block-raw-posix.c

   1 /*
   2  * Block driver for RAW files (posix)
   3  *
   4  * Copyright (c) 2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
  26 #include "qemu-timer.h"
  27 #include "exec-all.h"
  28 #include "qemu-char.h"
  29 #endif
  30 #include "block_int.h"
  31 #include "compatfd.h"
  32 #include <assert.h>
  33 #ifdef CONFIG_AIO
  34 #include <aio.h>
  35 #endif
  36
  37 #ifdef CONFIG_COCOA
  38 #include <paths.h>
  39 #include <sys/param.h>
  40 #include <IOKit/IOKitLib.h>
  41 #include <IOKit/IOBSD.h>
  42 #include <IOKit/storage/IOMediaBSDClient.h>
  43 #include <IOKit/storage/IOMedia.h>
  44 #include <IOKit/storage/IOCDMedia.h>
  45 //#include <IOKit/storage/IOCDTypes.h>
  46 #include <CoreFoundation/CoreFoundation.h>
  47 #endif
  48
  49 #ifdef __sun__
  50 #define _POSIX_PTHREAD_SEMANTICS 1
  51 #include <signal.h>
  52 #include <sys/dkio.h>
  53 #endif
  54 #ifdef __linux__
  55 #include <sys/ioctl.h>
  56 #include <linux/cdrom.h>
  57 #include <linux/fd.h>
  58 #endif
  59 #ifdef __FreeBSD__
  60 #include <signal.h>
  61 #include <sys/disk.h>
  62 #endif
  63
  64 #ifdef __OpenBSD__
  65 #include <sys/ioctl.h>
  66 #include <sys/disklabel.h>
  67 #include <sys/dkio.h>
  68 #endif
  69
  70 //#define DEBUG_FLOPPY
  71
  72 //#define DEBUG_BLOCK
  73 #if defined(DEBUG_BLOCK) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
  74 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (loglevel != 0)  \
  75     { fprintf(logfile, formatCstr, ##args); fflush(logfile); } } while (0)
  76 #else
  77 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  78 #endif
  79
  80 #define FTYPE_FILE   0
  81 #define FTYPE_CD     1
  82 #define FTYPE_FD     2
  83
  84 #define ALIGNED_BUFFER_SIZE (32 * 512)
  85
  86 /* if the FD is not accessed during that time (in ms), we try to
  87    reopen it to see if the disk has been changed */
  88 #define FD_OPEN_TIMEOUT 1000
  89
  90 typedef struct BDRVRawState {
  91     int fd;
  92     int type;
  93     unsigned int lseek_err_cnt;
  94 #if defined(__linux__)
  95     /* linux floppy specific */
  96     int fd_open_flags;
  97     int64_t fd_open_time;
  98     int64_t fd_error_time;
  99     int fd_got_error;
 100     int fd_media_changed;
 101 #endif
 102 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 103     uint8_t* aligned_buf;
 104 #endif
 105 } BDRVRawState;
 106
 107 static int fd_open(BlockDriverState *bs);
 108
 109 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 110 {
 111     BDRVRawState *s = bs->opaque;
 112     int fd, open_flags, ret;
 113
 114     s->lseek_err_cnt = 0;
 115
 116     open_flags = O_BINARY;
 117     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 118         open_flags |= O_RDWR;
 119     } else {
 120         open_flags |= O_RDONLY;
 121         bs->read_only = 1;
 122     }
 123     if (flags & BDRV_O_CREAT)
 124         open_flags |= O_CREAT | O_TRUNC;
 125 #ifdef O_DIRECT
 126     if (flags & BDRV_O_DIRECT)
 127         open_flags |= O_DIRECT;
 128 #endif
 129
 130     s->type = FTYPE_FILE;
 131
 132     fd = open(filename, open_flags, 0644);
 133     if (fd < 0) {
 134         ret = -errno;
 135         if (ret == -EROFS)
 136             ret = -EACCES;
 137         return ret;
 138     }
 139     s->fd = fd;
 140 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 141     s->aligned_buf = NULL;
 142     if (flags & BDRV_O_DIRECT) {
 143         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 144         if (s->aligned_buf == NULL) {
 145             ret = -errno;
 146             close(fd);
 147             return ret;
 148         }
 149     }
 150 #endif
 151     return 0;
 152 }
 153
 154 /* XXX: use host sector size if necessary with:
 155 #ifdef DIOCGSECTORSIZE
 156         {
 157             unsigned int sectorsize = 512;
 158             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 159                 sectorsize > bufsize)
 160                 bufsize = sectorsize;
 161         }
 162 #endif
 163 #ifdef CONFIG_COCOA
 164         u_int32_t   blockSize = 512;
 165         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 166             bufsize = blockSize;
 167         }
 168 #endif
 169 */
 170
 171 /*
 172  * offset and count are in bytes, but must be multiples of 512 for files
 173  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 174  *
 175  * This function may be called without alignment if the caller ensures
 176  * that O_DIRECT is not in effect.
