdrv_aio_cancel = raw_aio_cancel,
1182     .aiocb_size = sizeof(RawAIOCB),
1183 #endif
1184
1185     .bdrv_pread = raw_pread,
1186     .bdrv_pwrite = raw_pwrite,
1187     .bdrv_getlength = raw_getlength,
1188
1189     /* removable device support */
1190     .bdrv_is_inserted = raw_is_inserted,
1191     .bdrv_media_changed = raw_media_changed,
1192     .bdrv_eject = raw_eject,
1193     .bdrv_set_locked = raw_set_locked,
1194     /* generic scsi device */
1195     .bdrv_ioctl = raw_ioctl,
1196 };