 177  */
 178 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 179                      uint8_t *buf, int count)
 180 {
 181     BDRVRawState *s = bs->opaque;
 182     int ret;
 183
 184     ret = fd_open(bs);
 185     if (ret < 0)
 186         return ret;
 187
 188     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 189         ++(s->lseek_err_cnt);
 190         if(s->lseek_err_cnt <= 10) {
 191             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 192                               "] lseek failed : %d = %s\n",
 193                               s->fd, bs->filename, offset, buf, count,
 194                               bs->total_sectors, errno, strerror(errno));
 195         }
 196         return -1;
 197     }
 198     s->lseek_err_cnt=0;
 199
 200     ret = read(s->fd, buf, count);
 201     if (ret == count)
 202         goto label__raw_read__success;
 203
 204     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 205                       "] read failed %d : %d = %s\n",
 206                       s->fd, bs->filename, offset, buf, count,
 207                       bs->total_sectors, ret, errno, strerror(errno));
 208
 209     /* Try harder for CDrom. */
 210     if (bs->type == BDRV_TYPE_CDROM) {
 211         lseek(s->fd, offset, SEEK_SET);
 212         ret = read(s->fd, buf, count);
 213         if (ret == count)
 214             goto label__raw_read__success;
 215         lseek(s->fd, offset, SEEK_SET);
 216         ret = read(s->fd, buf, count);
 217         if (ret == count)
 218             goto label__raw_read__success;
 219
 220         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 221                           "] retry read failed %d : %d = %s\n",
 222                           s->fd, bs->filename, offset, buf, count,
 223                           bs->total_sectors, ret, errno, strerror(errno));
 224     }
 225
 226 label__raw_read__success:
 227
 228     return ret;
 229 }
 230
 231 /*
 232  * offset and count are in bytes, but must be multiples of 512 for files
 233  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 234  *
 235  * This function may be called without alignment if the caller ensures
 236  * that O_DIRECT is not in effect.
 237  */
 238 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 239                       const uint8_t *buf, int count)
 240 {
 241     BDRVRawState *s = bs->opaque;
 242     int ret;
 243
 244     ret = fd_open(bs);
 245     if (ret < 0)
 246         return ret;
 247
 248     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 249         ++(s->lseek_err_cnt);
 250         if(s->lseek_err_cnt) {
 251             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 252                               PRId64 "] lseek failed : %d = %s\n",
 253                               s->fd, bs->filename, offset, buf, count,
 254                               bs->total_sectors, errno, strerror(errno));
 255         }
 256         return -1;
 257     }
 258     s->lseek_err_cnt = 0;
 259
 260     ret = write(s->fd, buf, count);
 261     if (ret == count)
 262         goto label__raw_write__success;
 263
 264     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 265                       "] write failed %d : %d = %s\n",
 266                       s->fd, bs->filename, offset, buf, count,
 267                       bs->total_sectors, ret, errno, strerror(errno));
 268
 269 label__raw_write__success:
 270
 271     return ret;
 272 }
 273
 274
 275 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 276 /*
 277  * offset and count are in bytes and possibly not aligned. For files opened
 278  * with O_DIRECT, necessary alignments are ensured before calling
 279  * raw_pread_aligned to do the actual read.
 280  */
 281 static int raw_pread(BlockDriverState *bs, int64_t offset,
 282                      uint8_t *buf, int count)
 283 {
 284     BDRVRawState *s = bs->opaque;
 285     int size, ret, shift, sum;
 286
 287     sum = 0;
 288
 289     if (s->aligned_buf != NULL)  {
 290
 291         if (offset & 0x1ff) {
 292             /* align offset on a 512 bytes boundary */
 293
 294             shift = offset & 0x1ff;
 295             size = (shift + count + 0x1ff) & ~0x1ff;
 296             if (size > ALIGNED_BUFFER_SIZE)
 297                 size = ALIGNED_BUFFER_SIZE;
 298             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 299             if (ret < 0)
 300                 return ret;
 301
 302             size = 512 - shift;
 303             if (size > count)
 304                 size = count;
 305             memcpy(buf, s->aligned_buf + shift, size);
 306
 307             buf += size;
 308             offset += size;
 309             count -= size;
 310             sum += size;
 311
 312             if (count == 0)
 313                 return sum;
 314         }
 315         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 316
 317             /* read on aligned buffer */
 318
 319             while (count) {
 320
 321                 size = (count + 0x1ff) & ~0x1ff;
 322                 if (size > ALIGNED_BUFFER_SIZE)
 323                     size = ALIGNED_BUFFER_SIZE;
 324
 325                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 326                 if (ret < 0)
 327                     return ret;
 328
 329                 size = ret;
 330                 if (size > count)
 331                     size = count;
 332
 333                 memcpy(buf, s->aligned_buf, size);
 334
 335                 buf += size;
 336                 offset += size;
 337                 count -= size;
 338                 sum += size;
 339             }
 340
 341             return sum;
 342         }
 343     }
 344
 345     return raw_pread_aligned(bs, offset, buf, count) + sum;
 346 }
 347
 348 /*
 349  * offset and count are in bytes and possibly not aligned. For files opened
 350  * with O_DIRECT, necessary alignments are ensured before calling
 351  * raw_pwrite_aligned to do the actual write.
 352  */
 353 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 354                       const uint8_t *buf, int count)
 355 {
 356     BDRVRawState *s = bs->opaque;
 357     int size, ret, shift, sum;
 358
 359     sum = 0;
 360
 361     if (s->aligned_buf != NULL) {
 362
 363         if (offset & 0x1ff) {
 364             /* align offset on a 512 bytes boundary */
 365             shift = offset & 0x1ff;
 366             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 367             if (ret < 0)
 368                 return ret;
 369
 370             size = 512 - shift;
 371             if (size > count)
 372                 size = count;
 373             memcpy(s->aligned_buf + shift, buf, size);
 374
 375             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 376             if (ret < 0)
 377                 return ret;
 378
 379             buf += size;
 380             offset += size;
 381             count -= size;
 382             sum += size;
 383
 384             if (count == 0)
 385                 return sum;
 386         }
 387         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 388
 389             while ((size = (count & ~0x1ff)) != 0) {
 390
 391                 if (size > ALIGNED_BUFFER_SIZE)
 392                     size = ALIGNED_BUFFER_SIZE;
 393
 394                 memcpy(s->aligned_buf, buf, size);
 395
 396                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 397                 if (ret < 0)
 398                     return ret;
 399
 400                 buf += ret;
 401                 offset += ret;
 402                 count -= ret;
 403                 sum += ret;
 404             }
 405             /* here, count < 512 because (count & ~0x1ff) == 0 */
 406             if (count) {
 407                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 408                 if (ret < 0)
 409                     return ret;
 410                  memcpy(s->aligned_buf, buf, count);
 411
 412                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 413                  if (ret < 0)
 414                      return ret;
 415                  if (count < ret)
 416                      ret = count;
 417
 418                  sum += ret;
 419             }
 420             return sum;
 421         }
 422     }
 423     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 424 }
 425
 426 #else
 427 #define raw_pread raw_pread_aligned
 428 #define raw_pwrite raw_pwrite_aligned
 429 #endif
 430
 431
 432 #ifdef CONFIG_AIO
 433 /***********************************************************/
 434 /* Unix AIO using POSIX AIO */
 435
 436 typedef struct RawAIOCB {
 437     BlockDriverAIOCB common;
 438     struct aiocb aiocb;
 439     struct RawAIOCB *next;
 440     int ret;
 441 } RawAIOCB;
 442
 443 static int aio_sig_fd = -1;
 444 static int aio_sig_num = SIGUSR2;
 445 static RawAIOCB *first_aio; /* AIO issued */
 446 static int aio_initialized = 0;
 447
 448 static void qemu_aio_poll(void *opaque)
 449 {
 450     RawAIOCB *acb, **pacb;
 451     int ret;
 452     size_t offset;
 453     union {
 454         struct qemu_signalfd_siginfo siginfo;
 455         char buf[128];
 456     } sig;
 457
 458     /* try to read from signalfd, don't freak out if we can't read anything */
 459     offset = 0;
 460     while (offset < 128) {
 461         ssize_t len;
 462
 463         len = read(aio_sig_fd, sig.buf + offset, 128 - offset);
 464         if (len == -1 && errno == EINTR)
 465             continue;
 466         if (len == -1 && errno == EAGAIN) {
 467             /* there is no natural reason for this to happen,
 468              * so we'll spin hard until we get everything just
 469              * to be on the safe side. */
 470             if (offset > 0)
 471                 continue;
 472         }
 473
 474         offset += len;
 475     }
 476
 477     for(;;) {
 478         pacb = &first_aio;
 479         for(;;) {
 480             acb = *pacb;
 481             if (!acb)
 482                 goto the_end;
 483             ret = aio_error(&acb->aiocb);
 484             if (ret == ECANCELED) {
 485                 /* remove the request */
 486                 *pacb = acb->next;
 487                 qemu_aio_release(acb);
 488             } else if (ret != EINPROGRESS) {
 489                 /* end of aio */
 490                 if (ret == 0) {
 491                     ret = aio_return(&acb->aiocb);
 492                     if (ret == acb->aiocb.aio_nbytes)
 493                         ret = 0;
 494                     else
 495                         ret = -EINVAL;
 496                 } else {
 497                     ret = -ret;
 498                 }
 499                 /* remove the request */
 500                 *pacb = acb->next;
 501                 /* call the callback */
 502                 acb->common.cb(acb->common.opaque, ret);
 503                 qemu_aio_release(acb);
 504                 break;
 505             } else {
 506                 pacb = &acb->next;
 507             }
 508         }
 509     }
 510  the_end: ;
 511 }
 512
 513 void qemu_aio_init(void)
 514 {
 515     sigset_t mask;
 516
 517     aio_initialized = 1;
 518
 519     /* Make sure to block AIO signal */
 520     sigemptyset(&mask);
 521     sigaddset(&mask, aio_sig_num);
 522     sigprocmask(SIG_BLOCK, &mask, NULL);
 523
 524     aio_sig_fd = qemu_signalfd(&mask);
 525
 526     fcntl(aio_sig_fd, F_SETFL, O_NONBLOCK);
 527
 528 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 529     qemu_set_fd_handler2(aio_sig_fd, NULL, qemu_aio_poll, NULL, NULL);
 530 #endif
 531
 532 #if defined(__GLIBC__) && defined(__linux__)
 533     {
 534         /* XXX: aio thread exit seems to hang on RedHat 9 and this init
 535            seems to fix the problem. */
 536         struct aioinit ai;
 537         memset(&ai, 0, sizeof(ai));
 538         ai.aio_threads = 1;
 539         ai.aio_num = 1;
 540         ai.aio_idle_time = 365 * 100000;
 541         aio_init(&ai);
 542     }
 543 #endif
 544 }
 545
 546 /* Wait for all IO requests to complete.  */
 547 void qemu_aio_flush(void)
 548 {
 549     qemu_aio_poll(NULL);
 550     while (first_aio) {
 551         qemu_aio_wait();
 552     }
 553 }
 554
 555 void qemu_aio_wait(void)
 556 {
 557     int ret;
 558
 559 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 560     if (qemu_bh_poll())
 561         return;
 562 #endif
 563
 564     do {
 565         fd_set rdfds;
 566
 567         FD_ZERO(&rdfds);
 568         FD_SET(aio_sig_fd, &rdfds);
 569
 570         ret = select(aio_sig_fd + 1, &rdfds, NULL, NULL, NULL);
 571         if (ret == -1 && errno == EINTR)
 572             continue;
 573     } while (ret == 0);
 574
 575     qemu_aio_poll(NULL);
 576 }
 577
 578 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 579         int64_t sector_num, uint8_t *buf, int nb_sectors,
 580         BlockDriverCompletionFunc *cb, void *opaque)
 581 {
 582     BDRVRawState *s = bs->opaque;
 583     RawAIOCB *acb;
 584
 585     if (fd_open(bs) < 0)
 586         return NULL;
 587
 588     acb = qemu_aio_get(bs, cb, opaque);
 589     if (!acb)
 590         return NULL;
 591     acb->aiocb.aio_fildes = s->fd;
 592     acb->aiocb.aio_sigevent.sigev_signo = aio_sig_num;
 593     acb->aiocb.aio_sigevent.sigev_notify = SIGEV_SIGNAL;
 594     acb->aiocb.aio_buf = buf;
 595     if (nb_sectors < 0)
 596         acb->aiocb.aio_nbytes = -nb_sectors;
 597     else
 598         acb->aiocb.aio_nbytes = nb_sectors * 512;
 599     acb->aiocb.aio_offset = sector_num * 512;
 600     acb->next = first_aio;
 601     first_aio = acb;
 602     return acb;
 603 }
 604
 605 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 606 static void raw_aio_em_cb(void* opaque)
 607 {
 608     RawAIOCB *acb = opaque;
 609     acb->common.cb(acb->common.opaque, acb->ret);
 610     qemu_aio_release(acb);
 611 }
 612 #endif
 613
 614 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 615         int64_t sector_num, uint8_t *buf, int nb_sectors,
 616         BlockDriverCompletionFunc *cb, void *opaque)
 617 {
 618     RawAIOCB *acb;
 619
 620     /*
 621      * If O_DIRECT is used and the buffer is not aligned fall back
 622      * to synchronous IO.
 623      */
 624 #if defined(O_DIRECT) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
 625     BDRVRawState *s = bs->opaque;
 626
 627     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 628         QEMUBH *bh;
 629         acb = qemu_aio_get(bs, cb, opaque);
 630         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 631         bh = qemu_bh_new(raw_aio_em_cb, acb);
 632         qemu_bh_schedule(bh);
 633         return &acb->common;
 634     }
 635 #endif
 636
 637     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 638     if (!acb)
 639         return NULL;
 640     if (aio_read(&acb->aiocb) < 0) {
 641         qemu_aio_release(acb);
 642         return NULL;
 643     }
 644     return &acb->common;
 645 }
 646
 647 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 648         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 649         BlockDriverCompletionFunc *cb, void *opaque)
 650 {
 651     RawAIOCB *acb;
 652
 653     /*
 654      * If O_DIRECT is used and the buffer is not aligned fall back
 655      * to synchronous IO.
 656      */
 657 #if defined(O_DIRECT) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
 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 #endif
 669
 670     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 671     if (!acb)
 672         return NULL;
 673     if (aio_write(&acb->aiocb) < 0) {
 674         qemu_aio_release(acb);
 675         return NULL;
 676     }
 677     return &acb->common;
 678 }
 679
 680 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 681 {
 682     int ret;
 683     RawAIOCB *acb = (RawAIOCB *)blockacb;
 684     RawAIOCB **pacb;
 685
 686     ret = aio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 687     if (ret == AIO_NOTCANCELED) {
 688         /* fail safe: if the aio could not be canceled, we wait for
 689            it */
 690         while (aio_error(&acb->aiocb) == EINPROGRESS);
 691     }
 692
 693     /* remove the callback from the queue */
 694     pacb = &first_aio;
 695     for(;;) {
 696         if (*pacb == NULL) {
 697             break;
 698         } else if (*pacb == acb) {
 699             *pacb = acb->next;
 700             qemu_aio_release(acb);
 701             break;
 702         }
 703         pacb = &acb->next;
 704     }
 705 }
 706
 707 # else /* CONFIG_AIO */
 708
 709 void qemu_aio_init(void)
 710 {
 711 }
 712
 713 void qemu_aio_flush(void)
 714 {
 715 }
 716
 717 void qemu_aio_wait(void)
 718 {
 719 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 720     qemu_bh_poll();
 721 #endif
 722 }
 723
 724 #endif /* CONFIG_AIO */
 725
 726 static void raw_close(BlockDriverState *bs)
 727 {
 728     BDRVRawState *s = bs->opaque;
 729     if (s->fd >= 0) {
 730         close(s->fd);
 731         s->fd = -1;
 732 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 733         if (s->aligned_buf != NULL)
 734             qemu_free(s->aligned_buf);
 735 #endif
 736     }
 737 }
 738
 739 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 740 {
 741     BDRVRawState *s = bs->opaque;
 742     if (s->type != FTYPE_FILE)
 743         return -ENOTSUP;
 744     if (ftruncate(s->fd, offset) < 0)
 745         return -errno;
 746     return 0;
 747 }
 748
 749 #ifdef __OpenBSD__
 750 static int64_t raw_getlength(BlockDriverState *bs)
 751 {
 752     BDRVRawState *s = bs->opaque;
 753     int fd = s->fd;
 754     struct stat st;
 755
 756     if (fstat(fd, &st))
 757         return -1;
 758     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 759         struct disklabel dl;
 760
 761         if (ioctl(fd, DIOCGDINFO, &dl))
 762             return -1;
 763         return (uint64_t)dl.d_secsize *
 764             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 765     } else
 766         return st.st_size;
 767 }
 768 #else /* !__OpenBSD__ */
 769 static int64_t  raw_getlength(BlockDriverState *bs)
 770 {
 771     BDRVRawState *s = bs->opaque;
 772     int fd = s->fd;
 773     int64_t size;
 774 #ifdef _BSD
 775     struct stat sb;
 776 #endif
 777 #ifdef __sun__
 778     struct dk_minfo minfo;
 779     int rv;
 780 #endif
 781     int ret;
 782
 783     ret = fd_open(bs);
 784     if (ret < 0)
 785         return ret;
 786
 787 #ifdef _BSD
 788     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 789 #ifdef DIOCGMEDIASIZE
 790         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 791 #endif
 792 #ifdef CONFIG_COCOA
 793         size = LONG_LONG_MAX;
 794 #else
 795         size = lseek(fd, 0LL, SEEK_END);
 796 #endif
 797     } else
 798 #endif
 799 #ifdef __sun__
 800     /*
 801      * use the DKIOCGMEDIAINFO ioctl to read the size.
 802      */
 803     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 804     if ( rv != -1 ) {
 805         size = minfo.dki_lbsize * minfo.dki_capacity;
 806     } else /* there are reports that lseek on some devices
 807               fails, but irc discussion said that contingency
 808               on contingency was overkill */
 809 #endif
 810     {
 811         size = lseek(fd, 0, SEEK_END);
 812     }
 813     return size;
 814 }
 815 #endif
 816
 817 static int raw_create(const char *filename, int64_t total_size,
 818                       const char *backing_file, int flags)
 819 {
 820     int fd;
 821
 822     if (flags || backing_file)
 823         return -ENOTSUP;
 824
 825     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 826               0644);
 827     if (fd < 0)
 828         return -EIO;
 829     ftruncate(fd, total_size * 512);
 830     close(fd);
 831     return 0;
 832 }
 833
 834 static void raw_flush(BlockDriverState *bs)
 835 {
 836     BDRVRawState *s = bs->opaque;
 837     fsync(s->fd);
 838 }
 839
 840 BlockDriver bdrv_raw = {
 841     "raw",
 842     sizeof(BDRVRawState),
 843     NULL, /* no probe for protocols */
 844     raw_open,
 845     NULL,
 846     NULL,
 847     raw_close,
 848     raw_create,
 849     raw_flush,
 850
 851 #ifdef CONFIG_AIO
 852     .bdrv_aio_read = raw_aio_read,
 853     .bdrv_aio_write = raw_aio_write,
 854     .bdrv_aio_cancel = raw_aio_cancel,
 855     .aiocb_size = sizeof(RawAIOCB),
 856 #endif
 857     .protocol_name = "file",
 858     .bdrv_pread = raw_pread,
 859     .bdrv_pwrite = raw_pwrite,
 860     .bdrv_truncate = raw_truncate,
 861     .bdrv_getlength = raw_getlength,
 862 };
 863
 864 /***********************************************/
 865 /* host device */
 866
 867 #ifdef CONFIG_COCOA
 868 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 869 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 870
 871 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 872 {
 873     kern_return_t       kernResult;
 874     mach_port_t     masterPort;
 875     CFMutableDictionaryRef  classesToMatch;
 876
 877     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 878     if ( KERN_SUCCESS != kernResult ) {
 879         printf( "IOMasterPort returned %d\n", kernResult );
 880     }
 881
 882     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 883     if ( classesToMatch == NULL ) {
 884         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 885     } else {
 886     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 887     }
 888     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 889     if ( KERN_SUCCESS != kernResult )
 890     {
 891         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 892     }
 893
 894     return kernResult;
 895 }
 896
 897 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 898 {
 899     io_object_t     nextMedia;
 900     kern_return_t   kernResult = KERN_FAILURE;
 901     *bsdPath = '\0';
 902     nextMedia = IOIteratorNext( mediaIterator );
 903     if ( nextMedia )
 904     {
 905         CFTypeRef   bsdPathAsCFString;
 906     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 907         if ( bsdPathAsCFString ) {
 908             size_t devPathLength;
 909             strcpy( bsdPath, _PATH_DEV );
 910             strcat( bsdPath, "r" );
 911             devPathLength = strlen( bsdPath );
 912             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 913                 kernResult = KERN_SUCCESS;
 914             }
 915             CFRelease( bsdPathAsCFString );
 916         }
 917         IOObjectRelease( nextMedia );
 918     }
 919
 920     return kernResult;
 921 }
 922
 923 #endif
 924
 925 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 926 {
 927     BDRVRawState *s = bs->opaque;
 928     int fd, open_flags, ret;
 929
 930 #ifdef CONFIG_COCOA
 931     if (strstart(filename, "/dev/cdrom", NULL)) {
 932         kern_return_t kernResult;
 933         io_iterator_t mediaIterator;
 934         char bsdPath[ MAXPATHLEN ];
 935         int fd;
 936
 937         kernResult = FindEjectableCDMedia( &mediaIterator );
 938         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 939
 940         if ( bsdPath[ 0 ] != '\0' ) {
 941             strcat(bsdPath,"s0");
 942             /* some CDs don't have a partition 0 */
 943             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 944             if (fd < 0) {
 945                 bsdPath[strlen(bsdPath)-1] = '1';
 946             } else {
 947                 close(fd);
 948             }
 949             filename = bsdPath;
 950         }
 951
 952         if ( mediaIterator )
 953             IOObjectRelease( mediaIterator );
 954     }
 955 #endif
 956     open_flags = O_BINARY;
 957     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 958         open_flags |= O_RDWR;
 959     } else {
 960         open_flags |= O_RDONLY;
 961         bs->read_only = 1;
 962     }
 963 #ifdef O_DIRECT
 964     if (flags & BDRV_O_DIRECT)
 965         open_flags |= O_DIRECT;
 966 #endif
 967
 968     s->type = FTYPE_FILE;
 969 #if defined(__linux__)
 970     if (strstart(filename, "/dev/cd", NULL)) {
 971         /* open will not fail even if no CD is inserted */
 972         open_flags |= O_NONBLOCK;
 973         s->type = FTYPE_CD;
 974     } else if (strstart(filename, "/dev/fd", NULL)) {
 975         s->type = FTYPE_FD;
 976         s->fd_open_flags = open_flags;
 977         /* open will not fail even if no floppy is inserted */
 978         open_flags |= O_NONBLOCK;
 979     } else if (strstart(filename, "/dev/sg", NULL)) {
 980         bs->sg = 1;
 981     }
 982 #endif
 983     fd = open(filename, open_flags, 0644);
 984     if (fd < 0) {
 985         ret = -errno;
 986         if (ret == -EROFS)
 987             ret = -EACCES;
 988         return ret;
 989     }
 990     s->fd = fd;
 991 #if defined(__linux__)
 992     /* close fd so that we can reopen it as needed */
 993     if (s->type == FTYPE_FD) {
 994         close(s->fd);
 995         s->fd = -1;
 996         s->fd_media_changed = 1;
 997     }
 998 #endif
 999     return 0;
1000 }
1001
1002 #if defined(__linux__) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
1003
1004 /* Note: we do not have a reliable method to detect if the floppy is
1005    present. The current method is to try to open the floppy at every
1006    I/O and to keep it opened during a few hundreds of ms. */
1007 static int fd_open(BlockDriverState *bs)
1008 {
1009     BDRVRawState *s = bs->opaque;
1010     int last_media_present;
1011
1012     if (s->type != FTYPE_FD)
1013         return 0;
1014     last_media_present = (s->fd >= 0);
1015     if (s->fd >= 0 &&
1016         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1017         close(s->fd);
1018         s->fd = -1;
1019 #ifdef DEBUG_FLOPPY
1020         printf("Floppy closed\n");
1021 #endif
1022     }
1023     if (s->fd < 0) {
1024         if (s->fd_got_error &&
1025             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1026 #ifdef DEBUG_FLOPPY
1027             printf("No floppy (open delayed)\n");
1028 #endif
1029             return -EIO;
1030         }
1031         s->fd = open(bs->filename, s->fd_open_flags);
1032         if (s->fd < 0) {
1033             s->fd_error_time = qemu_get_clock(rt_clock);
1034             s->fd_got_error = 1;
1035             if (last_media_present)
1036                 s->fd_media_changed = 1;
1037 #ifdef DEBUG_FLOPPY
1038             printf("No floppy\n");
1039 #endif
1040             return -EIO;
1041         }
1042 #ifdef DEBUG_FLOPPY
1043         printf("Floppy opened\n");
1044 #endif
1045     }
1046     if (!last_media_present)
1047         s->fd_media_changed = 1;
1048     s->fd_open_time = qemu_get_clock(rt_clock);
1049     s->fd_got_error = 0;
1050     return 0;
1051 }
1052 #else
1053 static int fd_open(BlockDriverState *bs)
1054 {
1055     return 0;
1056 }
1057 #endif
1058
1059 #if defined(__linux__)
1060
1061 static int raw_is_inserted(BlockDriverState *bs)
1062 {
1063     BDRVRawState *s = bs->opaque;
1064     int ret;
1065
1066     switch(s->type) {
1067     case FTYPE_CD:
1068         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1069         if (ret == CDS_DISC_OK)
1070             return 1;
1071         else
1072             return 0;
1073         break;
1074     case FTYPE_FD:
1075         ret = fd_open(bs);
1076         return (ret >= 0);
1077     default:
1078         return 1;
1079     }
1080 }
1081
1082 /* currently only used by fdc.c, but a CD version would be good too */
1083 static int raw_media_changed(BlockDriverState *bs)
1084 {
1085     BDRVRawState *s = bs->opaque;
1086
1087     switch(s->type) {
1088     case FTYPE_FD:
1089         {
1090             int ret;
1091             /* XXX: we do not have a true media changed indication. It
1092                does not work if the floppy is changed without trying
1093                to read it */
1094             fd_open(bs);
1095             ret = s->fd_media_changed;
1096             s->fd_media_changed = 0;
1097 #ifdef DEBUG_FLOPPY
1098             printf("Floppy changed=%d\n", ret);
1099 #endif
1100             return ret;
1101         }
1102     default:
1103         return -ENOTSUP;
1104     }
1105 }
1106
1107 static int raw_eject(BlockDriverState *bs, int eject_flag)
1108 {
1109     BDRVRawState *s = bs->opaque;
1110
1111     switch(s->type) {
1112     case FTYPE_CD:
1113         if (eject_flag) {
1114             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1115                 perror("CDROMEJECT");
1116         } else {
1117             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1118                 perror("CDROMEJECT");
1119         }
1120         break;
1121     case FTYPE_FD:
1122         {
1123             int fd;
1124             if (s->fd >= 0) {
1125                 close(s->fd);
1126                 s->fd = -1;
1127             }
1128             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1129             if (fd >= 0) {
1130                 if (ioctl(fd, FDEJECT, 0) < 0)
1131                     perror("FDEJECT");
1132                 close(fd);
1133             }
1134         }
1135         break;
1136     default:
1137         return -ENOTSUP;
1138     }
1139     return 0;
1140 }
1141
1142 static int raw_set_locked(BlockDriverState *bs, int locked)
1143 {
1144     BDRVRawState *s = bs->opaque;
1145
1146     switch(s->type) {
1147     case FTYPE_CD:
1148         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1149             /* Note: an error can happen if the distribution automatically
1150                mounts the CD-ROM */
1151             //        perror("CDROM_LOCKDOOR");
1152         }
1153         break;
1154     default:
1155         return -ENOTSUP;
1156     }
1157     return 0;
1158 }
1159
1160 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1161 {
1162     BDRVRawState *s = bs->opaque;
1163
1164     return ioctl(s->fd, req, buf);
1165 }
1166 #else
1167
1168 static int raw_is_inserted(BlockDriverState *bs)
1169 {
1170     return 1;
1171 }
1172
1173 static int raw_media_changed(BlockDriverState *bs)
1174 {
1175     return -ENOTSUP;
1176 }
1177
1178 static int raw_eject(BlockDriverState *bs, int eject_flag)
1179 {
1180     return -ENOTSUP;
1181 }
1182
1183 static int raw_set_locked(BlockDriverState *bs, int locked)
1184 {
1185     return -ENOTSUP;
1186 }
1187
1188 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1189 {
1190     return -ENOTSUP;
1191 }
1192 #endif /* !linux */
1193
1194 BlockDriver bdrv_host_device = {
1195     "host_device",
1196     sizeof(BDRVRawState),
1197     NULL, /* no probe for protocols */
1198     hdev_open,
1199     NULL,
1200     NULL,
1201     raw_close,
1202     NULL,
1203     raw_flush,
1204
1205 #ifdef CONFIG_AIO
1206     .bdrv_aio_read = raw_aio_read,
1207     .bdrv_aio_write = raw_aio_write,
1208     .bdrv_aio_cancel = raw_aio_cancel,
1209     .aiocb_size = sizeof(RawAIOCB),
1210 #endif
1211     .bdrv_pread = raw_pread,
1212     .bdrv_pwrite = raw_pwrite,
1213     .bdrv_getlength = raw_getlength,
1214
1215     /* removable device support */
1216     .bdrv_is_inserted = raw_is_inserted,
1217     .bdrv_media_changed = raw_media_changed,
1218     .bdrv_eject = raw_eject,
1219     .bdrv_set_locked = raw_set_locked,
1220     /* generic scsi device */
1221     .bdrv_ioctl = raw_ioctl,
1222 };